ii..il...;! . . ' :-.. ., ' . . ...'-. .... -7, :Y,;; :,;7 ,j-ii , !I.: _ : ... I .1i$* ~ ~ ..... .;s! .: . NO41VqWIRO ~3S~ NVG 6DQ UNVNV- iHa .nl s- puv~~~~~~~~~~~~~~~~~~~~~~~. ., .... ...--..!."-t2-I.Eij E UO!llS'OdUIO~~~~~~~~~~~~~~~~~~~~~~~~~1 . ,q ,,- t- n:-! ;4jjjj j '' ' - -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~.-a . - a-.-C ..... .. ;,. PUT X1pnbou - t... .. .. r. ii.. ! . . jz .-#" _ *t j,;jj. i F i v , , _, ; _ _ 7s,, : 1 . ..1- - != - j;si, ,*4 ,+ .e- i t . i i - --. ! P:i: si, .... ., i.,j~~~~~~~~~~~~~~~~~~.... . . .. 89Z6 -- -- --w =n----g3t -w- ----i-u'sL-Nl!NryS Mm- i ;Mi 1 w - ii, it; 3 ---ia . ... ..... ----- - -- - ------ .......r Inequality and Poverty in Malaysia Measurement and Decomposition A World Bank Research Publication Peninsular Malaysia PES Regions, Metropolitan Towns, and Towns REGIONS: North \ 1- - 1| Nrmhwest lS m a\~~~~~~~~Z = c ielra! 0 T)\ 3 .M¢)olinoTwn, - Statr Boundar,i \ / b _._ International Boundanes PERIIS - &° '' 4 . ? ~~~~P.l BB 8h. KEDAH . ._6 TanjoogTokong6\$ i _9 { ~ § Gootrgorotook 0 lti rtt- PENANG| l / 7- / iKEL'NAN l '(REG , 'H X P E R A K-E SANU \ r. z ; - +, : tA PA 4N ' t SELANOR( roka 0~~~~ _- ' S Bi.L ' I INDON.ESI / ' (~~ ~ ~ 20 40 . 0Bnog o t 0 20 40 60 an I&) KILOMETERS S.9NGP i dE na _(. Inequality and Poverty in Malaysia Measurement and Decomposition Sudhir Anand Published for The World Bank OXFORD UNIVERSITY PRESS Oxford University Press NEW YORK OXFORD LONDON GLASGOW TORONTO MELBOURNE WELLINGTON HONG KONG TOKYO KUALA LUMPUR SINGAPORE JAKARTA DELHI BOMBAY CALCUTTA MADRAS KARACHI NAIROBI DAR ES SALAAM CAPE TOWN ©) 1983 by the International Bank for Reconstruction and Development / The World Bank 1818 H Street, N.W., Washington, D.C. 20433, U.S.A. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Manufactured in the United States of America. The views and interpretations in this book are the author's and should not be attributed to the World Bank, to its affiliated organizations, or to any individual acting in their behalf. The map serving as frontispiece has been prepared for the convenience of readers of this book; the denominations used and the boundaries shown do not imply, on the part of the World Bank and its affiliates, any judgment on the legal status of any territory or any endorsement or acceptance of such boundaries. Editor: Jane H. Carroll Figures: S. A. D. Subasinghe Map: Larry A. Bowring Book design: Brian J. Svikhart Binding design: Joyce C. Eisen Typeset by Macmillan India Ltd., Bangalore Library of Congress Cataloging in Publication Data Anand, Sudhir, 1946- Inequality and poverty in Malaysia. (A World Bank research publication) Bibliography: p. 1. Income-distribution-Malaysia. 2. Poor- Malaysia. 1. Title. II. Series. HC445. 5. Z91513 339.2'2'09595 81-14178 ISBN 0-19-520153-I Contents PREFACE viii 1. PERSPECTIVES ON MALAYSIA Fconomic Overview 4 Political Background 6 The New Economic Policy 9 Inequality, Poverty, and NEP 14 Reader's Guide to the Study 17 2. THE 1970 POST-ENUMERATION SURVEY AND COMPARISONS WITH OTHER SURVEYS 21 The 1970 Post-Enumeration Survey 22 The PES Household Income Distribution 34 Intertemporal Comparisons of Inequality in Malaysia 42 Appendix: PES 1970 Instructions to Field Interviewers 53 3. INEQUALITY IN LEVELS OF LIVING 63 PES Income as a Measure of Economic Welfare 63 The Population Unit and Appropriate Income Concept 65 The Joint Distribution of Households by Household Income and Size 67 The Distribution of Households by per Capita Household Income 79 The Distribution of Individuals by per Capita Household Income 81 Trhe Atkinson Index 82 The Methodology of Inequality Decomposition 86 Interracial and Interregional Inequalities 93 Rural-Urban Inequalities 99 l'olicy Considerations 101 4. THE DEFINITION AND MEASUREMENT OF POVERTY 111 IPrevious Attempts at Defining Poverty 111 The Definition of a Poverty Line 113 The Sen Poverty Measure 118 Estimates of Poverty in Malaysia 125 A Profile of Poverty in Malaysia 126 Sensitivity of the Poverty Profile 132 Appendix: A P'rofile of the Rich 135 V 1; CONTENTS 5. SUBGROUPS IN POVERTY 144 Rural Poverty 145 Two-digit PES Subgroups 151 Appendix: Urban Poverty 167 6. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 187 The Distribution of Income Recipients by Personal Income 192 The Interpretation of Decomposition for Three Inequality Measures 198 Decomposition by Race and Location 202 Decomposition by Sex of Income Recipient 206 Breakdown by Employment Status 207 Interracial Earnings Differentials among Rubber Tappers 215 Decomposition by Occupational Category 216 Decomposition by Sector of Employment 226 Multivariate Decompositions 227 7. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 237 The Earnings Function 238 The Return to Education 241 Some Problems of PES Data 243 Estimates by Race, Occupation, and Sex 247 Estimates by Age Cohort 255 Language of Instruction 257 Type of Degrcc 259 Breakdown by Region 261 Appendix: Some Properties of the Earnings Function 264 8. CONCLUSIONS AND SOME NOTES ON POLICY 271 Summary and Conclusions 272 Policies to Reduce Poverty in Malaysia 280 Some Implications of Employment Restructuring 294 The New Economic Policy: Concluding Comments 298 APPENDIXES: THE MEASUREMENT OF INCOME INEQUALITY 302 A. A Brief Review 303 Indices Based Directly on the Lorenz Diagram 304 Other Indices 306 B. The Gini Coefficient 311 Definition I (Geometric) 311 Definition 2 (Rao, 1969) 312 Definition 3 (Kendall and Stuart. 1963) 313 Definition 4 (Sen, 1973a) 314 Definition 5 (Fei and Ranis, 1974) 315 The Effect of Changes in Certain Incomes 316 The Disaggregation of Income by Factor Components 318 On the Decomposition of the Gini Coefficient 319 CONTENTS vii C. The Decomposition of Three Inequality Measures 327 The Theil Entropy Index T 327 The Theil Second Measure L 329 The Variance of Log-Income V 330 Comparison of the Decompositions 331 D. Lorenz Dominance and Inequality 333 E. Lemmas on Lorenz Dominance 341 The Redress of Poverty Rule 344 F. Mapping the Household to the per Capita Household Income Distribution 346 REFERENCES 355 INDEX 365 Preface THERE ARE FEW STUDIES which analyze primary data on income distribution in developing countries. Work in this area has consisted largely of broad generalizations made from secondary material with little attempt to examine the underlying data sources and concepts. This study analyzes primary data on income distribution in Malaysia collected in the 1970 Post- Enumeration Survey (PES); these data have not yet been systematically analyzed or tabulated. This book may be viewed as an anatomy of income distribution in Malaysia. It documents the state and nature of income inequality and of poverty, and develops a methodology for this purpose. Apart from detailed measurement, a decomposition by socioeconomic variables suggests the sources of inequality and poverty. In the course of the empirical work, several statistical and technical problems had to be faced-from the evaluation of the quality of data to conceptual issues of measurement and decompo- sition. In places I have developed general solutions or techniques of analysis that were not available in the existing literature. The final product may thus also be viewed as an application of a framework for analyzing income distribution in a developing country. It is hoped that this case study for Malaysia might serve as a benchmark for comparative studies elsewhere, as better income data begin to be collected and analyzed systematically. The statistical and descriptive analysis of Malaysian income distribution is presented in the context of the concerns expressed by the Malaysian government in its New Economic Policy (NEP) of 1971. This policy, announced in the Second Malaysia Plan, 1971-75, consists of the twin objectives of "eradicating poverty irrespective of race" and "restructuring society to correct racial economic imbalance." In the book I have attempted to go beyond the mechanics of measurement and the exploi- tation of the PES data to illuminate these broader policy questions. The issues addressed, however, are confined to the objectives and strategy of the New Economic Policy and to the framework for achieving it by means of the Outline Perspective Plan, 1970-90. Because the bulk of this book was complete before publication of the Third Malaysia Plan, 1976-80, which viii PREFACE lr actually incorporates some of the information contained in the first draft of this study, policy elaborated in documents published since 1977 is not discussed. Anyone who has handled very large bodies of data on tape will know the probliems of cleaming, editing, splicing, and recoding the original infor- mation to create new and usable data tapes. I have checked and rechecked the PES data (which cover 135,000 individuals by 55 distinct variables) for errors in coding and internal consistency and am confident that the edited tapes are largely 1ree of error. In this process I was helped by Alexander Meeraus and Vinli Le-Si of the Development Research Center of the World Bank, and I would like to thank them. I would also like to acknowledge the generous assistance given in 1973 by Ramesh Chander. Dorothy Fernandez, K. G. R. Nathan, and V. T. Palan of the Department of Statistics, Malaysia, on various technical aspects of the Post-Enumeration Survey and in helping to track down and answer questions about the only surviving records of the 1957-58 Household Budget Survey. rt will be obvious that the work described in this book has entailed considerable and painstaking computational effort. For this I owe large: debts of gratitude to Shail Jain and to Sam Pal; they provided outstanding research assistance in the early and later stages of this project, respectively. Thanks are also due to Meera Shah who in 1973 helped me write the initial computer program for decomposition of the three inequality measures, Theil T, Theil L, and varlog (see appendix C). The first draft of this study took the form of four lengthy mimeographed papers (Anand, 1973, 1974a, 1974b, 1975). Parts of these papers were presented at seminars in the Department of Statistics, Kuala Lumpur; the Development Research Center of the World Bank; the Universities of London (London School of Economics), Oxford, Sussex, and Warwick; and at the Fourteenth General Conference of the International Association for Research in Income and Wealth in Aulanko, Finland. I am grateful to the participants of these seminars for their comments. The second draft, in the form of this book, was substantially complete by 1977, but its publication has been delayed for various reasons. N/'any people have commented on, or discussed with me, chapters from the earlier draft. I would particularly like to thank Montek Ahluwalia, Anthony Atkinson, Clive Bell, Ramesh Chander, John Duloy, Ravi Gulhati, Heather Joshi, S. M. Kanbur, John Knight, Ian Little, Heather Milne, James Mirrlees, Graham Pyatt, Bruce Ross-Larson, Amartya Sen, T. N. Srinivasari, R. Thillainathan, A. Vaidyanathan, and Don Zagier. Hellpful comments were also received in 1978 from the three referees of this book: Irma Adelman, Gian Sahota, and Donald Snodgrass. Jean Ponchamni typed the drafts of this book with amazing speed and x PREFACE accuracy. She also provided valuable logistical assistance and acted as the principal contact between myself in Oxford and the World Bank in Washington, D.C. I am extremely grateful to her for undertaking these tasks with such dedication and efficiency. There are some people without whose support or encouragement in different ways this book would have remained in mimeographed form for even longer. They are Montek Ahluwalia, John Duloy, Kaval Gulhati, Ravi Gulhati, Edward V. K. Jaycox, Jean Ponchamni, and especially Heather Milne. SUDHIR ANAND St. Catherine's College Oxford, England Inequality and Poverty in Malaysia Measurement and Decomposition 1 Perspectives on Malaysia MALAYSIA IS A MULTIRACIAL SOCIETY. The main racial groups are the Malays, who account for just more than half the population of Peninsular Malaysia, the Chinese (36 percent), and the Indians (11 percent).1 With different religions, languages, cultures, and social customs, communal lines cut across several facets of life. Except for an outbreak of racial tension in 1969, communal harmony has been maintained in the country. The origins of ethnic pluralism in Malaysia go back to early contacts with Indian arid Chinese traders and with Portuguese and Dutch colonialists.2 Much of the present economic and political structure can be traced back to the era of British colonial rule. The British settled first on the island of Penang, later in the coastal city of Malacca and on the island of Singapore. Penang, Malacca, and Singapore came to be known as the Straits Settlements. Some Chinese were already involved in entrepot trade in the Straits Settlements at that time, but the arrival of the British probably stirnulated further Chinese immigration.3 Both communities began to exploit local resources, especially tin, and branched out into various cornmercial and economic ventures, including cash crops and spices in nearby areas of the mainland. The British also encouraged Indians to come to work on their new sugarcane and coffee plantations. 1. The data on which this study is based, and much of the discussion as a result, are for Peninsular Malaysia and do not cover Malaysia's insular states: Sabah and Sarawak. The perspective presented in this chapter does not go beyond the early 1970s. 2. See Lamb (1964) and Hirschman (1972). 3. This subcomniunity, called Baba Chinese, is thought to be descended from early Chinese imrnigrants to Malacca around the fifteenth century, when it was the center of an important trading empire (Purcell, 1967). The Baba or Straits Chinese now form about 15 percent of the Chinese population in Peninsular Malaysia. 2 INEQUALITY AND POVERTY IN MALAYSIA It was not until the second half of the nineteenth century that large-scale migration to the Malay Peninsula began.4 This coincided with the period in which the British expanded and consolidated their rule over the peninsula some hundred years after landing in Penang. An event of crucial importance in inducing large streams of migrants was the introduction of the rubber tree from Brazil in the late nineteenth century. Rubber and tin mining came to dominate the economy, with rubber soon supplanting all other commercial crops. The new economic activities required more labor, but the indigenous Malay community was largely bypassed in the satisfaction of this demand. Large numbers of foreign laborers were brought in from India to work on rubber plantations and from China to work in tin mines. Immigration continued and even increased through the first few decades of the twentieth century, until the depression and restrictive legislation began to slow it down in the 1930s. With this tapering off, the ethnic pattern of the population started to stabilize. The proportion of Malays in the population leveled off at about 50 percent after 1931 (Arl&s, 1971, p. 528). The shares of Chinese and Indians in the population stabilized, too, and the locally born proportion of these communities significantly increased. Between 1931 and 1947 the proportion of Chinese born in Malaysia rose from 30 to 64 percent, while that of Indians rose from 21 to 52 percent. At the time of independence in 1957, more than three-quarters of the Chinese population and almost two-thirds of the Indian population were locally born (Arles, 1971). Estimates put the population of Peninsular Malaysia at approximately 250,000 in 1800, 2 million in 1900, and 6.3 million in 1957.5 By 1970, the year to which the survey data analyzed in this study refer, the population of Peninsular Malaysia had grown to 9.2 million.6 The nonparticipation of Malays in the new plantation and mining sectors, and in modern activities generally, led to sectoral and geographical concentrations of the races. The Chinese and Indians became more prominent in the modern economy of tin, rubber, and commerce on the West Coast; the Malays remained in the subsistence sectors of paddy (rice) farming and fishing, mostly along the East Coast and in the North. Various reasons have been suggested for the nonparticipation of Malays in the 4. See Jackson (1961), Sandhu (1969), and Purcell (1967). As a point of reference, one could take 1850 as the beginning of large-scale Chinese immigration, and the 1880s as the beginning of large-scale Indian immigration (Hirschman, 1972). 5. Unadjusted total of the 1957 census. 6. Total of the 1970 census adjusted on the basis of the "1970 Census Post-Enumeration Survey" (PES) (Kuala Lumpur: Department of Statistics, n.d.. unpublished computer tapes) The population of Sabah in 1970 was about 650,000; that of Sarawak, about 975,000. PERSPECTIVES ON MALAYSIA 3 emergent modern economy: British colonial policy, covert discrimination by employers, poor conditions in the early tin mines and rubber estates, and the Malays' alleged lack of economic motivation and preference for the tradlitional peasant life. Obversely, the prominence of Chinese and Indians in modern sectors and occupations may in part be explained by the theory of "immigrant culture": that immigrants possess a stronger motivation to achieve and that they operate on the fringes of the established (in this case, rural) society.' VWhatever the reasons for the lack of Malay participation in modern activities, Malays continued to predominate in the rural sector while non- Malays came to dominate the urban sector. In 1970 Malays constituted 63 percent of the rural sector and 53 percent of the total population, but only 27 percent of the urban population. Non-Malays constituted 73 percent of the urban sector, and this in turn accounted for 29 percent of the total population of Peninsular Malaysia. The 1970 employment structure shows that nearly 80 percent of all Malays were employed in the rural sector, compared with only a little more than half of all non-Malays. Chapter 6 contains a more detailed picture of employment by sector and occupation. VVithin the rural sector, Malays predominated in the smallholder segment of agriculture, although they have begun to move more into estate agriculture.' A majority of the Indian population has continued to be engaged as estate workers since being brought to Malaysia by the British. Almnost half the Malay smallholders-about a seventh of the labor force- worked in paddy farming, which accounted for only 3 percent of the gross dornestic product. Most paddy farms were, and still are, owned by Malays. In the rubber smallholder sector, Malays outnumbered Chinese by two to one, but the average size of Chinese holdings (8.3 acres) was almost twice the average size of Malay holdings. About 42 percent of Malay smallhold- ings were less than three acres, with another 46 percent ranging from three to ten acres. In the urban areas, which accounted for about a third of total employment in 1970, the non-Malay share exceeded 75 percent. Chinese dorninated such modern or quasi-modern sectors as mining, manu- facturing, construction, and commerce. In most of the major sectors, Malays tended to be poorly represented among professionals, managers, supervisors, and clerical staff. The relative concentration of non-Malays in urban areas also gave them access to better educational facilities in the 7. Discussions of these issues may be found in Hirschman (1972), Alatas (1977), and Ross- Larson (1977). 8. Information on the ownership of the smaliholding sector is scanty. See Government of Malaysia (1973), p. 11. 4 INEQUALITY AND POVERTY IN MALAYSIA cities-a factor which has helped in perpetuating their edge over the Malays in the economy. The pattern of ownership in 1970 shows an extensive participation of European capital in Malaysia's economy. While non-estate agriculture was in the hands of Malay smallholders, more than three-quarters of Malaysia's estate rubber production came from European-owned plantations. The rest originated largely in estates owned by Chinese of Malaysian or Singaporean nationality. European-owned tin mines ac- counted for almost three-quarters of Malaysia's tin output, and Chinese the remainder. In terms of equity capital, foreign interests accounted for more than three-fifths of the share capital of limited companies in Malaysia in 1970. Economic Overview The development of the Malaysian economy has depended historically on the exploitation of land and minerals, with which the country is richly endowed.9 The main economic activities since the beginning of this century have been natural rubber production and tin mining, alongside traditional agriculture and fishing. This pattern persists. In 1970 tin and rubber accounted for 20 percent of net domestic output and 53 percent of merchandise exports; timber and palm oil accounted for another 21 percent of exports. There are abundant natural and land resources still to be exploited, ' and the extraction of petroleum is already beginning to play an important part in Malaysian development following recent discoveries of oil and natural gas. With such a high concentration of primary commodities in domestic production, Malaysia is heavily dependent on foreign trade: exports accounted for almost half its gross national product (GNP) in 1970. The consequent vulnerability of national income to the prices of a few exports is an important feature of the economy, although it is being reduced by progressive diversification. Palm oil production, negligible before World War II, has been developed on a large scale. In fact, Malaysia is now the world's largest exporter of palm oil, and it has consistently been the largest 9. The statistical data in this section have been obtained from recent World Bank economic reports on Malaysia. Wherever possible, economic data for 1970 are presented to allow comparison with the 1970 PEsdata, the analysis of which forms the subject matter of this book. 10. In addition to the 8 million acres now under cultivation, 4 million acres in Malaysia are still undeveloped and suitable for cultivation. PERSPECTIVES ON MALAYSIA 5 exporter of natural rubber and tin. Timber extraction has also expanded to take advantage of strong world demand. Apart from diversification, the government has embarked on major programs to raise productivity of the country's traditional crops: rice and rubber. Rice production has greatly expanded as a result of double- cropping made possible by government irrigation and drainage schemes, but the country still is a net importer. In rubber there has been a far- reaching and massive program to increase productivity by replanting rubber lands with high-yielding varieties of trees. Even the decline in rubber prices by more than a third during the 1960s was absorbed with relative ease because of the dramatic improvements in productivity. Export earnings from rubber managed to grow over that decade, but the share of rubber in total ex ports declined from about half in 1961 to a third in 1970.11 Nlanufacturing had a relatively late start in Malaysia. Given the openness of the economy and its large import capacity, most manufactured consumer goods were freely imported. But with the need to diversify and reduce the country's dependence on the export of a few primary products, the manufacturing sector has been given increasing attention. Helped in the 1960s by uncertainty about growth in world demand for such traditional products as natural rubber and palm oil, manufacturing came to be regarded as the leading growth sector of the economy. The main stimulus for industrialization has been the protection of domestic manufacturers by selective import tariffs and quotas, as well as various investment incentives. Import substitution, chiefly of consumer goods, caused value added in the sector to expand at almost twice the rate of GNP through the 1960s. Even so, it accounted for only 13 percent of Peninsular Malaysia's GNP in 1970. More recently, manufactured goods have begun to be exported and are fast becoming a substantial export item. The country's physical infrastructure has been adequate. Substantial investments in transport and communication facilities and in public utilities were made in the 1950s, and physical infrastructure has never represented a serious bottleneck for private development. The government's financial policies have also created and sustained favorable conditions for private initiative. There are few controls on foreign exchange, and the balance of payments position has been strong, with external reserves equal to about six months of imports in recent years. Conservative fiscal and monetary management, including a scrupulous avoidance of recourse to inflationary means for financing budget deficits, II. In 1976 the share of rubber in total exports was down to about a quarter. 6 INEQUALITY AND POVERTY IN MALAYSIA was in part responsible for the remarkable price stability in the 1960s. Between 1957 and 1970 the retail price index moved up by only 0.7 percent a year on the average. Financial resources do not appear to have constituted a serious constraint on economic development. The external resource position has remained comfortable ever since independence, reflecting in part the inherently strong export position, an adequate and fairly elastic tax system, and careful monetary and economic management. With respect to domestic resources, the long-run marginal savings rate-that is, the increment of GNP saved-has been around 25 percent. For the 1960s as a whole, total national saving exceeded domestic investment, with private saving greater than private investment and public deficits modest. Public sector revenue grew rapidly during the 1960s, mainly as a result of vigorous taxation efforts; the ratio of government revenue to GNP was 23 percent in 1970. The First Malaysia Plan, 1966-70, was drawn up on the premise that the principal constraint on the public sector was a financial bottleneck. But a shortfall of 7 percent in the expenditure target of the plan highlighted the administrative constraint on development: an inadequate capacity for planning and implementation. There have been shortages of technical and specialized manpower in nearly all government departments and agencies, a problem heightened by the preferential hiring of Malays over non-Malays, although the former have produced relatively few university graduates in science and engineering. Staff shortages have in turn led to problems and delays in the planning and implementation of government programs, especially in the newly established statutory bodies. Project preparation in the public sector is still weak though improving, and government staff support is sometimes inadequate. During the 1960s Malaysia's GNP increased at about 6 percent a year in constant market prices. When adjusted for the deterioration in the terms of trade, the growth in real income in Malaysia was about 4.5 percent a year, compared with a population growth of 2.6 percent a year. The GNP per capita in 1970 was about US$380, putting Malaysia behind only Hong Kong and Singapore among countries in Southeast Asia. Political Background Notwithstanding the relatively high level of economic performance compared with other Asian countries, there are tensions within the Malaysian polity which can be traced to the multiracial nature of the PERSPECTIVES ON MALAYSIA 7 society. In the last decade there has also been increasing concern with the problems of the poor. 12 The interests of the three major ethnic communities have always presented a challenge to political unity in Malaysia. Well before the transfer of power by the British in 1957, diflerences emerged between the communities over citizenship and political representation for non-Malays. The United Malays National Organization (UMNO), formed in 1946, successfully opposed a British constitutional plan for the Malayan Union, which was to institute common citizenship for Malays and non-Malays. The Malayan Chinese Association (MCA), formed in early 1949, sought to improve the political status of the Chinese under the Federation of Malaya Agreement of February 1948. The basic problem was "the division in political and economic power, with the Malays having most of the former and the Chinese most of the latter" (Morrison, 1949, p. 253). A modus vivendi was worked out over time, with concessions being macle by both sides. The non-Malays were granted citizenship rights and the freedom to pursue their economic objectives without interference. In return, the Malays became entitled to certain privileges. Simply put, a "social contract" was agreed upon, under which the Malays would have the paramount place in the political life of the country while the Chinese would continue to enjoy their economic position and religious freedom. This basic compact was generally understood and accepted at the time of independence and is reflected in the country's constitution. The constitution takes into account the multiracial nature of the society and the differences in economic status of the Malay and non-Malay communities. A balance is sought on the basis of a division of re- sponsibilities and functions: the economic prominence of the non-Malays is balanced by the political supremacy of the Malays. While the basically laissez-faire environment has allowed the continuation of non-Malay predominance in commercial and industrial spheres, the constitution has protected the political supremacy of the Malays by giving them preferential access to the civil service and by granting certain rights to Malay rulers. In addition to preserving the sultanates, the constitution has made Islam the national religion. Eighty percent of the senior positions in the administra- tive and diplomatic services are reserved for Malays (this is intended to ofl'set Chinese predominance in the commercial and industrial sectors). Quotas also apply in the granting of scholarships and in other public 12. A political cxpression of these problems occurred during the Baling disturbances in November 1974 and subsequent student demonstrations in Kuala Lumpur (see Peiris, 1975, pp. 29-31). 8 INEQUALITY AND POVERTY IN MALAYSIA training privileges. Permits and licenses needed to operate certain busi- nesses and trades have also been apportioned on an ethnic basis-for example, those for timber extraction and for bus, taxi, and trucking operations. Furthermore, certain land ("Malay reservation") has been earmarked for development by Malays only. I3 The three main political parties representing the three racial groups got together under one banner shortly before independence. They formed the Alliance, composed of UMNO, MCA, and the Malayan Indian Congress (MIc). 4 The Alliance won comfortable majorities in the national elections of 1955 (under the colonial administration), 1959, and 1964. During this time, it advocated a formula for racial harmony based on the earlier social compact, and it continued to rule after Sabah, Sarawak, and Singapore joined Malaya in 1963 to form Malaysia. But political polarization was increased and this formula was weakened by events such as the separation of Singapore and its large Chinese population from Malaysia in 1965 and the adoption of Bahasa Malaysia as the national language in 1967. In the 1969 elections the Alliance lost control of the state assembly of Penang and was stalemated in two others-Perak and Selangor-but it still obtained an overall majority. Chinese-based parties, such as the Dem- ocratic Action party and the Gerakan Rakyat Malaysia, scored sig- nificant gains. These parties, which draw support mainly from the Chinese working class, advocated racial integration and socialist reforms. The results were viewed by some as a victory for the Chinese who were seeking a greater political role for their community. The Pan Malayan Islamic party, which represents Malay nationalism with strong religious overtones, also increased its support at the expense of UMNO and the Alliance. This polarization indicated an erosion of the Alliance's strength and its formula for racial harmony. The situation led to race riots in Kuala Lumpur a few days after the elections in May 1969. The federal parliament and state assemblies were suspended and a state of emergency imposed, with the country run by a National Operations Council. The polarization of political sentiments and the growing representation by the Chinese in politics seriously undermined the established formula for racial harmony. With the inroads made by the Chinese in the political arena, the traditional Malay dominance in politics and Chinese dominance in the economic sphere no longer applied, and the division of responsi- bilities and functions became lopsided. Malay nationalist sentiment mounted, and leaders of the ruling party searched for an alternative 13. These Malay privileges are provided in Articles 3, 38. 89, 152, and 153 of the constitution. See Suffian (1972) and Government of Malaysia (1972). 14. The Malayan Indian Congress was started in 1946. PERSPECTIVES ON MALAYSIA 9 formula. The government responded by proposing steps to build up the economic position of the Malays. Malay rights were reaffirmed, and the Newv Economic Policy was proclaimed in the Second Malaysia Plan, 1971 - 75. This policy sought to "correct economic imbalances between the races" by expanding Malay participation in the modern commercial and indus- trial sectors. Presented largely in terms of national unity, the policy had another objective: to "eradicate poverty among all races." Normal political activity was restored and Parliament was reinstituted in February 1971. The partners of the ruling party preached moderation and unity, and each tried to increase ethnic support by appealing to the middle ground. The Malay leader of UMNO, Tun Abdul Razak, defended the New Economic Policy and Second Malaysia Plan as "the last chance for everybody's survival and the nation's survival." In the opening address of the 1971 MCA general assembly, he stated: The New E.conomic Policy must succeed. The stakes are too high to aLllow it to fail. It should not fail, because the Plan is not aimed at promoting any sectional interests. It is a blue-print for the progress and unity of our nation ... Let there be no misunderstanding as to the objectives of the Second Malaysia Plan. It will not deprive any one group of its legitimate rights. It is not aimed at promoting the interests of any particular community at the expense of others (Razak, 1971, pp. 5-7). The presidential address at the same assembly was delivered by Tun Tan Siew Sin, who explicitly recognized the other important facet of national unity: "In this day and age, it is not possible to maintain a small island of prosperity in a sea of poverty. The few who are rich cannot maintain their wealth in the midst of widespread and growing poverty. This is against the law of Nature" (Tan, 1971, p. 18). The proclamation of the New Economic Policy helped defuse the political situation after the racial disturbances in 1969. The political structure was stable after the resumption of Parliament in 1971, and the leadership of the Alliance broadened its support by forming coalitions with opposition parties from the 1969 elections. These came together in 1972 and 1973 as the Barisan Nasional (National Front), which won an overwhelming victory in the general elections in August 1974. The New Economic Policy The New Economic Policy (NEP) has as its overriding goal the promotion of national unity through the two-pronged strategy of: (1) eradicating poverty by raising income levels and increasing employment opportunities 10 INEQUALITY AND POVERT-Y IN MALAYSIA for all Malaysians, irrespective of race; and (2) accelerating the process of restructuring Malaysian society to correct economic imbalance, so as to reduce and eventually eliminate the identification of race with economic function. Announced in the Second Malaysia Plan (sMP), 1971-75, the New Economic Policy was presented largely in terms of national unity, which is seen to depend both on the reduction of racial economic imbalances and on the eradication of poverty irrespective of race. Poverty eradication and the correction of racial economic imbalances have thus been enshrined as the twin objectives of the New Economic Policy. Specifically, the objective of racial balance aims to "correct the imbalances in income distribution, employment, and ownership and control of wealth" between the Malays and non-Malays (SMP, p. 41). A fundamental premise of the New Economic Policy is that it will be undertaken in the context of rapid expansion of the economy so as to "ensure that no particular group will experience any loss or feel any sense of deprivation" (sMP, p. 1).15 The broad framework for achieving the objectives of the New Economic Policy was first provided in the Outline Perspective Plan (opp), 1970-90, contained in the Mid-Term Review of the Second Malaysia Plan (MTR). The opp provides a series of long-term targets designed to help restructure the racial composition of employment and the racial ownership of wealth. Such restructuring, it is believed, will help attain a better income balance between the races. Unlike the objective of racial balance, that of poverty eradication was not elaborated in the SMP documents, nor were any specific targets set. There was no attempt at quantifying or defining poverty, and only general policies were mentioned for its alleviation: for example, "The moderniza- tion of rural areas will help Malays and the overall eradication of poverty." The Third Malaysia Plan (TMP), 1976-80, published in late 1976, gave more explicit emphasis to poverty eradication. It goes some way toward quantifying the extent of poverty and specifying targets for its reduction during the course of the Outline Perspective Plan. It contains specific policies, programs, and projects to help the poor. Prong I, the eradication of poverty, needs little comment at this stage; it is discussed at length in chapters 4 and 5, where an attempt is made to define 15. Tun Tan Siew Sin believed that the key to the success of the New Economic Policy was "a much faster rate of economic growth, because a re-distribution of wealth to redress economic imbalances, both racial and between the haves and havenots of the same race, would be much easier if the economic cake were much larger. In this way . . . we do not have to take away from those who have in order to give to those who have not. Only the additional cake will be redistributed and this would benefit all" (Tan, 1971, pp. 20-21). The "redistribution with growth" philosophy of Chenery and others (1974) is similar in spirit to this approach. PERSPECTIVES ON MALAYSIA /I and measure poverty in Malaysia and to suggest policies for its alleviation. Although the sMP did not attempt to define or quantify the extent of poverty, the government visualized the following broad approach for its eradication: --Expansion of employment opportunities at a rate sufficient to bring about full employment of the labor force by 1990 --Increases in productivity of traditional smallholder agriculture and traditional urban activity --Encouragement of intersectoral movements of labor from low- to higher-productivity endeavors (Robless, 1975b, p. 7).16 Prong 2, the second objective, seeks to restructure society and narrow the diflerences between Malay and non-Malay incomes by reducing disparities in the ownership and control of wealth in the modern sectors and by diminishing the concentration of Malay employment in traditional agricul- ture while increasing it in the relatively high-income urban sector. The approach adopted by government to correct racial economic imbal- ances is: --Expansion of the share of Malays in the ownership of wealth --Restructuring of the racial pattern of employment between sectors as well as occupations, without decreasing employment opportunities for any race (Robless, 1975b, p. 7). The opp specifies targets for corporate ownership and employment by racial group in 1990. Targets are laid down for the ownership of equity capital only-that is, corporate assets-and noncorporate assets are ignored. One reason presumably is that the government wishes to expand Malay participation mainly in the modern nonagricultural sectors in which Malay ownership is particularly underrepresented. Another reason probably is the lack of data on noncorporate assets. '7 The most significant feature of the structure of equity capital in 1970 was a clear dominance by foreign interests. Of the total share capital of limited companies in Peninsular Malaysia, foreign interests accounted for 60.7 percent. Chinese held 22.5 16. Dr. C. L. Robless was then deputy director-general of the Economic Planning Unit, Prime Minister's Department, Malaysia. 17. There are some data, however, on the ownership of fixed assets in modern agriculture and industry in 1970 (MTR, table 1-4). These show the noncorporate sector to account for a relatively small proportion of fixed assets: in modern agriculture, 29.6 percent of estate acreage (planted); in industry, only 11.6 percent of the value of fixed assets. Malay ownership of corporate industrial assets was only 0.9 percent, whereas it was 47.; percent of noncorporate estate acreage. Estates are defined as landholdings of more than 100 acres. 12 INEQUALITY AND POVERTY IN MALAYSIA percent of the total, Malays 1.9 percent, and Indians 1.0 percent. The remainder-about 14 percent-was held by federal and state governments, statutory bodies, and other Malaysian residents (individuals, nominees, and locally controlled companies). 18 Apart from these figures in the Mid- Term Review of the Second Malaysia Plan, there are no other data on the ownership of wealth in Malaysia. The government's goal in relation to restructuring wealth is to "promote the creation of a commercial and industrial community among Malays and other indigenous people in order that, within one generation, they will own and manage at least 30 percent of the total commercial and industrial activities of the country in all categories and scales of operation."'9 The change in the Malay share of equity capital from 1.9 percent to 30 percent in twenty years obviously implies a considerable degree of restructuring. But government has emphasized that the restructuring is to be undertaken in the context of a rapidly expanding economy. There should, therefore, "be no grounds for fear or anxiety on the part of other Malaysians that government intervention in the private sector on behalf of the Malay community will lead to deprivation of the rights or prospects of non- Malays" (MTR, p. 85). The aim of increasing the share of Malay corporate ownership to 30 percent by 1990 does not entail a smaller share for the non- Malays, since the share of foreign investors is to drop from the present high level of 60.7 percent to 30 percent during the twenty-year period. The opp spells out the implications for increasing the Malay share of equity capital during 1970-90 when the total capital of limited companies is projected to grow at an average rate of 11.5 percent a year (MTR, table 4-8).20 The Malay:non-Malay:foreign proportions of equity capital in 1990 are targeted to be 30:40:30, which entail annual absolute growth rates for the three totals of 27.9 percent, 11.9 percent, and 7.6 percent, respectively. Thus, despite the decline in the foreign share in relative terms, a growth rate of 7.6 percent a year is still implied in absolute terms (Robless, 1975a, pp. 53-55). 18. MTR, table 4-7. This table presents a breakdown by race and sector of the ownership of share capital of limited companies in Peninsular Malaysia in 1970. The share of the Malays ranged from 0.7 percent in mining to 3.3 percent in banking and insurance, with 2.5 percent and 2.2 percent in manufacturing and construction, respectively. In transport and communications, however, the Malay share was 13.3 percent. The foreign share was 75.3 percent in agriculture and 72.4 percent in mining and quarrying. In manufacturing, commerce, and banking and insurance, it varied between 50 and 60 percent. 19. MTR, p. 62. The indigenous people of Malaysia include the Kadazans of Sabah and the Ibans of Sarawak in addition to the Malays of Peninsular Malaysia. 20. In the same period, gross domestic product (GDP) is projected to grow at an average rate of 7.0 percent a year (MTR, table 4-6). PERSPECTIVES ON MALAYSIA 13 It is generally recognized that the 30 percent target for Malay ownership is ambitious. One fundamental difficulty lies in the relative shortage of Malay savings. This shortage is probably due to lower average income levels and a diffcrent spending propensity among Malays compared with non-Malays.2' To supplement the role of individual Malay savers and entrepreneurs in restructuring the racial ownership of wealth, the govern- ment has established state-owned and -controlled enterprises and financial institutions that acquire share capital in existing and new companies to be held in trust for the Malays and other indigenous people until such time as they can acquire these shares from their own savings (MTR, pp. 14, 85). The publicly funded corporations are also expected to provide technical, financial, and management assistance to forthcoming Malay entre- preneurs, with the aim of restructuring wealth to "create a Malay entrepreneurial community equipped to play a full and wholesome role in the economic life of the nation" (Robless, 1975a, p. 53). Examples of such public corporations are the Council of Trust for the Indigenous People (MiARA), the National Corporation (PERNAS), the Urban Development Authority (UDA), and state economic development corporations (SEDC).22 Thcsc institutions arc taking up equity shares in joint ventures with the private sector, shares which will "eventually be transferred to individual ownership of Malays and other indigenous people" (sMP, p. 160). The problems associated with this transfer and the mechanisms for achieving it have still to be faced.23 The government's.long-term targets for the composition of employment have been set to "ensure that employment in the various sectors of the economy and employment by occupational levels will reflect the racial composition of the country" (MTR, p. 62). The targets for 1990 have been fixed for levels of racial employment by sector (MTR, table 4-5) and by occupational category (TMP, table 4-15). In 1970 the racial distribution of overall employment for Malays: Chinese:Indians was 52:37:11, which is close to the ratios of the three groups in the population of Peninsular Malaysia. But large deviations from this average existed in certain sectors: Malays were overrepresented in 21. There are as yet no systematic estimates of savings rates by race. 22. UDA is in charge of commercial and property development for Malays. PERNAS was created to promote Malay participation in insurance, construction, trading, proper- ties, engineering, and securities in addition to establishing joint ventures with the private sector. :23. Forexampl(, Puthucheary (1977, p. 10) doubts whether the Malayscan save enough by 1990 to acquire the shares being held in trust for them by the public corporations. He thinks that "a great part of the 30 % target for Malay ownership when achieved is likely to be owned by Malay interests which to all intents and purposes amounts to Government ownership " 14 INEQUALITY AND POVERTY IN MALAYSIA agriculture, where output per worker is lowest, and underrepresented in mining, manufacturing, construction, and commerce, where output per worker is two to three times that in agriculture. In agriculture, for instance, the racial composition of employment in 1970 was 68:21:11 for Malays:Chinese:lndians; in manufacturing it was 29:65:6.24 The opp seeks to change the racial pattern of employment in each sector (and each occupation) so that by 1990 the three groups are more nearly represented according to their population ratios. The targets for restructuring racial employment by sector have been chosen in such a way that "there is full employment for all races . . . in 1990" (MTR, p. 78). The Mid-Term Review of the Second Malaysia Plan underlines the need for rapid employment growth, particularly in the modern sectors; otherwise "the redistribution required in sectoral employ- ment shares of the various races would lead to the displacement of workers of one racial group or another from their present employment" (MTR, p. 78). In fact, the overall rate of employment growth during 1970-90 has been targeted at an average of 3.3 percent a year, with annual rates of 7.6 percent in manufacturing, 5.0 percent in construction, 4.7 percent in commerce, but only 0.5 percent in mining and 1.2 percent in agriculture.5 The annual rate of growth of the labor force in this period is estimated at 2.9 percent; that of population at 2.5 percent (MTR, p. 66). Although restructuring is intended to eliminate the identification of race with economic function, the government recognizes its effects in bringing about income balance. According to MTR (p. 9), "the differences in income between the races have their origin in the concentration of the various races in different sectors of the economy and differences in their occupational position in these sectors." Hence the envisaged restructuring of racial employment, "besides bringing about employment balance, will also remove racial income differentials arising from differentials in sectoral product per worker" (MTR, p. 80). Inequality, Poverty, and NEP Malaysia's efforts to reduce poverty and racial imbalances will obviously have an effect on income distribution among individuals. Personal income distribution, an important subject for study in its own right, has recently concerned many governments in developing countries and is increasingly 24. MTR, table 44. The racial composition of employment for Malays:Chinese:lndians was 25:66:9 in mining; 22:72:6 in construction; and 24:65:11 in commerce. 25. Computed from MTR, tables 4-4 and 4-5. PERSPECTIVES ON MALAYSIA 15 the subject of academic research.26 A study of the individual income distribution allows explicit analysis of poverty27 and racial income imbalances, and the testing of various hypotheses concerning the effect of racial income inequality on individual income inequality. The reduction of individual income inequality is not the chief concern of the New Economic Policy, which focuses instead on poverty and interracial inecluality. The commitment to poverty eradication, however, can be viewed as a desire to improve.the personal income distribution in Malaysia, at least up to the point that poverty has been eradicated. Moreover, the government itself indicates in opp its aim "to reduce the existing inequitable distribution of income between income classes and races" (MTR, p. 62; italics added).28 VVhat are the implications of the two prongs of NEP for individual income distribution? Figure 1-1 shows the pattern of individual income distri- bution for the Malays and non-Malays separately. Although intended to be illustrative, the diagram does embody most features of the actual distributions (see chapters 3 and 6). It shows a higher mean income for the nori-Malays than for the Malays and a considerable overlap between the twc, distributions. A similar pattern of inequality is also indicated in the two distributions around their respective means. The poverty line in the diagram corresponds to the definition of poverty in chapter 4. P'rong 1, the eradication of poverty among all Malaysians irrespective of race, obviously requires the specification of a poverty line, but this has not yet been defined explicitly in the Malaysian plans (including TMP). Since the government has not specified the income levels to which the poor are to be moved in the course of poverty eradication, I define a poverty line in cha,pter 4 and interpret Prong I as simply lifting all the poor up to this level. In figure 1-1 this means moving all poor individuals, Malay and non- Malay, from the left side of the poverty line all the way up to it (but not beyond it). P'rong 2, the correction of racial economic imbalances, is a separate and logically independent goal of NEP. To restructure the racial composition of employment and the racial ownership of wealth it seeks proportional representation of racial groups according to their population ratios in each sector and each occupation. These targets are specified in and of themselves 26. See, for example, Chenery and others (1974), and the references cited therein. 27. Indeed, "relative poverty" can be defined only with reference to the overall level of, and inequality in, individual incomes. 28. Robless (1975a) states that an operational aim of NEP is "to raise income levels of all the-se in the lowest 40% of the population and reduce the present inequality in the size distribution of income." 16 INEQUALITY AND POVERTY IN MALAYSIA Figure 1-1. Individual Income Distribution for Malays and Non-Malays Malays 30 - :2 -Malays 20 10 Poverty line 50 100 150 200 Per capita household income (M$ per month) to achieve the wider objective of restructuring, but they have clear implications for racial income distribution. With proportional racial representation at every layer of the economy, a proportional racial representation is also implied at each income level-in other words, thie racial groups are proportionately represented along the entire range of incomes. Thus Prong 2, racial balance, can be construed as implying a point-to-point equalization of the Malay and non-Malay income distributions. If non-Malay incomes are not to be reduced, this cor- responds to a rescaling of Malay incomes so that the Malay distribution coincides with the non-Malay distribution (see figure 1-1). As Dr. Mahathir Mohamad, the prime minister, once interpreted Prong 2: In trying to redress the imbalance it will be necessary to concentrate your effort on the Malays, to bring out more Malay entrepreneurs and to bring out, and to make Malay millionaires, if you like, so that the number of Malays who are rich equals the number of Chinese who are PERSPECTIVES ON MALAYSIA 17 ric'h, the number of Malays who are poor equals the number of Chinese who are poor, and the number of unemployed Malays equals the number of unemployed Chinese, then you can say that parity has been achieved (see Low, 1971, p. 74). Reader's Guide to the Study In this chapter I have attempted to provide a general introduction and perspective on Malaysia as a backdrop to this study of income distribution. In particular, I have tried to trace the importance of ethnic pluralism in the country, and how this has led to special distributional concerns in the Malaysian context. The correction of racial economic imbalances has been isolated as a major objective of government policy; hence there is a clear need to control for race in the analysis. A further organizing theme of this study is the eradication of poverty irrespective of race, the other principal goal of the New Economic Policy. Chapter 2 deals with the data source for this study: the Post- Enumeration Survey (PES) of the 1970 census. The chapter begins by describing certain technical aspects of the survey, such as sample design and the definition of income, which have not been documented elsewhere. There is then a discussion of the broad features of the PES household income distribution and a comparison of estimates of average household income based on PES and the national accounts. Next, it considers the possibility of inte:rnational comparisons of income inequality among countries at app:roximately Malaysia's level of development. Finally, an attempt is made at intertemporal comparisons of inequality in Malaysia itself, since two earlier surveys by the Malaysian Department of Statistics have reported income data. These surveys are analyzed in some detail to determine their comparability with PES. C'hapter 3 deals with inequality in levels of living in the country. The problems associated with using PES income as a measure of economic welfare are considered, as is the choice of population unit and income concept for measuring inequality. It is argued that the distribution of individuals by per capita household income is more suitable for this purpose than the distribution of households by household income. The joint distribution of households by household income and size is presented so that the relations among the various derived distributions can be determined. Several indices of inequality are computed for these distributions, and the corresponding within-race coefficients and racial disparity ratios are noted. There is a methodological discussion on the decomposition of inequality into within-group and between-group 18 INEQUALITY AND POVERTY IN MALAYSIA components, and this is applied to estimate the contributions to overall inequality of interracial, interregional, and rural-urban inequalities. The final section considers the policy implications of the results of racial decomposition in terms of the two prongs of NEP and shows that the redress of poverty is also an "efficient" way to redress inequality. Chapter 4 is concerned with the definition and measurement of poverty in Peninsular Malaysia. It seeks to explore the extent and nature of poverty in the country, so that policy measures for its alleviation can be considered. A poverty line is estimated after alternative approaches to its definition (absolute and relative) are examined. Various indices of poverty are ihen discussed, ranging from the simple incidence-of-poverty measure to others that take account of the poverty gap. A new index proposed by Sen (1976a) is derived, and alternative normalizations are suggested for it. Estimates of all these indices are presented for Peninsular Malaysia, and the simple incidence-of-poverty measure is used to construct a profile of the poor in Peninsular Malaysia. This profile identifies the poor in relation to such socioeconomic variables as race, location, employment status, occupation, and education. The chapter ends with a sensitivity analysis of the profile of poverty with respect to alternative definitions of the poverty line. An appendix contains a profile of "rich" households in Malaysia. Chapter 5 contains a detailed picture of rural poverty, since 87.7 percent of the poor in Peninsular Malaysia reside in rural areas. The chapter begins by investigating the broad characteristics of rural poverty at the one-digit level. For an efficient design of policies that reduce leakages to the nonpoor, smaller and more homogeneous subgroups that exhibit high rates of poverty are examined. This is done by disaggregating the rural poor according to operationally and analytically relevant categories by using the employment sector and occupational variables cross-classified at the two- digit level. From this matrix are selected five major subgroups which account for about 80 percent of the rural poor: paddy smallholders, laborers on paddy and mixed-agriculture farms, rubber smallholders, workers on rubber estates and smallholdings, and fishermen. The econornic problems of these subgroups, and measures to raise their productivity and incomes, are discussed with a view to identifying some major components of rural development policies and projects. The chapter has an appendix on urban poverty. Chapter 6 analyzes the personal income distribution; that is, the distribution of income recipients by personal income. This is the distri- bution most directly amenable to policy intervention; it also accounts for much of the inequality in levels of living. Various inequality indices are computed for the personal income distribution, and comparisons made with distributions considered in chapter 3. Racial-disparity ratios for PERSPECTIVES ON MALAYSIA 19 personal income are compared with those for household and per capita household income; the differences are explained by racial variations in average household size and participation rate. There is a decomposition of personal income inequality by race, urban or rural location, and region, and fiurther decompositions by employment status, occupational category, and industrial sector. These help in identifying sources of income inequality in the country and in examining the "association of race with economic function and geographical location" (NEP Prong 2). Combinations of the variables are used to perform multivariate decompositions, which help in evaluating the appropriateness of these categories for studying inequality in developing countries. Chapter 7 explores the empirical relations among age, education, and incorne of urban employees. This is done through a detailed regression analysis of earnings functions based on the human capital model. The subsample of urban employees is chosen because PES income is likely to be a good measure of earnings only for this group (data on earnings or labor income are not given separately in PES). Some properties of the earnings function are derived (in an appendix) which are useful in interpreting the regression results. The estimated equations thus provide rough orders of magnitude for private rates of return to education, for the percentage of inequality explained by the life-cycle factors of age and education, and for the peak and peakedness of the age-income profile. A disaggregation by racial group and occupation permits the testing of some interesting hypotheses about racial and occupational differences in average levels of and returns to education. Further disaggregations are by sex of employee, language of instruction, type of degree, age cohort, and region. The earnings functions illuminate various aspects of the relation between education and income and conveniently summarize much information about the urban labor market in Malaysia. Chapter 8, the concluding chapter of the study, starts with a review of the principal findings on inequality and poverty in Malaysia. The rest of the chapter is divided roughly into policy analyses of Prongs I and 2 of the New Economic Policy. First there is a general discussion of policies to reduce poverty in Malaysia. Four broad types of policy are isolated which seem particularly relevant in the Malaysian context: direct income transfers, fiscal policies, intervention in commodity markets, and rural development policies. These are discussed and evaluated in the light of information about the poor from the poverty profiles and subgroups (chapters 4 and 5). There is then a short review of the implications of the employ- mernt restructuring target of Prong 2. The chapter ends with a brief discus- sion of the interactions and complementarities between the two prongs of NEP. 20 INEQUALITY AND POVERTY IN MALAYSIA The six technical appendixes dealing with the measurement of income inequality are an integral part of the study. They are partly a unified review of recent literature relevant to this analysis of Malaysian data, and partly a presentation of results that have arisen out of measurement problems encountered in the course of this empirical work. Appendix A contains a brief review of inequality indices based on the Lorenz diagram, many of which have been estimated using data from the Post-Enumeration Survey (PES). Appendix B considers various definitions of the Gini coefficient and demonstrates their equivalence. It also derives some interesting properties of the Gini coefflcient, such as its nondecomposability. Appendix C derives the decomposition of three inequality measures used in this study: Theil's entropy index T, Theil's second measure L, and the variance of log-income. Definitions are given of the between- and within-group contributions for these measures, and their decomposition formulas are deduced. Appendix D presents Atkinson's theorem and proves the equivalence of two types of ranking of income distributions (Lorenz dominance and the principle of transfers) which are important for positive measurement of inequality. Appendix E contains new results on Lorenz dominance, which enable unambiguous comparisons of inequality between related distributions defined over different population units and income concepts. This ap- pendix also contains a statement and derivation of the redress of poverty rule. Appendix F discusses the mapping of the household to the per capita household income distribution. It derives the mathematical transformations required to effect the mapping and states conditions under which one distribution will be more or less equal than the other. 2 The 1970 Post-Enumeration Survey and Comparisons with Other Surveys THE 1970 POST-ENUMERATION SURVEY (PES) is the data base for this study of Malaysian income distribution. In this chapter, I examine and evaluate the PES in detail before proceeding to a full analysis of the data in later chapters. To enable an assessment of the quality of PES data, a documentation is provided of the survey design, sampling procedure, income definition, and other technical aspects. Such documentation on the survey is not available elsewhere, and an official report on PES has not been prepared. Thiis chapter also presents in summary form some estimates of the PES household income distribution. It is concerned with broad facts about inequality in Malaysia, such as the degree of overall inequality and rural- urban and racial income inequality, leaving detailed analyses to later chapters. The possibility of comparing inequality in Malaysia with inequality in other countries is discussed, and an attempt is made at intertemporal comparisons in Malaysia itself using income data reported by two previous surveys: the 1967-68 Malaysian Socio-Economic Sample Survey of Households and the 1957-58 Household Budget Survey. Recently there has been much interest and speculation about inter- national and in'te-temporal inequality in developing countries,' but little hard evidence is available on the subject. Even where the data exist, few studies have attempted to analyze the data sources and tackle the serious problems of comparability among them. I deal with this problem head-on 1. See Kuznets (1955 and 1963), Adelman and Morris (1973), Paukert (1973). and Ahluwalia (1974a and 1976). Following Kuznets, the latter authors have attempted to establish an inverse U-shaped relation between inequality and development on the basis of cross-country experience 21 22 INEQUALITY AND POVERTY IN MALAYSIA in the case of Malaysian survey data, bringing to bear all relevant evidence about the earlier surveys. Since the published report on the 1957-58 Household Budget Survey is deficient in information about sample coverage, definition of income, and the like, I have reconstructed an account of this survey from unpublished records and files of the Malaysian Department of Statistics. This account is documented in detail and used to demonstrate that the 1957 survey is not comparable with the 1970 survey, and that no conclusions can be drawn from these surveys about inter- temporal changes in inequality. The exercise illustrates the danger of comparing the data on inequality from different surveys after a superficial examination which does not first establish comparability. The chapter is divided into three major sections. The first section deals with the Post-Enumeration Survey as the data base for the entire study; it contains subsections on sample design, definition of PES income, and coding and estimation of PES income data. The second section deals with broad features of the PES household income distribution; it contains subsections on household income inequality, estimates of average household income based on PES and national accounts, and comparisons of inequality in Malaysia with that in other countries. The final section deals with intertemporal comparisons of inequality in Malaysia; it contains subsec- tions on the Malaysian Socio-Economic Sample Survey of Households (MssH), the Household Budget Survey (HBS), the sample design and income definition of HBs, and a detailed comparison between PES and HBS. The 1970 Post-Enumeration Survey The Post-Enumeration Survey (PES) is a very large sample survey covering some 135,000 individuals, or approximately 1.5 percent of the population of Peninsular Malaysia. It was conducted in September 1970, immediately after the 1970 Census of Population and Housing. Its chief purpose was to check on the undercoverage and content of the population census, but some family planning and income questions were included as well. The survey was thus primarily designed to check in detail such census items as age and education, rather than to measure income distribution accurately. Since the PES was not an income survey as such, it is possible that the usual errors associated with responses on income may have been aggravated. These are nonsampling errors which arise from incorrect or biased responses, including deliberate understatement of income at the upper end of the distribution and possible overstatement at the lower end. Such errors are generally minimized by asking a multiplicity of related questions which serve as indirect checks on the reported level of income. COMPARISONS OF PES WITH OTHER SURVEYS 23 It was recognized that a survey which includes only a few questions on income may suffer from poor and evasive responses. Therefore, con- siderable probing and prompting were attempted in the PES to obtain satisfa,ctory income data. The following instruction was issued on this subject to field interviewers: Conceptually and operationally, this is the most difficult information to obtain. Very often people are afraid of telling their true income, and are often prone to understate the amount they receive. They blow up the difliculties and underplay the receipts. Many even forget certain kinds of income that have to be incorporated in the overall monthly income. Hence it is necessary to be patient, tactful, and at the same time probe to ensure no major sources of income are missed out. Ideally, we would like to ask a multitude of questions and from that derive the correct and complete incorme of the earning members of the household. But this is not an Income Survey, hence we will have to obtain the correct responses with the least number of questions. A large part. of the accuracy of this data depends on your own ingenuity as an interviewer, and how best you can probe and probe. Let us try and explain this further. It is obvious that on visiting a particular household and talking to the respondent, you will have a feeling on two things-(a) if the household is in the very low income group, low income group, middle income group or high income group; and (b) the type of occupation in which the earning members are generally engaged. Using these two, you can judge if the reported income seems too high or too low. If you feel the reported income seems incorrect, keep probing . . . If you know a household is in a high income group, based oni the Radio, Television, Air-conditioner, Cars, etc. in the place, then clearly you should keep asking simple but effective questions to trace if th,ere are further sources of income which have not been mentioned (Department of Statistics, 1970b, p. 31). To assess the reliability of data collected, an elaborate procedure of checks and rechecks was instituted at the levels of field interviewer, regional office, and headquarters (see the appendix to this chapter for some of the checks). The checks were carried out by field interviewers2 and a small, 2. Field interviewers possessed at least a school certificate (that is, they had graduated from secondary school) and underwent intensive training for four weeks on the PES questionnaire. The average age of interviewers was around twenty-one years, and they were paid a monthly income of M$180 plus M55 outstation allowance (equivalent to US568 in 1970 U.S. dollars), which was well above the average personal income in Malaysia (see chapter 6). To avoid communication and other barriers, households were interviewed by persons of the same racial group. 24 INEQUALITY AND POVERTY IN MALAYSIA well-qualified group of full-time field enumerators supervised by full-time regional supervisors. Enumerators were sent back to the field if, in the opinion of supervisors, income data had not been collected satisfactorily. Although there are bound to remain inaccuracies in the data, the PES does appear a fairly reliable source of information on personal incomes in Malaysia. PES Sample Design The PES sample design is not yet officially documented. Much of the description here is based on discussions with V. T. Palan, senior statistician in the Malaysian Department of Statistics, who was responsible for the survey. Briefly, the sampling procedure for PES was as follows. The interview unit (or "ultimate" sampling unit) was the household, defined as a group of people who live together with a common budget for food and other essentials of living. The following instruction was given to lield enumerators on the definition: A household may be either a one-person household or a multi-person household. (i) A one-person household is one where a person lives alone in a separate room or rooms and is part of a dwelling unit but does not join with any of the other occupants of the dwelling unit to form part of a multi-person household as defined below. (ii) A multi-person household is a group of two or more persons who combine to occupy the whole or part of a housing unit and provide themselves with food or other essentials of living. The group may pool their incomes and have a common budget to a greater or lesser extent. The group may be composed of related persons only, or only of unrelated persons, or a combination of both; as for example, a family with servants who spend their whole lives with them. The general criteria which should be used in identifying members of a multi-person household are common house-keeping arrangements, sharing the principal meals in the sense that the household's food supply is obtained for common consumption or paid for out of a common budget and having common arrangements for supplying basic living needs and are normally living together (Department of Statistics, 1970a, pp. 4-5). Since the PES was a follow-up to the 1970 Population and Housing Census, its sample frame was naturally based on the latter. The country was partitioned into artificial geographical areas called enumeration blocks, which formed the "basic" (or "elementary") sampling unit. Altogether there were 15,594 enumeration blocks covering Peninsular Malaysia, with each one estimated to contain approximately sixty to eighty households. COMPARISONS OF PES WITH OTHER SURVEYS 25 A two-stage stratified sampling technique was employed to select the households. At the first stage the universe of 15,594 enumeration blocks was stratified into sixteen areas according to rural-urban and regional characteristics; for this purpose, Peninsular Malaysia was divided into three strata: Metropolitan towns: those with an estimated population in excess of 75,000 Towns: those with a population between 10,000 and 75,000 Rural areas: places with a population less than 10,000 and five regions: South: the states of Johore and Malacca Central: Selangor and Negri Sembilan Northwest: Perak North: Penang, Kedah, and Perlis East: Kelantan. Trengganu, and Pahang. The lthree strata and five regions were used to delineate the following sixteen areas (see the frontispiece map): Metropolitan towns in Peninsular Malaysia, that is, Johore Bahru, IMAalacca, Kuala Lumpur, Klang, Ipoh, and Georgetown (6 areas) Towns in each region (5 areas) Rural areas in each region (5 areas). A sample of 1,138 enumeration blocks was selected from the stratified universe of 15,594 using probabilities proportional to the size of the block. The blocks selected contained some 670,000 persons, or approximately 8 percent of the total population. The census books of house-listings provided the frame for the second stage of selection. The private households listed in the census book for each selected enumeration block were serially renumbered, and a random sample of households was drawn from this enumeration.3 This two-stage sampling procedure led to the ultimate selection for interview of about 27,000 households in the country. Owing to nonresponse,4 however, income information was obtained for only 25,023 3. The sample represented all segments of the total population with one small exception unattached males living in barracks in military encampments were not represented. They constituted an unknown fraction of the 45,903 military personnel that occupied military establishments. 4. Nonrespondents were those households or individuals within households who refused to provide information for PES, and those listed in the census frame who had moved out after the census. There was a gap of three to four weeks between the census and the PES. 26 INEQUALITY AND POVERTY IN MALAYSIA households, or 134,186 individuals.5 No estimates of sampling error have yet been computed. The absolute sample size is so large, however, that the sampling errors on income are likely to be quite insignificant.6 The check for undercoverage in the census consisted of relisting households in the enumeration blocks selected at the first stage. The first task was to relist actual houses (living quarters), which made possible an identification of those that had been missed in the census. The second task was to establish by means of interview the number of households and individuals in each house, and the results were compared with information obtained in the earlier census interviews. When there were unreconciled discrepancies, a further visit was made to the household in queslion. Through this process of repeated visits, individuals who had been overlooked in the census were discovered. The undercoverage check revealed an underenumeration in the census of 4.05 percent, which has a sampling error of 0.48 percentage points. Hence the PES adjusted popu- lation of Peninsular Malaysia in mid-1970 was 9,182,000, with a 95 percent probability that the true population lay within plus or minus 46,000 of this estimate. The content check on the census consisted of re-interviewing intensively the households selected at the second stage of the PES sample. The selected households were re-interviewed on such census items as age and education, and questions were also asked on individual and household incomes. The interview was usually conducted with the household head,' but for income questions individual recipients were interviewed. If they were not available during the first visit, the interviewer came back at a later time to see them. The percentage of the population of Peninsular Malaysia for whom "valid" income data were thus obtained is 1.46 percent (or approximately one person in sixty-eight). 5. Income information was actually obtained for 25,025 households or 134,192 individuals. However, the racial group of two households (six individuals) was not available, and the effective sample size was reduced to 25,023 households or 134,186 individuals. 6. Dr. Roe Goodman, the UNDP adviser on sample surveys in the Malaysian Department of Statistics, has prepared some preliminary estimates of standard error for the sample proportions (and cumulative proportions) in twenty-three PES income intervals. 7. The household head was identified largely by the members themselves. The following instruction was given to field enumerators: "By head of household we mean a person who is accepted by the rest of the members as being the person who makes a major decision in the household. By and large, this will be the husband or in very rare cases the wife. Generally this will offer very little problem since the households themselves will tell you who is the head. Obtain the names in full. In case the household consists of unrelated persons, write the name of the person the others accept as the head. If this is not possible, write any one of their names" (Department of Statistics, 1970a). COMPARISONS OF PES WITH OTHER SURVEYS 27 Definition oJ PES Income The PES concepl of individual and household income appears to be fairly comprehensive, including income received in kind as well as cash. A money value was imputed to receipts in kind, to own consumption from production, and to owner-occupied housing. Eleven major categories of income were distinguished to enable an accurate estimate of true income (see Department of Statistics, 1970c, sec. E on "Individual and Household Income"). They are: 1. Wages, salaries, and other receipts 2. Income from sale of produce 3. Income from jointly owned business or farm 4. Income from rent and investments (excluding capital gains) 5. Pensions, remittances, cash allowances, royalties, fees, and other receipts 6. Other periodic cash receipts, such as alimonies and scholarships 7. Money value of income in kind such as food, clothing, and housing '3. Money value of own consumption of produce 9. Money value of goods received from other sources 10. Imputed rent of owner-occupied house 11. Other concessions. T'wo separate questions on income were in fact asked of respondents (see the appendix to this chapter, Questions 5 and 6 from section E of the questionnaire). The first simply asked the average monthly income (over the previous year) of each person in the household in receipt of some form of income. The response was designated as "stated income" and recorded. This is the respondent's own perception of income and probably cor- responds to cash receipts (categories 1-6 above). The second question probed income recipients about each of the eleven sources of income and designated the total for each recipient as "computed income." The instruction given to field interviewers was: "This question is primarily airned at obtaining data on income by going through each and every source of income. By so doing we hope that the respondent will be able to think about each of ihe categories carefully, and say if he had any income from that source" ([)epartment of Statistics, 1970b, p. 37). In contrast to stated income, therefore, computed income includes all the imputed values on the checklist and is likely to be a better estimate of true income.8 Although 8. Some categories of income, such as capital gains and production for own investment, seom to have been omitted from the checklist. The latter category includes own farm improvements, such as fencing and embankments for irrigation ditches, which are not easy to value, in part because there is no market for them. 28 INEQUALITY AND POVERTY IN MALAYSIA information on each of the eleven components of income was collected and entered separately in the PES questionnaire form (Department of Statistics, 1970c), only the total, unfortunately, was coded onto the data tapes. The opportunity for many interesting analyses on income distribution has thereby been lost. PES income refers to gross or pretax income for the average month, and no particular reference month is specified. Data collected in this manner suffer from the drawback that recall lapse is increased (in effect, the respondent is being asked to estimate annual income and divide it by twelve), but there is the advantage that seasonal variations in income are ironed out over the year. The need to avoid questions that are too specific has been pointed out by the senior statistician responsible for designing the PES: Very specific answers to even very recent instances may be un- representative and misleading. For example, information on the pre- vious month's cash income may be completely misleading if it happened to be a festival month and workers had been paid bonuses. Similarly since many activities vary seasonally, questions relating to behaviour on a particular day or week may elicit far less accurate data than questions on the usual or "average" behaviour (Palan, 1968, p. 40). For some categories of income, the data were sought for the previous year and then converted into an average monthly figure. For example, to obtain a farmer's monthly income from sale of produce (category 2), the PES interviewer obtained data on farm output and costs on an annual basis and divided the difference by twelve. The following instruction was issued to field interviewers for the estimation of household business and farm incomes: If the earning members of the household are engaged in, say, a business activity, then the monthly income would be the annual profit, divided by twelve. Remember, if someone says that his wife and daughter are family workers, they will not have a separate income. The wages that should have been paid to them would have been included in the net profit. If, however, the wife and daughter are treated as employees, then their incomes would have to be accounted for separately. Again, in the case of self-employed persons, ask and compute the monetary value for the owner-occupied house (even if part of the shop), goods consumed from the shop, free electricity and water supply, etc. Similarly, in the case of a farm worker his income would be the sale of total produce for the year divided by twelve. Plus the value of the produce bartered, consumed, and stored for future use. Also include income from miscellaneous sources-like government aid, scholarships COMPARISONS OF PES WITH OTHER SURVEYS 29 to children, free books, subsidized items like fertilizers, etc., receipts from working sons and daughters elsewhere, etc. Farm expenses such as for seeds, fertilizers, etc. should be subtracted before obtaining the income. Generally, a family or unpaid helper will not receive any income. But even so, confirm this with the respondent. Agairi, unpaid helpers sometimes work as employees for others; you should obtain the average monthly income from this (Department of Statistics, 1970b, p. 31). Perhaps the most serious shortcoming of PES iS that prices for the impujtation of payments in kind and own consumption of produce were largely left to the discretion of respondents, with only minimal checks. This can introduce significant measurement errors in the income of subsistence and peasant farmers, as the difference between farm-gate (producer) and retail (consumer) prices is sometimes quite large. According to the senior statistician for PE-S, however, it is likely that market prices were used to value the own consumption of a farmer if he was a net buyer, and farm-gate prices if he was a, net seller.9 The part of a farmer's produce actually sold (income category 2) would necessarily have been valued at farm-gate prices. The method used to impute the rental value of an owner-occupied house (income category 10) was, whenever possible, simply to take the actual rent on a similar house in the neighborhood. If no neighboring houses were rented, an attempt was made to establish the current replacement value of the property from its owner or from some other knowledgeable person in the area. The replacement value was estimated by costing the building materials, labor services, and other inputs at local prices. A monthly rental value was estimated by converting the replacement value into an annual stream at 10 percent and dividing by twelve. The procedure was sometimes difficult to follow in rural areas, owing to imperfect markets for rented housing and limited knowledge about cost structures. Where no infor- mation could be obtained, a monthly rental value between M$ 10 and M$ 15, depending on condition, was assumed for the typical kampong (village) house with attap (thatched) roof. By contrast, the monthly rent for a typical two-storied shop-house in urban areas was about M$60. 9. According to the senior statistician for PES, if the farmer consumes his entire produce and is a net buyer, he is likely to have the market price in mind rather than the farm-gate price. But if he consumes a part of his produce and is a net seller, he is more likely to have the farm-gate price in mind, as this is the price he receives for his sales. Imputation at these prices does reflect con-ectly the marginal value to him of consumption. When the farmer consumes his entire produce and is neither a net buyer nor a net seller, the value to him of marginal consumption lies between the retail and farm-gate prices. The satisfaction of these marginal conditions does not, of course, resolve the problem of comparing standards of living across households when their consumption has been valued at different pnces. 30 INEQUALITY AND POVERTY IN MALAYSIA The Coding of PES Income Data The actual incomes received by individuals, in Malaysian dollars (M$) per month, were recorded on the completed questionnaries by interviewers. These data were then grouped and coded according to two different interval classifications with nine and ten intervals, respectively. The upper open- ended income class began at M$750 for one classification and at M$980 for the other.1" Subsequently, the Department of Statistics prepared another, more detailed coding of incomes into twenty-three intervals with the uppermost interval beginning at M$5,000. I have meshed the three different codings to create a new thirty-two- interval classification of PES income data. The result of combining the original two classifications is to make the intervals much finer at lower income levels. The third classification adds many more intervals at upper income levels above M$980 (see table 2-1). The new classification obviously permits more accurate estimates of Malaysian income distribution, which is especially important at the lower end of the distribution for a detailed analysis of poverty (chapters 4 and 5)." Estimating the Household Income Distribution from Coded Data To estimate the household income distribution it is necessary to know the actual income of each household. Unfortunately, this is no longer available because the income data were coded into intervals! 12 The problem, therefore, is to reassign an income level to each household, given the interval in which it falls. It is assumed that all households within an interval receive the mean income for that interval; thus it remains to estimate interval means for each income class, including the open-ended one. Interval means for the upper income classes were calculated by assuming that household incomes in this range follow a theoretical Pareto distri- bution."3 Interval means for the lower income classes, which are fairly 10. See Department of Statistics (1973), pp. 4-5 and 10-11. I1. The greater number of income classes also reduces income measurement error in the estimation of earnings functions (chapter 7). 12. The problem addressed in this subsection arises solely because the income data were grouped into intervals. It is not known why actual income figures were not coded onto the data tapes. 13. There are various economic models, stochastic and deterministic, which predict a theoretical Pareto distribution for the upper income ranges. This distribution is also known to fit extremely well at high income levels in a number of countries. COMPARISONS OF PES WITH OTHER SURVEYS 31 narrow, were simply chosen at the interval midpoints. The truncation point for the Pareto distribution, that is, the distinction between upper and lower income classes, Wits determined by an ad hoc goodness-of-fit criterion. The cumulative Pareto distribution in log-linear form can be written as log n(y) = log A - a log y, where y denotes household income, and n(y), the number of households with income greater than or equal to y. This equation was estimated sequentially for an increasing number of income classes, starting from the top down. The R2's and &'s were computed for each equation. Subject to the constraint that K'R exceeds 0.99,1 chose the equation which maximized the fittecl portion of the distribution (that is, the equation which included as many of the upper income classes as possible)."4 This somewhat arbitrary procedure led to the selection of the equation fitting the top eighteen income classes (more than M$500 per month), which account for 10.7 percent of households: log n(y) = 9.3-2.1455 logy t-ratio of a = 48.59 R2 = 0.993 Degrees of freedom = 16. M[ean household incomes for the top eighteen intervals were calculated using this equation with coefficient a = 2.1455. Given a continuous Pareto distribution with coefficient x, the mean income in any interval [a, b] can be shown to be Thus the mean income for the upper open-ended class (a = M$5,000, b = oc) was computed as M$9,365.15 The means computed for the other seventeen intervals at the top are shown in table 2-1. Owing to the shape of the Pareto distribution, it is clear that the means computed by this method will be smaller than the interval midpoints. The differences, however, turn out to be not large. Mean incomes for the fourteen remaining classes at the bottom of the disiribution were assumed to be the interval midpoints. Since these intervals are relatively narrow, any errors from this assumption are likely to be quite small. Hence this procedure was considered as acceptable as one in which a 14. Standard practice appears to be to determine the Pareto coefficient a by fitting such an equation to the top two income classes only. 15. See the later subsection "Household Income Inequality in Malaysia" for the sensitivity of the household income distribution to alternative assumptions concerning mean income of the upper open-ended class. Table 2-1. PES Income Intervals, Means, and Absolute Frequencies (Malaysian dollars per month) Estimated Estimated interval interval Absolute mean Jor Absolute mean Jor Jrequency Code Income Interval household income jrequencj personal income oJ income no. interval midpoint distribution oJ households distribution' recipients' 0 No income 0 0 341 0 0 I 1-39 20 20 1.336 20 6,018 2 40-49 45 45 711 45 2,103 3 50-79 65 65 2,751 65 6,889 4 80-99 90 90 1.898 90 4.273 5 100-129 115 115 2.712 115 5,693 6 130-149 140 140 1.596 140 2,373 7 150-179 165 165 2,251 165 3,217 8 180-199 190 190 1,204 190 1,395 9 200-279 240 240 3.625 233 3,716 10 280-299 290 290 581 290 471 11 300-399 350 350 2,175 343b 1.921 12 400-479 440 440 999 436b 846 13 480-499 490 490 168 490' 119 14 500-599 550 545' 693 545b 536 15 600-679 640 637' 412 637b 292 680-699 690 690' 69 690' 40 17 700-749 725 724' 160 724b 114 18 750-799 775 774' 122 774b 80 19 800-899 850 847' 246 847b 149 20 900-979 940 938' 142 938 101 2i 980-999 990 990n 29 990b 5 22 1,000-1.249 1,125 1,110 295 1,I1Ib 180 23 1,250-1,499 1,375 1,363a 163 1,363b 72 24 1,500-1,749 1,625 1,615 100 1,615b 48 25 1,750-1,999 1,875 1,866 44 1,867b 24 26 2,000-2,499 2,250 2,221- 83 2,222" 57 27 2,500-2,999 2,750 2,726 44 2,727b 27 28 3,000-3,499 3,250 3,230 26 3,230 17 29 3,500-3,999 3,750 3,733 ~ 19 3,733 13 30 4,000-4,999 4,500 4,442 11 4, 6 31 5,000+ - 9,365' 19 9,736 II All 0 + - 264 25,025 163 40,806 -Not applicable. a. Estimated by assuming Pareto distribution with coefficient 2.1455 valid in the range. b. Estimated by assuming Pareto distribution with coefficient 2.0559 valid in the range (see "The Distribution of Income Recipients by Personal Income" in chapter 6). c See chapter 6 for an analysis of the personal income distribution. 34 INEQUALITY AND POVERTY IN MALAYSIA separate theoretical distribution, such as the lognormal, is fitted to the lower tail and the interval means computed from its estimated parameters. Estimation of the Gini Coefficient Various indices of inequality have been used in income distribution studies (see appendix A, "A Brief Review"). The Gini coefficient, however, remains the most popular measure in use, and it is also estimated here. My definition of the Gini coefficient is based on the Lorenz curve for an income distribution (see appendix A), which is estimated here by linear interpo- lation between observed points. In general, there will be thirty-two observed points corresponding to the thirty-two-interval classification of PES income data. The Gini coefficient is defined as the ratio of the area between the Lorenz curve and the diagonal of perfect equality, to the area of the triangle below this diagonal. The actual formula used to compute the Gini coefficient G is equivalent to this definition (see appendix B, "The Gini Coefficient"): 31 G=1- E (FiI- Fi)(DP,I+ +D,) i=0 where F3 is the cumulative population share, and 'i is the cumulative income share, corresponding to the iph interval. It can also be expressed as 31 G = (Fjti+i, l-F,+ ,I (i). 5=1 The approximation of the Lorenz curve by a piecewise linear interjpo- lation of points on it underestimates the Gini coefficient. This is because the line segments in the approximation always lie above the true Lorenz curve, which is easily shown as convex (appendix A). Since there are as many as thirty-two well-spaced points on the curve, however, the piecewise linear approximation of it should be reasonably accurate. At any rate, a comparison between Gini coefficients of different distributions is unlikely to be biased when their Lorenz curves are similarly closely estimated."6 The PES Household Income Distribution A detailed discussion of inequality in Malaysia is deferred until chapter 3. This section presents summary information on average income levels and 16. Forthis reason, I did not feel itworthwhile toestimate the Lorenzcurve in astatistically more sophisticated fashion-for example, by the methods recently proposed in Kakwani and Podder (1973, 1976). Such a procedure would allow us to determine confidence limits for the parameters characterizing the Lorenz curve and Gini coefficient. COMPARISONS OF PES WITH OTHER SURVEYS 35 inequality, as revealed by the PEs, The information is presented in terms of the household income distribution, which is the distribution reported by most surveys in Malaysia and elsewhere. It is later argued that this distribution does not provide a good indication of inequality in levels of living because it takes no account of household size and composition, and more appropriate distributions are considered in chapter 3. Primarily for comparison with other surveys, I begin by setting out some relevant features of the: PES household income distribution. Household Income Inequality in Malaysia Table 2-2 summarizes the household income distribution from the PES and two previous Malaysian surveys (discussed in the next section). The PES shows an average household income in 1970 of M$264 (approximately US$100) per month and a Gini inequality coefficient of 0.5129. Since estimates of the PES household income distribution are all based on interval means computed from a specific Pareto distribution, their sensi- tivity should really be tested with respect to alternative assumptions about the underlying theoretical distribution. The only interval mean that can be affected significantly is the one for the upper open-ended class; the other interval means are likely to remain close to the interval midpoints for distributions with "reasonable" shape. Hence I have tried to test the sensitivity of the estimates for average household income and household income inequality with respect to alternative assumptions for mean income in the class M$5,000 and over. The results show that the estimates are not unduly sensitive: Mean oJ Mean assumed household Gini coefficient Theil T index Jor income class MS5,000 + income of household oJ household per momil distribution income distribution income distribution 5,000 261 0.5066 0.5028 9,365 264 0.5129 0.5378 10,000 265 0.5136 0.5434 15,000 268 0.5205 0.5900 20,000 272 0.5272 0.6399 As shown in appendix E, corollary 1, one expects inequality in the Lorenz sense to increase as the income of the richest person(s) in the distribution is incireased. Since both the Gini coefficient and the Theil Tindex satisfy mean independence, population-size independence, and the Pigou-Dalton con- dition, they should show more inequality for a distribution which is Lorenz- doiminated (see the proposition in appendix D); this is borne out by the figures above. Table 2-2. Household Income Inequality in Malaysia HBS, 1957-58d MSSH. 1967-68b PES, 1970 Mean house- Mean house- Mean house- Area hold income Gini hold income Gini hold income Gini and race (MS per month) coefficient (MS per month) coefficient (MS per month) coejficient Area Peninsular Malaysia 199 0.3705 217 0.5624 264 0.5129 Rural Malaysia 170 0.3549 114 0.4794 200 0.4689 Urban Malaysia 261 0.3514 283 0.5224 428 0.5037 Race Malay 144 0.3410 130 0.5072 172 0.4664 Chinese 272 0.3322 321 0.5081 394 0.4656 Indian 217 0.3117 253 0.4974 304 0.4722 Other n a n.a. 839 0.4912 813 0.6673 Sample size (number of households) 2,7601 30,000d 25,023 n.a. Not available. a. There are twenty-eight size intervals in the Household Budget Survey (HBS), with no open-ended income class. The topmost income class is M$900-M$1,000. There is also no separate zero-income class, zero incomes being lumped in the range of M$0-M$25 per month (see Department of Statistics [1961?], p. 39; hereafter cited as HBS Report). b. The income data for the Malaysian Socio-Economic Sample Survey of Houscholds (MssH) are presented in seven income classes (including a zero- income category), with the top open-ended class beginning at M$750 per month A Pareto distribution was fitted through the last two income classes alone, giving & = 1.7622 and a mean of M$1,735 for the open-ended interval. c. Of the 2,760 households sampled, 1,920 were in rural areas and 840 in urban areas. Since the HBS Report did not break down sample size by ethnic group and urban-rural location, I have chosen the same proportions as in the 1957 population census. d. Approximate. Sources: HBS Report; Department of Statistics (1970g); and PES, 1970. COMPARISONS OF PES WITH OTHER SURVEYS 37 The mean household income varies among racial groups from M$172 per month for Malays to M$813 per month for "other" communities (including Europeans, Thais, other Asians, and so on). The mean household income of Chinese is M$394 per month, and of Indians M$304 per month. Thus the oft-quoted racial disparity ratios, which are supposed to indicate average income differences between the races, are 2.29 for Chinese:Malay, and 1.77 for Indian:Malay. Despite these large differences in average household income, the Chinese and Malay distributions appear very similar about their respective means: for Malays the Gini coefficient is 0.4664, and for Chinese it is 0.4656. The Indian distribution is just a little more unequal, with a Gini coefficient of 0.4722. The "others" distribution is very much more unequal, and its high Gini coefficient of 0.6673 reflects the large income disparities among the Europeans, Thais, and other Asians who constitute this group.17 Further details of the PES household income distribution, with a breakdown by racial group, may be found in table 3-7. The PES disaggregates urban areas into metropolitan towns and towns,'h bul. sometimes I treat them together for uniformity of comparison with earlier surveys. rhe PEs average household income in urban areas is M$428 per month while that in rural areas is M$200 per month, implying a disparity ratio of 2.14. Rural household incomes are distributed signifi- cantly more equally than urban household incomes (Gini coefficient of 0.4689 compared with 0.5037), and the incomes within each location are distributed more equally than in Peninsular Malaysia as a whole (Gini coefficient of 0.5129). For the racial disparity ratios within each location, urban areas are split into metropolitan towns and towns (not shown in table 2-2). In metro- politan towns the average household income is M$495 per month, and interestingly the racial disparity ratios are quite small: 1.18 for Chinese:Malay and 1.15 for Indian:Malay. In towns the average household income is M$345 per month, but the racial disparity ratios are bigger: 1.52 for Chinese:Malay and 1.20 for Indian:Malay. In rural areas the racial 17. The relatively large European incomes account both for this inequality and for the high average income of this group (MS813 per month). 18. The PEsdefines metropolitan towns as Johore Bahru, Malacca, Kuala Lumpur, Klang, Ipoh, and Georgetown. Towns (with population of more than 10,000) are: Kula], Kluang, Segamat, Muar, and Batu Pahat (in Johore); Alor Star, Sg. Patani, and Kulim (in Kedah); Kota Bahru, Pangkat Kalong, Peringat, Tumpat, and Pasir Mas (in Kelantan); Seremban and Kuala Pilah (in Negri Sembilan); Kuantan Bentong, Raub, Temerloh, Mentakab, and Kuala Lipis (in Pahang), and Butterworth, Bukit Mertajam, Ayer Hilam, Tanjong Tokong, and Tanjong Bunga (in Penang). All other areas are rural areas (see Department of Statistics, 1973, p. 1, and the frontispiece map). 38 INEQUALITY AND POVERTY IN MALAYSIA disparity ratios are the largest of all locations: 2.22 for Chinese:Malay and 1.58 for Indian:Malay. The still higher overall racial disparity ratios [or Peninsular Malaysia reflect the disproportionate presence of Chinese and Indians in urban areas and of Malays in rural areas. (The between-location disparity ratios are 2.48 for metropolitan towns:rural areas and 1.73 lor towns:rural areas.) Comparison of Income Estimates: PES and National Accounts A comparison of the PES estimate of average household income in Malaysia with estimates derived from the national accounts may be suggested as a rough check on the quality of the survey data, but there are two problems associated with it. First, the correspondence is often tenuous between the income concepts of a survey and those of the national accounts. Second, the national accounts data cannot themselves be presumed sacrosanct. As is well known, they are subject to substantial error, particularly in developing countries, and this error is more pronounced for estimates of income level than for estimates of income growth, since national accounts data are at least collected on a systematic basis over time.19 The national accounts concept of personal income corresponds most closely to the concept of income used in surveys such as the PEs. Unfortunately, however, the national accounts in Malaysia do not provide direct information on aggregate personal income. This can be estimated approximately, starting with private consumption expenditure20 and adding to it various items such as personal saving and direct taxes on households. Aggregate personal income has been estimated in this manner by Snodgrass (1975), who calculates personal saving by a residual method (netting out from gross domestic capital formation the sum ofdepreciation, government saving, surpluses of public corporations, the current account deficit, and company saving). A further small correction was made to this figure for aggregate personal income to account for the part of it that accrues to persons not living in private households. Snodgrass thus obtained an estimate for average household income in 1970 of M$370 per month, which he called a "rough estimate only." Another estimate for Malaysian average household income, also based on the 1970 national accounts, has been derived in a somewhat different 19. They are collected on a commodity flow basis, but there is a large element of discretion in allocating estimated supplies to alternative uses. 20. As a comment on the accuracy of national accounts data, a 1976 World Bank economic report on Malaysia presents three different estimates for private consumption expenditure in 1970: M$7,151 million, M$6,O14 million, and M$6,349 million. COMPARISONS OF PES WITH OTHER SURVEYS 39 way from the Snodgrass estimate. Ahluwalia and others (1976, table IV-6)21 have constructed a social accounting matrix for Malaysia, which is essentially a presentation of the national accounts in a disaggregated framework. Resorting to "heuristic judgment" when data were inconsistent, they arrive at a figure of M$352 per month for average household income in 1970. The average household income computed from PES iS M$264 per month (table 2-2). This is 71.4 and 75.0 percent, respectively, of the estimates of Snoclgrass and Ahluwalia and others based on the national accounts. Thus, the PES income estimates seem to be about 25 percent less than those derived from the national accounts. This degree of understatement is not par- ticularly large by the standards of household surveys in developing countries and is only fractionally larger than that recently reported in surveys from developed countries. For six developed countries, Sawyer compares household survey estimates of income with national accounts estimates of the same categories of income. The survey estimate for pretax household income22 as a proportion of the national accounts estimate turns out at 70.5 percent for France (1970), 81.7 percent for Germany (1969), 83.9 percent for the United Kingdom (1973) Family Expenditure Survey, and 82.8 percent for the United Kingdom (1972-73) Blue Book data (see Sawyer, 1976, app. 2). The degree of underreporting varies between categories of income and is particularly high for income from investment and self-employment. As these categories of income are "disproportionately received by the upper deciles," Sawyer concludes that "inequality will tend to be underestimated everywhere." It is not obvious that similar underreporting of income in a country such as Malaysia (if underreporting were established) would bias inequality there. The reason is that the large class of self-employed in Malaysia is spread throughout the distribution, with a substantial rep- resentation at the lower end among subsistence farmers (see chapter 4 and appendix E). The understatement of income in PES in relation to the national accounts seem,s broadly in keeping with what is observed elsewhere. In Malaysia, though, there is less of a presumption that the estimates of household incorme based on the national accounts are more accurate than the direct estimates from PES, which are at least based on a statistically well-designed and well-implemented sample survey. The discrepancy in estimates from the two sources is not easy to resolve, but it cannot be taken as a reason for rejecting the PES. 21. A later versior. is Pyatt and Round (1977). 22. This is the same income concept as in PES, so that a uniform comparison can indeed be made between developed and developing countries. 40 INEQUALITY AND POVERTY IN MALAYSIA International Comparisons of Inequality Attempts at comparing income inequality in Malaysia with inequality in other countries will obviously encounter serious difficulties because income distribution data are typically noncomparable between countries. The careful study required to establish comparability and bring the data onto a common basis IS a large research project in itself and plainly outside the scope of this enquiry. Here I merely mention some countries, at similar stages of development, for which it would be interesting to pursue inequality comparisons. Table 2-3 lists some countries at similar stages of development for which income distribution data are reported in the compilation by Jain (1975). When income distribution data for a particular country are not available for 1970, the surveys carried out in the closest year are reported. Each country in the list is followed by a number in parentheses which shows the source number in Jain's compilation. According to Jain, the Gini coefficients in table 2-3 refer to the household income distribution and a sample coverage which is national. She employs a uniform method across countries to compute the Gini coeflicient.23 Table 2-3. Household Income Inequality in Selected Economies GNP per capita Year in 1970 Gini Economy of survey (USS) coefficient Malaysia (3) 1970 380 0.5179 Taiwan (4) 1972 390 0.2843 Philippines (4) 1971 210 0.4941 Korea, Rep. of (5) 1970 250 0.3719 Korea, Rep. of (6) 1970 250 0.3836 Thailand (I) 1962 200 0.5103 Brazil (4) 1970 420 0.5744 Brazil (7) 1970 420 0.6093 Mexico (5) 1969 670 0.5827 Mexico (4) 1968 670 0.6106 Mexico (3) 1967-68 670 0.5243 Turkey (I) 1968 310 0 5679 Zambia (1) 1959 400 0.5226 Note. Numbers in parentheses refer to source number in Jain (1975). Sources 1972 World Bank Atlas (Washington, D. C.); and Jain (1975). 23. Jain (1975) computes the Gini coefficient for each distribution from an estimate for the Kakwani and Podder (1976) functional form of its Lorenz curve. My procedure in the previous section for estimating the Lorenz curve by linear interpolation between observed points could lead to a differential underestimation of the Gini coefficient across countries. This is because the number and spacing of such points can vary substantially from survey to survey. COMPARISONS OF PES WITH OTHER SURVEYS 41 In table 2-3 1 attempted to standardize for population unit and sample coverage across surveys,24 but major problems still remain which could lead to noncomparability. These concern differences among the surveys in sampling and nonsampling errors and, perhaps most important, in the definitions of household and income. (Is income pre- or posttax? Are own consumption of production and payments in kind included? Is owner- occupied housing imputed?) The definitions and measurement techniques, however, are not easy to verify,25 since the published reports often do not provide adequate information on the subject (see Jain, 1975, p. xi). I have not checked the primary data sources to establish uniformity of the concepts used in the various surveys. Rather, the distribution data in table 2-3 are presented simply to give some idea of the estimates floating around as researchers try to discover "stylized facts" about inequality and development (such as the inverse U-shaped relation). It is indeed interest- ing to examine the extent to which levels of inequality can actually differ at similar stages of development. Table 2-3 shows a wide range in inequality levels for countries at approximately similar levels of development. The Gini coefficient varies from 0.28 for Taiwan to 0.61 for Brazil and Mexico. Malaysia, with a per capita GNP of US$380 in 1970, shows considerably more inequality than Taiwan with a per capita GNP of US$390, but less inequality than Brazil with a per capita G3NP of US$420. One estimate for Mexico shows inequality there at approximately the same level as in Malaysia, while two others show a markedly higher level of inequality. Both estimates for the Republic of Korea show less inequality than in Malaysia. The extent of inequality in Malaysia seems generally similar to that in Thailand, the Philippines, and Zambia. For what they are worth, then, these figures suggest that Malaysia displays a middle-to-high level of inequality by comparison with other countries at similar stages of development.26 Without a good deal of further research into the underlying surveys, however, this conclusion cannot be asserted with any certainty. 24. Even such basic comparability is sometimes not established by researchers undertaking cross-country studies of inequality. 25. My experienc.: with intertemporal comparisons in Malaysia confirms this, as shown in the iiext section. 26. By comparison with the three developed countries mentioned in the previous subsection, Malaysia shows significantly greater inequality. Using the distribution of household pretax income by decile shares, Sawyer (1976) computes the following Gini coefficients: France 0.416, Germany 0.396, United Kingdom Family Expenditure Survey 0.344, United Kingdom Blue Book data 0.373. 42 INEQUALITY AND POVERTY IN MALAYSIA Intertemporal Comparisons of Inequality in Malaysia Apart from PES, two previous surveys undertaken by the Malaysian Department of Statistics have reported income data: the Malaysian Socio- Economic Sample Survey of Households, 1967-68 (MssH) and the Household Budget Survey, 1957-58 (HBS). While MSSH is easily seen t) be not comparable with PES, it is more difficult to reach a judgment about the comparability between HBS and PES. The reason is that the published report on HBS (Department of Statistics, 1961?), hereafter referred to as HBS Report, gives totally inadequate information on the definition of income used. Several researchers have nonetheless adopted HBS as a benchmark for intertemporal inequality comparisons in Malaysia (for example, Lim, 1974; Hirschman, 1974; Lee, 1975; and Snodgrass, 1975) and reached a conclusion of worsening inequality between 1957 and 1970. Given the importance of this subject, I have attempted to construct an account of HBS income and other survey particulars along the lines of that provided earlier for PES. This account has been pieced together from unpublished records and files in the Malaysian Department of Statistics and from conversations with persons responsible for conducting the survey. It is valuable to document this information in detail, if only to apprise other researchers of the pitfalls in using HBS income data. In table 2-2 there are large differences in measured inequality between HBS, MSSH, and PES. The Gini coefficient of the household income distribution in Peninsular Malaysia is estimated to be 0.3705 for HBS, 0.5624 for MSSH, and 0.5129 for PES. In this section it is argued that these differences are not indicative of actual changes in inequality over time, and might be wholly due to differences between the surveys in income concept and coverage. MoTe fundamentally, I show that the three surveys are not comparable with each other, and that no conclusions can be drawn from them about changes in inequality. At the outset, it is well to point out that intertemporal comparison is a hazardous business owing to differences between surveys in sample design (and size), coverage, nonresponse errors, and so on. There are other technical differences, such as the number and width of income intervals used in coding and collecting data. These differences introduce significant measurement errors in attempting comparison. But the task of comparison is made truly formidable when the definition of income itself varies substantially from one survey to another. It is shown below how the different definitions used in the three surveys render comparisons between them meaningless. COMPARISONS OF PES WITH OTHER SURVEYS 43 MS'SH, 1967-68 The comparison of MSSH with the other two surveys is manifestly spurious. Its income concept is restricted to cash income, whereas the income definition of the other surveys is more comprehensive-even if that for HlBS iS partly elusive (see the later subsection, "Definition of HBS Income"). One would expect cash income to be distributed more un- equally than total income, since most of the poor (such as subsistence farmers) hardly enter the cash economy, whereas the rich derive most of their income in this form. Hence the Gini ratio of MSSH may be larger than that of the other surveys27 wholly because of differences in income concept. This explanation seems to be corroborated by examining the (nominal) mean incomes in HBS, 1957-58; MSSH, 1967-68; and PES, 1970. Table 2-2 shows that average household income in rural Malaysia, as estimated from the surveys, declined from M$170 in 1957-58 to M$114 in 1967-68, and rose again to M$200 in 1970.28 It is the exclusion of noncash income from MSSH, 1967-68, vwhich probably explains the lower average household inconme in the middle of the 1957-70 period compared with either end. The exclusion also accounts for a similar pattern observed in the mean household income: of Malays. It is clear that MSSH would have understated the income of rural and Malay households, which consist largely of self- employed groups receiving a substantial portion of their income in kind (such as owni consumption of produce). The Chinese and Indians, however, probably received a large enough fraction of their income in cash for the surveys to register a growth in their incomes throughout this period. The upshot of the preceding discussion is that MSSH should be dropped as a data source on intertemporal income distribution. It is too obviously not comparable with the other surveys, and its restricted income concept coulcl explain the observed differences in measured inequality and mean incornes. 27. In fact, it is an underestimate in relation to the other surveys. Since there are fewer income intervals in MSSH than in the other two surveys, there are fewer points on the Lorenz curve from which its Gini coefficient is estimated. 28. The income figures in table 2-2 are in nominal, not real, terms. Price inflation during 1957-70 averaged as little as 0.7 percent a year. 44 INEQUALITY AND POVERTY IN MALAYSIA HBS, 1957-58 While the comparison of MSSH with HBS or PES is easily dismissed as invalid, that between HBS and PES has been taken seriously by many researchers (for example, Lee, 1975, and Snodgrass, 1975). In my opinion, however, the published report on the survey is inadequate to allow such a comparison. There is little information in HBs Report on the income concept employed, sample coverage, and so on. Thus I have attempted to reconstruct an account of the HBS sample design and income definition. I then venture a comparison between PES and HBS and show that differences in income concept and coverage bias HBS inequality downward in relation to PES inequality.29 HBS SAMPLE DESIGN. The HBS was designed to obtain expenditure weights for the compilation of a retail price index. The collection of income data was incidental to the survey and was included only as a rough check on expenditure. The survey was based on a multistage sample which covered 2.760 households in the Federation of Malaya-l,920 in rural areas and 840 in urban areas.30 It was conducted over a period of one year, beginning in April 1957 and ending in March 1958. A one-year period was selected in order to take account of seasonal patterns of expenditure. The definition of household adopted in the survey may be obtained from "Course of Lectures for Training Investigators" (Department of Statistics, 1957?), hereafter cited as Course oJ Lectures: A household may consist of one or more persons who (a) live within some residential building or group of adjacent buildings (b) consume food from same kitchen, larder, cupboard or cookcing pot (c) incur common housekeeping expenditure or are charged on combined hotel, boarding or lodging house bill. 29. There is an enormous difference in sample size between PES and "ass: the HBS sample size is so small that its inequality estimate is liable to 'appreciable" sampling error (HBS Repjrl, p. 39). To make statistically significant comparisons of inequality, one needs to quantify confidence limits for the Gini coefficient estimates from the two surveys (for example, as in Kakwani and Podder, 1976). With the given sample sizes, the (95 percent) confidence band for the HBS estimate will obviously be much wider than for the PES estimate. 30. These figuresare from Ihe HisReport. Urban areasweredefined as municipallties, town councils, or town boards with a population of 10,000 or more COMPARISONS OF PES WITH OTHER SURVEYS 45 A domestic servant who satisfies (a) and (b) only is regarded as a member of employer's household (Course of Lectures, p. 8). A national sample: frame was used to select households:3" The rural households were chosen as follows: First the country was divided into 16 areas of approximately equal size by population, each area consisting of a number of administrative districts. One district from each area was chosen with probability proportional to size, the mukims [subdistricts] in the selected districts were listed, and three mukims per district were chosen with probability proportional to size. The investi- gators then visited each selected mukim and drew a sketch map showing the villages and kampongs, and the number of houses in each. These maps were then studied carefully and each mukim was divided into four sub-areas of approximately equal number of houses. From each sub- area ten houses were chosen at random and marked on the maps or, where the houses were numbered, the house numbers were listed. Reserve houses were selected in case of a house being empty, or the head of a household refusing to co-operate; substitutes in the case of non- cooperation were only allowed after the supervisor had visited the household and had also failed to gain any co-operation. iln the urban sample, Kuala Lumpur and Georgetown were included because of their size and importance; the remaining towns of 10,000 and over were strat.ified into towns with population between 10,000 and 25.000, and towns with population over 25,000. These towns were listed and since the over 25,000 group had twice the total population of the 10,000 to 25,000 group, two towns were selected from the former group and one from the latter. The towns chosen being lpoh and Seremban, and Teluk Anson respectively. All these towns were suib-divided into smaller areas of approximately equal size, and house- holds selected at random in a manner similar to that described for the rural areas. The sampling scheme had to be slightly modified when it was found that Indian households in rural areas had been inadequately represented. A list of rubber estates with Indian labour was drawn up and a sample was chosen at random. An investigator was instructed to visit these estates in turn throughout the period of the survey (HBs Report, app. 1, pp'. 61-62). 31. There is a small qualification to this-in Course of Lectures (p. 2): -[Owing to] the emergency . . . certain parts of Malaya are still unsafe for our investigators to work in, so we shall have to exclude all the families living in these unsafe areas from our survey. This would result in a slightly inaccurate population of households, but it cannot be helped, and luckily the nimber of areas which are unsafe is quite small." 46 INEQUALITY AND POVERTY IN MALAYSIA This procedure apparently led to the selection of 2,760 households,32 A breakdown of sample size between the major races was never published, nor is it now available. Where necessary, I have employed the 1957 census proportions. Other communities, such as the Europeans, were excluded from this survey. Apart from the noncoverage of European households in HBS, there is inadequate coverage generally of the higher income groups. According to the report of the Special Advisory Committee on Cost of Living Indices, chaired by H. A. Fell (the chief statistician for HBS), there is an "omission of European budgets from the survey; and [an] unsatisfactory coverage of the higher income groups among the Malays, Chinese and Indians" (F'ell and others, 1959, p. 8). In fact, there is no coverage of households with incomes above M$1,000 per month (see the later subsection "Comparison between PES and HBS"). While this may be unimportant in the construction of price indices for the lower income groups, which was the express purpose of HBS,33 it is clearly crucial for an accurate estimation of income distribution. No estimates of sampling error were made, nor is it now possible to make them. Since the absolute sample size was rather small, in fact, these may turn out to be fairly large. HBS Report (p. 2) itself draws attention to the small sample size of the survey and the possibility of error on this score: "A word of caution is . . . necessary to the user of the data . . . The sample was only one quarter percent of all the households in the Federation, and the sampling-errors must necessarily be large." It concludes with a warning note on the reliability of the income distribution data: 32. This figure is taken from the HBs Report, but it is not clear whether it refers to responding households or to those actually selected for interview. Either way, it contradicts slightly the figures in the only other source I could find on this subject in Malaysia, Chew (1968). According to Chew (table 1-5, p. 43), 2,770 households were actually visited (1,920 in rural areas and 850 in urban areas); but of these, 107 did not cooperate, and information was obtained for only 2,663 households (1,818 in rural areas and 845 in urban areas). 33. The following recommendation was made by the Special Advisory Committee on Cost of Living Indices: A separate index [should] be constructed and published for each ofthe three main races, i.e. the Malays, Chinese and the Indians. The weights should be based on the consumption patterns appropriate to each race as given in the household budget survey. In each case the pattern should be based on the combined consumption figures for the income groups "$1- $150 p.m." and "$151-$300 p.m." In addition, the Malay index should take account of the income group "income not clearly defined" which covers, in the main, padi planters whose average consumption was similar to the groups recommended to be covered by the indices (Fell and others, 1959, p. 3). COMPARISONS OF PES WITH OTHER SURVEYS 47 This sample gives sufficiently accurate results for the determination of weights to be used in the construction of retail price indices, but is subject to an appreciable significant margin of error as a source of information relating to income distribution. It does however give a rough broad indication of income distribution pattern and is therefore published as a matter of public interest (HBs Report, p. 39; emphasis in original). DEFINITION OF HBS INCOME. In any discussion on HBS, it is important constantly to bear in mind that HBS was an expenditure survey. The collection of income data was incidental to the survey, intended solely as a rough check on expenditure. It is not surprising, therefore, that no definition of income is given in the published report on the survey. Although I have gleaned certain aspects of it in my research, I still do not have a complete or precise definition. Since a comprehensive definition was neve r intended, the reconstruction of one is perhaps in any case impossible. The questionnaire for HBS consists of a booklet designed to record a household's dailv expenditure and consumption, item by item, over a period of one month. There is a separate section on "special periodical expenditure" for such items as electricity, rent, and income tax. On the last page of the questionnaire there is a small section on income, with columns for salary and wages; self-employed income; rent, profit, and dividend; and other income (including pension). A final column is supposed to show the total, income from all these sources.34 It is not knowrn if the published income figures of HBS are net or gross of income tax. It appears the data could have been reported to interviewers in either form, with the choice left to respondents. If gross income figures were recorded in the income section of the questionnaire, however, separate figures were collected on income tax payments under special periodical expenditure. The: following instruction was given to field interviewers for filling in the iricome tax item in the section on special periodical expenditure: "When income tax has already been deducted from the householder's income (i.e. he only gives his net income) the investigator need not put down anything for this, but where gross income is given, the amount he paid for Income Tax should be noted down here" (Course oJ Lectures, p. 10). Since the income distribution data published in HBS Report are thc figures that were recorded in the income section, they are likely to be a mixture of net and gross incomes. 11 is not clear whether gifts and allowances were included as income. An instruction to field investigators in Course of Lectures (p. 13) says: "When 34. This information is taken from the only surviving questionnaire of HBS in the files of the Department of Statistics, Kuala Lumpur. 48 INEQUALITY AND POVERTY IN MALAYSIA asking about income . . . sums obtained in the form of small monetary gifts, allowances, etc. need not be recorded in detail-just put down the amount under 'Other Income'." According to HBS Report (p. 3), however, "Items such as loans and gifts were not included as income in the table of income distribution."35 There is further ambiguity in Course of Lectures (p. 11): "As regards allowances for a mother or other relation, the sum should be noted under expenditure for one and income for the other, and the expenditure of the latter recorded." The rent on owner-occupied housing was not imputed as an itern of income (or expenditure). There are published statements to this effect, and it was confirmed in private communication with the chief statistician for HBS: Although some households pay a full or economic rent, other house- holds live in subsidized dwellings, or live rent-free-the latter are either owner-occupiers, or relatives of owner-occupiers who pay no rent, or employees on estates, etc. living in rent-free quarters. We can see the almost insurmountable difficulties in imputing an economic rent. An economic valuation would have to be made of a wide variety of owner-occupier dwellings in order to value the service derived from the dwelling. Original cost or present cost of building would be no guide in the case of the kampongs where labour is provided free by members of the household. Estates have indicated in another survey- national income survey-that they have little idea of the economic rent of their free-quarters (Fell and others, 1959, pp. 8-9). There was no imputation for income in the form of free meals at one's place of work. The rationale given in this case was that "since we are here only collecting income as a check on the expenditure this need not be done" (Course of Lectures, p. 12). It is worth reiterating here the principle that income questions in HBS were included only as a check on expenditure.36 For households whose expenditure was in excess of stated income, there may have been some upward adjustment of income at the checking stage. The following instruction to HBS supervisors is suggestive of this interpretation: 35. On the assumption that loans and gifts were excluded, there might be a slight overestimation of HBS inequality if such items represent transfers mainly from nch to poor households (see Atkinson's theorem in appendix D). 36. This point has been strongly emphasized in private conversations with the chief statistician for HBS. The design of the HBS questionnaire is also consistent with the use of income data to check on the balance between income and expenditure or between receipts and payments. COMPARISONS OF PES WITH OTHER SURVEYS 49 WVhen he [the supervisor] visits the investigator, he should check whether the income and expenditure roughly agrees, so far, and find out why if expenditure is greater than income. Also when the investigators hand in their record books at the end of month, the supervisor should again check to see that the stated total income and expenditure tally (Course of Lectures, p. 4). The income of self-employed households probably refers to gross incorme, because no instructions were given to field investigators to subtract from gross revenue any input costs (such as seeds, fertilizers, rent, electricity, and loan repayments). These costs were in fact itemized under special periodical expenditure, but the purpose there was simply to compute the total expenditure of a household. Since production costs were listedi under total household expenditure, the income-expenditure check would make sense only if recorded income referred to gross receipts.37 It is quite possible, therefore, that the income of the self-employed has been overestimated and is in fact their gross income. As regards the imputation of income in kind, the position is unclear- although actual consumption in kind does seem to have been noted under daily expenditure and consumption in the questionnaire:38 "If some of the households interviewed grow their own vegetables, fruit, etc., or rear their own poultry for food, the amounts of these items consumed should also be entered under expenditure on food, with their own estimate of quantity and value, and the word 'free' entered in the first column in the record book" (Course of Lectures, p. 11). It is not known whether retail or farm-gate prices were used to value nonmarket consumption, or indeed whether such "free" consumption was included under income. There does not appear to be any provision for such imputation in the income section of the questionnaire, and no instructions for field investigators were issued to this effect. It seems clear Ihat problems were encountered in valuing the income of subsistence farmers. In HBS Report average consumption patterns are shown according to four monthly income classes (M$1-M$150, M$151- M$300, M$301-M$500, and M$501-M$1,000) and a separate category, 37. When gross income information was recorded for a respondent, his income tax was treated as an item of expenditure-again, so that the income-expenditure check would remain valid. 38. This is confirmed in a press statement on HBS issued by the Department of Statistics, probably in 1957: "We are also interested in actual consumption, so we shall try to value and take down the amount of food the family consumes from their own garden and even gifts. besides just the expenditure." 50 INEQUALITY AND POVERTY IN MALAYSIA "income not definite," for which, presumably, there were difficulties in assigning an income ligure. It is known from HBS Report (pp. 6-8) that this group consisted entirely of Malay households in rural areas. Furthermore, they are likely to have been subsistence farmers mainly, as Fell and others (1959, p. 3) state: "The income group 'income not clearly defined' . . . covers, in the main, padi planters." Apart from the inclusions and omissions documented above, it is not known how some of the other categories of income were estimated. According to HBS Report (p. 3), there was some "subjective" estimation: The tables on income distribution of households are not as accurate and reliable as those for consumption and expenditure, and contain some subjective estimation where income was not entered on the questionnaire or was believed to be incorrect. The estimation of income had then to be deducedfrom the expenditure of the household or by reference to the type of occupation of the members of the household (italics added). This statement reinforces my earlier point concerning upward adjustment of income of households whose expenditure exceeded reported income. The evidence I have marshalled allows the building up of a partial picture of the income concept implicit in HBS. Although a fully comprehensive definition is likely to remain elusive, the following major limitations can be asserted with some confidence: 1. Income was sometimes recorded gross of tax and sometimes net of tax. It is not known how many households, and at what income levels, fall in each group. 2. No imputation was made for owner-occupied and free housing. This is fairly serious because the number of owner-occupied and free dwellings in Malaysia is large. According to HBS, 81 percent of the households lived in such dwellings; for Malay, Chinese, and Indian households separately the figures were 95, 56, and 75 percent, respectively (Fell and others, 1959, p. 9). 3. The income estimate for households with expenditure in excess of reported income may have been adjusted upward to the expenditure level. 4. The income of self-employed households was probably their gross income. 5. Own consumption of produce was entered under households' daily expenditure, and insofar as point 3 above was valid, a part of it may have been included under income. But the special category "income not definite" suggests this was not always practicable. It is not known how many households belong to this category. COMPARISONS OF PES WITH OTHER SURVEYS 51 The income distribution tables published in HBS Report contain twenty- eight income classes, with an upper limit on incomes of M$1,000 per month. The percentages in each income class were read from graphs of cumulative income distributions, shown in HBS Report (app. 111): "The numbers in each income group as derived from the sample were plotted on a graph and a free-hand line drawn to remove the irregularities and so give a smooth curve. This is believed to give a better distribution of income than that from the uncorrected data" (HBs Report, p. 3). Comparison between PES and HBS The lack of a proper definition of HBS income and the ambiguities persisting on imputation and coverage are of little consequence for the basic purpose of' HBS (expenditure weights for price indices), but they seriously undermine its credibility as a data source on income distribution. Furthermore, the major shortcomings of HBS income cannot be put right by adjustment or appeal to other sources. It would be necessary to go back to the completed HBs questionnaire for each household to construct a distribution which used a uniform definition of income. This is now impossible, however, because the records have long since been destroyed. It follows that HBS should not be used as a base from which to make inequality comparisons over time. Those authors who concluded that inequality worsened between 1957 and 1970 after comparing PES with HBS have: not probed sufficiently into problems leading to noncomparability. Their conclusion cannot be established because of technical differences between the surveys (for which adjustment is impossible) which bias HBS inequality downvvard in relation to PES inequality.39 I have isolated six such differences:40 1. The noncoverage in HBS of households with incomes of more than M$1,000permonth, thatis, theveryrich (HBsReport, p. 39).4" Infact, upper income groups were generally undersampled, and, as already 39. Some of these differences (points I to 3 below in the text) were already documented in Anand (1973). Lee (1975) and Snodgrass (1975) have taken account of these differences butdo not lecognize all the complexities. Given the importance of the subject, it is necessary to present a fuller account of the problems leading to noncomparability and to repeat my earlier admonishments. 40. Forsomedifferences between the surveys it is not possible to know thedirection of bias. For example, the difference in treatment of income tax in the two surveys could bias HBS inequality in either direction. If households in HBS reported net-of-tax income, there would be a downward bias of fIBS inequality in relation to PES, since PES incomes are gross of tax, and the tax system is progressive (see appendix E, corollary 2). 41. If corollary I in appendix E is applied to the top income class as a single group, it shows that HBS inequality is underestimated by underestimating such income. 52 INEQUALITY AND POVERTY IN MALAYSIA indicated, coverage of these groups among the Malays, Chinese, and Indians was unsatisfactory. 2. The noncoverage in HBS of households belonging to "other" communities. Exclusion of this very heterogeneous and unequal group biases HBS inequality downward.42 3. The nonexistence in HBS of a separate zero-income category. The lowest mean income in HBS begins at M$12.50 per month, whereas in PES 1.4 percent of the households belong to the zero-income category43 (see table 3-2). It is shown in lemma 2 in appendix E that the omission of households with zero income from a distribution always results in an underestimation of inequality. 4. The income of self-employed households in HBS may have been overestimated if it refers to their gross income. On the assumption that the self-employed are disproportionately represented among the poor (for example, subsistence farmers and urban households in the informal sector) this overestimate would tend to bias inequality downward. 5. The nonimputation of owner-occupied and free housing in HBS. It is reasonable to suppose the income elasticity of demand for housing to be greater than unity, so that omission of this component of income is likely to understate inequality (see corollary 2 in appendix E). 6. The upward adjustment of income in HBS for households whose expenditure exceeded income probably biases inequality downward. The reason is that the upward adjustment of income occurs mainly at the lower end of the distribution since expenditure is more likely to exceed income for lower income groups (see lemma I in appendix E).44 This list is not intended to be exhaustive. There might well be other differences that impart a downward bias to HBS inequality in relation to PES inequality. 42. This is certainly the case for an additively decomposable index (see chapters 3 and 6). The relative size of the "others" group was much larger in 1957 than in 1970, by which time many of the small subcommunities once classified as "others" (such as Indian Muslims, Ceylonese, and Thai Muslims) identified themselves with one of the major racial groups. Thus exclusion of the "others" is likely to have biased HBS inequality downward much more than it would have done PES inequality. 43. As noted in chapter 3, zero-income households could have supported their consurnp- tion expenditure in two ways which do not count as PES income by running down past savings or liquidating capital assets, and by borrowing. "Distress sales" and loans are a well-known source of finance for low-income households in developing countries. 44. If income were widely "deduced from the expenditure of the household," the ilBs income distribution would tend toward the (consumption) expenditure distribution, which is generally considered to be more equal. COMPARISONS OF PES WITH OTHER SURVEYS 53 It should be obvious from the evidence presented that HBS is not comparable with PI s in income concept and coverage. The downward biases rioted in HBS inequality could easily account for the entire difference in measured inequality between the two surveys. Furthermore, it is not possible to quantify the magnitude of these biases for points 4 to 6 above. Any attempt to adjust for points I to 3 (assuming this could be done satisfactorily) will, therefore, not be enough to prevent intertemporal comparisons of ineqluality from being seriously misleading. Accordingly, little significance can attach to conclusions about inequality changes between 1957 and 1970 based on these surveys. This analysis has served to highlight the danger of making intertemporal inequality comparisons without adequate research into the comparability of the underlying data. Such comparisons have become increasingly frequent in the literature of development economics, but there have been few attempts to verify comparability or to bring the data together in a common framework. Judgments about intertemporal inequality based on a superficial examination of published data are likely to be highly suspect, as shown by this analysis of two seemingly comparable Malaysian surveys. The same cautionary note needs to be struck when making judgments about international inequality, where surveys are not conducted by the same statistical department and there is even less reason to suppose uniformity of concepts and definitions. Appendix: PES 1970 Instructions to Field Interviewers Individual and Household Income [Extracted from Department of Statistics, 1970b, pp. 31-37.] The questions in this section should be asked as they are written- elaborate only after you have asked the question in the stated manner. Do not overlook even a single question. Quesuions I & 2: Now, I would like to ask a few more questions about persons who have been doing some job or other for pay, profit or family gains. Please tell me their namesand their relationship to the Head of Household? (ENTERTHIS IN COLUIMNS I AND 2) (a) Is there anyone else, like your wife, son or daughter, who has helped in the farm or field? (IF*YES'-WRITETHEIR NAMES IN COLUMN 1.) (b) Is there anyone else, what about (specify name)? (ENTER NAME IN COLUMN I.) (REPEAT THIS FOR OTHERS WHO SEEM ELIGIBLE) 54 INEQUALITY AND POVERTY IN MALAYSIA Here, first of all we want to determine the persons who are or were doing some work or other. Obviously these are the persons who will be mainly receiving some form of regular income. The names of these people should be entered in Column (1). Remember that the names mentioned here should be generally found in the household (i.e., the name must have been included with the household in Section (B), Column (9)). Very occasionally you may find someone who is normally a member of the household but did not sleep in the household on Census Night-and hence has been excluded from this household. In this case, include the person in this Section, even though he has not been included in the Household Composition. Give also a footnote to this effect. It is also possible that there were others who worked as family workers (i.e., helping in family enterprise or farm)-if there are any, also enter their names in Column (1). You should then flip back to Section (B) on Household Composition, and see if there are any eligible persons, whose names have not been mentioned. Then you should ask " , what about name)? Did he receive any income? or has he been working somewhere?" At the end of the question you are now reasonably sure you have not missed out any earning member in the household. Question 3: Let us look at those working, one by one. What type of job is (name) working. I would like to know the specific nature of his job, and where he is working. (REPEAT FOR ALL OTHERS) We want the type of occupation and the institution or firm in which each of the persons are working. It seems easy to get this information--but experience has shown that this is where most errors occur. This is either because the respondent does not know the precise designation of the job, or because the interviewer has not asked for sufficiently clear details. T o be sure you get the required details, we will give a book of the Occupation Codes. This is an alphabetical index of the occupations and will be helpful to you. It is important that you obtain the exact type of work done by the respondent, such as lorry driver, farm labourer, dress-maker, etc. Some of the respondents would say he is a helper or driver or mechanic, etc. In these cases you should find out what kind of a helper or driver or mechanic. For example, a driver may be a lorry driver, taxi driver, or even a chauffeur. It is COMPARISONS OF PES WITH OTHER SURVEYS 55 necessaiy for purposes of classification to know the exact nature of the occupation. Many of the occupations are found in an establishment. For example, in a school we have teachers, clerks, typists and watchmen. On the other hand, a given occupation may be found in several kinds of establishments, as in the case of, say, a typist. We want the exact nature of the job and the type of establishment. In obtaining the responses for this question, a few points must be borne in mind: a) Use the Occupation Classification wisely. You should make good use of the alphabetical index. For example if the respondent says he is a "driver" by occupation then refer to the occupation "driver" in the Occupational Classification Book. It shows 12 types of drivers; ask him what type does he fit into. Or again he may say that he is an engineer. If you refer to the index provided you will see that there are several types of engineers, and different levels of engineers: engineer- ing fitter, engine fitter, engine greaser, and engine oiler. There is a good deal of clifference between an engineer and an engine oiler. The basic level of the occupant can be sensed from the way he lives. Try and get at the correct description of the person's occupation using the occupation index as an aid. b) Respondents may reply in Chinese or Malay or Tamil. It is very difficult to translate some of the words into appropriate English equivalents. Look through the occupation index provided and try to get to know some of the fairly common occupations and their equivalents in the languages in which you are conversant. In some cases you may even write on the questionnaire the actual words used by the respoiidents, for example, words like "Mandore," 'Serang," "Amah," or "Hung Kong" can be translated in the office at a later stage. Sorne examples of occupation entries are listed below as a guide. This list is not exhaustive but is intended to give an idea of the description acceptable and the degree of details required to enable the proper classification of occupations. Attendant: Should state whether lift attendant, vehicle attendant, car park attendant, hospital attendant, woodworking machine attendant, or laboratory attendant, etc. Clerk: Should state whether accounts clerk, correspondence clerk, store clerk, solicitors' clerk, legal clerk, tally clerk, or general office clerk, etc. 56 INEQUALITY AND POVERTY IN MALAYSIA Clerical Assistant: Should also specify the type and clerical duties performed. Conductor: Should state whether bus conductor, railway train guard, or band conductor, etc. Driver: Should state whether lorry driver, bus driver, mail van driver, ambulance driver, taxi driver, motorcar driver, tractor driver, railway engine driver, crane driver, or bulldozer driver, etc. Electrician: Should state whether electrical fitter or wireman. If electrical fitter, state the type of machinery or equipment handled, e.g., motors and generators, transformer, switchgear and control apparatus, electrical elevators, etc. If electrical wireman, state whether building electrician, vehicle electrician, stage studio electrician, or an electrical repairman for domestic electrical equipment, etc. Engineer: Should state whether sanitary engineer, building construction or general construction civil engineer, general mechanical engineer, industrial machinery and tools engineer, marine and ship construction engineer, general chemical engineer, petroleum chemical engineer, mining engineer, general electrical or electronic engineer, power gener- ation engineer, telecommunication engineer, or industrial efficiency engineer. Factory Worker: Should state the kind of factory and the job performed, e.g., knitting machine operator, fruit press operator, lathe operator, rubber packer, rubber grader, machine labeller, etc. Farmer: Should state whether field-crop farmer (tobacco), planta- tion-crop farmer (coconut), fruit tree farmer, livestock farmer, dairy farmer, poultry farmer, nursery operator, or market garden farmer, etc. Farm Worker: Should state the type of farm as for farmer. Fisherman: Should state whether deep sea fisherman, kelong fisherman, coastal fisherman, fishpond or prawn-pond operator/worker, shellfish gatherer, etc. Fitter: Should state whether textile machinery fitter, internal combustion engine fitter, electrical motor and generator fitter, electronic, radio, and radar equipment fitter, gas pipe fitter, telephone fitter, or general pipe fitter, or even dress fitter, etc. Furnaceman: Should state whether ore smelting furnaceman, steel con- verting and refining furnaceman, metal melting or reheating furnaceman, copola furnaceman, electric arc furnaceman, or a rever- beratory furnaceman, etc. Inspector: Should state whether public health inspector, road transport inspector, railway service inspector, bus ticket inspector, inspector of police, motor vehicle inspector, etc. COMPARISONS OF PES WITH OTHER SURVEYS 57 Labourer: Should state whether dock labourer, railway and road vehicle loader, aircraft loader, warehouse goods handler, building construction labourer, demolition work labourer, or manufacturing labourer, etc. Machine Operator: If office machine operator, state whether accounting machine operator, comptometer operator, teleprinter operator. If metal- working machine operator, state whether lathe operator, metal-weaving machine operator, honing-machine operator. If woodworking machine operator, state whether precision sawyer, wood lathe setter-operator, wood-shaping and planing machine operator. If other industrial ma- chine operator, state whether bottle-washing machine operator, knitling-machine operator, etc. Manager: Should state whether production manager, sales manager, administration manager, finance manager, personnel manager, advertis- ing manager, farm manager, wholesale trade manager, retail trade manager, or manager of hotel, restaurant, etc. Mechanic: Should state whether motor-car mechanic, bus mechanic, radio and television mechanic, aircraft engine mechanic, office machines mechanic, plant maintenance mechanic, textile machinery mechanic, or metal-working machine tool mechanic, etc. Printer: Should state whether hand compositor, linotype operator, typecasting machine operator, cylinder pressman, pattern pressman, rotary pressman, stereotyper or electrotyper, etc. Repairman: Should state whether bicycle repairman, motor repairman, clock and watch repairman, electrical domestic appliance repairman, camera repairman, or shoe repairman, etc. Salesmnan: Should state whether wholesale trade salesman, retail trade salesman, technical salesman, manufacturer's agent, manufacturer's sales representative, insurance salesman, or advertising salesman, real estate agent, market stall-holder. Sewer/Seamstress: Should state whether hand sewer or machine operator anel the type of article the sewer specialises in, e.g., hand sewer, garments; hand sewer, house furnishings; etc. Supervisor: Should state whether general clerical supervisor, accounts section supervisor, typing pool supervisor, sales supervisor (wholesale or retail trade), road transport supervisor, supervisor of metal processing industry, supervisor of construction work, or supervisor of food processing industry, field supervisor, etc. Technician: Should state whether surveyor's technician, radio and tele- vision technician, air-conditioning engineering technician, chemical engineering teclnician, or mining technician, etc. ("Technician" should not be used to describe a Mechanic or a Repairman.) Welder: Should state whether gas welder or electric arc welder. 58 INEQUALITY AND POVERTY IN MALAYSIA Working Proprietor: Should state whether working proprietor of whole- sale trade or retail trade, working proprietor of coffeeshop, bar, canteen, nightclub, boardinghouse, etc. Question 4: What is (name) working as? Is he an Employer, Employee, Own Account Worker, or Family Worker? Please be careful; a Family Worker should be one who has contributed to family gain and not one doing housework. (ENTER IN COLUMN 4.) Here we want to know the earning member's Employment Status--and he may be an Employer, Employee, Own Account Worker, or Unpaid Family Worker. This is a way of describing the person's position in the business, factory, farm, etc. and his relationship to other persons in that organization. An Employer is a person who employs one or more persons and pays them to work (in cash and/or kind). Such workers (employees) may assist in operating his business, farm or enterprise. An Employee is a person who works for somebody else and is paid a fixed wage or salary or paid in kind. Included as Employees would be Government Servants, Managing Directors and Directors of Companies. and others in a similar position. Even though they can hire and fire people, they are still employees of the Government or the Company. Own Account Worker is a person who works for himself and does not employ anybody else. However, he may have a partner and still be counted as self-employed. The important point is that no labour is hired. Family workers are not to be regarded as hired labour. An Unpaid Family Worker is a person who works without pay in a business run by another member of the family. Question 5: We now come to an important part ofthe interview. I would like to know the average monthly income of each of the persons you have mentioned. (Name), what do you think is his average monthly income? Please think carefully. (REPEAT FOR OTHERS. ENTER IN COLUMN 5.) This is an exceedingly difficult question, and needs great tact in asking. We want to know what the respondent thinks is the monthly income of each of the persons living in the household and earning. While we want to know what he thinks is the income, you should prompt certain questions that he should bring into his calculation. Emphasise that you want the average monthly income. COMPARISONS OF PES WITH OTHER SURVEYS 59 Generally, this question needs to be carefully thought out by the respondent; hence be patient and don't rush him to give you a response. Question 6: Let us see, (name) is a (empl. status). What do you think would be his monthly income from: (a) WVages and salaries and other receipts. (b) From sale of produce. (Note: this should be an average for the month.) (c) From jointly owned businesses or farm. (d) 'From rent, investments, etc. (e) Pension, remittances and cash allowances, royalties, fees and other receipts. (f) Other periodic cash receipts-e.g., alimonies, scholarships, etc. (g) Monetary value for food, clothing, housing, etc. (h) Monetary value ($) of goods used for own consumption. (i) Monetary value ($) of goods for consumption received from other sources. (j) Value of owner-occupied house. (k) Other concessions. This question is primarily aimed at obtaining data on income by going through each and every source of income. By so doing we hope that the respondent will be able to think about each of the categories carefully, and say if he had any income from that source.... You should note that not all categories mentioned from (a) to (k) are applicable to all members. In fact only a few of them would be. You should apply your common sense to know what categories to ask and what not to ask. For example, if the respondent is a clerk-his monthly income would be mainly in wages, and he may have other sources of revenue from rent, investment, etc. Clearly, sale of produce generally will not be applicable to him though it is riossible. A farmer on the other hand will receive the bulk of his income from sale of own produce. You should know whom to ask what question. Report the details in the Second Block at the bottom of the page in Section (E). i'nterviewer's Assessment of Household [Extracted from Department of Statistics, 1970b, pp. 38-39.1 When all the five Sections (A to E) have been completed, you should now be in a position to assess the quality of the overall interview with the 60 INEQUALITY AND POVERTY IN MALAYSIA respondent. You should always attempt to answer the questions in this section as accurately as possible because we want to know the incon- sistencies in the respondents' answers, and also the reliability of the answers. Your assessment of the interview should refer to the overall interview, and not to a particular respondent. In case a particular resporndent's answers were particularly bad, indicate this under comments. Question 1-Reliability of Respondents' Answers: The problem here is how to assess the respondents' answers as reliable or otherwise. If the respondent answers your questions without much hesitation and in a relaxed manner, there is reason to believe the accuracy of the data given by him. When the respondent gives his answers, you should be able to tell whether his answers are reliable or not. Somel imes, you can sense the respondent's reluctance, especially when he tries to evade giving you a direct answer. If he understood fully the purpose of the survey and right from the start he was willing to co-operate, the chances are his answers would be reliable. It is also possible that right from the start, the respondent refuses to co- operate and he answers your questions with a "yes" or "no" and with indifference. It is here that you have to be careful for if he continues to respond in this manner, it is possible that his answers may be unreliable. This does not mean that a smiling and co-operative respondent will not give unreliable data. This is also possible. You have to look at the type of responses to decide the degree of reliability of the data you obtain for the household. Question 2-Respondent's Understanding of Questions: If the respondent is able to answer your questions directly and exactly, he must have comprehended the questions well. But if he needs simple explanations to the questions and ultimately he gives a correct answer, then he clearly has understood your questions only fairly well. Finally, if even after you have explained and re-worded your questions in greater detail, the respondent still gives some answers not pertaining to the questions, then it is obvious that his understanding is extremely bad. Another factor which can broadly indicate whether the respondent understands the question is "Time Factor." How long does he take to answer a question? The longer he takes to answer, the greater the likelihood he finds difficulty in following the question. This may not be so in all cases. He may take a long time if he wants to think carefully before answering. Generally speaking, understanding will be a function of the fluency in a language. Lack of understanding of a question may be because you are not COMPARISONS OF PE-S WITH OTHER SURVEYS 61 conducting the interview in the language the respondent understands best. If you get even the slightest doubt of language problem, excuse yourself politely aIid report this to your regional supervisor as soon as possible. Question 3-Respondent's Confidence in You: Confidence is the most important factor in helping you to obtain correct and accurate information, and also ensures a smooth interview. Therefore, it is your duty to establish rapport with the respondents. Once the respondent trusts the purpose of the Survey and he feels very free to answer your questions, then you can rest assured that he will be very co-operative in replying to your questions. Build this confidence through the way you perform your interview. A few useful hints would be: (a) Be sure that you are dressed properly and appear clean and smart (b) Be polite and pay respects to the elders if any, at the place of interview (c) If you use a motor-scooter or other vehicles, be sure that it is not jazzy and offensive to simple people (d) Do not engage in controversies and never discuss religion or politics even as a side line (e' Offer all common courtesies-e.g., if the respondent is a smoker, offer him a. cigarette if you have one. Don't ignore the respondent. (f) Be attentive to what he says, and be patient, and put on a smile as often as possible. Remember that if the respondent lacks the assurance and the confidence in thte contents of'the questions and the manner you ask the questions, he is bound to give you some problems in obtaining reliable information. Question 4-Laniguage: 11' right from the start you face language problems, then do not interview the household at all. You are to inform your regional supervisor as soon as possible. Try to find out the dialect of the respondent, for this will enable the regional supervisor to send an appropriate interviewer to do the job. Should the problem crop up during the interview, then you are to mark down the individual questions which give you trouble. This will enable the regional supervisor to check these questions and see whether a second visit to the household is necessary or not. If, however, you really get stuck in the middle of an interview and you are unable to proceed any further, excuse yourself politely and report to your regional supervisor. He will make some alternative arrangements. 62 INEQUALITY AND POVERTY IN MALAYSIA Question S-Comments: This column is meant for you to write all the problems faced by you in the interview and other field problems. Try to express your difficulties and if possible suggest a few solutions. Little comments can help a lot in planning the future Surveys, therefore, do not hesitate to contribute your ideas and solutions pertaining to field problems. Indicate also any customs or traditions or fears expressed by the persons. This will help future planning of Surveys. 3 Inequality in Levels of Living IN THE PREVIOUS CHAPTER, which presented a broad picture of inequality in Malaysia as revealed by the PES, the household income distribution was used to enable comparison with other surveys. In this chapter, to analyze inequality in levels of living or economic welfare among the population, the per capita household income distribution is considered more appropriate. Various inequality indices are estimated for this distribution, and there is a discussion on the decomposition of inequality, which is applied to estimate the contribution of interracial, interregional, and rural-urban inequalities to overall inequality. The final section discusses the implications of the racial decompositions for the two prongs of the New Economic Policy and shows how the redress of poverty rule (from appendix E) is also "efficient" for redressing inequality. PES Income as a Measure of Economic Welfare There appear to be three shortcomings in using PEs income as a proxy for levels of living or economic welfare. First, welfare levels at a point in time are likely to be better indicated by current consumption than by current income as defined in most surveys including PES. (Unfortunately, data on the consumption distribution are not available in sufficient detail in PES.) The problem is highlighted in PES by the existence of a small number (1.4 percent) of households with zero current income. Obviously. the consump- tion of such households cannot be taken as zero, and presumably their consumption expenditure was financed from sources that do not count as PES income. The two most important of these are likely to have been borrowing and dissaving,' which are omitted from the eleven income 1 The consumption expenditure could also have been financed by other unrecorded trarsfers or by asset decumulation. Surveys which use the net change in assets or balance-sheet approach to define income are likely to catch this aspect. 63 64 INEQUALITY AND POVERTY IN MALAYSIA categories listed in PES (see chapter 2, "Definition of PES Income"). Borrowing and dissaving are fairly common phenomenons among low- income households in developing countries, but surveys do not usually include these sources in their definition of income. PES current income is also an imperfect proxy if one is interested in a lifetime measure of welfare levels. This requires data on the profile of income over time, which irons out life-cycle variations associated with age and experience2 as well as purely random fluctuations around the iage profile. Again, current consumption might be a better proxy than current income, if a version of the "permanent income" hypothesis is accepted. The second problem associated with using PES income as a surrogate for economic welfare concerns its nonadjustment for tax incidence and public expenditure benefits (including transfer payments). 3 If it is to be a measure of welfare, PES income should ideally be adjusted for direct and indirect tax incidence and public goods and services, which have a differential impact according to income class. Unfortunately, it has not been possible to make such corrections here, and unadjusted PES income is used as a proxy for welfare levels. A potential third problem is that owing to geographical variation in prices, especially between rural and urban areas, the PES (money) income distribution could turn out to be somewhat different from the real income distribution in the country. Unfortunately, there are no regional price surveys in Malaysia to enable estimates of real income, but it seems that regional price differences are in fact fairly small for the average consump- tion bundle.4 Hence the error introduced by neglecting such price differences may not be particularly significant. 2. Income differences arising from age and experience are an important part of observed inequality among urban employees in Malaysia (see chapter 7). 3. There appears to be some adjustment on the subsidy side: the PES field interviewers were instructed to "include income from miscellaneous sources-like government aid, scholarships to children, free books, subsidized items like fertilizers, etc." (Department of Statistics, 1970b, p. 31; also quoted in chapter 2). 4. This was the opinion of senior officials in the Malaysian Ministry of Agricullure (including Mr. Selvadurai. the senior economist, and Mr. Haridas). For example, the difference in rice prices between states in 1970 was at most 5 percent because the price of paddy was nationally controlled. The prices of other food items apparently varied by about 10-- 15 percent, but there are likely to have been compensating variations across regions. On the one hand, clothing and imported goods (such as sugar and kerosene) are likely to have been more expensive in the rural hinterland because of transport costs (and also because goods are often bought on credit in rural areas) On the other hand, fruits and vegetables (which are locally produced) were probably cheaper in rural than in urban areas. For the average consumption bundle, therefore, the difference in rural-urban prices may not have been very large. This question is briefly discussed again in chapter 4, in connection with the definition of a poverty line. INEQUALITY IN LEVELS OF LIVING 65 The Population Unit and Appropriate Income Concept Income distributions have typically been defined over different types of population unit.5 The choice of unit obviously depends on the purpose at hand, which in this chapter is to analyze inequality in levels of living. The distribution of households by household income is most widely used for this purpose, but for two reasons this distribution does not provide a good indication of differences in levels of living in the population. First, the distribution is defined with respect to the household, whereas the primary interest here is in differences among individuals. Second, even when the household is the unit of concern, its living standard is not properly measured by household income. Household income needs to be adjusted to take account of variations in household size and composition, and economies of scale in consumption. For example, a large household with a certain household income is not as well-off as a small household with the same total income.6 Even if household size is the same, a unit of income can produce different amounts of household utility or well-being, depending on the age-sex composition of its members.7 Even though the primary interest is in inequality among individuals, it is still ntecessary to consider the household income of households to which the individuals belong. I This is because the household performs a redistribu- tive function among its members, with the income accruing to individual recipients getting pooled. Indeed, in some cases income accrues directly to the household as a unit and not to individual members within it. In family 5. Several different types of unit have been distinguished for the purpose of measuring inequality: households, individuals, income recipients, economically active population, and so on. To avoid confusion, throughout this book I have tried to describe distributions explicitly in terms of both the population unit and the income concept. For example, I refer to the distribution of households by per capita household income. This terminology arises naturally from the underlying frequency distribution, in which households are "distnbuted" across class intervals of per capita household income. The alternative terminology based on "distributing" per capita household income among households is less elegant: first, it is not per capita household income that is distributed among households, but a total (which, in this case, is not the total household income in the economy); second, mentioning this total does not obviate the need to mention the income concept according to which the total is distributed 6, An exception to this occurs if household income is a "public good." 7 The welfare level of a household depends both on the total income of the household and on its size and composition. Among those who have tried to derive equivalence scales for households by formalizing this dependence is Muellbauer (1976). 8. The income actually received by individuals does not provide much information about differences in levels of living. About 70 percent of them get zero income, including many well- off individuals belonging to rich households. 66 INEQUALITY AND POVERTY IN MALAYSIA farms and enterprises, for example, the income is jointly received by the household, and family workers are typically unpaid. The income from other jointly owned physical or financial assets can also be attributed only to the household collectively. For these reasons, then, the household must be regarded as the basic income-sharing unit. The welfare level of an individual depends on the size of his or her share of household income, but information about sharing among members is not easily obtainable. I make the assumption that it is shared equally, although its distribution undoubtedly depends on characteristics of individual members such as earner status, power, age, sex, weight, and so on.9 Hence the procedure is adopted of ranking individuals according to their per capita household income.'° The welfare level of a household is also measured by its per capita household income. " No allowance is thus made for the age and sex composition of its members, nor for economies of scale in household consumption. Some correction should have been made for household composition, but with present data this has proved impossible. An estimation of equivalence scales requires detailed household expenditure data, which are not yet available. 12 Neglect of composition effects 9 Little is known about the intrafamily income distribution. Some sociological literature from village India suggests considerable inequality in the sharing of household food. This is explained by the differing needs of working and nonworking members of the household, with the working members consuming more to sustain higher levels of physical activity 10 This income concept defined over the population of individuals certainly gives a better. idea of inequality in levels of living than does the distribution of income recipients by personal income. For certain purposes, such as an analysis of education and earnings (chapter 7), the personal income distribution would also be of interest. It is shown in appendix E, lemma 3, that the distribution of individuals according to per capita household income is more equal (in the Lorenz sense) than their distribution according to personal income. II.. If household income is strictly a "public good," however, then for welfare measurement individuals and households should be ranked according to household income, not per capita household income. 12. Such scales would allow estimates of the household income per equivalent adult For inequality measurement, however, there is still the problem of assigning this income concept to the appropriate population unit: households, individuals, or equivalent adults (including fractional ones). In other words, which of the three distributions should be chosen to measure inequality: the distribution of households by household income per equivalent adult, the distribution of individuals by household income per equivalent adult, or the distribution of equivalent adults by household income per equivalent adult? The present methodology for estimating equivalence scales is based on income-scaling factors that keep household utility constant for variations in household size and composition. For consistency with the derivation of equivalence scales, therefore, the distribution of households by household income per equivalent adult may be regarded more appropriate for the measurement of inequality. INEQUALITY IN LEVELS OF LIVING 67 probably leads to a relative underestimation of welfare leveis of households with a high proportion of children and female members.13 In Malaysia, these are not necessarily the large households, owing to the prevalence of the joint family system and other institutions.14 Three separate distributions are considered: 1. The distribution of households according to household income 2. The distribution of households according to per capita household income 3. The distribution of individuals according to per capita household income. The first distribution has already been discussed in the previous chapter. It is briefly included again to facilitate comparison with the other distributions: table 3-1 shows the distribution of households by household income class and racial group. (For a disaggregation of this distribution by urban-rural location, see tables 3-12 and 3-13 at the end of the chapter.) The second and third distributions will generally differ from the first and from each other, except when household size is constant in the economy. Their derivation requires full information on the joint distribution of households by household income and size (see appendix F). The Joint Distribution of Households by Household Income and Size The joint distribution of households by household income and size is presented in table 3-2. Each cell refers to a particular household income and size class and shows the absolute frequency of households in it. The rows of the table show Ihe household size distribution within each income class. The columns of the table show the income distribution within each household size class, together with the average household income, Gini coefficient, average number of income recipients, and so on. 13. This assumes that children and women count for less than a "standard adult" or, in other words, their "equivalent adult" factors are smaller than unity But it is not clear how different from unity these might be for children in certain age groups. While children eat less food on average than adults (those. at any rate, who are not yet teenagers), they incur larger expenditures on other items such as education (including books), health, toys, and so on. 14. The welfare level of members from large households is. however, underestimated if economies of scale in consumption are neglected. But I have not seen a convincing treatment that allows for scale economies In fact, in the context of existing models that have been used to estimate equivalence scales, it seemns impossible to distinguish scale effects from composition effects. 68 INEQUALITY AND POVERTY IN MALAYSIA Table 3-1. Distribution of Households by Household Income Class and Racial Group (number of households) Household All races income class including (MS per month) Malay Chinese Indian 'other" No incomc 190 110 32 341 1-39 1,145 104 56 1,336 40-49 617 46 34 711 50-79 2,316 270 141 2,751 80-99 1,525 245 122 1,898 100-129 1,834 537 333 2,712 130-149 996 373 218 1,596 150-179 1,204 661 379 2,251 180-199 576 409 217 1,204 200-279 1,478 1,564 573 3,625 280-299 238 267 74 581 300-399 744 1,163 264 2,175 400-479 317 555 120 999 480-499 54 97 17 168 500-599 201 398 88 693 600-679 123 228 55 412 680-699 17 43 9 69 700-749 46 92 20 160 750-799 27 76 19 122 800-899 47 169 22 246 900-979 32 91 16 142 980-999 7 18 3 29 1,000-1,249 59 180 43 295 1,250-1,499 29 97 27 163 1,500-1,749 12 67 17 100 1,750-1,999 11 29 3 44 2,000-2,499 8 49 13 83 2,500-2,999 3 25 7 44 3,000-3,499 2 15 3 26 3,500-3,999 2 8 4 19 4,000-4,999 1 3 5 11 5,000+ 2 14 2 19 All income classes 13,863 8,003 2,936 25,025a Note: See tables 3-12 and 3-13 for a disaggregation by urban-rural location. a. Includes two households for which the racial group was not available. The row showing average household income for different household sizes indicates a steady rise in average income with household size, except for a slight fall for the nine-member group. The row showing the Gini coefficient for each household size indicates inequality to be highest among Table 3-2. Joint Distribution of Households by Household Income and Size (number of households) Household Household size All income class household (MI per month) 1 2 3 4 5 6 7 8 9 10+ sizes No income 29 54 53 46 57 39 24 13 8 18 341 1-39 495 352 163 127 82 53 32 18 8 6 1,336 40-49 124 148 156 102 87 43 26 11 11 3 711 50-79 354 407 489 462 374 279 181 102 52 51 2,751 80-99 175 217 288 301 261 244 195 96 64 57 1,898 100-129 269 260 360 393 397 368 257 189 113 106 2,712 130-149 128 121 193 237 228 232 180 119 78 80 1,596 150-179 187 168 254 313 296 290 249 2i7 144 133 2,251 180-199 71 82 119 151 182 151 137 137 77 97 1,204 200-279 173 233 340 440 463 491 406 385 287 407 3,625 280-299 22 35 45 56 59 81 79 68 59 77 581 300-399 92 112 176 230 240 250 270 255 185 356 2,175 400-479 38 47 65 119 116 128 116 106 85 179 999 480-499 6 12 10 18 13 19 23 14 10 43 168 500-599 24 33 57 64 75 75 82 69 59 155 693 600-679 13 25 34 46 46 38 39 35 30 106 412 680-699 3 4 9 9 7 7 5 3 3 19 69 700-749 2 7 12 15 18 18 14 20 14 40 160 750-799 2 6 6 10 16 18 17 9 10 28 122 800-899 3 12 19 26 21 36 25 22 20 62 246 900-979 4 6 9 15 20 22 12 7 10 37 142 980-999 0 1 2 4 1 3 2 2 5 9 29 (Table continues on the following page.) Table 3-2 (continued). Household HouLvehold size All income class household (MS per monih) 1 2 3 4 5 6 7 8 9 10+ sizes 1,000-1,249 7 20 25 32 28 47 28 30 17 61 295 1,250-1,499 5 5 12 17 22 18 16 20 12 36 163 1,500-1,749 1 6 4 12 12 15 12 7 7 24 100 1,750-1,999 1 1 1 5 2 12 3 4 1 14 44 2,000-2,499 3 9 10 8 8 11 8 4 4 18 83 2,500-2,999 0 2 2 3 9 4 5 5 2 12 44 3,000-3,499 0 3 5 3 0 4 I 5 0 5 26 3,500-3,999 0 2 0 4 2 0 3 2 1 5 19 4,000-4,999 0 1 0 0 2 1 1 2 0 4 11 5,000 + 0 0 2 0 0 3 2 3 0 9 19 All income classes 2,231 2,391 2,920 3,268 3,144 3,009 2,450 1,979 1,376 2,257 25,025 Average household income 137 181 207 231 243 282 289 326 315 496 264 Gini cocfficient 0.5111 0.5621 0.5178 0.4927 0.4837 0.4920 0.4634 0.4592 0.4031 0.4728 0.5129 Average number of income rec.p:e-nts' 0.99 1.24 i.37 1.46 1 54 1.64 1.73 1.90 2.02 2.76 1.63 Standard deviat;on of number of income recipientsa 0.11 0.50 0.64 0.75 0.82 0.91 0.99 1.07 1.18 2.09 1.10 Average partici- pation ratea b 0.99 0.62 0.46 0.37 0.31 0.27 0.25 0.24 0.22 0.23 0.40 Standard deviation of participation ratea 0.11 0.25 0.21 0.19 0.16 0.15 0.14 0.13 0.13 0.20 0.28 Average dependency ratio' c.d 1.00 1.72 2.47 3.18 3.83 4.44 5.03 5.38 5.82 5.97 3.76 Standard deviation of dependency ratioa,d 0 0.45 0.76 1.08 1.40 1.73 2.07 2.41 2.76 3.57 2.37 a. Per household. b. The participation rate for a household is the ratio of number of income recipients to household size. c. The dependency ratio for a household is the ratio of household size to number of income recipients. d. The average dependency ratio per household and the standard deviation of dependency ratio per household were calculated by excluding the zero- income households; otherwise they are not calculable. 72 INEQUALITY AND POVERTY IN MALAYSIA two-member households (Gini coefficient of 0.5621).15 Inequality seems to fall as one moves to larger-sized households, dropping sharply between the eight-member and the nine-member groups, and rising again for the largest size (ten or more members). For one-member households inequality is fairly high with a Gini coefficient of 0.51 1 1, which is only slightly lower than the overall Gini coefficient of 0.5129 for all households. Tables 3-3 to 3-5 show the average number of income recipients, the average participation rate, and the average dependency ratio"6 by household income class and racial group; table 3-2 shows these averages by household size class. The average number of income recipients rises uniformly with household size, from 0.99 for one-member households to 2.76 for households of ten or more. This rate of increase, however, does not prevent the average participation rate falling from 0.99 for one-member households to 0.23 for households with ten or more members. The obverse of this finding is the increase in average dependency ratio from 1.01 to 5.97 over the same range. Across income classes, the average number of income recipients per household rises from 1. 10 in the lowest positive class (M$1- MS39) to 8.37 in the uppermost (M$5,000 and over), although it fluctuates somewhat above the M$600 level. Over the same range, the average participation rate displays a nonmonotonic relationship, which might be described as approximately U-shaped.'" The participation rate first falls from 0.62 in the lowest class, then fluctuates between 0.36 and 0.44 at middle income levels, and finally rises back up to 0.52. The average dependency ratio follows a more or less inverse pattern to this. For all households, the average number of income recipients is 1.62, the average participation rate is 0.40, and the average dependency ratio is 3.76." 15. Ignoring composition effects, the Gini coelficient for any household size does give a valid indication of inequality in living standards within it. since there is no variation in the number of members. 16. The participation rate for a household is defined here as the ratio of number of inconie recipients to household size. The dependency ratio for a household is the reciprocal of the participation rate. This definition of participation rate is somewhat different from that used in labor economics generally. The latter refers to the ratio of actual to potential labor force participants in a household (or a population). My definition is of an "income" participation rate, where everyone is viewed as a potential income recipient (for example, through remittances or gilts) 17. These data make it possible to conduct analyses explaining the participation rate in terms of economic and demographic variables. This would be important for making projections about labor supply and unemployment. 18. The average dependency ratio is not equal to the reciprocal of the average partici- pation rate: in general, the average of a reciprocal is not equal to the reciprocal of the average INEQUALITY IN LEVELS OF LIVING 73 Table 3-3. Average Number of Income Recipients per Household by Household Income Class and Racial Group Household All races income class including Standard (MS per month) Malay Chinese Indian '.other' deviationa No income 0 0 0 0 0 1-39 1.11 1.02 1.02 1.10 0.33 40-49 1.17 1.04 1.09 1.16 0.39 50-79 1.24 1.07 1.15 1.22 0.50 80-99 1.33 1.18 1.17 1.30 0.59 100-129 1.41 1.21 1.17 1.34 0.60 130-149 1.49 1.31 1.34 1.43 0.65 150-179 1.52 1.35 1.47 1.46 0 68 180-199 1.63 1 54 1.69 1.62 0.80 200-279 1.70 1.70 1.82 1.72 0.85 280-299 1.72 2.01 2.16 1 91 0.95 300-399 1.81 2.07 2 24 2.00 1.04 400-479 2.00 2.33 2.21 2.20 1.26 480-499 2.32 2 23 2.29 2.26 1.20 500-599 2.02 2.61 2.02 2.36 1.36 600-679 2.42 2.61 2.07 2.47 1.51 680-699 2.00 2.35 2.22 2.25 1.18 700-749 2.07 2.79 2 65 2.58 1.38 750-799 2.41 2.82 1.95 2.59 1.47 800-899 2 51 2.92 2.77 2.79 1.77 900-979 2.41 3.01 2.06 2.75 1.74 980-999 2.00 3.11 2.67 2.72 1.31 1,000-1,249 2.49 2.95 2.63 2.76 1.68 1,250-1,499 3.10 3.13 3.11 3.03 1 71 1,500-1,749 4.00 3.31 4.29 3.52 2 16 1,750-1,999 3.64 3.59 3.33 3.52 1.69 2.000-2,499 3.13 2.94 3.85 2.86 2.28 2,500-2,999 3.67 3.44 4.71 3.30 2.73 3,000-3,499 3.00 2.93 3.33 2.65 1.41 3,500-3,999 2 00 4.75 2.50 3.53 2.59 4,000-4,999 1.00 2.67 6.60 4.18 4.00 5,000+ 2.00 10.50 2.50 8 37 13 43 All income classes 1.46 1.90 1.72 1.63 1.10 Nlote: See tables 3-12 and 3-13 for a disaggregation by urban-rural location. a. Standard deviation of number of income recipients per household for all races including "other." 74 INEQUALITY AND POVERTY IN MALAYSIA Table 3-4. Average Participation Rate per Household by Household Income Class and Racial Group Household All races income class including Standard (MS per month) Malay Chinese Indian "other" deviallona No income 0 0 0 0 0 1-39 0.61 0.79 0.76 0.62 0.33 40-49 0.44 0.80 0.71 0.48 0.30 50-79 0.38 0.71 0.57 0.42 0.28 80-99 0.34 0.59 0.45 0.38 0.27 100-129 0 33 0.50 0.44 0.38 0.27 130-149 0.33 0.44 0.42 0.37 0.26 150-179 0.34 0 39 0.41 0.37 0.26 180-199 0.35 0.40 0 36 0.37 0.26 200-279 0.35 0.38 0.38 0.36 0 25 280-299 0.34 0.40 0.37 0.37 0.25 300-399 0.35 0.39 0.41 0.38 0.25 400-479 0.36 0.41 0.43 0.40 0.26 480-499 0.44 0.36 0.49 0.40 0.26 500-599 0 40 0.40 0.45 0.41 0.27 600-679 0.42 0.41 0.45 0.42 0.25 680-699 0.54 0.43 0.32 0.44 0.30 700-749 0.35 0 40 0.57 0.41 0.24 750-799 0.41 0.42 0.37 0.41 0.25 800-899 0.44 0.41 0.54 0 43 0 25 900-979 0.43 0.43 0.38 0.43 0.25 980-999 0.32 0.36 0.58 0.37 0.22 1,000-1,249 0.44 0.47 0.56 0.47 0.27 1,250-1,499 0.42 0.48 0.59 0.48 0.25 1,500-1,749 0.67 0.47 0.63 0.52 0.25 1,750-1,999 0 51 0.51 0.53 0.50 0.20 2,000-2,499 0 45 0.45 0.59 0.49 0.26 2,500-2,999 0.52 0.42 0.50 0.44 0 23 3,000-3,499 0.44 0.44 1.00 0.48 0.26 3,500-3,999 0.43 0.52 0.38 0.50 0.27 4,000-4,999 0.09 0.41 0.49 0.47 0.29 5,000+ 0.28 0 56 0.58 0.52 0.27 All income classes 0.37 0.43 0.43 0.40 0.28 Note: The participation rate for a household is the ratio of number of income recipients to household size. a. Standard deviation of participation rate per household for all races including "other." INEQUALITY IN LEVELS OF LIVING 75 Table 3-5. Average Dependency Ratio per Household by Household Income Class and Racial Group Household All races income class including Standard (Mg per month) Malay Chinese Indian other" deviationa No income - - - - 1-39 2.43 1.99 1.88 2.40 1.76 40-49 3.22 1.64 2 21 3.07 1.90 50-79 3.74 2.20 2.97 3.55 2.13 80-99 4.11 2.78 3.64 3.91 2.30 100-129 4.24 3.23 3.75 3.98 2 35 130-149 4.21 3.70 3.82 4.04 2.40 150-179 4.22 4.00 3.79 4.09 2.42 180-199 4.08 4 01 4.07 4.05 2.45 200-279 4.17 4.03 3.80 4.06 2 47 280-299 4.27 3.83 3.66 3.98 2.48 300-399 4.14 3.79 3.68 3.90 2.44 400-479 4.11 3.62 3.70 3.78 2.46 480-499 3.16 4.17 2.76 3.70 2.46 500-599 3.74 3.73 3.55 3.70 2.48 600-679 3.65 3.51 3.42 3.53 2 55 680-699 2.78 3.73 4.50 3.60 2 43 700-749 4.02 3.38 2.51 3.45 2.27 750-799 3.31 3.23 3.76 3.33 1.79 800-899 3.30 3.40 2.92 3.31 2.33 900-979 3.31 3.20 4.11 3.31 2.10 980-999 3.57 3.53 2.75 3.44 1 62 1,000-1,249 3.17 3.13 2.56 3.07 2.13 1,250-1,499 3.75 3.05 2.39 3.02 2.52 1,500-1,749 2.01 2.76 2.14 2.53 1.57 1,750-1,999 2.24 2.43 2 64 2.41 1.31 2,000-2,499 2.72 2.91 2.45 2.68 1.51 2,500-2,999 3.78 3.44 2.18 3.14 2.05 3,000-3,499 3.13 2.85 1.00 2.69 1.53 3,500-3,999 2.83 2.43 2.90 2.59 1.24 4,000-4,999 11.00 2.50 2.73 3.24 2.79 5,000+ 9.00 2.78 1.75 3.37 3.94 All income classes 3.88 3.61 3.60 3.76 2.37 -- Not applicable. Nlote: The dependency ratio for a household is the ratio of household size to number of income recipients. a. Standard deviation of dependency ratio per household for all races including "other." 76 INEQUALIT Y AND POVERTY IN MALAYSIA The average household size by household income class and racial group is shown in table 3-6. Excluding the zero-income class, average household size increases with income up to the M$980-M$999 class, after which it fluctuates a little. Households in the M$1 - M$39 class have an average of Table 3-6. Average Household Size by Household Income Class and Racial Group Household All races income class including Standard (MS per month) Malay Chinese Indian 'other' deviationa No income 4.24 5.16 4.47 4.51 2.52 1-39 2 62 2.01 1.93 2.57 1.87 40-49 3.54 1.70 2.35 3.36 1.94 50-79 4.28 2.36 3.33 4.04 2 22 80-99 4.90 3.12 4.04 4.61 2.37 100-129 5 29 3.67 4 23 4 84 2.51 130-149 5.51 4.42 4.69 5.15 2.54 150-179 5.62 4.85 5.10 5.31 2.63 180-199 5.79 5.33 5.95 5 67 2 70 200-279 6.02 5 85 6.32 6.00 2.87 280-299 6.24 6.40 6.92 6.39 2.84 300-399 6.33 6.54 6.94 6.51 3.08 400-479 6.66 6 84 6.36 6 71 3.34 480-499 6.43 7.54 6.59 7.08 3.63 500-599 6.31 7.69 5.93 7.05 3.63 600-679 7.14 7.60 6 22 7.25 4,12 680-699 5.06 7.51 8.33 7.01 4.57 700-749 6.67 8.07 5.65 7 34 3.50 750-799 7.30 7.66 6.47 7.39 3 85 800-899 6.45 8.17 5 77 7.49 3.91 900-979 7.25 8.02 6.56 7.61 4.80 980-999 6.71 10.28 6.67 8.79 4.79 1.000-1,249 7.05 7.46 5.63 7.02 3.94 1.250-1.499 8 17 7.91 6 22 7 42 4.28 1,500-1,749 6 75 7.88 7.47 7.53 4 57 1.750-1,999 8.27 7.52 7.33 7.59 3 43 2,000-2,499 8.13 7 20 6.85 6.53 4.02 2,500-2.999 7.33 9.12 8 14 7 82 4.33 3,000-3.499 11.00 7.40 3 33 6.54 4.14 3.500-3,999 4 50 10.13 6.50 7.47 4.18 4.000-4,999 11.00 6.33 13.20 9.36 5.36 5.000+ 11.00 14.79 4.50 13.11 1255 All income classes 5.084 5 839 5.453 5.363 3.09 Note: See tables 3-12 and 3-13 for a disaggregation by urban-rural location. a. Standard deviation of household size for all races including "other." INEQUALITY IN LEVELS OF LIVING 77 2.57 members, while those in the M$980-M$999 class have an average of 8.79 members. For incomes above M$1,000, the average household size drops slightly and varies before climbing again to reach its maximum of 13.11 for those with incomes of M$5,000 and over; the average household size for the entire sample is 5.36 members.'9 Across income classes, a correlation coefficient of 0.81 has been computed between average household size and income level. Table 3-6 indicates a large variation in household size at each income level, measured by the standard deviation of household size within each income class. The variability in size at a given income level causes large- sized households to be ranked below small-sized ones in the ordering by per capita householcl income. Apart from the variability within income classes, size variability across income classes can also cause a reordering of households. Thus a large-sized high-income household can turn out to have a lower per capita income than a small-sized low-income household. Both factors can lead to substantial reordering of households when ranked by per capita household income rather than by household income. Starting from the joint distribution functionf (y, m) of households with household income y and size m, I derive in appendix F the theoretical dist-ribution of households and of individuals according to per capita household income. I show there the exact mathematical transformations required to effect these mappings. Sufficient conditions are specified for the per capita household income distribution to be identical to the household income distribution. Furthermore, I disprove (by counterexample) the widely held but erroneous belief that if average household size increases with household income, the per capita household income distribution is mcore equal than the household income distribution. The last proposition needs some elaboration. As noted above, it is not merely the variation of average household size across income classes that causes the household and per capita household income distributions to diverge. They can diverge even if average household size is constant across income classes, so long as there is a nonzero variance around this average within income classes.20 Except in certain special cases, however, it is dilficult to predict the direction of divergence (in terms of inequality) when one moves from the household to the per capita household income distribution. 19. Since the sample size thins out in upper income classes, statistics for these groups are subject to increasingly large sampling error. For example, there were only 19 households in the tol) income class (M$5,000+), of which one had 56 members, 53 of whom were income recipients. 20. The larger this variance, other things being equal, the larger is the inequality in the per capita household income distribution (see appendix F). Table 3-7. Distribution of Households by Household Income and per Capita Household Income A verage household Percentage share of or per capita inconme by percentile group household Average Sample size income household Gini Lowest Lowest Highest Highest (number ofj Households (MS per month) size coefliient quintile 40 percent quinuile 5 percent households) Household income All 264 5.363 0.5129 3.5 11 5 55.7 28.3 25,023 Malay 172 5.084 0.4664 4.3 13.2 51.6 24.0 13,863 Chinese 394 5.839 0.4656 4.8 13.8 52.6 25.5 8,003 Indian 304 5.453 0 4722 5.0 14.8 54.0 28.2 2,936 Other 813 4.416 0.6673 0.5 2.2 68.2 26 0 221 Per capita household income All 62 0.5374 4.9 10.9 59.2 30.5 25,023 Malay 41 0.4926 4 5 13.0 56.0 26.0 13,863 Chinese 86 0.4834 4.7 13.0 54.0 26 5 8,003 Indian 79 0.5382 4.0 11 3 59.2 31.0 2,936 Other 249 0.6998 0.5 1.9 72.0 29.2 221 Note: Further intormation on these distributions is provided in tables 3-14 to 3-16. a. The standard deviations of household income are: all, 417; Malay, 226, Chinese, 548, Indian. 451; and other. 1,190. INEQUALITY IN LEVELS OF LIVING 79 The Distribution of Households by per Capila Household Income For Malaysia, there is an increase in inequality among households in going from the household income distribution to the per capita household income distribution. The Gini coefficient is 0.5129 for the household income distribution, and 0.5374 for the per capita household income distribution (table 3-7; see also tables 3-15 and 3-16). The Lorenz curves for the two distributions are drawn in figure 3-1. Since they intersect, their Figure 3-1. Lorenz Curves for the Household and per Capita Household Income Distributions -- Distribution of households by household 90 income Mean = M$264 per month Gini coefficient = ().5129 80 … Distribution of households by per cbpita household income Mean = MS62 per month Gini coefficient = 0.5374 70 - --- Distribution of individuals by per capita household income Mean = MS50 per month ° . - 60Gini coefficient = 0 4980 E -o o 0 50 2 5C - -~40 u~~~~~~ 30 20 - 30 20 0 10 20 30 40 50 60 70 80 90 100 Cumulative population (percent) 80 INEQUALITY AND POVERTY IN MALAYSIA inequality ranking is not unambiguous in the sense that different meastures of inequality could rank the distributions differently. The use of Lorenz comparability in this study is basically positive, however, not normative. It relies on the proposition that if the Lorenz curves of two distributions are nonintersecting (that is, one Lorenz curve dominates, or lies above, the other), then all measures of inequality which satisfy mean independence, population-size independence, and the Pigou-Dalton condition will show less inequality for the Lorenz-dominant distribution (see appendix D for a proof). Without drawing any welfare implications about the underlying distributions, I use Lorenz dominance simply to establish an unambiguous ranking of inequality by all indices belonging to this wide class. Indeed, in most cases it is impossible to undertake meaningful welfare comparisons between the underlying distributions because they refer to different population units or income concepts. It is still useful, however, to show Lorenz dominance where possible, as this allows automatic comparison by most well-known inequality measures. The increase in inequality among households in going from the household income distribution to the per capita household income distribution occurs despite the fact that average household size increases with household income, and at a rate which is less than proportionate with income (see table 3-6). In these circumstances an unambiguous decrease in inequality would be expected if household size were constant at each income level (see appendix F, proposition 1). Hence I conclude that it is the variation of household size within income classes which produces the observed increase in inequality by substantially reordering households. Table 3-7 presents summary information on the distribution of house- holds by per capita household income (see also table 3-16). The average per capita household income in Malaysia is M$62 per month, but the importance of this number seems limited: it is simply the unweighted mean of per capita household incomes. The true per capita income (Pci) in Table 3-8. Distribution of Individuals by per Capita Household Income Percentage share of income Per capita by percentile group Sample size income (number (MI per Gini Lowest Lowest Highest Highest oJ ,ndr- Individuals month) coefficient quintile 40 percent quinule 5 percent riduahI) All 50 0.4980 4.3 12.3 54.8 28.5 134,186 Malay 34 0.4553 5.2 14.8 52.2 24.6 70.474 Chinese 68 0.4542 5.3 14.3 52.8 26.8 46.720i Indian 57 0.5003 5.0 13.7 56.7 29.5 16,010 Other 185 0.7071 0.5 2 3 75.5 31.0 976 INEQUALITY IN LEVELS OF LIVING 81 Malaysia is the mean of distribution 3, which has been computed as M$50 per month. This is the weighted mean of per capita household incomes, where the weights are the household sizes. It is easy to see that Pci is equal to avera,ge household income divided by average household size, but there is obviously no reason for average per capita household income to equal this. The relations among the different mean incomes of the three distributions in tables 3-7 and 3-8 can be expressed as follows: Let Yh be the income of household h and mh its size, for h = 1, 2, . .. , H. Then 1 H Average household income, j=- E Yh = M$264. H h =l l Hq Average per capita household income, (y/m) =- E (Yh/mh) = M$62. H h = I Per capita incoDme, PC] = - (1/ Em EMh(YhlMh) = M$50 I1H = 3/Ir, where mh = Y- M = 5.363. H h = I The Distribution of Individuals by per Capita Household Income Since inequality in levels of living among individuals is of ultimate interest, it is necessary to examine the distribution of individuals by per capita household income. In further discussion of discrepancies in living standards in Malaysia, I restrict attention to this distribution, the summary characteristics of which are set out in table 3-8. The Gini coefficient of the distribution is 0.4980, which implies a fairly high degree of inequality in the country. Individuals in the 40 percent of the population with the lowest incomes receive only 12.3 percent of total income, whereas those in the top 5 percent receive 28.5 percent of total income. The lowest quintile get a mere 4.3 percent, and the ratio of the shares of the highest to the lowest quintile (a frequent measure of inequality21) is almost 13. The extent ol inequality in distribution 3 seems lower than that indicated in distributions I and 2. Since the Lorenz curve for distribution 3 intersects 21. An obvious defect of this measure is that it satisfies only the "weak" principle of trEnsfers, which states that transfers from rich to poor should either reduce the measure or leave it unchanged. For example, transfers among the 60 percent of the population'in the middle income range (or within the top or bottom quintiles) leave the index unchanged 82 INEQUALITY AND POVERTY IN MALAYSIA the Lorenz curves for the other two distributions, however, an unambigu- ous comparison is not possible; compare the shares of the highest 5 percent in I and 3, and the shares of the lowest quintile in 2 and 3 (see tables 3-7 and 3-8). In fact, it might have been expected that distribution 3 would Lorenz- dominate distribution 1 (see appendix F, proposition 2); but, as noted earlier, the reordering of households when they are ranked by per capita household income prevents this outcome (because the hypothesis of proposition 2 is not met). A breakdown by racial group of the distribution of individuals by per capita household income has been effected in table 3-8. It shows the per capita income of the Chinese community (M$68 per month) to be twice that of the Malay community (M $34 per month). The per capita income of the Indians (M$57 per month) is around 1.7 times that of the Malays. These disparity ratios are lower than those calculated on the basis of average household income.22 The reason for this is evident from table 3-7, since the per capita income of a group is simply its average household income divided by its average household size. As the Malays have a lower average household size (5.084 members) than the Chinese (5.839 members) and the Indians (5.453 members), the disparity ratios in per capita income will be smaller than the disparity ratios in household income. Hence disparity ratios in Malaysia are lower than is indicated by researchers who neglect racial differences in household size. The distribution of individuals by per capita household income within racial groups shows that Malay and Chinese incomes are distributed very similarly around their respective means. The corresponding fractile shares are close to one another, and the Gini coefficient for the Malay distribution (0.4553) is very similar to that for the Chinese distribution (0.4542). Indian incomes are distributed somewhat more unequally (Gini coefficient of 0.5003), and "other" incomes are distributed extremely unequally (Gini coefficient of 0.7071). The Atkinson Index So far I have attempted to measure inequality largely by positive or descriptive indices such as various fractile shares. Another type of index (Atkinson, 1970) is based explicitly on a social welfare evaluation of income 22. On the basis of average household income. the Chinese-Malay disparity ratio is 2.29i, and the Indian-Malay ratio is 1 77 (see chapter 2 or table 3-7). INEQUALITY IN LEVELS OF LIVING 83 distribution.23 Let y = (Y1, Y2, . . , yn) denote an income distribution among n individuals, where y, ) 0 is the income of individual i = 1, . . . , n). Denote the mean income level by p, so that n n i= Z1. Given a social welfare function, the Atkinson index is constructed by computing the equally distributed equivalent income, YEDE. of the distri- bution. This is defined as the level of income per head which, if equally distributed, would give the same level of social welfare as the existing distribution. Atkinson's index is then defined as the difference between YEDE and the mean income p of the distribution, in proportionate terms. This definition is independent of the actual welfare function chosen, but in practice Atkinsori restricts himself to the class of additively separable and symmetric functions of individual income. Formally, YEDE is defined through n nU(yEDE)= E U (yi), ,= I which is the level of social welfare associated with the existing distribution y, and the Atkinson index I is defined as I-( YEDEIP-) If I is required to be a mean-independent inequality index, the function U(y) must be limited to the constant elasticity marginal utility form, that is, U(y)= I -)E logy, z:= 1 up to a positive linear transformation. Only values of E ) 0 are considered so that U(y) is concave (that is, it displays nonincreasing marginal utility). This condition implies YEDE p. The choice of a particular value for E is obviously a value judgment; E measures the degree of inequality aversion. As E rises, more weight is attached to transfers at the lower end of the distribution and less weight to transfers at the top. As E -a we get the so- called Rawlsian function min, {yj }, and as E - 0 we get the linear utility function which ranks distributions solely according to total income. 23 This is the pioneering contribution on normative measurement of inequality. Roughly speoLking, Atkinson shows that normative inequality comparisons between two distributions can be made without reference to a prespecified social welfare function (as long as it is strictly con,cave) if, and only if, the corresponding Lorenz curves are noninteresecting except at the end points (that is, one Lorenz curve dominates the other). A formal statement and proof of this theorem are provided in appendix D. 84 INEQUALITY AND POVERTY IN MALAYSIA The choice of e within the range 0 to oo remains arbitrary. In the empirical sections of his paper, Atkinson (1970) restricts himself to values for e between 1.0 and 2.5. Others advocate a value of about 2.0. Stern (1977) reviews the literature on the elasticity of marginal utility of income and presents a number of arguments in support of values between 1.5 and 2.5. It seems, therefore, that a value for e of 2.0 might be reasonable and broadly acceptable. Unfortunately, the Atkinson index cannot be computed on the Malaysian data for values of e > 1 because of the presence of zero-income individuals in distribution 3.24 Welfare falls to minus infinity, and the equally distributed equivalent income YEDE is not defined in this case.25 This is not simply a technical problem which can be got around by assigning a small positive income to zero-income individuals, as is sometimes sug- gested. The value of the inequality index will depend crucially on the particular income assigned, and it can be made arbitrarily close to unity (perfect inequality) by choosing a small enough level. When there are zero-income individuals, the Atkinson index can perhaps be extended to a lexicographic ordering of inequality. Thus distributions might be ranked by first comparing the number of zero-income individuals in each. If the number of zero-income individuals is the same in both distributions, the usual Atkinson index computed for the positive- income individuals can be used to rank them. The natural index for this extended ordering is one which shows the number of zero-income individuals before the decimal point, and the Atkinson index for positive- income recipients after the decimal point. However, the value judgment implicit in this extension is open to serious question. For example, it might be difficult to accept that a distribution with a greater number of zero- income individuals is necessarily worse, independently of the number of near-zero-income individuals in the other distribution. The difference between a zero-income level and an infinitesimally small income level is unlikely to be a decisive one in economic terms.26 With zero-income individuals, therefore, it seems that we can properly use the Atkinson index only if e is chosen to be less than (or equal to) 24. There are 1.366 zero-income individuals (corresponding to 341 zero-income house- holds) in this sample of 134,186 individuals. 25. For the case of E = 1. YEDE iS simply the geometric mean income of the distribution, and the Atkinson index is I minus the ratio of the geometric to the arithmetic mean incorne. With a single zero-income individual, the geometric mean is zero and the Atkinson index takes the value unity. 26. A discontinuity at the subsistence level of income might be perfectly defensible, however, if the difference is interpreted as one between life and death. INEQUALITY IN LEVELS OF LIVING 85 Table 3-9. Equally Distributed Equivalent Income and Atkinson Inequality Index for the Distribution of Individuals by per Capita Household Income (MS per month) Sample E size (number of Individuals 0.00 0.25 0.50 0.75 0.90 0.99 individuals) All 49.9 44.1 39.3 34.8 309 11 9a 134,186 (0.1162) (0.2124) (0.3026) (0.3807) (0.7615) Malay 34.3 31.0 28.2 25.3 22.6 8.1 70.474 (0.0962) (0.1778) (0.2623) (0.3411) (07638) Chinese 68 4 61.8 56 2 50.9 45 9 18.7 46,726 (0.0964) (0.1783) (0.2558) (0.3289) (0.7266) Indian 56.5 49 7 44.3 39.6 35.9 17.7 16,010 (0.1203) (0.2159) (0.2991) (0.3646) (0.6867) Other 185.4 143.1 104.4 71.5 52.0 10.3 976 (0.2281) (0.4368) (0.6143) (0.7195) (0.9444) Note: Figures in parentheses are the Atkinson inequality index. a. When the 1,366 5'ero-income individuals are omitted, the equally distributed equivalent income YEDE for e = 0.99 is as high as M$33.1. unity.27 Table 3-9 presents the equally distributed equivalent income YEDE for six different values of £ less than unity (from 0.0 to 0.99) and the corresponding Atkinson inequality index I (in parentheses). For E = 0, YEDE is simply the mean income (p) of the distribution, which has earlier been called Pci. For E := 0.5, YEDE= M$39.3, which means that if incomes were equally distributed, it would require only M$39.3 per person to achieve the same level of social welfare as the existing distribution with a mean income of M$49.9. Thus a proportionate income "loss" of (p - YEDE)/P = 21.24 27. The restriction might lead some to reject values for E which are greater than unity-or U(y) functions which are unbounded below. The problem raised by zero-income individuals is not simply that current income is an imperfect proxy for levels of living It could arise even with current consuniption as the proxy, if the reference period of the survey were very short. This leads naturally to an interest in levels of living over the longer run, but for a normative measurement of their inequality, essentially the same underlying problem of welfare economics would have to be faced. How does one evaluate in welfare terms a level of living which falls below the long-run subsistence minimum and implies starvation and premature death? Such levels of living are certainly consistent with a positive level of long-run or "permanent" consumption. This suggests the original problem is not in essence the empirical one of zero incomes, but it raises deep questions of welfare economics which are difficult to handle in a dynamic context. 86 INEQUALITY AND POVERTY IN MALAYSIA percent arises from the inequality in the distribution, which gives a value of 0.2124 for I. The table also shows YEDE and I for each racial group separately. For £ = 0.5, the Atkinson inequality index for Malays, Chinese, Indians, and others, respectively, is 0.1778, 0.1783, 0.2159, and 0.4368. Again, the Malay inequality coefficient is close to that of the Chinese. This reinforces the earlier observation about the similarity of the Malay and Chinese distributions about their respective means. As the inequality aversion parameter E increases, YEDE decreases and I increases; with zero-income individuals, neither is defined for E > 1. In this case, YEDE can be made arbitrarily close to zero, and I to unity, by choosing E sufficiently close to unity. Witness the sharp drop in YEDE as E approaches unity in table 3-9.28 When E = 1, in fact, YEDE iS the geometric mean of the distribution, which is zero with zero-income individuals, and in this case I is equal to unity. The Methodology of Inequality Decomposition Traditionally the study of between-group inequality in Malaysia has been conducted in terms of differences in mean income between the groups. Thus, for example, the previously noted racial disparity ratios between non- Malays and Malays simply reflect mean income differences between the groups. Such ratios completely ignore income differences within racial groups, which could turn out to be quite large. It is interesting, therefore, to ask how much of total inequality in Malaysia consists of between-group inequality and how much of within-group inequality. The obvious way to answer this question is to decompose total inequality into between-group and within-group inequality. The between-group contribution might then be defined as the ratio of between-group to total inequality (and similarly the within-group contribution). The inequality indices considered so far, however, are not neatly decomposable into between- and within-group terms. For example, the Gini coefficient canrot in general be written as the sum of a between-group component and a within-group component, where the two components have a natural meaning (see "On the Decomposition of the Gini Coefficient" in appendix B). I attempt to give a consistent definition for these terms. 28. When zero-income individuals are omitted from the distribution, the value of YEDE for = 0.99 is M$33.1 (note to table 3-9). With zero-income individuals omitted, YEDE is defined for all E > O. Its values for £ = 0.5, 1.0,2.0, and 3.0 are M$40. 1, M$33.0, M$23.2, and M$lf.0, respectively. Since there are 1,366 fewer individuals in this case and the same total income, the arithmetic mean ji is slightly higher at M$50.4 (compared with M$49.9). INEQUALITY IN LEVELS OF LIVING 87 The between-group component can be defined as the value of the inequality index when all within-group income differences are artificially suppressed. In other words, it is the value of the inequality index for the hypothetical income distribution which assigns to each person within a group the mean income of the group. In this way, within-group inequality is eliminated, and the resultant distribution shows only the inequality arising from between-grouip income differences. Likewise, the wiihin-group component can be defined as the value of the inequality index when all between-group income differences are suppressed. Thus a hypothetical income distribution is constructed in which the group mean incomes are equalized to the overall mean through an equipropor- tional.e change in the income of every person within a group. In this way, between-group inequality is eliminated, but the inequality within each group remains constant (assuming the inequality index is mean- independent).29 The within-group component is then the value of the index for this hypothetical distribution. An inequality index may be said to be additively decomposable if for any grouping total inequality can be written as the sum of between-group and within-group inequality. This property allows the unambiguous measure- ment of the contribution of a particular grouping (or variable) to overall inequality.30 The Gini coefficient does not satisfy the property of additive decompos- abiliity as shown by a racial grouping of distribution 3. The between-race component according to the above definition is 0.1648, which is the value of the Gini coefficient when each member of a group receives the mean income of that group; that is, all 70,474 Malays get M$34 per month, all 46,726 Chinese get M$68 per month, all 16,010 Indians get M$57 per month, and all 976 others get M$ 185 per month. The within-race component according to the above definition has not been computed, but it is at least as large as 0.4621, which is the population-weighted average of Gini coefficients of the four racial groups.3' Hence the sum of the between-group and within- group components is at least as large as 0.6269, whereas the overall Gini 29. With a mean-independent inequality index, in fact, it does not matter to what income level all the group mean incomes are equalized in constructing the hypothetical distribution, provided the level is non-zero. 30 Notice the analogy with regression analysis, where the total variance is decomposed into the sum of the "explained" variance and the "unexplained" variance, and R2 is the fraction of the total variance explained by the independent variable(s). The analogy is discussed further in chapter 6. 31. The section "On the Decomposition of the Gii Coefficient" in appendix B proves that the Gini coefficient of a composite group is always greater than or equal to the population- weighted average of subgroup Gini coefficients. 88 INEQUALITY AND POVERTY IN MALAYSIA coefficient is only 0.4980 (table 3-8). The Gini coefficient is not de- composable into the sum of between- and within-group inequality and therefore cannot provide an adequate basis for assessing the relative importance of these two sources of inequality.32 It would be an interesting exercise to characterize the complete class of inequality measures decomposable according to the strict definition given above. I suspect the class is quite restrictive, although Theil's second measure L (discussed below) belongs to it as does the variance of the logarithm of income (variance of log-income).33 The latter satisfies the above decomposition property if the means in the definitions of between- group component and within-group component are interpreted to be geometric means.34 The decomposition formulas for Theil's second measure L and the variance of log-income are shown in appendix C, "The Decomposition of Three Inequality Measures." The within-group component for these measures is a weighted average of the inequality indices for each group, where the weights are simply the population shares of the groups.35 Unfortunately, neither Theil's second measure L nor the variance of log- income can be used as inequality measures in this chapter, although I will use them in chapters 6 and 7. The reason is a technical one, namely, the existence of individuals with zero per capita household income. This causes log-income and the logarithm of geometric mean income to blow up to minus infinity, so that the measures become uncomputable. It is exactly the 32 By the above definition, the between-group contribution of racial inequality is 33.1 percent (0 1648 divided by 0.4980). Yet if interracial inequality were to be eliminated, with intraracial inequalities kept constant, overall inequality as measured by the Gini coefficient would reduce by less than 7.2 percent (0.4980 minus 0 4621, all divided by 0.4980). In this sense, the between-group contribution to overall inequality would be less than 7.2 percent! This inconsistency in interpretation arises because the Gini coefficient is not decomposable. The two interpretations are consistent for any inequality index which satisfies the above decomposition property. See the section "The Interpretation of Decomposition for Three Inequality Measures" in chapter 6. 33. The variance of log-income satisfies mean independence and population-size independence, but it does not satisfy the Pigou-Dalton condition over the entire range of incomes (see appendix A). 34. In other words, the between-group component is the variance of log-income for the hypothetical distribution in which each person within a group gets the geometric (not arithmetic) mean income of the group. Likewise. the within-group component is the varalice of log-income for the hypothetical distribution in which the group geometric mean incomes are equalized (to the overall geometric mean) through an equiproportionate change in the income of every person within a group. 35. The definition given earlier for within-group component does not restrict it to be of this form a priori. INEQUALITY IN LEVELS OF LIVING 89 same reason which prevented the use of the Atkinson index for values of E greater than unity. I relax slightly the definition of within-group component so that other measures of inequality, which are computable when there are zero incomes, can be classified as decomposable. Keeping the definitions of between- group component and additive decomposability the same as before, a weaker definition is provided for within-group component, which extends the class cf decomposable measures. It requires only that the within-group component be constructed from the individual inequality indices, population sizes, and total incomes of each group, as an additively separable function over groups so that the contribution of each to overall inequality can be identified. The only functional form possible, if the measure is also to satisfy the basic properties of mean independence and population-size independence,36 seems to be a weighted sum of inequality indices for each group, where the weights depend only on the population share and income share of the groups. This form of decomposability is clearly desirable in that overall inequality is built up as a sum of its constituent parts, *vhich isolates the contribution of each; thus knowledge of changes in constituent parts allows us directly to predict changes in overall inequality. There are several measures which satisfy this definition of decompos- abilitv as well as other desirable properties for an inequality index, namely, mean independence, population-size independence, and the Pigou-Dalton condition. Among such inequality measures are: -Theil's entropy index T (see Theil, 1967, pp. 91-95) -Theil's second measure L, which is the logarithm of the ratio of arithmetic mean income to geometric mean income (see Theil, 1967, pp. 125-27) -The squared coefficient of variation C2 -.Any index A of the form A=I Y~EDE for I- nt. The proportion of the population in poverty is (qln), and the poverty gap is q where gi = (t - y,) is the income gap of person i. Thus the poverty gap is equal to q E (r- yi) = q (7r -v). Therefore, the average poverty gap is (7i - v); the proportionate average income shortfall from the poverty line is (7r - v)/it; and the normalized value of the Sen index is (q/n) (7r - v)/n. The rank-order-weighting scheme irnplies a weight of (q + I - i) on the income gap .y, of person i, since there are (q + I - i) persons among the poor with incomes at least as large as that of person i. The Sen index P is then q A E (q+l-i)(ir-y1) where A is a parameter depending on the normalization selected. The 120 INEQUALITY AND POVERTY IN MALAYSIA normalized value of the index, when each yi = v, is q nt-v q(q+l) q q(q+1) n = A (i - v) 2 since Z (q + 1 - i) = nii 2 t*= 1 2 Thus A = 2/[(q+l)n7n]. Now the Gini coefficient Gp of the income distribution among the poor can be written as: Gp = - 2 (q+1-i)y, (see definition G4 of the Gini coefficient in appendix B).'7 Therefore, qp = 7 as 1 [n + q v For large q, q/(q + 1) f 1, and the index P reduces to qlI P = - - [7i- v (I - Gp n i The effect of the weighting scheme is to augment the average poverty gap by the Gini coefficient times mean income of the poor. Thus an additional income "loss" arises when inequality in the income distribution among the poor is taken into account. The correction for this loss involves deflating the mean income v of the poor by (1 - Gp), which yields the familiar equally distributed equivalent income (see Atkinson, 1970) corresponding to the rank-order welfare function. Hence the weighted income gap is calculated by taking the difference, not between the poverty line and the mean income of the poor, but between the poverty line and the equally distributed equivalent income of the poor. The index P lies between 0 and 1. It assumes the value 0 when everyone's income is above the poverty line it (that is, when q = 0), and the value 1 when everyone has zero income (implying v = 0 and q = n). The rank-order welfare function is rather special in that the relative weight on a person's income depends only on the rank of the person in the income ordering and not on the amount of the person's income as such. 17. If G is the Gini coefficient for the whole population, and G,P the Gini coefficient of the income distribution among the nonpoor, the following relation holds: G = (q)qv)G,+ ( q)(nu v )G.,, + (q)(-v) n npx n njunp (see note 5 in the section "On the Decomposition of the Gini Coefficient" in appendix B). Thus the Gini coefficient for the nonpoor G. can be inferred from a knowledge of G (table 3-8) and Gp (table 4-2). DEFINITION AND MEASUREMENT OF POVERTY 121 Other welfare functions with relative weights that do depend directly on the size of a person's income may be found more acceptable. The weighting schemes implied by them produce different expressions for equally distributed equivalent income and, by the same token, different indices of inequality. It is evident that the weighted income gap under any welfare function is simply the difference between the poverty line and the corresponding equally distributed equivalent income. Hence for each different welfare function there corresponds by this approach a different index of poverty. An obvious consequence of using such income-weighting schemes should be mentioned. In the Sen index P, for example, there is clearly a tradeoff between the mean income (v) of the poor and equality (I -Gp) in their income distribution, the tradeoff being given by v(1 - GP). Thus it is perfectly possible for the Sen index to register a decline in poverty when the poor have become poorer in absolute terms (that is, v has decreased) so long as equality in their income distribution (1 - GP) has increased more than proportionately. Put another way, the index implies a reduction in poverty even if there are transfers of income from the poor to the nonpoor (or the amount taken from the poor is simply thrown away) so long as the remaining incomes of the poor are sufficiently better distributed. A maximum reduction of (1 - GP) percent can be made in the total income of the poor, yet an improvement in distribution can still neutralize the effect of this income loss on the Sen index. These implications, while acceptable when weighting the incomes of the entire population as in the Atkinson inequality index, may be more difficult to swallow when applied only to those below the poverty line, especially if this is interpreted as an absolute minirmum. In this case, one may not wish to weight the incomes of poor people differently, preferring instead the value judgment of equal or unit weights on all their income gaps. Treating the incomes of the poor similarly yields a poverty measure which is simply the normalized value of the Sen index. It was noted earlier that a commonly used index of poverty is the percentage of GNP needed to close the poverty gap. A slightly different normalization from the one used by Sen produces a poverty measure which generalizes this index to correct for income inequality among the poor. The normalization can be modified so that when incomes below the poverty line are equal the measure reduces to the poverty gap expressed as a fraction of the total income of society; that is, (qln) (7r - v)/,u. With this normalization, A takes the value: A = 2/[(q + l)ny].18 The rank-order weighting pro- 13. Sen himself (1973b) alludes to this kind of normalization, but his equations (8) and (9) imply a different value for A, namely, A = 2/n2p, 122 INEQUALITY AND POVERTY IN MALAYSIA cedure now yields the modified Sen measure given by qi1 M = q- [7r- v(l-GI)] n l The relation between P and M is M = (7t/pu)P, and the measure Ml lies between the limits 0 and 7t/p.'9 The measure M reduces to the proportion of total income needed to close the poverty gap in either of two circumstances: (1) incomes below the poverty line are equally distributed (implying Gp = 0), or (2) the same weight of unity attaches to the income gap of every person below the poverty line. Instead of expressing the income required to close the poverty gap as a fraction of total income, define an index F (after Fishlow, 1972, 1973) which expresses the gap as a fraction of the income of the nonpoor: F = l [i [- v(l -Gp)] nju- qv = P. q n The idea behind this index is the elimination of poverty through a direct transfer of income from the nonpoor to the poor.20 The ratio reflects the burden on the nonpoor since it represents the proportionate reduction in their income if the poverty gap is to be closed through redistribution alone. Three comments are appropriate about these indices M and F. First, they are not so much measures of poverty as indicators of the economy's capacity for its alleviation. Failure to distinguish the measurement of poverty from the prospects for its alleviation can lead to the following 19. If one adopts Atkinson's suggestion for defining a relative poverty line (discussed earlier in this chapter), nt and p stand in constant relation to each other, namely, n/, = 1/2. In this case, M and P are related as M = P12, and M lies between the limits of 0 and 1/2. 20. If poverty is to be eliminated by transfers alone, the income of the nonpoor must be large enough not to drag the nonpoor themselves into poverty in the course of income transfers. This motivates yet another normalization for the Sen index, in which the poverty gap is expressed as a fraction of the income of the nonpoor in excess ofthe poverty level. This measure is easily seen to reduce to n n n where P is the Sen index. DEFINITION AND MEASUREMENT OF POVERTY 123 anomalous consequence. With no change in the number or the incomes of the poor, an increase in the incomes of people above the poverty line will lead to a fall in both the indices M and F, Yet no reduction in poverty has actuallyt occurred since the position of the poor remains unchanged.2' What has happened is that a smaller fraction of society's income is now required to eliminate poverly, and to that extent the task may be regarded as potentially easier. The measurement of poverty thus needs to be conceptu- ally separated from the possibilities for its alleviation. Second, the values of F and M could exceed unity if the poverty line happens to be drawvn at a level higher than the mean income of society. Then, the augmented poverty gap could exceed the income of the nonpoor (or even the total income of society), implying a value of F(or of M) larger than unity. A sufficient condition for the poverty problem to be tractable through transfers is that the mean income of society exceed the poverty-line income. There is then enough income to bring everyone in the population above the poverty line. With this condition satisfied, both the indices M and F are bounded above by unity. Third, the assumption implicit in the indices M and F is that the redistribution of income to close the poverty gap does not affect the size of total income in the economy. Yet any transfer scheme based on the taxation systemn is likely to influence both the (pretransfer) poverty gap and the (pretransfer) incorne of the nonpoor through its disincentive effects. A full incomne support program implies a 100 percent marginal rate of taxation at and below the poverty line, as well as changes in taxation at higher income levels (to raise the required revenue). The work disincentive effects of such changes in the tax schedule could substantially increase the pretransfer poverty gap and reduce the size of pretransfer income of the nonpoor. This, of course, would alter the values of the indices M and F. Indeed, for each different tax schedule/transfer scheme that is contemplated to eliminate poverty, there will correspond different values for M and F. To calculate these and the optimum tax schedule requires information on individual labor supply functions, the distribution of skills, and so on.22 21. There might even be an increase in poverty if one takes a relative view ofpoverty. The Sen poverty measure Pis unchanged in this case since the income gap of the poor is normalized on tthe poverty line n and not on the mean income u of the entire community. Only an actual reduction in the num,ber of poor, or an increase in their incomes, or an improvement in their income distribution, or an increase in the number of nonpoor can lead to a fall in the Sen poverty index. 22. See Mirrlees 11971) for the pioneering contribution in the theory of optimum income taxation. The optimum tax schedule mentioned here is different from that of Mirrlees because of il]e additional constraint that no one's posttax income should fall below an exogenously chosen poverty line. Table 4-2. Estimates of Poverty by Racial Group A verage Rank-order poverty Weights of unity weights on gap per on income gaps Gini income gaps oJ Proportion person, of the poor coefficient the poor of persons (MS per of income Racial in poverty month) Index Index Index distribution Index Index Index group (q/n) (7-,u) p M F among the poor P M F Peninsular Malaysia 0.402 9.05 0.145 0.073 0.083 0.2126 0.200 0.100 0 115 Malay 0.562 9.74 0.219 0.161 0.215 0.2200 0.294 0.216 0.290 Chinese 0.183 6.80 0.050 0.018 0.019 0.1677 0.072 0.026 0.028 Indian 0.334 7.28 0.097 0.043 0.048 0.1658 0.137 0.060 0.067 Other 0.433 12.44 0.215 0.029 0.030 0.3328 0.288 0.039 0.040 DEFINITION AND MEASUREMENT OF POVERTY 125 Est:imates of Poverty in Malaysia The indices discussed above have been estimated for the Malaysian population in poverty. It is assumed first that the weights attaching to each person's income gap are unity. The indices then reduce to the poverty gap expressed as a fraction of various income aggregates. Estimates of these measu res, as well as the proportion in poverty and the average poverty gap, are presented in table 4-2 for Peninsular Malaysia and each ethnic group separately. The percentage of the population in poverty was calculated as 40.2 percent, and the average poverty gap as M$9.05 per month. The poverty gap as a fraction of the total income needed to support everyone in the population at the poverty level is 14.5 percent. The index M for the country was estimated as 0.073, which implies that the poverty gap in Malaysia stands at 7.3 percent of total personal income. If poverty were to be elimin,ated by transfers from the nonpoor to the poor, the nonpoor would need to sacrifice 8.3 percent of their income (or 12.7 percent of their income in excess of the poverty level).23 These indices have also been computed separately for each ethnic group. The average income gap is largest for the small and heterogeneous community of "others" (Europeans, Thais, other Asians, and so forth), while the incidence of poverty is highest among Malays. The product of these two measures divided by poverty-line income gives the Sen index in the case of unit weights, which shows that poverty is more acute among Malays than among "others." The values of M' and F for the communities show the poverty gap of each racial group as a fraction of various income aggregates for that group. For policy purposes, however, it is probably more useful to express the poverty gap of each racial group as a fraction of the overall poverty gap. Of the overall income shortfall, the Malays account for 79.0 percent, the Chinese for [1.9 percent, the Indians for 8.0 percent, and other races for 1.1 percent.24 Of the overall number in poverty, however, the Malays account for 73.5 percent, the Chinese for 15.8 percent, the Indians for 9.9 percent, and other races for 0.8 percent. The difference between these two sets of figutes obviously reflects the difference between races in their average poverty gaps. 23. See note 20 above. 24. In a poverty relief program, allocation to communities in these percentages will reduce their income gaps equiproportionately. 126 INEQUALITY AND POVERTY IN MALAYSIA Assume now that rank-order weights attach to the income gaps cf poor persons. The average poverty gap then needs to be augmented by the mean income times the Gini coefficient of the income distribution among the poor. This adjustment yields values for P, M, and F shown in the last three columns of table 4-2. The Sen poverty measure takes the value 0.200 for Peninsular Malaysia. It is difficult to judge whether the degree of poverty which this represents is large or small in the absence of estimates for other countries."5 In fact, one of the main reasons for evaluating the Sen measure in the unit weights (or distribution-free) case is that it has a straightforward interpretation there. Its value under rank-order weighting indicates the magnitude of the correction to the unit weights case, which arises from the inequality in incomes among the poor. According to the Sen index, poverty is greatest among Malays, followed by "others," Indians, and Chinese, respectively. Although the Sen index points to the severity of the problem within each racial group, it cannot be used to indicate the contribution of a group to overall poverty. Yet, in the design of policies to redress poverty, it would seem important to be informed of the extent to which a particular group accounts for overall poverty. An index which does permit poverty to be "decomposed" as a weighted average of poverty in each group is the simple incidence-of- poverty measure (unlike the case for an inequality index, there is obviously no between-group component for a poverty index).26 In the next section, this index is adopted to characterize the nature of poverty in Malaysia. A Profile of Poverty in Malaysia A characterization of poverty requires answers to questions such as: Who are the poor? Where are they located? In which sectors do they work? What 25. Like most indices of inequality, the Sen poverty measure is useful mainly for comparisons across countries or over time. Unlike indices of inequality, which are relative, however, indices of poverty (including the one by Sen) are sensitive to the choice of poverty line. This needs to be borne in mind when making intercountry comparisons of poverty. So far I have seen an estimate of the Sen index for only one other country: from National Sample Survey data on consumption in 1970-71, Ahluwalia (1977) estimates the Sen index for rural India as 0.176 and the incidence of rural poverty as 47.5 percent. 26. The overall incidence of poverty can be written as a weighted average of the poverty incidence in each group, where the weights are the population shares of the groups. The income gap measures (aggregate and proportionate) are similarly decomposable (with income share weights), but the Sen index, which uses the Gini coefficient to correct for inequality among the poor, is not. Because the Gini coefficient is not decomposable (see chapter 3 and "On the Decomposition of the Gini Coefficient" in appendix B), the Sen index cannot indicate the proportion of poverty accounted for by a particular group. DEFINITION AND MEASUREMENT OF POVERTY 127 are the characteristics of the poor that are different from those of the nonpoor? The profile of poverty in this section identifies the poor in terms of socioeconomic variables such as race, location, employment status, occupation, sector of employment, and education. It also indicates the extent of poverty accounted for by separate values of each variable (or characleristic}-made possible by the decomposability of the incidence-of- poverty measure. Since the household is the basic income-sharing unit, it appears more appropriate for policy purposes to describe the population in poverty in terms of households, rather than individuals. Accordingly, the population unit in the poverty profile is chosen to be the household, and poor households are those with per capita household incomes below M$25 per month. While the percentage of persons in poverty is 40.2 percent (table 4- 2), the percentage of households in poverty is 36.5 percent (table 4-3). This is because the poor have a larger average household size than the nonpoor [see point 8(3) below].27 Table 4-3 shows two distinct aspects of poverty. In column 2 the percentage distribution of poverty among the values of each variable is shown, which indicates concentrations of poverty. Column 4 shows which groups suffer from a particularly high incidence of poverty; these are so- called high-risk groups which may, in fact, account for only a small proportion of overall poverty. Clearly, both types of information are important in the design of policies to redress poverty. The following picture of the poor emerges from an examination of the characteristics of households and their heads in table 4-3. 1. The problem is overwhelmingly a Malay one, with 78.1 percent of poor households being Malay. There are six Malay households in poverty for every one Chinese. Malays, who constitute 55.4 percent of all households, are overrepresented among poor households by a factor of 1.41 (see column 5 on the relative incidence of poverty). More than half (51.4 percent) of Malay households suffer from poverty, while the incidence among Chinese is 14.7 percent, and among Indians, 24.8 percent. 2. Poverty is also overwhelmingly a rural phenomenon, with 87.7 percent of poor households living in rural areas.28 27. Whereas there is a negative correlation between average household size and per capita household income, there is a positive correlation between average household size and household income (see chapter 3). 28. Separate profiles of the urban and rural poor reveal that the ethnic distribution of urban poverty is quite different from that of rural poverty (see chapter 5). The Chinese form the rnost numerous group among the urban poor, but the incidence of urban poverty among Malays is almost twice that among Chinese (table 54). 128 INEQUALITY AND POVERTY IN MALAYSIA Table 4-3. Profile of Poverty at a Poverty Line of M$25 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among of incidence Selected among all poverty nonpoverty poverty of poverty characteristic households households households (percent) (2),/t() of household (1) (2) (3) (4) (5) Race Malay 55.4 78.1 42.4 51.4 1.41 Chinese 32.0 12.9 42.9 14.7 0.40 Indian 11.7 8.0 13.9 24.8 0.68 Other 0.9 1.0 0.8 40.3 1.11 Total 100.0 100.0 100.0 Location Urban 28.4 12.3 37.6 15.8 0.43 Rural 71.6 87.7 62.4 44.6 1.22 Total 100.0 100.0 100.0 State Johore 13 4 12.1 14.1 32.9 0.91) Kedah 11.3 15.1 9.2 48.6 1.34 Kelantan 9.5 17.0 5.2 65.2 1.79 Malacca 4.6 4.1 4.9 32.0 0.89 Negri Sembilan 4.7 4.2 5.1 32.1 0.89 Pahang 5.8 4.9 6.3 30.7 0.84 Penang 8.5 6.9 9.4 29.7 0.81 Perak 17.8 16.8 18 3 34.5 0.94 Perlis 1.5 2.4 1.0 58.9 1.60 Selangor 18.2 9.5 23.2 19.1 0.52 Trengganu 4.7 7.0 3 3 54.6 1.49 Total 100.0 100.0 100.0 Employment status of head Employer 2.7 0.4 4.0 5.1 0.51 Employee 51.8 38.2 59.3 26.3 0.74 Own-account worker 39.3 55.3 30.4 50.1 1.41 Housewife or houseworker 2.6 2.2 2.8 30.5 0.85 Unemployed 3.6 3.9 3.5 38.0 1.08 Total 100.0 100.0 100.0 Occupation of heada Professional and technical 5.7 1.1 8,2 6.7 0.19 Administrative and managerial 3.3 0.4 4.9 4.4 0.12 Clencal and related 4.0 0.3 6.1 2.7 0.08 DEFINITION AND MEASUREMENT OF POVERTY 129 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among of incidence ';elected among all poverty nonpoverty poverty of poverty characteristic households households households (percent) (2)1(1) oJ household (1) (2) (3) (4) (5) Sales 11.5 6.4 14.3 20.0 0.56 Service 8.3 3.4 10.9 14.9 0.41 Farmers 27.6 47.9 16.4 61.9 1.74 Farm laborers 21.6 29.5 17.3 48.6 1.37 Production 18.0 11.0 21.9 21.9 0.61 Total 100.0 100.0 100.0 Sector of employment of headb Agricilture 24.1 41.7 14.4 61.5 1.73 Agricultural products 25.7 33.4 21.4 46.2 1.30 Mining and quarrying 1.8 0.9 2.3 18.1 0.50 Manufacturing 8.6 5.2 10.4 21.8 0.60 Construction 3.2 2.0 4.0 21.5 0.63 Public utilities 1.6 1.0 2.0 21.0 0.63 Commerce 12.6 7.2 15.6 20.2 0.57 Transport and communications 5.5 3.3 6.7 21.2 0.60 Services 16.9 5.3 23.2 11.1 0.31 Total 100.0 100.0 100.0 Education ol head None 32.1 43.2 25.7 49.0 1.35 Some primary 33.1 35.6 31.7 39.1 1.08 Completed primary 20.4 18.4 21.6 32.8 0.90 Lower secondary (forms 1-111) 6.7 2.1 9.3 11.7 0.31 Some upper second- ary (forms IV-V) 3.0 0.4 4.4 5.2 0.13 School certificate or higher 4.7 0.3 7.3 2.1 0.06 Total 100.0 100.0 100.0 Sex of head Male 81.7 77.5 84.2 34.6 0.95 Female 18.3 22.5 15.8 44.9 1.23 Total 100.0 100.0 100.0 (Table continues on the following page.) 130 INEQUALITY AND POVERTY IN MALAYSIA Table 4-3 (continued). Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among of incidence Selected among all poverty nonpoverty povert)' of poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Age oJ'head Under 20 1.5 1.3 1.6 31.5 0.87 20-29 15.3 11.5 17.5 27.4 0.75 30-39 25.2 26.7 24.4 38.5 1.06 40-49 22.9 25.4 21.5 40.4 1.11 50-59 19.1 18.0 19.7 34.3 0.94 60+ 16.0 17.1 15.3 39.0 1.07 Total 100.0 100.0 100.0 Household size 1 8.9 5.7 10.7 23 5 0.64 2 9.5 6.1 11.6 23.2 0.64 3 11.7 9.4 12.9 29.5 0.80 4 13.1 11.4 14.0 31.8 0.87 5 12.6 13.8 11.9 40.0 1.10 6 12.0 13.8 11.0 41.8 1.15 7 9.8 12.5 8.2 46.7 1.28 8 7.9 9.9 6.8 45.6 1.25 9 5.5 6.1 5.2 40.3 1.11 10+ 9.0 11.3 7.7 45 7 1.26 Total 100.0 100.0 100.0 Number oJ'children under age /5 0 24.7 14.0 30.9 20.7 0 57 1 17.0 14.3 18.5 30.7 0.84 2 15.5 14.9 15.8 35.2 0.96 3 13.6 15.5 12.5 41.5 1.14 4 11.5 15.3 9.3 48.4 1.33 5+ 17.7 26.0 13.0 53.5 1.47 Total 100.0 100.0 100.0 Number oj income recipients 0 1.3 3.4 0.0 99.0 2.62 1 57.8 66.5 52.9 41.9 1.15 2 26.4 22.5 28.6 31.1 0.85 3 9.1 5.6 11.1 22.4 0.62 4+ 5.4 2.0 7.4 13.3 0.37 Total 100.0 100.0 100.0 Note The poverty line isdefined at a percapita household income ofM$25 per month; 36.5 percent of all households fall below this line. a. See note to table 6-11 for a definition of this one-digit occupation classification. b. See note to table 6-9 for a definition of this one-digit employment sector classification DEFINITION AND MEASUREMENT OF POVERTY 131 3. The four northern states of Kedah, Kelantan, Perlis, and Trengganu stand out as having above average incidences of poverty. Together they account for 41.5 percent of poverty households, but for only 27.0 percent of all households. 4. Of poor households, 93.5 percent are headed by employees or own- account (self-employed) workers.29 The incidence of poverty among households whose heads are own-account workers is 50.1 percent, which is higher than that among households whose heads are employees (26.3 percent). The unemployment rate among heads of poverty households is a mere 3.9 percent, and the rate of poverty among households with unemployed heads is 38.0 percent.30 5. Farmers head 47.9 percent of poor households, and farm laborers, 29.5 percent. The incidence of poverty among households headed by farmers and farm laborers is 61.9 percent and 48.6 percent, respectively. The higher incidence of poverty among households headed by farmers reflects something of a dual economy in the rural sector. The category of farm laborers includes both relatively well-paid estate workers, who form a significant proportion of the rural labor force, as well as casual and other laborers. The calegory of farmers includes all peasants and smallholders. 6. The incidence of poverty is well above average among households with heads in the agricultural sector (61.5 percent) and in the agricultural products sector (46.2 percent). These two sectors account for three-quarters (75.1 percent) of poverty households but less than half (49.8 percent) of all households.3' 7. Persons whose schooling does not extend beyond primary school head 97.2 percent of all poverty households. Of these, 43.2 percent have had no education at all. There is a strong negative correlation between education and poverty incidence, and the decline in incidence is particularly marked for households whose heads have acquired even some secondary education. Of households whose heads have received some upper secondary 29. For definitions of the various employment status categories, see the appendix to chapter 2. 30. Thus the problem of poverty needs to be distinguished from the problem of un- employment. A policy of absorbing unemployed heads of households into the labor market will not make a significant dent in poverty. Of course, a general expansion of formal sector employment, which also absorbs subsistence sector own-account and family workers, could contribute appreciably to a reduction in poverty. :31. This is in accordance with the finding of 77.4 percent of poverty households headed by farmers or farm laborers. The breakdown by employment status of head of household includes categories (own-account workers and employees) that overlap with the occupations of farmer and farm laborer. The overlap orjoint distribution of categories must obviously be borne in mind when policy interpretations are attempted. 132 INEQUALITY AND POVERTY IN MALAYSIA education, only 5.2 percent are poor, and of those whose heads have received the school certificate, only 2.1 percent are poor. 8. The distribution and incidence of poverty as a function of houschold composition show the following features: (1) Households headed by females are somewhat more poverty-prone (44.9 percent) than those headed by males (34.6 percent). (2) The incidence of poverty does not show wide variation according to age of household head, but it is lowest for the 20-29 age group. (3) The incidence of poverty increases with household size up to seven-member households, after which the relation is unclear (owing to the effect of additional income recipients); but the incidence is above average for all size classes above five and below average for all lower size classes. A comparison of the percentage distribution of household size among poverty and nonpoverty households shows a larger average household size for the poor. (4) The incidence of household poverty increases steadily with the number of children under age fifteen (who are unlikely to be income recipients).32 (5) Almost all (99 percent) of the households with no income recipients are in poverty, and the incidence rate falls with the number of recipients. Among poor households, 66.5 percent have just one income recipient, and 22.5 percent have two. When several of the characteristics associated with high degrees of poverty are taken together, the chances of being poor can become extremely high. Thus, for example, a Malay farmer in rural Kelantan has a worse than three-fourths probability of being poor. In order to design policies and projects to help the poor selectively and with minimal leakage, it is necessary to identify smaller, more homogeneous groups such as these, with particularly high incidences of poverty (see chapter 5). Sensitivity of the Poverty Profile Before zeroing in on subgroups in poverty, it should be established that the picture of poverty in table 4-3 does not hinge crucially on the chosen poverty line of M$25 per capita household income per month. Accordingly, I consider two further poverty lines, at M$15 and M$33 per month, and conduct sensitivity analysis on the profile of poverty.33 This is partially 32. This relation might weaken if the per equivalent adult household income had been used instead of per capita household income to measure living standards. But the relation between poverty incidence and household size need not be affected by this very much. Because of the prevalence of the joint family system in Malaysia, larger households do not necessarily have a larger proportion of children. 33. This is an obvious but somewhat neglected exercise in the growing empirical literature on poverty. As attempts to estimate poverty in developing countries get under way and cross- DEFINITION AND MEASUREMENT OF POVERTY 133 possible through table 4-4, which shows the percentage of households falling below these poverty lines with a breakdown by location and racial group. The M$15 poverty line cuts off roughly the bottom 20 percent of the population, and may be said to identify the "very poor" in Malaysia. Since the redress of pover ty rule requires filling the poverty gap from the bottom upward, this poorest group may be of special interest. Some 15.5 percent of houselholds, containing 17.3 percent of individuals, receive a per capita houselhold income of less than M$15 per month. The average per capita income of these individuals is slightly less than M$ 10 per month, while that of the other 82.7 percent is M$58 per month. A transfer ofjust 1.8 percent of the income of this upper group can bring all the very poor up to the income level of M$15 per month. The M$33 poverty line corresponds to that of the second method used earlier in this chapter to define an absolute poverty line. The proportion of households falling below M$33 per month is 49.3 percent. 34 Thus when the poverty line is raised from M$25 to M$33, the incidence of poverty rises from 36.5 to 49.3 percent, and when it is lowered from M$25 to M$15, the incidence of poverty falls from 36.5 to 15.5 percent. Hence, the incidence of poverty is fairly sensitive to changes in the poverty line, showing an elasticity of 1. I for upward movement from M$25 and an elasticity of 1.4 for downward movement from M$25. The profile of poverty is much less sensitive to variations in the poverty line than is the incidence of poverty.35 As the poverty line is lowered from M$33 to M$25 and then to M$15, Malays account for 75.3, 78. 1, and 85.3 percent, respectively, of poverty households, while the reverse trend is in evidence for the Chinese, who account for a decreasing percentage of poverty households.36 As the poverty line is lowered, the rural concentra- tion of poverty also increases a little: 85.2 percent of M$33 poverty country comparisons begin to be made, it becomes important to establish the robustness of estimates through sensitivity analysis. In some countries it is possible that the estimates are highly sensitive to small variations in the poverty threshold. 34. This is the figu e for poverty incidence mentioned in the Third Malaysia Plan (TMP) (see Govemment of Malaysia, 1976, p. 160). In addition to this, the TMP(tables 9-1,9-2,9-3, and 9- 6) quotes detailed poverty incidence figures from my M$33 poverty profile (see Anand, 1974a, or table 4-6 below). The plan document, however, does not mention that these figures correspond to a poverty line of M$33 income per month. 35. It is the presence of bunching, or excessive inequality, in the lower portions of the income distribution that could cause profile characteristics and the incidence of poverty to move discontinuously as a function of the poverty line. Since the Sen index specifically incorporates inequality among the poor, it may be relatively immune to such Jumps. 36. These findings are consistent with the racial pattem of income distnbution depicted in figure 1-1. Table 4-4. Household Percentages in Four per Capita Household Income Classes by Racial Group and Location Percentage of Percentage of Per capita household Percentage of Percentage distribution among racial groups urban households rural households income class households in in income in income (MS per month) income class Malay Chinese Indian Other Total class class 0-15 15.5 85.3 7.8 5.4 1.5 100.0 5.0 19.6 Urban 1.4 42.6 34.7 21.6 1 1 100.0 Rural 14.1 89.6 5.1 3.8 1.5 100.0 0-25 36 5 78.1 12.9 8.0 1.0 100.0 15.8 44.6 Urban 4.5 374 41.9 19.6 1 1 100.0 Rural 32.0 83.9 8 8 6.3 1.0 100.0 0-33 49.3 73.0 16.9 9.3 0.8 100.0 25.5 58.6 Urban 7.3 34.5 47.0 17.7 0.8 100.0 Rural 42.0 79.6 11 7 7.9 0.8 100.0 185+ 5.0 21.5 55.8 16.9 5.8 100.0 11.4 2.6 Urban 3.2 14.9 61.9 17.3 5.9 100.0 Rural 1.8 33.2 44.9 16.3 5.6 100.0 All income classes (0+) 100.0 55.4 32.0 11.7 0.9 100.0 100.0 100.0 Urban 28.4 25.9 57.9 14.9 1.3 100.0 Rural 71.6 67.1 21.7 10.5 0 7 100.0 DEFINITION AND MEASUREMENT OF POVERTY 135 households reside in rural areas, compared with 87.7 and 91.0 percent, respectively, of M$25 and M$15 poverty households. It seems that racial and rural-urban features of poverty are not very sensitive to the choice of poverty line within a reasonably wide range, although they are accentuated as the poverty line is dropped. The distribution of poverty according to other characteristics also remains relatively stable as the poverty line rises from M$15 to M$33 (see tables 4-5 and 4-6)."7 Kedah, Kelantan, Perlis, and Trengganu remain the states worst affected by poverty. As the poverty line rises, the relative incidence of poverty in these states drops slightly but remains well above unity. Households whose heads are own-account workers continue to be overrepresented among the poor as the poverty line is raised, with the relative incidence dropping a little from 1.61 to 1.32. The relative incidence of poverty among households whose heads are employees increases (from 0.61 to 0.81) but remains well below unity. With respect to sectors of employment, the relative incidence among households with heads in agriculture declines from 2.12 to 1.56, while for those in agricultural products it remains virtually stationary around 1.30. By occupational category, the relative incidence of poverty for households headed by farmers declines from 2.07 to 1.58 as the poverty line is raised, while for those headed by farm laborers it oscillates from 1.34 to 1.37 to 1.32. Finally, the educational and demographic characteristics of poverty households display similarly small variations as the poverty line is altered. Thus the picture of poverty in table 4-3 does not seem very sensitive to the variations considered in the poverty line. One can remain fairly confident about the profile it depicts, and in subsequent discussion on poverty I adhere to the original line of M$25 per month. Such changes in emphasis as there are in the poverty profile suggest that households headed by farmers and those in agriculture form the hard core of the poor. This highlights the need to study rural poverty in greater detail-which is the subject of the next chapter. Appendix: A Profile of the Rich For comparativre purposes, it is interesting to look briefly at the top end of the per capita household income distribution. Defining rich households as those belonging in the top 5 percent of this distribution, that is, those 37. Separate rural and urban poverty profiles for the M$15 and MS33 poverty lines were presented in Anand (1974a); they are included as tables 5-7, 5-8. 5-9. and 5-10 in the next chapter. 136 INEQUALITY AND POVERTY IN MALAYSIA Table 4-5. Profile of Poverty at a Poverty Line of M$15 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among of' inciadence Selected among all poverty nonpoverty poverty of poverty characteristic households households households (percent) (2)/'(1 ) oJ'household (1) (2) (3) (4) (5) Race Malay 55.4 85.3 49.9 23.9 1.54 Chinese 32.0 7 8 36.4 3.8 0.24 Indian 11.7 5.4 12.9 7.2 0.46 Other 0.9 1.5 0.8 25.3 1.67 Total 100.0 100.0 100.0 Location Urban 28.4 9.0 32.0 5.0 0.32 Rural 71.6 91.0 68.0 19.6 1.27 Total 100.0 100.0 100.0 State Johore 13.4 11 7 13.7 13.5 0.87 Kedah 11.3 15 8 10.5 21.7 1.40 Kelantan 9.5 21.4 7.4 34.8 2.25 Malacca 4.6 3.4 4.8 11.4 0. 74 Negri Sembilan 4 7 3.9 4.9 12.7 0.83 Pahang 5.8 4.6 6.0 12.4 0.79 Penang 8.5 5.1 9.1 9.3 0.60 Perak 17 8 15.5 18.2 13.6 0.87 Perlis 1.5 3.0 1.2 31.1 2.00 Selangor 18.2 7.1 20.2 6.0 0.39 Trengganu 4.7 8.5 4.0 28.2 1.31 Total 100.0 100.0 100.0 100.0 Employment status oJ'head Employer 2.7 0.3 3.1 1.8 0.11 Employee 518 31.6 55.2 8.7 0.61 Own-account worker 39.3 63.1 35 3 23.1 1.61 Housewife or houseworker 2.6 1.7 2.8 9.1 0.65 Unemployed 3.6 3.3 3.6 13.2 0.92 Total 100.0 100.0 100.0 Occupation of'head Professsional and technical 5.7 0.6 6.5 1.6 0.11 Administrative and managerial 3.3 0.1 3.9 0.5 0.03 Clerical and related 4.0 0.1 4.7 0.2 0.03 DEFINITION AND MEASUREMENT OF POVERTY 137 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among oj incidence ielected among all poverty nonpovertv poverty of poverty characteristic households households households (percent) (2)/(1) of household (1) (2) (3) (4) (5) Sales 11 5 4.6 12.7 6.0 0.40 Serviee 8.3 2.3 9.3 4 0 0.28 Farmers 27.6 57.0 22.5 30 6 2.07 Farm laborers 21.6 28.9 20.4 19.8 1.34 Production 18.0 6.4 20.0 5.3 0.36 Total 100 0 100.0 100.0 Sector of employment of head Agriculture 24.1 51.1 19.5 30.9 2 12 Agricultural products 25.7 33.5 24.4 18.9 1.30 Minrig and quarrying 1.8 0.3 2.1 2.4 0 17 Manufacturing 8 6 3.5 9.4 6.0 0 41 Construction 3 2 1.0 3.6 4.3 0.31 Public utilities 1.6 0.2 1.9 1.5 0.13 Commerce 12.6 5.1 13.9 5 9 0 41 Transport and communications 5.5 2.2 6.0 5.7 0.40 Services 16.9 3.1 19.2 2.7 0.18 Total 100.0 100.0 100.0 Education oJ head None 32.1 48.2 29.1 23.2 1.50 Some primary 33 1 34.0 33.0 15.8 1.03 Completed primary 20.4 15.1 21.4 11.4 0.74 Lower secondary (forms 1-Ill) 6.7 1.9 7.5 4.5 0.28 Some upper second- ary (forms IV-V) 3.0 0.4 3 5 2.0 0.13 School certifi- cate or higher 4.7 0.4 5.5 1.2 0.09 Total 100.0 100.0 100.0 Se., of head Male 81.7 73.5 83 2 13.9 0 90 Fernale 18.3 26.5 16.8 22.5 1.45 Tol al 100.0 100.0 100.0 (Table continues on the jollowing paye.) 138 INEQUALITY AND POVERTY IN MALAYSIA Table 4-5 (continued). Percentage Percentage Percentage distribution distribution Incidence Relarive distribution among among of incidence Selected among all poverty nonpoverty poverty oj'poveriv characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Age oJ head Under 20 1.5 1.4 1.5 14.9 0 93 20-29 15.3 11.6 16.0 11.6 0 76 30-39 25 2 28.4 24.7 17.4 1.13 40-49 22.9 25 2 22.5 17.0 1.10 50-59 19.1 17.7 19.3 14 3 0 93 60+ 16.0 15.7 16.0 15.2 0.98 'Total 100 0 100.0 100.0 Household size 1 8.9 0.7 10.4 1 3 0.08 2 95 10.5 9.4 17.0 1.11 3 11.7 96 12.1 127 0.82 4 13.1 7.1 14 2 8.4 0 54 5 12.6 15.5 12.0 19.1 1.23 6 12.0 17.0 11.1 21.9 1.42 7 9 8 11.8 9.4 18.7 1.20 8 7.9 11.0 7.3 217 1.39 9 5.5 6.6 5.3 19.6 1.20 10+ 9.0 10.2 8.8 17 6 1.13 Total 100.0 100.0 100.0 Number oj children under age 15 0 24.7 10.5 27.3 6.6 0.43 1 17.0 14.3 17.5 13 1 0.84 2 15.5 13 2 15.9 13 2 0 85 3 13.6 18.4 12.7 21.0 1.35 4 115 17.0 10.5 22.8 1.40 5+ 17.7 26.6 16.1 23.3 1.50 Total 100 0 100.0 100 0 Numberq o incoine recipients 0 1.3 8.0 0.0 98.7 6.15 i 57.8 67 7 56 1 18.1 1.17 2 264 18.9 27.8 11.1 0.72 3 9 1 4.0 10.0 6 7 0.44 4+ 5.4 14 6.1 4.1 0.26 Total 100.0 100.0 100.0 Note: The poverty line isdefined at a percapita household income ofM$15 per month; 1I 5 percent of all households fall below this line. DEFINITION AND MEASUREMENT OF POVERTY 139 Table 4-6. Profile of Poverty at a Poverty Line of M$33 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among oJ incidence Selected among all poverty nonpoverty poverty oJ poverti characteristic households households households (percent) (2)1(1) oJ household (1) (2) (3) (4) (5) Race Malay 55.4 73.0 38.4 64 8 1.32 Chinese 32.0 16.9 46 6 26.0 0.53 Indian 11.7 9.3 14.0 39.2 0.79 Other 0.9 0.8 1.0 44.8 0 89 Total 100.0 100 0 100.0 Locaton Urban 28.4 14.8 41.6 25.5 0.52 Rural 71.6 85.2 58.4 58.6 1.19 Total 100.0 100.0 100.0 State Johore 13.4 12.4 14.3 45.7 0 93 Kedah 11.3 14.6 8.2 63.2 1.29 Kelaritan 9.5 14.8 4.5 76.1 1.56 Malaxza 4.6 4.2 5.0 44.9 0 91 Negri Sembilan 4.7 4.3 5 1 44.8 0.91 Pahang 5.8 5.1 6.5 43.2 0.88 Penang 8.5 7.5 9 4 43.7 0.88 Perak 17.8 17.5 18.0 48 6 0.98 Perlis. 1.5 2.3 0.8 73 9 1.53 Selangor 18.2 10.8 25.4 29 2 0.59 Trengganu 4.7 6.5 2.8 68.9 1.38 Total 100.0 100.0 100.0 Employment status of head Employer 2.7 0.5 4.8 8.8 0.19 Employee 51.8 41.8 61.2 39.1 0.81 Own-account worker 39.3 51.9 27.4 64.0 1.32 Housewife or houseworker 2.6 2.3 2.9 42.8 0.88 Unemployed 3.6 3.5 3.7 47.2 0 97 Total 100.0 100.0 100.0 Occupation of head Professional and technical 5.7 1.2 9.8 10.6 0.21 Administrative and managerial 3.3 0.7 5.8 9.5 0.21 Clerical and related 4.0 0.9 6.9 10.7 0.23 (Table continues on the following page.) 140 INEQUALITY AND POVERTY IN MALAYSIA Table 4-6 (continued). Percentage Percentage Percentage distribution distribution Incidence Rela/ive distribution among among of incide nce Selected among all poverty nonpoverty poverty of poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sales 11 5 7.2 15.5 30.5 0.63 Service 8.3 4.6 11.7 27.0 0.55 Farmers 27.6 43.5 12.7 76.4 1.58 Farm laborers 21.6 28.5 15.2 63.8 1 32 Production 18.0 13.4 22.4 36.0 0.74 Total 100.0 100.0 100.0 Sector of employment of head Agriculture 24.1 37.5 11.5 75.4 1.56 Agricultural products 25.7 32.7 19.1 61.7 1.27 Mining and quarrying 1.8 1 3 2.3 34.0 0.72 Manufacturing 8.6 5.7 11.2 32.3 0.66 Construction 3.2 2.4 4.0 36.6 0.75 Public utilities 1.6 1.3 2.0 37.0 0.81 Commerce 12.6 7 9 17.1 30.3 0.63 Transport and communications 5.5 4.2 6.8 36.6 0.76 Services 16.9 7.0 26.0 20.3 0.41 Total 100.0 100.0 100.0 Education of head None 32.1 40.6 23.8 62.3 1.26 Some primary 33.1 36.2 30.2 53.7 1.09 Completed primary 20.4 19.8 21.0 47.7 0.97 Lower secondary (forms 1-111) 6.7 2.6 10.6 19.0 0.39 Some upper second- ary (forms IV-V) 3.0 0.5 5.4 7.6 0.17 School certificate or higher 4.7 0.3 9.0 3.3 0.06 Total 100.0 100.0 100.0 Sex of head Male 81.7 79.2 84.2 47.7 0.97 Female 18.3 20.8 15.8 56.1 1.14 Total 100.0 100.0 100.0 DEFINITION AND MEASUREMENT OF POVERTY 141 Percentage Percentage Percentage distribution distribution Incidence Relative distribution among among oJ incidence Selected among all poverty nonpoverty poverty of poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Age of head Under 20 1.5 1 2 1.8 38.3 0.80 20-29 15.3 11.8 18.7 379 0.77 30-39 25.2 26.6 23.9 51.9 1.06 40-49 22.9 25.4 20.5 54.5 1.11 50-59 19.1 18.1 19.9 46.9 0.95 60+ 16.0 16.9 15.2 51.9 1.06 Total 100.0 100.0 100.0 Household size 1 8.9 4.3 13.4 23.5 0.48 2 9.5 7.8 11.3 40.2 0.82 3 11.7 9.3 13.9 39.3 0.79 4 13.1 11.6 14.5 43.8 0.89 5 12.6 14.5 10.7 56.7 1.15 6 12.0 13.8 10.3 56.5 1.15 7 9.8 10.4 9.2 52.3 1.06 8 7.9 10.4 5.5 65.0 1.32 9 5.5 7.3 3.7 65.5 1.33 10+ 9.0 10.6 7.5 57.8 1.18 Total 100.0 100.0 100.0 Number of children under age 15 0 24.7 14.4 34.8 28.6 0.58 1 17.0 14.8 19.1 43.0 0 87 2 15.5 15.2 15.3 40.2 0.99 3 13.6 15.8 11.4 57.3 1 16 4 11.5 14.5 8.6 62.2 1.26 5+ 17.7 25.3 10.3 70.4 1.43 Total 100.0 100.0 100.0 Number of income recipients 0 1.3 2.5 0.0 99.0 1.92 1 57.8 64.0 51.9 54.4 1.11 2 26.4 24.2 28.6 45.1 0.92 3 9.1 6.7 11.4 36.0 0.74 4 + 5.4 2.6 8.1 23.9 0.48 Total 100.0 100.0 100.0 Note: The poverty line isdefined at a percapita household income of M$33 per month,49.3 percent of all households fall below this line. 142 INEQUALITY AND POVERTY IN MALAYSIA with per capita household income above M$185 per month, certain characteristics of "richness" are apparent from tablc 4-4. Of the 5.0 percent rich households, 3.2 percent reside in urban areas and 1.8 percent in rural areas. As many as 11.4 percent of urban households, but only 2.6 percent of rural households, are rich. Only 2.0 percent of Malay households are rich, compared with 8.8 percent of Chinese, 7.3 percent of Indian, and 33.5 percent of other races.38 Of rich households 55.8 percent are Chinese, compared with 21.5 percent Malay, 16.9 percent Indian, and 5.8 percent others. Further characteristics (not shown in table 4-4) of the top 5 percent of households are:39 1. Almost three-quarters (73.8 percent) of the rich are concentrated in the four states of Johore, Penang, Perak, and Selangor, with Selangor itself accounting for 37.7 percent. 2. Households whose heads are employers are overrepresented among the rich by a factor of 4.26 (that is, they show a relative incidence of richness of 4.26), and those whose heads are own-account workers are under- represented (with a relative incidence of 0.33). Among rich households 2.3 percent have heads who are unemployed, and 3.4 percent of households with unemployed heads are rich. 3. With respect to sector of employment, heads of rich households are highly concentrated in services (46.7 percent), commerce (20.3 percent), and manufacturing (12.0 percent). Few of them are to be found in agriculture (3.1 percent) or agricultural products (5.4 percent), which also accords with the occupational finding that few are farmers (2.3 percent) or farm laborers (1.2 percent). Indeed, an extremely small percentage of households whose heads are farmers (0.5 percent) or farm laborers (0.3 percent) are rich.40 In contrast, the professional and technical and the administrative and managerial categories each have an incidence of richness above 30 percent. 4. Education of the household head seems positively correlated with the household's being rich. Only 1.2 percent of households whose heads have no education belong to the M$185 + group, compared with 45.5 percent of 38. This last fact, togetherwith 40.3 percent of others" beingpoor(table 4-3), confirms the wide income inequality within this racial group noted in chapter 3. 39. A detailed profile of the rich is contained in Anand (1 974a), pp. 29-37, which also has a disaggregation according to separate urban and rural profiles. 40. The incidence of richness among urban households whose heads are farmers is 2.8 percent, whereas that among rural households whose heads are farmers is 0.4 percent. This is to some extent explained by different cropping patterns in urban and rural areas, with urban farmers probably growing high-value perishables such as fruits and vegetables for nearby markets. DEFINITION AND MEASUREMENT OF POVERTY 143 those whose heads have a school certificate (form V) or more.4' The chances of being rich improve significantly for households whose heads have completed even some secondary education. 5. Other, demographic features of richness stand in mirror image to those associated with poverty. 41. The strong relation persistseven when rural areas areconsidered separately from urban areas' 36.9 percent of rural households whose heads have a school certificate or higher education are rich (51.1 percent of urban households). While I have partially tested for the effect ol education on inicome in urban areas (see chapter 7), no such test has been conducted for rura I areas. The evidence here, however, suggests that even in rural areas living standards are positively correlated with education. 5 Subgroups in Poverty WITH 87.7 PERCENT OF POOR HOUSEHOLDS located in rural areas, it seems probable that rural development will form a major instrument for the redress of poverty in Malaysia. The possibilities for absorption of the r ural poor into the modern industrial sector are obviously limited by its small absolute base and growth, whereas several million acres of land are still available for cultivation and development in the country. Chapter 8 contains a general analysis of alternative policy instruments to alleviate poverty in Malaysia, including tax, price, and transfer policies. But in the present chapter I anticipate the conclusions of that discussion to a large extent, and assume that rural development through land (re-)settlement and the further development of existing agriculture will form a major means for raising the income levels of the poor. Hence it is clear that a detailed statistical picture of rural poverty is needed. This chapter begins by investigating the broad characteristics of rural poverty at the one-digit level of detail for the variables of industry a-nd occupation.' From the general features of this poverty map, I then zero in on specific subgroups that display especially high rates of poverty. The purpose is not merely better diagnosis of the problem, but more efficient design of rural development policies-which reduce leakages to the nonpoor. Improvement in the targeting of such policies requires the identification of smaller and more homogeneous groups than those shown in the one-digit profile. This is done here by two methods: (1) increasing the selected level of detail to characterize relevant variables (when this is possible), and (2) cross-classifying these variables to obtain a multidimen- sional profile of the poor. The rural poor in Malaysia are thus disag- gregated into operationally and analytically meaningful subgroups with selected values of the industrial sector and occupational variables cross- classified at the two-digit level.2 From the two-digit matrix which is 1. See notes to tables 6-9 and 6-11. 2. A caveat must be entered here. The PES classification is determined by the occupation or industry which accounts for the bulk of the individual's income, although the individual may 144 SUBGROUPS IN POVERTY 145 generated, I select the five largest subgroups in poverty, which together account for as many as 79 percent of poor rural households with known occupations. These subgroups are further disaggregated regionally to determine their concentration by state. The five subgroups are paddy farmers, laborers on paddy and mixed- agriculture farms, rubber smallholders, workers on rubber estates and smallholdings, and fishermen. The economic problems of these subgroups, and measures to raise their productivity and income, are discussed case by case. This allows the identification of some major components of rural development policies and projects in Malaysia. The chapter ends with a brief cliscussion of other rural subgroups in poverty. Urban poverty is examined separately in the appendix to this chapter. Although urban poverty accounts for a relatively small proportion of overall poverty (12.3 percent of all poor households are urban), its dimensions are quite different from those of rural poverty and con- sequently antipoverty policies are likely to be different in the two locations. Furthermore, given the attention recently placed on the growing problem of urban poverty in developing countries (see, for example, McNamara, 1975), a separate discussion of this question seems warranted in the Malaysian context. The emphasis placed in this chapter on particular subgroups takes for granted the importance of the microapproach to the alleviation of poverty. Such an approach helps us not only to understand better the causes and circumstances of poverty, but also to suggest policy packages in specific cases (such as land settlement and certain types of agricultural devellopment). In a recent survey paper on distributional issues in development planning, Michael Bruno (1977) concludes: With all my natural bias in favour of a general equilibrium approach I believe some of the most vexing problems in this area had best be studied by concentrating on specific social or regional groups in which poverty is most concentrated. This of necessity entails a partial equilibrium or micro approach to the problem . . . With our present knowledge and resource limitations this may be the avenue of highest marginal social p!roduct. Rural Poverty T'he broad characteristics of rural poverty are examined in this section at the one-digit level of detail for the industrial sector and occupational have income from ot her occupations or industries. This can affect policy conclusions premised on a classification by a single occupation or industry. 146 INEQUALITY AND POVERTY IN MALAYSIA variables. Since rural poverty accounts for the bulk of overall poverty, considering the rural poor in isolation does not give a picture very different from that of overall poverty. Table 5-1 presents a full profile of rural poverty at the one-digit level.3 The incidence of poverty in rural Malaysia (44.6 percent) is higher than in the country as a whole (36.5 percent), and it is higher for each racial group considered separately. Malays constitute an even larger fraction of rural poor households (83.9 percent) than they do of all poor households (78.1 percent). In this sense, rural poverty seems largely synonymous, with Malay poverty, and government policies to redress rural poverty will also help to narrow racial income imbalances slightly (see chapter 8). Households whose heads are farmers or farm laborers account for 82.8 percent of the rural poor, with the former accounting for 51.6 percent and the latter for 31.2 percent. Households headed by farmers show a higher poverty incidence (62.5 percent) than those headed by farm laborers (49.3 percent). This phenomenon is probably attributable to the dual economy noted earlier within the rural sector.4 Farm laborers are both those who work on paddy and rubber smallholdings and vegetabte farms, and those employed in the plantation sector, which is largely unionized and relatively well paid. Farmers include not merely owner-operators but also tenant farmers (sharecroppers), who pay half (bagi dua) to two-thirds (bagi tiga) of their gross income as rent-a fact often invoked to explain their poverty. Furthermore, landlessness is not widespread, and those employed at relatively low wages on smallholdings constitute a small fraction of all farm laborers.5 Hence, the farm laborer category is less prone to poverty than the farmer category.6 3. Table 5-10 in the appendix to this chapter presents the rural poverty profile corresponding to the M$33 per month poverty line implicit in the Third Malaysia Plan (T.MP). 4. Another possible explanation would be that farmers suffered temporary poverty owing to adverse price movements in 1970, whereas the majority of farm laborers (estate workers) were shielded against price fluctuations through their unions' sliding-scale agreements with the estates. This suggestion must be rejected, however, because 1970 was not an abnormal year for rubber prices, and paddy prices continued to be supported at a stable level. 5. In theMuda project area, for instance, less than iOpercent ofthe work forceconsisted of landless laborers. A possible explanation for the limited extent of landlessness in Malaysia may be found in Muslim inheritance law. Under this law, land tends to be equally divided among the heirs, in contrast to the practice ofprimogeniture, and often results in uneconomic holdings and endemic poverty. 6. In India, however, landlessness is prevalent, and agricultural laborers show the highest incidence of poverty in rural areas (Dandekar and Rath, 1971). SUBGROUPS IN POVERTY 147 Interestingly, this finding does not hold for each racial group separately. For the: Chinese, the incidence of poverty is higher among households headed by farm laborers (22.5 percent) than among those headed by (Text continues on page 150.) Table 5-1. Profile of Rural Poverty at a Poverty Line of M$25 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence Selected among all among rural among rural of rural of rural characteristic rural poverty nonpoverty poverty poverty of household households households households (percent) (2)1(1) (1) (2) (3) (4) (5) Race Malay 67.1 83.9 53.5 55.8 1.25 Chinese 21.7 8.8 32 2 18.1 0.41 Indian 10.5 6.3 13.8 27 0 0.60 Other 0.7 1.0 0.5 59.7 1.43 Total 100.0 100.0 100.0 State Johore 13.8 11.9 15.2 38.6 0.86 Kedah 13.6 16.3 11.5 53.3 1.20 Kelantan 11.5 17.7 6.5 68.6 1.54 Malacca 5.2 4.3 5.9 37.0 0.83 Negri Sembilan 5.6 4.4 6.6 35 0 0.79 Pahang 6.4 5.3 7.2 37.2 0.83 Penang 6.1 5.5 6.6 40.4 0.90 Perak 17.5 16.7 18.1 42.6 0.95 Perlis 2.1 2.8 1.6 58.7 1.33 Selangor 13.1 8 1 17.2 27.5 0.62 Trengganu 5.1 7.0 3.6 61.2 1.37 Total 100.0 100.0 100.0 Empkoyment status oJ'head Employer 1.6 0.4 2.6 10.3 0 25 Employee 47.8 36.3 56.8 33.3 0.76 Own-account worker 45.6 58.1 35.8 55.9 1.27 Housewife or houseworker 2.1 1.8 2.2 38.9 0.86 Unemployed 2.9 3.4 2.6 50.6 1.17 Total 100.0 100.0 100 0 (Table continues on the Jollowing page.) 148 INEQUALITY AND POVERTY IN MALAYSIA Table 5-1 (continued). Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among rural among rural of rural of rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (2)/(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 4.1 0.9 6.6 9.5 0.22 Administrative and managerial 2.5 0.4 4.1 7.4 0.16 Clerical and related 2.1 0.2 3.6 3.9 0.10 Sales 7.9 5.1 10.2 28.4 0.65 Service 5.5 2 1 8.2 16.8 0.38 Farmers 36.4 51.6 24.4 62.5 1.42 Farm laborers 27.8 31.2 25.2 49 3 1.12 Production 13.7 8.5 17.7 27.4 0.62 Total 100.0 100.0 100.0 Sector of employment of head Agriculture 31.1 44.6 20.6 62.7 1.43 Agricultural products 34.1 36.7 32.1 46.9 1.08 Mining and quarrying 1.9 0.8 2.7 19.3 0.42 Manufacturing 5.8 4.2 7.1 31.7 0.72 Construction 2.2 1.4 2.8 28.3 0.64 Public utilities 1.1 0.6 1.5 23 1 0.55 Commerce 8.7 5.8 11.0 29.2 0.67 Transport and communi- cations 3.9 2.4 5.0 27.3 0.62 Services 11.2 3 5 17.2 13.6 0.31 Total 100.0 100.0 100.0 Education of head None 35.0 43.7 280 55.6 1.25 Some primary 34.6 35.5 33.9 45.7 1.03 Completed primary 21.5 18 7 23.8 38.8 0.87 Lower second- ary (forms 1-111) 4.7 1.7 7.0 16.5 0.36 Some upper second- ary (forms IV-V) 1.6 0.2 2.7 7.0 0.13 School certificate or higher 2.6 0.2 4.6 2.8 0.08 Total 100.0 100.0 100.0 SUBGROUPS IN POVERTY 149 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among rural among rural of rural of rural Selected rural. poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sex of head Male 82.5 78.8 85.4 42.6 0 96 lemale 17.5 21.2 14.6 53.8 1.21 Total 100.0 100.0 100.0 Age of head Under 20 1.4 1.2 1.5 39.1 0.86 20-29 15.2 12.0 17.7 35.4 0.79 30-39 2'.1 26.5 23.9 47.1 1:06 40-49 22.6 24.8 21.0 48.8 1.10 50-59 19.0 18.2 19.7 42.6 0.96 60+ 16.7 17.3 16.2 46.1 1 04 Total 10(.0 100.0 100.0 Household size 1 '3.0 5.8 9.8 32.3 0.73 2 9 9 6.3 12.7 28.6 0.64 3 12.1 9.9 13.8 36.7 0.82 4 13.3 12.0 14.4 40.1 0.90 5 13.0 14.1 12.1 48.4 1.08 6 12.6 13.9 11.5 49.2 1.10 7 9.9 12.5 7.9 56.2 1.26 8 7.9 9.6 6.5 54.3 1.22 9 5.3 6.0 4.7 50.6 1.13 10+ 8.0 9.9 6.6 54.8 1 24 Total 100.0 100.0 100.0 Numbe, of children under age 15 0 22.8 14.4 29.6 28.1 0.63 1 17.5 15.0 19.5 38.3 0.86 2 15.8 15.6 15.8 44.3 0.99 3 13.8 15.6 12.4 50.3 1.13 4 12.0 14.9 9.7 55.4 1.24 5+ 18.1 24.5 13.0 60.3 1.35 Total 1)0.0 100.0 100.0 Number of income recipients 0 1.3 2.8 0.0 98.7 2.15 I 359.7 67.2 53.7 50.2 1.13 2 :26.9 23.0 30.2 38.0 0.86 3 8.1 5.3 10.3 29.4 0 65 4+ 4.0 1.7 5.8 19.5 0.43 Total 100.0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M$25 per month, 44.6 percent of all rural households fall below this line. 150 INEQUALITY AND POVERTY IN MALAYSIA farmers (18.5 percent), although the reverse holds for the other communities. These figures probably reflect the economic situation in the predominantly Chinese "New Villages,", in which a majority of the residents are thought to be underemployed laborers engaged on smallhold- ings or nonunionized estates where wages are low.' The obverse of this situation for farm laborers is the fact that Chinese farmers have larger landholdings than Malay farmers. In the rubber sector, for which some data are available, the average Chinese smallholding is almost twice as large as the average Malay smallholding. In accordance with the results for the occupational variable, households whose heads are in the one-digit industrial sectors of agriculture and agricultural products account for 81.3 percent of the rural poor. House- holds headed by those in agriculture account for 44.6 percent of the rural poor and show a poverty rate of 62.7 percent, while households headed by those in agricultural products account for 36.7 percent of the rural poor and show a poverty rate of 46.9 percent. The importance of these sectors in accounting for relatively high rates of poverty survives the breakdown by racial group (and, in fact, the breakdown by location; see the appendix to this chapter, "Urban Poverty"). Given that the industrial sector and occupational variables are both economically and operationally relevant in identifying target groups, the one-digit categories agriculture and agricul- tural products, and -farmers and farm laborers, should be disaggregated further, especially since they each account for more than 80 percent of rural poverty. This is done in the next section, which considers the occupational and industrial sector variables at the two-digit level and, further, cross- classifies them to obtain a still more precise definition of target groups. The breakdown of rural poverty by the employment status of the household head is consistent with the above findings. Households headed by own-account workers and employees make up 58.1 and 36.3 perocnt, respectively, of the rural poor and show poverty incidences of 55.9 and 33.3 percent, respectively. Households with unemployed heads account for a very small percentage (3.4 percent) of the rural poor and show a poverty incidence (50.6 percent) which is lower than for households headed by own- account workers. In fact, the phenomenon of open unemployment is insignificant (2.9 percent) among household heads in rural areas. Underemployment, both among households (in terms of low participation rates and number of income recipients) and among individuals within households, is much more important as a factor determining the degree of 7. These New Villages, now some 465 in number with a total population of about I million, were established in 1950-52 to contain the communist insurgency. Resettled with Chinese, the villages suffer from a severe land shortage, and many of the farm families earn a liveliliood through market gardening, the cultivation of tapioca, and the like. SUBGROUPS IN POVERTY 151 poverty. Thus, the incidence of poverty drops markedly with additional income recipients, from 98.7 percent for households with no recipients to 19.5 percent for households with four or more recipients. The incidence of poverty tends to rise, however, with household size.8 Together with the previous finding, this suggests that poverty incidence is likely to be negatively correlated with the participation rate of a household. A manifestation of the increase in poverty with the household dependency ratio is the fact that the poverty percentage rises with the number of children under age fifteen, from 28.1 percent for no children to 60.3 percent for five or more children. Other features of rural poverty remain largely unaltered in comparison with overall poverty.9 The four poorest states of Kedah, Kelantan, Perlis, and Trengganu account for 43.8 percent of rural poverty, compared with 41.5 percent of overall poverty. The educational characteristics of the rural poor show that 97.9 percent of rural poverty, compared with 97.2 percent of overall poverty, is concentrated in households where the head has primary education or less. The demographic characteristics of poverty, that is, those pertaining to age or sex of household head, are virtually the same in rural areas as in Peninsular Malaysia as a whole. The one-digit profile presented of the rural poor gives a broad indication of groups which account for a significant proportion of poverty and also suffer a high incidence. But these groups are too widely defined for the purpose of designing effective poverty action programs. Further targeting is needed within these categories by relevant variables. For example, farmer households, which account for more than half of rural poverty, need to be narrewed down at least by region and sector (or commodity) at the two- digit level. Such cross-classifications will allow subgroups such as paddy farmers in Trengganu and rubber smallholders in Kelantan to be singled out. The cross-tabulations will also permit estimates of poverty among subgroups traditionally thought to be poor, such as fishermen along the East Coast of Peninsular Malaysia and rubber farmers in Perak. Two-digit PES Subgroups In the following subsections the PES is used to isolate specific subgroups in poverty with a view to discussing policy measures to raise their incomes. 8. Among poverty households in rural areas the average number of income recipients is 1.36 and the average household size, 5.77. The comparable figures for all households in Malaysia are 1.63 and 5.36, respectively (tables 3-3 and 3-6). 9. The incidence of rural povcrty, however, is greater than that of overall (or urban) poverty in each category. owing to rural incomes being lower than overall (or urban) incomes and the degree of incquality being fairly similar in the two locations. 152 INEQUALITY AND POVERTY IN MALAYSIA To minimize leakages, groups selected should be homogeneous and show high rates of poverty. To keep the number of separate groups and policies down to a manageable level, the subgroups selected should account for a significant fraction of total poverty. First, the one-digit classification of rural poor is refined according to the industrial sector and occupational variables. Selected values of these variables are disaggregated to the two- digit level (the finest level of detail available in PEs) and then cross-classified to generate a matrix of rural poverty (table 5-2). Although this procedure does not allow estimates of poverty among all the subgroups in the population believed to be poor, it does isolate the major target groups in the rural sector by commodity and occupation." Table 5-2 shows the incidence of poverty, the number of households in poverty, and the total number of households for each occupational- industrial subgroup. The two-digit industrial groups are a disaggregation of the one-digit categories, agriculture and agricultural products; and the two-digit occupational groups are a disaggregation of the one-digit categories, farmers and farm laborers. 1 i Some of the subgroups defined by the cells of the matrix correspond to groups identified by the Economic Planning Unit (EPU) of the government of Malaysia in its "Survey of Rural Poverty."12 In that survey, EPU estimated poverty among six rural subgroups in 1970 using the one-digit figures contained in Anand (1974a),\ supplemented where possible by information from existing socioeconomic surveys of the Ministry of Agriculture. Now that I have computed a two- digit matrix of poverty, it can be compared with EPU'S estimates for some of these groups. According to table 5-2 the five largest subgroups in poverty are: Paddy smallholders and livestock and mixed-agriculture farmners (01 x 61) Laborers on paddy and livestock and mixed-agriculture farms (01 x 62) Rubber smallholders (11 x 61) Laborers on rubber estates and smallholdings (11 x 62) Fishermen (04 x 64). 10. Aziz (1975b), for example, intuitively singled out the following subgroups as poor: Chinese rubber and vegetable farmers and residents of the New Villages, Teochow farmers in mainland Penang, several fishing communities along the West Coast of Peninsular Malaysia, vegetable farmers in the Cameron Highlands, pineapple farmers in South Johore, and Trengganu fishermen. I1. The one-digit categories, agriculture and agricultural products, make up codes 01-09 and 11-19, respectively, of the PES two-digit industrial classification (see table 6-9). The one- digit categories, farmers and farm laborers, make up codes 61 and 62-69, respectively, of'the PES two-digit occupational classification (see table 6-1) 12. This survey (EPU, 1975) was prepared as a background paper to the Third Malaysia Plan. SUBGROUPS IN POVERTY 153 These five subgroups together account for as much as 79 percent of impoverished rural households whose heads have known occupations: 5,123 out of 6,481 households.'3 A few small subgroups in sectors such as coconut, oil palm, forestry, tea, and coffee can also be singled out by means of table 5-2. The economic problems of the five largest subgroups in poverty, located by state, and measures to raise their incomes, are discussed in the following subsections. Paddly Farmers The paddy farmer subgroup has probably attracted greatest attention in the context of Malaysian poverty. The relatively low incomes in paddy are revealed by the faci that whereas 20 percent of the working population is engaged in this subsector, only 5 percent of GDP originates there (Selvadurai, 1972a). Socioeconomic surveys of paddy farms in different parts of the country distinguish between specialized paddy growers and others.'4 According to Selvadurai (1972a), specialized paddy farms (those with rnore than three-quarters of the farm acreage in paddy) constitute about 50 percent of all farms cultivating paddy. The nonspecialized paddy farms are generally engaged in mixed agriculture (including fruits and vegetables), livestock, and miscellaneous crops. The two-digit commodity classification in PES does not distinguish be- tween purely paddy farmers and those who grow paddy along with other crops (table 5-2). For the combined category of paddy smallholders and livestock and mixed-agriculture farmers, the poverty incidence is an alarmning 65.8 percent-in other words, two out of three such farmers are poor.'5 The group is predominantly Malay (88.5 percent), with an ethnic composition and poverty incidence as follows: Malay Chinese Indian Other All races Incidence of poverty (percent) 69.8 24.8 37.9 87.3 65.8 Number of poor households 2,204 79 11 55 2,349 Total number of liouseholds 3,159 319 29 63 3,570 13. For some I,500 rural households in poverty, either the head is not in the labor force or the occupation is nol available. 14. Pureal (1971) contains a valuable description of several socioeconomic surveys of paddy households. Some of the important surveys are Selvadurai (1972b, 1972c, and 1975); Narkswasd, and Selvadurai (1968); Selvadura, and Ani (1969); Selvadural, Ani, and Nik Hassan (1969). 15. EPU (1975) estimated poverty among paddy cultivators separately from that among mixed-agriculture farmers. Its estimates for these grouns are 75 and 78 percent. respectively. Table 5-2. Incidence of Poverty by Two-digit Industrial Sector and Occupational Group Two-digit occupational groupa 61 62 63 64 Fishermen, hunters, and Subtotal of Code Farm Forestry related occupations Total of all number Two-digit industrial sector a Farmers laborers workers workers 61-64 occupations 01 Paddy, livestock, and mixed agriculture Incidence of poverty (percent) 65.8 63.9 _b 65.5 63.9 Number of poor households 2,349 327 0 - 2,676 2,710 Total number of households 3,570 512 1 0 4,083 4,238 02 Forestry Incidence of poverty (percent) _ b _b 33.3 _b 31.4 30.5 Number of poor households 0 0 20 2 22 36 Total number of households 1 4 60 5 70 118 04 Fishing Incidence of poverty (percent) _b - - 50.9 50.8 50.7 Number of poor households 0 - - 315 315 317 Total number of households I 0 0 619 620 625 11 Rubber Incidence of poverty (percent) 55.6 47.5 - - 50.9 47.3 Number of poor households 981 1,151 - - 2,132 2,196 Total number of households 1,764 2,423 0 0 4,187 4,641 12 Oil palm Incidence of poverty (percent) _b 23.5 - - 23.1 20.9 Number of poor households 1 44 - - 45 49 Total number of households 8 !87 0 0 195 235 13 Coconut, copra, and coconut oil Incidence of poverty (percent) 49.3 61.9 _ - 56.0 55.8 Number of poor households 68 99 - - 167 174 Total number of households 138 160 0 0 298 312 14 Tea Incidence of poverty (percent) - 15.4 - - i5.4 !0.0 Number of poor households - 2 - - 2 2 Total number of households 0 13 0 0 13 20 15 Coffee Incidence of poverty (percent) 44.4 60.0 - - 50.0 46.7 Number of poor households 8 6 - - 14 14 Total number of households 18 10 0 0 28 30 Subtotal of industries 01-15 Incidence of poverty (percent) 61.9 49.2 32.8 50.8 56.6 53.8 Number of poor households 3,407 1,629 20 317 5,373 5,498 Total number of households 5,500 3,309 61 624 9,494 10,219 Total of all industries Incidence of poverty (percent) 61.9 48.5 32.4 50.8 56.1 Number of poor households 3,423 1,767 23 318 5,531 Total number of households 5,529 3,641 71 626 9,867 - Not applicable. a. The two-digit occupational classification is from Department of Statistics (1971 a), and the two-digit industrial classification is from Department of Statistics (1971b). b. Denotes sample size too small for statistically valid inference. 156 INEQUALITY AND POVERTY IN MALAYSIA Many of the Chinese farmers are so-called market gardeners growing fruits and vegetables for market sales. The distribution by state of paddy smaliholders and mixed-agriculture farmers is shown in table 5-3. Although paddy is grown in all eleven states of Peninsular Malaysia, the northern states are the most important producers of this crop. Whereas for Peninsular Malaysia as a whole, 14 percent of cultivated land is under paddy, the corresponding percentages for the northern states of Perlis, Kedah, Kelantan, and Trengganu are 75, 36, 35, and 26 percent, respectively.'6 These four states account for 65 percent of the paddy land in Peninsular Malaysia and are sometimes called the rice bowl of the country. From table 5-3 it is clear that these states account for the bulk of poverty in paddy and mixed agriculture (70.1 percent). The poverty incidence in the rice-bowl states varies from 64.8 percent for Perlis to 79.0 percent for Trengganu. The average size of paddy farms in Malaysia has been estimated at 3.1 acres (Selvadurai, 1972a, p. 41). About 55 percent of the farms are less than three acres, and 80 percent are less than five acres. The size of landholding is a major factor affecting the income level of paddy farmers."' Small and fragmentary holdings are to a large extent the result of the system of inheritance in the country. Consolidation of existing land and resettlement on larger holdings of land newly developed by the Federal Land Development Authority (FELDA) are possible solutions to this problem. Another major factor contributing to poverty among paddy farmers is the low productivity of paddy in relation to other crops. 8 To some extent the problem of low incomes can be tackled by increasing yields on paddy land. An important government program to assist this subgroup is 16. See Selvadurai (1972a), table 2-2. Except forPenang, which has 24 percent, otherstates have a considerably smaller percentage of land under paddy. 17 The problemsofpaddy smallholders. and measures to raise theirincomes, are discussed extensively in FAo/World Bank (1975), annex E. 18. The tenurial status of farmers is sometimes said to affect the incomes they earn. About 40 percent of paddy land in Peninsular Malaysia is leased to tenant farmers under fixed rent (in cash or paddy) or crop-sharing leases. Two alleged features of sharecropping in developing countries are: (1) reduced incentives for the use of inputs and for on-farm investment (such as to improve land for double-cropping); and (2) very high rentals which cause poverty among tenant farmers. Recent empirical work by Huang, however, casts doubt on the validity of these arguments for Malaysia. Huang (1975, table 5) finds that productivity is not adversely affected by tenancy, and that tenants and owner-tenants (those who both own and rent land) have significantly higher yields than owner-cultivators. Tenants also appear to use more fertilizer than owner-cultivators and to adopt double-cropping just as readily. Finally, Huang finds that the poverty of paddy tenants is due not to exorbitant rentals but to the relatively low profitability of paddy farming compared with other crops or wage employment SUBGROUPS IN POVERTY 157 therefore irrigation and drainage to provide for double-cropping. The extension of irrigation facilities was initiated during the Second Malaysia Plan,"' when it was intended to increase the double-cropped area from one- third of all paddy land in the country to two-thirds. The scope for double- cropping is, however, limited by the availability of surface water and the topography.20 In any case, double-cropping needs to be supplemented by direct measures to improve yield.2' The FAO/World Bank (1975) report suggests that average yields fall short of best-practice yields on the order of 40 percent; it emphasizes the use of higher-yielding varieties and more fertilizer and pesticides. The adoption of these inputs clearly depends on the quality and intensity of extension services and the availability of credit. Policies aimed at alleviating poverty by raising productivity also help fulfill the government's target of self-sufficiency in rice. There is a residual category of paddy farmers who cannot be lifted out of poverty by productivity-raising measures: those whose holdings are too small to generate sufficient income even with double-cropping and best- practice yields. The long-term solution to this problem lies in encouraging such smallholders to switch to more remunerative crops or occupations. Laborers in Paddy and Mixed Agriculture The two-digit occupation-industry matrix shows a poverty incidence of 63.9 percent for farm laborer households in paddy, livestock, and mixed agriculture (table 5-2). 2 This small subgroup consists mainly of employees on the larger smallholdings who, unlike estate workers, are neither unionized nor covered by formal wage agreements. Some farm laborers may possess fractional parcels of land, but these make only a modest contribution to their total income. On the whole, the subgroup is landless and relies on wage employment for support. The conditions of employment are diverse, and many work as agricultural laborers on vegetable, poultry, and mixed-agriculture farms. Many are also employed as seasonal laborers in paddy, especially in the double-cropping areas.23 19. Large irrigation works have begun in Muda (the states of Kedah and Perlis), Kemubu (Kelantan), and Besut (Trengganu). 20. The limit has been estimated by EPU (1975) at about 75 percent of total paddy area. Remunerative off-season crops need to be identified for the 25 percent which cannot be double-cropped. Tobacco has been successfully tried among Kelantan farmers; other crops such as fruits and vegetables could also be tried. 21. See Selvadurai (1972a), table 2-6, p. 21, for estimates of yield variation across states. 22. The EPU (1975, p. 33) estimate of poverty incidence for this subgroup is 80 percent, but this is stated to be based on "'shaky" information. 23. For tasks such as transplanting and harvesting of paddy, family labor is often insufficient, and it is necessary to hire outside labor. Table 5-3. Incidence of Poverty among Five PES Subgroups by State Paddy smallholders Laborers on and livestock paddy, livestock, Laborers and mixed- and mixed- on rubber agriculture agriculture Rubber estates or farmers farms smallholders smaliholdings Fishermen State (01 x 61) (01 x 62) (11 x 61) (11 x 62) (04'x 64) Johore Incidence of poverty (percent) 39.2 44.6 49.5 42.1 53.1 Number of poor households 38 29 164 246 43 Total number of households 97 65 331 584 81 Kedah Incidence of poverty (percent) 66.8 77.7 49.6 51.9 59.1 Number of poor households 588 87 70 182 39 Total number of households 880 112 141 351 66 Kelantan Incidence of poverty (percent) 78.7 84.3 63.4 64.6 79.7 Number of poor households 702 75 123 93 51 Total number of households 892 89 194 144 64 Malacca Incidence of poverty (percent) 57.6 -a 38.7 41.9 35.3 Number of poor households 34 8 29 54 6 Total number of households 59 11 75 129 17 Negri Sembilan Incidence of poverty (percent) 55.1 _ a 58.7 37.7 -a Number of poor households 49 4 105 81 2 Total number of households 89 8 179 215 3 Pahang incidence of poverty (percent) 62.0 -a 54.2 48.8 59.3 Number of poor households 134 5 110 60 16 Total number of households 216 15 203 123 27 Penang Incidence of poverty (percent) 57.3 57.9 -a 58.7 36.9 Nu -ber of poor households 110 22 2 37 38 Total number of households 192 3O 9 63 103 Perak Incidence of poverty (percent) 56.4 44.1 66.2 50.5 43.8 Number of poor households 216 26 278 249 64 Total number of households 383 59 420 493 146 Perlis Incidence of poverty (percent) 64.8 80.6 _-a -a 44.1 Number of poor households 142 25 0 5 15 Total number of households 219 31 2 6 34 Selangor Incidence of poverty (percent) 44.9 40.7 39.8 42.0 46.2 Number of poor households 122 22 51 86 6 Total number of households 272 54 128 205 13 Trengganu Incidence of poverty (percent) 79.0 80.0 59.8 52.7 53.8 Number of poor households 214 24 49 58 35 Total number of households 271 30 82 110 65 Peninsular Malaysia Incidence of poverty (percent) 65.8 63.9 55.6 47.5 50.9 Number of poor households 2,349 327 981 1,151 315 Total number of households 3,570 512 1,764 2,423 619 a. Denotes sample size too small for statistically valid inference. 160 INEQUALITY AND POVERTY IN MALAYSIA This subgroup accounts for only 5.8 percent of households in all five subgroups (512 households out of 8,888). Its ethnic composition and poverty breakdown are: Malay Chinese Indian Other All races Incidence of poverty (percent) 69.6 34.4 31.6 - 63.9 Number of poor households 297 22 6 2 327 Total number of households 427 64 19 2 512 Table 5-3 shows that over half the subgroup is found in the rice-bowl states, which suggests that a large proportion of it probably consists of paddy laborers. Landlessness in these areas is also higher than in other states, and the incidence of poverty there is about 80 percent. There are few direct policy measures which would specifically benefit this subgroup. One measure would be to select them for new land development schemes in paddy, oil palm, rubber, and the like. Even before they are settled, they could be employed as unskilled labor for preparatory work on develop- ment sites. A drawback to this measure, however, is that it might remove a valuable supply of agricultural labor at peak periods. Moreover, land-rich states are frequently reluctant to accept residents of other states as settlers. It might be preferable to create nonagricultural employment in the off-peak season through rural public works and the stimulation of local small-scale industry. Rubber Smallholders The two-digit occupation-industry matrix (table 5-2) allows us to isolate the important rural subgroup of rubber farmers. Since only a negligible proportion of the rubber farmers (less than 0.5 percent) are estate owners24 (Barlow and Chan, 1968), the subgroup can be taken as consisting of smallholders. The incidence of poverty among rubber smallholders is thus 55.6 percent, with Malays constituting an overwhelming majority (96.1 percent) of rubber smallholders in poverty. The ethnic composition and breakdown of poverty for this subgroup are: Malay Chinese Indian Other A/l races Incidence of poverty (percent) 63.8 11.8 22.7 - 55.6 Number of poor households 943 31 5 2 981 Total number of households 1,477 262 22 3 1,764 Information on the distribution of smallholdings by size is scanty. EPU (1975) estimates 45 percent of smallholdings to be less than five acres, and 90 percent to be less than ten acres.25 The average size of rubber 24. In Malaysia, an estate is defined as a farm larger than 100 acres. 25. FAO/World Bank (1975, p.70), however, estimates 28 percent of smallholdings to be less than five acres. SUBGROUPS IN POVERTY 161 smallholdings has been put at 6.6 acres by Barlow and Chan (1968). Of the 2.7 million acres in the country under smaliholder rubber, about 60 percent are planted with high-yielding stocks which yield an average of 800 pounds per acre compared with 500 pounds per acre elsewhere. EPU (1975) has calculated poverty percentages separately for the two categories of smallholding: it estimates a 30 percent incidence for owners with high- yielding stock and a 70 percent incidence for those with low-yielding stock.26 Overall il has estimated a poverty percentage for rubber small- holders of 50 percent, which is reasonably close to my own estimate of 55.6 percent from PES. Table 5-3 shows that the largest concentration of poor rubber smallhold- ers is in Perak, followed by Johore, Kelantan, Pahang, and Negri Sembilan. These five states account for 79.5 percent of rubber smallholder households in poverty. The poverty incidence is highest in Perak (66.2 percent), followed by Kelantan (63.4 percent), Trengganu (59.8 percent), Negri Sembilan (58.7 percent), and Pahang (54.2 percent). The states with the heaviest concentrations of poor rubber smallholders are different from those with the heaviest concentrations of poor paddy smallholders. The extent to which smallholder acreages are replanted with high- yielding stock atTects the level of poverty. Those with very small-size holdings have been reluctant to take part in government replanting programs, apparently because-although those with larger smallholdings can afford to replant in stages-small growers find it difficult to forgo rubber income during the six to seven years needed for replanting. Government grants are not geared to the maintenance of smallholder income during the long immaturity period but only to the estimated cash outLays required for replanting (cutting and burning old trees, buying new trees, fertilizer, pesticides, and so forth). If smallholders are not to be deterred from replanting, grants need to be set at levels which maintain current income.27 The Rubber Industry Smallholders Development Authority (RISDA) has initiated schemes for block replanting that can also increase the financial inoentives to replant. Apart from reducing the unit cost of replanting, such schemes might allow smallholders to spread replanting in the block so that sorne rubber income is maintained during the gestation period. Since the group replanting schemes cannot be implemented, however, without the 26. Estimates of nonrubber income were based on the only available socioeconomic survey of rubber smaliholdings: Selvadurai (1972d). :27. Some smailholders are ineligible for replanting grants because they are squatters Although they have no title to their lands they are unlikely ever to be removed from them; if the status of such people were legitimized, they could be included in replanting programs. 162 INEQUALITY AND POVERTY IN MALAYSIA consent of the multiple owners of plots, some measures to make partici- pation obligatory for unwilling co-owners might have to be considered. There appears to be scope for raising smallholder incomes further by improving production practices. Smallholder yields (800 pounds per acre for high-yielding stock) are still well below estate yields (about 1,200 pounds per acre) or yields on organized settlement projects with intensive technical guidance. RISDA could help to raise these yields by increased extension services, particularly beyond the replanting period.28 Some holdings are too small to provide poverty-line incomes, even if they are replanted with high-yielding stock. EPU (1975) calculations suggest that holdings of four to five acres in high-yielding materials are sufficient to generate the 1970 poverty-line income for an average-size household (earning a "typical" amount of nonrubber income). The Barlow and Chan (1968) study comes up.with a goal of six to eight acres under high-yielding materials, which it considers "optimum."29 To the extent that farm size is the constraint, fringe alienation schemes might be supported near present holdings. A government agency such as RISDA could acquire and con- solidate existing land to create viable-size plots. In the longer term, those with subeconomic holdings could be encouraged to switch to alternative crops which generate higher and more stable incomes. Although this policy might conflict with the government's desire to protect rubber production, it may nonetheless be the best way of alleviating poverty among this group of smallholders. A change in the indirect taxation of rubber is another policy that might be considered to alleviate poverty among smallholders (discussed in chapter 8 under general policies30). The farm-gate price of smallholder rubber could be raised significantly by reducing or eliminating the rubber export tax, which falls almost wholly on producers. This measure would also stimulate the production of rubber. A general reduction of rubber taxes would benefit all producers, however, and not just those in poverty. To minimize leakages, one needs to consider ways of revising the export duty on rubber so that its incidence is reduced on small producers, say, 28. Various institutional and administrative reforms, such as improved marketing of smallholder rubber and better coordination of responsible agencies, are discussed in the FAo/World Bank (1975) report. 29. Such acreage figures suggest that the present allotment of twelve acres on government (FELDA) land settlement schemes is considerably above that needed to lift a family out of poverty. It would seem desirable to reduce the land allocation on these schemes to, say, six acres, and increase the number of families settled. 30. See the subsection entitled "Fiscal Policies" in chapter 8 The possibility of stabilizing the price received by rubber producers is discussed in the subsection entitled "Intervention in Commodity Markets." SUBGROUPS IN POVERTY 163 those with acreages below ten to fifteen acres. One method might be to adopt a system of rebates below the acreage limit fixed, possibly in the form of subsidized inputs or higher replanting grants. Another method is to change the base of the tax from a product to a special revenue or production tax on estates, which might be redefined to begin at fifteen to twenty acres instead of a hundred acres as at present. Laborers on Rubber Estates and Smallholdings The intersection of the two-digit farm laborer group with the rubber subsector in table 5-2 generates the subgroup of laborers on rubber estates and smallholdings. The definition of a rubber smallholding used by the Malaysian Department of Statistics is "an area contiguous, aggregating less than 100 acres, planted with rubber, or on which the planting of rubber is permitted under a single ownership." Rubber estates account for 1.6 million acres out of the 4.3 million acres in Malaysia planted with rubber in 1970.3' Smallholdings account for 2.7 million acres, or 63 percent of acreage, but produce only 49 percent of total rubber output. Smallholdings of fewer than ten acres probably account for half the rubber produced by all smallholdings. Most smallholdings between ten and a hundred acres can be expected to hire significant numbers of nonfamily laborers. The PES, unfortunately, does not distinguish between estates and smallholdings. For the combined subgroup of laborers on estates and smallholdings, it shows a poverty incidence of 47.5 percent. The ethnic composition and poverty breakdown for this subgroup are: falay Chinese indian Other All races Incidence of poverty (percent) 64.0 18.6 30.1 - 47.5 Number of poor households 889 79 183 0 1,151 Total number of households 1.390 425 607 1 2,423 The relatively large percentage of Indians in this subgroup reflects their dominant position as estate workers (Ministry of Labour and Manpower, 1972). Malays are not far behind Indians in estate employment, with a share in 1972 of 33 percent compared with the Indians' 40 percent. When laborers on smallholdings are included, Malays form the largest ethnic community (57.4 percent of all households), which reflects their over- whelming majority as laborers on the smallholdings. Fairly complete information is available on the cash incomes of rubber estate workers. Monthly cash incomes, as recorded in the Ministry of Labour's annuial survey, are shown at about M$115 for 1970. Subsidized 31 Rubber land in turn accounts for two-thirds of total planted area in the country. 164 INEQUALITY AND POVERTY IN MALAYSIA housing and fringe benefits (such as electricity, water, and medical care) are additional and have been estimated by EPU (1975) at approximately M$30 per month for 1970. With further assumptions about household size and number of income recipients, the EPU estimates a poverty incidence of 20 percent among estate worker households. Households below the poverty line are those with few recipients or a large household size, or both, and include some living off the estates who receive negligible nonmonetary benefits. Contract workers and workers on nonunionized estates also constitute some of the poor in this subgroup. About half of all estate workers are covered by a wage agreement between the National Union of Plantation Workers (NuPw) and the Malayan Agricultural Producers Association (MAPA). This agreement includes a basic wage, an output incentive, and a sliding-scale adjustment for rubber prices. It has succeeded in raising the incomes of unionized workers considerably above those of workers on smallholdings. Even among estate workers not covered by the agreement, wages have clearly been influenced by union scales (EPU, 1975, p. 19). The EPU'S poverty estimate of 20 percent for estate workers together with a PES poverty incidence of 47.5 percent for all rubber laborers implies a rather high incidence for workers on smallholdings.32 One policy measure that might help is to extend union-management agreements and Ministry of Labour regulations to smallholdings,33 but the employment costs of this measure would need to be considered. Another problem that faces rubber workers is one of redundancy. Estate work has become increasingly mechanized with a general improvement in rubber technology. There has also been a structural shift in recent years from rubber to the less labor-intensive oil palm. The problem of redundancy is likely to continue in the future, yet there are indications of relative immobility, especially among estate workers.34 One approach is to provide vocational and other training for younger estate workers to prepare them for job mobility. Another approach is to settle poor estate workers on government (FELDA) land schemes.35 Their previous experience on estates is likely to facilitate the transition. 32. EPU (1975) estimates a poverty incidence of something like 80 percent for all nonestale agricultural laborers. 33. Ministry of Labour regulations might be extended by reducing the minimum acreage defining an estate from the present hundred acres to fifteen to twenty acres. 34. According to FAo/World Bank (1975, p. 39), immobility is "supplemented in many cases by the reluctance of rubber tappers to take on the much harder work of oil palm harvesting." 35. FAo/World Bank (1975, p. 42) notes that special efforts should be made to settle more Indians on public land schemes. It points to the fact that whereas non-Malays have SUBGROUPS IN POVERTY 165 Fishermen In table 5-2 the subgroup of fishermen is generated by the intersection of the two-digit occupational code for fishermen, hunters, and related workers with the two-digit industrial code for fishing. It shows that the subgroup of fishing households in Peninsular Malaysia is relatively small (accounting for 7.0 percent of households in all five subgroups), and that its poverty incidence is 50.9 percent. The EPU has also estimated poverty among fishermen by using two socioeconomic surveys of this group.36 From these surveys EPU (1975) estimates an overall poverty incidence of 65 percent in 1970, with separate incidences of 90 and 40 percent, respectively, for the East and West coasts of the peninsula. This implies that more than two-thirds of poor fishing households reside on the East Coast. My estimate of overall poverty diverges from that of EPU, and its distrilbution across states also turns out to be somewhat different. Whereas according to EPU the largest number of poor fishermen reside in Trengganu, followed by Kelantan and East Johore, table 5-3 indicates a different ordering: Perak, Kelantan, and Johore-with heavy concentrations in Kedah and Penang. The PES results also question the view that East Coast fishermen are much poorer (poverty incidence 90 percent) than West Coast fishermen (poverty incidence 40 percent). According to table 5-3, the East Coast states display poverty incidences around 55 percent (Trengganu 53.8 percent and Pahang 59.3 percent); but almost equally high incidences seem to obtain in the West Coast states (Kedah 59.1 percent, Perlis 44.1 percent, and Perak 43.8 percent), with Johore, which is on both coasts, at 53.1 percent. The subgroup of fishermen shows an ethnic composition and poverty breakdown as follows: Malay Chinese Indian Other All races Incidence of poverty (percent) 60.3 25.9 - - 50.9 Nurnber of poor households 269 43 2 1 315 Total number of households 446 166 6 1 619 The ethnic breakdown for the subgroup has also been computed for each state separately, although this is not shown here. Between the two coasts there are significant differences in the racial composition of the fishing constituted about 4 percent of settlers in FELDA schemes, their share of applications has run at around 25 percent. 36. See Fisheries Division (1971), and Universiti Sains Malaysia (1972). In fact, these are the only socioeconomic surveys of this group at present available. 166 INEQUALITY AND POVERTY IN MALAYSIA population. In the eastern states almost all fishing households are Malay, while in the western states, particularly Penang and Perak, about half the fishermen in poverty are Chinese. On the whole, however, poverty among fishermen in Malaysia is predominantly a Malay problem. The poorest fishermen are usually small-time operators engaged in inshore fishing, who use traditional gear and net low catches. One policy measure to help this subgroup might consist in modernization: better boats and engines, a greater variety of nets and gear for different types of fish and weather conditions, and practical training in the use of this equipment. Extension of bank credit, grants and subsidies, and training courses would need to be provided by the government. There are dangers, however, in excessive modernization. An expansion in trawling, for instance, will not necessarily benefit traditional fishermen. Trawlers are relatively capital intensive and their crews need not be experienced fishermen; thus traditional poor fishermen may not be provided with employment. Moreover large-scale trawling poses a threat to inshore fishing resources, on which small fishermen rely for their livelihood. The danger of depleting fishing resources suggests that poverty allevi- ation programs should not concentrate on increasing output from existing fishing activities (see Khoo, 1976). Rather, the possibilities of alternative employment and sources of income for poor fishermen should be considered. Fishermen can be encouraged to emigrate to other occupations by providing vocational training for those willing to move (for example, through MARA, the new Rural and Industrial Development Authority). Fishermen could also be accommodated on the state and federal land settlement schemes.37 Another measure would be to help fishing house- holds obtain more part-time employment, for example, in agricultural crops such as cashew nuts and tobacco. Any such program would naturally have to be tailored to the situation in particular fishing, villages.3" In addition, the possibilities of aquaculture could be explored. Aquaculture would have the advantage of allowing fishermen continuity with the old way of life; it would also help boost fish supplies. 37. Unlike rubber estate workers, fishermen appear ready to move, but few so far have been absorbed in such schemes. 38. It is customary for many fishing households in the traditional sector to augment their incomes by working as part-time laborers. In some seaside places, notably Penang, the growth of tourism has brought new opportunities for earning cash through various services to tourists. In most rural fishing areas, however, agriculture is the only extra source of income available to the community, but fishermen often do not possess enough land. In Kelantan, a number of fishermen have managed to raise their family incomes substantially by cultivating tobacco on previously unused plots of bris soil. SUBGROUPS IN POVERTY 167 Other Subgroups A few other rural subgroups in poverty can be isolated by commodity in table 5-2. The two-digit industrial code for coconut, copra, and coconut oil isolates the coconut subsector as a whole. It shows- a poverty incidence of 56.0 percent and accounts for about 2 percent of total rural poverty. When the subsector is disaggregated by two-digit occupational group, a poverty incidence of 49.3 percent is found among coconut farmers (mainly smallholders), and a poverty incidence of 61.9 percent among coconut farm laborers. EPU (1975) has identified coconut smallholders as a rural poverty group, and from a single survey on coconut smallholdings (Selvadurai, 1968), it has estimated a poverty incidence of 45 percent for this group in 1970. l'he problems of coconut smallholders, and measures to raise their incomes, are discussed in the FAO/World Bank (1975) report. Chief among these measures are the replanting and rehabilitation of old coconut trees to r aise yields, and intercropping with suitable plants such as cocoa, pineapples, and colfee. The other subgroups isolated through table 5-2 are even smaller, with the oil paln subsector next in size. More than 95 percent of households in the oil paln subsector are headed by laborers on estates or smallholdings and show aL poverty incidence of 23.5 percent. The oil palm subsector has not been singled out by EPU as a target group; this seems justified in view of its comparatively low rate of poverty. Individual estimates of poverty in the subsectors of forestry, coffee, and tea have also been computed in table 5-2. The EPU (1975) study has estimated poverty for two of these three subsectors: forestry and coffee. In table 5-2 the estimate of poverty incidence among forestry workers is 33.3 percent, compared with EPU'S 29 percent. In table 5-2 the estimate of poverly incidence in the coffee subsector is 50.0 percent and among coffee smallholders 44.4 percent; EPU'S estimate for coffee smallholder poverty is 45 percent. Finally, the estimate of poverty incidence in the tea subsector is 15.4 percent. The PES sample size for the last subgroup is rather small (thirteen households), however, and begins to stretch the limits of reasonable statistical inference. Little confidence can attach to estimates based on an absolute sample size this small. Appendix: Urban Poverty Although quantitatively small (12.3 percent of all poor households are urban), urban poverty could become an increasingly serious problem in 168 INEQUALITY AND POVERTY IN MALAYSIA Malaysia. With the present emphasis on job creation in the modern sector, it is likely that large numbers of rural poor will migrate to the cities. 39 These migrants will probably join the urban informal or unorganized sector while searching and waiting for modern sector job openings. Thus the urban informal sector is a potential receptacle for the poor. Table 5-4 presents a one-digit profile of the urban poor in Malaysia.40 It shows urban poverty to be characterized by features quite different from those associated with rural poverty. The incidence of urban poverty is considerably lower than that of rural poverty (15.8 percent compared with 44.6 percent). In addition, several salient features stand out. The ethnic distribution of urban poverty is quite different from that of rural poverty, with the Chinese (41.9 percent) rather than the Malays (37.4 percent) as the most numerous racial group among the urban poor. This reflects the larger proportion of Chinese in urban areas, where they constitute as much as 57.9 percent of the household population, while the Malays make up only 25.9 percent. The incidence of poverty among Malays (22.9 percent), however, is twice that among Chinese (11.5 percent). The Indians, too, show a high incidence of poverty (20.9 percent) in urban areas. The distribution of urban poverty by region shows Selangor with the heaviest concentration although a low relative incidence. Selangor contains almost a third of all urban households in Malaysia, owing to the large urban conurbation around Kuala Lumpur. Penang and Perak also account for sizable percentages of the urban poor; this simply reflects their share in the urban population since the two states show near-average rates of poverty. Employees make up 53.9 percent of the heads of urban poor households, whereas own-account workers make up only 32.7 percent. This is the reverse of the situation for the country as a whole, in which a greater 39. A theoretical explanation for such migration is provided by Todaro (1969) and Harris and Todaro (1970). An econometric test of the Harris-Todaro model for Indian clata is contained in Anand (1971). 40. This is the profile corresponding to the poverty line of M$25 per capita household income. Tables 5-7 and 5-8 present urban poverty profiles corresponding to the poverty lines of MS15 and M$33. respectively. The povertyline implicit in TMPis M$33 forboth urban and rural areas, and no urban-rural differential in the cost of living has been assumed (despite the claim to the contrary in TMP, para. 507, p. 166). Note that an urban poverty line of M$33 together with a rural poverty line of M$25 implies an urban-rural cost of living differential of 32 percent, which is the approximate figure assumed for some developing countries. SUBGROUPS IN POVERTY 169 proportion of heads of poverty households are own-account workers. The relative incidence c,f poverty among these groups is broadly similar in urban and rural areas, respectively: 0.87 and 0.76 for employees and 1.43 (Text continues on page 172.) Table 5-4. Profile of Urban Poverty at a Poverty Line of M$25 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban among urban of urban oJ urban Selec red urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Race Malay 25.9 37.4 23.8 22.9 1.44 Chinese 57.9 41.9 60.9 11.5 0.72 Indian 14.9 19.6 14.0 20.9 1.32 Other 1.3 1.1 1.3 13.0 0.85 Total 100.0 100.0 100.0 State Johore 12.5 13.4 12.3 17.1 1.07 Kedah 5.5 6.5 5.3 18.7 1.18 Kelantan 4.6 12.7 3.1 43.6 2.76 Malacca 3.1 2.0 3.3 10.6 0.65 Negri Sembilan 2.5 2.4 2.5 15.3 0.96 Pahang 4.4 2.0 4.9 7.0 0.45 Penang 14.4 16.5 14.0 18.2 1.15 Perak 18.5 17.8 18.6 15.3 0.96 Perlis 0.0 0.0 0.0 0.0 0.00 Selangor 31.0 19.9 33.1 10.2 0.64 Trengganu 3.5 6.8 2.9 30.6 1.94 Total 1(10.0 100.0 100.0 Employment status of head Employer 5.6 0.5 6.4 1.1 0.09 Employee 62.2 53.9 63.5 12.3 0.87 Own-account worker 22.9 32.7 21.3 20.3 1.43 Housewife or houseworker 4.0 5.5 3.8 19.3 1.38 Unemployed 5.3 7.4 5.0 19.8 1.40 Total 100.0 100.0 100.0 (Table continues on the following page.) 170 INEQUALITY AND POVERTY IN MALAYSIA Table 5-4 (continued). Percentage Percentage Percentage 'elative distribution distribution distribution Incidence incidence among all among urban among urban of urban oJf urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 10.0 2.8 11.1 3.6 0.28 Adminstrative and managerial 5.6 0.3 6.3 0.7 0.05 Clerical and related 9.3 1.5 10.4 2.0 0.16 Sales 21.3 19.0 21.7 11.3 0.89 Service 15.9 16.5 15.8 13.1 1.04 Farmers 3.3 11.4 2.2 43.3 3.45 Farm laborers 4.6 13.0 3.3 36.4 2.83 Production 30.0 35.5 29.2 15.0 1.18 Total 100.0 100.0 100.0 Sector of employment of head Agnculture 4.9 15.3 3.3 40.8 3.12 Agricultural products 2.8 5.1 2.5 23.6 1.82 Mining and quarrying 1.6 1.6 1.5 14.0 1.00 Manufacturing 16.1 14.8 16.3 12.0 0.92 Construction 6.0 6.8 5.9 14.7 1.13 Public utilities 3.1 4.4 2.9 18.9 1 42 Commerce 23.3 19.4 23.9 10.9 0.83 Transport and communi- cations 9.9 11.2 9.8 14.7 1.13 Services 32.3 21.4 33.9 8.7 0.66 Total 100.0 100.0 100.0 Education of head None 24.7 39.6 22.0 25.4 1.60 Some primary 29.5 36.8 28.1 19.7 1.25 Completed primary 17.6 16.0 17.8 14.5 0.91 Lower second- ary (forms 1-111) 11.7 5.1 13.0 6.9 0.44 Some upper secondary (forms IV-V) 6.4 1.4 7.3 3.5 0.22 School certif- icate or higher 10.1 1.1 11.8 1.7 0.11 Total 100.0 100.0 100.0 SUBGROUPS IN POVERTY 171 Percenlage Percentage Perceniage Relatitve distribution disiribution distribution Incidence incidence among all among urban among urban of urban of urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sex of head Male 79.9 68.1 82.1 13.5 0.85 Female 20.1 31.9 17.9 25.2 1.59 Total 100.0 100.0 100.0 Age of head Under 20 1.7 1.7 1.7 15.8 1.00 20-29 I5.8 7.9 17.2 8.0 0.50 30-39 25.6 28.1 25.2 17.4 1.10 40-49 23.5 29.7 22.3 20.1 1.26 50-59 19.2 16.5 19.7 13.6 0.86 60+ 14.2 16.1 13.9 17.9 1.13 Total 100.0 100.0 100.0 Household size 1 11.2 5.3 12.3 7.5 0.47 2 8.8 4.3 9.7 7.8 0.49 3 10.7 5.9 11.6 8.7 0.55 4 12.4 7.1 13.4 9.1 0.57 5 11.4 11.4 11.4 15.8 1.00 6 10.6 13.3 10.1 19.9 1.25 7 9.4 12.6 8.8 21.3 1.34 8 8.0 11.8 7.2 23.6 1.48 9 6.0 6.7 5.9 17.6 1.12 10+ 11.5 21.6 9.6 29.7 1.88 Total 100.0 100.0 100.0 Number of children under age 15 0 29.5 11.4 32.9 6.1 0.39 1 15.7 9.3 16.9 9.4 0.59 2 14.8 10:0 15.7 10.8 0.68 3 13.0 14.7 12.7 18.0 1.13 4 10.2 17.7 8.8 27.5 1.74 5+ 16.8 36.9 13.0 34.9 2.20 Total 100.0 100.0 100.0 Number of income recipients 0 1.3 8.0 0.0 100.0 6.15 1 53.1 61.2 51.6 18.3 1.15 2 :!5.1 19.9 26.1 12.6 0.79 3 11.5 7.3 12.3 10.0 0.63 4+ 9.0 3.6 10.0 6.5 0.40 Total 100.0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M$25 per month; 15.8 percent of all urban households fall below this hne. 172 INEQUALITY AND POVERTY IN MALAYSIA Table 5-5. Distribution of Urban Poverty by Two-digit Occupational Classification Percentage distribution among urban povertv Selected occupation of head of household households Working proprietors in whosesale and retail trade 3.6 Salesmen, shop assistants, canvassers, news vendors, and street vendors (such as hawkers and cold-drink sellers) 15.0 Cooks, waiters, maids. and housekeeping service workers (including amahs and ayahs) 5.8 Protective service workers (such as doormen, porters, ushers. and attendants) and other service workers (including undertakers and embalmers) 7 9 Farmers, farm laborers, and fishermen 24.4 Bricklayers, carpenters, and other construction workers 6.2 Material handlers, dockers and freight handlers, transport equipment operators 13 8 Total 76.7 and 1.27 for own-account workers. As might be expected, open unemploy- ment of household heads is greater in urban areas (5.3 percent) than in rural areas (2.9 percent); and unemployment among household heads is a larger contributor to urban poverty (7.4 percent) than to rural poverty (3.4 percent).4t The incidence of urban poverty is slightly lower among households with unemployed heads (19.8 percent) than among households whose heads are own-account workers (20.3 percent). This indicates the importance of secondary earners in determining the size of household income. The occupational breakdown of heads of urban poverty households shows, somewhat surprisingly, that as many as 24.4 percent of them are farmers or farm laborers.42 In fact, farming households in urban areas have substantially higher rates of poverty than any other occupational group, clustering around the national avarage of 36.5 percent rather than the urban average of 15.8 percent. Groups with which urban poverty is generally thought to be associated are informal or unorganized sector workers, such as hawkers, stailkeepers, shoe repairers, trishaw pedalers, and houseboys. Workers in traditional manufacturing are also considered poverty prone, but those in organized manufacturing are regarded as on the whole immune. The two-digit occupational classification in table 5-5 supports the view that the urban poor are engaged largely in the informal 41. The fact that this contribution is nonetheless small tends to support the obserxation that the poor in developing countries cannot "afford" to be unemployed. 42. Urban areas are defined as towns with a population in excess of 10,000 SUBGROUPS IN POVERTY 173 Table 5-6. Distribution of Urban Poverty by Two-digit Industrial Classification Percentage distribution among urban poueriy Selected industrial sector of head of household households Agriculture and agncultural products 20 4 Retail trade (in food, drink, tobacco, footwear, and so on, inclucling hawkers and temporary stall-holders) 18.6 Government and conmmunity services (including local administration, education, medical and health services, religious and welfare organizations) 10.8 Personal services (including domestic and miscellaneous laundry services; eating and drinking stalls) 8 1 Water a nd sanitary services (including garbage collection and disposal and sewage disposal) 3.2 Transport (including taxi and trishaw services) 10 8 ConstrJction (contractors and services) 6 8 Manufacture of wood, rattan, and attap products (except furniture and footwear) 3.6 Total 82.3 sector, and are underemployed or self-employed on a casual basis.43 Consistent with the occupational breakdown, the sector of employment of heads of urban poverty households shows that 20.4 percent of them are in agriculture and agricultural products (table 5-4). The two-digit indus- trial classification in table 5-6 shows another 40.7 percent in commerce and services, 10.8 percent in transport, and 10.4 percent in construction and light manufacturing. The distribution of urban poverty by two-digit occupational and industrial categories (tables 5-5 and 5-6) confirms that, apart from agriculture and related activities, the poor are engaged in a variety of services, many of which are supplied in the informal sector. Agriculture and agricultural products account for about a quarter, and the informal service sector for about a half, of urban poor households. The rest of the urban poor are found in miscellaneous nonservice activities, including construction and manufacturing. The strong inverse relation between education of household head and poverty incidence persists in urban areas, and, as for overall poverty, there is a mtarked decline in incidence for households whose heads have acquired even some secondary education. The percentage of heads of poor (Text continues on page 186.) 43. Their underemployment may in fact represent a quasi-voluntary queuing for the attractive but infrequent modern sector jobs. Material on urban unemployment in Malaysia (Mazumdarn 1975), which shows the unemployed as long-time urban dwellcrs with some education, lends support to the queuing hypothesis. 174 INEQUALITY AND POVERTY IN MALAYSIA Table 5-7. Profile of Urban Poverty at a Poverty Line of M$15 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban among urban of urban of urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)/(1) of household (1) (2) (3) (4) (5) Race Malay 25.9 42.6 25.1 8.3 1.64 Chinese 57.9 34.7 59.1 3.0 0.60 Indian 14.9 21.6 14.5 7.3 1.45 Other 1.3 1.1 1.3 4.3 0.85 Total 100.0 100.0 100.0 State Johore 12.5 13.2 12.4 5.3 1.06 Kedah 5.5 7.0 5.4 6.4 1.27 Kelantan 46 17.1 4.0 18.6 3.72 Malacca 3.1 1.4 3.1 2.3 0.45 Negri Sembilan 2.5 1.7 2.5 3.4 0.68 Pahang 4.4 1.9 4.6 2.2 0.43 Penang 14.4 13.7 14.4 4.8 0.95 Perak 18.5 18.5 18.5 5.0 1.00 Perlis 0.0 0.0 0.0 0.0 0.00 Selangor 31.0 17.1 31.8 2.8 0.55 Trengganu 3.5 8.4 3.3 11.9 2.40 Total 100.0 100.0 100.0 Employment status of head Employer 5.6 1.0 5.7 0.6 0.18 Employee 62.2 44.9 62.8 2.4 0.72 Own-account worker 22.9 38.5 22.3 5.5 1.68 Housewife or houseworker 4.0 6.8 4.0 5.5 1.70 Unemployed 5.3 8.8 5.2 5.4 1.66 Total 100.0 100.0 100.0 SUBGROUPS IN POVERTY 175 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban among urban of urban of urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 10.0 3.2 10.3 0.9 0.32 Administrative and managerial 5.6 0.6 5 7 0.3 0.11 Clerical and related 9.3 1.3 9.5 0.4 0.14 Sales :21.3 16.8 21.5 2.3 0.79 Service 15.9 18.7 15.8 3.4 1.18 Farmers 3.3 14.2 3.0 12.4 4.30 Farm laborers 4.6 20.0 4.1 12.8 4.35 Production 30.0 25.2 30.1 2.4 0.84 Total 100.0 100.0 100.0 Sector of employment of head Agriculture 4.9 26.1 4.3 15.4 5.33 Agricultural prcducts 2.8 6.2 2.7 6.4 2.21 Mining and quarrying 1.6 1.9 1.5 3.5 1.19 Manufacturing 16 1 8.7 16.3 1.6 0.54 Construction 6.0 3.1 6.1 1.5 0.52 Public utilities 3.1 1.2 3.1 1.2 0.39 Commerce 23.3 16.2 23.6 2.0 0.70 Transport and communi- cations 9.9 11.8 9.9 3.5 1.19 Services 32.3 24.8 32.5 2.2 0.77 Total 100.0 100.0 100.0 Education of Head None 24.7 45.9 23.6 9.4 1.85 Some primary 29.5 30.1 29.5 5 1 1.02 Cornpleted pnmary 17.6 13.5 17.8 3.9 0.77 Lower secondary (forms 1-111) 11.7 7.1 12.0 3.0 0.61 Some upper secondary (forms IV-V) 6.4 1.7 6.6 1.3 0.27 School certificate or higher 10.1 1.7 10.5 0.8 0.17 Total 100.0 100.0 100.0 (Table continues on the fbilowing page.) 176 INEQUALITY AND POVERTY IN MALAYSIA Table 5-7 (continued). Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban among urban of urban of urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sex of head Male 79.9 50.1 81.5 3.2 0.63 Female 20.1 49.9 18.5 12.5 2.48 Total 100.0 100 0 100.0 Age of head Under 20 1.7 3.1 1.6 9.2 1.82 20-29 15.8 10.9 16.0 3.5 0.69 30-39 25.6 26.6 25.6 5.2 1.04 40-49 23.5 24.1 23.4 5.2 1.03 50-59 19.2 17.9 19.3 4.7 0.93 60+ 14.2 17.4 14 1 6.1 1.23 Total 100.0 100.0 100.0 Household size 1 11.2 2.8 11.7 1.3 0.25 2 8.8 9.5 8.8 5.4 1.08 3 10.7 8.7 10.7 4.1 0.8 1 4 12.4 8.1 12.6 3.3 0.65 5 11.4 12 0 11.4 5.3 1.05 6 10 6 14.3 10.4 6.8 1.35 7 9.4 8.4 9.5 4.5 0.89 8 8.0 10.9 7.8 6.9 1.36 9 6.0 65 6.0 5.4 1.08 10+ 11.5 188 11.1 8.2 1.63 Total 100.0 100.0 100.0 Number oJ children under age 15 0 29.5 11.5 30.5 2.0 0.39 1 15.7 13.4 15.8 4.3 0.8' 2 14.8 10 6 15.0 3.6 0.72 3 13.0 17.7 12.8 6.8 1.36 4 10.2 15 4 9.9 7 6 1.51 5+ 16.8 31.4 16.0 9.4 1.87 Total 100.0 100.0 100.0 Number of income recipients 0 1.3 25.2 0.0 100.0 19.38 1 53.1 48.4 53.4 4.6 0.91 2 25.1 18.2 25.5 3 6 0.73 3 11.5 4.8 11.9 2.1 0.42 4 + 9.0 3.4 9.2 1.9 0.38 Total 100 0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M$1 5 per month; 5 0 percent of all urban households fall below this line. SUBGROUPS IN POVERTY 177 Table 5-8. Profile of Urban Poverty at a Poverty Line of MS33 Percenrage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban among urban of urban of urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Race Malay 2;.9 34.5 23.0 33.9 1.33 Chinese 57.9 47.0 61.6 20.8 0.81 Indian 14.9 177 13.9 30.4 1 19 Other 1.3 0.8 1.5 16.3 0.62 Total 100.0 100.0 100.0 State Johore 12.5 13.4 12.1 27.5 1.07 Kedah 5 5 6.7 5.1 31.3 1 22 Kelantan 4.6 9.4 3.0 52.1 2.04 Malao-a 3.1 2.1 3.4 17.5 0.68 Negri Sembilan 2.5 2.0 2.6 20 9 0.80 Pahang 4.4 2.7 5.0 15.6 0.61 Penang 14.4 17.0 13.5 30.2 1.18 Perak 18.5 19.3 18.2 26.7 1.04 Perlis 0.0 0.0 0.0 0.0 0.00 Selangor 31.0 21.2 34.4 17.5 0.68 Trengganu 3.5 6.1 2.7 43.7 1.74 Total 100.0 100.0 100.0 Employment status of head Employer 5.6 0.6 7.1 2.9 0.11 Employee 62.2 58.5 63.3 22.4 0 94 Own-account worker 22.9 29.6 20.8 30.8 1.29 Housewife or hcuseworker 4.0 5.0 3.8 29.5 1.25 Unemployed 5.3 6.3 5.0 28 1 1.19 Total 100.0 100.0 100.0 (Table continues on the following page.) 178 INEQUALITY AND POVERTY IN MALAYSIA Table 5-8 (continued). Percentage Percentage Percentage Relative distribution distribution distribution Incidence tncidence among all amnong urban among urban of urban o0 urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 10.0 2.7 12.1 6.0 (0.27 Administrative and managerial 5.6 0.6 7 0 2.4 0 11 Clerical and related 9.3 3.6 10.8 8.5 0 39 Sales 21.3 19.1 22.0 19 6 0.90 Service 15.9 17 2 15 6 23 6 1 08 Farmers 3.3 8.8 1.8 57.3 2 67 Farm Laborers 4.6 10.0 3.0 47.9 2.17 Production 30.0 38.0 27.7 27.8 1.27 Total 100.0 100.0 100.0 Sector of employment of head Agriculture 4.9 11.9 2.9 54.8 2 43 Agricultural products 2.8 4.2 2 4 33.1 1.50 Mining and quarrying 1.6 2 2 1.3 32.6 1 38 Manufacturing 16.1 15.1 16.4 21.3 0.94 Construction 6.0 7.6 5.6 28.4 1.27 Public utilities 3.1 4.5 2.6 33 1 1 45 Commerce 23.3 19.5 24.5 18.9 0.84 Transport and communi- cations 9 9 12 2 9.3 27.8 1 23 Services 32.3 22 8 35.0 16.0 0.71 Total 100.0 100.0 100.0 Education of head None 24.7 35.8 20 9 37.0 1.45 Some primary 29.5 37.6 26.7 32.5 1.27 Completed primary 17.6 18.2 17.4 26.3 1 03 Lower secondary (forms 1-111) 11.7 5.7 13.8 124 0.49 Some upper secondary (forms IV-V) 6 4 1.6 8.0 6.4 0.25 School certificate or higher 10.1 1.1 13.2 2.8 0.11 Total 100.0 100.0 100 0 SUBGROUPS IN POVERTY 179 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among urban aniong urban oJ'urban oJ'urban Selected urban poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sex of iead Male 79.9 72 9 82.3 23.3 0.91 Female 20.1 27.1 17.7 34.4 1 35 Total 100.0 100.0 100.0 Age of head Under 20 1.7 1.4 1 8 21.7 0.82 20-29 15.8 8.7 18.2 14.2 0 55 30-39 :25.6 28.8 24.6 28.7 1.13 40-49 '23.5 29.7 21 3 32.3 1.26 50-59 19.2 16.5 20.1 21.9 0 86 60 + 14.2 14.9 14.0 26.7 1.05 Total 100.0 100.0 100.0 Household size 1 11.2 3.3 139 7.5 0.29 2 8.8 5.6 9.9 16.3 0.64 3 10.7 5.6 12.4 13.4 0.52 4 124 7.4 14.1 153 060 5 11.4 13.2 10.8 297 1.16 6 10.6 13.4 9.7 32.1 1.26 7 9.4 9.8 9.3 26.6 1.04 8 8.0 13.2 6.1 426 1 65 9 6.0 10.2 4 6 43.1 1.70 10+ 115 18.3 9.2 40.5 1.59 Total 100.0 100.0 100.0 Number of children inder age 15 0 29.5 10.6 36.0 9.2 0 36 1 15.7 9 6 17.8 15 6 0.61 2 14.8 11.7 15.8 203 0.79 3 13.0 16.2 12.0 31.7 1 24 4 10.2 16 6 8.0 414 1 63 5+ 168 35.3 10.4 53.8 2 10 Total 100.0 100.0 100.0 Number of inconie r ecipients 0 1.3 5.0 0.0 100.0 3 85 1 53.1 60.5 50.6 29.1 1.14 2 25.1 21.7 26.3 22.0 0 86 3 11.5 8.6 12.5 19.1 0.75 4+ 9.0 4.2 10 6 12.0 0.47 Total 100.0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M $33 per month; 25 5 percent of all urban households fall below this line. 180 INEQUALITY AND POVERTY IN MALAYSIA Table 5-9. Profile of Rural Poverty at a Poverty Line of M$] 5 Percentage Percentage Perceniage Relative distribution distribution distribution Incidence incidence among all among rural among rural of rural of rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Race Malay 67.1 89.6 61 6 26.3 1.33 Chinese 21.7 5.1 25.8 4.6 0.24 Indian 10.5 3.8 12 1 7.1 0.36 Other 0.7 1.5 0.5 40.3 2.14 Total 100.0 100.0 100.0 State Johore 13.8 11.5 14 3 16 5 0.83 Kedah 13.6 16.7 12.9 24.1 1.23 Kelantan 115 21.8 9.0 37.3 1.90 Malacca 5.2 3.6 5.6 13.5 0.69 Negn Sembilan 5.6 4.1 6.0 14.3 0.73 Pahang 6.4 4 9 6.7 15.2 0.77 Penang 6.1 4.2 6.6 13.6 0.69 Perak 17.5 15.3 18.0 17.1 0.87 Perlis 2.1 3.3 1.8 31.1 1 57 Selangor 13.1 6.1 14.8 9.1 0.47 Trengganu 5.1 8.5 4.3 32.6 1.67 Total 100.0 100.0 100.0 Employment status of head Employer 1.6 0.3 1.9 3.4 0.19 Employee 47.8 30.7 51.7 12.0 0.64 Own-account worker 45.6 64.7 41.2 26.5 1.42 Housewife or houseworker 2.1 1.3 2.2 11.8 0.62 Unemployed 2.9 3.0 3.0 18.7 1.03 Total 100.0 100.0 100.0 SUBGROUPS IN POVERTY 181 Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among rural among rural of rural of rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 4.1 0.5 4.9 2.2 0.12 Administrative and managerial 2.5 0.1 3.0 0.5 0.04 Clerical and related 2.1 0.0 2.6 0.0 0.00 Sales 7.9 3.9 8.9 9.5 o.49 Service 5.5 1.4 6.5 4.7 0.25 Farmers 36.4 59.3 31.0 31.2 1.63 Farm laborers 27.8 29.4 27.5 20.2 1.06 Production 13.7 5.4 15.6 7.6 0.39 Total 100.0 100.0 100.0 Sector of employment of head Agriculture 31.1 52.6 26.1 31.7 1.69 Agricultural products :14.1 35.0 33.8 19.3 1.03 Mining and quarrying 1.9 0.2 2.3 2.1 0.11 Manuiacturing 5.8 3.3 6.4 10.6 0.57 Construction 2.2 0.8 2.5 7.1 0.36 Public utilities 1.1 0.1 1.4 1.8 0.09 Comnuyerce 8.7 4.5 9.7 9.7 0.52 Transport and communi- cations 3.9 1.6 4.4 7.8 0.41 Services 11.2 1.9 13.4 3.2 0.17 Total 100.0 100.0 100.0 Education of head None 35.0 48.5 31.7 27.1 1.39 Some primary 34.6 34.4 34.6 19.4 0.99 Completed primary 21.5 15.2 23.1 13.8 0.71 Lower secondary (forms 1-111) 4.7 1.5 5.4 6.0 0.32 Some upper secondary (forms IV-V) 1.6 0.2 2.0 2.4 0.13 School certificate or higher 2.6 0.2 3.2 1.7 0.08 Total 100.0 100.0 100.0 (Table continues on the following page.) 182 INEQUALITY AND POVERTY IN MALAYSIA Table 5-9 (continued). Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among rural among rural of rural oJ rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (-)/(1) of household (1) (2) (3) (4) (5) Sex of head Male 82.5 75.9 84.1 18.1 0.92 Female 17.5 24.1 15.9 27.0 1.38 Total 100.0 100.0 100.0 Age of head Under 20 1.4 1.3 1.4 17.7 0.93 20-29 15.2 11.6 16.0 15.0 0.76 30-39 25.1 28.5 24.3 22.3 1.14 40-49 22.6 25.4 22.0 21.9 1.12 50-59 19.0 17.7 19.3 18.2 0 93 60+ 16.7 15.5 17.0 18.2 0.93 Total 100.0 100.0 100.0 Household sizc 1 8.0 0.5 9.8 1.3 0.06 2 9.9 10.6 9.7 21.1 1.07 3 12.1 9.7 12.7 15.8 0.80 4 13.3 7.0 14.9 10.3 0.53 5 13.0 15.8 12.3 23.9 1.22 6 12.6 17.2 11.4 26.9 1.37 7 9.9 12.1 9.4 24.0 1.22 8 7.9 11.1 7.1 27.6 1.41 9 5.3 6.6 5.0 24.6 1.25 10+ 8.0 9.4 7.7 23.0 1.18 Total 100.0 100.0 100 0 Number of children under age 15 0 22.8 10.5 25.8 9.0 0.46 1 17.5 14.4 18.3 16.2 0.82 2 15.8 13.4 16.3 16.8 0.85 3 13.8 18.5 12.7 26.3 1.34 4 12.0 17.1 10.8 28.0 1.43 5+ 18.1 26.1 16.1 28.4 1.44 Total 100.0 100.0 100.0 Number of income recipients 0 1.3 6.3 0.0 98.2 4.85 1 59.7 69.6 57.3 22.9 1.17 2 26.9 19.0 28.9 13.9 0.71 3 8.1 3.9 9.1 9.4 0.48 4 + 4.0 1.2 4.7 6.0 0.30 Total 100.0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M$ 15 per month; 19.6 percent of all rural households fall below this line. SUBGROUPS IN POVERTY 183 Table 5-10. Profile of Rural Poverty at a Poverty Line of M$33- Percentage Percentage Percentage Relative distribution distribution distribution Incidence incidence among all among rural amnong rural of rural of rural Selected ruwal poverty nonpovertwi poverty povertyn characteristic households households households (percent) (2)j(1) of household (1) (2) (3) (4) (5) Race Malay 67.1 79.6 49.3 69.6 1.19 Chinese 21.7 11.7 36.0 31.5 0 54 Indian 10.5 7.9 14.1 44.1 0.75 Other 0.7 0.8 0.6 65.1 1.14 Total 100.0 100.0 i00.0 State Johore 13.8 12.2 15.8 52.3 0.88 Kedah 13.6 15.9 10.4 68.3 1.17 Kelantan il.5 15.7 5.6 79.9 1.37 Malacca 5.2 4.6 6.2 51.3 0.88 Negri Sembilan 5.6 4.7 6.9 49.1 0.84 Pahang 6.4 5.5 7.6 50.8 0.86 Penang 6.1 5.9 6.5 56.2 0.97 Perak 17.5 17.2 17.8 57.8 0.98 Perlis 2.1 2.7 1.3 73.9 1.29 Selangor 13.1 9.0 18.9 40.2 0.69 Trengganu 5.1 6 6 3.0 75.9 1 29 Total 100.0 100.0 100.0 Employment status of head Employer 1.6 0.5 3.2 16.8 0.31 Employee 47.8 39.1 59.7 47.5 0.82 Own-account worker 45.6 55.4 32.0 70.4 1.21 Housewife or houseworker 2.1 1.9 2.3 52.8 0.90 Unemployed 2.9 3.1 2.8 60.4 1.07 Total 100.0 100 0 100.0 (Table continues on the Jollowing page.) 184 INEQUALITY AND POVERTY IN MALAYSIA Table 5-10 (continued). Percentage Percentage Percentage Relntive distribution distribution distribution Incidence incidence among all among rural among rural of rural of rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (2),'(1) of household (1) (2) (3) (4) (5) Occupation of head Professional and technical 4.1 1.0 8.3 14.6 0.24 Administrative and managerial 2.5 0.7 5.0 15.4 0.28 Clerical and related 2.1 0.5 4.3 14.2 (.24 Sales 7.9 5.6 11.1 41.1 0.71 Service 5.5 2.9 9.1 30.5 0.53 Farmers 36.4 48.3 20.0 77.0 1.33 Farm laborers 27.8 31.0 23.5 64.7 1.12 Production 13.7 10.0 18.7 42.5 0.73 Total 100.0 100.0 100.0 Sector of employment of head Agnculture 31.1 41.1 17.3 76.6 1 32 Agricultural products 34.1 36.8 30.2 62.6 1.08 Mining and quarrying 1.9 1.1 3.0 34.5 0.58 Manufacturing 5.8 4.4 7.8 43.5 0.76 Construction 2.2 1.7 2.9 44.6 0.77 Public utilities 1.1 0.8 1.6 40.8 0.73 Commerce 8.7 6.3 12.1 41.5 0.72 Transport and communi- cations 3.9 3.0 5.1 44.8 0.77 Services 11.2 4.8 20.0 24.8 0.43 Total 100.0 100.0 100.0 Education of head None 35.0 41.5 25.8 69.4 1.19 Some primary 34.6 36.0 32.6 60.9 1.04 Completed primary 21.5 20.1 23.6 54.6 0.93 Lower secondary (forms 1-111) 4.7 2.0 8.4 25.5 0.43 Some upper secondary (forms IV-V) 1.6 0.2 3.6 8.7 0.13 School certificate or higher 2.6 0.2 6.0 4.1 0.08 Total 100.0 100.0 100.0 SUBGROUPS IN POVERTY 185 Percentage Percentage Percentage Relative distribulion distribution distribution Incidence incidence among all among rural among rural oJ rural of rural Selected rural poverty nonpoverty poverty poverty characteristic households households households (percent) (2)1(1) of household (1) (2) (3) (4) (5) Sex of head Male 82.5 80.3 85.5 57.0 0.97 Female 17.5 19.7 14.5 65.9 1.13 Total 100.0 100.0 100.0 Age of head Under 20 1. 1.1 1.8 46.4 0.79 20-29 15.2 12.3 19.2 47.7 0.81 30-39 25.1 26.3 23.4 61.3 1.05 40-49 22.6 24.7 19.9 63.7 1.09 50-59 19.0 18.4 19.8 56.8 0.97 60+ 16.7 17.2 15.9 60.4 1.03 Total 100.0 100.0 100.0 Household size 1 8.0 4.4 13.1 32.3 0.55 2 9.9 8.2 12.2 48.6 0.83 3 12.1 10.0 15.0 48.4 0.83 4 13.3 12.3 14.7 54.2 0.92 5 11.0 14.7 10.7 66.0 1.13 6 12.6 13.9 10.7 64.6 1.10 7 9.9 10.5 9.2 61.9 1.06 8 7.9 10.0 5.0 74.0 1.27 9 5.3 6.8 3.1 75.6 1.28 10+ 8.0 9.2 6.3 67.6 1.35 Total 100.0 100.0 100.0 Number of children under age 15 0 22.8 15.0 33.9 38.5 0.66 1 17.5 15.7 20.0 52.7 0.90 2 15.8 15.7 15.7 58.6 0.99 3 13.8 15.8 11.1 66.8 1.14 4 12.0 14.2 9.0 69.1 1.18 5+ 18.1 23.6 10.3 76.5 1.30 Total 100.0 100.0 100.0 Number of income recipients 0 1.3 2.1 0.0 98.7 1.62 1 59.7 64.6 52.8 63.4 1.08 2 26.9 24.7 30.2 53.7 0.92 3 8.1 6.3 10.7 45.5 0.78 4+ 4.0 2.3 6.3 34.5 0.58 Total 100.0 100.0 100.0 Note: The poverty line is defined at a per capita household income of M$33 per month; 58.6 percent of all rural households fall below this line. 186 INEQUALITY AND POVERTY IN MALAYSIA households with no education is smaller in urban (39.9 percent) than in rural areas (43.7 percent), and a larger percentage have some upper secondary education or better (2.5 percent compared with 0.4 percent). The demographic features of urban poverty are a little different from those of rural poverty. Women head a significantly larger percentage of poor households in urban areas (31.9 percent) than in rural areas (21.2 percent). The age profile of poverty is similar in the two locations, but the "alternating periods of want and comparative plenty" are more pro- nounced in urban areas.44 The highest relative incidence of urban poverty occurs for the 40-49 year age group. The household size distribution shows a larger average size for urban poor households (6.77 members) than for rural poor households (5.77 members).45As many as 21.6 percent of urban poor households have ten or more members. The relative incidence of poverty among large households is higher in urban than in rural areas, which might reflect the fact that unpaid family helpers can make more of a contribution to household income in rural areas (by working on their own land) than they can in an urban environment. Another manifestation of the heavier concentration of poverty in large households in urban than in rural areas is the higher relative incidence of urban poverty in households with three or more children under age 15. Finally, the incidence of urban poverty declines in a predictable manner with the number of income recipients in the household. As many as 8.0 percent of urban poverty households have no recipients compared with 2.8 percent of rural poverty households. These findings suggest higher average dependency ratios and lower average participation rates for urban than for rural poor households. 44. Five alternating periods of want and comparative plenty in the life of a laborer were indicated by Rowntree (1922) in his study of poverty in nineteenth-century Britain. The cross- section data in table 5-4 tend to corroborate the cycle of poverty noted by Rowntree An initial period of want when [the laborer] is too young to work [under 20 years] is followed by a period of comparative prosperity when he is earning money which continues [20-29 years] until he has two or three children. The next period of poverty [30-49 years] will last until his first child begins to earn. While the children are earning, the man enjoys another period [50-59 years] of prosperity . only to sink back again into poverty when his children have married, and he is too old to work. 45. There is, however, a larger percentage of one-member households in urban (11.2 percent) than in rural areas (8.0 percent). 6 Inequality in the Personal Income Distribution IN THE PRECEDING CHAPTERS inequality in levels of living and poverty have been measured by the distribution of individuals or households according to per capita household income. I now consider the distribution of income recipients according to personal income, also referred to as the personal income distribution. An income recipient is defined here as a person who receives income from any of the eleven sources listed in the PES question- naire (see "Definition of PES Income" in chapter 2); personal income is defined as the sum of the income received from these eleven sources.' The category of income recipient thus includes both working and nonworking persons, and personal income includes both earned and unearned income.2 Household income in PES is the sum of personal income received by all the income recipients in a household. The distribution of income recipients according to personal income provides little information on inequality in levels of living in a country,3 which is usually the eventual concern of analysis. Nevertheless, it is of some importance to analyze the personal income distribution. It comes closest to the distribution generated by the production and payments system of the 1. Personal income thus includes wages and salaries plus income from rent, property, gifts, and sD on. The terminology is borrowed from the national accounts notion of personal income to which this PEs concept of income approximately corresponds. 2. There is a problem with imputing income from jointly owned assets such as land or a family house. In practice, such income is likely to have been attributed wholly to the household head, and unpaid family workers, for example, would not count as income recipients unless they were in receipt of income from other sources This should be borne in mind in the interpretation of some of the findings in this chapter (see, for example, note 9). 3. Nor does this distribution even provide accurate information on inequality in levels of living among recipients themselves. The reason is that income is redistributed from high- inccme recipients to low- (including zero-) income recipients in each household, and the percentage transfer varies from recipient to recipient according to household circumstances. 187 188 INEQUALITY AND POVERTY IN MALAYSIA economy and hence is relevant for understanding the distributional nature of that system. It is also the distribution most directly amenable to policy intervention, because most instruments of economic policy operate via the production and payments system and affect household or per capita household income only through personal incomes. Ultimately, of course, the personal income distribution needs to be translated into the household or per capita household income distribution so that welfare implications can be considered.4 But this mapping is not an easy task since it requires matching income recipients (from different parts of the distribution) who belong to the same household, as well as information on the size of the household. Without a satisfactory theory that would allow a prediction of the per capita household income distribution from the personal income distribution, it is possible to make only the following suggestive type of calculation. Its purpose is to describe the inequality across households in per capita income in terms of the inequality in income per recipient and the proportion of recipients (or participation rate). Since only the distribution of positive income recipients is considered, households with no recipients, that is, zero-income households, are excluded. Using standard notation, for household h let yh denote the household income, mh the household size, and nh the number of income recipients. As households with no recipients are excluded, the per capita income of household h can be expressed as yh/mh = Yh/nh nh/mh where yh/nh is the income per recipient of household h, and nh/mh the proportion of recipients in it. Taking logarithms of both sides, log yh/mh = log yh/lnh + log nh/Mi. Now forming the variance across households, var (log Yh/Mh) = var (log Yh/nh) + var (log nh/mh) + 2cov (log yh/nh, log nh/mh). (1) The variance of the logarithm of income (or any positive variable) is a well- known measure of inequality (appendix A); I call this measure "varlog." Therefore this equation expresses the inequality across households in per capita income as a sum of three terms: the inequality in income per recipient; the inequality in proportion of income recipients, or participation rate; and 4. Mapping the personal to the household income distribution is important for determin- ing not only welfare levels among the population, but also consumer demands in the economy-the household being the appropriate decisionmaking unit in both cases. INEQlJALITY IN THE PERSONAL INCOME DISTRIBUTION 189 twice the covariance between the logarithms of income per recipient and proportion of recipients. Table 6-1 presents the means and varlogs of these variables for Malaysian households, with a disaggregation by racial group. The inequality level, or varlog, of 0.8453 in per capita income across all households is the sum of inequality levels of 0.7154 in income per recipient and 0.4251 in partici- pation rate, minus a number (0.2952) that reflects the negative correlation between the logaritlhms of these variables. A large part of inequality in per capita household income is thus accounted for by inequality in income per recipient. Moreover, changes in economic policy that reduce inequality in income per recipient are also likely to reduce inequality in per capita household income.5 The negative correlation between the logarithm of income per recipient and the logarithm of participation rate is not strong enough to compensate for the reduction of inequality in income per recipient. The correlation coefficient p between log YA/lnh and log nh/mh is defined as: p[var(log y/f/nj] 1/2 [var (log nh/mrn)] 1/2 = cov(log Yh/nh, log nh/mh). Substituting in the variance equation (1) above, and differentiating partially with r espect to var(log yh/nh) under the assumption of a constant p,6 we get: avar (log ylmh) I [var(log nh/mh)]1 /2 Svar(log Yh/nh) [var (log yh/Inh)] 1l2 > 0 if var(logyhlnh) > -cov(logyh/nh, log nh/mn). The latter condition is easily satisfied in Malaysia (0.7154 > 0.1476), as can be verified from table 6-1. Provided the negative correlation coefficient persists, therefore, a change of inequality in income per recipient should lead (other things being equal) to a change of inequality in per capita household income in the same direction. In Malaysia the average number of income recipients per household is 1.651, and the average participation rate, or proportion of income recipients, per household is 40.3 percent (table 6-1).7 Malay households 5. However, since poor households have significantly fewer income recipients than nonpoor households (table 4-3), there might be considerable scope for alleviating poverty by raising the per capita incomes of poor households through policies which affect their participation rate in paid employment. 6. This might be a reasonable prediction to make on the basis of a household-utility- ma.imizing theory of participation rates, in which the participation rate turns out to be negatively related to income per recipient. 7. The reciprocal of the participation rate is the dependency ratio, which shows the number of household members mb supported by income recipients nh. But, since the average value of a reciprocal is not equal to the reciprocal of the average value, the average dependency ratios in Malaysia cannot be inferred from the average participation rates shown in table 6-1. Table 6-1. Means and Varlogs of per Capita Household Income, Household Income per Recipient, and Household Participation Rate 2cov (log Racial Var (log Var (log Var (log yh/th, Sample group yh if h h (yh/mh) (Yh/fln) (nh/m.) yhlmh) yAhn,) nh/m.) log nhIm,) p size, All households 2679 5.381 1.651 61.9 169.1 0.403 0.8453 0.7154 04251 -0.2952 -0.2676 24,682 Malay 174.5 5.097 1.481 41.0 124.2 0.378 0.6921 0.6624 0.3931 -03634 -0.3545 13,673 Chinese 399.6 5.863 1.919 86.7 228.8 0.431 0.6651 05371 0.4572 -0.3292 -0.3322 7.893 Indian 307.8 5.472 1.731 78.6 188.2 0.434 0.8050 0.5373 0.4631 -0.1954 -0.1959 2.904 Other 847.3 4.528 1.542 256.8 584.1 0.442 3.1277 2.5857 0.4020 +0.1400 +0.0687 212 a. The 341 zero-income households have been excluded from the calculations in this table. See the related estimates in tables 3-3 to 3-5 based on all 25.023 households. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 191 have a significantly lower participation rate (37.8 percent) than either Chinese households (43.1 percent) or Indian households (43.4 percent).8 There is also less inequality in the participation rate among Malay households tvar(lognh/nih) =0.3931] than among Chinese households (0.4572) or Indian households (0.4631). But Malay households display a stronger negative association between participation rate and income per recipient (correlation coefficient of -0.3545) than the other races.9 If correlation coefficients remain stable, participation rates for Malay house- holds are likely to fall more than those for other households in response to an increase in income per recipient. For each racial group separately, a large part of the inequality in per capita household income arises from inequality in income per recipient- notwithstanding the negative covariances (table 6-1). The inequality in income per recipient is therefore likely to play an important role in the explanation of differences in levels of living among households. But the distribution of households by income per recipient is not a tractable distribution for purposes of analysis. Since personal income is received directly by individual recipients, the individual is clearly a more satisfactory popu,'iation unit than the household. Two types of individual unit may be distinguished in an analysis of personal income inequality: the actual recipients and all individuals in the population. The difference between the distribution of actual recipients by perscinal income and the distribution of all individuals by personal income is the addition of dependents who receive no personal income.'" The addition of zero-income individuals to the distribution of recipients leads to an increase in the percentage of people falling below each income level, but it leads to no change in aggregate income. Thus to each previous cumulative income share there now corresponds a larger cumulative population share, and the new Lorenz curve lies horizontally to the right of the old one (see appendix E, lemrna 2 for a formal proof). This implies that the distribution 8. The significance of these racial differences in average participation rate has been determined at the 95 percent level of confidence, using a 1-test based on the standard deviation of the estimates. 9. This finding may be related to the possibility that the PES attributed the income from a family farm or business entirely to the household head, and that unpaid family workers were not in receipt of income from other sources and thus do nol count as income recipients. The participation rate in such households would tend to be underestimated and the income per recipient correspondingly overestimated. The larger negative correlation between the logarithms of these variables for Malay households may thus be partly due to the disproportionate number of family farm or family business households among the Malays. 10 Although such individuals receive no personal income (that is, income from one or more of the eleven sources listed in chapter 2. the section "Definition of PES Income"), they obviously get that part of household income which is transferred to them as household members. 192 INEQUALITY AND POVERTY IN MALAYSIA of income recipients by personal income Lorenz-dominates the distribution of all individuals by personal income. Thus the former distribution shows less inequality than the latter distribution according to any index that satisfies mean independence, population size independence, and the Pigou- Dalton condition." Since the relation between the two distributions is straightforward, we can drop the zero-income individuals and confine our attention to the distribution of actual recipients. Before doing so, however, note the relation between the distribution of all individuals by personal income and the distribution of all individuals by per capita household income (chapter 3). It can be shown that the distribution of all individuals by per capita household income always Lorenz-dominates their distribution by personal income (see appendix E, lemma 3). This follows from a direct application of Atkinson's theorem (appendix D), because the first distribution is obtained from the second distribution by a sequence of progressive transfers from rich to poor. The transfers in this case are within each household from the high-income recipients to the low- or zero-income recipients. Figure 6-1 shows the Lorenz curves for (a) the distribution of income recipients by personal income and (b) the distribution of all individuals by per capita household income. Although both distributions Lorenz- dominate the distribution of all individuals by personal income, in general there is no relation between the two distributions themselves."2 The Distribution of Income Recipients by Personal Income PES data for the personal income distribution have been coded as absolute frequencies in thirty-two income intervals (see table 2-1 and "The Coding of PES Income Data" in chapter 2). As in the case of the household income distribution, I estimated interval means for the upper-income classes of the personal income distribution by fitting a theoretical Pareto distribution. The same procedure was adopted as in chapter 2 of maximizing the portion II In this sense, Lorenz dominance implies unambiguous" ranking of inequality (see the proposition in appendix D). 12. Although in figure 6-1 distribution (b) appears to Lorenz-dominate distribution (a), this is not the case In fact. for a short stretch at the bottom, the Lorenz curve for distribution (a) lies above that for distribution (b). The reason is that about I percent of individuals in distribution (b) have no income (see chapter 3) whereas everyone in distribution (a) receives positive income. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 193 Figure 6-1. Lorenz Curves for the Personal and per Capita Household Income Distributions 100 (a) Distribution of income recipients by 90 personal income Mean = MS163 per month Gini coefficient = 0.5063 -- (b) Distribution of individuals by per cap- 8,0 - ita household income Mean = MS5O per month Gini coefficient =0.4980 70 2 E 0 50/ *~40 E 30 20 10 0 10 20 30 40 50 60 70 so 90 100 Cumulative population (percent) of the distribution fitted from above, subject to an k2 greater than 0.99. This led to the following equation: log n(y) = 8.8 - 2.0559 log y (56.48) where n(y) denotes the number of recipients with personal income greater than or equal to y; the t-ratio is shown in parentheses, R2 = 0.994, and the degrees of freedom = 21. This equation refers to the top twenty-three income intervals (more than M$200 per month) and accounts for 21.7 percent of all recipients. Mean incomes for the top twenty-three intervals were calculated by assuming that a Pareto distribution with the coefficient of 2.0559 was valid in this range; hence the mean calculated for the open- Table 6-2. Distribution of Income Recipients by Personal Income Percentage share of A rithmetic Sample size mean income Gini Lowest Lowest Highest Highest (number of income Racial group (MS per month) coefficient quintile 40 percent quintile 5 percent recipients) All income recipients 163.0 0.5063 3.5 11.3 55 0 28.5 40,806 Malay 118.2 0.4751 3.3 12.8 51.9 23.5 20,321 Chinese 209.1 0.4908 4.0 12.6 54.6 27.5 15,190 Indian 180.4 0.4693 5.0 14.8 54.2 29.0 4,994 Other 573.5 0.7048 0.5 3.9 75.5 31.0 301 INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 195 ended class M$5,000 and over per month was M$9,736. Mean incomes for the remaining eight intervals at the bottom of the distribution were taken to be at interval midpoints. The means for each income interval are shown in table 2.-. I have also conducted a sensitivity analysis of the personal income distribution with respect to alternative assumptions for interval mean in the upper open-ended class M$5,000 and over per month (compare table 2-1 and the section "Household Income Inequality in Malaysia" in chapter 2). The results are: Interval mean assumedfor Mean of Gini coefficient Theil T index incomdn class personal of personal of personal M$5,000 and income income income over per month distribution distribution distribution 5,000 162 0.5024 0.5103 9,736 163 0.5063 0.5360 10,000 163 0.5065 0.5375 15,000 164 0.5105 0.5686 20,000 166 0.5145 0.6019 As the assumed mean for the top income interval (that is, the income of the richest persons) increases, so does inequality in the Lorenz sense (see appendix E, corollary 1), and hence inequality as measured by the Gini coefficient and the Theil Tindex, which belong to the Lorenz class of indices (see the proposition in appendix D). Table 6-2 summarizes the distribution of income recipients by personal income in terms of traditional inequality indices such as the Gini coefficient and the shares of various fractile groups. For the same distribution, table 6-3 presents the equally distributed equivalent income and Atkinson inequality index for values of E between 0 and 3. In each table there is a breakdown by ethnic group. The arithmetic mean of the personal income distribution in Malaysia is M$ 163 per month, and the Gini coefficient is 0.5063. The income share of the lowest quintile is 3.5 percent while that of the top quintile is 55.0 percent; this implies a ratio of almost 16 between the top and the bottom quintiles. For different values of the inequality aversion parameter £, the Atkinson index is as followvs: 0.2196 for £ = 0.5,0.3785 for £ = 1, 0.5975 for e = 2, and 0.7159 for E = -3.' These indices all point to a high degree of personal 13. In chapter 3 the Atkinson index could not be computed for any value of E greater than unity because of the presence of zero-income people in the distribution of individuals by per capita household iicome There are, however, no zero-income people in the personal income distribution, which is defined over actual recipients, and the Atkinson index can be computed for all values of a > 0. In table 6-3 notice the comparatively,small change in going from E = 0.9 to C = 0.99 or E 1.00 ln table 3-9, however, there is a sharp drop in 3'EDE and a sharp rise in I as e goes from 0.9 to 0.99; further, when E = I 00. FEDE = 0.0 and I 1.0. Table 6-3. Equally Distributed Equivalent Income and Atkinson Inequality Index for the Distribution of Income Recipients by Personal Income (M$ per month) Sample size F (number of Income income recipients 0.00 0.50 0.90 0.99 1.00 1.50 2.00 2.50 3.00 recipients) All 163.0 127.2 105.9 101.7 101.3 81.1 65.6 54.3 46.3 40,806 (0.2196) (0.3503) (0.3760) (0.3785) (0.5024) (0.5975) (0.6668) (0.7159) Malay 118.2 95.7 81.0 78.1 77.7 63.4 52.5 44.6 39.1 20,321 (0.1903) (0.3147) (0.3392) (0.3426) (0.4636) (0.5558) (0.6226) (0.6692) Chinese 209.1 166.1 140.1 134.9 134.3 108.6 87.5 71.1 59.0 15,190 (0.2056) (0.3299) (0.3548) (0.3577) (0.4806) (0.5815) (0.6599) (0.7178) Indian 180.4 145.2 126.0 122.3 121.9 103.8 88.3 74.8 63.5 4,994 (0.1951) (0.3015) (0.3220) (0.3242) (0.4246) (0.5105) (0.5853) (0.6480) Other 573.5 330.2 196.2 174.7 172.5 98.4 66.6 51.4 43.1 301 (0.4242) (0.6578) (0.6953) (0.6992) (0.8284) (0.8838) (0.9103) (0.9248) Note: The Atkinson inequality index is shown in parentheses. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 197 income inequality in Malaysia: this inequality is higher than the inequality in per capita household incomes (chapter 3). It should be stressed that all the indices used here, including the Atkinson index, refer to a statistical and not a normative measurement of inequality. The underlying distribution is in any case not the relevant one for a measurement of inequality based on social welfare. Thus even the Atkinson index is used here in a purely positive sense to summarize inequality in the personal income distribulion.'4 The ethnic breakdown of the personal income distribution shows a Chinese mean income (M$209) which is 1.77 times the Malay mean income (M$1 18). At the same time, the Indian-Malay ratio of mean incomes is 1.53. These racial disparity ratios for personal income are smaller than for per capita household income, where they are 2.00 and 1.65, respectively. The reason is that the participation rates of the Chinese and Indian communities are higher than that of the Malay community (table 6-1).'5 Table 6-4 presents the racial disparity ratios according to the three different income concepts. The disparity ratios were reduced in going from household income to per capita household income because of a larger average household size (n-) for the Chinese and Indians than for the Malays (chapter 3). They are further reduced in going from per capita household income to personal income because of a higher ratio of average number of Table 6-4. Racial Disparity Ratios Per e apita Household income household income Personal income' (Y&) (YhlThs) (YJb/nl (Zhinese-Malay 2.29 200 1.77 Indian-Malay 1 77 1.65 1.53 a. There are negligible discrepancies between these disparity ratios and the same ones calculated from table 6-1. The discrepancies arise from independent estimation of the Pareto coeilficients and interval means for the household income distribution and the personal income distribution (see table 2-l). For exarnple, the mean personal income in Malaysia as calculated from the household income distribution is yh/ni, = M5162.3 (table 6-1), but M$163.0 when calculated directly (table 6-2). Source: Tables 3-7, 3-9, and 6-2. 14. The Atkinson index for constant elasticity of marginal utility e > 0 satisfies all the desirable properties for a "positive" inequality measure specified in appendix A: mean independence, population-size independence, and the Pigou-Dalton condition. 15. This should be interpreted with somecaution. Theparticipation rate of Malays may be underestimated and their mean personal income overestimated, relative to that of Chinese and Indians, because olfthe nonimputation of income to unpaid family workers (see note 9 above) Hence racial disparity ratios for personal income may be underestimated. 198 INEQUALITY AND POVERTY IN MALAYSIA income recipients to average household size (hh/inih) for the Chinese and Indians than for the Malays. Hence, in terms of incomes that can be influenced by economic policy, the disparity ratios are a good deal smaller than is popularly believed. These considerations also suggest that the between-race contribution to personal income inequality should be smaller than the between-race contribution to per capita household income inequality, unless, of course, it happens that within-race inequality in personal income is dispropor- tionately small. But this is not the case, as can be seen from tables 6-2 and 6-3. Inequality within each ethnic group is high, with the Chinese showing a consistently higher level than the Malays, who in turn show more inequality than the Indians. (Among recipients of other races, inequality is, not surprisingly, extremely high.) The between-race contribution to personal income inequality does indeed turn out to be smaller than the between-race contribution to per capita household income inequality; this can be verified in the later section "Decomposition by Race and Location." The Interpretation of Decomposition for Three Inequality Measures In chapter 3 the decomposition of per capita household income inequality was carried out by means of the Theil entropy index T. This index is decomposable in the weak sense only, however, and not according to my strict definition (see "The Methodology of Inequality Decomposition" in chapter 3). Thus, although the T index can be written as the sum of a between-group component and a within-group component, the within- group component is not the value of the T index when all between-group income differences are suppressed. The within-group component for the T index is a weighted average of the T indices for each group where the weights are the income shares of the groups in total income. The interpretation of decomposition in the weak and strict senses is immediate if one writes out the decomposition formulas for the weakly decomposable Theil Tindex and the strictly decomposable Theil L measure. Adopting the notation of chapter 3, divide the population into two groups, say, Malays and non-Malays, with mean incomes P1, P2 and population sizes n,, n2, respectively."6 Let the overall mean income bep and the overall population size n. Let the subscripts 1, 2 on an inequality index denote the value of the index for groups 1, 2, respectively. 16. There is no loss of generality in partitioning the population into two groups only: the extension for more than two groups is obvious and follows by induction. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 199 The Theil T index can be expressed as the sum: 7 = TW + TB where Tw = nly, T, + 2.02 T2 nil nfu and TB = np log IO ! + n2,2 log! 2. n,u du ny T8f is the between-group component, since it is the value of the Tindex when all within-group income differences are suppressed. If everyone in group I gets income ,, and everyone in group 2 gets income u2, the Tindex is easily seen to be equal to TB. (This follows directly from the definition of the index, or by putting T, = T2 = 0 in the above expression.) However, T, is not the value of the T index when all between-group income differences are suppressed but inequality within each group remains constant. If the group mean incomes At, P2 are equalized to the overall mean p by equipropor- tionate changes in the income of every person within a group, the T index reduces to (n I/n) T, + (n2 In) T2. (This follows directly from the definition of the index, or by putting j1u = u, = p in the above expression for T.) BecaLuse it is mean-independent, the same value of the T index results if the group means are equalized to any level other than u, for example, P2. Thus the elimination of between-group inequality -(for example, by multiplying all group I or Malay incomes by a factor of 127111) results in a value for the T index which is not Tw [= (nlp,lnp) T, + (n2p2/np) T2]. Hence T is not decomposable in the strict sense, and therefore T, does not measure the reduction in inequality if between-group income differences are elim- inated." Of course, TB does still measure the extent of inequality arising from the between-group (Malay/non-Malay) differences in mean income. This is because r is additively decomposable in the weak sense. T he Theil L measure can be expressed as the sum: L = LW+ LB where L,n,-LI +-L2 n n and LB= nl log y + n2 log 1. n u, n P2 LB is clearly the between-group component, since it is the value of the L measure when all within-group income differences are suppressed. If 17. The elimination of between-group income differences leads to a reduction in overall inequality of T- (ni/n)T, -(n2.n)T2 = Tq+[T;w-(tn;/n)T, -(n2n)TO2] # 7B 200 INEQUALITY AND POVERTY IN MALAYSIA everyone in group 1 gets income hi, and everyone in group 2 gets income P2. the L measure is equal to LB. (Again, this follows directly frorn the definition of the L measure, or by putting L, = L2 = 0 in the above expression.) Furthermore, Lw is indeed the within-group component according to the strict definition. When all between-group income dif- ferences are suppressed by the equalization of group mean incomes, but inequality within each group is kept constant, the L measure reduces to (n1/n)L, + (n2/n)L2 = L w (Again, this follows directly from the definition, or by putting A, = p, = u in the above expression for L.) Since L is the sum of the between-group component, LB, and the within-group component according to the strict definition, L, the measure is additively decomposable in the strict sense. This property allows us to interpret unambiguously the between-group component LB. By one interpretation, LB measures the extent of inequality which arises if between-group differences in mean income are the only source of income variation. In other words, it is the value of the L measure if inequality is eliminated within each group, but the group mean incomes remain constant. By another interpretation, LB measures the reduction in overall inequality if between-group differences in mean income are eliminated, but inequality within each group remains constant. These Iwo interpretations of the between-group component give the same answer for an inequality index decomposable in the strict sense. The two interpretations are not consistent for measures that are decomposable in the weak sense only. For these measures the within-group component is defined to be a weighted sum of the inequality indices for each group, where the weights are a function of the population share and income share of the groups. If the weights depend on the population shares alone, then the measure is decomposable in the strict sense, as verified for L above. Otherwise, the two interpretations of the between-group component lead to different answers. The elimination of inequality within each group while holding group mean incomes constant yields the between-group corn- ponent. But the elimination of between-group differences in mean income while holding inequality within each group constant does not yield the within-group component. Although the inequality indices for each group remain constant, the weights on these indices change if group mean incomes are equalized. When the mean income for each group is the same, the income share of a group collapses to its population share, which changes the weight on its inequality index (unless the income and population shares are the same to begin with, in which case there is no between-group inequality). When the two interpretations lead to the same answer, there is an unambiguous meaning to the between-group contribution to overall inequality. For example, the contribution of between-race differences in INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 201 mean income to overall inequality can be measured either as the inequality which arises when these differences are the only source of income variation, or as the amount. by which inequality falls when these differences are eliminated. While it is convenient to have an index whose between-group component gives the answer to both questions, thereby removing any ambiguity, it is clearly not essential. In fact, either question can be answered using any index, decomposable or not. Decomposability is an attractive property which helps circumvent some of the computations required to answer the different questions, including the extent of inequality within groups. The variance of log-income is another index which can be built up as the sum of inequality in its constituent parts. Let V be the overall variance of log-income, and VI, V2 the variance of log-income for groups 1, 2, respectively. Let A be the overall geometric mean income, and Al, A2 the geornetric mean incomes of groups 1, 2, respectively. Then V can be expressed as the sum V VK + VB where V_,,= IVI + _2 V2. n n and V_ nI (log -_log p)2+ ±2 (log10 2 -log p)2. n n V". is a weighted average of the variance of log-income within each group where the weights are population shares of the groups. However, VB is not the variance of log-income when all within-group income differences are suppressed but group arithmetic mean incomes are held constant. Rather, VB iS the value of V when everyone in group 1 gets the geometric mean A, an(d everyone in group 2 gets the geometric mean A2. Thus VB measures the inequality which arises from between-group differences in geometric mean income, not in arithmetic mean income. ][t is easy to verify that the variance of log-income V is, in fact, decomposable in the strict sense around the group geometric means. But it is not decomposable around the group arithmetic means. The geometric mean income is not equal to the arithmetic mean income, and the equalization of group geometric mean incomes through equiproportionate changes in the income of every person within a group does not lead to the equalization of group arithmetic mean incomes. Nevertheless, the between- group component VB provides the answer to either question concerning between-group inequality in geometric mean incomes. VB measures the extent of inequality if the between-group differences in geometric mean income are the only source of income variation and inequality within each group is set to zero. VB also measures the reduction in overall inequality if 202 INEQUALITY AND POVERTY IN MALAYSIA inequality within each group is held constant but the between-group differences in geometric mean income are eliminated through equipropor- tionate changes in the income of every person within a group. This is because the weights on the within-group variances of log-income are population shares, which do not change with changes in group mean incomes. Decomposition by Race and Location Personal income inequality in Malaysia has been decomposed by using all three inequality measures discussed in the previous section: Theil's entropy index T, Theil's second measure L, and the variance of log-income V, or varlog. Unlike the per capita household income distribution in chapter 3, the personal income distribution allows all three measures to be computed because it has no zero-income recipients. 18 Table 6-5 presents a breakdown of the personal income distribution by race as well as location. The between-race contribution to personal income inequality is 9.2 percent by Theil T, 9.6 percent by Theil L, and 7.9 percent by varlog. 9 The fact that Theil L yields a between-race contribution of'9.6 percent implies that if the between-race differences in arithmetic mean income were eliminated, but inequality within each race remained the same, the reduction in overall inequality would be 9.6 percent. The fact that varlog yields a between-race contribution of 7.9 percent implies that if the between-race differences in geometric mean income were eliminated, but inequality within each race remained the same, the reduction in overall inequality would be 7.9 percent. The NEP'S racial balance objective can be taken to imply the elimination of inequality in arithmetic mean incomes between the races, leavilg unchanged the inequality within each race (see chapters I and 3). The implementation of this objective would therefore bring about a reduction of 9.6 percent in the strictly decomposable Theil L measure.20 This result supports the conclusion in chapter 3 about the limited effect of raci..al balance on individual income inequality, especially since policy can influence only the personal income distribution. 18 As noted in chapter 3, the Theil L measure and varlog blow up to infinity when there are zero incomes. 19. For the index A(E) based on the Atkinson equally distributed equivalent income (see chapter 3), the between-race contribution to personal income inequality is: 10.1 percent for e = 0.5; 9.8 percent for c = 0.9; 9.6 percent for E = 0.99 (computed from table 6-3). 20. The reduction in varlog and the weakly decomposable Theil Tindex have to be specially computed; this can be done from the data given in table 6-5. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 203 The Theil entropy index Tis a measure which can be used to decompose both personal income inequality and per capita household income inequality whereas PES data make computation of Theil L and varlog impossible for the per capita household income distribution. As predicted earlier, the between-race contribution to personal income inequality is indeed smaller than the between-race contribution to per capita household income inequality: 9.2 percent compared with 13.0 percent according to the Theil T index.21 Thie between-location contribution to personal income inequality in Malaysia is approximately the same as the between-race contribution: 9.5 percent by Theil r, 10.0 percent by Theil L, and 7.3 percent by varlog. Thus if rural-urban inequalities were completely removed, but intrarural and intraurban inequality remained the same, the reduction in overall in- equality would be 10 percent. Hence it cannot be claimed that rural-urban differences in income are responsible for more than a small fraction of overall inequality in the country.22 The racial disparity ratios for the country as a whole are distinctly larger than those for urban areas. For instance, the Chinese-Malay disparity ratio for Malaysia is 1.77, whereas for metropolitan towns it is 1.05 and for towns 1.26. A disproportionate presence of Malays in rural areas where the average income is relatively low and of Chinese in urban areas where the average income is relatively high, together with a Chinese-Malay disparity ratio of 1.70 for rural areas, 23 accounts for the higher overall disparity ratio of 1.77. To reduce such overall racial income disparities, the government has declared its desire to promote migration flows and urbanization of Malays (MTR, chap. 1). The smaller disparity ratios for urban areas also bring about low bet-ween-race contributions there. Race accounts for less than 2 percent (1.9, 1.8, and 1.5 percent by the three measures, respectively) of income inequality in towns, and less than 5 percent (4.5, 3.6, and 1.7 percent, respectively) of inequality in metropolitan towns. It follows that the elimination of racial disparities within urban areas would by itself have a negligible effect on inequality. The effect should be somewhat larger (8.4 percent) in rural areas, however, since they have higher racial disparity ratios. A regional decomposition of the personal income distribution was also undertaken, though it is not shown here. There are significant differences in 21. See tables 6-5 and 3-10, respectively. 22. Contrast the views on this subject in Lipton (1968) or (1977). 23. The higher disparity ratio in rural than in urban areas perhaps arises in part from greater racial disparities in asset-holding in rural areas. Table 6-5. Decomposition by Race and Location of Inequality in the Personal Income Distribution 8etween-location contribution to Metropolitan Rural Peninsular inequality Racial group Item towns Towns areas Malaysia (percent) Malay Gini 0.4610 0.4838 0.4509 0.4751 Theil T 0.4724 0.4134 0.3800 0.4370 9.4 Theil T Theil L 0.3961 0.4583 0.3765 0.4193 8.3 Theil L Varlog 0.7418 1.0057 0.7589 0.8293 6.1 Varlog Geometric mean 163.4 107.7 71.2 77.7 Arithmetic mean 242.8 170.3 103.7 118.2 Sample size 1,297 1.711 17,313 20,321 Chinese Gini 0.5246 0.4936 0.4487 0.490S Theil T 0.5772 0.4848 0.3945 0.4958 2.7 Theil T Theil L 0.5034 0.4527 0.3734 0.4430 3.0 Theil L Varlog 0.9213 0.8896 0.7503 0.8428 1.3 Varlog Geometric mean 154.7 136.4 121.9 134.3 Arithmetic mean 255.8 214.4 177.1 209.1 Sample size 4,665 3,194 7,331 15,190 Indian Gini 0.5242 0.4481 0.3853 0.4693 Theil T 0.6069 0.3506 0.3335 0.4998 11.4 Theil T Theil L 0.4943 0.3794 0.2722 0.3925 13.7 Theil L Varlog 0.8465 0.8286 0.4812 0.6571 7.5 Varlog Geometric mean 173.6 139.5 104.1 121.9 Arithmetic mean 284.5 203.9 136.7 180.4 Sample size 1,244 514 3,236 4,994 Other Gini 0.5858 0.5809 0.7765 0.7048 Theii T 0.6206 0.6242 1.2736 0.9442 9.3 Theil T Theil L 0.7647 0.6887 1.4578 1.20;6 7.0 The!l L. Varlog 1.8184 1.4565 2.3611 2.5426 19.4 Varlog Geometric mean 432.1 210.5 91.8 172.5 Anthmetic mean 928.3 419.1 394.3 573.5 Samnle size 99 44 158 301 All racial groups Gini 0.5259 0.4912 0.4655 0.5063 Theil T 0.5945 0.4642 0.4320 0.5360 9.5 Theil T Theil L 0.5047 0.4576 0.4035 0.4763 10.0 Theil L Varlog 0.9041 0.9392 0.7918 0.8967 7.3 Varlog Geometric mean 161.5 127.4 85.7 101.3 Arithmetic mean 267.5 201.3 128.4 163.0 Sample size 7,305 5,463 28,038 40,806 Between-race- and-location contribution to inequality (percent) Between-race Theil T 4.5 1.9 8.6 9.2 15.0 Theil L contribution to Theil L 3.6 1.8 8 4 9.6 15.7 Theil L inequality Varlog 1.7 1.5 7.3 7.9 12.2 Varlog (percent) Note: Metropolitan towns are Johore Bahru, Malacca, Kuala Lumpur, Klang, Ipoh, and Georgetown. Towns are all others of more than 10,000 population (see chapter 2, note 18, for a listing). All other areas are considered rural. 206 INEQUALITY AND POVERTY IN MALAYSIA mean income between the states, but these account for only about 8 percent of inequality in the country. Government strategy for regional development, "to reduce the marked economic disparities which currently exist between States" (MTR, p. 18), is thus unlikely to reduce personal income inequality significantly. Development of the rice-growing northern states (Kedah, Kelantan, Perlis, and Trengganu) is likely to reduce poverty, however, which is heavily concentrated there (see chapter 4). It is also likely to reduce racial income disparities because the population of these states is predominantly Malay.2" The policy of developing these largely Malay backward states is therefore consistent with both prongs of the NEP, provided the gains do in fact accrue uniformly among their populations and are not captured disproportionately by the upper-income groups (such as Chinese traders, millers, and middlemen) at the expense of Malay peasants. Decomposition by Sex of Income Recipient A disaggregation of the personal income distribution by sex of income recipient has also been carried out (see table 6-6). It shows that for personal income the male-female disparity ratio in Malaysia is 1.96, which is even higher than the racial disparity ratio of 1.75 between the non-Malays and Malays. The male-female disparity ratio within each racial group is 2.08 among Malays, 1.94 among Chinese, 1.88 among Indians, and 3.66 among other races. Of the total number of income recipients in the sample, only 30.2 percent are females and 69.8 percent males. The male-female breakdown of income recipients within each racial group reveals an interesting pattern: females constitute 28.6 percent of Malay income recipients. 32.9 percent of Chinese. 28.8 percent of Indian, and 20.6 percent of other races. These figures suggest a significant difference in female participation rates between Chinese on the one hand, and Malays and Indians on the other-presumably because of religious and cultural factors. 25 The between-sex contribution to personal income inequality in the country is 7.2 percent by Theil T, 9.1 percent by Theil L, and 10.4 percent 24. Of the three major ethnic groups, regional disparities are largest among the Malays. The between-state contribution to Malay inequality is about 9 percent, while it is only about 2 percent for the Chinese, and about 4 percent for the Indians. 25 This phenomenon probably helps explain the lower average number of income recipients among Malay and Indian households than among Chinese households (table 6-1), which in turn explains the higher racial disparity ratios for household income than for personal income (table 6-4). INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 207 by varlog. Hence the elimination of male-female differences in mean income will bring about a reduction in individual income inequality of about 10 percent. The between-sex contribution to personal income inequality seems to be of the same order of magnitude as the between-race contribution. The combined between-race-and-sex contribution to overall income inequality is 16.8 percent by Theil T, 19.3 percent by Theil L, and 19.3 percent by varlog. This indicates the proportion of inequality attributable to mean income differences among the race-and-sex groups, however these differences may have come about. Breakdown by Employment Status As defined earlier, an income recipient is any person who obtains income from one or more of the eleven sources listed in PES. For instance, the person could be an employer receiving profits from his or her enterprise, or an employee receiving a wage or salary, or even an unemployed person receiving a transfer. In this section income recipients are considered according to their employment status. Since income data in PES were not coded by source of income, the employment status breakdown might be used to shed some light on particular factor income distributions. lThe PES perrrmits disaggregation into eight categories of employment status (Department of Statistics, 1973). Here the PES categories of unpaid family worker, student, housewife or houseworker, unemployed, and other have been aggregated into a single category called "other." Table 6-7 thus shows four major categories; two of these, employees and own-account workers, account for more than 90 percent of all income recipients. It seems reasonable to assume that personal income of employees consists largely of wage income, though some investment or interest income may also accrue from accumulated savings; thus the personal income distribution among employees may be a reasonably good surrogate for the wage income distribution. The income of other groups, however, contains significant elements of both labor and property income, though the actual proportion varies according to the activities of the group. In the case of employers, it seems plausible to assume that their income consists mainly of returns to capital invested. But in the case of own-account (that is. self-employed) workers, it is difficult to determine the proportion of income that is a return to capital and the proportion that is a return to labor and other factors. (Even in theory this imputation is difficult if the markets for capital, labor, and other factors are not perfect.) For own-account workers, capital assets typically consist of land (rural smallholders), shop (urban shopkeepers), or transport equipment (transport operators). The Table 6-6. Decomposition by Race and Sex of Inequality in the Personal Income Distribution Between-sex contribution to inequality Racial group Item Male Female All persons (percent) Malay Gini 0.4393 0.4870 0.4751 Theil T 0.3853 0.4424 0.4370 9.7 Theil T Theil L 0.3537 0.4128 0.4193 11.6 Theil L Varlog 0.6929 0.7229 0.8293 15.4 Varlog Geometric mean 97.5 44.2 77.7 Arithmetic mean 138.9 66.8 118.2 Sample size 14,503 5,818 20,321 Chinese Gini 0.4773 0.4470 0.4908 Theil T 0.4783 0.3671 0.4958 8.1 Theil T Theil L 0.4169 0.3615 0.4430 10.0 Theil L Varlog 0.7831 0.7168 0.8428 9.7 Varlog Geometric mean 164.0 89.4 134.3 Arithmetic mean 248.9 128.3 209.1 Sample size 10,187 5,003 15,190 Indian Gini 0.4672 0.3874 0.4693 Theil T 0.4973 0.3246 0.4998 6.6 Theil T Theil L 0.3873 0.2767 0.3925 9.4 Theil L Varlog 0.6415 0.5024 0.6571 8.5 Varlog Geometric mean 141.5 84.1 121.9 Arithmetic mean 208.5 110.9 180.4 Sample size 3,558 1,436 4,994 Other Gini 0.6856 0.6247 0.7048 Theil T 0.8751 0.7659 0.9442 8.1 Theil T Theii L 1.1746 0.7953 1.2016 8.8 Theil L Varlog 2.6326 1.5261 2.5426 5.4 Varlog Geometric mean 208.4 83.2 172.5 Arithmetic mean 674.5 184.3 573.5 Sample size 239 62 301 All racial groups Gini 0.4885 0.4850 0.5063 Theil T 0.5108 0.4370 0.5360 7.2 Theii T Theil L 0.4355 0.4270 0.4763 9.1 Theil L Varlog 0.7969 0.8177 0.8967 10.4 Varlog Geometric mean 123.8 63.6 101.3 Arithmetic mean 191.4 97.5 163.0 Sample size 28,487 12,319 40,806 Between-race-and- sex contribution to inequality (percent) Between-race Theil T 10.3 10.9 9.2 16.8 Theil T contribution Theil L 11.1 11.5 9.6 19.3 Theil L to inequality Varlog 7.8 14.6 7.9 19.3 Varlog (percent) 210 INEQUALITY AND POVERTY IN MALAYSIA proportion of income that can be attributed to capital and other factors is likely to be different in each case. and there is little that can be concluded about the nonlabor income distribution in general. In the case of employers the personal income distribution might give some idea about the capital asset distribution. If it is assumed that employer income consists entirely of returns on capital (or that the proportion of capital income to total income is constant), and also that the rate of return on capital is constant, then the income and wealth distributions for employers will be identical. Generally, however, imperfec- tions of various kinds imply a variable return on wealth. In this situation, mapping the wealth distribution to the income-from-wealth distribution involves the addition of extra variance (or "noise"). If W is wealth, r the rate of return on it, and y the income from wealth, then y = rWand logy = logr+log W. Hence, var(logy) = var(log W) +var(logr)+2cov(log W, log r). With a zero or positive correlation between the logarithm of wealth and the logarithm of the rate of return, the income-from-wealth distribution will be more unequal than the wealth distribution. Extra information such as this is needed on the joint distribution of wealth and its rate of return before anything can be inferred about one distribution from a knowledge of the other. (Compare the exactly analogous problem in chapter 3 and appendix F of mapping the household income distribution to the per capita household income distribution.) The PES data on personal income distribution disaggregated by race and employment status are presented in table 6-7, which shows the standard inequality measures, geometric and arithmetic mean incornes, and sample size. The arithmetic mean income of employers (M$738.8) is almost four-and-a-half times larger than that of employees (M$166.6). The mean incomes of the two other groups are closer together and only slightly below the mean income of employees. This last fact, together with there being few employers with high incomes, explains the relatively small contribution of employment status to personal income inequality26 (1 1.9 percent by Theil T, 9.0 percent by Theil L, and 6.5 percent by varlog). By any of the measures computed, employer incomes are distributed more unequally than self-employed incomes, and these are distributed 26. The dependence of the between-group contribution to inequality on the mean income and population share of a particular group can be examined from the decomposition formula for the measure given in chapter 3 or appendix C. For example, other things being equal, the higher the mean income of a group whose mean is above the overall mean, the greater the INEQUALITY EN THE PERSONAL INCOME DiSTRlBUTION 21i more unequally than employee incomes. Thus nonlabor incomes seem less equally distributed than labor incomes in Malaysia. To move to the wealth distribution requires making assumptions about the rate of return. If there is relatively small variability in this, and little correlation between it and wealth, the wealth distribution in Malaysia should also be more unequal than the labor income distribution. The ethnic decomnpositions of these distributions are more interesting. Racial disparity ratiios are largest for own-account workers, and so also is the between-race contribution to inequality (24.3 percent by Theil T, 25.2 percent by Theil L, and 20.5 percent by varlog). This undoubtedly reflects wide disparities among the races in asset ownership, with the Malays, by and large, owning and cultivating paddy and rubber smaliholdings in rural areas, and the Chinese owning shops and other sales or service establish- ments in the urban informal sector.27 Among employers, too, the racial disparity ratios are fairly large. The Chinese, who account for 80 percent of employers, receive significantly more than the Malays, who account for only 10 percent of employers. Indian employers, who account for another 10 percent, receive slightly less than the Chinese. These disparities indicate larger average wealth holdings for the Chinese and Indians than for the Malays; the differences in income seem much too large to be explicable by differential rates of return.28 Although there are no data on the wealth distribution among individuals in Malaysia, MTR (table 1-4) reports a racial breakdown for the ownership of corporate and noncorporate assets in modern agriculture and industry. between-group contribution to inequality according to the Theil L measure. Adopting the notation of chapter 3, we can partially differentiate (L8/L) with respect to P to give 8(L8/L) L wn)w/ 1 I\ ap, L2 \n/ p f p The dependence of (LB/L) on the population share (n1/n) of group I can be simiLarly examined (the answer is somewhat messier to report). The same exercises can also be conducted for the other inequality measures which are decomposable. 27. Most own-account workers in rural areas probably cultivate paddy or rubber smaliholdings. As seen in chaptcr 5, paddy farms are owned mainly by Malays, while rubber smallholdings are distnbuted more evenly between Malays and non-Malays. Most own- account workers in urban areas probably operate small-scale family enterprises such as retalt establishments in the commercial sector and service or trucking businesses in the transport sector. These are owned to a large extent by the Chinese 28. Some of the differences in rates of return might arise from differences in entrepreneurial ability. More generally, in seeking to explain racial income dispanties for own-account workers and employers. factors such as initial endowment, savings propensity, and access to loans are likely to be important. Table 6-7. Decomposition by Race and Employment Status of Inequality in the Personal Income Distribution Between-employment- Own- All status contribution account income to inequality Racial group Item Employers Employees workers Othera recipients (percent) Malay Gini 0.6764 0.4788 0.3965 0.5139 0.4751 Theil T 1.1274 0.4323 0.2892 0.5005 0.4370 8.0 Theil T Theil L 0.9484 04371 0 2935 0.4735 0.4193 8.0 Theil L Varlog 1.6328 0.9067 0.6064 0 8621 0.8293 5.0 Varlog Geometric mean 163.6 92.6 66.5 51.5 77.7 Arithmetic mean 422.4 143.4 89.2 82.8 118.2 Sample size 68 10,660 7,952 1,639 20,319 Chinese Gini 0.4900 0.4537 0.4305 0.4993 0.4908 Theil T 0 4966 0.3965 0.3687 0.4553 0.4958 17.9 Theil T Thell L 0.4204 0.3729 0.3557 0.4782 0.4430 14.2 Theil L Varlog 0.7267 0.7275 0.7488 0.9988 0.8428 10.4 Varlog Geometric mean 505.0 123.3 164.6 93.1 134.3 Arithmetic mean 768.9 179.0 234.9 150.2 209.1 Sample size 548 10,280 3,076 1,286 15,190 Indian Gini 0.5617 0.4391 0.5114 0.5175 0.4693 Theil T 0.6916 0.4185 0.6444 0.4577 04998 7.4 Theil T Theil L 0.5568 0.3408 0.4834 0.5298 0.3925 6.5 Theil L Varlog 0.8724 0.5781 0.8101 1.1560 0.6573 3.7 Varlog Geometric mean 412.7 119.6 138 7 93.9 121.9 Arithmetic mean 720.2 168.1 224.9 159.6 180.4 Sample size 63 4,180 509 240 4,992 Other Gini 0.3283 0.6276 0.5391 0.6232 0.7048 Theil T 0.1980 0.7093 0.7330 0.6882 0.9442 28.2 Theil T Thei; L 0 2843 0.9621 0.5199 0 8780 1.2016 33.4 Theil L Varlog 0.7775 2.3924 0.7105 i 9840 2.5426 29 7 Varlog Geometric mean 1,333 7 322.1 52.2 150.2 172.5 Arithmetic mean 1,772.3 843.0 87.8 361.3 573.5 Sample size 6 169 95 31 301 All racial groups Gini 0.5205 0.4764 0 4821 0.5361 0 5063 Theil T 0.5612 0.4568 0.4685 0.5352 0.5360 11.9 Theil T Theil L 0.5024 0.4205 0 4283 0.5389 0.4763 9.0 Theil L Varlog 0.9537 0.8160 0.8225 1 0434 0.8967 6.5 Varlog Geometric mean 447.0 109.4 87.1 69 1 101.3 Arithmetic mean 738.8 166.6 133.6 118.4 163.0 Sample size 685 25.289 11,632 3.196 40.802 Between-race-and- employment-status contribution to inequality (percent) Between-race Theil T 3 1 7 2 24.3 10.6 9.2 21.7 Theil T contribution to Theil L 3.6 5.2 25.2 10.1 9.6 19.3 Theil L inequality Varlog 12.9 3.3 20.5 9.0 7.9 14.8 Varlog (percent) a. The category of "other" here consists of the five PES categories of unpaid family workers, students, housewives or houseworkers, unemployed persons, and others. 214 INEQUALITY AND POVERTY IN MALAYSIA Modem agriculture covers only estates under rubber, oil palm, coconut, and tea, with ownership specified in terms of total planted acreage; industry covers only manufacturing, construction, and mining, with ownership in terms of fixed assets (in millions of Malaysian dollars). (Other branches such as commerce, banking, and transport are not included under industry here.) I have converted the planted acreage figures for modem agriculture into money values at M$1,000 per acre, and added these to the fixed asset figures for industry. Foreign (that is, non-Malaysian) capital, which in fact accounts for 52 percent of estate acreage and for 51 percent of fixed assets in industry, is excluded from the following breakdown of asset ownership in Malaysia: Percentage share of Percentage share of assets in modern employer income from agriculture and all sources zndustry (table 6-6) Malay 19.4 5.7 Chinese 63.0 83.2 Indian 4.5 9.0 Other 13.1 2.1 Total 100.0 100.0 Employees are the most homogeneous group among the categories considered. Not only is the level of inequality in their incomes smallest, but so also are the racial disparity ratios and the between-race contribution to inequality. Among employees, the Chinese-Malay disparity ratio is only 1.25, and the Indian-Malay ratio is only 1.17. The between-race contri- bution to inequality among employees is 7.2 percent by Theil T, 5.2 percent by Theil L, and 3.3 percent by varlog. These disparity ratios for employee incomes indicate that only a small part of the overall disparity ratios for personal income are attributable to racial differences in labor income. It is the relatively large differences in nonlabor income and the dispropor- tionate distribution of the races by employment status that account for the higher overall Chinese-Malay disparity ratio of 1.77 and Indian-Malay ratio of 1.53. It is possible to list various factors that could account for observed disparities in labor income (see Atkinson, 1975). Because of the limitations of PES data, only two factors can be examined to any extent, namely, education and age (or experience). The next chapter contains a detailed analysis of education, age, and employee incomes in urban areas. Here I briefly mention some other explanations that are sometimes advanced for interracial disparities in labor income. INEQUALITY IN THE PERSONAL INCOME DIST RIBUTION 215 Interracial Earnings Differentials among Rubber Tappers Apart from differences in education and training generally, earnings differentials could also arise from intergroup differences in ability, opportunities, and tastes. While it is common to observe interpersonal differences in ability, there is no prima facie reason to expect any interracial differences in this attribute. In any case, it seems rather difficult to agree upon an economically relevant and operationally usable definition of ability which can permit empirical testing. The role played by opportunity also raises problems. For example, it is difficult to estimate the extent to which income differences among groups arise from discrimination, if any, in education or employment. Differences in tastes would appear to relate to differences in one of at least three major characteristics: propensity to save, risk aversion, and leisure. preference.29 The first two characteristics probably affect the incomes of employers and self-employed more than of employees, and are of no concern here.30 In regard to differences in leisure preference, there are some isolated data which allow a partial test. Thillainathan (1976) has compiled an interesting time series on the daily and monthly earnings of rubber tappers on estates.3" Since the activity is rather narrowly defined and payment is by piece rate, any differences in monthly earnings can probably be attributed to differences in productivity and in the number of days worked per month, assuming there are no systematic differences in employment practice, estate yields, or work-force composition. The data, covering a fourteen-year period, 1960-73, show a Chinese-Malay disparity 29. Consideration of these factors immediately reminds one of the inadequacy of current incomfe as a measurc: of welfare, static or dynamic. Current welfare depends on current consumption as well as leisure, among other things. Ignoring leisure generally overstates inequality in current-period utility, unless leisure and income are highly complementary To compensate for variations in leisure time among individuals and groups, a measure such as Kolni's *'leisurely equivalent income" is needed (see Kolm, 1969, pp. 181-82). Lifetime welfare is also not adequately captured by current income: different savings propensities can lead to different amounts accumulated (and bequeathed), and hence to different income and utility profiles over the life cycle. 30. To the extent that employee incomes include returns to capital, however, the propensity to save and risk aversion are indeed relevant factors. But there are not yet reliable data on savimgs, and testing for differences in risk aversion is likely to be even more difficult 31. The data have been compiled from the Annual Reports and Handbook of Labour Statistics of the Ministry of Labour and Manpower, Kuala Lumpur. 216 INEQUALITY AND POVERTY IN MALAYSIA ratio in average monthly earnings of 1.25, and an Indian-Malay ratio of 1.14; for average daily earnings, the corresponding disparity ratios are 1. 1 5 and 1.07, respectively (Thillainathan, 1976, pp. 47-48). Thus, in terms of latex output per day, the Chinese appear to be about 15 percent more productive than the Malays, and the Indians about 7 percent more productive than the Malays. When these differences in daily earnings are netted out, the rest of the monthly earnings disparities should be due to differences in the number of days worked per month.32 On average, therefore, the Chinese would seem to work about 9 percent more days per month than the Malays, and the Indians about 7 percent more than the Malays. This could be some indication of the relative preference between leisure and income among the three ethnic groups, at least in one major industry in the country.33 Decomposition by Occupational Category One of the objectives of NEP Prong 2 is to reduce (and eventually eliminate) the identification of race with economic function. Although economic function could be defined by a variety of characteristics, it has generally been associated with occupation and sometimes sector of employment. This section examines the extent to which the major racial groups are identified with particular occupations, and the effects of this on personal income inequality. Table 6-8 presents a breakdown of income recipients according to eight broad (one-digit) occupational categories, cross-classified by racial 32. In a quite different context, I have analyzed the effects of differences in leisure preference among workers on the shape of the optimal earnings schedule offered by an employer. These differences give rise to an unequal earnings distribution, which is less or more equal than the leisure preference distribution according to the convexity or concavity of the earnings schedule. See Anand (1970). 33. A body of data for the gravel pump mining industry during 1960-73 (Thillainathan, 1976, pp. 50-52) enables a comparison of the earnings differentials between Chinese and Indian unskilled workers. The jobs performed by the two groups are not specified exactly, and the data seem generally less robust than those for the rubber tappers. Since 1970 there is an average monthly earnings differential of 20 percent between the Chinese and the Indians, and a daily earnings differential of 17 percent. Thus, on average, the Chinese would seem to wo rk 3 percent more days per month than the Indians. The daily earnings differential of 17 percent is likely to be due at least in part to differences in personal characteristics between the groups. The differential could arise from differences in physical productivity as well as in daily overtime, but differences in skill mix between the groups are also possible, as this is an industry with a gradation of skills within the unskilled labor category. INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 217 group. " It shows the usual statistics for this breakdown, including the four inequality measures (Gini, Theil T, Theil L, and varlog), the geometric and arithmetic mean income, and sample size. The racial pattern of employment for all occupations shows the Malays accounting for 50 percent, Chinese 37 percent, Indians 12 percent, and others I percent (calculated from table 6-8). This pattern is almost identical to that af the racial proportions in the population of the country as a whole. The racial groups, however, are not represented in these proportions withiJn the various occupations, and severe imbalances exist in some. For instance. the Malays constitute 85 percent of all farmers but only 28 percent of all sales workers; and in the administrative and managerial, clerical and related, and production worker categories, the Malays account for little more than 30 percent of the work force. In contrast, the Chinese constitute less than 13 percent of farmers but some 60 percent of sales workers and 57 percent of production workers; further, they account for 46 percent of administrative and managerial petsonnel and also of clerical staflf but for only 25 percent of farm laborers (of whom 55 percent are Malays). Indians are over- represented in the administrative and managerial (21 percent), clerical and related (17 percent), and farm laborer (20 percent) categories, but are markedly underrepresented in the farmer (I percent) category. These imbalances in the racial pattern of employment, with the Malays concentrated in the less well-paid occupations and the Chinese in the better-paid ones, account for a large part of the income disparities between the races. Further, within each occupation there are sometimes significant differences in mean income among the races, with the Malays often, but not always, receiving less than the Chinese and Indians. This imbalance, which is examined below, compounds the effect on income disparities of the already uneven occupational distribution among the races. To reduce the wide differentials "generated by existing racial inequalities in the job- hierarchy, especially in the modern sectors of the economy," the govern- ment stresses that "increased efforts will be needed to progressively expand the professional, managerial and technical skills of the Malays so that their position in the occupational structure within each of the major sectors of the economy is enhanced" (MTR, p. 80). The labor market policies (such as racial quotas in employment and admission to institutes of higher education) associated with the specified restructuring of high- and middle- level Malay manpower are examined in chapter 8, in the context of a wider discussion of NEP Prong 2. 34. The note to table 6-11 at the end of this chapter contains a definition of the one-digit occupational categories. These were regrouped from the more detailed two-digit classification available in PES. Table 6-8. Decomposition by Race and Occupation of Inequality in the Personal Income DistribuLion Between- Pro- Adminms- occupation Jessional -rative Clerical contribution Racial and and and Sales Service Farm Production All to inequaluit group Item technical managerial related workers workers Farmers laborers workers occupations (percent) (1) (2) (3) (4) (5) (6) (7) (8) Malay Gini 0 3445 0 5206 0.3671 0.4747 0 4018 0.3652 0.3680 0.4039 0.4771 Theil T 0.2102 0.5962 0.2540 0.4295 0.2961 0.2318 0.2357 0.2782 0.4429 37.0 Thecl T Theil L 0 2560 0.4715 0.2811 0.4197 0.3435 0 2509 0.2582 0 3392 0.4209 32 0 Theil L Varlog 0.6555 0.7543 0.6808 0.8312 0.8211 0.5408 0.5567 08099 0.8267 23.2 Varlog Geometric mean 246.8 358 5 179.7 77.7 121.8 65.3 56.9 94 3 80 0 Arithmctic mean 318.8 574.5 238.0 118.2 171.7 83.9 73 7 132.3 121.9 Sample size 1,001 237 682 1,079 1,325 5,791 5,191 2,340 17,646 Chinese Gini 0.3808 0.4804 0.3638 0 5074 0.4899 0.4309 0.3070 0.3996 0.4903 Theil T 0.2690 0.4211 0.2389 0.5681 0.4890 03453 0.1753 03227 04960 24.7 Theil T Theil L 0.2713 0.4041 0.2451 0.4761 0.4359 0.3515 0.1836 0.3113 04396 24.9 Theil L Varlog 0.5812 0.7628 0.5326 0.8749 0.8065 0.7473 04013 0.6651 0.8290 19.7 Varlog Geometric mean 349.7 494.0 228.7 166.9 107.9 350.4 89 9 125.0 142.4 Anthmetic mean 458.7 740.0 292 2 268 7 166.9 213 7 108.0 170 7 221 0 Sample size 797 364 879 2,346 1,458 860 2,341 4,045 13.090 Indian Gini 0.4937 0.5006 0 3949 0.5175 0 4331 0.4750 0.2189 0.2972 0.4744 Theil T 0.4846 0.4561 0.2616 0.7049 03610 0.3870 00923 0.1811 05142 34,9 Theil T Theil L 0.4504 0 4180 0.3086 0.4883 0.3498 0.4353 0.0984 0 1680 0.3963 40.4 Theil L Varlog 0.8837 0.7076 0 7469 0.7526 0.7069 0 9490 0.2211 0 3403 0.6425 27 5 Variog Geometric mean 361.4 270.9 2!2.! 126 u iO4.7 103 3 89.5 152 4 125 7 Anthmetic mean 567 0 411.5 288.8 206.4 148 5 159.7 98 8 180.3 186 8 Sample size 288 177 329 457 501 78 1,885 680 4,395 Other Ginm 0.5119 0.3088 0.3593 07192 0.5243 0.3616 0.3422 0.6195 0.7032 Theil T 0.4357 0.1830 0.2277 1.0625 04818 0.2272 0.2422 07631 0.9363 600 Theil T The,l L 0.58!9 0.1735 0.3412 1.1401 0.6663 0.2294 0.2119 0.7452 1.2194 63.1 Theil L Varlog 1.4646 0.3318 0.9996 1 9189 1.6626 0.4502 0.3453 1!1738 2.6269 62.6 Varlog Geometric mean 573.3 2,235.7 189.1 101.2 341.6 43.7 28.6 220.1 181.4 Anthmetic mean 1,025.8 2,659.1 265.9 316.4 665.1 54.9 35.3 463 7 613.9 Sample size 60 21 32 14 23 72 17 23 262 All racinI Gini 0.4176 0 5291 0.3742 0.5277 0.4590 0.4234 0.3339 0.3979 0.5093 groups Theil T 0.3423 0.5363 0.2520 0.6!5! 0.4118 0.3404 0.2003 0.3104 0.5441 31.8 Theil T Theil L 0 3366 0.4911 0.2749 0.5216 0.3996 0.3286 0.2233 0.3184 0.4798 31.7 Th.i! L Varlog 0.7240 0.8716 0.6415 0.9636 0.8158 0.6513 0.5033 0.7093 0.8953 23.8 Varlog Geometnc mean 302.7 409 1 206.6 130.5 113.7 726 69.7 116.3 105.4 Arithmetic mean 423.8 668.6 271.9 219.9 169 5 100.9 87.2 159.9 170.2 Sample size 2,146 799 1,922 3,896 3,307 6.801 9,434 7,088 35,393 Between-race-and- occupation contn- bution to inequality (percent) Between-race Theil T 12.8 16.9 1 8 8.1 44 22.1 7.8 3.6 9.7 38.5 Theil T contribution Theil L 11 8 14.3 1.7 11.0 3.1 19.2 7.0 3.2 10.2 38.0 Theil L to inequality Varlog 5.9 15.5 1.9 11.6 1 5 12.4 10.5 3.7 8.6 29.9 Varlog (percent) Note- See note to table 6-Il for thedefinition oftheeight one-digit occupationalcategories in termsofthe two-digitcode used in the PES. It can be verified from table 6-11 that there were 5,413 income recipientswho were not in the labor forceor whose occupation was notavailable (codenumbersOO, 10,and 99)-hence the samplesize of 35,393 for all occupations in table 6-8 compared with a sample size of 40,806 for all income recipients in table 6-2, 6-3, 6-5, or 6-6. 220 INEQUALITY AND POVERTY IN MALAYSIA An examination of the mean incomes of the major racial groups within each occupation (table 6-8) reveals that the Chinese receive higher incomes on average than the Malays and Indians, and that the Indians receive higher incomes than the Malays. With the exception of farmers and sales workers, racial income disparities within occupations are not as large as overall racial income disparities. There are some notable exceptions to the general ranking of the three racial groups by income level: -In the professional and technical occupations, the Indians receive the highest mean incomes. -In the administrative and managerial category, the Malays receive significantly higher incomes than the Indians. -In the category of service workers, the Malays receive the highest mean incomes, but the difference between the racial groups is not large. -In the category of production workers, the Indians receive the highest mean incomes, but again the difference between the three racial groups is not very large. More detailed information on income by racial group and occupation at the two-digit level is contained in table 6-11 at the end of this chapter. The table presents for each subcategory the mean income, standard deviation of income, and sample size.35 Thus, the disaggregation of the one-digit administrative and managerial category in table 6-8 shows that the Malays receive significantly more than the Chinese in the subcategory of govern- ment administrators and legislative officials (two-digit code 20 in table 6-11), but the Chinese receive more than the Malays in the subcategory of managers (two-digit code 21). Higher Malay incomes in the administrative subcategories are probably a reflection of the preferential treatment in hiring and promotion given to Malays in the government sector. In addition, the level of earnings in the public sector is about 50 percent higher than in the private sector (Mazumdar, 1975), so that government's public sector employment policies (four-to-one quotas in favor of Malays at some levels of the hierarchy) should contribute to a narrowing of between-race income differentials. This deployment of the government machine to help the Malays is meant to counterbalance the natural advantage enjoyed by the Chinese in the business sector. The racial disparity ratios among farmers are greatest, with Chinese farmers receiving more than two-and-a-half times as much as Malay 35 The mean and standard deviation of income immediately allow one to compute a well- known measure of inequality, the coefficient of variation. This is defined as the standard deviation of income divided by the arithmetic mean income. INEQlJALITY IN THE PERSONAL INCOME DISTRIBUTION 221 farmers, and Indian farmers (constituting only I percent of the category) receiving almost twice as much as Malay farmers. These differentials strongly suggest that average Chinese landholdings are much larger than average Malay landholdings (see the previous section also), but there are no independent data on land distribution in Malaysia.36 The disparity ratios among farmers tally with the large between-race contribution to inequality in their incomes (22.1 percent by Theil T, 19.2 percent by Theil L, and 12.4 percent by varlog). The between-race contribution to inequality within other occupations is not large, except for the administrative and managerial category, where it is about 16 percent. As one might expect, the incomes of clerical, service, and production workers are fairly homogeneotis across racial groups, and this is reflected in the small between-race contributions to inequality. These are occupations in which the degree of labor unionization is high and uniform wage contracts are likely to prevail.37 Higher between-race contributions to inequality within other occupations (table 6-8) probably arise from greater. variation in skill levels and wealth owned. Interoccupation differentials (within each racial group) are much larger than interracial differentials (within each occupational category). The betwecn-occupation contribution to personal income inequality for all racial groups is 31.8 percent by Theil T, 31.7 percent by Theil L, and 23.8 percent by varlog, whereas the between-race contribution is 10.2 percent or less by the three measures.38 In computing this between-occupation contribution, race was not held constant in any sense, and it might be claimed that some of the between-occupation contribution observed is really the effect of race. As noted earlier in this section, there is indeed some multicollinearity between occupation and race, so that a part of the between-occupation contribution could be a between-race effect.39 But 36. Some data on nibber smallholdings imply a ratio of 2:1 for Chinese-to-Malay average size of holding. An agricultural census of the country has recently been taken, but the results are not yet available. 37. For sales workers and farm laborers, the between-race contributions are not quite so small. Apart from nonunionization, this might reflect differential ownership of assets (see, for example, occupational code 41, working proprietors, in table 6-11) and preferential hiring within one's own race. 38. As discussed in appendix C. one expects different between-group contributions from different measures of inequality. It is interesting, however, that the differences are not wildly large (as, theoretically, they could be). 39. If, for instance, there were just as many occupations as racial groups, and each racial group were perfectly identified with one occupation, then the between-occupation contri- bution would be indistinguishable from the between-race contribution. This is obviously not the case in Malaysia. where the racial groups are distributed over every occupation, albeit unevenly (see table 6-8). 222 INEQUALITY AND POVERTY IN MALAYSIA from table 6-8, the between-occupation contribution within each racial group is just as high as the between-occupation contribution overall. For Malays this is 37.0 percent by Theil T, 32.0 percent by Theil L, and 23.2 percent by varlog; for Chinese it is 24.7, 24.9, and 19.7 percent, respectively; and for Indians it is 34.9, 40.4, and 27.5 percent, respectively. This suggests that the overall contribution is genuinely one of occupation. Another way to attempt to disentangle the contributions of the variables is by a two-way decomposition of inequality by race and occupation (4 times 8 separate groups). With this finer partition, the between-group contribution becomes 38.5, 38.0, and 29.9 percent according to Theil T, Theil L, and varlog, respectively. It can be shown that any two-way contribution is always greater than or equal to each of the one-way contributions, and less than or equal to their sum. Had the between-race- and-occupation contribution been equal to the sum of the between-race and between-occupation contributions, the variables might be called uncor- related or orthogonal. The divergence of the two-way contribution from the sum of the one-way contributions might then be said to indicate the degree of collinearity. The sum of the between-race and between-occupation contributions is 41.5 percent by Theil T, 41.9 percent by Theil L, and 32.4 percent by varlog, while the between-race-and-occupation contributions are 38.5, 38.0, and 29.9 percent by the three measures, respectively. This indicates only slight dependence between occupation and race in their contributions to overall inequality. The analogy with regression analysis may be carried further. The between-race contribution to overall income inequality is analogous to the proportion of variance of, say, income (or log-income) explained by race in a simple regression between income (or log-income) as the dependent variable, and race as the explanatory variable. Similarly, the between- occupation contribution to overall income inequality is analogous to the R2 of a single-variable regression between income (or log-income) and occupation. The between-race-and-occupation contribution to overall income inequality is analogous to the proportion of variance of income (or log-income) explained by both race and occupation in a multiple regression between income (or log-income) as the dependent variable, and race and occupation as the explanatory variables. The R2 of the multiple regression is greater than or equal to the R2's of each of the simple regressions, and less than or equal to their sum-just as the two-way between-group contri- bution is greater than or equal to each of the one-way contributions, and less than or equal to their sum. When the explanatory variables are uncorrelated or orthogonal, the R2 of the multiple regression is equal to the sum of the R2's of the simple regressions. Otherwise, with multicollinearity, the proportion of variance of income (or log-income) explained by race and INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 223 occupation cannot be split up into a part explained by race alone and another part explained by occupation alone. In this case, the R2's (and coefficient estimates) of the simple regressions include the effect of the omitted variable to the extent that it is collinear with the included variable. The R2 of a multivariate regression equation is like the between-group contribution of a niultivariate partition (that is, a partition defined by multiple variables). Just as the R2 increases (does not decrease) as variables are added, the between-group contribution increases (does not decrease) as the partition is made finer with additional defining variables. But an a priori comparison of R2's, or between-group contributions, cannot be made if the variables or the groups are not nested. Thus it cannot be said whether the R2, or between-group contribution, of an arbitrary set of independent variables, or partition of the population, will be greater or smaller than that of another set of variables or partition-even if one has many more variables or groups than the other."0 It is only when variables are actually added, or the partition is actually made finer, that the R2, or between-group contribution, increases. In the limit, however, just as R2 tends to unity as the number of variables tends to the number of observations (that is, the degrees of freedom shrink to zero), so also does the between-group contribution tend to unity as the number of groups tends to the number of people in the population (that is, the partition becomes so fine that each group contains only one person). The methodology of inequality decomposition does not allow the measurement of the pure contribution of a factor to overall inequality: the effect of all variables correlated with the factor is also included." Just as the R2 of a simple regression between income (or log-income) and race does not hold other variables constant, so the between-race contribution to overall income inequality does not hold the effect of other variables constant. The between-race contribution to inequality measures the effect of race both on its own and in association with other factors, such as occupation, education, and so on.42 Thus 8.6 to 10.2 percent is the total contribution of race to inequality, not the contribution of race with other factors held 40. For example, if each group has the same mean income, then no matter how many groups there are, the between-group contribution will be zero. 41. In linear regression analysis, however, the partial correlation coefficient can be used to measure the correlation between two variables holding the effect of other variables constant, that is, removing their influence. There is no analogy to the partial correlation coefficient in inequality decomposition analysis. 42. Everything that makes the racial income distributions what they are is thus taken into account: the b.tween-race contribution simply measures the fraction of overall inequality attributable to racial differences in mean income, however these came about Table 6-9. Decomposition by Race and Employment Sector of Inequality in the Personal Income Distribution Transport, Between-sector Agri- storage, and contribution to Racial Agri- cultural Mining and Manu- Con- Public communi- All inequalitY group Item culture products quarrying Jacturing struction utilities Commerce cation Services sectors (percent) (1a) (2) (3) (4) (5) (6) (7) (8) (9) Malay Gini 0.3937 0.3590 0.3624 0.4906 0.4101 0.2336 0.4717 0.3377 0.4277 0.4709 Theil T 0.2749 0.2305 0.3422 0.4454 0.5865 0.1117 0.4130 0.2022 0.3563 0.4313 265 Theil T Theil L 0.2893 0.2460 0.2607 0.4537 0.3312 0.0945 04174 0.2067 0.3730 0.4114 25.7 Theil L Varlog 0.6062 0 5294 0.4653 0.9057 0.4674 0 1602 0.8510 0.4365 0.8541 0.8140 20.6 Varlog Geometric mean 61.1 63.0 168.9 66 0 143.6 176.3 85 3 150.4 166.7 80.9 Arithmetic mean 81.5 80.6 219.3 103.8 200.0 193.7 129.4 184.9 242.0 122.0 Sample size 5,458 5,695 183 1,326 343 224 1,301 725 3,045 18,300 Chinese Gini 0.4499 03787 04108 0.5056 04069 04439 0.5178 0.2982 0.4964 0.4888 Theil T 0.3870 0.2928 0.5051 0.5319 0.3799 0.3762 0.5821 0.1614 0.4476 0.4939 6 7 Theil 7' Theil L 0.3806 0.2581 0.3166 0.4768 0.3016 0.3672 0.4934 0.1552 0.4692 0.4365 7.9 Theil L Varlog 0.7823 0.4797 0.4660 0.9091 0.5189 0.7510 0.8982 0.3061 0.9934 0.8212 5 7 Varlog Geometric mean III 6 107.6 161.3 118.1 157.1 208.9 177.2 192.8 166.3 141.1 Arithmetic mean 163.3 139.2 221.4 190 2 212.5 301.5 290.3 225 1 265.8 218.3 Sample size 1,119 2,398 438 2,698 646 77 2,631 601 2,720 13,328 Indian Gini 0.4449 0.2969 0.2843 0.3999 0.2534 0.3622 0.5255 0.3150 0.5477 0.4662 Theil T 0.3482 02312 0.1879 03424 0.1174 0.3183 0.6618 0.1728 0.6126 0.5014 142 Theil 7T Theil L 0.3743 0.1743 0.1448 0 2982 0.1166 0.2341 0.4981 0.1715 0.5573 0.3844 19.2 Theil L Varlog 0.7985 0.2938 0.2210 0.5631 0.2369 0.3467 0.8084 0.3475 1.0444 0 6246 11.2 Varlog Geometric mean 85.0 95.9 158.6 132.1 161.1 192 1 140.2 193.9 159.3 124.7 Arithmetic mean 123.6 114.2 183.3 178.1 181.1 242 8 230.7 230.2 278.2 183.1 Sample size 112 2,062 62 234 145 146 538 287 1,077 4,663 Other Gini 0.3826 0.6705 0.2941 0.5735 0 1354 0 274i 0.6834 0.5489 0.5872 0 7037 Theil T 0.2674 0.8846 0 2111 0.5730 0.0451 0.2216 0.8780 0 5774 0.6190 0 9387 32 1 Theil T Theil L 0.2509 1.3648 0.3166 0.6869 0.0505 0.3461 1.2733 0 5673 0.7692 1.2192 46.3 Theil L Varlog 0.4543 3.0720 0.8494 1.4277 0.1108 0.9535 2.8787 1.0279 1.8320 2.6217 44.2 Varlog Geometric mean 38.5 254.6 1063.6 383.6 184.1 557 0 178.4 290 3 440.8 180 3 Arithmetic mean 49.5 996.6 1459.7 762.3 193.7 787 3 637.4 512.0 951.2 610.3 Sample size 82 i9 3 !5 3 3 20 15 104 264 All racial Gini 0.4330 0.3802 0.4021 0 5156 0.3918 0 3430 0.5350 0.3301 0.3937 0.5040 groups Theil T 0.3540 0.5359 0.4612 09727 04683 0 3391 0.9867 0.3960 0.9672 0 8773 16 I Theil T Theil L 0.3463 0 2770 0.3038 0.5035 0.2877 0.2103 0 5352 0.1950 0.4667 0.4698 19.0 Theil L Varlog 0.6913 03158 0.4615 0.5484 0.4118 0.2694 0.6220 0.2006 0.4785 0.5347 15.6 Varlog Geometric mean 67.5 78.0 164.4 99.6 153.5 188.0 139.4 1734 - 167.8 105.3 Arithmetic mean 95.4 102.9 222.8 164.8 204.7 232.0 238.1 210.8 267.6 168.4 Sample size 6,771 10,174 686 4.273 1,137 450 4,490 1.628 6.946 36,555 Between-race- and-sector con- tribution to inequality (percent) Between- Theil T 12.9 16.9 6.7 8.0 0.3 9.7 8.6 5.9 66 9.5 23.5 Ttieil T race con- Theil L 11.8 14.6 5.8 8.5 0 5 10.6 11.2 5.2 4 6 10.0 26.8 Thetl L tribution Varlog 7.9 11.4 3.4 8.4 0.4 3.5 10.6 4.5 1.5 8 2 21 7 Varlog to inequality (percent) a The industrial classification used in PES is the standard two-digit code from Department of Statistics. Malaysian Industrial Classification (Kuala Lumpur.1971). The nine one-digit sectors in this table have been regrouped from the two-digit PES code as follows Two-digit PES code One-digit cods Deicriplion 00.10,99 - Not in labor force, or sector not available 01-09 1 Agriculture 11-19 2 Agricultural products 20-29 3 Mining and quarrying 30-49 4 Manufacturing 50-59 5 Construction 60-69 6 Public utilities 70-79 7 Commerce 80-89 8 Transport, storage, and communication 90-98 9 Services 226 INEQUALITY AND POVERTY IN MALAYSIA constant.43 Similarly, 23.8 to 31.8 percent is the total contribution of occupation to inequality. The conclusion from the race-occupation de- compositions is that occupational differences contribute much more to personal income inequality than do racial differences. Decomposition by Sector of Employment One of the government's aims in restructuring the racial pattern of employment is "to achieve greater equalization of income among the different races in the country" (Robless, 1975a, p. 44). Two factors underlying racial income disparities are seen as: "(i) the concentration of Malays in the agricultural sector where output per worker is the lowest; (ii) the concentration of Chinese in the mining, manufacturing and construction sectors where output per worker is two to three times higher than in agriculture" (Robless, 1975a, p. 42). Imbalances also exist within sectors of employment: within each sector Malays are concentrated in the "lower occupational levels of the job hierarchy, especially in the unskilled and semi-skilled categories" (MTR, p. 9). The sectoral imbalances are confirmed in table 6-9, which also shows a significant concentration of Chinese in commerce. For the one-digit industrial classification into nine sectors shown in table 6-9, the between-sector contribution to personal income inequality is considerably less than the between-occupation contribution. The between- sector contribution stands at 16.1, 19.0, and 15.6 percent by the three measures, respectively, which is half to two-thirds the between-occupation contribution. Sectoral differences in mean income, therefore, do affect overall inequality, but not by very much. The result casts doubt on the emphasis that planners often place on differences in sectoral productivity in explaining personal income inequality. Since 80 to 85 percent of inequality arises within sectors, it becomes difficult to agree, for example, that "a primary reason for the degree of inequality . . . is the sectoral distribution of employment among income classes and racial groups" (Robless, 1975a, p. 41). The sectoral distribution of employment contributes a variable amount to income inequality within each racial group. The between-sector contribution to Chinese income inequality is only 6 to 8 percent, to Malay income inequality about 25 percent, and to Indian income inequality about 43. If it were possible to hold other factors constant, the extent of inequality attributable to racial discrimination could be measured. INEQIJALITY IN THE PERSONAL INCOME DISTRIBUTION 227 15 pernent-while the between-sector contribution to overall income inequality is about 17 percent. The between-race contribution to overall income inequality is about half as much as the between-sector contribution. Table 6-9 shows the between- race contribution to income inequality in each sector. This is especially small in the construction sector (less than I percent), and also fairly small in mining, transport, and services. In general, these sectors use manual and unskilled labor for which the rate of wage payment is fairly uniform. In other sectors, the between-race contribution is higher, which probably reflects differential ownership by racial groups of physical and human capital. For example, in agriculture, agricultural products, and commerce, it is probably the superior possession by Chinese of land and other assets that accounts for the large income disparity ratios. In manufacturing, the relatively large disparity ratios are probably the result of a superior average level of skills among the Chinese. In the next chapter, I examine the extent to which skills acquired through education and experience are associated with the incomes that people receive. Multivariate Decompositions With the possible exception of occupation, the decompositions so far have not accounted for much of personal income inequality. These results are of value in a negative sense, for they cast doubt on some current beliefs concerning the factors responsible for income inequality in developing countries. They also suggest that the variables and the degree of disaggregation for which data are commonly available might have limited power to explain personal income inequality. Some multisector distributional planning models assume an income generation process with a two-way classification of income recipients by sector and skill (or occupation).44 All workers in each sector-skill group are assumed to receive the same wage, the mean income of the group.45 Thus income inequality is generated in these models solely by between-group differences in mean income; any within-group inequality is superimposed 44. Some of the better-known models of this type are those of Weisskoff (1973), Morley and Williamson (1973), Thorbecke and Sengupta (1972). and Adelman and Robinson (1977). A morecomplete bibliography of such models may be found in Chenery and others (1974) or Blitzer, Clark. and Taylor (1975). 45. In some modlels, this wage is fixed exogenously, but in the more sophisticated general equilibrium model;, such as those of Adelman and Robinson (1977) or Adelman and Tyson (1974), the wage for each sector-skill category is determined endogenously. 228 INEQUALITY AND POVERTY IN MALAYSIA exogenously. The decompositions for Malaysia suggest that a multisector (input-output) model of this type would explain only about 30 percent of personal income inequality (see table 6l10).46 Other models assign "welfare weights" to particular groups of people defined not by income level but by race or region (see, for example, Marglin, 1967). These weights, if based solely on the mean income of the group, can be misleading indicators of the relative social value of income accruing to the group. The reason is that they completely ignore within- Table 6-10. Multivariate Decomposition of Personal Income Inequality Between-group contri bution to personal income Number inequaliti (percent) oJ Description oJ multirariate grouping groups T/ieil T Theil L Varlog 3 locations x 9 employment sectors (as in table 6-5) (as in table 6-9) 27 18.8 21.6 17 1 9 employment sectors x 4 employment-status (as in table 6-9) categories (as in table 6-7) 36 26 0 25 4 19.9 8 occupations x 9 employment sectors (as in table 6-8) (as in table 6-9) 72 34.5 33 5 26 9 6 educational x 9 employment sectors categoriesa (as in table 6-9) 54 31.1 31.4 25 9 6 educational x 8 occupations categoriesa (as in table 6-8) 48 38 1 36 4 29.3 13 age groupsh x 8 occupations (as in table 6-8) 104 40.3 40.9 34.1 13 age groupsb x 6 educational categoriesa 78 42.2 38 7 35.2 13 age groupsb x 6 educational categoriesa x 8 occupations (as in table 6-8) 624 54 1 49.6 44.1 a. The 6 educational categories (as in table 4-3) are: none, some primary, completed primary, lower secondary (forms 1-111), some upper secondary (forms IV-V), and school certificate or higher. b. The 13 age groups (in years) are: 1-14 35-39 60-64 15-19 40-44 65-69 20-24 45-49 70 + 25-29 50-54 30-34 55-59 46. This does not detract from the usefulness of multisector models in indicating areas where policy intervention can, in fact, reduce inequality, by however small an amount. INEQUALITY IN THE PERSONAL INCOME DIST RIBUTION 229 group income inequality, which can be substantial.47 This is what prompted, in the previous chapter, the detailed cross-classification of certain income groups (the poor) in preference to a broader approach which simply discriminates on the basis of region or race. In an attempt to explore the sources of inequality and the determinants of income in Malaysia, some further decompositions, based on two-way and three-way partitions by relevant variables, are presented in table 6-10. The rnultivariate decompositions do not add very much to the single- variable decompositions. For instance, the partition of 8 occupations times 9 employment sectors yields a between-group contribution of 34.5 percent by Theil T, 33.5 percent by Theil L, and 26.9 percent by varlog. This is not much more than the one-way between-occupation contribution of 31.8, 31.7, and 23.8 percent, respectively, by the three measures (table 6-8). The hypothetical distribution which assigns to each person in an industry- occupation cell the mean income of the cell will thus account for only about 30 percent of overall income inequality. This suggests that the existing categories need to be disaggregated or other variables added, or both, to simulate better the personal income distribution.48 Other decompositions do yield higher between-group contributions. Interestingly, education, age, and occupation seem to be the most important variables in explaining inequality (table 6-10). The decompo- sition by education and occupation accounts for 38.1, 36.4, and 29.3 percent of inequality by Theil T, Theil L, and varlog, respectively; the decomposition by age and occupation accounts for 40.3, 40.9, and 34.1 percent of inequality; the decomposition by age and education accounts for 42.2, 38.7, and 3'i.2 percent of inequality. Finally, when all three variables are combined in a three-way decomposition, the between-group contri- bution to inequality is 54.1 percent by Theil T, 49.6 percent by Theil L, and 44.1 percent by varlog. In absolute terms, this explanatory power is not huge, but it suggests that life-cycle factors such as age and education should figure prominently in any model seeking to explain personal income inequality. The human capital model in the next chapter uses exactly these variables to explain income inequality. 47. A model which takes some account of within-group income inequality is that of Squire and van der Tak ( 1975). They estimate welfare weights by assuming a particular distribution for incremental income among the members of a group. In a different context, Sen (1 976b) estin.lates indices of regimnal welfare by combining the mean income of a region with a measure (the Gini coefficient) of the income inequality within it. 48 Examples of disaggregation for the occupational category are large-, medium-. and small-scale farmers, and skilled and unskilled production workers. Any additional variable which can distinguish asset ownership among individuals is also likely to improve performance. Other variables should incorporate personal characteristics such as age and education. 230 INEQUALITY AND POVERTY IN MALAYSIA Table 6-11. Arithmetic Mean Income of Income Recipients by Two-digit Occupational Category and Race (M$ per month) All Two-digit racial occupation code Item groups Malay Chinese Indian Other 01. Physical Mean income 465 323 375 743 1,615 scientists and Standard deviation 485 229 250 871 0 related Sample size 41 10 22 8 I technicians 02. Architects Mean income 1,337 775 1,288 1,394 1,t54 and engineers Standard deviation 713 444 678 667 830 Sample size 46 4 26 6 10 03. Engineering Mean income 409 351 473 413 309 assistants, Standard deviation 260 203 321 206 183 draftsmen, etc. Sample size 177 72 69 32 4 04. Aircraft Mean income 1,289 196 343 - 2,353 and ships' Standard deviation 1,191 66 0 - 595 officers Sample size 10 4 1 0 5 05. Life scientists Mean income 608 447 660 1,169 1,369 and related Standard deviation 690 356 439 815 2,048 technicians Sample size 39 29 3 4 3 06. Medical Mean income 821 420 713 1,288 472 doctors, Standard deviation 1,104 305 802 1,663 1'0 dentists, Sample size 123 21 64 35 3 vets, etc. 07. Nurses Mean income 260 192 320 256 545 and midwives Standard deviation 178 163 178 145 0 Sample size 169 70 76 22 1 08. Statisticians, Mean income 1,796 - - 1,796 - mathema- Standard deviation 1,701 - - 1,701 - ticians, etc. Sample size 3 0 0 3 () 09. Economists Mean income 965 818 1,111 1,111 - Standard deviation 358 507 0 0 - Sample size 6 3 2 1 (I I1. Accountants Mean income 970 980 1,033 479 1,508 Standard deviation 602 775 506 183 945 Sample size 24 3 13 5 3 12. Jurists Mean income 1,431 1,867 986 1.767 - Standard deviation 656 0 452 683 - Sample size 9 1 4 4 0 13. Teachers Mean income 365 311 414 422 875 Standard deviation 256 171 219 366 954 Sample size 1,275 705 416 135 19 14. Workers in Mean income 173 212 181 112 147 religion Standard deviation 197 264 145 94 218 Sample size 87 33 26 21 7 INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 231 All Twvo-digit racial occupation code Item groups Malay Chinese Indian Other 15. Authors and Mean income 519 544 531 343 - journalists Standard deviation 311 315 336 286 - Sample size 20 10 8 2 0 16. Commercial Mean income 197 165 199 343 - artists and Standard deviation 173 86 185 0 - photographers Sample size 53 8 44 1 0 17. Performing Mean income 282 259 349 90 - artists and SIandard deviation 190 121 247 0 - composers Sample size 24 12 10 2 0 18. Athletes and Mvlean income 303 255 343 115 690 sportsmen Standard deviation 172 49 0 0 0 Sample size 8 5 1 1 1 19. ProZessional Mlean income 544 613 459 618 481 and technical Standard deviation 354 490 270 268 273 workers n.e.c. Sample size 32 11 12 6 3 20. Government Niean income 1,268 1,049 705 1,537 3,628 adrninistrators Standard deviation 1,753 1,481 604 727 3,187 and legislative Sample size 68 53 6 3 6 officials 21. Managers Mean income 898 721 897 633 2,191 Standard deviation 1,097 1,580 1,004 761 832 Sample size 287 36 217 24 10 30. Clerical Mean income 723 669 751 711 1,111 supervisors Standard deviation 394 209 439 454 0 Sample size 56 14 22 19 1 31. Government Mv4ean income 513 444 834 402 - executive Standard deviation 445 335 708 203 - officials S'ample size 114 76 22 16 0 32. Stenogra- M4ean income 343 369 334 350 231 phers, typists, Standard deviation 352 545 275 209 93 and card- Sample size 181 45 106 26 4 punch operators 33. Bookkeepers Mh4ean income 342 342 324 435 313 and cashiers Standard deviation 296 211 306 319 304 Sample size 367 61 254 48 4 34. Computing Mean income 258 252 203 441 - machine Standard deviation 134 95 108 170 - operators Sample size 20 8 9 3 0 35. Transport Mean income 471 351 518 545 - and communi- Standard deviation 218 140 264 196 - cation super- Sample size 36 12 12 12 0 visors (Table continues on the following page.) 232 INEQUALITY AND POVERTY IN MALAYSIA Table 6-11 (continued). All Twvo-diqu racial occupation code hlem groups Malay Chinese Indian Other 36. Transport Mean income 181 182 183 128 - conductors Standard deviation 88 114 57 88 - Sample size 72 34 36 2 0 37. Mail distri- Mean income 151 151 176 139 217 bution clerks Standard deviation 71 69 75 66 179 Sample size 190 117 20 51 2 38 Telephone Mean income 233 263 210 195 191 and telegraph Standard deviation 106 107 115 76 141 operators Sample size 39 20 8 8 3 39. Clerical Mean income 258 232 275 270 280 and related Standard deviation 178 163 183 191 173 workers n.ec Sample size 1,017 385 434 179 19 40 Managers Mean income 700 174 893 424 (wholesale and Standard deviation 730 83 851 217 -- retail trade) Sample size 22 2 14 6 0 41. Work Mean income 390 179 503 381 347 proprietors Standard deviation 721 200 830 885 434 (wholesale and Sample size 1,206a 367 698 135 5 retail trade) 43. Technical Mean income 384 366 413 209 -- salesmen and Standard deviation 283 402 249 263 - commercial Sample size 79 13 58 8 0 travellers 44 Insurance and Mean income 357 195 401 272 1,144 real estate Standard deviation 478 109 504 325 1,525 salesmen and Sample size 67 14 39 12 2 auctioneers 45. Salesmen and Mean income 131 79 156 112 58 shop assistants Standard deviation III 62 124 77 49 Sample size 2,410 636 1,491 276 7 49. Sales Mean income 134 79 121 267 -- workers n.e c. Standard deviation 203 38 85 421 Sample size 135 49 60 26 0 50. Managers Mean income 343 - 438 153 -- (catering and Standard deviation 224 - 215 53 - lodging Sample size 6 0 4 2 0 services) 51. Work Mean income 375 185 472 287 1,111 proprietors Standard deviation 452 227 511 356 t) (catering and Sample size 298 71 177 49 1 lodging services) INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 233 All Tavo-digt racial occupation code Item groups Malay Chinese Indian Oiher 52. Housekeeping Mean income 166 290 155 165 190 supervisors Standard deviation 128 0 138 0 0 Sample size 16 1 13 1 1 53. Cooks, Mean income 116 87 123 115 455 waiters, Standard deviation 88 68 80 56 578 bartenders, etc. Sample size 518 110 331 74 3 54. Maids and Mean income 77 58 90 57 98 housekeeping Standard deviation 56 53 57 37 29 service Sample size 1,003 271 590 139 3 workers n.e.c. 55. Building M1ean income 126 111 136 128 90 caretakers and Standard deviation 82 60 100 55 0 cleaners Sample size 96 31 48 16 1 56. Launderers Mean income 161 III 155 194 - and dry- Standard deviation 170 84 181 150 - cleaners Sample size 84 5 61 18 0 57. Barbers and Mean income 173 116 185 226 - beauticians ,Standard deviation 163 47 194 158 - Sample size 209 59 112 38 0 58. Protective Mean income 250 242 266 209 985 service workers Standard deviation 207 163 228 148 753 Sample size 847 666 - 68 101 12 59 Service Mean income 148 152 159 130 217 workers n.e.c. Standard deviation 75 62 105 57 179 Sample size 236 111 58 65 2 60. Farm Mean income 301 199 272 282 2,763 managers and Standard deviation 424 77 184 350 1,020 supervisors Sample size 245 56 78 107 4 61. Farmers Mean income 101 84 214 160 55 Standard deviation 108 62 220 158 41 Sample size 6,801 5,791 860 78 72 62. Agricultural Mean income 83 70 100 99 35 and animal Standard deviation 53 51 55 47 31 husbandry Sample size 8,282 4,453 1,943 1,871 14 workers 63. Forestry Mean income 184 175 228 87 - workers Standard deviation 148 151 136 101 - Sample size 107 71 29 7 0 64? Fishermen, Mean income 107 87 143 119 37 hunters, and Standard deviation 83 52 112 S0 14 related workers Sample size 1,046 667 369 7 3 (Table continues o0} the following page.) 234 INEQUALITY AND POVERTY IN MALAYSIA Table 6-11 (continued). All Two-digit racial occupation code Item groups Malay Chinese Indian Other 70. Production Mean income 322 255 396 199 - supervisors Standard deviation 219 117 258 103 - and general Sample size 119 37 64 18 0 foremen 71. Miners and Mean income 180 160 186 144 1,867 quarrymen Standard deviation 486 81 576 40 0 Sample size 411 90 283 37 1 72. Metal Mean income 162 157 163 171 -- processors Standard deviation 121 50 137 59 -- Sample size 46 8 35 3 0 73. Wood and Mean income 158 130 175 158 paper workers Standard deviation 125 72 141 183 -- Sample size 271 97 157 17 0 74. Chemical Mean income 112 III 112 113 78 processors Standard deviation 61 63 67 45 18 Sample size 346 113 155 76 2 75. Spinners Mean income 46 41 68 115 -- and weavers Standard deviation 37 35 39 0 Sample size 165 140 24 1 0 76. Tanners Mean income 110 140 102 - -- and Standard deviation 78 0 88 - -- fellmongers Sample size 5 1 4 0 0 77. Food Mean income 143 84 174 143 -- and beverage Standard deviation 255 73 317 74 processors Sample size 375 122 230 23 0 78. Tobacco Mean income 97 58 166 149 preparers and Standard deviation 113 64 175 69 - tobacco prod- Sample size 86 53 21 12 0 uct workers 79. Tailors and Mean income 105 64 114 131 - dressmakers Standard deviation 132 67 144 94 - Sample size 603 122 466 15 () 80. Shoemakers Mean income 106 107 106 128 - and leather Standard deviation 78 34 82 18 - goods makers Sample size 85 6 77 2 C0 81. Cabinet- Mean income 158 104 164 343 - makers and Standard deviation 137 89 141 0 - wood workers Sample size 107 13 93 1 0 82. Stone cutters Mean income 143 - 143 - - and carvers Standard deviation 90 - 90 - - Sample size 13 0 13 0 0 INEQUALITY IN THE PERSONAL INCOME DISTRIBUTION 235 Ali 7'wo-digit racial occupation code ltem groups Malay Chinese Indian Olher 83. Blacksmiths Mean income 178 126 203 236 65 and machine- Standard deviation 148 69 173 126 0 tool operators Sample size 99 52 61 5 1 84. Machinery Mean income 206 221 192 262 715 fitters and Standard deviation 169 143 145 194 1,007 machine Sample size 668 130 484 50 4 assernblers 85. Electrical Mean income 221 211 208 242 1,156 fitters and Standard deviation 182 117 146 136 1,508 workers Sample size 287 71 164 50 2 86. Broadcasting Mean income 218 254 189 103 343 station and Standard deviation 121 121 123 53 0 sound Sample size 13 6 4 2 1 equipment operators 87. Plumbers, Mean income 166 195 152 230 233 welders, etc. Standard deviation 101 110 94 114 0 Sample size 134 15 103 15 1 88. Jewelry and Mean income 253 121 271 219 - precious metal Standard deviation 357 90 391 250 - workers Sample size 86 4 64 18 0 89. Glassformers Mean income 108 66 157 146 - and potters Standard deviation 92 36 121 38 - Sample size 60 32 24 4 0 90. Rubber and Mean income 132 161 128 65 - plastics Standard deviation 95 108 94 0 - producis Sample size 34 6 27 1 0 rnal:ers 91. Paper and Mean income 46 47 45 - - paperboard Standard deviation 36 25 37 - - products Sample size 42 5 37 0 0 makers 92. Printers Mean income 156 128 146 325 - and related Standard deviation 205 114 108 577 - workers Sample size 143 51 79 13 0 93. Painters Mean income 164 133 168 165 - Standard deviation 92 77 95 43 - Sample size 121 12 106 3 0 94. Production Mean income 55 33 121 121 140 and related Standard deviation 79 35 128 87 0 workers n.e.c. Sample size 240 179 56 4 1 (Table continues on the following page.) 236 INEQUALITY AND POVERTY IN MALAYSIA Table 6-11 (continued). All Two-digit rt zoal occupation code Item groups Moala Chinese Indian Other 95. Bricklayers Mean income 149 130 154 136 739 and other Standard deviation 98 74 93 57 882 construction Sample size 783 204 542 35 2 workers 96. Stationary Mean income 179 167 180 194 engine Standard deviation 72 78 71 61 -- operators Sample size 116 40 52 24 0 97. Dockers Mean income 170 172 173 161 146 and freight Standard deviation 116 82 162 69 24 handlers Sample size 334 139 123 68 4 98. Transport Mean income 188 168 213 185 157 equipment Standard deviation III 106 120 87 72 operators Sample size 1,296 612 497 183 4 00. No Mean income 118 83 150 160 361 occupation Standard deviation 161 III 182 172 496 Sample size 3,196 1,639 1,286 240 31 10. Not Mean income 65 65 - 65 applicable Standard deviation 0 0 - 0 Sample size 2 1 0 1 0 99. Not Mean income 112 112 112 116 74 available Standard deviation 61 61 64 54 54 Sample size 2,215 1,035 814 358 8 - Not applicable. n e.c. Not elsewhere classified. a. The racial affiliation of one person in this two-digit category was not available. Note: The occupational classification used in the PES is a standard two-digit code from Department of Statistics, Index of Occupations (Kuala Lumpur, January 1971). Its coding scheme is based on the more comprehensive Dictionary oJ Occupattonal Classf ication (K uala Lumpur, 1968). In the present study the two-digit classification of the PES has been recoded into eight broad, one-digit occupational categories as follows. Two-digit PES code One-digit code Description 00, 10, 99 - Not in labor force, or occupation not available 01-09, 11-19 I Professional and technical 20-31, 40, 50, 60 2 Administrative and managerial 32 -39 3 Clerical and related 41-49 4 Sales workers 51-59 5 Service workers 61 6 Farmers 62-69 7 Farm laborers 70-98 8 Production workers 7 Earnings Functions for Urban Employees THE GOVERNMENT OF MALAYSIA regards education as an important policy instrument for redressing income imbalances: The lack of education is a major factor adversely affecting the ability of an individual to enhance the quality of his life and to advance his economic position. Consequently, the lack of education becomes both a symptom as well as a significant factor contributing towards poverty. Ed acation is thus a major vehicle for the achievement of the objectives of the New Economic Policy (MTR, P. 189). Education is an important dimension of the nonhomogeneity of labor, and it is part of the aim of this chapter to examine the relation between education and income, and between educational inequality and income inequality. I shall not present a life-cycle model of education and income distribution, but rather draw upon the existing theory of human capital to explain incomes in terms of education and age (experience). The theory applies only to earnings or labor income, data on which are not available separately in PES. Hence I have restricted the sample to urban employees, whos.e income may be assumed to derive entirely, or largely, from labor. Rural employees have been excluded because the theory is not directly applicable in the context of rural labor markets, which tend to be imperfect. Also, owing to the nature of work in rural areas, a significant part of the income of rural employees may derive from nonhuman capital. that is, assets such as land. Earnings functions based on the so-called human capital model have thus been estimated for urban employees in Peninsular Malaysia. But it should be emphasized at the outset that these earnings functions are consistent with other theories of earnings (such as "screening" and "job competition"), although the latter have not been developed to the same extent. In any case, simple reduced-form equations between income and the 237 238 INEQUALITY AND POVERTY IN MALAYSIA variables of education and age are consistent with too many different hypotheses to allow for sharp tests between alternative models (see Anand, 1976b). The earnings function estimated here quantifies the response of income to education and experience; the relation has been estimated using the PES cross-sectional data on urban wage earners. Even if one rejects the formal interpretation of the estimated equations in terms of human capital theory, the approach turns out to be useful for a purely descriptive analysis of the data. There is much merit in adopting this framework simply to describe age-income profiles for different levels of education. The estimated equations are a convenient method of summarizing labor market infor- mation in terms of earnings differentials for individuals at different age- education levels. Furthermore, the regression equations automatically generate a measure of income inequality used in this study, the variance of log-income, which facilitates inequality comparisons between urban labor and other incomes. Under restrictive assumptions the earnings functions can also allow rough orders of magnitude to be obtained for what may be called individual or private rates of return to education. The assumptions required to interpret the estimated coefficients as true rates of return are admittedly heroic, but since the main purpose here is to compare the returns to education across various groups-occupational, racial, and other--this problem is perhaps less important. The resulting comparisons among groups could be suggestive of the role of educational policy in narrowing racial, occupational, or regional income inequalities-though further information about the demand side of the labor market would be required for any conclusions.' For small changes in supply, however, the income differentials for different levels of education and experience can be assumed constant; hence the earnings function analysis can at least guide the choice of educational projects, if not policies. The Earnings Function A formal derivation of the human capital earnings function is outside the scope of this chapter. It is in any case unnecessary since a detailed I. For example, a policy of simultaneously upgrading overall skill levels may well affect the income premiums commanded by education. Only if the macrodemand curves for labor of different skill levels are perfectly elastic (that is, only if the different types of labor are perfectly substitutable in production) can conclusions be drawn about educational policy, or large changes in supply, simply on the basis of the estimated earnings equations. For a discussion of the human capital model in the context of the supply of and demand for labor of different types, see Anand (1976b). EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 239 exposition of the model can be found in any of the standard works on the subject.2 I will therefore restrict myself to a statement of the estimating form used and will provide only a heuristic justification for it. The estimating form employed is: logy = f0 +31 S + 2T- f3 T2 pi , 0, i = 0, 1, 2,3 where y = annual income; S = number of years of formal schooling; and T = number of years of labor force experience. Years of labor market experience, T, are assumed to be measured by age A, minus schooling, S, minus 5; that is, T = A - S - 5, where six is assumed to be the age at the commencement of schooling. This definition of labor market experience raises obvious problems if employment is not continuous and there are periods of unemployment and job search. It is particularly unsatisfactory for women, because their participation in the labor force is often intermittent for various reasons. The model of optimal investment in human capital which underlies the earnings function for an individual was initiated by Ben-Porath (1967). It predicts a declining rate of investment in human capital with age. The intuitive reasoning behind this result is that most of the investment is made at younger ages to give individuals a longer period in their finite lifetimes over which they can receive returns. But the entire investment is not made instantaneously (before beginning the working life) because the marginal cost of acquiring human capital rises within each period, so that it pays to sprea(d the investment over time.3 The investment declines over time both because marginal benefits decline4 and because the marginal cost curve itself shifts upward with advancing age. There is also the depreciation of human capital with age (owing to obsolescence and physiological factors), which accentuates this decline in investment. Qualitatively, the Ben-Porath analysis implies three distinct phases of investment in human capital over the life cycle. In the initial phase, all available time is spent acquiring human capital. This period of complete specialization is one of full-time schooling and no earnings, and it can end before the completion of schooling. In the second phase, there is positive but declining investment in human capital. This is a period of on-the-job training and includes part-time schooling, when a declining fraction of 2. See, for example, Mincer (1970) and the references contained therein. The particular estimating form employed here is derived in Chiswick and Mincer (1972) and Mincer (1974, 1976). 3. Attempts to increase investment within a given period run into diminishing returns: costs increase with the speed of production of human capital. 4. T he marginal benefit of investment is measured by the discounted present value of increases in earning power over the remaining lifetime. 240 INEQUALITY AND POVERTY IN MALAYSIA available labor time is spent on the further acquisition of human capital. In the final phase, all available time is spent earning, and none is spent acquiring additional human capital-indeed, there is a net loss arising from depreciation. These results become clear if a formal optimizing model of investment in human capital is set up with its associated phase diagram (not included here). These considerations lead to a declining rate of investment in human capital over the life span, which becomes negative in the final phase. The decline itself implies that earnings rise to a peak (at zero net investment) and then begin to fall off. But the exact shape of the earnings function depends on the particular rate of decline assumed, that is, on the shape of the life-cycle investment schedule. A linear decline in the postschool investment schedule generates the following quadratic earnings function (shown in figure 7-1 in the appendix to this chapter):5 logy = #30+P.S±+#2T-f#lT2 3i > 0, i = 0, 1, 2, 3. Apart from those mentioned, the main assumptions subsumed in the derivation of this earnings function are: (I) a constant labor market return (/3,) for every year of schooling, and (2) independence between the return to formal schooling and to postschool investment (that is, no interaction effect between education and on-the-job experience). Furthermore, other relevant determinants of earnings have been omitted, such as ability and the proportion of the year actually spent working (the "weeks worked" variable in Mincer, 1974, 1976).6 The simplicity and econometric tractability of this earnings function make it agreeable to work with. As stated above, quite apart frorn its interpretation in terms of human capital theory,7 it furnishes some useful by-products. Since the dependent variable is the logarithm of income, the estimated regression equation explains the variance of log y-a familiar index of inequality. The computed R2 can then be interpreted as the percentage of inequality (measured by the variance of log-income) that is explained by the model. Further properties of this earnings function, which turn out to be useful in interpreting the regression results, are derived in the 5. An exponentially declining schedule would have generated a different earnings function, namely, the Gompertz curve (see Mincer, 1974). 6. See also Bhalla (1973). 7 For example, a quadratic experience (or age) - log-income profile is also implied by the stochastic model of income determination proposed by Aitchison and Brown (1957); this posits proportionate growth of an individual's income at a rate depending on experience and a (normally and independently distributed) random variable. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 241 appendix to this chapter. Some of these appear to have gone unnoticed by the human capitalists. The Return to Education The coefficient /31 in the earnings function log1y = [0 + /3S+ /32T-/33T2 shows the effect of schooling on log-income if experience T is held constant. If, however, age A is held constant, the effect of schooling on log-income is given by differentiating partially with respect to S the expression log Y = 130 + 3l. S+ f32(A - S- 5)- J33(A -S - 5)2. Hence: O 3S ]A const =1 -2 + 2/3 (A-S-5) / - [ f 2 - (A A-S-S)] Thus iif T = (A - S -5) < /2/233, the quantity in square brackets is positive, and :, overestimates the return to education with age constant. The reason, of course, is that with age held constant, an additional year of schooling is at the expense of a.year's experience. The reduction in experience offsets to some extent the effect /,3 of schooling on log-income. The extent to which it offsets Pu depends on the marginal return to a year's experience, oT = /2-2/33T = #2-2#3 (A-S-5), which is precisely as shown in the equation above. The coefficient P, on S in the earnings function must be interpreted with care. In certain very special circumstances, it can be taken to be the rate of return to education. With T held constant, partial differentiation with respect to S gives fl, = (1/y) (Ay/AS). Thus 3,1 indicates the percentage increa,se in annual income Ay/y for an additional period of schooling AS. For one extra year of schooling, AS = 1, and ,B, Ay/y. Thus 3,B measures the in:ternal rate of return to an investment of y which raises income in perpetuity by Ay. If the investment cost of an extra year's schooling for an individual happens to be exactly y, /3, measures the rate of return to it.8 8. Strictly speaking, fl, is the internal rate of return if the investment of y yields the flow of Ayforever. With a finite working life, the rate of return will be less than fl,, but the difference will be negligible provided the working life is of normal duration (say, thirty years). 242 INEQUALITY AND POVERTY IN MALAYSIA To an individual, the investment cost of a year's schooling is the sum of the opportunity cost and the direct (or out-of-pocket) cost, which includes tuition fees (if applicable), books, and so forth. The opportunity cost is the earnings forgone while at school; the direct cost is additional, but possibly small in relation to earnings forgone. In any event, if the total cost is some factor k times y, the rate of return to education is given by deflating f5, by k. With detailed information on k, therefore, the earnings functions can be used to assess the private return to education.9 To calculate the social return to education, further information and assumptions are necessary about government subsidies to education, wages' reflection of the marginal productivity of labor, and so on. 1 Important caveats attach to estimates of the private or social return to education derived from a single year's cross-sectional data. There is the usual problem of drawing time series inferences from cross-sectional regression results. Apart from that, it is questionable how long values measured for a particular year will retain their validity. This may be because technical progress affects the productivity of different types of labor differently or because supplies of different types of labor change rapidly. 1I Thus, even if the first problem is ignored, the return estimated from a cross-sectional earnings function will be valid only for a marginal expansion of education, which does not disturb the general equilibrium of skilled wage rates very much. Otherwise, with downward sloping demand curves for skilled labor, a large expansion in the supply of educated manpower could reduce wage rates significantly. Even though education commands a large income premium when it is scarce, it will not necessarily continue to do so when it has become much more plentiful. If, however, the macrodemand curves for skilled labor are perfectly elastic (reflecting infinite substitutability in production between different types of labor), the calculated returns will prevail even for large educational programs or policies (see Anand, 1976b). 9. It may happen that direct costs roughly match student vacation earnings, in which case k = I approximately. For example, if the earnings forgone are those for the academic year of, say, nine months (as opposed to the whole year), and if direct costs match earnings dunng the remaining two or three months, then k = 1. 10. PES incomes are already pre- rather than posttax (see chapter 2), so no adjustment is required on the benefit side to calculate the social return. See Hoerr (1973) for estimates of educational costs in Malaysia and a social cost-benefit analysis of education based on the 1967-68 Malaysian Socio-Economic Sample Survey of Households. I1. In view of the government's policy of employment restructuring, supplies of different types of educated labor are indeed likely to change rapidly during the period of the Outline Perspective Plan, 1970-90 (see the sections "The New Economic Policy" in chapter I and "Some Implications of Employment Restructuring" in chapter 8). EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 243 Two further points need to be noted in discussing rates of return to education. First, if education is used merely as a screening device and does not contribute to the productivity of labor, the calculation of the private return is unaffected, but that of the social return will have to be based on factors such as the v'alue of the information obtained through screening. 2 Second, since the possibility of unemployment has not been taken into account, the expected (in the statistical sense) rate of return should be considered. Thus, if the alternative to schooling is unemployment, the opporlunity cost or earnings forgone will be zero. Therefore the return to education for an unemployed person will be higher than that for an employed person contemplating the same increment in schooling. Similarly, the return expected from schooling at a particular level will be affected by unemployment at the next level. In other words, k will vary according to the individual's actual and prospective employment status. If the probability of unemployment decreases with education, the unadjusted regression estimate fi, will underestimate the expected rate of return. For in that case the ratio between expected income gained and expected income forgone will be larger than the ratio between actual income gained and actual income forgone. Although the coefficient fi1 is only a crude version of rate of return, for convenience I shall hereafter call j3, the "return" (not rate of return) to schooling: remember that /l1 does at least measure the proportionate income increase associated with schooling. For policy purposes it is sometimes useful to consider the absolute income increase, in which case #I needs to be multiplied by the appropriate base income level y. Some Problems of PES Data Since income data in PES are not coded by source, it is not possible here to distinguish earned income from other types such as income from business or property (see the subsection "Definition of PES Income" in chapter 2). In order to fit the earnings function, therefore, it was necessary to choose a subset of income recipients for whom earned orlabor income was believed a priori to be coterminous with total income. Thus the sample was narrowed down to employees in the urban sector only, a subsample of '8,263 individuals (5,843 males and 2,420 females).'3 12. See Stiglitz (1975) for a discussion of the theory and social benefits of screening. 13. The returns to education for this subsample of individuals may yet be overestimated owing to the likelihood of higher-income (and older) employees' deriving a larger share of their income from nonhuman capital (property and accumulated saving). In the regressions that follow, the income interval means assigned to these individuals are the same as those assigned to income recipients in general (see table 2-1). Table 7-1. Earnings Functions for Urban Male Employees by Occupation and Race R2 and F-ratto standard and Estimated regression error of degrees of equation estimate Jreedom Mean and variance of log y, S, T, and y Occupation 1: Professional and technical lOgYM = 6.74401+0.08705S +0.0551IT-0.00072T2 R2 = 0.324 F = 31.64 logy = 8.435 Y= 12.027 T= 17 394 y= 5,340 (0.00934) (0.00950) (0.00019) SEE = 0.454 DF = 198 var(logy) = 0.301 var(S) = 19279 var(T) = 134.647 var(y) = 9.856 106 logyc =6.35481 +012691S+0.04895T-0.00066T2 R2 = 0.469 F = 8267 logy = 8.512 S= 13.170 T= 14.202 y= 6,372 (000815) (0.01022) (0.00024) SEE = 0.539 DF = 281 var(logy) = 0.542 var(S) = 19.919 var(T) = 115.629 var(y) = 24404 10' logyj = 589927 +0.14896S+0.07980T-000129T2 R 2= 0.670 F = 72.26 logy= 8.612 S= 12.820 T= 18.550 y= 8.190 (0.01331) (0.01297) (0.00023) SEE = 0.545 DF = 107 var(log y) = 0.875 var(S) = 28.749 var (7) = 181.159 var(y) = 69 739 iO6 logyo = 6.81433+0.11814S+0.08517T-0.00197T2 R2 = 0.280 F = 2.85 logy = 9.196 S= 15.422 T= 13.519 y= 14,379 (0.05662) (0.04993) (0.00135) SEE = 0.860 DF = 22 var(logy) = 0.904 var(S) = 11667 var(T) = 124.110 var(y) = 140.682 106 log YTOT = 6.30867 + 0.12408S + 0.06305T-.00096T2 R2 = 0.461 F = 176.81 logy = 8.534 3-= 12.833 r= 15.980 y= 6,695 (000579) (0.00602) (0.00013) SEE = 0.549 DF = 620 var/logy) = 0.557 var(S) = 21 391 var(T) = 136.642 var(y) = 35 767 106 Occupation 2: Administrative and managerial log YM = 609361 + 0.12568S + 0.08452T- 0.00088T' R' = 0.464 F = 22.21 log y = 8.749 S = 11.710 T = 20.068 3 = 9,610 (0.01956) (0.01630) (0.00032) SEE = 0.646 DF = 77 var(logy) = 0.750 var(S) = 20.018 var(T) = 179.973 var(y) = 192 918 106 log Yc 6.03720 + 0.14174S + 0.08995T- O.O0098T1 RI = 0.485 F = 38.65 log y = 8.933 S' = 10.496 T= 23.543 3 = 10,455 (0.01477) (0.01583) (0.00030) SEE = 0.607 DF = 123 var(logy) = 0.699 var(S) = 19625 var(T) = 171.290 var(y) = 74,512 106 logyl = 5.95274 +0.15920S +006370T-000081T' R' = 0464 F = 12.99 logy = 8.494 S= 10.122 r= 25 551 y = 6,611 (0.02779) (0.02468) (0.00044) SEE = 0.558 DF = 45 var(log y) = 0.545 var(S) = 11 453 var(T) = 207.450 var(y) = 41 999 106 log yo = 6.98792 + 0.16349S + 0.06891T- 0.00102T' R' = 0371 F = 0 79 log y = 10.274 T= 14.125 7= 25.250 y = 37,470 (0.11222) (0.13823) (0.00296) SEE = 0752 DF = 4 var(logy) = 0.514 var(S) = 10.125 var(T) = 126.786 var6.) = 1.154.248 106 1ogyToT=596794+0.14145S+0.09120T-000108T' R2 =0453 F = 72.19 logy = 8.836 S' 10908 T= 22904 Y= 10,302 (0.01115) (0.01072) (0.00020) SEE = 0.650 DF = 261 var(logy) = 0765 var(S) = 18.490 var(T) =181.501 var(y) = 157318 106 Occupation 3: Clerical and related logyM =6.00163+0.10015S+0.08555T-0.00116TI R' = 0341 F = 42.70 logy= 7818 Y= 9.502 T= 14.808 y= 3.212 (0.01320) (O.OiO42) (0.00023) SEE = 0.617 LF = 248 var(logy) = 0.570 var(S) = 10.724 var(T) = 126.649 var()) = 10.475 106 log yc = 5.96774 + 0.09660S + 0.09828T7- 0.00139T' R' = 0.368 F = 80.34 log y = 8.052 Y= 10.078 7= 19.195 3 = 4,027 (0.00949) (0.00730) (0.00014) SEE = 0.580 DF = 414 var(logy) = 0.528 var(S) = 14.050 var(T) = 190.459 var(y) = 11885 106 logy1 =5.43999+0.11824S+0.11640T-0.00165T' R' = 0.516 F = 71.49 logy = 7.858 S= 9.766 J- 17:239 y= 3,613 (0.01316) (0.01160) (0.00025) SEE = 0.620 DF = 201 var(log y) = 0.783 var(S) = 12.072 var(T) 154.349 var(y) = 8.926 106 logyo = 3.89049+0.09287S+0.29097T-0.00505T2 R2 = 0.657 F= 9.56 logy = 7.776 S= 9184 T= 19.132 y = 3,32i (0.08029) (0.05657) (0.00108) SEE = 0.649 DF= 15 var(log y) = 1.022 var(S) 4.950 var(T) 142.968 var(y) = 4.497 106 log YTOT= 5-79988+0.10511S+0.10116T-0.00141T2 R=Q0.407 F=203.26 logy = 7.935 S= 9825 T= 17.508 y= 3,688 (0.00655) (0.00525) (0.00011) SEE 0.607 DF = 890 var(logy) = 0.618 var(S)= 12.502 var(T) 166.140 var(y) = 10.746 106 Occupation 4 Sales' logyM =5.23758+0.10067S+0.11222T-0.00147T2 R2 =0330 F = 9.69 logy= 6.898 S= 7.079 T- 11.841 Y= 1,581 (0.03176) (0.02774) (0.00057) SEE= 0.755 DF = 59 var(log y) = 0.810 var(S) = 12.486 var(T) 121.660 var(y) =5.477 106 log yc = 5.14879 + 0.13317S + 0.12290T-0.00168T' R' = 0.502 F = 18.42 logy = 7.339 S= 6.430 T= 17.187 ,=2,180 (0.00887) (0.00638) (0.00012) SEE = 0.598 DF = 552 var(log y) = 0.714 var(S) = 10.727 var(T) 168.235 var(y) = 5.944 106 logy1 = 5.73251 + 0.08007S + 0.07022T- 0.00088T2 RI = 0.377 F = 20.36 logy = 7.073 S = 6210 T= 20.714 y = 1,521 (0.01765) (0.01395) (0.00026) SEE = 0.569 DF = 101 var(log y) = 0.504 var(S) = 11.436 var(T) = 267.571 var(y) = 1.920 106 logyTOT= 5.21236+0.12212S+0.11912F-0.00166T' RI=0.466 F =210.22 logy= 7.260 S = 6.466 T= 17.202 y= 2,059 (0.00787) (0.00583) (0.00011) SEE = 0.626 DF = 724 var(log y) = 0.730 var(S) = 10.995 var(T) = 182.112 var(y) = 6.197 106 t Occupation 5: Service ; logym = 5.34974+0.13303S+0.10608T-0.00142T2 R2 = 0.465 F = 103.82 logy = 7.717 S= 6.844 7= 21.924 j = 2,753 (0.01034) (0.00940) (0.00020) SEE = 0.488 DF = 358 var(log y) = 0.441 var(S) = 9.216 var(T) =131.243 var(y) = 4.537 106 logyc =6.08223+0.07743S+0.06183T-0.00076T2 R2=0.233 F= 27.87 logy= 7.257 S- 5.137 T= 24.623 y= 1.780 (0.01329) (0.00741) (0.00011) SEE =0.615 DF= 276 var(logy) = 0.488 var(S) = 12.518 var(T) 374.735 var(y) = 2.028 106 logyj = 6.28622+0.08078S+0.03476T-0.00032T2 R' =0.177 F = 12.52 logy = 7.355 S= 5.168 T= 29.117 y=1,881 (0.01518) (0.00803) (0.00011) SEE = 0.557 DF = 175 var(1ogy) =0.370 var(S) = 11.621 var(T) = 281.079 var(y) = 1.971 106 logyo =3.07565+0.40362S+0.13175T-0.00172T2 R-=0.520 F= 2.17 logy = 8.408 S= 9.800 T== 15.900 y=8,762 (0.15973) (0.14114) (0.00299) SEE = 1.308 DF = 6 var(logy) - 2.379 var(S) = 11.733 var(T) 181.878 var(y) = 49.673 106 logyTOT=5.82197+0.11204S+0.06981T-0.00082T2 R2 =Q0325 F= 132.67 logy =7.492 S= 5.943 T.= 24.310 y=2,310 (0.00747) (0.00419) (0.00007) SEE = 0.591 DF 827 var(logy) = 0.516 var(S) = 11.717 var(T) = 253.652 var(y) =4.338 106 Occupation 7: Farm Iaborersa logyM =6.18609+0.05126S+0.04171T-0.00055T2 R2=0=073 F= 338 logy=7.027 S= 5.023 r= 23.659 = 1,385 (0.02377) (0.01634) (0.00028) SEE = 0.692 DF = 128 var(log y) = 0.505 var(S) = 8.049 var(T) = 173.940 var(y) = 0.684 106 logyc = 5.33538+0.09833S+0.11264T-0.00149T' R 2= 0.458 F= 2053 logy =7.148 S= 4.649 7= 25.273 j=1,728 (0.02632) (0.01443) (0.00020) SEE = 0.608 DF = 73 var(log y) = 0.655 var(S) = 12.382 var(T) = 367.392 var(y) = 2.350 106 logy, =6.29225+0.05935S+0.04978T-0.00077T2 R2 =0.107 F= 1.68 logy =7.148 S = 3.609 7= 31.565 p=1,649 (0.04321) (0.02904) (0.00043) SEE = 0.690 DF = 42 var(log y) = 0.498 var(S) = 8.410 var(T) = 210.729 var(y) = 3.465 106 log YTOT = 5.95603 + 0.06063S+ 0.06386T-0.00088T2 R2 = 0.154 F = 15.28 logy = 7.086 S = 4.648 T= 25.621 y = 1,536 (0.01674) (0.00977) (0.00015) SEE = 0.684 DF = 252 var(log y) = 0.546 var(S) = 9.588 var(T) = 245 046 var(y) - 1.689 106 Table 7.1 (continued). R2 and F-ratio standard and Esiimated regression error oJ degrees oJ equation estimate Jr-edom Mean and varianice of log y. S. T. and y Occupation 8. Production workers logyM =5.79966+0.08216S+0.09019T-0.00123T2 R2 =0326 F= 60.89 logy = 7504 3= 5938 T= 22413 = 2,174 (0.01022) (0.00812) (0.00015) SEE = 0.523 DF = 377 var(logy) = 0.403 var(S) = 9.203 var(T) = 151.710 var(y) = 1807 106 logyc = 5.60324+0.08086S+0.11354T-0.0015572 R2 =0447 F= 42244 logy = 7.396 S= 5.120 T= 20.402 y=2.040 (0.00569) (000357) (0.00006) SEE = 0 545 DF = 1.569 var(log y) = 0535 var(S) = 9 191 var(/() = 189.250 var(j) = 1 845 10° log y = 6.52732+0.06709S+0.04647T-000053T2 R2 = 0.315 F = 6809 logy = 7.644 3= 5.731 T= 25.262 y =2,351 (0.00683) (0.00688) (0.00012) SEE = 0.400 DF = 275 var(logy) = 0231 var(S) = 17.086 var(T) = 188.176 var(y) = 1.780 106 logyo = 6.89637+0.16673S+0.06364T-0.00201T2 R2 = 0503 F = 2.70 logy= 8281 3= 9.625 T= 15458 = 7,574 (0.08402) (0.07988) (0.00181) SEE = 0.919 DF = 8 var(logy) = 1.236 var(S) = 27278 var(T) = 154 112 var(y) = 96.127 106 log YTOT= 5.73677+0.07933S+0.10148T-0.00137T2 R2 =0.403 F = 50327 logy = 7.450 S= 5.359 T= 21.321 Y=2,131 (0.00423) (0.00298) (0.00005) SEE = 0541 DF = 2,241 var(logy) = 0.489 var(S) = 10.459 var(T) = 185.244 var(y) = 2461 1O6 All occupations together logyM =542422+014240S+009301T-0.00120T' R2=0451 F= 40264 logy =7.738 S= 7890 T= 19.834 j=3,241 (0.00451) (0.00445) (0.00009) SEE = 0.613 DF = 1.469 var(logy) = 0.684 var(S) = 17.044 var(T) = 153.706 var(y) = 19.104 106 logyc = 5.32333+0.13874S+0.11040T-0.00143T2 R2 =0.521 F = 1.199 62 logy = 7606 S= 6.853 T= 19.768 y=2,979 (000283) (0.00246) (0.00004) SEE = 0.612 DF = 3,312 var(logy) = 0780 var(S) = 18.779 var(T) = 205.500 var(y) = 12.466 106 log yi = 5.69940 + 0.13495S + 0.07047T - 0.000847T2 R' = 0.476 F = 293 74 logy = 7.704 S= 7.457 T = 23.339 y = 3,287 (0.00486) (0.00441) (0.00007) SEE = 0.612 DF = 970 var(logy) = 0712 var(S) = 22.413 var(7) = 228.454 var(y) = 17.462 106 log yo = 4.68325 +0.23471S +0 10384T-000138T2 R2 = 0.520 F = 27.43 logy = 8.590 S = 11.744 T= 16.819 Y = 11,804 (002701) (003273) (0.00075) SEE = 0.989 DF = 76 var(logy) = 1961 var(S) = 20 525 var(T) = 150.128 var(y) = 266112 106 logYTOT = 5.41823 +0.14005S+0.097937T-0001257T2 R2 = 0.492 F = 1.881.79 logy = 7.669 3= 7282 T= 20339 y = 3,217 (0.00215) (0.00193) (0.00003) SEE = 0.628 DF = 5.839 var(logy) = 0775 var(S) = 19437 var(T) = 197.338 var(y) = 19.437 106 Note The subscripts M, C. 1, 0, and TOT refer to Malay. Chinese, Indian, other, and total, respectively Numbers in parentheses are the standard error of the estimated coefficient The variables S and T are measured in years. and the vanable y in Malaysian dollars per year a The lack of observations on "others" in this occupation prevented the estimation of a regression equation for them. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 247 Workers in the rural sector were excluded for two reasons. First, it is likely that a significant fraction of their income takes the form of returns to nonhuman capital, in particular, land and other physical assets. Second, the earnings model as developed may not apply equally well to farm labor for various institutional reasons, such as capital and labor market imperfections and the nature of rural work. If it were possible to hold constant the ownership of assets other than human capital, an analysis of the effects of education on agricultural productivity and income would be particularly important, given the objective of raising smallholder pro- ductivities analyzed in chapter 5. This points urgently to the need to collect data on physical assets such as land, livestock, and houses in future income and expenditure surveys in the country. Estimates by Race, Occupation, and Sex Table 7-1 shows estimates of the earnings function for a sample restricted to urban male employees; there is also a breakdown by broad occupational category and racial group.14 Table 7-6 in the appendix to this chapter shows estimates of the earnings function for urban female employees. The last panel of table 7-1 (and the penultimate panel of table 7-6) shows the estimates for all occupations together, fitted separately by racial group. The last panel of table 7-6 provides earnings function estimates for all urban employees, male and female, disaggregated by race. In the tables the subscripts M, C, 1, O, or TOT on the dependent variable log y indicate the particular subsample of Malays, Chinese, Indians, others, or total, respectively, for which the equation has been estimated. The numbers in parentheses below each estimated coefficient (except the constant) of the r egression equation give the standard error of the coefficient. ' 5 SEE and Frefer to the standard error of estimate and F-ratio, respectively, and D F denotes degrees of freedom. The next four columns in the panel show the mean and variance of log y, S, T, and y, respectively (y is here measured in Malaysian dollars per year). The mean of logy is the 14. T he same definition of broad occupational categories is used here as in earlier chapters (see note to table 6-11). There were no observations for occupation 6 (farmers) in the sample of urban employees, farmers being either self-employed or employers. But there was a small number of employees (256 males and 217 females) in occupation 7 (farm, forestry, and fishing workers). 15. The computer program used to estimate the regression equations unfortunately does not provide standard errors for the constant term. 248 INEQUALITY AND POVERTY IN MALAYSIA logarithm of the geometric mean income, and the variance of logy is a familiar measure of inequality (the variance of log-income).'6 For regressions on microdata such as the PES sample, the overall fits obtained are rather good. They are certainly comparable to fits of the same equation estimated for developed countries (see Chiswick and Mincer, 1972; Mincer, 1974; and Psacharapoulos, 1973). Almost all the equations and almost all the coefficients of independent variables are significant at the I percent level. The variables S and T possess positive coefficients, and the coefficient of T' has the expected negative sign. In general, the results are fairly encouraging. The goodness-of-fit of the equation is shown by an R2 of 0.492. This means that age and education by themselves explain about half the observed inequality in earnings, as measured by the variance of log-income. This level of explanation is really quite impressive in light of the many important income-determining variables that have been omitted, such as the number of weeks in the year actually worked, ability differences, and the quality of schooling received."7 The parabolic age-log-income profile is confirmed for each level of educatiofi by the negative and significant coefficient on T2.8 The coefficient on S substantiates a positive association between education and income. There is a 14.01 percent coefficient on education for all males in the urban sector, with a t-ratio of more than 65.19 The arithmetic mean income of the 5,843 = DF + 4) urban male employees in the sample is M$3,217 per year (or M$268 per month), and the variance of log-income is 0.775.2o Their average level of schooling is 7.282 years and of experience, 20.339 years. Hence, the average age of males in the sample is 32.621 years (since X = T+Y+ 5 = 20.339 + 7.282 + 5.000). As 16. Note also that the square root of vanance, divided by the mean, of S and T furnishes measures of inequality (the coefficient of variation) for schooling S and experience T (see chapter 3 and appendix A). 17. The omitted variables are on the so-called supply side; vanables on the demauid side have been ignored altogether. 18. The equation for the racial group "others" in some occupations is not significant, and neither are the 1-ratios of some coefficients. In general there are very few degrees of freedom for regression equations for this group (see table 7-1). 19. The coefficients f, fl, and 3 will be unbiased if the omitted variables are uncorrelated with the included ones of education, experience, and experience squared. This may not be a good assumption for the omitted variable of ability (or father's income), which might well be correlated with education. In that case, the estimated coefficient fl, will be biased upward. 20. This degree of inequality is smaller than that among all employees in Malaysia. From table 6-6. the vanance of log-income for the latter group is 0.816. But all urban employees, male and female, display a larger degree of inequality in their incomes (0.868. from the last panel of table 7-6) than this. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 249 shown in the appendix to this chapter, the peak of the experience-income profile occurs at an experience level of #2/2fi3 = 0.09793/2(0.00125), or 39.172 years. For the average level of schooling (7.282), therefore, the age- income profile peaks at 51.454 years. It peaks later for those with more schooling and earlier for those with less, and the later peak occurs at a higher level of income (see the appendix to this chapter). Racial Differences Of the three major racial groups in Malaysia, the Malays obtain the hiighest return to education. Each additional year of schooling (holding experience constant) raises the annual income of a Malay by 14.24 percent, of a Clhinese by 13.87 percent, and of an Indian by 13.50 percent.2' (The following discussion of racial differences excludes consideration of the other races.) Although each of these coefficients has extremely high t-ratios, paired t-tests suggest that they are not significantly different. The overall return of 14.01 percent falls within the confidence interval of the three separate coefficients.22 At the 95 percent significance level one cannot reject /, = 0.1401 for any racial group.23 The higher coefficient for Malays is consistent with education's giving them a better entry into occupations with high incomes, such as government jobs. The arithmetic mnean income of Indian male employees (M$3,287) is higher than that of Malays (M$3,241), which in turn is higher than that of Chinese (M$2,979). The Indian:Malay:Chinese labor income disparities for males stand at 1.10:1.09:1.00.24 Since the returns to education between the groups are highly similar, differences in mean income should be associated with differences in levels of schooling and experience.25 This is 21. Every additional year of schooling raises the income of a member of the other races by 23.47 percent. 22. The 95 percent confidence interval is the coefficient value, plus or minus approximately two standard errors. 23. Akt any rate, differences between occupations within a race seem to compensate for between-race differences within an occupation, so that when all occupations are considered together, no racial differences seem to emerge. 24. The sample excluldes rural workers, and urban workers are relatively highly represented in govemment. 25. Since, however, the experience-income profiles are different for the three groups (significantly different P2 and P,), so also are the age-income profiles. Corresponding to the average level of schooling for each group, the age-income profile peaks at 51.64 years for the Malays, at 50.45 years for the Chinese, and at 54.41 years for the Indians. The profile is most sharply peaked for the Chinese, as measured by the sLze of the coefficient f3 (see the appendix to this chapter). 250 INEQUALITY AND POVERTY IN MALAYSIA indeed borne out. Since A =T+5+5, the combination of average schooling and experience levels for a group implies the average age level. In the sample, the average age of Indians is 35.80 years, of Malays 32.72 years, and of Chinese 31.62 years. This is the same as the relative income position of the three communities. The ranking of communities according to average schooling levels, however, is different: the Malays have the highest average level of education (7.89 years), followed by the Indians (7.46 years), and then the Chinese (6.85 years). Inequality in earnings, as measured by the variance of log-income, is smallest for the Malays at 0.684; the Indians are next with an inequality level of 0.712, and the Chinese have the highest inequality of earnings at 0.780. This inequality in earnings is associated exactly with inequality in education and experience for the races. The Malays have the smallest inequality in years of schooling as measured by the coefficient of variation, (var S)"2/23 The coefficients of variation of schooling for Malays, Indians, and Chinese are 0.52, 0.63, and 0.63, respectively. Inequality in years of experience displays the same ranking, indicated by a coefficient of variation of 0.63 for Malays, 0.65 for Indians, and 0.73 for Chinese.26 Two further questions concerning these findings are of some policy interest. First, how does preferential government policy toward the Malays in employment, promotions, and university quotas bear on these results?27 Second, what part of the inequality between Malays and non-Malays can be explained by differences in educational attainment? Some observers claim that the higher Malay returns are linked to the government's discrimination in employment and promotion policies. But it is difficult to quantify the extent to which the application of pro-Malay policies might have affected the above results. Nevertheless, certain occupations, in which the application of pro-Malay policies is likely to have been most intense, do display higher returns and average incomes for the Malays (see, for example, occupation 20, government administrators and legislative officials, in table 6-1 1). A countervailing factor, however, may be operating in the case of the Chinese. They control a large share of private business and probably practice preferential hiring of other Chinese in their businesses (for reasons of language or clan). This fact is consistent with the lower average levels of schooling among Chinese employees but com- 26. It can be shown that inequality in earmings is affected not only by inequality in S and T (and their covariance), but also by the average levels of S and T. To demonstrate this, simply take the variance of both sides of the earnings function. 27. There is, for example, aguarantee of places and jobs for the Malays in the public sector. See article 157 of the constitution of Malaysia. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 251 parably high rates of return.2" On balance, it is not possible to deduce a bias in any particular direction. The second question presents a larger set of issues, which requires a comparison of labor and nonlabor incomes between the races. This was done in the previous chapter by use of the surrogate variable, employment status. In regard to labor incomes, it is not unreasonable to assert that differences in educational attainment do account for much of the difference in earnings. This follows from the statistically insignificant difference in returns to education among the major racial groups. Other things being equal, it would seem that labor income differences are associated with differences in education received. If causation can be attributed, and there is high substitutability in production between labor of different levels of skill or education, these findings suggest that education could be an instrument of mobility for Malays in the category of employees. (But see the caveats on drawing time series inferences from cross-sectional estimates of the earnings function in the earlier section "The Return to Education.") Occupational Differences Substantial differences exist in income inequality for urban male employees across occupations. The degree of inequality in income is fairly small among production workers [var(logy) = 0.489], service workers (0.516), and professional and technical staff (0.557). The highest degree of inequality is found among administrative and managerial personnel (0.765). These results are again explained by differences in the level and inequality of schooling and experience. One of the main findings is that returns to education are generally high in modern sector occupations and generally low in the traditional sector. From table 7-1, it is seen that high returns obtain in the administrative and managerial (14.15 percent), professional and technical (12.41 percent), and sales (12.21 percent) occupations; and low returns obtain among farm laborers (6.06 percent) and production workers (7.93 percent). As expected, the return to formal schooling is highest in white-collar activities and relatively low in blue-collar (production worker) and traditional sector (farm laborer) jobs. Schooling seems to suit better the job requirements of modern sector (tertiary) occupations. It is also the case that the (human capital) earnings model performs least well in explaining the incomes of traditional sector employees (see the 28. Mlany Chinese start work in business (a family-owned shop or commercial establish- ment) siraight after primary school. 252 INEQUALITY AND POVERTY IN MALAYSIA earlier section "Some Problems of PES Data"). It explains only (R2 =) 15.4 percent of the variance of log-income of farm laborers. At the other end, the best explanations of inequality by the model are for the sales (R2 of 0.466), professional and technical (R2 of 0.461), and administrative and managerial (R2 of 0.453) categories.29 The average levels of schooling are also lower in the traditional than in the modern sector occupations. The average level of schooling for farm laborers is only 4.65 years, which is lower than that in any other occupation. For professional and technical personnel the average level of schooling is as high as 12.83 years, and for administrative and managerial personnel it is 10.91 years, but these job classifications undoubtedly require more education. Average income levels across occupations are positively cor- related with average schooling levels. Mean incomes for the administrative and managerial and the professional and technical categories are at the top of the scale-while an average monthly income of only M$128 puts farm laborers at the bottom of the scale. These results are consistent with another explanation. It could be that certain occupations have higher rewards associated with them purely for institutional or class reasons. Entrants into these occupations might be screened by means of educational qualifications. Then higher schooling levels would be associated with the higher paid occupations. Thus, even if its relation with productivity is tenuous, it may be that education is deployed as a screening device for certain high-income occupations, which would explain the observed positive relation between income and education.30 The ethnic breakdown within each occupation shows that Indians seem to benefit more than other groups from schooling in occupations 1-3 (professional and technical, administrative and managerial, clerical and related). The explanation for this is probably sociocultural (see Sandhu, 1967 and 1969). At the same time, the Malays display lower returns than other racial groups in the professional and technical and the ad- ministrative and managerial occupations. As service workers, however, they obtain the highest return to schooling. The relation between starting salary and overall lifetime income in an occupation can be illustrated by the separate age-income profiles for 29. Of course, the amount of variation in log y is itself different for each occupational category, so the R2es would be different even with the same standard error of estimate. But the R2's still represent the proportion of inequality explained by education and experience in each occupational category. 30. It is not clear how the effects of experience can be explained in these termis-by gerontocratic pnnciples, perhaps, or a "fiist in last out" system for nontransferable skills. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 253 occupalions. If lifetime income (undiscounted) in an occupation is approximated by the mean income of employees belonging in it (at different stages of their life cycle), it can be seen that starting salary is imperfectly correlated with expected total income.3" Administrative and managerial personnel, for instance, have a lower starling salary but 50 percent more lifetime income than professional and technical persons. The former's earnings peak at a later age and a higher income level than the latter's (the earnings peak is given by an income level of exp[30+ #hS+(±(1/4#3)]; see the appendix to this chapter). At the other end, production workers and farm laborers have a low earnings peak-which also occurs at an earlier stage in their life cycle. These results shed some light on an interesting question concerning urban unemployment in Peninsular Malaysia. There is currently substan- tial unemployment among secondary school-leavers in the country (see Mazumdar, 1975), who are said to be queuing up for white-collar jobs in clerical and related occupations. It is suggested that starting pay scales in government create unrealistically high earnings expectations in such jobs, and this has caused an oversupply of, and unemployment among, secondary school-leavers. It turns out, however, that clerical and related occupations have not only relatively high initial salaries, but also high lifetime incomes (irt relation, for example, to comparable occupations such as sales and services). Thus it is not clear that starting salary rather than lifetirne income has influenced people's job choice in this case. Male-Female Differences Regression equations parallel to those in table 7-1 have been estimated for female employees in the urban sector and are included as table 7-6. In the last panel of thi s table male and female employees in all occupations are pooled. As in the case of males, the earnings function for females is highly significant and displays the predicted signs on S, T, and T2. The return to schooling is slightly lower for females (13.77 percent) than for males (14.01 percent), but a paired t-test shows the difference to be not statistically significant. The earnings model explains a smaller fraction of income inequality among females than among males (R2 of 0.430 compared with 0.492). A major reason is probably that Tis an inaccurate measure of labor 31. Starting saiaries are the anti-log of the term , + Jf,S in the earnings function. The estimate obtained ther-eby will be biased, though consistent; the earnings function will give an unbiased estimate only of the logarithm of the starting salary. 254 INEQUALITY AND POVERTY IN MALAYSIA market experience for women (see the earlier section entitled "The Earnings Function").32 Females generally obtain a lower return to schooling within each occupation than males.33 A notable exception is the case of occupation 3, clerical and related workers, in which females obtain a higher return (13. 25 percent) than males (10.51 percent). This is perhaps due to their superior skills in this occupation and the lack of opportunities for employment in others. Women in clerical and related occupations have an average of 12. 11 years of schooling compared with 9.83 years for men. The average levels of experience (17.35 years), schooling (6.59 years), and therefore age (28.94 years) are all lower for females than for males. The age-income profile is flatter for females than for males, as measured by the size of the coefficient 33,34 which is 0.00073 for females compared with 0.00125 for males. This may reflect the fact that women have more limited opportunities for promotion than do men. The female experience-income profile peaks at an experience level of 43.62 years, which exceeds the level at which the male experience-income profile peaks (39.17 years). It should be reiterated, however, that the measure T of labor market experience is not satisfactory for women. Female mean incomes are lower than male mean incomes within every occupation; overall, the mean income of females is slightly more than half that of males (M$1,784 per year compared with M$3,217). Within each occupation, income inequality among females is lower than among males, but overall it is higher (varlog of 0.837 compared with 0.775). This reflects greater between-occupation differentials for women. For all 8,263 urban employees together, male and female, there is a 14.77 percent return to schooling. Their mean income is M$2,798 per year, and variance of log-income is 0.868. Their average level of schooling is 7.08 years; of experience, 19.46 years; and of age, 31.54 years. The age-income profile corresponding to the average level of schooling for all urban employees peaks at 52.41 years. The coefficient of variation of schooling is 0.677 and of experience 0.734. The variables S, T, T2 explain 48.5 percent of the inequality in income among urban employees. 32. Perhaps T is a better measure of labor market experience in developing than in developed countries. The institution of the joint family may allow working women to leave children at home to be looked after by relatives, so that women employees need not withdraw from the labor force for very long in Malaysia. 33. The equation for occupation 2, administrative and managerial personnel, does not have enough degrees of freedom to allow valid statistical inference. 34. See property (6) and figure 7-2 in the appendix to this chapter. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 255 Estimates by Age Cohort The sample of urban male employees is disaggregated into three successive age cohorts (table 7-2) to study their separate earnings functions. The three groups consist of those under 30 years, those between 30 and 49 years, and those over 50 years. The motivation for this disaggregation is to identily possible differences in the return to education among cohorts. The regression results indicate significantly different returns for the three groups. The youngest age group obtains the highest return to schooling of 18.10 percent. This is followed by the middle age group (30-49 years), which obtains a return of 11.91 percent, while the oldest age group (50+ years) obtains a return of only 6.94 percent. A possible explanation for these findings is that there is a "vintage" or "obsolescence" effect in education. In other words, the quality of schooling has improved over time or previously acquired education has become less relevant or effective for present day (1970) job requirements.35 It is clear that significant changes have indeed taken place in the level and nature of education in Malaysia. For example, there is a markedly higher average level of schooling among the younger age groups (approximately eight and seven years, respectively) than among the oldest group (five years). Another explanation for the findings is that the independence assumption between the return to formal schooling and to experience is not valid (see the earlier section "The Earnings Function"). For example, if older people earn higher incomes because of their experience, an extra year of schooling may yield a lower return as a proportion of their present income. The fraction of inequality explained by the model (R2) is largest for the under 30 age group. Significantly weaker fits are associated with the older age groups, possibly because they derive larger fractions of their income in the form of returns to accumulated savings, that is, from nonhuman capital.36 The level of inequality in income is smallest for the middle group, with var(log y) = 0.519. Part of the reason for this may be the greater homogeneity of this group (the size of its defining interval, 30-49 years, is relatively small). 35. Another explanation could be that older people just happen to occupy jobs for which education is less important, or that education has been increasingly used as screening for jobs. 36. In fact, this would be a reason for observing higher returns to education in the estimates for the older age groups. The fact that returns are actually lower seems to reinforce the explanation offered above. Table 7-2. Earnings Functions for Urban Male Employees by Age Cohort Age: Under 30 years lg YTOT = 4.74180 + 0.18102S + 0.14101T- 0.00158T2 (0.00378) (0.00804) (0.00037) R2 = 0.470 F = 813.11 logy = 7.301 S= 7.977 T= 9.153 y=2,115 SEE = 0.619 DF = 2,749 var(log y) = 0.723 var(S) = 15.422 var(T) = 27.303 var(y) = 5.396 106 Age: 30-49 years log YTOT = 6.44727 + 0.11910S + 0.03987T- 0.00038T2 (0.00367) (0.00907) (0.00016) R2 = 0.390 F = 513.23 log y = 8.041 S = 7.083 T = 25.759 y = 4,235 SEE = 0.563 DF = 2,408 var(log y) = 0.519 var(S) = 21.939 var(T) = 62.786 var(y) = 28.869 106 Age: 50-98 years log YTOT = 10.72164 + 0.06939S - 0.10498T + 0.00073T2 (0.01000) (0.02800) 0.00026) R2 = 0.393 F = 145.40 logy = 7.840 S = 5.166 T = 46.480 y= 4,072 SEE = 0.710 DF = 674 var(log y) = 0.826 var(S) = 20.312 var(T) = 70.431 var(y) = 33.619 106 All ages log YTOT = 5.41823 + 0.14005S + 0.09793T- 0.00125T2 (0.00215) (0.00193) (0.00003) R2= 0.492 F = 1,881.79 log y = 7.669 S = 7.282 T = 20.339 y = 3,217 SEE = 0.628 DF = 5,839 var(logy) = 0.775 var(S) = 19.437 var(T) = 197.388 var(y) = 19.437 106 Note: See note to table 7-1. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 257 The average age of persons in the under 30 age group is 22.13 years, and in the middle and older groups it is 37.84 years and 56.65 years, respectively. The mean income level is highest for the middle age group (M$4,235 per year): a reflection of the fact that this group is closer to its earnings peak than are other groups. Predicted peaks in the age-log- income (or age-income) profile differ among the three groups. A peak at 57.60 years is forecast for the under 30 group, and at 64.54 years for the 30- 49 age group. Most employees in the 50+ age group are past their true income peak, and their earnings function is estimated mainly from incomes to the right of this point. The function fitted for this group turns out to be convex with a trough at 82.07 years, suggesting that a more gradual decline in income may occur than is indicated by the original concave functional form. But the function is obviously not valid over the entire range beyond the turning point, for instance after a person's withdrawal from the labor force. Not much significance should therefore be attached to the convex shape of the age-log-income profile for this group. For female employees, much the same results are in evidence (not shown here). The -smaller return to schooling acquired earlier is less marked, however, and the explanatory power of the function is rather weak for the oldest age group of females (R2 = 0.198). Language ol Instruction Returns to education are highest for males who have attended English- language schools (the category of "other" languages of instruction is excluded).37 For males who have been to such schools, the return to schooling is 16.38 percent (table 7-3). Average levels of income and schooling are also highest for these persons. The inequality in their incomes is largest, and the variables S, T, and T2 account for 59.4 percent of it. An education in the Tamil language receives the lowest return of only 6.71 percent. (Thie average level of schooling is also smallest for people who have attended Tamil schools.) Returns to education in the Chinese or Malay language are significantly higher (table 7-3)38 One explanation for these findings is the possibility that Tamil school qualifications are 37. Afterthe "other" races, theChineseobtain the highest return (not shown here)of 17.33 percent to schooling in the English language of instruction, followed by the Indians (15.18 percent), and the Malays (14.12 percent). 38. In the sample, Tamil-language schools were attended only by Indians, Chinese- language schools only by Chinese (except for one non-Chinese), and Malay-language schools mainly by Malays (847 out of 875). Table 7-3. Earnings Functions for Urban Male Employees by Language of Instruction English 1og YTOT = 5.08280 +0.16376S +0.120007-0.0016772 (0.00368) (0.00402) (0.00010) R' = 0.594 F = 1,068.62 iog y = 7.964 S = 10.569 T 13.888 9 4,671 SEE - 0.637 DF = 2,188 var(logy) = 0.999 var(S) = 14.592 var(T) = 117.017 var(y) = 39.442 101 Malay 10gYTOT = 5.33086 + 0.12168S + 0.104337'- 0.00137T2 (0.00635) (0.00558) (0.00011) R2 0.469 F = 256.46 logy = 7.511 S = 6.346 7 21.537 y = 2,364 SEE 0.552 DF = 871 var(logy) = 0571 var(S) = 10.631 var(T)- 146.532 var(y) = 4.695 106 Chinese log YTOT = 5.45793 + 0( 10707S + 0.1 1395T- 0.0016172 (0.00446) (0.00338) (0.00006) R2 0.452 F = 549.61 iog)y = 7.479 S = 5.587 1 21.164 = 2,298 SEE = 0.560 DF = 1,998 var(log y) = 0.572 var(S) = 9.092 var(T) = 167.278 var(y) = 4.140 106 7amil 1og YTOT = 6.14768 + 0.06711S + 0 065047T- 000092T2 (0.00989) (0.00801) (0.00013) R' = 0.250 F = 36.77 log y = 7.446 S = 4.981 T = 29.978 9 = 1,977 SEE = 0.483 DF = 330 var(log y) = 0.308 var(S) = 7.529 var(T) - 173.724 var(y) = 1.334 106 Other log YTOT = 6.07881 + 0.19314S + 0.01496T - 0.000t272 (0.02381) (0.02795) (0.00048) R2 = 0.691 F = 38.71 logy = 8 236 S = 9.616 / = 27.616 j= 7,490 SEE 0.715 DF = 52 var(logy) = 1.562 var(S) = 33.418 var(f) =185564 var(y) = 80.467 106 Noue: Owing to 384 individuals for whom ihe language of instruction was not available, the total number of observations here (5,459) does not tally with that in table 7-1 (5,843). See also the note to table 7-1. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 259 insufficiently recognized. In Malaysia, Tamil is neither an official working language nor a business language (unlike English, Malay, or Chinese). Another is that the effective demand for Tamil-educated workers is low. and many are employed in occupations (such as farm laborers and production workers) which show low rates of return to schooling. The high returns realized in English-language schools could be the result of two factors. First, it is likely that the quality of instruction in the English schools is superior to that in the vernacular schools-as manifested by higher teacher-student ratios and better facilities such as books and libraries. Second, the higher returns in the English schools might reflect a screening effect. A knowledge of the language is a definite advantage in many of the more highly paid occupations in government and business (for example, in multinational firms); in some, it might even be a prerequisite.39 Type of Degree In commenting on the age-income profiles of persons at the highest level of education, those who hold a degree or diploma, I consider first the shape of the income profile for males with the same degree and thus the same level of schooling. Then I look into the differences between such profiles for different degrees. The S term is absent from these equations because for each one the number of years of schooling is the same. The starting salary, that is, the salary with no experience (T = 0), is indicated by the size of the constant term in the regressions.40 It is evident from table 7-4 that the starting salary is highest for engineers and is closely followed by that for doctors; for teachers (those with an education degree or diploma), it is about two-thirds the level for doctors. The average lifetime income (again undiscounted) is highest for doctors, not engineers. So the earnings of medical degree holders rise faster and overtake those of engineers after starting at a lower level. 39. The average level of schooling is highest for those who have attended English-language schools, at approximately ten years. Similarly high average levels of schooling prevail in the professional and tl chnical and the administrative and managerial occupations-which is consistent with people in these occupations having been to English-language schools. 40. Unfortunately, the computer program used for estimation does not calculate the standard error for the constant term in the regression. Thus tests of statistical significance of difrerences between intercepts cannot be made. Note that the intercept in any case provides an unbiased estimate only of the logarithm of the starting salary. Table 7-4. Earnings Functions for Urban Male Employees by Type of Degree or Diploma Education degree or diploma lg YTOT = 8.34768 + 0.03216T - 0.00043T2 (0.01185) (0.00038) R2 = 0.141 F = 14.47 log y = 8.636 S = 16.000 T= 11.480 Y = 6,151 SEE = 0.365 DF = 176 var(logy) = 0.153 var(S) = 0.000 var(T) = 55.004 var(y) = 10.729 106 Medical degree or diploma log YTOT = 8.81623 + 0.04690T + 0.00050 T2 (0.03570) (0.00137) R2= 0.537 F = 11.59 logy = 9.313 S= 19.000 7= 8.913 = 14.406 SEE = 0.525 DF = 20 var(log y) = 0.541 var(S) = 0.000 var(T) = 82.538 var(y) = 128.270 106 Engineering degree or diploma log YTOT = 8.88111 +0.07041T-0.00152T2 (0.02128) (0.00061) R2 = 0.185 F = 6.82 log y = 9.306 S = 17.500 T = 9.500 1 12,935 SEE = 0.513 DF = 60 var(log y) = 0.313 var(S) = 0.000 var(T) = 71.742 var(y) = 65.622 106 Other degrees or diplomas" log YTOT = 7.79079 + 0.10650T- 0.00185T2 (0.01204) (0.00036) R2 = 0.313 F = 76.66 log y = 8.533 S= 17.000 T= 10.776 = 3,601 SEE = 0.813 DF = 336 var(logy) = 0.957 var(S) = 0.000 var(T) 102.784 var(y) = 14.427 10' Note: Although a breakdown by racial group was effected within each degree type, frequently there were not enough degrees of freedom to warrant it. The breakdown is not reported here. See also the note to table 7-1. a. Agriculture, pure and natural science, government and political science, management and administration, economics and commerce, and others EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 261 Inequality in the income of teachers is very small indeed, with a variance of log-income of just 0. 153. This largely reflects pay scales in state schools where, for example, the rate of annual pay increases is constrained to about 3 percent. The degree of inequality in the earnings of engineers is also fairly low, with var(log y) = 0.313. Again, this is probably explained by the absorption of a majority of engineers into government departments at predetermined salary scales (which are more equal than in the private sector). Doctors' incomes are more unequal. An important reason for this is probably the variation among doctors in private practice in the number of hours worked, which depends partly on different choices between income and leisure. Unit rates also vary, of course, owing to such factors as skill and reputation. Interestingly, concavity is not verified in the age-log- income profile for doctors (the coefficients on T and 7'2 not being significant). This may be due to labor-supply effects over the life cycle or the absence of institutionally determined salary scales. The main point is that the profession allows considerable flexibility in the number of hours workced. Female employees with equivalent qualifications have somewhat lower starting salaries both as teachers and as doctors-a reflection of possible discrimination against women. They also have significantly lower average income levels, possibly because of lower average levels of experience or because of part-time work, that is, differences in the "weeks worked" variable. Even with the same measured T, interruptions in post-training work experience could reduce average incomes.4 EBreakdown by Region The earnings functions were estimated separately for each state in Peninsular Malaysia (table 7-5).42 This allows an examination of inter- regional differences in education and incomes and may also shed some light on the extent to which regional inequality can be explained by differences in schooling and experience.43 41. The male-fernale difference in annual teacher salaries is not very large (M$6,151 conmpared with M94,563), nor is the difference in their average experience levels (11.48 years conmpared with 10.16 years). 42. Since, as noted in the appendix to chapter 5, there is no urban population for Perlis state, the estimates here refer to ten Peninsular Malaysian states only. 43. Interregional analyses using this framework are reported for the United States in Chiswick (1975). Thiere are too few observations here (ten states) to explain regional inequality in Lerms of the means, variances, and covariances of schooling, experience, and experience squared. Table 7-5. Earnings Functions for Urban Male Employees by State Johore log YTOT= 5.28116 +0.14021S+0.10718T-0.00146T2 (0.00638) (0.00574) (0.00011) R2= 0.474 F = 222.09 log y = 7.562 S= 7.255 T= 19.561 1 2.774 SEE = 0.634 DF = 739 var(log y) = 0.761 var(S) = 18.096 var(T) = 188.762 var(y) = 8.517 10' Kedah log YTOT = 5.19841 + 0.13604S + 0.10467T-0.00132T2 (0.01174) (0.00952) (0.00017) R2=0.433 F= 7314 logy =7.517 = 7.242 T= 21.095 = 2,770 SEE = 0.697 DF = 287 var(log y) = 0.848 var(S) = 16.761 var(T) =218 025 var(y) = 10.720 106 Kelantan 1og YTOT = 4.78393 + 0.17886S + 0.09575T - 0.00107T2 (0.01545) (0.01516) (0.00027) R2= 0.481 F = 48.88 log y = 7.433 S = 7.225 T 21410 V = 2,992 SEE = 0.781 DF = 158 var(logy) = 1.153 var(S) = 21.478 var(T)= 191.615 var(y)= 13.966 106 Malacca 1og YTOT = 4.87935 + 0.16398S + 0.11803T -0.00152T2 (0.01225) (0.01028) (0.00019) R= 0.560 F = 84.16 logy = 7513 S= 7.235 T= 20.270 V =2,835 SEE = 0.636 DF = 198 var(log y) = 0.906 var(S) = 17.834 var(T) = 214.125 var(y) = 10.787 106 Negri Sembilan log1 'ToT = 5.79133 + 0.14144S + 0.07413T-0.00116T2 (0.01278) (0.01254) (0.00023) R2 = 0 522 F = 53.84 log y = 7.721 S = 7.970 T = 20.503 y = 3,365 SEE = 0.616 DF = 148 var(log y) = 0.778 var(S) = 20.991 var(T) = 201 392 var(y) = 14.876 106 Pahang 10g YTOT = 5.76133 + 0.12102S + 0.09174T-0.00123T2 (0.00938) (0.00785) (0.00014) R2= 0-475 F = 81.41 Iogy Y 7.740 Y- 7.443 T= 19 513 y = 2,941 SEE =0.542 DF = 270 var(logy) = 0-554 var(S) = 16.952 var(T) = 198.567 var(y) = 4.541 106 Penanq 10g y TOT = 5.38412 + 0.12682S + 0.10299T- 0.00138T2 (0.00543) (0.00507) (0.00009) R2 = 0.501 F = 274.84 logY = 7.578 S= 7.066 T- 21.455 = 2,858 SEE = 0.583 DF = 822 var(log y) = 0.678 var(S) !7.720 var(T) = 202.397 var(y) = 13.027 106 Perak 1og yTOT = 5.39270 + 0.12466S + 0. 10675T - 0.001 39T2 (0.00500) (0.00438) (0.00008) R2= 0.499 F = 330.00 IOgY 7.623 S = 7.012 T= 21 343 Y = 2,868 SEE = 0.605 DF = 993 var(log Y) = 0.729 var(S) 19.592 var(T) = 206.390 var(Y) = 7.494 106 Selangor lg YTOT = 5.53159 + 0.14742S + 0.09323T - 0.001 IOT2 (0.00321) (0.00291) (000005) R2 = 0.566 F - 879.22 log Y = 7.819 = 7.374 T 19.745 = 3,863 SEE = 0.578 DF = 2,024 var(log y) = 0.768 var(S) = 21.019 var(T) = 190.495 var(y) = 36 804 106 Trenggyi 1ii log YTOT = 5.31429 + 0-1 6569S + 0.07339T - 0.0009IT2 (0.01599) (0.01711) (0.00033) R2= 0.396 F = 35.795 og-y= 7.573 3= 8.241 7T= 18.461 = 2.997 SEE = 0.752 DF = 164 var(logy) = 0918 var(S) = 19.886 var(T) = 167.904 var(y) = 12324 106 Peninsular Malaysia log y-101- = 5 41823 + 0.14005S + 0.09793T-0.00125TI (000215) (0.00193) (0.00003) R2 = 0.492 F = 1,881.79 logy - 7.669 T= 7.282 T= 20.339 V = 3,217 SEE = 0.628 DF = 5,839 var(log Y) = 0.775 var(S) = 19.437 var(T) = 197 338 var(y) = 19 437 106 Nole Peris is omitted from the table 5ince it has no urban population. Sec also the note to table 7-1, 264 INEQUALITY AND POVERTY IN MALAYSIA There are fairly wide interstate disparities in the average level of income and education for urban male employees (table 7-5). Mean incomes are lowest in Kedah (M$2,770 per year) and highest in Selangor (M$3,863 per year), the disparity ratio being 1.0: 1.4.44 Inequality in income, however, is relatively low in Selangor, but high in Kedah. In general, there appears to be a negative correlation across states between the level and inequality of urban employees' incomes. (This contrasts with the result in chapter 3 on the positive association across states between the level and inequality of individual per capita incomes.) The differences among states in returns to schooling are also fairly wide. They vary from 12.10 percent in Pahang to 17.89 percent in Kelantan,45 and there is a further negative association between the income level of a state and the return to s,hooling in it. The generally higher returns in poorer states can probably be explained by the fact that high-wage (skilled) labor has greater mobility nationally than does low-wage (unskilled) labor.46 This leads to a wider income gap in the poorer states than in the richer ones. It is also consistent with the negative correlation observed earlier between the level and inequality of income across states. The earnings functions fit best in the most developed state, Selangor, where they explain 56.6 percent of the inequality in incomes. The coefficient of determination (R2) is in fact more than 40 percent in every state except Trengganu (where it is 39.6 percent). Appendix: Some Properties of the Earnings Function The earnings function used in this chapter, log1y = 0± + #IS + ±B2T- #3T2, generates many of the observed characteristics of individual earnings profiles.47 These characteristics are depicted by a life cycle that begins with a period of no earnings, followed by a period in which earnings rise and then eventually fall off. Earnings are also observed to be a roughly concave 44. These interstate differences in the incomes of urban male employees are much smaller than the interstate differences in incomes of all employees (see chapter 6) 45. A paired 1-test shows that the two coefficients are significantly different from one another at the 99.5 percent level of confidence. 46. In other words, it is posited that there is a greater degree of equilibrium geographically in the market for skilled labor than for unskilled labor. 47. The properties of individual eamings profiles were originally derived in Anand (19 74b, pp. 10-14) in the section "Implications of the Human Capital Model of Earmings." EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 265 function of age, especially in the neighborhood of the peak. As a function of the level of schooling, earnings peak at a later age for those with more schooling, and the peak itself occurs at a higher level of income. This earnings function, borrowed from the theory of optimal investment in hurnan capital, turns out to fit the above characteristics rather well. This is demonstrated here, and some further properties of the earnings function are derived which prove useful in interpreting the regression results in this chapter. These properties appear to have gone largely unnoticed by the human capitalists. 1. Income begins to be earned only upon the completion of schooling at age A = S + 5, and the starting log-income level is I3o + f3IS. 2. The maximum of the age-log-income (or age-income) profile occurs when the slope a log y/lA = (I/y) (ay/lA) = 0, that is, when f2 - 2P3 (A - S - 5) = 0. Hence the age at which the age-log-income (or age-income) profile peaks, for schooling level S, is A = S + 5 + (/32/2/3). It follows, therefore, that the profile corresponding to a higher S peaks at a later age A. In other words, the age-income curve turns down later in life for those with more schooling. 3. The log-income corresponding to the peak of the S profile is obtained by substituting T = (A - S - 5) = (32/2/33) into the earnings function. This gives logy = pf + ±IS + (/2/33) - /3(f2/4/3) - /3o+ThS 2/2#4I3).-# / = fio + pIS + (#2/43 This is also evident from rewriting the earnings function as log y = [#0 + p,S + (l2/4/3)] -3[T-( 2/2/3)]2. For a given S, the second term of this expression is always nonpositive, so the maximum of log y must occur when it is zero (that is, when T = /2/2/33). The maximum value is then given by the first term, as in the previous expression. Hence log-income, and therefore income, peaks at a higher level for the profile corresponding to a higher S. 4. For any S, the age-log-income profile is concave because 02 logy/lA2 =-2/33 < 0 In fact its shape is an inverted parabola since log y is a quadratic function of A. Conclusions (1) to (4) can be illustrated diagrammatically. Figure 7-1 depicts the age-log-income profiles for three different levels of schooling. Some further properties of the earnings function are given below. 5. The experience (T)log-income profile is also concave and parabolic in shape. 266 INEQUALITY AND POVERTY IN MALAYSIA Figure 7-1. Age-Log-income Profiles for Three Levels of Schooling log y [Po + I13S + ( /43)]----- [S+51 [s+5+(,2/2,3)] A 6. The flatness of the age-log-income (or age-income) profile is indicated by the magnitude of the coefficient /33 in the quadratic earnings function. A small coefficient (in absolute terms) indicates a flat profile, whereas a large coefficient indicates a peaked profile.48 Thus, a glance at the estimated coefficients of T2 in the various subsamples considered will establish the relative peakedness of their earnings profiles. (For example, the age-income profiles of males can be compared with those of females.) 7. In contrast to the age-log-income profile (which is parabolic), the age- income profile is not everywhere concave. By differentiation, 2 logy/lA2 = -(llogy/aA)2 + [(l/y) (a2y/aA2)]. Hence (I/y) (a2y/aA2) = - 233 + (32 - 233T)2 and thus a2y/aA2 < 0 only in a neighborhood around the peak at age A = S +5+(#2/2#3) given by [f32-233(A-S-5)]2 < 2/3.49 Outside this neighborhood, that is, for values of A more than (2133)- 1/2 away from the peak at S + 5 + (/32/2/33), the age-income profile is convex. 48. This is clear from the expression for the earnings function in (3) above, in which the square has been completed on T. 49. Note that at the peak itself, (I/y)(012v/aA2) =-2#3 < 0; that is, the age-income profile is locally concave at the peak. EARNINGS FUNCTIONS FOR URBAN EMPLOYEES 267 Figure 7-2. Age-Income Profile for a Given Level of Schooling exp lo,, + 3,S + (2 /4p3)] …-------------- [S + 5 + (2/2p,) A 8. The latter implications also follow by rewriting the earnings function as: y, = exp { [f0 + #IS + (Pi3/4#3)]- _ 3[T-(fi2/2#3)] 2} = exp { [fo+ + +3S ± (2 /4f3)]-3[A-[S + 5 + (/2/2/3)] ]2 } This is the expression for a normal (or Gaussian) distribution in T, or A, given S.50 Therefore, the experience-income and age-income profiles are bell-shaped, like the curves of a normal distribution. Properties (6), (7), and (8) of the earnings function are illustrated in figure 7-2, which shows the age-income profile for a given level of schooling. 50. The modal value for the Tdistribution is f,2/2fl3, while that for the A distribution is S + 5 + (fl,/2fl3). Table 7-6. Earnings Functions for Urban Female Employees by Occupation and Race and for Total of Male and Female Employees in All Occupations by Race Occupation 1. Professional and technical logyM = 6.79243+0.06841S+0.04536T-000103T2 R2 = 0.237 F - 9.73 log y = 7.946 S 12.245 r= 13.051 y= 3,296 (0.01428) (0.01867) (0.00048) SEE = 0567 DF - 94 var(log y) = 0 408 var(S) =24.553 var(T) = 99.724 var(y) = 2.835 10' log yc = 6.52519 + 0.09312S + 0.04671T-0.00067T2 R2 = 0.341 F - 39.10 log y = 8.225 S'- 14.143 T= 11.316 y = 4,319 (0.00895 (0.00827) (0.00019) SEE = 0.462 DF = 227 var(log y) =0.320 var(S) 16.136 var() = 89.717 var(y) - 5.524 106 logy, = 5.86618+0.12938S+0.06874T-0.00120T2 R2 = 0.527 F - 2079 logy =8211 3=14025 T= 11.925 y = 4,918 (0.02286) (0.01819) (0.00035) SEE = 0.544 DF e 56 var(log y) = 0.594 var(S) -13.478 var(T) = 99.914 var(y) = 21.160 106 log yo = 6.99854 + 0.071435 +0.02866T- 0.00218T2 R' = 0.705 F - 5.57 log y = 8.390 S e 16.091 7== 10.727 j = 6,303 (0.06546) (0.04181) (0.00120) SEE 0.536 DF - 7 varQog y) = 0.682 var(S)= 7.291 var(T)= 151 018 varly) = 50.709 106 log yTOT = 643237 + 0.09597S+ 0.05217T-0.00088T2 R2 - 0351 F- 71.51 log y = 8.159 S = 13.714 r= 11.816 9 = 4,213 (0.00706) (0.00680) (0.00016) SEE = 0.514 DF - 396 var(log y) = 0.404 var(S) = 18.253 var(T) = 95.073 var(y) = 8.681 106 Occupation 2: Administrative and manageriala log YTOT= 5.62930+0.18664S+0.01965T+0.00004T R' = 0.899 F - 8.888 logy = 7.676 S= 9.500 T= 13.357 9=3,706 (0.03844) (0.06899) (0.00162) SEE = 0.478 DF - 3 var(log y) = 1.130 var(S) = 35.083 var(7) = 157.560 var(y) = 18.919 106 Occupation 3. Clerical and related logym =5.40377+0.11547S+0.16802T-0.00367T2 R = 0288 F= 10.63 logy=7.648 S= 12.295 T= 6572 9=2,661 (0.03244) (0.03290) (0.00110) SEE = 0.689 DF - 79 var(log y) = 0.641 var(S) = 9 439 var(T) = 33 562 var(y) = 2.666 10' log ic = 5.22096+0.13773S+0.1261V-0.00202T2 R' = 0.431 F e 53.28 logy -7 661 3 = 12.130 7= 8.163 )9 = 2,701 (0.01391) (0.01418) (0.00042) SEE = 0561 DF = 211 var(logy) = 0.546 var(S) - 12.275 var(7) = 62.915 var(y) = 3.979 l0' logy, = 5.30169+0.12446S+0.13428T-00020472 R -0.443 F= 10.09 logy= 7.817 5=. 11.905 T = 10.048 9=3,185 (0.03570) (0.03853) (0.00119) SEE = 0.645 DF = 38 var(log y) = 0.693 var(S)= 9344 var(T) = 54.839 var(y) = 4 116 106 logyo = 1.96313+0.38153S+0871387-0.02662T' R5=0537 F= 1 l60 logy=7.160 S= 10.571 T= 12714 9 =1.978 (0.24565) (0.49161) (0.01556) SEE = 1.147 DF = 3 var(log y) = 1.421 var(S) = 9619 var(T) = 73 905 var(y) = 2,067 106 logYTOT= 5.26085+0-13250S+O133607-0002327T R = 0.364 F = 6549 logy = 7.666 5= 12.111 T= 8102 y = 2,735 (0.01249) (0.01269) (0.00038) SEE = 0.623 DF = 343 var(log y)= 0605 var(S) - 11.165 var(T)= 56 100 var(y) = 3655 106 Occupation 4: Sales' logyM =5.49770+014356S+0030717-0.000447" R = 0.764 F= 8.635 logy- 6.734 S= 6.958 T= 17.625 y=915 (0.03194) (0.02562) (0.00048) SEE = 0.273 DF = R var(logy) = 0.230 var(S) 13.021 var(S) = 417.142 var(y) = 0110 106 log yc -5.75586 + 0-07619S + 0.06944T- 0.00121T' R = 0.160 F = 6.79 logy = 6.840 S = 7.450 7= 11 748 9 = 1,104 (001950) (0.01748) (0.00034) SEE = 0.548 DF = 107 var(logy) = 0.347 var(S) = !4.454 .ai(TS)= li 431 var(y) = 0496 106 logyTOT= 5.79335+0.07767S +0.06469T-0.001 14T R2 - 0.7177 F= 8.75 logy=6.827 S= 7313 T= 12.313 y= 1,081 (0.01742) (0.01535) (0.00029) SEE = 0.527 DF = 122 var(logy) = 0330 var(S) = 14.475 var(T) = 138 931 var(y) = 0.451 106 Occupation 5: Service ' -gyi = R5.31354+0.05898S+0.08771T-O.0147T2 R2 =0.139 F= 9.07 Fogy= 6.502 S= 3.956 T 20.392 y=908 (0.02084) (0.01873) (0.00038) SEE = 0.761 DF = 168 var(iug y) = 0.661 var(s =10.318 var(TI) 152.978 var(y) = 0517 10' logyc =6.41334+0.02881S+0.02207T- O00021T2 R2 =0041 F= 7.49 bogy=6.853 S= 2.744 T= 25.656 y=1,150 (0.01326) (0.00690) (0.00010) SEE = 0.652 DF = 520 var(log y) =0.441 var(S) = 8.693 var(T) 324.460 var(y) = 0.530 10' logy, =5.41470 + 0.12619S + O.05089T- 0.00063T2 R2 = 0.169 F = 609 logy =6.465 S= 2.920 T= 21.771 = 863 (0.03623) (0.02022) (0.00038) SEE = 0.745 DF = 90 var(log y) = 0.646 var(S) 5.727 var(T) - 198.907 var(y) = 0.398 106 log y18 i = 6.1446 n-0.n4448S +0.02786T-0.00025T2 R2 = 0.059 F = 16.61 logy = 6.734 S= 3.059 T = 24.023 y = 1,069 (0.01069) (0.00601) (0.00009) SEE = 0.716 DF = 79t) var(iogy) = 0.544 v-a(S' 9.073 ar(T) = 276,397 var(y) = 0Q536 106 Occupation 7. Farm laborers iogym =5.15026+0.00781S+0.035647-0.00053TI R2 = 0.074 F= 0.56 bgy=5.640 S= 2.520 T = 29.520 y=307 (0.02335) (0.02902) (0.00044) SEE = 0.391 DF = 21 var(logy) = 0,144 var(S) = 13.260 var(T) = 173.510 var(y) = 0.027 106 log yc = 6.09509 + 0-04606S + 0.04083T-0.00044T' R2 = 0.147 F = 10.06 log y = 6.833 S = 2.246 T= 23.581 j = 1,030 (0.01754) (0.00987) (0.00015) SEE = 0.464 DF= 175 var(logy) =0.248 var(S) = 8.015 var(T) 197.326 var(y) = 0.190 106 logy, = 6.84629+0.09683S +0.00018T- 000005TO R2 = 0246 F = 0.98 logy = 7,049 3 = 1.615 T= 32.615 j = 1,223 (0.06195) (0.14970) (0.00232) SEE = 0.372 DF = 9 var(logy) = 0.138 var(S) = 3.923 var(T) = 51.090 var(y) = 0.182 106 log YTOT =6.37131+O.00566S+0.0212T-'0.0002572 R2 =0.031 F= 2.27 logy =6.709 S= 2.240 T= 24.806 i=959 (0.01859) (0.01135) (0.00018) SEE =0,611 DF = 213 var(logy) =0.380 var(S)= 8329 var(T)= 192.215 var(y)=0.228 106 Occupation 8: Production workers' logyM = 5.44079+0.09078S+0.04434T-0.00073T2 R'=0.185 F= 5.23 iogy'-6.269 S= 4.740 T= 19.192 j=700 (0.03518) (0.02168) (0.00031) SEE = 0.692 DF= 69 var(log y) = 0564 var(S) =14.404 var(T) = 252.310 var(y) = 0.327 106 logyc = 5.68518+0-05878S+0.06278T-0.00077T2 R2 =0.138 F= 23.62 logy =6.602 S= 4412 T= 15063 yj=902 (0.01206) (0.00831) (0.00013) SEE = 0.591 DF = 443 var(log y) = 0.402 var(S) = 10.506 var(T) = 144.804 var(y) = 0.553 106 log y, = 4.73171 + 0.15598S + 0.08263T- 0.00065T2 R2 = 0.488 F - 1.59 log y 6.774 3 = 3.944 T = 23.278 i = 1,020 (0.08311) (0.11877) (0.00221) SEE = 0.561 DF = 5 var(logy)=0.384 var(S) = 22.278 var(T)= 255.694 var(y)=0.343 106 logYTOT =5.69058+0.05973S+0.05973T- 000082T3 R2 = 0.101 F = 1955 logy =6.559 Y= 4.449 T= 15.772 y =876 (0.01163) (0.00794) (0.00013) SEE = 0.628 DF = 525 var(logy) = 0.436 var(S) = 11.193 var(T) = 163.599 var(y) = 0.522 106 Table 7-6 (continued). All occvpations together og!yM = 5.20134+0.14738S+0.07082T-000115T2 R2= 0.421 F = 111.46 logy = 6.935 S= 7.329 T= 16.593 y = 1,663 (0.00895) (0.00962) (0.00018) SEE = 0.795 DF = 459 var(log y) = 1.085 var(S) = 29.999 var(T) = 177.996 var(y) = 2.532 106 logyc = 5.46229+0.13305S+0.06785T-0.00075T2 R2 = 0.437 F = 442.23 logy = 7.073 = 6.167 T= 17.602 y = 1,697 (0.00366) (000372) (0.00006) SEE = 0.631 DF= 1,709 var(logy) = 0706 var(S) = 30.110 var(T) = 229.071 var(y) = 3.007 106 logyl = 5.12670+0.16787S+0.07747T-0.00106T2 R2 = 0.628 F = 122.17 logy =7.261 S'= 7.695 T= 17.400 y = 2,482 (000924) (0.01212) (0.00024) SEE = 0.670 DF = 217 var(logy) = 1.191 var(S) = 36.107 var(T) = 176026 var(y) = 10.014 106 logyo =5.73004+0.14582S+0.00674T-0.00103T2 R2 = 0.476 F = 5.76 logy = 7.767 g = 12.478 T= 13.217 y = 4,012 (0.04283) (0.05851) (0.00159) SEE = 0850 DF = 19 var(log y) = 1.191 var(S) = 23.715 var(T) = 127 632 var(y) = 29.029 106 10B YTOT = 5.43235 + 0.13772S+ 0.06369T-0.00073T2 R2 = 0430 F = 608 28 log y = 7.070 S = 6.589 T= 17 349 y = 1,784 NJ (0.00326) (0.00341) (0.00006) SEE = 0.691 DF= 2,416 var(log y) = 0.837 var(S) = 31.212 var(T) = 213.603 var(y) = 3.886 106 All occupations together, male and female logyM =5.13026+0.15751S+0.10063T-0.00138T2 R2 =0.442 F = 509.57 logy= 7.546 = 7.756 T= 19.059 y= 2,864 (0.00426) (0.00430) (0.00008) SEE = 0.708 DF = 1,932 var(log y) = 0.897 var(S) = 20.186 var(T) = 161.338 var(y) = 15.591 106 logyc = 5.25608 +0.14504S+0 101837-0.00126T2 R2 = 0.501 F = 1.679 81 logy = 7.425 9= 6619 T= 19.030 y = 2,542 (0.00226) (0.00208) (0.00004) SEE = 0 639 DF = 5,025 var(log y) = 0.818 var(S) = 22.739 var(T) = 214.539 var(y) = 9 612 106 logy1 = 5.51637 +0.14685S+0.07240T-0.00084T2 R2 = 0.510 F = 412.52 logy = 7.622 S= 7.501 T= 22.241 y = 3,138 (0.00436) (0.00422) (0.00007) SEE = 0.639 DF = 1,191 var(log y) = 0.829 var(S) = 24.926 var(T) = 223.923 var(y) = 16173 106 logyo = 4,82882+0.21605S+0.9l00T -0.001137T2 R2 = 0.465 F = 28.70 logy = 8.406 S = 11.908 T= 16015 y = 10,064 (0 02391) (0.02913) (0.00069) SEE = 1.022 DF = 99 var(logy) = 1.894 var(S) = 21.107 var(T) = 146.076 var(y) = 222.999 106 log YTOT = 5.29352 + 0.14772S + 0.09437T-0.00117T2 R2 = 0.485 F = 2,592.85 log y = 7.494 S = 7.079 T = 19 463 y = 2,798 (0.00181) (0.00171) (000003) SEE = 0.669 DF =8.259 var(logy) = 0868 var(S) = 22.981 var(T) = 203.929 var(y) = 15307 106 a. The lack of observations on particular racial groups in this occupation prevented the estimation of regression equations for them. See also the note to table 7-1. 8 Conclusions and Some Notes on Policy THIS STUDY HAS EXAMINED PATTERNS of income inequality in Malaysia from the data generated by the 1970 Post-Enumeration Survey. It has described a large number of results on income distribution and hinted at the broad policy implications that follow from such a diagnostic analysis. Income data from the Post-Enumeration Survey have not yet been analyzed or even tabulated, and a major purpose of this study has been to document the state of income inequality in the country. Naturally, I have attempted to go beyond this to suggest the sources of income inequality and of poverty. In fact, I have trie.i to build up an anatomy of income distribution for Malaysia. In an area where the quality of the data is notoriously poor, it is hoped that this study will contribute a benchmark for comparative studies elsewhere, as better income data begin to be collected and analyzed systematically. The chapter begins with a brief reappraisal of PES data and a summary of the imain findings on income distribution in the country. There is a review of the methodology, the major results on inequality decomposition, and the main findings on poverty. The rest of the chapter is concerned with a more general policy analysis of Prongs I and 2 of the NEP. Policies to reduce poverty in Malaysia are surveyed, and four broad types are identified that seern especially relevant: direct income transfers, fiscal policies, interven- tioin in commodity markets, and rural development. These policies are discussed and evaluated in the light of knowlcdge about the poor derived from the detailed poverty maps in chapters 4 and 5. There is next a brief review of the employment restructuring targets to achieve racial economic balance, and some implications and costs of implementing them are noted. The chapter concludes with a cursory discussion of the underlying aims and mutual consistency of Prongs I and 2 of the NEP. 271 272 INEQUALITY AND POVERTY IN MALAYSIA Summary and Conclusions The most frequently used variable in this study has been race. Chapter 1, which contains a general introduction to and perspective on Malaysia, traces the development and importance of ethnic pluralism in the country. This pluralism has led the government to show special concern for racial income distribution and, more generally, for racial economic disparities. Thus the New Economic Policy of 1971 proclaimed the major objective of "restructuring Malaysian society to correct racial economic imbalances." It is in the context of this historical and economic situation that I have disaggregated by race in most of the analyses in this study. The two prongs of the NEP have been examined in chapter I in relation to income distribution among individuals. The study of individual income distribution, which forms the central framework of this book, has permitted a detailed analysis of poverty and racial income distribution. The two prongs of the NEP have been characterized by their separate effects on the individual income distribution (figure 1-1). Prong I implies drawing a poverty line and moving all the poor above it, irrespective of race. Prong 2 corresponds to a proportionate rise in all Malay incomes which keeps within-race inequality constant but eliminates between-race inequality completely. The income distribution data used in this study were collected through the Post-Enumeration Survey of 1970. Since no official report on the survey has been prepared, chapter 2 presents a detailed description and evaluation of the PES, including its survey design, sampling procedure, and income definition. A comparison with two previous surveys conducted in Malaysia shows that the PES, while by no means perfect, is the best source of income data to date on household and individual incomes in Malaysia. The principal sources of error are likely to have been the usual ones of response associated with the collection of income data, rather than sampling errors arising from survey design and sample size (which was approximately 135,000 individuals). Although it was not primarily an income survey, an elaborate system of probing, prompting, and other checks was followed to obtain satisfactory income data. The definition used for income also appears to have been fairly comprehensive, including income received in both cash and kind. A very rough guide to the quality of PES income data is suggested by a comparison with estimates of aggregate personal income based on the national accounts. Average household income shown in the PES was approximately 75 percent of that estimated via the 1970 national accounts data. This degree of understatement is not particularly large, even CONCLUSIONS AND NOTES ON POLICY 273 by the standards of household income surveys in developed countries (such as France, Germany, and the United Kingdom). At any rate, it cannot be taken as a reason for dismissing the PES data as unsuitable for analysis. The broad features of the PES household income distribution show overall inequality in Malaysia to be fairly high (Gini coefficient of 0.5129), especially in comparison with other economies, such as Taiwan and the Republic of Korea, at similar stages of development. But the careful examination required of these different data sources to establish full comparability is outside the scope of this study, and reliable international comparisons of inequality must await detailed empirical research. More important, however, are intertemporal comparisons of inequality in Malaysia itself, since other researchers have concluded from a superficial examination of the earlier Household Budget Survey (HBS, 1957-58) that inequality in the country has sharply worsened. Because the published report on the HBS is deficient in information on income definition, sample coverage, and so on, I have attempted to reconstruct an account of this survey, along the lines of that given for PES, from unpublished records and files in the Malaysian Department of Statistics and conversations with persons responsible for conducting it. This account is documented in detail and used to denionstrate that the 1957 survey is simply not comparable with the 1970 survey, and that no conclusions can be reached from these surveys about intertemporal changes in inequality. The exercise is valuable both for researchers hoping to use the HBS, 1957-58, and as an illustration of the dangers in making inequality comparisons without first establishing comparability. A detailed examination of PES income data that uses different population units and income concepts shows overall inequality in Malaysia to be high (tables 3-7 and 3-8). The distribution of individuals by per capita household income reveals that the poorest 40 percent of the people in Malaysia get only 12.3 percent of total income while the richest 5 percent get 28.5 percent, and the ratio of income shares between the highest quintile and lowest quintile is almost 13 to 1. The Gini coefficient and Theil T index for this distribution are 0.4980 and 0.5161, respectively. The distribution of income recipients by personal income (that is, the personal income distribution) shows that the poorest 40 percent of income recipients obtain 11.3 percent of total income while the richest 5 percent obtain 28.5 percent, and the ratio of income shares between the highest quintile and the lowest quintile is almost 16 to 1. The Gini coefficient, Theil T index, Theil L measure, and variance of log-income (varlog) for this distribution are 0.5063, 0.5360, 0.4763, and 0.8967, respectively (table 6-5). Racial income disparities have traditionally been measured in terms of the ratios between the mean income of one racial group to that of another. 274 INEQUALITY AND POVERTY IN MALAYSIA Again, these vary according to the distribution considered, that is, the population unit and income concept used. For the distribution of households by household income, the racial disparity ratios are 2.29 Chinese: Malay and 1.77 Indian: Malay. But for the distribution of individuals by per capita household income, the disparity ratios are 2.00 Chinese: Malay and 1.65 Indian: Malay. These ratios are lower than those often bandied in public, which neglect racial differences in average household size (Chinese, 5.839 members; Indians, 5.453 members; Malays, 5.084 members). For the distribution of income recipients by personal income, the disparity ratios are even smaller: 1.77 Chinese: Malay and 1.53 Indian: Malay. The reason is that the Chinese and Indians have higher average participation rates (strictly, average number of income recipients divided by average household size) than the Malays (table 6-1). Hence, in terms of incomes that can be directly influenced by economic policy, the racial disparity ratios are a good deal smaller than is popularly believed. The individual income distributions show large inequalities within the racial groups. Ranked by per capita household income, Malay and Chinese individuals are distributed very similarly about their respective means (although it is sometimes suggested that Malay incomes are less dispersed). The corresponding fractile shares are close to one another, as corroborated by various relative inequality measures (including the Atkinson index when this is computable; see table 3-9). The Gini coefficient for the Malays and Chinese is 0.4553 and 0.4542, respectively, and the Theil T index is 0.4114 and 0.4228, respectively. Indian incomes are distributed somewhat more unequally, with a Gini coefficient of 0.5003 and a Theil T index of 0.5448. The incomes of "other" races are distributed extremely unequally (Gini coefficient of 0.7071 and Theil Tindex of 0.9371), which is not surprising given the heterogeneity of this group (rich Europeans and very poor Thais and other Asians). For the personal income distribution, the Gini coefficient, Theil T index, Theil L measure, and varlog, respectively, are: 0.4751, 0.4370, 0.4193, and 0.8293 for the Malays; 0.4908, 0.4958, 0.4430, and 0.8428 for the Chinese; 0.4693, 0.4998, 0.3925, and 0.6571 for the Indians; and 0.7048, 0.9442, 1.2016, and 2.5426 for the others. These large income inequalities within racial groups suggest that racial income disparities may be only part of a much wider problem of income inequality in the country. It is thus interesting to ask how much of overall inequality is explained by such racial income disparities. To do this in an unambiguous manner involves using an index which is decomposable in the strict sense defined in chapter 3: an index is strictly decomposable if the amount by which inequality reduces when mean income differences between the groups are eliminated (keeping within-group inequality constant) is the same as the amount of inequality which arises when each CONCLUSIONS AND NOTES ON POLICY 275 member of a group gets the mean income of that group and within-group inequality is eliminated. Of the inequality indices used in this study, two are thus decomposable: the TheiI L measure and varlog (the latter only around group geometric mean incomes). These two measures are computable for the personal income distri- butiori: the between-race contribution to inequality is 9.6 percent according to the Theil L measure and 7.9 percent according to varlog. In other words, the Prong 2 policies designed to achieve racial balance (keeping within-race inequality unchanged but eliminating between-race inequality altogether) will reduce overall income inequality by less than 10 percent. This result contradicts the claim that racial disparities explain much of the income inequality in Malaysia: more than 90 percent of it arises from the large income disparities within each racial group. Thus it is somewhat misleading to quote racial disparity ratios in the context of the redress of individual inconie inequality, although this is often done in public debate. Taking the redress of individual income inequality as the objective of policy, I have derived the redress of poverty rule (filling in the poverty gap from the bottom up) as the most "efficient" rule for distributing incremental inco.me (chapter 3 and appendix E). This rule yields a distribution which Lorenz-dominates the distribution of any other rule and hence shows less inequality for any index which is mean-independent, is population-size independent, and satisfies the principle of transfers. The best strategy to redress inequality turns out to be to redress poverty irrespective of race, which is in fact the aim of Prong I of the NEP. Chapters 4 and 5 explore the extent and nature of poverty in Malaysia, so that policy measures for its alleviation can be better informed. After consideration of both the absolute and the relative approaches to the definition of a poverty line, a poverty line for Malaysia is set at a per capita household income level of M$25 per month. Some 36.5 percent of households and 40.2 percent of individuals fall below this level of per capita household income (as the poor have larger than average size households). Other indices of poverty are also examined, which in- corporate the income gap of the poor, normalized by some measure of aggregate income available for redistribution. A recent index proposed by Sen uses rank-order weights on the income gaps of the poor. This index turns out to be the same as computing the difference between the poverty line and the equally distributed equivalent income of the poor with the use of a rank-order welfare function. The aggregate poverty gap in Malaysia stands at 7.3 percent of total personal income, and a transfer to the poor of 8.3 percent of the income of the nonpoor (or 12.7 percent of their income in excess of the poverty line) is required to close the gap. Estimates of other measures are also presented (table 4-2), but the simple incidence-of- poverty measure is adopted for the decomposition of poverty. 276 INEQUALITY AND POVERTY IN MALAYSIA A profile of poverty in Malaysia is constructed, which identifies the poor in terms of socioeconomic variables such as race, location, employment status, occupation, and education (table 4-3). Such information is useful not only in understanding better the correlates and circumstances of poverty, but also in identifying areas of government intervention for the redress of poverty. One-way classifications reveal that 87.7 percent of poor households are in rural areas; 78.1 percent are Malay; 41.5 percent are concentrated in the four northern states of Kedah, Kelantan, Perlis, and Trengganu; 93.5 percent are headed by employees or own-account workers; 77.4 percent are headed by farmers or farm laborers; 75.1 percent are headed by workers in agriculture or the agricultural products sector; and 97.2 percent of poor household heads have not gone beyond primary school. Groups which suffer particularly high rates of poverty (above the average of 36.5 percent) are households in rural areas (44.6 percent); Malays (51.4 percent) and "other" races (40.3 percent); households in Kedah (48.6 percent), Kelantan (65.2 percent), Perlis (58.9 percent), and Trengganu (54.6 percent); households whose heads are own-account workers (50.1 percent) and unemployed (38.0 percent); households whose heads are farmers (61.9 percent) and farm laborers (48.6 percent); households whose heads are in agriculture (61.5 percent) and agricultural products (46.2 percent); households whose heads have no education (49.0 percent) or only some primary education (39.1 percent); and households headed by females (44.9 percent), households with more than five members (40.0 percent or more), households with three or more children under the age of 15 (41.5 percent or more), and households with fewer than two income recipients (41.9 percent or more); see table 4-3. When several of the characteristics associated with high rates of poverty are taken together, the chances of being poor can become extremely high. To design minimum leakage policies and projects to redress poverty, it is necessary to zero in on high-risk subgroups which are homogeneous but nevertheless account for a significant fraction of overall poverty. This has been done in chapter 5 by increasing the selected level of detail of relevant variables and cross-classifying them to obtain a multidimensional profile. From the two-digit matrix thus generated, five subgroups of rural poor are isolated, which account for 79 percent of total rural poverty (table 5-1). These subgroups are then further disaggregated regionally to identify special pockets of poverty (table 5-2). The five subgroups (with their poverty incidences in parentheses) are households headed by: paddy farmers (65.8 percent); rubber smallholders (55.6 percent); laborers on paddy and mixed-agriculture farms (63.9 percent); workers on rubber estates and smallholdings (47.5 percent); and fishermen (50.9 percent). The economic problems of these homogeneous subgroups, and measures to raise their CONCLUSIONS AND NOITES ON POLICY 277 productivity and incomes, are examined individually (double-cropping of paddy, land settlement and consolidation, revision of the export duty on rubber, rubber replanting with high-yielding stock, and other programs to assist traditional srnallholders). The regional disaggregation also allows identification of special (regional) components of rural development policies which, as argued later in this chapter, are a major means for raising the income levels of the rural poor in Malaysia. Whereas such variables as race, region, location, employment status, and industrial sector are fairly good guides for identifying poverty in Malaysia, they turn out to be relatively poor at explaining overall income inequality. Chapter 6 decomposes personal income inequality in terms of these variables, in an attempt to explore the sources of inequality. The purpose here is to describe the contribution of particular variables to inequality, but there is no attempt at an explicit model or theory of income determination. For example, there is no explanation of why the Chinese mean income is 1.77 times the Malay mean income, but merely a description of the effect of this on overall inequality. When it is found that the between-race contribution to personal income inequality is less than 10 percent. variables such as assets, educational level, and occupation are not being held constant in the decomposition.' The decomposition analysis simply amounts to an ex post accounting of the sources of inequality which does. however, yield some important insights; it also throws light on hypotheses and perceptions of inequality which have a bearing on policymaking. One of the significant sources of inequality in developing countries generally is alleged to be the large rural-urban differences in mean income (Lipton, 1968 or 1977). But in Malaysia rural-urban differences explain only 10.0 percent of personal income inequality by the Theil L measure and 7.3 percent by varlog; most of the inequality arises within rural and wilhin urban areas. Higher stages of urbanization appear to be associated with higher levels of inequality (table 6-5). And a regional decomposition shows that between-state differences in income (which are quite large) account for only 8 percent of inequality in the country. Racial disparity ratios in the country as a whole are higher than those in any of the three locations (metropolitan towns, towns, and rural areas) because of the disproportionate presence of Chinese and Indians in urban areas and of Malays in rural areas. Thus the NEP perception of racial economic imbalances owing to the "identification 1. We cannot, therefore, infer that this gives some idea of the inequality attributable to racial discriminationi, which, if anything, is likely to be overestimated because other variables are not held constant. 278 INEQUALITY AND POVERTY IN MALAYSIA of race with geographical location" seems largely valid; however, its removal is unlikely to reduce personal income inequality very much. An obvious determinant of personal income inequality is inequahty in the distribution of assets, both physical and human. Although there are no data on the distribution of physical wealth among individuals in Malaysia, the breakdown of the personal income distribution by employment status does shed some light on the distribution of capital assets. Employers' incomes are distributed more unequally than own-account workers' incomes, which are in turn distributed more unequally than employees' incomes. There are also very large disparities between the mean income of employers and that of other groups (which are fairly close to the overall mean), but the contribution of employment status to income inequality is relatively small (9.0 percent by the Theil L measure and 6.5 percent by varlog) because, in fact, there are few employers with high incomes.2 The racial disaggregation for employers and own-account workers, employ- ment categories in which capital or property income is likely to be significant. reveals large disparities between the races (especially among own-account workers, for whom the between-race contribution to in- equality is more than 20 percent by both measures). This undoubtedly reflects wide disparities between the racial groups in asset ownership. Employees form the most homogeneous category, with racial income disparity ratios of only 1.25 Chinese:Malay and 1.17 Indian:-Malay. (These differentials could arise from between-group differences in skill level and, possibly, leisure preference.) The decomposition by occupational category also suggests that average wealth among the Chinese is greater than that among the Malays. For 2. By the way it is defined, the between-group component of inequality depends only on the population share and mean income of each group (see also the formulas in chapter 3 or appendix C). Other things being equal, the higher the mean income of the richest group, the greater the between-group component (this follows by direct manipulation of the formula or by applying corollary I in appendix E); further, the lower the population share of the richest group, the smaller the between-group component provided its population share is not "too large" initially. For the Thetl L measure, for example. if there are only two groups, with Ju > pu, we have i13(n1/n) -~~~~~log (PI/P2). (n1 /n2) + Hence a'LB>0 provided the initial population share (nin) is less than b(n, In) lop9 ,,, /112 IP 1 / P2I CONCLUSIONS AND NOTES ON POLICY 279 instance, the racial disparity ratios and between-race contribution to income inequality among farmers is largest of the eight one-digit occu- pational categories (19.2 percent by Theil L and 12.4 percent by varlog). Table 6-8 also shows that the major racial groups are not proportionately distributed across the occupations, with the Chinese usually overrepre- sented in the better-paid occupations and the Malays overrepresented in the less well-paid ones. Mean incomes within occupations are generally highest for Chinese and lowest for Malays, with the Indians occupying a middle position. There are some significant exceptions to this ranking at the two- digit level of disaggregation (table 6-11) where, for example, Malay government administrators and legislative officials receive more than Chinese, possibly reflecting the preferential treatment in hiring and promotion given lo Malays in the public sector. The between-occupation contribution to overall income inequality in Malaysia is greater than that of any other single variable examined: 31.7 percent by the Theil L measure and 23.8 percent by varlog. Thus the personal income distribution is more closely related to major categories of occupation than to any other variable including industrial sector, whose contrilbution is 19.0 percent and 15.6 percent, respectively (by the two measures).3 Furthermore, multivariate decompositions by the variables so far considered do not add very much to the single-variable decomposition by occupation. For instance, the two-way decomposition by eight occu- pational categories and nine industrial sectors yields a between-group contribution of 33.5 percent by the Theil L measure and 26.9 percent by varlog. These results suggest that the standard categories of sector, occupation, and employment status may be too broad for the purposes of simulating personal income distribution. Other combinations of variables which include personal characteristics such as age and education perform significantly better (see table 6-10), and if personal wealth data were available the performance would improve even more. Given these decomposition results, I have attempted to explain income inequality in terrns of age and education in chapter 7. This has been done by means of earnings function regressions for urban employees, whose income 3. Given the large intrasectoral variation in incomes (80 to 85 percent of overall inequality), the government's policy of specifying Malay employment quotas within broadly defined industnal sectors may not be very effective in rectifying even the racial disparity ratios. The large intrasectoral income differences arise from the dualism which prevails wirhin sectors (such as formal-informal subsector variations in productivity and other forms of regional- occupational dualism). However, if employment quotas were imposed within subsectors such as rnodern manufacturing, they would probably be even more disruptive than sectoral quotas because the Malay base is likely to be smaller there. 280 INEQUALITY AND POVERTY IN MALAYSIA may be assumed to accrue mostly as wages or labor earnings. Although the particular estimating form used for the earnings function has some basis in the so-called human capital model, the purpose is not to test this model but to use it to describe age-income profiles at different levels of education. The estimated equations are a convenient way of summarizing labor market information in terms of earnings differentials for individuals at different age-education levels, and the regressions automatically generate a measure of income inequality used in this study, the variance of log- income, or varlog. The life-cycle factors of experience and education explain (in the R2 sense) almost 50 percent of the income inequality arnong urban employees as measured by varlog. This level of explanation is quite impressive in light of the omission of many important variables for individuals (such as ability, time worked, and proportion of unearned income in total income) on which there are no data. The private rate of return to education (coefficient on the schooling term in the log-income regression) is 14.01 percent for all urban male employees, and there are no significant differences in this between the races. Differences in urban male employee incomes between the races stand at 1.10:1.09:1.00 for Indian:Malay:Chinese, and these are matched by differences in their average levels of education and experience. Interpersonal inequality in urban male employee incomes is also highly correlated with inequality in education and experience, but policy implications about the role of education in reducing income inequality cannot be deduced without further stringent assumptions, for example, about the demand side of the labor market and the validity of the cross-sectional estimates for time series inferences. Returns to education are higher in modern sector (tertiary) occupations and for those educated in English, possibly because of screening. The return to schooling for all urban female employees is 13.77 percent, and females generally obtain a lower return to education within each occupation than males, and their age-log income profiles are flatter. A disaggregation by age cohort shows significantly higher returns for the younger age groups, which perhaps reflects an obsolescence effect in education. Policies to Reduce Poverty in Malaysia The first prong of the New Economic Policy seeks to eradicate poverty in Malaysia. In the light of the detailed information about poverty generated from the PES (see chapters 4 and 5), it is possible to speculate on policies that could be used to influence the welfare levels of the poor. These vary from direct income maintenance to indirect taxation and fiscal policies, includ- CONCLUSIONS AND NOTES ON POLICY 281 ing state provision of public goods and services. A survey of the typical areas of intervention and instruments of policy may be found in Ahluwalia (1974b). In this section four broad types of policy are briefly reviewed which seem particularly relevant in the Malaysian context: direct income transfers, fiscal policies, intervention in commodity markets, and rural developrnent policies. Direct Income Transfers Implicit in the poverty indices estimated in chapter 4 is the idea of closing the poverty gap by means of direct transfers from the nonpoor to the poor. The poverty gap of 7.3 percent of total personal income in Malaysia could be eliminated by a transfer of 8.3 percent of the income of the nonpoor.4 Such a proportional income transfer from all the nonpoor would place a heavy burden on those just above the poverty line (and those not very much above it). It is perhaps more meaningful to consider transfer schemes whereby the entire burden is borne by the rich (top 5 percent or some other appropriately chosen upper-income class). As a fraction of the income of the top 5, percent, the poverty gap is 25.6 percent; of the top 10 percent, it is 18.7 percent; of the top 20 percent, it is 13.3 percent; and of the top 40 percent, it is 9.8 percent. These figures give some idea of the scale of the problem if direct income transfers were used to eliminate poverty.5 The Ministry of Welfare Services in Malaysia is currently considering a public assistance program "to provide financial assistance to those with little or no income So that they can enjoy a minimum level of living" (Departrment of Social Welfare, 1976, p. 1). The idea of social security is not new in Mvlalaysia, and in the past the social welfare departments of states have allocated small amounts for general welfare (including natural disaster relief, school aid, and relief for the needy, the destitute, and the elderly).6 The government also seems to recognize the responsibility of the state to assure a minimum income level for all its citizens: "The national interest requires that all people have sufficient income to maintain a living conducive to health and well-being. Today this philosophy has added significance when it is borne in mind that the twin objectives of the S.M.P. are the eradication of poverty and the restructuring of society" (Department of Social Welfare, 1976, p. 1). 4. If the extreme poverty line of M$15 per month were adopted, it would require a transfer of only 1.3 percent of the income of the nonpoor to eradicate poverty. 5. These figures do not indicate the total cost of implementing such transfers. 6. The allocation for welfare services in the Second Malaysia Plan (SMP), 1971-75, was M$13.47 million, or 0.2 percent of the total plan allocation. Of this, M$4.36 million was actually spent between 1971 and 1973 (MTR, p. 101). Table 8-1. Taxes Paid by Households below the Poverty Line. Taxes (MS thousand) paid by Taxes paid by Perceniage of all house- Perceniage All poor Rural Urban lax borne by holds of total house- poor poor poor house- Tax (MS thousand) federal taxes holds households households holds Export taxes 250,320 13.9 4,876 4,876 . .. 1.9 Rubber 76,190 4.2 4,876 4,876 ... 6.4 Tin 129,712 7.2 ... ... ... ... Other 44,418 2.5 ... ... ... ... Import duties, excises, licensesa 884,662 49.3 75,463 66,310 9,153 8.5 Foodb 51,316 2.9 7,605 6,870 735 14.8 Beverages and tobacco 213,626 11.9 27,985 25,360 2,625 13.1 Textiles 33,139 1.8 2,293 2,091 202 6.9 Rent, fuel, and power 74,209 4.1 8,334 7,217 1,117 11.2 Transport 310,658 17.4 14,072 12,382 1,690 4.5 Other 201,714 11.2 15,174 12,390 2,784 7.5 Inland Revenue 661,889 36.8 ... ... ... ... Income taxc 560,997 31.2 ... ... ... ... Other 100,892 5.6 ... ... ... ... Totalfederal taxes 1,796,871 100.0 80,339 71,186 9,153 4.5 Land taxesd n.a. - n.a. 3,247 n.a. n.a. Zakat n.a. - 556 556 ... n.a. Total plus land taxes and zakat - 84,142 74,989 -Not applicable. .. . Zero or negligible. n.a. Not available. a. Based on Alternative I in Andic (1975), which allocated import duties, excise taxes, and licenses according to the expenditure pattern of poor households in relation to that of all households in each one-digit category of goods and services of the HES sample. b. Sugar represents 94 percent (import duty M$29,803 thousand, excise tax M$ 18,372 thousand of the total tax of M $51,316 thousand collected on food (Andic, 1975, table 12). c. Corporate taxes represent nearly 75 percent of income tax collections. d. Does not include land taxes paid by coconut smallholders. There is also insufficient information to estimate any land taxes paid by or shifted to urban poor households. Sources: Andic (1975), tables 12, 14, 16; EPU (1975), table 9; and Government of Malaysia (1974). CONCLUSIONS AND NOTES ON POLICY 283 The question therefore arises whether full income maintenance is a feasible policy in Malaysia. There are two obvious constraints to consider. First, does the administrative machinery exist to implement a very large cash transfer scheme (even if the government could afford it)? Second, can the public exchequer raise the resources to finance this scheme? It is also necessary to ask whether income maintenance, even if feasible, would necessarily be optimal as a long-run policy. Given the size of the state government machinery which runs the current (inadequate) public assistance program, it is doubtful if it could handle anything approaching a full income maintenance program.' A register does not exist of the economically active, nor is it easy to devise one with the present extremely limited coverage of the income tax net (a complete register does not exist even for the unemployed). Further, since such a large proportion of heads of households (39.3 percent), and an even larger proportion of heads of poor households (55.3 percent), are self-employed (see table 4-3), verification of income to determine the magnitude of assistance to be given becomes an extremely difficult task. In addition, vested interests and power realities at the local level could prevent the funds from reaching the intended beneficiaries.8 These problems are not insur- mountable, but they do imply heavy additional expenditures for training and administration if an effective social assistance program is to be mounted. The financial commitment by the government, solely for transfers to fill the income gap of the poor, works out to approximately M$401 million in 1970." This takes no account of the sums required to administer the program, nor of the additional commitment resulting from the reduction in work effort and earnings by those faced with such a negative income-tax schedule. (For instance, if all the poor were to earn zero incomes, the poverty gap would rise from M$401 million to M$1,107 million!) The federal tax revenue in 1970 has been estimated at M$1,797 million from all sources (table 8-1). The income gap of the poor therefore amounts to some 22.3 percent of total tax revenue. This figure seems very high inasmuch as it excludes admi'nistrative costs, disincentive effects on labor 7. 'The size of the government machinery may also be a constraint for other types of policies, such as rural development (see the later subsection "Rural Development Policies"). - 8. The transfer process could be largely thwarted by those who wield power or who can exert influence on those who wield power. The poor, as is well known, exert a political influence that Is woefully incommensurate with their number. 9. From table 4-2. the average poverty gap is M59.05 per month or M$108.6 per year. and approximately 3.69 million people (0.402 x 9.182 million) are in poverty in Peninsular Malaysia. 286 INEQUALITY AND POVERTY IN MALAYSIA Fiscal Policies A second way to improve the welfare of the poor is through fiscal policies that alter the distribution of the tax burden or the benefits of public spending. Ideally, tax policy recommendations should be based on a full- fledged study of optimal taxation and public expenditure. In the absence of such a study for Malaysia, I consider taxation separately from expenditure and review in particular the estimates of taxes paid by the poor. Tax policies to benefit the poor can be better established if it is lcnown which taxes are borne by them and to what extent. A special study was commissioned by the Economic Planning Unit (EPU) on tax incidence (Andic, 1975), the results of which are briefly described here. Little work, however, has so far been done on the incidence of public spending. 1 7 The methodology of the EPU tax incidence study is fairly standard and consists of allocating the total revenue from each tax between poor and nonpoor households. It assumes that personal income taxes and land taxes are borne entirely by the payer, export taxes are shifted fully backward (to labor and owners of capital),'8 and import duties and excise taxes are shifted fully forward (to consumers). Import duties, excise taxes, and licenses have been allocated on the basis of preliminary information from the 1973 Household Expenditure Survey (HES). 9 Export taxes are assumed either not to fall on the poor (as in the case of tin, timber, or palm oil) or, as in the case of rubber, to fall on smallholders in proportion to their production.20 The incidence of state land taxes and zakat (a religious tax levied on Muslim paddy growers)2" has been calculated in a rough.-and- ready manner by applying the average of state rates to an estimate of the 17. The Distnbutive Effects of Public Spending (DEPS) survey was, however, conducted by Jacob Meerman during 1974-75. But its results became available only after this study had been completed. 18. The assumption here is that of perfectly elastic world demand for Malaysian exports. 19. See Andic (1975), tables 7, 8, and 9. The preliminary HES data indicate the rough pattern of expenditure by household expenditure class according to eight broad categonies of goods and services. A poverty line of MS 150 household expenditure per month was adopted to define the poor in Andic (1975). 20. Production has been assumed proportional to acreage. All figures on acreage and production are taken from Department of Statistics (1972). 21. The rates of zakat vary from state to state: a'certain minimum production is exempted from the tax and the remainder is taxed ad valorem. In the case of another religious tax, fitrah, levied on Muslims, insufficient information prevented its allocation between the poor and the nonpoor. CONCLUSIONS AND NOTES ON POLICY 287 land cull.ivated by poor farmers.22 Taxes within the purview of the Inland Revenue, namely, personal and corporate income taxes,23 are reckoned not to affect poor households because of the high exemption limit for personal income tax (M$2,000 a year for residents) and the judgment that corporate taxes are not shifted to an extent which affects the poor. The results of the tax incidence calculations are summarized in table 8-1.24 Of total federal taxes, export taxes represent 13.9 percent, import duties 26.4 percent, excises 13.8 percent, licenses 9.1 percent, and Inland Revenue 36.8 percent (table 8-1 and Andic, 1975, table 11). Accordingly, indirect taxes (foreign trade and excise) account for almost two-thirds of tax revenue whereas income tax accounts for about one-third. And of the income tax collected, nearly 75 percent is corporate and only 25 percent personal. Table 8-1 shows that the poor bear a disproportionate burden of the import duties and excise taxes on food (94 percent of food revenue derives from sugar), beverages, and tobacco-items that make up a relatively large proportion of their budget. While their share of income is 10.2 percent (Andic, 1975, table 3A), they pay 14.8 percent of the total taxes on food, 13.1 percent of the taxes on beverages and tobacco, and 1 1.2 percent of the taxes on rent, fuel, and power. Import duties and excises as a whole, however, are moderately progressive: the poor pay -8.5 percent of the revenues collected from such taxes.25 Categories which account for large tax collections from the poor are beverages and tobacco; transport (bus and taxi services); rent, fuel, and power; and the miscellaneous category "other." Export taxes are borne only to a small extent by the poor (1.9 percent), owing to the negligible impact of tin, timber, and palm oil duties on them. But the rubber export tax does fall significantly on rubber smallholders in 22. The average annual tax rate on rubber land was estimated at M$6 per acre, on paddy land M$2 per acre, and on coconut land M$5 per acre (Andic, 1975, table 13). A paddy farmer was assumed to be in poverty if he held less than five acres, and a rubber smallholder if he held less than eight acres. On this basis, 78 percent of paddy farmers (holding an average of 3.1 acres) and 13 percent of rubber smallholders were estimated as poor in 1970. No information was available on landholding and poverty among coconut smallholders. 23 The Inland Revenue also collects the tax on tin and timber profits and development taxes. (Department of Inland Revenue, 1974). 24. These results are subject to the usual caveats, such as the partial equilibrium nature of the znalysis; see de Wulf (1975) for a detailed critique of incidence studies. 25. To determine the progressivity of the tax system, one should also estimate how prices of nontaxed items are affected; for example, protective tariffs might raise the pnce of other, domestically produced goods which are substitutes. Such general equilibrium effects should be included in comparing the burden with and without the tax system. 288 INEQUALITY AND POVERTY IN MALAYSIA poverty, who end up paying 6.4 percent of the tax revenue collected from this duty and surcharge (Government of Malaysia, 1974). As long as world demand for Malaysian rubber remains elastic, the tax will be shifted backward and act as a direct levy on producers. The estates may be able partly to shift it further backward to their employees, but smallholders cannot escape the brunt because they supply land and labor inelastically. The only other taxes borne by the poor are state land taxes and zakat, which is paid by Muslim paddy farmers. Together, these account for about 5 percent of the total taxes paid by poor households. Direct taxes in Malaysia are judged not to fall on households below the poverty line. The results of this fiscal incidence study suggest that the federal tax system is reasonably progressive: poor households pay 4.5 percent of all federal taxes, but they receive 10.2 percent of total household income. The scope for redistribution via the tax system is therefore limited, even though particular taxes might be reduced to improve the relative position of the poor (with offsetting increases elsewhere to restore revenue). By far the greatest burden of taxation on the poor stems from import duties, excises, and licenses. Two-thirds of the taxes paid by the poor under this heading are accounted for by beverages and tobacco, petroleum products, and vehicles. These items, however, also account for three-quarters of the total revenue from this source, and for as much as a third of total federal tax revenue. A reduction in tax rate for these items thus seems unlikely since the revenue loss would be substantial: its social cost would probably be judged to exceed the welfare gain to the poor (and nonpoor). Items for which a tax reduction is likely to be socially beneficial are goods that are consumed wholly or largely by the poor. A price reduction on such goods is an effective means of improving their welfare levels.26 For example, an exemption of duties on coarse cotton fabrics might be of considerable benefit, since the taxes on this commodity are likely to be regressive and to fall particularly heavily on the poor. But before specific changes in the tax system can be recommended, it is necessary to identify with greater precision the demand patterns of poor households.27 An obvious candidate for tax revision is the rubber export duty.28 Smallholders bear their share of the burden of this tax, and its alleviation 26. The instrument is somewhat blunted if demand from the nonpoor increases more thian proportionately as a result of the price reduction. 27. This should be possible through a detailed analysis of the HES 1973 data. 28. The standard rationale for an export tax is the "optimal-tariff" argument in international trade theory: if a country can exercise monopoly power in world trade. it can improve social welfare over the free-trade situation by levying an export tax (which improves its terms of trade) and redistributing the tax revenue to citizens in a nondistortionary way CONCLUSIONS AND NOTES ON POLICY 289 should significantly (and quickly) improve the welfare of a sizable subgroup among the rural poor. Since it is a product tax, however, it will not be easy to discriminate in favor of the poorer producers. The size of holding seems the natural distinguishing variable, and one approach would be to adopt a rough rebate system for smallholders with less than some given acreage, say, ten acres.29 Another approach would be to revise the basis for the rubber tax and shift it from the export to the production stage. A progressive special revenue or production tax on rubber land would enable the entire tax burden to be shifted to nonpoor smallholders and to estates.330 The state land taxes also ought to be reconsidered. At present the states levy the land tax at a flat rate, although they differentiate according to land use in agriculture. They should be persuaded to introduce progressivity into the system and to exempt owner-cultivated holdings below a certain acreage. The land tax offers substantial scope for increasing govemment revenues and is an efficient instrument that can be levied in a progressive manner. Since people cannot easily pass it on or avoid it by a change in behavior, the tax creates minimal distortions (deadweight loss). It should be exploited much more as a revenue source than it has been in the past. Although a restructured land tax would have only a small effect in relieving the present burden on the rural poor (since the present tax base and rates are small), it would boost the tax yield significantly owing to much higher average rates through progressivity. The general conclusion emerging from this investigation of the distri- bution of the tax burden is that since the poor pay a relatively small fraction of total taxes, the benefits to them from tax relief are likely to be limited.3" This does not, however, detract from the need for revision of the tax system to reduce the incidence on the poor still further and to introduce greater progressivity to maintain or increase tax revenues. Raising the posttax incomes of the poor by even a few percentage points generates a large increase in social welfare. The other aspect of fiscal policy, the distribution Malaysiacannot exert monopoly power in the world rubbermarket, however, because it faces a highly elastic demand curve for the product (since other countries supply large quantities of syrithetic rubber, which is a near-perfect substitute for natural rubber). The argument for the rubber export duty in Malaysia, therefore, is not an optimal-tariff argument, and its revision will not affect the country's terms of trade. 29. The rebate might be given, for example, in the form of subsidized inputs or higher replanting grants. The rubber replanting cess might also be reconsidered at the same time. 30. The imposition of such a tax requires a cadastral survey to bring land records up to date. 31. According to Andic (1975), removal of all taxes would increase the incomes of the poor by only about 16 percent. 290 INEQUALITY AND POVERTY IN MALAYSIA of public expenditure benefits financed by taxation, is likely to be a more powerful instrument for benefiting the poor. Unlike taxation relief, expenditure programs are not limited to the amount of revenue collected from the poor; they are also easier to focus on beneficiaries belovw the poverty line. Intervention in Commodity Markets Closely related to indirect tax policies are government policies that intervene in commodity markets through price or quantity controls. For instance, price support policies seek to establish a floor price for a commodity, with the government undertaking to buy from producers at the guaranteed minimum price. The stocks purchased are subsequently sold by the government through ("fair price") ration shops or in international markets. The government thus supports the producer price of a commodity by intervening in the domestic market as a buyer, a move which in turn supports producer incomes. In Malaysia compensatory price policies need to be considered for two commodities: paddy and rubber. As seen from table 5-2, paddy and rubber smallholders account for more tihan half the rural poor. There is already a guaranteed minimum price (GMP) for paddy in Malaysia, which was M$16 per picul in 1970.32 The government finances the GMP by exercising a quantitative control over imported rice, which is cheaper than local rice and constitutes about 10 percent of domestic consumption. Rice may be Imported only by means of a license, which is granted on condition that the importer purchase a certain fixed ratio of the higher-priced local rice from the government stockpile (see Grant, 1970). The ratio is regulated in accordance with government policy on the optimal size of the stockpile. The objectives of the price support program are not only to raise the incomes of paddy planters and assure them of a ready market, but also to "encourage the cultivation of food, as this is part of the agricultural diversification programme" (Selvadurai, 1972a, p. 8). Programs to improve yields per acre and reduce unit production costs are being implemented, but it is recognized that Malaysia will remain a relatively high-cost producer of rice for some time (Selvadurai, 1972a, p. 5). The government's policy of price support coupled with import controls raises both domestic producer and consumer prices above the international price. This policy avoids the problem of budgetary finance for the subsidy 32. See Selvadurai (1972a)* pp. 70-71. The conversion factor for piculs into pounds (avoirdupois) is: I picul = 133.3 pounds. CONCLUSIONS AND NOTES ON POLICY 291 to producers (via the GMP) since it is largely self-financing. It does imply, however, that the consumer bears the cost of the subsidy to domestic producers.33 Unfortunately, many of the poor, who are net buyers of rice, therefore end up paying for the price support program-including those in the poverty subgroups of rubber smallholders, agricultural workers, fishermnen, and urban poor. One subgroup of the poor is thus taxed to support another subgroup, the paddy farmers, and it is not at all clear that the net social benefits of this policy are positive. Without a careful quantification of the welfare gains to paddy farmers and the welfare losses to other consumers, it is difficult to comment on the desirability of the government's price support program (see Goldman, 1975). The government's price support scheme appears to have similar effects to those of an import tariff. An import tariff, however, is equivalent to a production subsidy and consumption tax at the same rate (see Corden, 1971), whereas the rice price support scheme implies a production subsidy at a higher r ate than the consumption tax.34 Both policies subsidize domestic production but both create a distortion on the consumption side.3' The welfare loss inflicted by a consumption tax is especially high since rice constitutes a large fraction of the typical poor household's budget. To minimize the burden on the nonpaddy poor it would be preferable to finance the production subsidy by some means other than a consumption tax on rice. Funding the subsidy from general government taxation is likely to create a much smaller burden (see table 8-1). In the previous section, the case for price intervention in the market for rubber. was considered to some extent. Since 93.6 percent of the benefits of a rubber. price increase are likely to accrue to nonpoor households (see table 8-1), price support for rubber is clearly an inefficient income transfer policy. There is, however, some point in the government's attempting to stabilize the price received by producers. This varies significantly with changes in demand for Malaysian rubber as a result of booms and recessions in the automnobile industry, changes in the price of synthetic rubber, and so on. A more stable price for rubber seems especially important for poorer smallholders, who are likely to find it difficult to withstand periods of abnormally low income. The government might consider a buffer-stock scheme to iron out some of the short-term fluctuations in price, accumulat- 33. T'his is not shown in the tax incidence estimates in table 8-1. 34. tinder an import tariff, the implicit consumption tax goes partly to finance the produci ion subsidy and partly to increase net government revenue; under the price support scheme. the implicit consumption tax is used entirely to finance the production subsidy. 35. T'he same effect on domestic production and farm incomes could be achieved by subsidizing input use, but this policy would distort input prices. 292 INEQUALITY AND POVERTY IN MALAYSIA ing surpluses in years when the international price is low and disposing of them in years when it is high.36 Another approach might be to negotiate longer-term commodity agreements with importing countries. Rural Development Policies Rural development policies attack poverty through various measures which help to raise the productivity of specific subgroups in the population. A rural development package might include improvements in irrigation facilities, provision of credit and input subsidies, marketing improvements, land development, and education and extension services.37 These agricul- tural supports will vary with the particular subgroup for which they are intended, and individual packages can be designed to suit each project. Two prominent features of rural development policies distinguish them from other policies to redress poverty. First, unlike fiscal and agricultural price policies, they appear easier to direct toward the poor. Many public expenditure policies can be aimed explicitly at low-income groups-for example, the selection of settlers for new land development schemes, or an irrigation scheme to permit double-cropping in a poor paddy-farming region. In contrast, population-wide policies can result in substantial leakages to the nonpoor. Even policies oriented toward specific target groups are unlikely to be completely efficient, however, and actual implementation is bound to result in leakages. For instance, middlemen and administrators may take illegal cuts in administering them ancl thus prevent the benefits from reaching only the intended subgroups. Second, rural development policies are distinguished by their effects on output growth. If it is decided for equity reasons to transfer a fixed amount of income from nonpoor to poor, is it better to do so through tax-price (or income maintenance) policies or through development expenditure policies 36. Although rubber producers acting on their own are capable of operating this policy, they cannot be expected to do so for a variety of reasons. Chief among these are an inability to borrow when the price is low (because of imperfect capital markets), lack of infornation about market trends, and incapacity to bear risk. Government intervention is socially desirable since it reduces price and therefore income uncertainty for individual smallholders. 37. A useful survey of rural development policies is provided in World Bank (1975). In a speech to the World Bank's Board of Governors, Robert McNamara (1973, p. 17) identified the following essential elements of a strategy to increase smallholder productivities: "acceleration in the rate of land and tenancy reform, better access to credit, assured availability of water, expanded extension facilities backed by intensified agricultural research, greater access to public services, new forms of rural institutions and organizations that will give as much attention to promoting the inherent potential and productivity of the poor as is generally given to protecting the power of the privileged." CONCLUSIONS AND NOTES ON POLICY 293 which raise the productive capacity of the poor'? In the former case the budgetary transfer is akin to a straight redistribution of income, whereas in the latter case the transfer should yield a permanent gain in output and income for the poor. A dynamic comparison between the two types of policy is possible only if additional assumptions are made about the productivity of poverty-oriented investment (rural development) in re- lation to investment by the nonpoor, and about the effects of the transfer on aggregate saving and investment. Suppose that the productivity of poverty-oriented investment is the same as that of investment by the nonpoor. Then rural development is dynamically superior if the transfer increases the aggregate volume of investiment. Obversely, suppose that the budgetary transfer leaves ag- gregate saving and investment unaffected. Then rural development is again a dynamically superior policy if its rate of return exceeds the rate of return on the displaced investment. Otherwise, it is better to let the higher return obtain on investment by the nonpoor and to transfer a proportion of this to the poor every year. On both counts, there are reasons to suppose that rural develcipment might be more efficient. First, the increase in productivity of the poor may itself be sufficiently high (because of double-cropping, improved fertilizers, and the like) to raise the average rate of return on aggregate investment. Second, a rural development policy should imply a higher aggregate level of saving and investment than a straight income transfer policy, since the former redirects the entire transfer to investment whereas the latter allows consumption out of the transfer.33 Taken together the two counts do suggest that rural development may be more efficient than other policies to redress poverty. Moreover, such policies, if properly designed, can be the most effective means of zeroing in on the. poor. By an appropriate matching of target groups and project packages, the leakage of benefits can be minimized. For these reasons I attempted to identify homogeneous subgroups of rural poor in chapter 5. The components of direct-benefit productivity-raising programs can be put together for such subgroups with relative ease, but actual projects may be selected only after in-depth cost-benefit studies.39 38. Inideed, unless the marginal propensity to save of the nonpoor is unity, a rural development policy should imply a higher overall level of investment than would occur in the absence of transfers. By contrast, since the nonpoor are likely to have a higher savings propensity than the poor, a straight income transfer policy should imply a lower overall level of investment than would occur without any transfers. 39. Income distribution weights should be incorporated into such cost-benefit analyses. In an appraisal of a Malaysian highway project (Anand, 1 976a) before this study was complete, it was not possible to incorporate distributional weights because of difficulties in identifying the ultimate beneficiaries of the project, but these problems do not arise for the typical rural development project. 294 INEQUALITY AND POVERTY IN MALAYSIA Some Implications of Employment Restructuring The second prong of the New Economic Policy seeks to ensure that employment in the various sectors of the economy and employment by occupational levels will reflect the racial composition of the population. The Outline Perspective Plan (opp) in the Mid-Term Review oJ the Second Malaysia Plan (MTR) specifies targets for employment by racial group which are to be achieved by 1990 (see the section "The New Economic Policy" in chapter 1).4 Various instruments in both the public and private sectors are used to achieve these targets. Direct hiring in the public sector, including criteria that favor the selection of Malays in government schemes-for example, in land settlement, timber concessions, licensing, and government tenders- helps ensure the fulfillment of racial employment targets for the public sector. In the private sector, a combination of instruments seems to be used to regulate employment and asset ownership. Permits and licenses to operate certain businesses and trades-such as bus, taxi, and trucking operations-are granted on an ethnic basis.4" The licensing of industries through such legislation as the Industrial Coordination Act provides control over racial employment in the manufacturing sector.42 Employment restructuring is also implemented through ad hoc checks, controls, and moral suasion whenever government permission or approval of any kind is needed. There is also a directive to the banks that 20 percent of all credit be given to Malays. Finally, admission quotas are applied in universities and other training institutes to increase the supply of qualified Malays.43 The sectoral employment targets for the Malay share of employment in 1990 have been set at approximately 50 percent plus or minus 3 percent (see 40. The o contains racial employment targets by sector only, but the Third Malaysia Plan (TMP) also specifies the targets by occupational category. 41. This has led to the so-called Ali-Baba phenomenon, in which a Malay lends his name to a Chinese to obtain the business permit and is then treated as a sleeping partner in the firm. Ali is a common Muslim name among Malays; Baba is another name for the Chinese. 42. The Industrial Coordination Act of 1975 is designed to ensure "the orderly development of industries and to facilitate the collection of industrial information." The act makes a license mandatory for manufacturing firms employing twenty-five or more workers and having share capital of more than M$250,000. It thus allows a close check on the racial pattern of employment in manufacturing enterprises. 43. The special provisions for Malays are set out in Articles 3, 38, 89, 152, and 153 of the constitution. CONCLUSIONS AND NOTES ON POLICY 295 MTR, table 4-5). The exception is agriculture, for which the Malay share in the terminal year is 60 percent. The 1990 racial employment targets for Malays:Chinese:lndians are 60:29:11 in agriculture, 50:40:10 in manu- facturing and construction, 50:39:11 in mining, and 48:40:12 in commerce. Sizable intersectoral labor movements will be needed to achieve these targets., with Malay employment in the modern sectors having to grow much faster than average. Particularly interesting are the implications of the targets for Malay employment in the manufacturing and construction sectors, and for Chinese employment in the agricultural and mining sectors. During 1970-90 Malay employment will have to grow at an average rate of 10.5 percent a year in manufacturing and 9.5 percent a year in construction. Both requirements will be difficult to achieve. At the same time, the growth of Malay employment in agriculture is to be kept down to a mere 0.6 percent a year, while Chinese employment in agriculture has to expand at 2.7 percent a year. The growth of Chinese employment in agricul Lure is at more than twice the sectoral rate, which implies that nearly 60 percent of all new agriculturaijobs in this period will have to be taken up by the Chinese. In the mining sector, Chinese employment will have to fall at 1.5 percent a year. The opp does not specify how this intersectoral and geographic mobility is to be brought about. For example, the targets for the agricultural sector necessitate a significant amount of reverse, or urban-rural, migration by the Chinese. But because of the substantial urban-rural differences in incomes and other amenities, such migration is unlikely to take place. Most new agricultural jobs are likely to arise on the land development schemes for rubber and palm oil, and to increase the participation of the Chinese in these schemes, land would have to be reserved for them and sufficiently attractive bonuses and subsidies offered. Such discrimination in favor of non-Malays would be difficult to defend when poverty criteria clearly favor resettlement by Malays, who account for 83.9 percent of rural households in poverty. The unequal racial allocation of new job opportunities by sector is likely to affect the new generation of non-Malays disproportionately and lead to a sense of deprivation. It has therefore been suggested that the racial employment targets might be implemented in relation to flows rather than stocks (Thillainathan, 1970, 1975a; and Moore, 1975). In other words, the racial allocation of new labor force entrants should be in proportion to the population ratios of the racial groups. Ultimately, of course, this policy would also achieve the stock target, but the time span would be stretched out over a period considerably longer than twenty years. Employment restructuring by sector alone will help to narrow income differences between the races but not to eliminate them. As shown in 296 INEQUALITY AND POVERTY IN MALAYSIA chapter 6, there are large intrasectoral differences in output and income per worker. If the occupational structure of employment within each sector is also balanced racially, the income differences between the races will be reduced more effectively. The racial employment targets by occupational level in the Third Malaysia Plan (TMP) imply that Malays are particularly underrepresented in the professional and technical and in the administrative and managerial occupations (see also chapter 6). Simple calculations show that achieving proportional racial representation within each occupation by 1990 will call for even greater redirection of manpower than the sectoral composition targets.44 The government, of course, recognizes the crucial role of expanding the supply of skilled manpower: "The current stock of qualified Malays at the managerial, professional, and technical levels requires substantial expan- sion and this will necessitate increasing the proportion of Malays pursuing courses in science, technology, economics, and business administration and other professional courses" (MTR, p. 13). Some progress has already been made in this direction by vigorously expanding vocational education, by accelerating the training of skilled workers, by constructing special secondary schools for rural areas, by placing increased emphasis on science and technology at the university level, and by expanding the MARA Institute of Technology which provides technical and vocational training to Malays. Programs are also being designed to provide Malays with basic knowledge of business management and administration. Methods other than formal training in institutes are also envisaged, particularly for the development of Malay managerial and entrepreneurial talent. Malay participation in business and commerce is to be supported by government finance, technical assistance, and other facilities. Non-Malay and foreign enterprises are being encouraged to participate in the development of Malay and other indigenous executives, managers, and entrepreneurs. Such on-the-job training is to be supplemented by having the government set up enterprises and train Malays to take them over in due course. What are the difficulties with the NEP'S targets for employment re- structuring? Some commentators feel that Malay managerial talent cannot be created as rapidly as the NEP implies and that excessively speedy implementation of the policy may be a mistake (Rafferty, 1975a,b). It is 44. Under one set of assumptions, Moore (1975) estimates that during the 1970-90 period 63.3 percent of all new jobs in four leading occupations (professional and technical, administrative and managerial, clerical, and sales) will have to be reserved for Malays, and 57 percent of all new Malay entrants to the labor force will have to be directed to these four occupations. CONCLUSIONS AND NOTES ON POLICY 297 probable that pursuing the racial participation ratios for skilled occu- pations will require increasing the percentage of Malays in many training and educational institutes well beyond their share in the population. This may imply a lowering of standards and the admission of underqualified Malays into institutions of higher education (Moore, 1975). But this is likely to be a temporary cost, borne only during the transition period until the system adjusts to its new equilibrium. Questions have been raised, however, as to whether even the overall output of Malaysian vocational training institutes and universities will be sufficient to meet the manpower needs of the opp, and whether the present flow (and past stock) of technologically skilled Malays will be adequate to fill the Malay quotas.45 It is also possible that efficiency could be jeopardized by excessively rapid employment restructuring. Given the present lack of trained Malays, some authors believe there will be "unfilled slots" in the higher occupational levels or "slots which can at best be filled by underqualified persons" (Thillainathan, 1975a; Moore, 1975). This outcome could itself constrain growth of overall output and employment because of the long lags associated with training high- and middle-level manpower and providing appropriate educational and training facilities. Yet the rapid growth of output is a necessary condition for the success of the NEP, for it is "only through such growvth that the objectives of the NEP can be achieved without any particular group in Malaysian society experiencing any loss or feeling any sense of deprivation" (MTR, p. 63). The potential for the medium-term growth of output and employment could be affected by the emigration of skilled manpower if racial employment quotas are rigidly enforced. Some highly skilled non-Malays might choose to emigrate if faced with limited opportunities for promotion and career advancement.46 Given the scarcity of high-level manpower in Malaysia, growth prospects could be adversely affected by unforeseen reductions in supply. In addition, growth prospects could be dampened by a reluctance of the business community to invest if quotas make investment less profitable at the margin. Thus rigid enforcement of the restructuring targets could entail tradeoffs with medium-term output and employment growth; such growth would itself be instrumental in any effort to restructure and to eradicate poverty. 45. According to von der Mehden (1975). the difficulties of developing Malay managerial pools in the private sector arise out of cultural and educational factors which may take more than twro decades to reverse. From interviews conducted in 1973, he concludes that trained Malays have traditionally found the civil service and politics more attractive fields of endeavor than commerce and industry-but this may be changing. 46. Indeed, Shaplen (1977) claims that some Chinese professionals are already emigrating 298 INEQUALITY AND POVERTY IN MALAYSIA While it is not easy to quantify the tradeoffs, one would expect some cost in output and efficiency if employment quotas impose a constraint on economic activity.47 But such losses may be interpreted as short- and medium-term adjustment costs, which could be offset partially or even outweighed by longer-term gains. Once the stock of skilled Malay manpower has been increased in line with the Malay population ratio, improved long-run efficiency could ensue from the wider pool of talent. The New Economic Policy: Concluding Comments The reduction of individual income inequality is an objective of considerable importance in most developing countries, and is one about which the government of Malaysia has itself expressed concern (see MTR, p. 62, and TMP, p. 51). But it is not an explicit objective of the New Economic Policy. The eradication of poverty and the correction of racial imbalances are the NEP'S twin objectives. It has been shown in chapter 3 and appendix E that the redress of poverty is the most efficient method of redressing individual income inequality. More precisely, if decreases in individual incomes are ruled out, the maximum reduction in inequality is obtained through a policy of distributing incremental income to fill the poverty gap from the bottom upward. Hence reducing racial income imbalance-by increasing Malay incomes so that the Malay distribution is scaled up toward the non-Malay distribution-will be less efficient in redressing individual income in- equality. In fact, as seen in chapters 3 and 6, the complete elimination of racial income imbalance will reduce individual income inequality by only about 10 percent; at the same time, it will entail a far greater expenditure than that required for the eradication of poverty.4" It is useful to compare the reduction in inequality for the same expenditure on the two prongs. Since a smaller amount is required to eradicate poverty than to eliminate racial income imbalance, let us assume that the expenditure on each prong is the total income gap of the poor. This 47. Quotas in proportion to population ratios can produce economic inefficiencies of vanous kinds. These inefficiencies could turn out to be particularly acute when quotas are applied in favor of a majority community such as the Malays in Malaysia-in contrast with a minority community, such as the blacks in the United States or the scheduled castes and tribes in India. 48. In terms of the PES distribution of individuals by per capita household income, the poverty gap works out to M5401 million a year in 1970, while M$1,879 million a year is required to raise all Malay incomes by a factor of 1.96 to eliminate the racial disparity ratio between the Malays and non-Malays. CONCLUSIONS AND NOTES ON POLICY 299 amount allows all poor individuals to be brought up to the poverty line and allows all Malay incomes to be raised by a factor of 1.20 (Malay incomes need to be raised by a factor of 1.96 to eliminate racial income imbalance completely). The expenditure on the first or poverty prong achieves a reduction in individual income inequality of 17.1 percent as measured by the Gini coefficient and of 20.5 percent as measured by the Theil Tindex.49 The same amount spent on the second or racial balance prong achieves a reduction in individual income inequality of 2.5 percent as measured by the Gini coefficient and of 4.5 percent as measured by the Theil T index.50 As an instrument for individual inequality reduction, therefore, the first prong is much more efficient than the second. Despite the differences between the two prongs of the NEP, there turns out to be a common area of policy in which the two are mutually consistent. The first prong seeks to help the poor irrespective of race; the second prong seeks to help Malays, including middle- and upper-income Malays. As it happens, the intersection of the two target groups is rather large, with almost four-fifths (78.1 percent) of poor households Malay, and just over half (51.4 percent) of Malay households poor (see table 4-3). Hence policies that raise the incomes of poor Malays will help promote both prongs. This interdependence can be explored further. Suppose that M$ I is spent on each prong at the margin; what is the indirect effect on the other prong? Clearly this effect will depend on the particular way the M$1 is distributed over the target population. For example, if M$ I is spent on the first prong, the amount accruing to Malays will depend on whether it is spent on redressing poverty from the bottom upward or on alleviating poverty close to the poverty line. Similarly, if M$1 is spent on the second prong, the amount accruing to the poor will depend on how it is distributed over different parts of the Malay income distribution. It is assumed that the marginal M$ 1 is spent on each prong just as the average M$ I would be spent. In other words, the marginal M $1 is distributed over the poor so that 41. Owing to zero-income individuals, varlog and the Theil L measure are not computable for the distribution of individuals by per capita household income. After all poor individuals have been brought up to the poverty line, inequality in this distribution would be given by: Malay Chinese Indian Other Total Gini coefficient 0.3023 0.4325 0.4463 0.6681 0.4126 Theil T index 0.2493 0.3971 0.4799 0.8411 0.4104 Mean income (MS 39.42 68.95 58.33 189.51 53.08 per month) 50. When all Malay incomes are raised by a factor of 1.20, the new Gini coefficient is 0.4857 and the new Theil T index is 0.4929. (See the section "Policy Considerations" irn zhapter 3 for the results when all Malay incomes are raised by a factor of 1.96.) 300 INEQUALITY AND POVERTY IN MALAYSIA the poverty gap of each individual is reduced equiproportionately, and the marginal M$1 is distributed over Malays so that all Malay incomes are raised equiproportionately. Under these assumptions, the indirect effect on Prong 2 of M $1 spent on Prong I is M$0.79 (see chapter 4); that is, 79 cents of every dollar spent at the margin on reducing poverty serve also to reduce racial income imbalance. The indirect effect on Prong I of M$1 spent on Prong 2 is MS0.25; that is, 25 cents of every dollar spent at the margin on reducing racial income imbalance serve also to reduce poverty. Thus the indirect effect of focusing on poverty is greater than the indirect effect of focusing on racial imbalance.5' Ultimately, however, the indirect effects of one prong on the other may be of limited consequence. The two are not only logically independent, but also defensible in and of themselves. In addition, the two prongs have their basis in considerations that go beyond the narrowly economic ones. Prong I can be justified in terms of two moral principles. The first is humanitarian: people's basic needs should be satisfied. The second is egalitarian: income inequality should be reduced.52 Under certain con- ditions, both principles can be derived from utilitarianism; both can also be defended as nonutilitarian principles of distributive justice. Prong 2 raises other moral issues. One is equality of opportunity. The association of race with economic function may be seen as evidence of unequal opportunities in the past. Although formal equality of oppor- tunity may now be said to exist, the perceptions and aspirations of Malays may still be constrained by historical and social tradition. Quota systems that place Malays in jobs hitherto unfilled by them may help to provide role models and widen the perception of choices available to them.53 Another moral issue is compensatory justice. For example, if there have been past injustices against the Malay community, there is a prima facie case for compensation. Such arguments for compensatory justice have recently 51. The indirect effect of focusing on rural poverty would be even greater: if the marginal MSI on Prong I were distributed only among the rural poor, Malay incomes would be raised by more than 79 cents. 52. This defense of Prong lin terms of egalitarianism is contingent on the fact (established in appendix E) that the redress of poverty leads efficiently to the redress of inequality. 53. They may also enhance the self-esteem of younger generations of Malays (although in the short run they could have the opposite effect psychologically). Malays might thus be encouraged to aim for positions they previously ruled out, and this would help create real equality of opportunity. Once equal access has been achieved, however, discrimination on grounds of race becomes difficult to defend. CONCLUSIONS AND NOTES ON POLICY 301 been advanced to defend policies of reverse discrimination or affirmative action in favor of blacks and women in the United States.54 Political considerations also need to be taken into account. If racial imbalances persist and race is seen to be associated with economic function, tensions could develop between the racial groups which might threaten the countiy's stability--a danger highlighted by the 1969 riots. It is thus possible to adduce strong prudential reasons in support of the government's policy of restructuring. The NEP has been justified in terms of the "overriding objective to promote national unity in the country." The government recognizes that national unity has several facets, and indeed stresses "there must be no delusion that national unity can be achieved by purely economic means." Nevertheless, the redress of racial economic imbalances is a necessary condition for maintaining harmony in this multiracial society. The redress of poverty, which cuts across ethnic lines, is also a necessary condition for maintaining national unity. The ultimate success of the NEP will depend on the achievement of an equitable balance between the two prongs. 54. ';ee Cohen, Nagel, and Scanlon (1977) for discussions of the philosophical basis of such policies. Some people reject reverse discrimination on the ground that liability for the actions of past generations cannot be located within the present generation. Even if liability could be so located, reverse disenmination may be thought unfair because a few individuals end up bearing the costs of compensation, which should properly be borne by the community as a whole. In the case of Malaysia, there are further complications. First, it might be argued that the economic backwardness of Malays is not itself proof of past injustices. Second, even if there were such proof, the problem of locating liability would be compounded by the former presence of a colonial power in the country. To the extent that non-Malays have been the beneficiaries of colonial policy, however, it might be held appropriate that they should bear some of the costs of compensation. Appendixes The Measurement of Income Inequality THE FOLLOWING SIX APPENDIXES present some recent results on the measure- ment of income inequality. In part they are a selective survey of work in the area, which has been adapted to shed light on problems that concern us in the text. In part they also present some new results (see appendixes B, E, and F) which have arisen out of measurement problems encountered in the empirical analysis of the Malaysian data. For excellent surveys of the literature on inequality measurement in general, see Theil (1967, chapter 4), Atkinson (1970), and Sen (1973a). 302 A A Brief Review MOST OF THE IMPORTANT RESULTS in inequality measurement, and many inequality indices themselves, are based on the Lorenz curve for an income distribution. This brief review begins with a definition of the Lorenz curve for a continuous income distribution specified as an income density function f(y);' the definition for a discrete distribution is provided later (see appendix D). Let F (x) = Sf(y)dy be the cumulative population share corresponding to income level x, so that F(x) is the proportion of the population that receives income less than or equal to x. Let tp(x) = (l/p)fxyf(y)dy be the cumulative income share corresponding to income level x, where p = i yf(y)dy is the mean of the distribution. This defines an implicit relation between F and 4) in terms of the parameter x. The graph F(x), 4)(x) is said to be the Lorenz curve of the income distribution f(y). Alternatively, starting with the pth percentile in the income distribution, we can define x as the income level which cuts off the bottom p percent, that is, p = F(x) or x = F '(p). The income share of the bottom p percent in the distribiution is then tP[F- -(p)]. This function gives the Lorenz curve of the distribution, L(p), which shows the cumulative income share corresponding to percentile p(O S p < 1). Thus L(p) = tP[F -l(p)] on the support of F(x). It is easy to check the following propositions, which are illustrated in figure A-1. 1. 0 < F < 1, 0 >. 4D < 1; F(O) = (D(O) = 0, F(oo) = D(co) = 1. 2. T:he Lorenz curve L(p) (O < p 1) is convex, and its derivative L'(p) is given by ,F'(p) L'(p) = - p = - p pt 1. As f(y) is a density function for income, y > Ojf(y) > 0, andf' f(y)dy = 1. 303 to tne Lorenz curve. Now UF) is clearly a triangle ot maximal area that can be inscribed in the larger set OQRD. A fortiori, it is a triangle of maximal area that can be inscribed in the required subset. The area of triangle OPD is (1/2)(OD)(AP/ 12) = [p* -L(p*)]/2, while that of triangle OBD is 1/2. Hence the value of the inequality measure is [p* - L(p*)], or M/2. 304 INEQUALITY AND POVERTY IN MALAYSIA Figure A-1. The Lorenz Diagram (D~~~~~~~~~~~~~~~~11 (0, 1) D _____________________ 11 306 INEQUALITY AND POVERTY IN MALAYSIA Other Indices There are three basic properties that one would like an inequality index to satisfy: (1) mean or scale independence, that is, the index remains invariant if everyone's income is changed by the same proportion; (2) population- size independence, that is, the index remains invariant if the number of people at each income level is changed by the same proportion;3 and (3) the Pigou-Dalton condition, that is, any transfer from a richer to a poorer person that does not reverse their relative ranks reduces the value of the index (Sen, 1973a, p. 27). The properties of mean independence and population-size independence together imply that the index can be computed directly from the Lorenz curve of the income distribution: knowledge of mean income and total population size are unnecessary (see appendix D). Conversely, an index that can be computed directly from the Lorenz curve obviously satisfies the properties of mean independence and population-size independence. The inequality indices in the previous section are defined in terms of the Lorenz diagram and thus satisfy the properties of mean independence and population-size independence. The Gini coefficient satisfies the Pigou- Dalton condition also,4 but the other indices, which are equivalent to the relative mean deviation up to a scalar multiple, do not. From the definition of the relative mean deviation, it is clear that it is insensitive to income transfers between people on the same side of the mean.5 There are two common statistical measures of dispersion for a distri- bution: the range and the variance. The range can be defined as the absolute difference between the highest and lowest income levels divided by the mean income. Since it ignores the distribution inside the extremes, the range obviously violates the Pigou-Dalton condition. The variance, however, does satisfy this property-but it violates mean independence. A way around this deficiency is to deflate the variance by the square of the mean. This gives the squared coefficient of variation, which satisfies all three 3. With this property, the index depends only on the relative population frequencies at each income level, not the absolute population frequencies. In the continuous case this is equivalent to the index being computable from the income density function alone. 4. This is plain from the geometrical definition of the Gini coefficient. A transfer of income from a richer to a poorer person raises the entire Lorenz curve between the corresponding percentiles; hence it reduces the Gini coefficient. 5. Again this is evident from the geometrical representation of the relative mean deviation. In figure A-I, the length of AP is unaltered by income transfers on one side only of p* = F(p) (see also Atkinson, 1970). APPENDIX A A BRIEF REVIEW 305 The Gini coefficient G is defined as the area between the Lorenz curve and the line of equality divided by the area of the triangle OBD below this line The Gini coefficient varies between the limits of 0 (perfect equality) and I (perfect inequality), and the greater the departure of the Lorenz curve from the diagonal, the larger is the value of the Gini coefficient. An alternative definition for the Gini coefficient can be specified in algebraic terms as G = 2(A/p) where A = v x -yjf(x)f(y)dxdy is the absolute mean difference (see Kendall and Stuart, 1963). Thus G can also be defined as one-half the relative mean difference. In appendix B these two definitions are shown to be equivalent. Another measure of inequality based on the Lorenz diagram is the value of the maximum discrepancy between the line of perfect equality and the Lorenz curve (see Schutz, 1951). The distance between the diagonal and the Lorenz curve is evidently maximized at the point p* in figure A-I where the slope of the Lorenz curve is equal to unity; hence the value of the maximum discrepancy is AP or [p* - L(p*)]. The value of the maximum discrepancy turns out to be equal to one-half the relative mean deviation M, which is another measure of inequality (see Sen, 1973a). It is not difficult to prove that [p* -L(p*)] = M/2 = 2 (6/p) where 3 = I Iy-uIf(y)dy is the absolute mean deviation. Yet another measure has been proposed which tries to capture the divergence between the Lorenz curve and the line of perfect equality. This is defined as the area of the largest triangle that can be inscribed between the Lorenz curve and the line of equality, divided by the area of the triangle OBD below this line. As shown below, this measure reduces to the value of the maximum discrepancy, or M/2! The triangle with the largest area that can be inscribed in the convex set defined by the Lorenz curve and the line of equality is OPD. This is seen by constructing the quadrilateral OQRD which contains the convex set; OQRD has OD as base and opposite side QR parallel to OD and tangential to the Lorenz curve. Now OPD is clearly a triangle of maximal area that can be inscribed in the larger set OQRD. A fortiori, it is a triangle of maximal area that can be inscribed in the required subset. The area of triangle OPD is (1/2)(OD)(AP/ /2) = [p* -L(p*)]/2, while that of triangle OBD is 1/2. Hence the value of the inequality measure is [p* - L(p*)], or M/2. 306 INEQUALITY AND POVERTY IN MALAYSIA Other Indices There are three basic properties that one would like an inequality index to satisfy: (1) mean or scale independence, that is, the index remains invariant if everyone's income is changed by the same proportion; (2) population- size independence, that is, the index remains invariant if the number of people at each income level is changed by the same proportion;3 and (3) the Pigou-Dalton condition, that is, any transfer from a richer to a poorer person that does not reverse their relative ranks reduces the value of the index (Sen, 1973a, p. 27). The properties of mean independence and population-size independence together Imply that the index can be computed directly from the Lorenz curve of the income distribution: knowledge of mean income and total population size are unnecessary (see appendix D). Conversely, an index that can be computed directly from the Lorenz curve obviously satisfies the properties of mean independence and population-size independence. The inequality indices in the previous section are defined in terms of the Lorenz diagram and thus satisfy the properties of mean independence and population-size independence. The Gini coefficient satisfies the Pigou- Dalton condition also,4 but the other indices, which are equivalent to the relative mean deviation up to a scalar multiple, do not. From the definition of the relative mean deviation, it is clear that it is insensitive to income transfers between people on the same side of the mean.5 There are two common statistical measures of dispersion for a distri- bution: the range and the variance. The range can be defined as the absolute difference between the highest and lowest income levels divided by the mean income. Since it ignores the distribution inside the extremes, the range obviously violates the Pigou-Dalton condition. The variance, however, does satisfy this property-but it violates mean independence. A way around this deficiency is to deflate the variance by the square of the mean. This gives the squared coefficient of variation, which satisfies all three 3 With this property, the index depends only on the relative population frequencies at each income level, not the absolute population frequencies In the continuous case this is equivalent to the index being computable from the income density function alone. 4 This is plain from the geometrical definition of the Gini coefficient A transfer of income from a richer to a poorer person raises the entire Lorenz curve between the corresponding percentiles, hence it reduces the Gini coefficient. 5 Again this is evident from the geometrical representation of the relative mean deviation In figure A- t,the length of AP is unaltered by income transfers on one side only of p*=F(,) (see also Atkinson, 1970) APPENDIX A: A BRIEF REVIEW 307 properties. In addition, the squared coefficient of variation satisfies a (weak) decomposability property (see chapter 3, the section "The Methodology of Inequality Decomposition"). The other inequality indices considered are defined in terms of a discrete income distribution. Let the vector y = (YI, Y2'... , Yd denote an income distribation among n persons, where yi > 0 is the income of person i(i = I., 2 .. , n). Let the arithmetic mean income of the distribution be p, so that ln p n = I Y The variance of income var(y) can then be written as I n var(y)=- E (yj_e)2 n _ If all incomes are multiplied by the factor A, the variance of income changes by the factor A2. It is easily checked that for a positive scalar A var(Ay) = A2var(y). Unlike the variance of income, the variance of the logarithm of income var(logy) is a mean-independent measure of inequality. Let fi be the geometric mean income of the distribution, so that by definition I n log p =- E logyi. n i , Then the variance of log-income or var (logy) can be written as I n var(logy) = - E (log yi-log )2. n.= Now if all incomes are multiplied by the positive factor A, the variance of log-income does not change at all. It is easily checked that for a scalar A > 0, var (log Ay) = var(log y). The variance of log-income also obviously satisfies the property of population-size independence. However, it does not satisfy the Pigou- Dalton condition for the entire range of incomes.6 The deviation of the logarithms of income is sometimes taken from the logarithm of the arithmetic mean logp rather than the logarithm of the 6. Tlhe Pigou-Dalton condition is not satisfied for incomes above fie, where e is the base of the natiral logarithms. 308 INEQUALITY AND POVERTY IN MALAYSIA geometric mean logp (see Sen, 1973a). This yields a slightly different measure which is also mean-independent: n It can easily be verified that v = var(log y) + (log p - log p)2. But (log p- log i) is itself a measure of inequality, namely, Theil's second measure L (see below). Thus v is really the sum of two distinct inequality measures, var(log y) and the square of the Theil L measure. Bo,th var(logy) and v suffer from a rather serious practical defect. The measures are not defined if there is a person in the distribution with zero income. Unf\ortunately, this happens to be the case with some of the Malaysian disiributions considered. To overcome this problem, some have suggested that the zero-income recipients be assigned a small positive income (such as 1). But the choice of the amount assigned makes all the difference to the value of the measure. The sensitivity of the measure to this arbitrary procedure, and the inability to defend the particular amount assigned, render the measure unusable in such situations. One attractive feature of the variance of log-income, however, is that it is decomposable around group geometric mean incomes (see chapters 3 and 6 and appendix C). Another derives from its relation to the lognormal distribution (see Aitchison and Brown, 1957) and to the estimating form of the human capital model (see chapter 7). Finally, two inequality measures of Theil (1967) are considered. The first is Theil's entropy index T based on the notion of entropy in information theory. It is defined as I = Yi log- ni=1p p where np = 1 I y, = Y is the total income. Note that (yi/p) is simply the slope of the Lorenz curve at the percentile corresponding to income level y,. Hence T, like the measures in the previous section, can be computed directly from the Lorenz curve of the income distribution. The motivation for T, however, derives from the "entropy" H(y) associated with the income shares (y / Y), . . . , (y/ Y): n 1 H(y) = y (yi/Y)log . i_ (yi / Y) The closer are the income shares (y,/Y) to the population shares (I/n), the greater is H(y); and when each (yi/Y) equals (l/n), H(y) attains its maximum APPENDIX A A BRIEF REVIEW 309 value of log n. On the other hand, if one income share tends to unity and all the others tend to zero, H(y) tends to its minimum value of zero. Thus the entropy H(y) of an income distribution can be regarded as a measure of income equality. Theil obtains a measure of income inequality by subtract- ing Hly) from its maximum value, logn.7 This inequality measure is T, which can be written as T= logn-H(y) = n (yi/Y)logl Y) =1 ' (I In) As (I/n) is the population share and (yj/Y) the income share of person i, Theil interprets T as "the expected information of a message which transforms population shares into income shares" (1967, p. 95). When there is perfect equality, each person's income share (y,/Y) and population share (1/n) are equal, and T assumes the value zero.8 When there is perfect inequality, however, a single person receives all the income and everyone else receives zero income: one of the y1's is then equal to Y, and all other y,'s are equal to zero. In this case, T assumes its maximum value of log n; all terms with a zero income share tend to zero, since x logx - 0 as x - 0. The Theil entropy index T fulfills most of the desirable properties specified for a measure of inequality. It is mean-independent and population-size-independent; it satisfies the Pigou-Dalton condition; it is defined for distributions with zero-income recipients; and, finally, it is additively decomposable in the weak sense, with income-share weights for the within-group component-which sum to unity (see chapters 3 and 6 and appendix C). Another inequality index of Theil (1967)1 call Theil's second measure L. It is arialogous to the entropy index T except that it reverses the roles of income share (y,/Y) and population share (1/n) in the formula for T. Thus Theil's second measure can be written as L = E (1/n)log (I(1/n) Theil interprets L as "the expected information content of the indirect message which transforms the income shares as prior probabilities into the population shares as posterior probabilities" (1967, p. 125). Like T, the index L attempts to measure the divergence between income shares and 7. The difference between the maximum entropy value, log n, and the actual entropy value, H(y), is called redundancy in communication theory. 8. T may be thought of as a general distance function which measures the divergence between income shares and population shares. 310 INEQUALITY AND POVERTY IN MALAYSIA population shares, but it uses a somewhat different distance function. Since Y = np, Theil's second measure L can also be written as L=- l iogff. n i I Yi As (y,/p) is the slope of the Lorenz curve at the percentile corresponding to income level y,, the measure L can be computed directly from the Lorenz curve of the income distribution. Rewriting the expression for L, 1 n L = log p-- YE log y = log p -logfi = log - where p is the geometric mean income of the distribution. In other words, L is the logarithm of the ratio of the arithmetic mean income of the distribution to the geometric mean income.9 Theil's second measure L obviously satisfies the properties of mean independence and population-size independence; it also satisfies the Pigou- Dalton condition. Moreover, L is additively decomposable in the strict sense, with population-share weights for the within-group component- which sum to unity (see chapters 3 and 6 and appendix C). One disadvantage, however, is that it is not defined for distributions with zero incomes, since log x - - oo as x - 0. 9. L is also a simple monotonic increasing transform of Atkinson's (1970) index I when the inequality aversion parameter E is equal to unity. In this case, the Atkinson equally distributed equivalent income is just the geometric mean income p of the distribution, and I = I -- (j/p). Hence, L = -log(1 -I). B The Gini Coefficient THE MOST COMMON DEFINITION of the Gini coefficient is in terms of the Lorenz diagram-as the ratio of the area between the Lorenz curve and the line of equality, to the area of the triangle below this line (see appendix A). Various other definitions have also been discussed in the literature and are useful for different purposes. Here several definitions of the Gini coefficient are reviewed, and their equivalence is demonstrated. Suppose there are n individuals (or households) who are labeled in nondescending order of income as: y < Y2 - . yn. Denote this (ordered) income distribution by the vector y = (yr, Y2 ..... yj, and let p be its mean. Let Fi be the cumulative population share and 0P the cumulative income share corresponding to individual i(i= 1, 2, ... n). Define F = 0 = 0. Thus Fi=- and 0,=- y y, for i =0,1, . . ,n. n nY k= I The first definition is the one used in this study to estimate Gini coefficients for Malaysia. Defi[nition 1 (Geometric) n1-l CG, =I- E (Fi ,-F j) (0i +, + 0,). i=o It is shown that G, is equivalent to the geometric definition of the Gini coefficient given above. Figure B-1 illustrates the Lorenz curve for the discrete income distribution y = (YI, Y2, , ynj where Yi < Y2 < . Yn- The shaded part shows a typical segment of the area below the Lorenz curve. The total area below the Lorenz curve In-l =- E (F,+,-Fj)(D,+, +±j). 311 312 INEQUALITY AND POVERTY IN MALAYSIA Figure B-1. The Lorenz Curve for a Discrete Income Distribution E 0 0F, F, F Cumulative population share Therefore the Gini coefficient n-I(Fi,I i)1 D, 1/2 2 2i=(o l s(i++P) n-I = I - Y (Fi + I-F, )((Di , + Oi) i=o = GI . || Definition 2 (Rao, 1969) n-I G2= (Fi 0,+ 1- Fi + I Dj ,=1 It is shown that GI = G2. n-I G = 1- 2 (Fi + I-Fi) ((i+I + +i) i=O n-I n,I = 1I+ = (F0@,+,-Fi+,bi)- (Fi+,Oi+,-Fi¢,). i=O I1=0 APPENDIX B: GINI COEFFICIENT 313 n-i But Y (Fs+l4>i+l- Fj ) = F.On-Fo(o = 1, i=o since F, = (D. = 1, and Fo = , = 0. Therefore n-I G= (F,41,+-Fi+14D) .=1 =G2.II Defiinition 3 (Kendall and Stuart, 1963) Kendall and Stuart define the Gini coefficient as one-half the relative mean difference, that is, one-half the average value of absolute differences between all pairs of incomes divided by the mean income. Thus, 1 n n G3 = 12n28j- j- =Y 1 This definition implies that 2n2pG3 is the sum of every element of the symmetrical n x n matrix whose (i,j )th element is | -y |. It is shown that G3 = G;.. Since individuals are labeled in nondescending order of income Y I Y 2 Y., G3 can be written as: G3 = 2 (Yi ] I .,IY . YI I 2 n21 E iyi E Y- Yj] = 28 2 [iyi-njqi0j. Substituting Y = (0i -)i 1) and - = Fj, np n one has G3 = Y_Y nI = Y (Fi.- I , - Fi'Di-1) sinceF,F_~ n-I = Y (FiDi+I-Fi+,I i) i= G 314 INEQUALITY AND POVERTY IN MALAYSIA Definition 4 (Sen, 1973a) With individuals labeled in nondescending order of income so that Yi < Y2 . s y,, Sen defines the Gini coefficient as 1 2 G4 = + --2 [ny, + (n - OY2 + **+ 2Yn,- + Yn] n n + 1 2i -2 u, (n +1I-i)yi. This form makes clear the income-weighting scheme in the welfare function behind the Gini coefficient. Rank-order weights are applied to different people's income levels so that the poorest person receives a weight of n, the ith poorest person a weight of (n + 1 - i), and the richest (or nth poorest) person a weight of unity. It is shown that G3 = G4. As before, G3 can be written as 1n G3 = 2 Yi yj)l which is the sum of all the elements of the lower triangular matrix: (Y, -Y,) 0 ... 0 (Y2 -Y1)(Y2- Y2) .. ° 0 (Y. lYl)(Y. Y2) .. (Yn Yn)j Summing the first element in each bracket horizontally by row gives 'InI iyi; summing the second element in each bracket vertically by column gives -E = I (n + 1 -j)yj. Hence G3= n2[ iyi- (n + -i)yi] = n2 [ E(n + 1)Yj -2 Y (n + I1- Oyi] = n__ - 2- E (n + 1-i)yi = G4.II APPENDIX B GINI COEFFICIENT 315 Definition 5 (Fei and Ranis, 1974) Fei and Ranis express the Gini coefficient as a linear transform of the rank index of the income distribution. The rank index R is defined as a weighted average of the ranks of persons in the income distribution, where the weights are their income shares. With individuals labeled in non- descending order of income as y, Y2 . . y Y, the rank index R can be written: n I n R = E iy;I /Z , j=1 I ,=, The Gini coefficient G5 is then defined as 2 n+ 1 G5 =-R- n n = 2 [lv +2v2+. . + 'n n It is shown that G4 = G5. n_ - 2 E (n+1-i)y n= - 2 L E (n+])y;- iy,] 2 n2 . _n+ = 2 n n +Y- I nn = G5sI Up to a multiplicative constant, G5 can also be expressed as the covariance of incorne and its rank: 2 G5 =-cov(i, yj). n,u This is easily checked as follows. By definition of covariance, cov (i, I~ -Z(i -I)(y1 -jr) I n n i = I 316 INEQUALITY AND POVERTY IN MALAYSIA where a bar above a variable denotes its mean. But since -1 E i_(n+1) y 2,and u=, n=1 2 cov(i,y1)=- E iyi-( 2p. 2 2 n (n +1 Therefore -cov(i, yf) = 2 - _(n+1 nu n 1 i= 1 n = G5 .I The equivalence of five alternative definitions of the Gini coefficient has thus been demonstrated. The Effect of Changes in Certain Incomes It is often useful to predict the effect on inequality of changes in certain incomes in the distribution. In appendix E general results on this question are proved in terms of Lorenz dominance, but here attention is restricted to the Gini coefficient. It is clear that if the income of every individual is raised by the same proportionate amount, the Gini coefficient, being mean-independent, will remain unchanged. But if the income of everyone is raised by the same absolute amount, say £ > 0, the Gini coefficient will decrease as a straightforward function of E. This is easily seen. Starting with the distribution y = (yI, Y2, . y .), let x = (xl, x2, . . . , x,) be the distri- bution with x, = (yi + c), i = 1, 2, . . , n. Then if the mean of distribution y is u, the mean of distribution x is (p + e). Using definition 3 of the Gini coefficient, one has G(x)= 22(pE) kxi-XjI 2n' (ju + e) j, j = (p G) (y) < G(y). Indeed the Lorenz curve for distribution x lies above that for distribution y (see corollary 3 in appendix E). Distribution x evidently tends to perfect equality as the absolute amount E by which everyone's income is raised APPENDIX B GINI COEFFICIENT 317 becomes indefinitely large. From the expression for G(x), it follows that G(x) -*0 as E -+* o. Now suppose the income of only certain individuals in the distribution is changed. For example, suppose the income of the richest person is raised and the income of the poorest person is reduced. In this case, the ranks of these two persons in the distribution are unaltered and, from the rank index definition, the Gini coefficient can be seen to increase. Indeed the new distribution will show more inequality in the Lorenz sense (see corollary I in appendix E). In the general case, let the income of thej th poorest person in the distribution be changed without altering this person's rank; the effect on the Gini coefficient can then be measured by using the rank index definition. By definition 5 of the Gini coefficient, G = -R n + n n where R = E iyj yi i= I j= is the rank index of the income distribution. Differentiating G partially with respect to yj, one has aG 2 2(R) ay, -n tay, ) -- (j j-R)/ Yi which is' 5 0 as j 5 R. Hence if the jth poorest person's income is raised without altering the rank order of individuals, G falls if j < R and G rises if j > R. If thej'h poorest person's income is reduced without altering the rank order, then G rises if j < R and G falls ifj > R. It can be shown that (n + 1)/2 < R n.' Hence if the income of anyone below the median income level2 is raised (reduced) without altering the rank order of individuals, the Gini coefficient will fall (rise). 1. Since it is always the case that G > 0 (see definition 3 of the Gini coefficient), G =(2/n)R-(n+ l)/n -O, or R Ž (n+ 1)/2. And since i < n, it follows that R = E iy,l E y, -< E ny,| E y, = n. Therefore, (n + 1)/2 S R s n. 2. The median income level may be defined as Y./2 or y(f + l)/2 depending on whether n is even or odd. 318 INEQUALITY AND POVERTY IN MALAYSIA The Disaggregation of Income by Factor Components The personal income of an income recipient is typically the sum of income from several sources (see, for example, the definition of PES income in chapter 2). For simplicity, suppose that personal income is made up of two factor income components only, corresponding to labor and capital in functional income terms. Thus, let personal income yi be the sum of wage income w; and property income ri; that is, y, = wi + ri, for i = 1,. . . , n. What is the relation between the Gini coefficient G(y) for total personal income and the Gini coefficients G(w) and G(r) for wage income and property income, respectively? Let a bar above a variable denote its mean over the population, so that n I n I n n, yi, wv - Y wi, T - Y, ri; and y P. + Then the following result of Rao (1969) can be stated: G(y) < _ G(w) +=:G (r). y y Definition 3 of the Gini coefficient turns out to be the most convenient one for proving this result. (The proof given by Rao is a little complicated because he uses definition 2.) Thus, 2n2 vG (y) = E_ I y -yj I i j = E lwi -Wj+r i-rjl . J Y l wi -wj I+ E' Iri - rj i,j with equality if and only if w, w wj whenever ri 5 r, for all i,j, that is, if and only if the rank order of individuals by wage income is identical to their rank order by property income. Applying the definitions of G(w) and G(r), it follows that 2n2jG(y) A 2n2ivG(w)+2n2FG(r). Hence G(y) =- G(w)+ _G(r). II y y Thus the Gini coefficient for total income is less than or equal to a weighted average of the Gini coefficients for wage income and property APPENDIX B. GINI COEFFICIENT 319 income, where the weights are the shares of wage and property income in total income, respectively. The result obviously generalizes to more than two factor income components. Information on the distribution of factor income components therefore allows an upper bound to be placed on the distribution of total personal income. It can be shown that the larger the rank correlations between factor income components and total income, the closer is the overall Gini to the weighted average of factor income Ginis. Equality obtains if and only if the rank ordering of individuals is the same by every component of income. On the Decomposition of the Gini Coefficient An inequality index is defined to be additively decomposable in the weak sense (see chapter 3) if it can be written as the sum of a between-group component and a within-group component where (1) the between-group component is the value of the inequality index when each member of a group receives the mean income of the group; and (2) the within-group component is a weighted sum of the inequality indices for each group where the weights depend only on the population or income shares of the group, or both. It is shown that, in general, the Gini coefficient is not decomposable according to this definition. Divide the population into two subgroups, labeled x and y, respectively.3 Let there be n. individuals in the x subgroup with incomes xi(i = 1, . . ., nx) and mean income pu; let there be ny individuals in the y subgroup with incomes yj(i = 1, . . ., n.) and mean income py. Further, let n be the size of the whole population and p its mean income. Then 1 n' 1 n, X =-E xi, uy =- y1, n = n.+ny, and ny = nXjIIX+nyUY nx i= I nvj= I What is the relation between the Gini coefficient G for the whole population and the (Gini coefficients Gx and Gy for the two subgroups? To study this, definition 3 of the Gini coefficient is most useful: I n, n, Gy = 2 Y 2 |Yi-YZ | It is convenient to think of the absolute values I yj-yjr I as the elements of a symmetrical matrix. For the whole population of n = (nX + ny) individuals, 3. Results for partitions of the population into more than two subgroups follow by induction. 320 INEQUALITY AND POVERTY IN MALAYSIA this matrix can be written in an obvious notation as follows: xi, Yj,I Xj Ixi-x,Yl Xj-yjj Yj y, -xi, I I Yi - Yj l The expression 2n2pG for the whole population is the sum of all the elements of this matrix. This sum can be disaggregated as follows: 2n2,uG = Sum of absolute differences between all pairs of incomes = E Ixj-x I+ E Iyj-yj .I+2yIx-y,l i. il J.J' i,j = 2nuxGx + 2n',uyGy + 2 Ixi - Y i,, Thus, G = () (nP j) X (n)(Ž) G x (B. 1) The first two terms in this disaggregation of the Gini coefficient can indeed be taken to represent the within-group component: they are a weighted sum of the subgroup Gini coefficients, where the weights -are the subgroup population share times income share.4 But it is by no means clear that the third term measures the between-group component. By definition, the between-group component Go is the value of the Gini coefficient for the distribution in which nx individuals receive income yx and ny individuals receive income juy. It is easily verified that Go n2p1 4 Thus the weights on the subgroup Gini coefficients sum to a number less than unity; the definition of weak decomposability does not require the subgroup weights to sum to unity. APPENDIX B: GIN] COEFFICIENT 321 The tlhird term in the disaggregation (B.l) is in general greater than Go: n n n n, E I xi-y >1 E I E (X.-y )I =I j= I i=I j=1 with equality if and only if each xi is either larger or smaller than all the yj's nF n = Z 5nyx;-nyH| since EZ Yi=f nylY i=l j=l n, =ny E Ixj-PyI if= I n, i = I with equality if an d only if all the xi's are either larger or smaller than py =nItnh.-hnxeryeo since re xi=n x = nxny I/X -PyI nzn, Thus pE E |xi -ayj I equ nxny t 1x- nmy we and equality obtains if and only if all the xi's are either larger or smaller than all the yj's. Therefore, n2 E E X,-J I) n Y I ^px |,y = Go. Hence the third term in the disaggregation of the Gini coefficient is generally greater than the between-group component Go. Only in very specialI circumstances is it actually equal to Go, namely, when all xj(i = 1_ . . , nx) ar-e either larger or smaller than all y,(j = 1_ . . ., n,). In other words, the x and y subgroups must be nonoverlapppng for the third term in the Gini disaggregation (B.l) to measure the between-.group component of inequality. When the subgroup distributions do not overlap, the Gini coefficient. can indeed be decomposed into the sum of a within- group component and a between-group component.5 But in general all that 5. An obvious exarrple of nonoverlapping subgroup distributions occurs when the population is divided into the two subgroups of the poor and nonpoor by means of a poverty line (see "The Sen Poverty Measure" in chapter 4). In this case, the overall Gini G can be built up from the subgroup Ginis G, and G,, for the poor and nonpoor, respectively, and the subgroup population and income shares. Using the notation of chapter 4, we can substitute 322 INEQUALITY AND POVERTY IN MALAYSIA can be asserted is that G( n-) ( n-, ) Gx ( n ) ( ny'u ) Gy + n" n x-y | (B.2) It has been shown that the Gini coefficient is not additively decomposable in the weak sense. Hence it is also not decomposable with the within-group component defined as a population or income-weighted average (not just sum) of the inequality indices for each group-unlike the two Theil measures (see appendix C). But can a lower bound at least be placed on the overall Gini in terms of a population or income weighted average of subgroup Ginis? In other words, is it the case that G > (n) G± ()nY Gy (B.3) or G > (Inxu) G± + (nyPu) GY. (B.4) These lower bounds for G can be better than that in (B.2). For instance, when j = py and all inequality arises from inequality within subgroups, the right-hand side of (B.2) is clearly smaller than the right-hand side of either (B.3) or (B.4). Since the weights in (B.2) are population share times income share, which is less than either population share or income share by itself, the lower bounds in (B.3) and (B.4) will be better than the lower bound in (B.2) when px = p.y The bound in (B.2) will be better, however, when Gx = Gy = 0, and all inequality arises from between-group differences in income. In this case, the right-hand side of (B.2) gives the correct value for the Gini coefficient, that is, G = 28 I| x-i' I, while the right-hand sides of (B.3) and (B.4) both give something worse, namely zero. To prove the inequalities (B.3) and (B.4), definition 3 of the Gini n = q, pu =v, G. = Gp; n, = (n-q), p, = (np-qv)/(n-q), Gy = G., Hence, = (q)() n (n-q)(np -qv (q) (qv). G = (-tv)Gp+ )G.)G" +n-nJ From this decomposition, the Gini coefficient for a particular subgroup (for example, G..) can be inferred from the overall Gini (G), the other subgroup Ginis (here simply Gp). and the subgroup population and income shares. Thus the Gini coefficient for the nonpoor in Peninsular Malaysia can be computed from the information in tables 3-8 and 4-2. APPENDIX B: GINI COEFFICIENT 323 coefficient is used together with the identity yj-y, I yj+ yj-2min(yj,iyw,). Thus GI = T,,2 2 E E I|Y,- yj'I 2ny py JE, [yj + yj-2 min (yj, yj ,) = l I- 2 Z min(yj, yj ). ny jay j, j It follows that 2, min (y,, yj.) = n 2py(1 - )G (B.5) j. ji Similarly, , min(xi, xi,) = n 21(1 -GQ). (B.6) Again the terms min(yj, yj.) can be thought of as the elements of a symmetrical matrix. For the whole population of n = (n, + ny) individuals this matrix can be represented as: . . Xi' Yi, xi min (xi, ) min (xi, yj,) Yj min (yj, xi,) min* (yj. yj) The sum of the minima over all pairs of incomes in the population is just the sum of the elements of this matrix. Thus n2p(1 -G) = ? min(xi, xi,.)+ min(yj, yj ) H- j min(xi, yj.) + E min(xi, yj) i,j i.j n n2 px(I - G.) + n 2,a (I - GY) + 2 min (xi, yj). (B.7) IC y~~~~~~~~~~. 324 INEQUALITY AND POVERTY IN MALAYSIA Hence, 2 min(xi,yj) = n2p(1 -G)-n'pu(l -G.)-n2py(I -G,). .. J To prove the inequalities (B.3) and (B.4), I make use of the following general result (Zagier, 1977): Lemma: 2aBflZ min(xi, yj) < a2 E min(xi, x,) + 132 min(yj, yj.) for i, j i,i, j j' any real numbers a, P. Proof: For t > 0, define the characteristic functions y (t) and 6j(t) corresponding to each income level xi in the x subgroup and each income level yj in the y subgroup as follows: y(t) = x[O,](t) = I if t > for i=1,.., n, [Ox, 0 if t> Xi j =(t) = x (t) = {O if t > yj forj =1, . . ., ny. ro 10 t if t > y n Let 4 (t) = Y, (t) nf and P7(t) = Y 6j(t). j=, Thus ((t) is the number of individuals in the x subgroup with income greater than or equal to t, and l(t) is the number of individuals in the y subgroup with income greater than or equal to t. It is now shown that f ((t)q (t)dt = E min(xi, yj). 0 a. ) By definition, ((t)jq(t) = Zyi(t)j( i., j and yi(t)6 (l)= if t < xi and t < yj, that is, if t min (xi, yj) 0o otherwise. Thus J ((t)q(t)dt = [Y y,(Obi(t) dt = f f y1(t)6j(t)dt, reversing the operations of i*j 0 integration and summation, = Zmin(xi, yj). i. j APPENDIX B GINI COEFFICIENT 325 Similarly, it follows that I 4(t)2dt = E min(xi, x,.) o c0 and f ?I(t)2dt = E min(yj, yj.). 0 j j. Therefore, a2 E min (x,, xi,) + f2 E min (y,, yj,) -2a: min (x,, y,) j,EJ i, j =J[cic(t)-_flq(t)]2 dt 0 ~oI 0°.1 Note that the value of the integral is zero if and only if 4(t)/1(t) = f,/a for t ? 0. This happens if and only if at each income level the number of individuals in the x subgroup is a constant multiple (#/a) of the number of individuals in the y subgroup. Hence the integral is zero if and only if one subgroup distribution is a replica of the other and, apart from total population size, the two are identical. Substituting the expressions (B.5), (B.6), and (B.7) in the statement of the lemma, the inequality becomes: at2n2ux x(I -Gx) +2 n2n' ^a(I - G) >, a,B[n 2,u(1G) -n 2,ax(l- Gj - n 21juy( l-GY)] or a±(xz+)n2pX(I -Gx)±fl@±fl)n2p'(1 - GY) L ann2p(l - G) and hence, -Qnx (I-Gj + fn 2pY(1-GY) n2 p(l-G). This general inequality relating subgroup Gini coefficients and the overall Gini coefficient is valid for all real numbers a and ft. Substituting a = and ,B =-, nxux nyly one gets nx(I -Gx)+ny(l -Gy) Ž n(l -G), that is, G > (x) Gx + (n) GY. (B.3) 1n 1 Substituting a = -- and / -, one gets n, ny n..x -G) nyp)( Y -n( 326 INEQUALITY AND POVERTY IN MALAYSIA that is, G > (nxpx) G + (n y ) GY (B.4) Hence it has been shown that the overall Gini is greater than or equal to both a population-weighted average of subgroup Ginis and an income- weighted average of subgroup Ginis.6 Strict equality obtains in these Gini inequalities only if the subgroup distributions are identical. Otherwise, even with u, = py, that is, with between-group inequality zero, the overall Gini will be strictly greater than the population- or income-weighted average of subgroup Ginis. 6. Zagier (1977) provides various other bounds for the Gini coefficient and a characteriza- tion of the class of decomposable inequality indices. c The Decomposition of Three Inequality Measures IN THIS APPENDIX the decomposition of three inequality measures is considered: the Theil entropy index T, the Theil second measure L, and the variance of log-income V. Empirical decompositions have been performed for these three measures in the main text. The exposition here is based on the PES data format actually used in the computations. The basic data array is in the form of a table (similar to table 3-2) showing the joint distribution of individuals by income and some other variable (such as race) according to which I he decomposition is desired. In other words, there is a matrix which shows the absolute frequencies nij of individuals in each cell (i, j). Let the columns j of the matrix refer to different income classes, and the rows i to different values of the decomposition variable (such as race). Assume that each individual in the j th income class (column) receives the mean income yj of that class (this is the standard assumption in all the tables of the text). The Theil Entropy Index T The T heil entropy index Tcan be defined in terms of this data matrix and notation. Since there are nij persons in cell (i, j), each assumed to be receiving income yj (the mean of the]j i income class), total income y j in cell (i,j) is njiy. Hence the total income of all persons is Y E E nijyj i J = zY where Y; = Ij nijyj is the total income of the ith group. The total population 327 328 INEQUALITY AND POVERTY IN MALAYSIA size n is given by n = E ii = z nj where ni = .j nij is the population size of the i h group. The Theil index Tfor this distribution is then given by (see appendix A) T = i _ Yj log Yyj/ i j Y n,jln where yij/Y is the income share of cell (i, j), and ni/n the population share. Consider the decomposition of the Theil index T into between-group i and within-group i components: T= z y Y+g log Y1i] i Y, Y. I nijlni Y/n Y_ i ~ yj og__!] + ; Y. gyi/y i E-Y [ Yi nfln. i Y ni/n since Y Yij I for each i. J Yi Hence, T = Y, [y] T + y- yi lg nj/ (C.l1) where Ti = Yi logJ/Z j Yi nijlni' Equation (C.1) says that the Theil entropy index Tcan be decomposed into two terms: T = Tw + TB where Tw = YT is a weighted average of within-group i Theil indices T,, the weights being equal to the income shares YJY of the groups, and TB = yi logyi/Y , Y ni/n APPENDIX C DECOMPOSITION 329 is the between-group i Theil index of group income and population shares Y}/Y and n,/n, respectively. Tw is called the within-group component, and TB is called the between-group component. The between-group contribution is then defined as the ratio of the between-group component TB to the overall Theil index T. The within- group contribution is defined as (TwIT). Tlhe Theil Second Measure L Theil's second measure L simply reverses the roles of population share and income share in the formula for the entropy index T (see appendix A). Using the same notation as for T, the formula for L can be written as L E nj og n/n I n~ YijJ/ This reduces to L = log _ nij log yj, which is the logarithm of the arithmetic mean income minus the logarithm of the geometric mean income. Now consider the decomposition of the Theil measure L into between- group i and within-group i components: i n jni Yyij/Yi Yi /Y i= ni[- °Y'j1Y ] n y1°yy, since E ij = I for each i. j ni Hence, L =E [ni ]Li + y, ni log niln (C.2) where Lj = y nilog n/n wherc:~~~~~~ Li Equation (C.2) says that the Theil second measure L can be decomposed into two terms: L = Lw+ LB 330 INEQUALITY AND POVERTY IN MALAYSIA where L W = n] Li is a weighted average of within-group i Theil measures L,, the weights being equal to the population shares ni/n of the groups, and LB = E lnoj0g n/ is the between-group i Theil measure of group population and income shares niln and Yi/Y, respectively. Lw is called the within-group component, and LB is called the between-group component. The between-group contribution is then defined as the ratio of the between-group component LB to the overall Theil measure L. The within- group contribution is defined as (Lw/L). The Variance of Log-Income V The variance of the logarithm of income, just like the variance of any variable, can be decomposed into the sum of a between-group and a wit hin- group component. Defining x,j = logyj (the same for all i), the total variance of xi, can be written as v= - £_ Ynij (Xij _x..)2 where x = -XETnijx,j is the mean of xi, over i and j. n Let x, =E nijx,,I-j ni, be the mean of x,j over j, and rewrite the expres- sion for V as 1 V = - En,,[ (Xl,-x, ) + (xi -X.) 2 nj = - E E [nij(X, -xi)2 + niJ(x, -x )2 + 2n,j(x,,-xi.)(xi,-x)] En nJ (Xi, Xi.)2 + E-(x, -x = E [ n ] Vi + En (x, x..)2 (C.3) where V, = j (nij/ni) (xij - xi.)2. The cross-product term in the expansion vanishes because (xi - x*) is APPENDIX C DECOMPOSITION 331 constant in thej summation, and £j nij (xij - xi) = 0 for each i by definition of the mean xi.. Equation (C.3) says that the variance V can be decomposed into two terms: V= VW + VB where VW [n; V is a weighted average of within-group i variances Vi, the weights being equal to the population shares n1/n of the groups, and VB = Ei (n1/n)(x - x)2 is the between-group i variance of group means x;..' Vw is called the within-group component, and VB is called the between-group component. Thie between-group contribution is then defined as the ratio of the between-group component VB to the total variance V. The within-group contribution is defined as Vwl V. The variance decomposition (C.3) holds for any variable x whether or not this is equal to log-income. The reason for choosing x = log y was to obtain a mean-independent measure of inequality. Another transformation of y which gives a mean-independent measure is x = yip, where p is the overall arithmetic mean. T he variance of (y/p) is known as the squared coefficient of variation, and its decomposition formula can be obtained from (C.3) (see also chapter 3). The between-group component is the squared coefficient of variation of the group arithmetic means, and the within-group component is a weighted sum of squared within-group coefficients of variation-but the weights, which depend on group population and income shares, no longer add up to unity. Comparison of the Decompositions In the foregoing it was seen that the two Theil measures and the variance of log-income are nicely decomposable into between- and within-group components. But the between-group contributions are generally different according to the three measures. The measures emphasize different aspects of the distribution, and there is no reason to expect the contribution by one measure to be the same as that by another measure. The Theil L measure, for instance, is very sensitive to changes at low income levels, whereas the Theil T index is not. Thus the within-group component according to the 1. Note that when x,J = log yJ, xi is the logarithm of the geometric (not arithmetic) mean income of group i. 332 INEQUALITY AND POVERTY IN MALAYSIA Theil L measure can be made indefinitely large relative to the between- group component by redistributing income within a group so that one person's income tends to zero.2 Hence the between-group contribution by the Theil L measure can be made as small as one likes by letting the income of one person in a multiperson group tend to zero. There is a further point concerning the decompositions. As derived, the contributions are purely descriptive. There is no attempt to base the decompositions on statistical theory. Nor is it easily possible to do so for the two Theil measures, since they do not follow known statistical distri- butions. By contrast, the decomposition of the variance of log-income can be based on standard analysis of variance, since the ratio of the between- group to the within-group variance is known to follow an F distribulion. Instead of merely quantifying the between-group contribution, therefore, it is also possible to test it for statistical significance. (With the large number of degrees of freedom in the empirical decompositions in the text, however, all the between-group contributions turn out to be statistically significant.) 2. This transfer from poor to rich within a group leaves the between-group component constant, while increasing the within-group component and the overall value of the measure. D Lorenz Dominance and Inequality IN THIIS APPENDIX [ first prove Atkinson's celebrated theorem about the ranking of income distributions in terms of welfare, Lorenz dominance, and the principle of transfers. Using Atkinson's theorem, I then go on to show that Lorenz dominance is a useful property to establish even when welfare comparisons are not possible between the underlying distributions- but the purpose is purely a positive or descriptive comparison of inequality.' It is easier here to deal with discrete distributions, so let the vector Y = (Yl Y2' - * , yn) denote the ordered income distribution y: 0 < Yi - Y2 < . . < yn among n individuals (or households). The Lorenz curve LY(.) of this distribution can be defined at the discrete points (i/n), for i = 0, 1, . . ., n, as follows: LI(O) = 0 i / n Ly(i/n)= , Yk! Yk for i = 1,2,_ . . ,n. k = I k= I For all other points p in the interval [0, 1], LY(p) is defined by linear interpolation. Now suppose x and y are two ordered distributions with the same number of individuials n and the same total income n n I Xk = E Yk. k = I kk= I 1. See Sen (1973a) for a distinction between positive and normative comparisons and measures of inequality. 333 334 INEQUALITY AND POVERTY IN MALAYSIA Consider three criteria for ranking these income distributions: (i) xS LY, that is, x Lorenz-dominates y, which means that L,(i/n) ) L,(i/n) for i = 0, 1, . n. (ii) X > Ty, which means that distribution x can be obtained from distribution y by a finite sequence of income transfers from richer to poorer individuals, where each transfer preserves the relative ranks of the two individuals affected. This criterion is the ranking of distributions x and y according to the principle of transfers. Given an ordered distribution y, a single progressive transfer d from a richer individualj to a poorer individual i(y, < yj) which preserves their relative ranks leads to a new distribution x defined as: Xk = Yk k i,j xi =yi+d xj =yj-d where xi xj, that is, d < (yj -y)2 2 (iii) x > uY, which means that EZk U(xk) > E-I U(y,) for all non- decreasing concave functions U(y). This criterion says that distribution x yields at least as much welfare as distribution y for any additively separable, symmetric, nondecreasing, concave social welfare function XU(y). It may not always be possible to rank two arbitrary distributions x and y by criterion (i), (ii), or (iii).3 However, when it is possible to rank them by one criterion, it will also be possible to rank them by the other criteria, and in that case all three criteria will give the same ranking. This is Atkinson's theorem. Theorem (Atkinson, 1970): The ranking of income distributions x and y by criteria (i), (ii), and (iii) is identical. Formally, the following statements are equivalent: (i) X > LY (ii) X > 7Y (iii) X > UY. 2. Given symmetry, nothing would be altered if, instead of imposing the condition that the income transfer preserves the relative ranks of individuals i and j, that is, d (y, - y,)!2, the size of the transfer was limited to their income difference, that is, d S (yj - yJ,. This is the more usual way of defining x > TY (see Atkinson, 1970; Sen. 1973a, and Dasgupta. Sen, and Starrett, 1973). 3 In other words, the criteria > L. u, and > T provide only a partial ordering among distributions with the same number of people and the same total income. APPENDIX D LORENZ DOMINANCE 335 Proof: It is shown that (i) => (ii) a (iii) => (i). The proof given here is adapted from Rothschild and Stiglitz (1973). (i) => (ii) X >, y > X >TY Since x and y have the same total income, x >- LY implies that X Xk > Y Yk for i = 1,2,. . . , n. k= 1 k= I Let i be thefirst integer for which xi > yj, so that Xk = Yk for k < (i-1). Define a new distribution x(i) from x as follows. Transfer an amount (xi - y,) from individual i to individual (i + 1); this lowers i's income to yi and raises (i + 1)'s income to xi+ I + (xi - y,). Then the new distribution x(i) has tlhe properties: Xk(i) = Xk = Yk for k (i-1) xi(i) = xi - (xi-yi) = Yi xI + I (i) = xi + I + (Xi- Yi) Xk (i) = Xk for k > (i + 1). Thus x (i) agrees with y in one more place than x-the first i places instead of the first (i - 1) places-and it is still true that x(i) > Ly since i+1 i+1 E Xk (i) =Ex*. k= I k= I The same procedure may be applied to x(i) to find an income transfer from poor to rich which produces a new distribution which agrees with y in at least one more place than x(i) and still Lorenz-dominates y. Continuing in this manner, the distribution y is eventually obtained from x by a sequence of at most (n - 1) transfers from poor individuals to rich. Hence x > TY. (ii) => (iii) X >- TY X > UY Since x >- TY means that x can be obtained from y by a finite sequence of income transfers from rich individuals to poor, it will suffice to show that a single transfer from rich to poor does not lower welfare. Without loss of generality, suppose that x is obtained from y by a single transfer d from individual 2 to individual 1, where d < (Y2 - yl )/2. Then, xi = y, +d x2 = Y2 -d Xk = Yk for k > 3. Let U (y) be any nondecreasing concave function, as shown in figure D-1. 336 INEQUALITY AND POVERTY IN MALAYSIA Figure D-1. The Concave Function U(y) U(y) d d yi Xi X2 Y2 Income, y Given the relation between xI, x2 and Yl, Y2, it is obvious from the concavity of U(y) that [U(x,) - U(y,)] > [U(y2)- U(x2)]. Thus EU(X1)+ U(X2)] ) [U(YI)+ U(Y2)], and therefore E U(xi) > U(yi). Hence x >- uY* (iii) > (i) x )> UY 77 X > LY x > uY implies that k = IU(xk) > k = IU(yk) for any nondecreasing concave function U (y). Consider the function: m, (y) = min (y - z, 0). For each z, m. (y) is a nondecreasing concave function of y, as shown in figure D-2. Hence n n Z mZ(xk) ) Y m.(yk) for each z. k = I k= I That is, Y (xk-z) > X (Yk-Z). (D.1) Xh<2 Y& z APPENDIX D LORENZ DOMINANCE 337 Figure D-2. The Concave Function m,(y) mAY) -z / Now suppose that x > L y is not true. Then it is not the case that i i Y Xk >- yk for each i= 1, 2, . . , n. k = I k Let i be the first integer such that i i Xk < Z Yk (D.2) k= 1 k=1 i-I i-I Since Xk >_ Yk' k= I k= I it follows that xi <: y. Putting z = y; in (D.1) gives Y (Xk Yi -) > (Yk Y) XS y, Y. Y, = (yk-yi). (D.3) k= 1 Now (Xk Y,)= (Xk-Yi) + Y (xk-yd) x J, x X. x,x, < X, y, since xi < y, 1< (xk - yJ) since the second term is xX X, negative = Y (Xk - Yi) (D.4) k= I 338 INEQUALITY AND POVERTY IN MALAYSIA Stringing together (D.4) with (D.3), it follows that L (Xk Y) >E (Yk Z Y,) k= I k= 1 or X k > E YkE which contradicts (D.2). So it must be true that x >- L y, and the proof is complete. 11 In fact, a stronger theorem than this can be proved by adopting a weaker criterion for the welfare ranking (iii). The welfare function n Z U(yi) can be replaced by a symmetric nondecreasing quasi-concave function of individual incomes W(y, Y2,. ,YJ) (see Dasgupta, Sen, and Starrett, 1973). Defining the criterion w as x > wY if W(x) > W(y) for all symmetric nondecreasing quasi-concave functions W, the theorem can also be proved with Ž u replaced by > ,in (iii). The quasi-concavity restriction on the welfare function can be weakened still further. For the theorem to go through, it is clear that the weakest requirement on the function is that welfare does not decrease by a transfer of income from a richer to a poorer individual, where the size of transfer is less than or equal to their income difference. I call a function E(y1, Y2. yn) which satisfies this property "egalitarian"; it is called "locally equality preferring" by Rothschild and Stiglitz (1973). Defining the criterion > E as X > EY if E(x) > E(y) for all symmetric nondecreasing egalitarian functions E, the theorem is also valid with > u replaced by > E in (iii). Since the class of additively separable concave functions is contained in the class of quasi-concave functions, which in turn is contained in the class of egalitarian (or locally equality preferring) functions, it follows that X > EY E> X S wY > x - uY But by the very definition of > E, x >- TY > x >- EY, and from Atkinson's theorem, x > uY > x >- TY; therefore, x > uY > x >- EY. The chain of implications is complete and X >- EY => X >- wY => X >- uY => X >- EY. Thus the rankings of distributions by the three criteria > u, L w, and E are equivalent, and each is equivalent to > L and >-7. Given two distributions with the same number of individuals and the same mean income, Atkinson uses Lorenz dominance to establish an unambiguous welfare ranking between them in terms of all welfare APPENDIX D LORENZ DOMINANCE 339 functions from a wide class. But in any actual comparison of income distri butions (for example, between countries or over time), the population size and mean income are likely to be different. With different population sizes but the same mean income, a simple extension of Atkinson's theorem shows that Lorenz dominance still gives an unambiguous ranking between the distributions in terms of per capita welfare.4 With different mean incomes, the unambiguous welfare ranking survives only when it is the Lorenz-dominant distribution that has the higher mean income. Thus Lorenz dominance cannot always be used for normative comparisons of inequality between two distributions. I put the Lorenz ranking to a somewhat different use in this study. Even if normative comparisons are not possible, for example, because the under- lying distributions refer to different population units or income concepts, the Lorenz ordering can still reveal a good deal about inequality in a positive or descriptive sense. The next result shows that Lorenz dominance provides an unambiguous ranking of distributions by all positive inequality indices from a wide class. Prolposition: Let L be the class of inequality indices that satisfy three basic properties: mean independence, population-size independence, and the Pigou-Dalton condition. If the Lorenz curve of a distribution x(,up, n,) with mean pu and population size nx dominates the Lorenz curve of another distribution y(,yu n,) with mean p3, and population size ny, then all indices from the class L will show less inequality for x(pu, n.) than for y(ply, nj). Proof: Let I be a typical inequality index from the class L of inequality indices. Replicate n ' times the number of people at each income level in x(,iU, n ,), and replicate n,, times the number of people at each income level in y(pju, n)). Then multiply each person's income in x(pu, n n,) by uy and each person's income in y(pu, n.ny) by u,. Then the new distributions x(p.,Uy, n,ny) and y (tt,u n,n,) have the same population size n = nn nyand the same mean p = ;ItHy; hence the hypotheses of Atkinson's theorem hold for them. Furthermore, x(u, n) Lorenz-dominates y(p, n) because the Lorenz curve of x (p, n) is identical to the Lorenz curve of x (M., nj), and the Lorenz curve of y(p, n) is identical to the Lorenz curve of y(p,, n,). Therefcre, by Atkinson's theorem, x(,u, n) can be obtained from y(p, n) by a sequence of transfers from rich individuals to poor. Now as the index I satisfies the Pigou-Dalton condition, it will show less inequality for x(p, n) than for yu(p, n). (A transfer from rich to poor decreases the value of such an 4. If the criterion > w or Ž E is being used instead of Atkinson's > u, then a "symmetry axiom for population" is needed (see Sen, 1973a). 340 INEQUALITY AND POVERTY IN MALAYSIA index.) But since the index I is also mean-independent and population-size- independent, its values for x(pu, n.) and y(p/, n,) are identical to its values for x (ju, n) and y (p, n), respectively. Hence the index I shows less inequality for x(pu, n.) than for y(jiy, ny). I I All the positive indices considered in this study belong to the class L, except for the variance of log-income (at high levels of income). *rhus Lorenz dominance automatically implies less inequality according to such indices as the Gini coefficient, the two Theil measures, and the squared coefficient of variation. It also implies less inequality according to so-called normative indices belonging to this class, such as Atkinson's index, which may be used in a positive or descriptive sense. I call the class L the Lorenz class of inequality indices. E Lemmas on Lorenz Dominance IN THIS APPENDIX conditions are sought under which one distribution will be l,orenz-preferred to another. Lorenz preference implies an unambiguous ranking of inequality according to any index which satisfies mean independence, population-size independence, and the Pigou-Dalton con- dition. In other words, if one distribution Lorenz-dominates another, then it will show less inequality for any measure from the Lorenz class. Also explored here are the effects on inequality of certain transformations to the underlying distribution, including the under- or overestimation of incomes. The results allow unambiguous comparisons of inequality in terms of the Lorenz partial ordering. Suppose there are two ordered distributions x and y with the same number n of population units.' The following lemma provides a sufficiency condition for x to Lorenz-dominate y. Lemrna 1: Let x be the ordered income distribution X:O 1< XI 1< X2 1< .. * *< Xn, and let y be the ordered income distribution Y:O 0 y, Y < Y2 1< - -- Yn- If x,/yI Ž x2/y2 >.. > x.1y,, then x FLY Proof: Since x,/y, ) X21Y2, x, 71+x2 x2 ( 3, by hypothesis). Yi Yl +Y2 Y2 Y3 X X2 I+X2 +X3 X3 'C4 Therefore x > x3 ( X-, by hypothesis). Yl +Y2 Y± +Y2 +Y3 Y3 Y4 1. The population unit itself need not be the same for the two distributions: for example, it could be individuals in one distribution and households in the other. 341 342 INEQUALITY AND POVERTY IN MALAYSIA Proceeding in this way, the hypothesis of the lemma implies the following sequence of inequalities: X> XI +X2> > X+X2 + *+X-l X- + X2 + +nX> X- YI YI +Y2 Y1 +Y2 + +Yn-I Yl +Y2 + * +Y. Y. Hence, x+X2+ +X >Y+Y2+ for = n. xI +x2 + . . Xn YI +Y2 + +Y,, In other words, x >- Ly.11 The converse of lemma 1 is obviously not true. Corollary 1: Let x be an ordered income distribution. Increase the income of the richest person (n) by an amount e > 0, and reduce the income of the poorest person (1) by 3 > 0, and call the new distribution y. Then x Lorenz-dominates y. Formally, let X: O <1 Xl :S X2 :S . .. 1< Xn- 1 S Xn and y: O < (xI -6) x2 < ... xn -1< (xn+ ) where e, t >, 0. Then x> -Ly. The proof of this result is a simple application of lemma 1, since X1 > X2 > > X-I > X. X1 -6 x2 X.-I X,,+ The result says that if the income of the richest person is underestimated, or the income of the poorest person is overestimated (or both), then inequality is unambiguously underestimated. This conclusion does not necessarily follow if incomes at more than one income level are changed at the top and bottom ends of the distribution. It is easy to construct counterexamples to the corollary when such changes are permitted. With certain restrictions on these changes, however, Lorenz comparability is indeed possible. For example, the following result can be proved. Corollary 2: If proportional underestimation increases with income, then inequality is unambiguously underestimated. Formally, let X: O :< XI1 < X2 1< .. * * Xn and y: O 1< X(I +1) < x2(1 +E2) < ....< XX + where El1< 82 < . .. . &,,. Then x > LY. This result is again an easy application of lemma 1, since for s, < +, 1' xi (l + Ej X;i+ I ( + £j+ 1) For absolute underestimation of incomes, there is the following result. APPENDIX E: LEMMAS ON LORENZ DOMINANCE 343 Corollary 3: Let y be an ordered income distribution. Increase the incorne of every person in y by the same absolute amount E > 0, and call the new distribution x. Then x Lorenz-dominates y. Formally, let y: O A< YI < Y2 1< .. Y. and x: O ( (y, + E) < (Y2 + E) A ... .< (Yn + where E > 0. Then x > LY. This follows from lemma 1, since for Y. ~ ~ ~ ~ Y+ Y '+i+l Yi += , . . ., (n-1). Yi Yi + I Therefore increasing everyone's income by the same absolute amount reduces inequality. Put another way, an equal absolute underestimation of incomes leads to an unambiguous overestimation of inequality. It is obvious that the distribution x tends to perfect equality when the absolute amount £ by which everyone's income is increased becomes indefinitely large. (As shown in "The Effect of Changes in Certain Incomes" in appendix B, the Gini coefficient G (x) - 0 as E - OO.) The effects. on inequality of underestimating or overestimating the incomes of particular persons have been discussed. Can anything be said now about the effects on measured inequality of undersampling or oversampling persons at particular income levels? In other words, how does the Lorenz curve shift if persons are added to the existing distribution at various income levels? The general answer to this question is quite unilluminating, but for certain frequently encountered types of undersam- pling l_orenz dominance can be demonstrated. Lemma 2: Let x(r) be an ordered income distribution among r people. Add on a number q of persons at income level zero to x(r), and call the new distribution among the q + r = n people y(n). Then x(r) t Ly(n). Proof: Let x (r): O xl x < .x.. x,. Then y(n): 0YI=Y2 ... YYq(YqI1 Ly(n).11 With the addition of persons with positive income at the bottom end of the distribution x (or indeed anywhere along x), Lorenz dominance cannot be established, despite the fact that when an increasing number of persons (each with the same positive income) is added, the Lorenz curve tends to the diagonal of perfect equality. Lemma 3: The distribution of individuals according to per capita household income z(n) Lorenz-dominates their distribution according to personal income y(n). Proof: z(n) and y(n) have the same number n of individuals. The distribution z(n) can be obtained from the distribution y(n) by a series of rich to poor transfers: redistribute income within each household from members receiving more than its per capita income to members receiving less (which includes members who are zero-income recipients). These are a series of transfers from rich individuals to poor, and hence by Atkinson's theorem z(n) Lorenz-dominates y(n). 11 The Redress of Poverty Rule Given an ordered income distribution y, if an additional amount of income A becomes available for distribution among the population but the existing 2. The previous results which established Lorenz dominance concern a vertical movement of the Lorenz curve. Here, instead of holding the points i/n (the population shares) fixed, I have held the ordinates (income shares) fixed and considered horizontal movements of the abscissas (population shares). It is difficult to prove results about Lorenz dominance when the movement is different from one of these two types. APPENDIX E. LEMMAS ON LORENZ DOMINANCE 345 income of any person cannot be reduced, how should A be distributed to maximize social welfare? For an egalitarian social welfare function, the answer is obvious:3 Give A to the poorest person 1 until his or her income reaches that of person 2. Distribute the remainder equally between them until their incomes reach that of person 3. And so on. Label the resulting distribution x. This distribution policy, with incomes raised from the bottomn upward, is called the redress of poverty rule. The question arises whether the distribution x Lorenz-dominates the original distribution y. This is immediate from the extension of Atkinson's theorem for egalitarian social welfare functions (see appendix D). It can also be shown by the following two-stage procedure. First, increase each person's income by the same absolute amount e = A/n. The resulting distribution Lorenz-dominates y, by corollary 3. Then obtain the distri- bution x from this by a sequence of transfers of the c's from rich persons to poor. By Atkinson's theorem (the principle of transfers), x Lorenz- dominates this distribution, which in turn has been shown to Lorenz- dominate y; hence, by transitivity, x >- y. || The redress of poverty rule is most "efficient" for the redress of inequality too. That is to say, if redistribution is permitted only out of the additional income A, then the maximum reduction in inequality (for any index from the Lorenz class) is secured by the redress of poverty rule. Any other distribution of the incremental income can always be improved upon in the Lorenz sense (as long as it is not x already) by a transfer from rich to poor. Thus x Lorenz-dominates all other distributions of the incremental income A; hence it yields the maximum reduction in inequality for any index from the Lorenz class. 11 3. A social welfare function is egalitarian (see appendix D) ifa transfer of income from rich to poor, which does not reverse their relative ranks, leads to an increase in welfare. F Mapping the Household to the per Capita Household Income Distribution IN THIS APPENDIX, I attempt to show how the distribution of individuals by per capita household income can be obtained from the joint (bivariate) distribution of households by household income and size. The general mathematical mapping between the household and the per capita house- hold income distributions is derived here. Furthermore, for certain special joint distributions of households by household income and size, I obtain analytical expressions for the distribution of individuals by per capita household income. To simplify the analysis, consider first a constant household size within each income class, but allow the size to be different across income classes. This in itself causes the per capita household income distribution to diverge from the household income distribution. I shall later allow a non-zero variance of household size within each income class,' and show that even if average household size is constant across income classes there will be a divergence between the household and the per capita household income distributions. I shall also show that the direction of divergence in terms of inequality cannot generally be predicted, except in some very special cases. If household size is constant within each income class, the following statement is obviously true: If household size is constant across income classes, then the household and per capita household income distributions are identical. The next statement is intuitively plausible, and a variant of it is often invoked in empirical work on income distribution. It is, however, false: I. That is, a positive variance in the household size distribution conditional on each income level. 346 APPENDIX F MAPPING DISTRIBUTIONS 347 If household size is positively correlated with household income, then the disi ribution of individuals by per capita household income is more equal than the distribution of households by household income. The falseness of the second statement is readily demonstrated by counter- example. Let the joint distribution of households by household income and size be as follows: Household size Household income 1 4 (MS per month) 200 1 0 300 o 1 Number of households I I Number of individuals 1 4 Per capita household income (MS per month) 200 75 Household size is positively correlated with household income, but the distribution of households by household income Lorenz-dominates the distribution of individuals by per capita household income; hence it will show less inequality for any index from the Lorenz class.2 The Lorenz curves for the two distributions are shown in figure F-1. The reason the second statement fails in this example is that the ranking of households is reversed in going from household income to per capita household income. If the ranking is preserved under this transformation, it is possible to prove some general results by comparing the two distri- butiorns. I first compare the two distributions when the household is maintained as the population unit in both. Proposition 1: Suppose there are H households with incomes Yh, for h = 1, . . . , H. Let y(H) be the ordered distribution of households by household income; that is, y(H): 0- Ly(H). Proof: It is given that ml ( Ly(H).II Now consider the more general situation in which household size varies within each income class. The relation between the household and per capita household income distributions then becomes more complex. Before examining this in any detail, compare relative inequality in the two distributions with the variance of log-income measure. Using the previous notation of y for household income, m for household size, and z for per capita household income, z = y/m. Therefore log z = log y - log m. Hence, var (log z) = var (logy) + var (log m) - 2cov (logy, log m). Now if cov(logy, log m) < 'var(log m), then var(logz) > var(logy). Thus a positive correlation between logy and log m is not enough to ensure that inequality in z is smaller than inequality in y (measured by the variance of log-income). For this result to follow, the covariance between them must be sufficiently large [ > 'var(log m)]. If in fact log y and log m are uncorrelated, z will be distributed more unequally than y. In this case, deflation of household income by household size is like adding extra variance (or noise).4 4. For instance, a distribution of households according to household income can be made as unequal as one likes according to per capita household income by increasing the variance of household size around a fixed average. APPENDIX F MAPPING DISTRIBUTIONS 351 Note, however, that var (log z) refers to the variance of log-per-capita- household-income across the population of households, not individuals. Since the variance operator above has been applied over the population of households, the results concern inequality in the distribution of households by per capita household income. In fact, it is the distribution of individuals by per capita household income that is of interest. This distribution needs to be derived rather carefully from the joint distribution of households by household income and size. Letf(y, m) be the joint density function of households with household income y and household size m. Then g(y) = Jf(y, m)dm is the (marginal) densitv function of households with household income y, and h(m) Jf (y, m)dy is the (marginal) density function of households with household size m. Hence the total number of individuals in the population is | mh (m)dm. Assuming that household income is equally divided among the members of a household, how is per capita household income z = y/m distributed, where y and m range over all possible values? Let +(z) be the density function of individuals according to per capita household income z, so that 4(z)dz represents the number of individuals with income in the interval [z, z + dz]. Evidently, ¢ (z) = Jmf (mz, m)dm m. Note that dm m represents the inverse of the Jacobian of z = y/m, the transformation used to change variables from y to z. Further,f(mz, m) gives the number of households with characteristics (mz, m), and mf(mz, m) gives the number of individuals with these characteristics. Multiplyingf(mz, m) by m thus represents the step in going from ihe distribution of households according to per capita household income z to the distribution of individuals according to per capita household income z.5 From the distribution +(z), the total number of individuals in the population is J f(z)dz = JJ m2f (mz, m)dm dz. 5. If J(mz, m) had not been multiplied by m, the household (rather than individual) distribution of z would have been obtained. The distribution of households by per capita household income z is given by the density function O(z) where (z) = ff(mz,nm)dm m. The total number of households according to the distribution O(z) is JIO(z)dz = JIf(mz, m)dm mdz. But from the original bivariate distribution f(y, m), the total number of households is Jff(y, m)dy dm = Jff(mz, m)dm-mdz, using y = mz. This is consistent with the total number of households according to the distribution o(z). 352 INEQUALITY AND POVERTY IN MALAYSIA But, from before, fmh(m)dm should also equal the total number of individuals in the population. By definition, h(m) = ff(y, m)dy. Put y = mz, so that dy = mdz (for a given value of m). Then h(m) = f mf(mz, m)dz. Hence, J mh (m)dm = I f m2f(mz, m)dz dm = JO (z)dz. The derivation of + (z) is therefore consistent! The process of deriving + (z) from f(y, m) is illustrated geometrically in figure F-2. I now derive the distribution of individuals by per capita household income in three cases for which the mapping 4)(z) yields an analytical solution. Figure F-2. The Derivation of 4f(z) from f(y,m) fmy,m) Integ,rate over this inefr difrent values of its slope z in the (y, m) m plane. Case 1: Suppose the household income and size distributions are independent, that is,f(y, m) = g(y) h(m). Then + (z) = fm2g(mz) h(m)dm. Under what conditions is +(z) the same distribution as g(y) = Jf(y, m)dm, the distribution of households by household income. If g(mz) is multipli- catively separable as g(mz) = a(m) g(z), then 4(z) =g(z)fm2a(m) h(m)dm = g(z) up to a scalar multiple.6 6. In this case, the distribution of households by per capita household income z also has density function g(z), up to some other scalar multiple. APPENDIX F. MAPPING DISTRIBUTIONS 353 For example, if g(y) is a Pareto distribution with coefficient a, g(y) = Ay-a where A is a constant. Then, g(mz) = A(mz)- = m-Az- = a(m) g(z), and in this case g(mz) is multiplicatively separable. Therefore, if the household income and size distributions are independent, and the house- hold income distribution g(y) is a Pareto distribution with coefficient a, then the distribution of individuals by per capita household income +(z) is also a Pareto distribution, with the same coefficient a. Case 2: Still assuming that the household income and size distributions are independent, that is, f(y, m) = g(y) h(m), suppose g(y) is a negative exponential instead of a Pareto distribution, that is, g(y) = Ae-'Y. Then +(Z) = ')Am2e-mzh(m)dm, where the limits of integration a and b represent the minimum and maximum household sizes, respectively. If h(m) = m-1, and k is an integer, we can integrate by parts to obtain an explicit expression for +(z). In the special case when k = 2, -A'az - kbz (z) [e- _ Z] z For other values of k the expression for +(z) is more complicated, but it is never a negative exponential distribution. Hence, in this case, the distri- butiori of individuals by per capita household income is different from the distribution of households by household income. Case 3: Suppose now that the household income and size distributions are not independent. For instance, let f(y, m) = Ay - h(m). Then +(z) = A Jm2 (mz) - h(m)dm = A Jm2 -mh(m) - e-malosz dm. The value of this integral depends on the form of the function h(m). If h(m) happens to be h(m) = 1/(m2 ), then +b(z) = A ) [e-malog]b A ot logz _ a - X log z 354 INEQUALITY AND POVERTY IN MALAYSIA where a and b are the minimum and maximum household sizes, respectively. The distribution of households by household income, g(y), is in this case given by g(y) = J. f(y, m) dm which, although difficult to integrate, is easily seen to be different from the above distribution 0(z). I have tried to illustrate in this appendix the mapping from the household to the per capita household income distribution. Starting with a joint distribution of households by household income and size, I have shown that quite different distributions of individuals by per capita household income may be obtained depending on the nature of interaction between household income and size. In particular, distributions of individuals by per capita household income can be derived which are more or less equal than the underlying distribution of households by household income. References Adelman, Irma, an,i C. T. Morris. 1973. Economic Growth and Social Equity in Developing Couniries. Stanford, Calif.: Stanford University Press. Adelman, Irma, and Laura D. Tyson. 1974. "A Regional Microeconomic Model of Yugoslavia-Factors Affecting the Distribution of Income in the Short Run." Development Research Center, World Bank, Washington, D.C. Adelman, Irma, and Sherman Robinson. 1977. Income Distribution Policy in Developing Countries: A Case Study of Korea. Stanford, Calif.: Stanford University Press. Ahluvwalia, Montek S. 1974a. "Income Inequality: Some Dimensions of the Problem." In Chenery and others (1974). -.1974b. "The Scope for Policy Intervention." In Chenery and others (1974). -. 1976. "Inequality, Poverty and Development." Journal of Development Economics, vol. 3, pp. 307-42. -. 1977. "Rural Poverty and Agricultural Performance in India." Journal of Development Studies, pp. 298-323. Ahluwalia, Montek S., and others. 1976. "A Social Accounting Matrix for Malaysia." Development Research Center, World Bank, Washington, D.C. Aitchison, J., and J. A. C. Brown. 1957. The Lognormal Distribution. Cambridge, Eng.: Cambridge University Press. Alatas, Syed Hussein. 1977. The Myth of the Lazy Native. London: Frank Cass. Anand, Sudhir. 1970. "An Optimal Earnings Schedule." Paper presented to the Second World Congress of the Econometric Society, Cambridge, England. -. 1971. "Rural-Urban Migration in India: An Econometric Study." Harvard University, Cambridge, Mass. . 1973. "The Size Distribution of Income in Malaysia, Part I." Development Research Center, World Bank, Washington, D.C. .1974a. "Addendum to the Size Distribution of Income in Malaysia, Part I." Development Research Center, World Bank, Washington, D.C. . 1974b. "The Size Distribution of Income in Malaysia, Part II." Oxford: St. Catherine's College. .1975. "Aspects of Poverty in Malaysia." Paper presented to the Fourteenth General Conference of the International Association for Research in Income and Wealth, Aulanko, Finland. 3S5 356 INEQUALITY AND POVERTY IN MALAYSIA . 1976a. "The Little-Mirrlees Appraisal of a Highway Project in Malaysia." Journal of Transport Economics and Policy, vol. 10, no. 3, September. . 1976b. "Discussion on 'Personal Characteristics and Income' by J. Tinbergen." In The Personal Distribution of Incomes. Anthony B. Atkinson, ed. London: Allen and Unwin. Anand, Sudhir, and S. M. Kanbur. 1978. "Inequality and Development: A Reconsideration." Oxford: St. Catherine's College. Andic, F. M. 1975. "The Impact of the Tax System on the Poor: With Special Reference to Rural Poor." Kuala Lumpur: Economic Planning Unit. Arles, J-P. 1971. "Ethnic and Socio-Economic Patterns in Malaysia." International Labour Review, vol. 104, no. 6, December. Atkinson, Anthony B. 1970. "On the Measurement of Inequality." Journal of Economic Theory, vol. 2. no. 3, September. . 1975. The Economics of Inequality. Oxford: Clarendon Press. Awang Had Salleh. 1975. "Translation of National Agenda into Educational Strategies: A Proposal for Malaysia." Paper presented at the Second Malaysian Economic Convention, Kuala Lumpur. Aziz, Ungku A. 1964. "Poverty and Rural Development in Malaysia." kajian Ekonomi Malaysia, vol. 1, no. 1, June. . 1965. "Poverty, Proteins and Disguised Starvation." Kajian Ekonomi Malaysia, vol. 1, no. 1, June. . 1975a. "Footprints on the Sands of Time: The Malay Poverty Concept over Fifty Years from Za'ba to Aziz and the Second Malaysia Five Year Plan." In Chee and Khoo (1975). . 1975b. "Recent Thoughts on Poverty." Paper presented at the Second Malaysian Economic Convention, Kuala Lumpur. Balassa, Bela. 1977. "The Income Distribution Parameter in Project Appraisal." In Economic Progress, Private Values, and Public Policy. B. Balassa and R. Nelson, eds. Amsterdam: North-Holland. Barlow, C., and C-K. Chan. 1968. "Towards an Optimum Size of Rubber Holding." Paper presented to National Rubber Conference, Kuala Lumpur. Ben-Porath, Y. 1967. "The Production of Human Capital and the Life-Cycle of Earnings." Journal of Political Economy, vol. 75, no. 4, pt. 1, August. Bhalla, Surjit S. 1973. "The Education-Income Connection-An Investigative Report." Research Program in Economic Development Discussion Paper no. 38. Princeton University, Princeton, N. J. Bliss, Christopher J., and Nicholas H. Stern. 1976. "Economic Aspects of the Connection between Productivity and Consumption." University of Essex, Discussion Paper no. 67. Blitzer, Charles R., Peter B. Clark, and Lance Taylor, eds. 1975. Economy-Wide Models and Development Planning. New York: Oxford University Press. Brown, C. 1973. "Rice Price Stabilization and Support in Malaysia." Developing Economies, vol. 11, no. 2, June. REFERENCES 357 Bruno, Michael. 1977. "Distributional Issues in Development Planning-Some Reflections on the State of the Art." Paper presented at the Workshop on Analysis of Distributional Issues in Development Planning, Bellagio, Italy. Chaim, B. N. 1975. "Class and Communal Conflict in Malaysia." Journal of Contemporary Asia, vol. 5, no. 4. Chandra Muzaffar. 1977. "Some Political Perspectives on the New Economic Policy." Paper presented at the Fourth Malaysian Economic Convention, Kuala Lumpur. Chee, S., and S. M. Khoo, eds. 1975. Malaysian Economic Development and Policies. Kuala Lumpur: Malaysian Economic Association. Chenery, Hollis B., Montek S. Ahluwalia, C. L. G. Bell, John H. Duloy, and Richard Jolly. 1974. Redistribution with Growth. London: Oxford University Press. Chew, A. F. Y. 1968. "Index Numbers in West Malaysia." M. A. thesis, Division of Statistics, Faculty of Economics and Administration, University of Malaya. Chisvvick, B. R. 1975. Income Inequality: Regional Analyses within a Human Capital Framework. New York: National Bureau of Economic Research. Chiswick, B. R., and J. Mincer. 1972. "Time-Series Changes in Personal Income Inequality in the United States from 1939, with Projections to 1985." Journal oJ Political Economy, vol. 80, supplement, May/June. Cohen, M., T. Nagel, and T. Scanlon, eds. 1977. Equality and Preferential Treatment. Princeton, N. J.: Princeton University Press. Corden, W. M. 1971. The Theory of Protection. Oxford: Clarendon Press. Dandekar, V. M., and N. Rath. 1971. Poverty in India. Poona: Indian School of Political Economy. Dasgupta, Partha S., Amartya K. Sen, and David Starrett. 1973. "Notes on the Measurement of Inequality." Journal of Economic Theory, vol. 6, no. 2, April. Department of Inland Revenue. 1974. Annual Report of the Department of Inland Revenue, 1970. Kuala Lumpur: Government Press. Department of Social Welfare. 1975. "A Joint State and Federal Govern- ment Public Assistance Programme." Kuala Lumpur: Ministry of Welfare Services. -. 1976. "Public Assistance as a State/Federal Government Responsibility." Kuala Lumpur: Ministry of Welfare Services. Department of Statistics. [1957?]. "Course of Lectures for Training Investigators" [for the Householdi Budget Survey]. Kuala Lumpur, undated. Cited in text as Course oJ Lectures. - [1961?]. Household Budget Survey of the Federation of Malaya 1957/58: Report. Kuala Lumpur, undated. Cited in text as HBS Report. 1968. Dictionary of Occupational Classification. Kuala Lumpur. 1970a. PES 1970 Instructions to Field Enumerators (Re-listing oJ Households-PES Form 2). Instruction Booklet 2. Kuala Lumpur. . 1970b. PES 1970 Instructions to Field Interviewers (Demographic, Family 358 INEQUALITY AND POVERTY IN MALAYSIA Planning and Income Particulars-PES Form 3). Instruction Booklet 3. Kuala Lumpur. - 1970c. PES 1970 Instructions to Field Supervisors. Instruction Booklet 4. Kuala Lumpur. - 1970d. PES 1970 General Instructions (Including Financial Instructions) to Field Enumerators and Interviewers. Instruction Booklet 5. Kuala Lumpur. - 1970e. PES 1970 Sampling Instructions. Instruction Booklet 6. Kuala Lumpur. - 1970f. PES Form 3 (Post-Enumeration Survey Questionnaire Booklet). Kuala Lumpur. . 1970g. Socio-Economic Sample Survey of Households-Malaysia 1967/68. Kuala Lumpur. . 1971a. Index of Occupations. Kuala Lumpur. - 1971b. Malaysian Industrial Classification. Kuala Lumpur. . 1972. Rubber Statistics Handbook 1970. Kuala Lumpur. 1973. PES Coding Manual (Post-Enumeration Survey 1970-71 "Code Book"). Kuala Lumpur. de Wulf, L. 1975. "Fiscal Incidence Studies in Developing Countries: Survey and Critique." IMF Staif Papers, vol. 22, no. 1, March. Economic Planning Unit (EPU). 1975. "Survey of Rural Poverty." Kuala Lumpur. Food and Agriculture Organization (FAO)/World Bank. 1975. "Problems of Rural Poverty in Malaysia." World Bank Report no. 838-MA. Washington, D.C. Restricted circulation Fei, John C. H., and Gustav Ranis. 1974. "Income Inequality by Additive Factor Components." Economic Growth Center, Yale University. Fell, H. A., and others. 1959. Report on Proposalsfor New Retail Price Indices. Kuala Lumpur: Special Advisory Committee on Cost of Living Indices, Government Press. Fisheries Division. 1971. "An Economic Survey of the Fishing Communities on the East Coast of Malaysia." Kuala Lumpur: Ministry of Agriculture and Lands. Fishlow, Albert. 1972. "Brazilian Size Distribution of Income." American Economic Review, vol. 62, May. - 1973. "Brazilian Income Size Distribution-Another Look." Berkeley: University of California. Goldman, R. 1975. "Staple Food Self-Sufficiency and the Distributive Impact of Malaysian Rice Policy." Food Research Institute Studies (Stanford, Calif.), vol. 14, no. 3. Government of Malaysia. 1971. Second Malaysia Plan 1971-1975. Kuala Lumpur: Government Press. Referred to as SMP in the text. . 1972. Parliameniary Debates on the Constitution Amendment Bill, 1971. Kuala Lumpur: Government Press. REFERENCES 359 -. 1973. Mid-Term Review of the Second Malaysia Plan, 1971-1975. Kuala Lumpur: Government Press. Referred to as MTR in the text. -. 1974. Estimates of Federal Revenue and Expenditure, 1972. Kuala Lumpur. -. 1976. Third Malaysia Plan, 1976-1980. Kuala Lumpur: Government Press. Referred to as TMP in the text. Grant, W. R. 1970. "Application of an Economic Model for Evaluating Government Program Cost for Rice." American Journal of Agricultural Economics, vol. 52, no. 2, May. Harris, J. R., and M. P. Todaro. 1970. "Migration, Unemployment and Development: A Two-Sector Analysis." American Economic Review, vol. 60, March. Hirschman, Charles 1972. "Ethnic Stratification in West Malaysia." Ph.D. dissertation, Department of Sociology, University of Wisconsin. Ann Arbor, Mich. University Microfilms, Inc. -. 1973. "The Determinants of Ethnic Inequality in Peninsular Malaysia." Paper presented at the annual meeting of the Rural Sociological Society, College Park, Maryland. -. 1974. "Economic Progress in Malaysia: How Widely Has It Been Shared?" UMBC Economic Review (Kuala Lumpur), vol. 10. Hoerr. 0. D. 1973. "Education, Income and Equity in Malaysia." Economic Development and Cultural Change, vol. 21, no. 2, January. Reprinted in Lim (1975). Huang, Y. 1975. "Tenancy Patterns, Productivity, and Rentals in Malaysia." Economic Development and Cultural Change, vol. 23, no. 4, July. Jackson, R. M. 1961. Immigrant Labour and the Development of Malaya: 1786- 192G'. Kuala Lumpur: Government Printer. Jain, Shail. 1975. Size Distribution of Income: A Compilation of Data. Baltimore, Md.: Johns Hopkins University Press. Kakwani, N. C., and N. Podder. 1973. "On the Estimation of Lorenz Curves from Grouped Observations." International Economic Review, vol. 14, no. 2, June. . 1976. "Efficient Estimation of the Lorenz Curve and Associated In- equality Measures from Grouped Observations." Econometrica, vol. 44, no. 1, January. Kamal, Salih. 1975. "Urban Strategy, Regional Development, and the New Economic Policy." Paper presented at the Second Malaysian Economic Convention, Kuala Lumpur. Kendall, M. G., and A. Stuart. 1963. The Advanced Theory of Statistics. Vol. 1, Distribution Theory. 2d ed. London: Griffin. Khoo, Khay Huat. 1976. "Optimal Utilization and Management of Fishery Resources." Kajian Ekonomi Malaysia, vol. 13, nos. I and 2, June/December. Kolm, S. Ch. 1969. "The Optimal Production of Social Justice." In Public 360 INEQUALITY AND POVERTY IN MALAYSIA Economics. J. Margolis and H. Guitton, eds. London: Macmillan. Kuznets, Simon. 1955. "Economic Growth and Income Inequality." American Economic Review, vol. 45, no. 1, March. .1963. "Quantitative Aspects of Economic Growth of Nations: Distribution of Income by Size." Economic Development and Cultural Change, vol. 11, no. 2, pt. 2, January. - . 1976. "Demographic Aspects of the Size Distribution of Income: An Exploratory Essay." Economic Development and Cultural Change, vol. 25, no. 1, October. Lamb, A. 1964. "Early History." In Malaysia: A Survey. Wang Gangwu, ed. London: Pall Mall Press. Lee, E. L. H. 1975. "Income Distribution in a Developing Economy: A Case Study of West Malaysia." D.Phil. dissertation, Oxford University. Lim, D., ed. 1975. Readings on Malaysian Economic Development. Kuala Lumpur: Oxford University Press. Lim, Ln Lean. 1974. The Pattern of Income Distribution in West Malaysia, 1957- 1970. ILO Working Paper. Geneva: International Labour Office. Lipton, M. 1968. "Strategy for Agriculture: Urban Bias and Rural Planning." In The Crisis of Indian Planning. P. P. Streeten and M. Lipton, eds. London: Oxford University Press. - . 1977. Why Poor People Stay Poor: Urban Bias in World Development. London: Temple Smith. Low, P., ed. 1971. Trends in Southeast Asia. No. 2. Proceedings and Background Paper of Seminar on Trends in Malaysia. Singapore: Institute of Southeast Asian Studies. Marglin, S. A. 1967. Public Investment Criteria. London: Allen and Unwin. Mazumdar, Dipak, assisted by P. Videt-Yontrakich. 1975. "Education and Employment in Urban Malaysia." Development Economics Department, World Bank, Washington, D.C. McNamara, Robert S. 1972. Address to the Board of Governors. Washington, D.C.: World Bank. Reprinted in The McNamara Years at the World Bank.: Major Policy Addresses of Robert S. McNamara, 1968-1981. Baltimore, Md.: Johns Hopkins University Press, 1981. 1973. Address to the Board of Governors. Washington, D.C.: World Bank. Reprinted in The McNamara Years at the World Bank: Major Policy Addresses of Robert S. McNamara, 1968-1981. Baltimore, Md.: Johns Hopkins University Press, 1981. - . 1975. Address to the Board of Governors. Washington, D.C.: World Bank. Reprinted in The McNamara Years at the World Bank: Major Policy Addresses of Robert S. McNamara, 1968-1981. Baltimore, Md.: Johns Hopkins University Press, 1981. Milne, R. S. 1976. "The Politics of Malaysia's New Economic Policy." Pacific Afjairs, vol. 49, no. 2, Summer. REFERENCES 361 Mincer, J. 1970. "The Distribution of Labor Incomes: A Survey with Special Reference to the Human Capital Approach." Journal oJ Economic Literature, vol. 8, no. 1, March. --. 1974. Schooling, Experience and Earnings. New York: National Bureau of Economic Research. --. 1976. "Progress in Human Capital Analyses of the Distribution of Earnings." In The Personal Distribution of Incomes. Anthony B. Atkinson, ed. London: Allen and Unwin. Ministry of Labour and Manpower. 1972. Handbook of Labour Statistics. Kuala Lumpur. Mirrlees, James A. 1971. "An Exploration in the Theory of Optimum Income Taxation." Review of Economic Studies, vol. 38, April. Moore, Basil J. 1975. "Restructuring Society." School of Comparative Social Sciences, Universiti Sains Malaysia, Penang. Morley, S. A., and J. G. Williamson. 1973. "The Impact of Demand on Labor Absorption and the Distribution of Earnings: The Case of Brazil." Paper no. 39. Program of Development Studies, Rice University. Morrison, I. 1949. 'Aspects of the Racial Problem in Malaya." Pacifc Afjairs, no. 22, September. Mue,llbauer, John 1976. "Can We Base Comparisons of Welfare between Households on B3ehaviour?" Birkbeck College, London. Narkswasdi, U., and S. Selvadurai. 1968. Economic Survey of Padi Production in West Malaysia. Kuala Lumpur: Ministry of Agriculture and Co-Operatives. Organisation for Economic Co-operation and Development (OECD). 1976. Public Expenditure on Income Maintenance Programmes. Paris. Orshansky, M. 1965. "Counting the Poor: Another Look at the Poverty Profile." Social Security Bulletin, vol. 28. Oshirna, Harry T. 1970. "Income Inequality and Economic Growth: The Postwar Experience of Asian Countries."Malayan Economic Review, vol. 15, no. 12, October. Palan, V. T. 1968. "Problems of Questionnaire-Design with Particular Reference to Household Surveys." Kajian Ekonomi Malaysia, vol. 5, no. 1, June. Paukert, Felix. 1973. "Income Distribution at Different Levels of Development: A Survey of Evidence." International Labour Review, vol. 108, nos. 2-3, August- September. Peiris, Denzil. 1975. "The Emerging Rural Revolution." Far Eastern Economic Review (Hong Kong), January 10, 1975. Psacharapoulos, G. 1973. Returns to Education-An International Comparison. San Francisco: Jossey-Bass/Elsevier International Series. Purcal, J. T. 1971. Rice Economy: A Case Study of Four Villages in West Malaysia. Kuala Lumpur: University of Malaya Press. Purcell, V. 1967. The Chinese in Malaya. Kuala Lumpur: Oxford University Press. Ist ed., 1948. 362 INEQUALITY AND POVERTY IN MALAYSIA Puthucheary, J. 1977. "Changes in Ownership and Control in Malaysia." Paper presented at the Fourth Malaysian Economic Convention, Kuala Lumpur. Pyatt, Graham. 1976. "On the Interpretation and Disaggregation of Gini Coefficients." Economic Journal, vol. 86, no. 342, June. Pyatt, Graham, and Jeffrey 1. Round, assisted by Jane Denes. 1977. "The Distribution of Income and Social Accounts: A Study of Malaysia in 1970." Development Research Center, World Bank, Washington, D.C. Rafferty, Kevin. 1975a. "Clouds Over Eden's Horizon." Financial Times (London), February 24, 1975. . 1975b. "Pitfalls of the New Economic Policy." Financial Times (London), February 24, 1975. Rajakumar, M. K. 1977. "Conflict Resolution and National Progress." Paper presented at the Fourth Malaysian Economic Convention, Kuala Lumpur. Rao, V. M. 1969. "Two Decompositions of Concentration Ratio." Journal of the Royal Statistical Society, ser. A, vol. 132, pt. 3. Rawls, John. 1971. A Theory of Justice. Cambridge, Mass.: Harvard University Press. Razak, Abdul. 1971. "With One Mind toward One Objective." Opening address to the MCA general assembly, Kuala Lumpur, August 21, 1971. Robless, C. L. 1975a. "Some Notes on the Feasibility and Consistency of the New Economic Policy." In Chee and Khoo (1975). . 1975b. "Policy Analysis and Development Planning in Malaysia." Kuala Lumpur: Economic Planning Unit, Prime Minister's Department. Roff, W. R. 1967. The Origins of Malay Nationalism. Singapore: University of Malaya Press. Ross-Larson, Bruce, ed. 1977. Issues in Contemporary Malaysia. Kuala Lumpur: Heinemann Educational Books (Asia) Ltd. Rothschild, Michael, and Joseph E. Stiglitz. 1973. "Some Further Results on the Measurement of Inequality." Journal of Economic Theory, vol. 6, no. 2, April. Rowntree, B. S. 1922. Poverty: A Study of Town Life. New ed. London: Longmans, Green. Sandhu, K. S. 1967. "Indians in the Modern Malayan Economy." India Quarterly (New Delhi), vol. 23, April/June. . 1969. Indians in Malaya: Immigration and Settlement, 1786-1957. Cambridge, Eng.: Cambridge University Press. Sawyer, Malcolm. 1976. "Income Distribution in OECD Countries." OECD Economic Outlook, no. 19, July. Schutz, R. R. 1951. "On the Measurement of Income Inequality." American Economic Review, vol. 41, no. 1, March. Selvadurai, S. 1968. A Preliminary Report on the Survey of Coconut Smallholdings in West Malaysia. Kuala Lumpur: Ministry of Agriculture and Co-operatives. - . 1972a. Padi Farming in West Malaysia. Kuala Lumpur: Kementerian Pertanian dan Perikanan. REFERENCES 363 1972b. Padi Survey: Sungei Manik and Changkat Jong Irrigation Areas. Perak. Kuala Lumpur: Ministry of Agriculture and Fisheries. --. 1972c. KIrian Padi Survey. Kuala Lumpur: Ministry of Agriculture and Fisheries. I. 1972d. Socio-Economic Survey oJ Rubber SmalIholdings in West Johore. Kuala Lumpur: Ministry of Agriculture and Fisheries. - -. 1975. Padi Survey: Sebarang Perak (Stage I). Kuala Lumpur: Ministry of Agriculture and Rural Development. Selvadurai, S., and Ani b. Arope. 1969. Socio-Economic Study of Padi Farms in Province Wellesley 1968. Kuala Lumpur: Ministry of Agriculture and Co- operatives. Selvadurai, S., Ani b. Arope, and Nik Hassan b. Mohammad. 1969. Socio- Economic Study of the Padi Farms in the Kemubu Area oj Kelantan 1968. Kuala Lumpur: Ministry of Agriculture and Co-operatives. Selvaratnam, V. 1975. "Some Aspects of Race and Class in Colonial and Post- Ccolonial Peninsular Malaysia." Institute of Development Studies, University of Sussex. Sen, Amartya K. 1973a. On Economic Inequality. Oxford: Clarendon Press. -. 1973b. "Poverty, Inequality and Unemployment: Some Conceptual Issues in Measurement." Economic and Political Weekly, Special Number, August. -. 1976a. "Poverty: An Ordinal Approach to Measurement." Econometrica, vol. 44, no. 2, March. -. 1976b. "Real National Income." Review oJ Economic Studies, vol. 43. Shaharuddin bin Flaron. 1977. "Restructuring Ownership and Control: Some Recollections on Policy." Paper presented at the Fourth Malaysian Economic Convention, Kuala Lumpur. Shaplen, Robert. 1977. "Letter from Malaysia." New Yorker, April 18, 1977. Snodgrass, Donald R. 1975. "Trends and Patterns in Malaysian Income Distribution, 1957-70." In Lim (1975). Squire, Lyn, and Herman G. van der Tak. 1975. Economic Analysis of Projects. Baltimore, Md.: Johns Hopkins University Press. Stenson, Michael. 1976. "Class and Race in West Malaysia." Bulletin oJ Concerned Asian Scholars, April-June. Stern, Nicholas H. 1977. "Welfare Weights and the Elasticity of the Marginal Valuation of Income." In Current Economic Problems. M. Artis and R. Nobay, eds. Oxford: Basil Blackwell. Stiglitz, Joseph E. 1975. "The Theory of Screening, Education, and the Distribution of Income." American Economic Review, vol. 65, no. 3, June. Suffian, Mohamed bin Hashim. 1972. An Introduction to the Constitution of Malaysia. Kuala l_umpur: Government Press. Tan, 13. T. 1975. "Agricultural Subsidy and Price Support Policies." Paper presented at the Second Malaysian Economic Convention, Kuala Lumpur. 364 INEQUALITY AND POVERTY IN MALAYSIA Tan, Siew Sin. 1971. "With One Mind toward One Objective." Presidential address to the MCA general assembly, Kuala Lumpur, August 21, 1971. Theil, Henri. 1967. Economics and Jn.frmalion Theory. Amsterdam: North- Holland. Thillainathan, R. 1970. "The Second Malaysia Plan-Notes on the Objectives of Balanced Distribution of Wealth and Employment." Kajian Ekonomi Malaysia, vol. 7, no. 2, December. - 1975a. "The Public Enterprise as an Instrument for Restructuring Society: The Malaysian Case." In Chee and Khoo (1975). - . 1975b. "Planning for Economic Equality and the Role of the Public Sector: The West Malaysian Case." In Lim (1975). - .1 975c. "Distributional Issues and Policies in Malaysia-A Review." Paper presented at the Second Malaysian Economic Convention, Kuala Lumpur. - . 1976. "An Analysis of the Effects of Policies for the Redistribution of Income and Wealth in West Malaysia, 1957-75." Ph.D. dissertation, London School of Economics and Political Science. Thorbecke, E., and J. K. Sengupta. 1972. "A Consistency Framework for Employment, Output and Income Distribution: Projections Applied to Colombia." Development Research Center, World Bank, Washington, D.C. Todaro, Michael P. 1969. "A Model of Labour Migration and Urban Unemployment." American Economic Review, vol. 59, March. Universiti Sains Malaysia. 1972. "Fishermen and Fisheries Development in Penang and Kedah." Penang. von der Mehden, F. R. 1975. "Communalism, Industrial Policy and Income Distribution in Malaysia." Asian Survey, vol. 15, no. 3, March. von Vorys, Karl. 1975. Democracy without Consensus. Princeton, N. J.: Princeton University Press. Welsskoff, R. 1973. "A Multi-Sector Simulation Model of Employment, Growth and Income Distribution in Puerto Rico: A Re-Evaluation of Successful Development Strategy." Economic Growth Center, Yale University, New Haven, Conn. World Bank. 1975. Rural Development. Sector Policy Paper. Washington, D.C. Zagier, D. B. 1977. "On the Decomposability of the Gini Coefficient and Other Indices of Inequality." Unpublished manuscript. Index Age, 279, depreciation of human capital household income distribution from. 30- with, 239; earnings function and, 265-67; 34 education and earnings and. 255-57; Commodity market intervention, 290-92 human capital model and, 237; personal Constitution, 7 income inequality and, 229; poverty and, Consumption, 291; economies of scale neg- 132, 151; urban poverty and, 186 lected in, 67 nl4, farmer's own (of pro- Age-income profile, 64, 248, 249, 254, 265 duce), 29,43, 49. 50; welfare levels and. 63, Agriculture. 11, 172, 276, 295; ethnic com- 64 position and, 3, 4; poverty measures and, Corporate ownership. II - 12 131, 157-64. 167; racial breakdown of Council of Trust for the Indigenous People modern, 211-14. See also Farmers (MARA), 13 Ahluwalia, Montek S., 39, 97, 126 n25 Course oJ Leciures (Department of Alliance (political party). 8, 9 Statistics), 44, 47, 48, 49 Anand, Sudhir. 97, 111 n I. 135 n37, 142 n39, 152 Andic, F M.. 286 nl9, 287 Data: assessing reliability of PES income, 22- Ani B. Arope, 153 n14 24; coded personal income distribution, Atkinson index, 82-86. 89, 90 n38, 92, 103 192; coded PES income, 30: estimating n56, 122 n19, 192, 195, 197, 202 n19, 274, household income distribution from 310 n9, 333, 334, 345 coded income, 30-34, fits obtained by, Aziz, Ungku A., 112-13, 152 nlO 248: income distribution. 40, 41, 46, 51: income (PiEs. problems with), 243-47; national accounts, 38 n20: noncompar- Baling disturbances (1974), 7 nl2 ability (between countries) of, 40; PES Barisan Nasional (National Front political (1970) as base for study, 21-22, 271, 273; party), 9 study, I nI, 4 n9 Barlow, C .,... 161, 162 Decomposition: of Gini coefficient, 319-26, Ben-Porath, Y., 239 incidence-of-poverty measure and, 126, Between-group component (racial) See 127; of log-income variance, 330-31, Ethnic composition, between-group methodology of inequality, 86-92; of pov- component erty, 275; racial inequality and, 99-101; Budgets (food and nozifood), poverty line regional inequality and, 97, rural-urban definition and, 114-17 inequality and, 99; strictly decomposable indices and, 88. 102. 274. of Thell sccond Capital, European (in Malaysia). 4 measure L, 329 -30; Theil Tindex and, 93. Census of Population and Housing (1970), 327-29, weakly decomposable indices 22, 24, 25 and. 89-92, 322 Chan. C-K., 161, 162 Decomposition of personal income distri- Chew, A.. F. Y., 46 n32 bution. 277: inequality measures and, Children. 67 nl3, 132, 151, 276 198-202; multivariate, 227-29; occu- Chinese See Ethnic composition pational category and, 216-26; race and. Chiswick, B. R, 239 n2. 261 n43 202-03; region and, 203-06: sector of Coconut subsector, 167, 214, 287 n22 employment and, 226-27. sex of income Coding of PES income data. 30: estimating recipient and, 206-07 365 366 INDEX Degree (educational), eamings and, 259-61 and, 214; ownership of fixed assets and, I I Democratic Action party. 8 n17; rubber, 160 163-64 Department of Statistics, 22, 42, 115, 117 Ethnic composition: Atkinson index of in- Dependency ratio, 72. i51, 186 equality and, 86; between-group com- Developing countries: income inequality com- ponent and, 86-92, 96. 97, 102, 198, 200- parisons and. 40. 41; income inequality 01, 221, 222. 223. 229. 279, 329; 330. 331; factors and, 227. inequality and. 21 education estimates and earnings and, Development. 21 n ]; policy for rural. 292-93 247, 249-5 i, 252: employment restructur- de Wulf. L., 287 n24 ing and, 295-98: food budget analysis Diploma, earnings and. 259-61 and. 115: government efforts to balance, Disaggregation of tncome (Gini coefficient), 15-17; -ts household choice and. 45-46; 318-19 HoBs and PES comparison and. 52: Doctors, earnings of. 259, 261 household income and, 37-38, 72, 76. 82: incidence of poverty and, 125-26, 127, 133, 146-50, 153-56, 160; interracial in- Earnings function, 280, age and, 255-57, equality and, 93-96. 97-99, 273-74, 277; 265-67; age-income profile and. 248. 249, language in schooling and earnings and, 254. 265: degree type and, 259-61; estim- 257-59: New Economic Policy and, 10- ation of (by race, sex, occupation). 249- 14. 272; overview of. 1-4. personal income 53;humancapital,238-41,humancapital distribution (racial disparity) and, 197- model and. 237-38; language used in 201, 202-03. 206-07, 211, 214, 215-16. schooling and, 257-59: properties of, 217-23, 226-27, 228-29; political activity 264-67; regional breakdown of, 261-64; and, 6-9; restructuring of'society and. 10, return to education and. 241-43 11-14, 15-17, 101-03. 298-301: of rich Economic Planning Unit (EPU), 152. 161, households. 142; rural household income 162, 164, 165, 167, 286 and, 43; urban-rural inequality and, 100; Economic welfare, PES income as measure of, within-group component and. 86-92, 96, 63-65 See also Social welfare; Welfare 97. 102, 200. 228-29, 329, 330, 331 comparisons Expenditures, 49, 52. consumption, 63 Education, 279, age cohort and earnings Exports Malaysian economy and. 4-5: and, 255- 57; age-income profile and, 248, taxes on, 286. 287: tax on rubber, 162, 249, 254; degree type and earnings and, 288-89 259-61: estimates of earnings by occu- pation and, 251-53. estimates of earnings by race and. 247, 249-51, estimates of Farmers: personal income and race of, 217, earnings by sex and, 253-54; ethnic quotas 220-21; PES income definition and, 28-29; and. 7-8. 294: human capital model and, poverty measures and, 131, 135, 146, 157- 237-38; language of instruction and earn- 64, 167. 276; subsistence, 39, 49-50 ings and, 257-59; personal income in- Federal Land Development Authority equality and. 229; poverty measures and, (FELDA). 156. 165 131-32, 173, 186; rate of return to. 241 - Federation of Malaya Agreement (February 43: rich households and, 142-43: rural 1948), 7 poverty and, 276 Fei. John C. H.. 315- 16 Employment: ethnic composition of. 3: New Fell, H. A.. 46. 50 Economic Policy and. 10, 1, 13-14; non- First Malaysia Plan (1966-70), 6 agricultural, 160; personal income in- Fishermen, 165-66, 276 equality and. 207-14; personal income Food budget, poverty iine definition and, inequality and sector of, 226-27; poverty 114-16. 117 measures and. 131, 135. 150, 172; re- structuring of. 294-98, rich households Gerakan Rakyat Malaysia (political party), 8 and, 142. See also Self-employed indi- Gini coefficient, 88, 93, 96 n43, 97. 273, 274, viduals; Underemployment; Unemploy- 299, 306; decomposition of, 319-26; de- ment; Urban employees fined, 34, 311-26; as index based on Engineers, earnings of. 259, 261 Lorenz diagram. 304-05; personal income Enumeration blocks (PEs), 24-25 calculation and, 195, Sen's poverty Estates. 157; HBS and. 48. modern agricuiture measure and, 119, among states, 97 INDEX 367 Goodman, Roe, 26 n6 inequality and, 80, 81, poverty and. 132, Goodness-of-fit, equations and, 248 151: urban poverty and, 186 Government: employment policy in public Houses HBS and Pi-S compared and, 52, HBS sector and, 220; minimum income and, sample and, 45, 50; PES income definition 281; poverty policy and. 10-11, 14-15, and rental of, 29 public assistance and, 283, 285, replanting Huang, Y , 156 n18 (rubber) program of, 161-62, 163; re- Humancapital. 100;earnings unction, 238- structuring of society and, 9-14, 15-17 41: income and theory of, 237-38, 265: Gross national product (GNP): growth of, 12 traditional sector employees and model of, n20; increase in, 6; savings rate and, 6 251-52 Hariclas, Mr.(MinistiyofAgriculture),64n4 Import duties, 286. 287 Hiring: of outside farm labor, 157 n23; Incidence-of-poverty measure, 126, 127, 275; preferential (of Malays), 6, 220, 250, 300 household composition and, 132; large Hirschman, Charles. 42 urban households and, 186; rural sector Hoer-, 0. D., 242 nlO and, 146 See also Poverty Household Budget Survey (HBS, 1957-58), Income: coding Of PES data on, 30, data in 21, 22, 42, 273, compared with PES, 51-53; PES, 22-23, 243-47, 273; definition ofHBs, definition of income and. 47-51; sample 47-51; definition of PES, 27-29. 272, econ- design of, 44-47 omic welfare and PES, 63-65; equally Household Expenditure Survey (HEs, 1973), distributedequivalent,83-85,92, 120-21, 286, poverty budget analysis and, 116 194. 196, 202 n19; estimating household Household heads: PES interviews and, 26; income distribution from PES coded data poverty measures and, 131, 135, 146, 150, on, 30-34; HBS concept of, 44; MSSH 151, 168, rich, 142-43; self-employed, 283 concept of, 43; human capital model and, Household income: distribution of house- 237-38; per capita household, 66, 67, 77, holcis by. 65; distribution of households by 79-82; poverty measures and, 118; study per capita, 66, 69, 77, 79-82; distribution choice of concept of, 65-67; surveys of households by household size and, 67- and definition of, 41, 42; underesti- 77; estimating distribution (PES) of, 30-34. mation of, 342, 343; underreporting of, 39. 272--73; HBS and distribution of, 50-51; See also Household income; Income inequality and, 35-38; inequality in per distribution; Personal income distri- capih a, 188-92; international comparisons bution and, 40-41; mapping the household to the Income distribution. data in PES, 22-23; per capita, 346-54; MSSH and, 43; PES and estimating household (coded PES data). national accounts comparison and, 38- 30-34; government efforts and, 14-15; 39; poverty and. 133; poverty definition HBs and household, 50-51, HBs and PES and, 114 comparison and. 51-52, household Households: as basic income-sharing unit, income inequality and, 35-38; inter- 66-67; definition of (survey comparisons) national comparisons of inequality and 41; distribution of(by household income), household, 40-41; PES and national ac- 65; distribution of (by household income counts comparisons and household, 38- and size), 67-77; distribution of (by per 39. population unit choice and, 65; pov- capita household income), 66, 67, 77, 79- erty measures and, 118-23 82; HBs and definition of, 44-45; HBS and Income inequality. Atkinson index and. 82- rural and urban, 45-46; incidence of pov- 86; decomposition of (methodology), 86- erty and, 132, 276; income distribution 92; distribution of individuals by per estimation (PES coded income data) for, capita household income and, 81-82; 30-34; mapping to per capita household household, 35-38; household income and income distribution and, 346-54; PES in- per capita household income distribution come definition and farm. 28-29; PES and, 79-81; internationalcomparisonsof, sample definition of, 24; rEs check on under- 40-41; intertemporal comparisons of, 39, coverage of, 26; poverty and large urban, 42-53; Lorenz diagram and indices of, 186, 276; rich, 135-43; urban farm, 172 303-10; personal income and, 192-98; Household size: distribution of households policy analysis and, 101-03; racial dis- by, 67-77; food budget analysis and, 116; parity and. 197-201, 202-03. 206-07, 368 INDEX Income inequality (continued) capita household income, 79-80: in- 211, 214, 215-16, 217-23, 226-27, 228- equality in distribution of incomes and. 29; rural-urban, 99-101 81-82: inequality indices based on, 303- Indians. See Ethnic composition 10; inequality in personal income calcu- Individual per capita household income dis- lation and, 191-92. 195: lack of welfare tribution. 66, 67, 77, 79-82, 102 comparisons and, 333-40; lemmas on, Industrial Coordination Act (1975), 294 341-44; redress of poverty and. 344-45 Industrial sector, 5, 11 nl7, 226, 276; licens- McNamara, Robert, 113 n5. 292 n37 ing on ethnic bases and, 294. See also One- Mahathir Mohamad, Dr., 16 digit industrial and occupational level Malayan Agricultural Producers Associ- (poverty): Two-digit industrial and occup- ation (MAPA), 164 ational level (poverty) Malayan Chinese Association (MCA), 7, 8, 9 Inequality. See Income inequality Malayan Indian Congress, 8 Inflation, 5-6, 43 n28 Malay Mail, 112 Institute of Medical Research (Malaysia). Malays. See Ethnic composition; Resiruc- 114' MaasSeEtm copsto;Rsr- Intervals (income classification in PES) 30- turing of society 31 ' Malaysia: economic overview of, 4-6; es- Interviews (HBs), 4749 timates of poverty in, 125-26: ethnic Interviews (HBS): farmer's income and, 28- pluralism in, 1-4; household income in- 29:ehousPEhol fasmentb interview. er equality in, 35-38; indigenous people of, 29: household assesSment by intervidwer 12 n19; intertemporal comparisons of in- and, 59-62; questions on income dunng. equality in, 42; New Economic Policy in, 23-24, 53-59: sampling procedure and, 9-14; politics in, 6-9; profile of poverty in, 24-26 126-32; profile of poverty sensitivity and, Irrigation, 157 132-35. See also Peninsular Malaysia; Jain, Shail, 40 Regions (states) Malaysian Socio-Economic Sample Survey Kak,an EN.onomi Malaysia, 112 of Households (MSSH, 1967-68). 21, 22, Kakwani. N. C.. 34 n16, 40 n23 42; income concept in, 43 Kendall, M. G4., 33 Mean independence, 89, 103, 192, 275, 306, Kendall, M. G., 31334 Kolm. S Ch., 215 n2934 Men: estimates of earnings function and Laborers: farm. 252. paddy, 157: landless education and. 247, 249. 253-54. 257: PES paddy, 160; rubber, 163-64 sample and military, 25 n3 Labor market experience, definition of, 239 Metropolitan towns See Urban sector, Land, 100, 287 n22; poverty and size of metropolitan towns holdings in, 156, 157; rubber. 163 n31 Mid-Term Reiaew oJ the Second MalaYsia Language Bahasa Malaysia as national, 8: Plan (MTR), 10, 12-15, 294-98 of school instruction (earnings function), Migration: to Peninsular Malaysia, 2, rural 257-59 to urban, 168; urban to rural. 295 Lee, E. L. H., 42, 44 Mincer. J.. 239 n2, 240 Leisure preference, 215, 216 Mining, 4, 216 n33, 226, 295 Lim. Lin Lean, 42 Ministry of Agriculture, 64 n4, 152 Log-income variance, 88, 261, 273. 275. 280, Ministry of Labour and Manpower. 163-64 350; age and education and, 248. 249, 254, Ministry of Welfare Services, 114; public 265; decomposition of, 330-31; discrete assistance program and, 281 income distribution (inequality indices de- Mirrlees, James A., 123 n22 fined in), 307-08: inequality across Moore, Basil J., 296 n44 households (personal income calculation), Muellbauer, John, 65 n7 188-89; inequality measure (personal income) and, 201-02: inequality measures Narkswasdi. U., 153 n14 compansons and. 331-32 National accounts, PES income estimates Lorenz class (of inequality indices), 339-40 comparison and, 38-39 Lorenz curve. 275; Gini coefficient definition National Corporation (PERNAS). 13 and, 34. 316- 17. for household and per National Operations Council, 8 INDEX 369 Natioial Union of Plantation Workers by, 192-98: employment status and. 207- (NUl'W), 164 14, income recipient definition and, 187- Negative exponential distribution, 353 88, inequality across households in per NEP Piong I See Poverty capita terms (calculation) and, 188-92, NEP Prong 2. See Restructuring of society interracial differentials (rubber tappers) New Economic PolIcy (NEP, 1971): analysis and, 215-16, male and female decompo- of study results and observations on, 298- sition and, 206-07; multivariate de- 301; employment and, 294; objectives of, composition and. 227-29: occupational 9-14, 272 decomposition and, 216-26, racial de- New Villages (Chinese), 150 composition and, 202-03, regional de- Nik b. Mohammad, 153 n14 composition and, 203-06, sector of em- Nonfood budget, poverty line definition and, ployment decomposition and. 226-27 116- 17 Pigou-Dalton condition, 80, 88 n33, 89, 103, Normalization axiom (Sen), 119-20, 121. 192, 306, 339 122 n20 Podder, N., 34 nl6, 40 n23 Nutrition, poverty budget and, 114-16, 117 Policy: commodity market intervention, 290-92; direct income transfers and, 281- Occupation, 279; education estimates and 85; employment restructuring, 294-98, earnings and, 251-53; HBS incomeconcept inequality measures and restructuring of and, 50; inequality in personal income society and, 101-03; interracial inequality and, 216-26, 229, 279; PES income ques- and, 298-301; poverty and fiscal, 286-89, tions and, 23 See also One-digit industrial redressing poverty and. 298-301. rural and occupational level (poverty); Two- development and, 144, 292-93 digit industrial and occupational level Political issues, 6-9 (poverty) Political parties, 8-9 Oil paln subsector, 4, 164, 167, 214 Population adjusting data on, 26, 1970 One-digit industrial and occupational level census and, 22, 24; of Peninsular Malaysia. (poverty): definition of. 236n, personal 2; percentage in poverty, 125 income inequality and, 216, 220; rural Population-size independence. 89. 103. 192. poor and, 145-46, 150, 151, 152: urban 275, 306, 341 poor and, 168 Population unit, 273. 339, choice of, 65. 127, Orshansky. M., II 6 nl 2 personal income inequality calculation Outline Perspective Plait (opp, 1970-90), 10. and, 191; standardizing (international 11, 294, 297 comparison), 41 Post-Enumeration Survey (PES, 1970), 271, Paddy (rice), 5, 287 n22; poverty in sector for, coding of income data and, 30; as data 157-60, 276; price support and. 290 base for study, 21-22. 271, 273; design of. Palan, V. T., 24, 28 24-26, 272; estimating household income Palm oil, 4, 164, 167, 214 distribution and, 30-34; general analysis Pan Malayan Islamic party, 8 of, 22-24; Gim coefficient estimation and, Parabolic age-log-income profile, 248. See 34; HBS comparison and, 51-53: also Age; Age-income profile household income inequality and, 35-38: Pareto distribution: personal income calcu- income definition and. 27-29. income esti- lations and, 192, 193, PES income classifi- mates compared with those of national cation and, 30-31, 33, 353 accounts, 38-39, incomequestions in, 22- Participation rate, 72, 186; personal income 23; international comparisons of in- inequality calculation and, 189-90, 197; equality and, 40-41: intertemporal com- underemployment and, 150-51 parisons of inequality and. 42-51, sample Payments in kind, 29, 43, 49 design of. 24-26 Peninsular Malaysia: indigenous people Poverty: commodity market intervention (Malays) of, 12 n9; migration to, 2; rural and, 290-92; defining, II 1-13; definition and urban ethnic groups and, 3-4; study of poverty line and, 113-18: direct income data and, I nl. See also Malaysia transfers and, 281-85. distribution of tax Personal income distribution, 277-78; de- burden and, 286-90; education and. 131- composition for inequality measures and. 32, 173, 186, employment and. 131, 135. 198-202, distribution of income recipients employment restructuring and, 294-98, 370 INDEX Poverty (continued) Rent: HBS and. 48. PES income definition estimating, 125-26: ethnic composition and, 29 of, 125-26, 127. 133, 146-50, fishermen Replanting program (rubber). 161-62. 163 and, 165-66, government focus on. 14- Restructuring ofsociety. 275. 281: as govern- 15, New Economic Policy and, 10-11, ment goal. 15-17: inequality measures 281; profile of Malaysian, 126-32. 276, and policy for, 101-03; New Economic redressing. 103. 275, 276, 298-301. 344- Policy and. 10, 11-14. 272 45: rural development and. 144. 292-93: Rice. See Paddy (rice) rural sector and. 127. 131. 133-35. 144- Riots, 7 nl2, 8, 9, 301 57: rural sector subgroups and. 157-64. Robless, C. L., II nl6 167, 276. sarong index of, 112. sensitivity Rothschild. Michael. 335. 338 of profile of, 132-35: urban sector and. Round. Jeffrey 1., 39 n2l 127, 145. 167-68, 172-73, 186 Rowntree, B S., 186 n44 Poverty budget. analysis of. 114- 17 Rubber, 4, 5. 214, 287 n22: interracial earn- Poverty gap. 123, 275. racial groups and, ings differentials among tappers of. 215- 125-26 16. poverty in sector for, 160-64. 276: Poverty line. 275: defining, 113- 18 govern- price intervention and. 291. tax on export ment plans and definition of'. 15, poverty of, 162, 288-89 profile and. 132-33; Sen index and, 125: Rubber Industry Smallholders Development variations considered in, 135 Authority (RISDA), 161. 162 "The Poverty of the Malays" (Za'ba). 112 Rural and Industrial Development Auth- Prices. food budget analysis and, 115. HBS ority, 166 and, 49: PES income definition and. 29, 64. Rural sector: development policies for. 144, of rubber. 162, support of. 290-91 292-93; ethnic composition of'. 3: food Prong I (NEP). See Poverty budget analysis and. 115. HBS sampLe and, Prong 2 (NEP). See Restructuring of society 45, 50, household income in MSSH and. 43. Public expenditure, 286. 290 household income in PES and, 37. in- Public sector: employment restructuring equality measurement and, 99-101; in- and, 294: government employment policy equality of personal income and, 203, 277. in. 220 PES income data problems and. 247, PES Purcal, J. T.. 153 n14 sample design and, 25; poverty and, 127, Puthucheary, J.. 13 n23 131, 133-35, 144-51, 276: poverty and Pyatt, Graham. 39 n21 subgroups in agriculture and fishing and, 157-72: rich households in, 142, 143 n41 Race. See Ethnic composition Racial disparity ratios, 82, 86, 197-98, 273- Sabah (state), I nI, 8, 12 nl9 74 Sampling procedure: HBs, 45-47: PES, 24-26. Racial disturbances, 7 nI2, 8. 9 272 Ranis, Gustav. 315-16 Sarawak (state), I nl, 8, 12 nl9 Rank-order welfare function. 118. 120-22. Sarong index of poverty, 112 314 Savings, 6, 38 Rao, V. M., 312-13 Sawyer. Malcolm, 39, 41 n26 Rawls, John, 113 Screening, education and, 243, 280 Redress of poverty rule, 103, 275, 276. econ- Seasonality off-season crops and. 157 n20: omic policy and. 298-301. Lorenz domin- PES income definition and. 28 ance and. 344-45 Second Malaysia Plan (sMP, 1971-75). 9. 10. Regions (states), 277, education and income 12, 157, 281 n6. 294 and. 261-64; ethnic composition and. 2: Self-employed individuals. 283: HBS income interregional inequality and. 93. 97. 99, concept and, 49, 50; HBs and PrES com- personal income inequality and. 203-06: parison and, 52: personal income in- poverty and. 131, 135, 156. 161, 168; pov- equality and, 207, poverty measures and. erty of fishermen and. 165: public assist- 131, 135. 172, understatement of income ance and. 283; rich households in. 142 and. 39 Regression analysis (race, occupation, and Selvadurai. S., 64 n4, 153, 161 n26 income), 222-23 Sen. Amartya K.. 229 n47, 314 Religion Islam as national, 7. taxation and. Sen poverty measure, 1 18-23, 125, 126, 133 286 n35. 275 INDEX 371 Sex: education estimates (eamings function) 60; rubber smallholders and, 160-63, by, 253-54; personal income inequality rural poor and. 150, 151-57, urban poor and, 206-07; poverty and, 151. See also and, 172 Men; Women Shaplen, Robert, 297 n46 Underemployment, 150-51 Singapore, 8 Unemployment earnings function and edu- Skilled workers, training of, 296-97 cation and, 253; poverty measure and, 131, Snodgrass, Donald R., 38, 39, 42, 44 rate of return to education and, 243: urban Social welfare, Atkinson index and, 82-86 poverty and, 172 n43 See also Economic welfare; Welfare Union of rubber workers, 164 comparisons United Malays National Organization Social welfare function (egalitarian), 338 (UMNO), 7, 8. 9 345 Urban Development Authority (UDA), 13 Special Advisory Committee on Cost of Urban employees, 279; age cohort and, 255- Living Indices, 46 57; age-income profile and, 248, 249, 254; Squire. Lyn, 229 n47 degree or diploma and, 259-61; esti- States. See Regions (states) mation of earnings function of education Sten, Nicholas H., 84 (by race, sex, and occupation) and, 249- Stiglitz, Joseph E., 243 nI2. 335. 338 53, human capital model and, 237-38; Stuart, A , 313 language used in schooling and earnings of, 257-59; PES income data and, 243; Tan, Siew Sin. 9, 10, nl5 regional breakdown of, 261-64 Taxes, 6, 38; direct income transfers and, Urban sector, I1; ethnic composition of. 3- 284-85; distribution of burden of, 286; 4; food budget analysis and, 115, HBS HBS income concept and, 50, improvement sample and, 45; household income in PES of welfare of poor and, 286-90; negative and, 37; inequality measurement and, 99- income-tax schedule, 283, 284; PES income 101; inequality of personal income and, and. 64; rubber and, 162, 288-89 203, 277; metropolitan towns definition Tax schedule (Sen's poverty measure), 123 (PEs) and, 25, 37 nI8; metropolitan towns Tea, 167, 214 and towns as area of, 100; poverty in, 127, Teachers, earnings of, 259, 261 145, 167-68, 172-73,186; rich households Tenure (land), 156 nl8 in, 142, 143 n41; towns definition (PES) Theil entropy index T, 93, 96, 102, 273, 274, and, 25, 37 n18 299; clecomposition of, 327-29; inequality van der Tak. Herman G., 229 n47 measures analysis and, 308-09; inequality Variance of logarithm of income. See Log- measures comparisons and, 331-32; in- income variance equality measures (personal income) and. -Varlog" measure. See Log-income variance 198-99, 202. 203, 206 von der Mehden, F. R., 297 n45 Theil's second measure L, 88, 90, 273, 275; deconnposition of, 329-30; inequality Wealth distribution vs. income-from-wealth measures analysis and, 309- 10, inequality distribution, 210 measures comparisons and, 331-32; in- Welfare. See Economic welfare; Social equality measures (personal income) and, welfare 199-200, 202, 203, 2()6 Welfare comparisons (lack of) and Lorenz Thillainathan, R., 215 dominance, 333-40 Third Malaysia Plan (TMP, 1976-80). 10. 133 Within-group component (racial). See n34, 152 nl2, 296 Ethnic composition, within-group com- Towns. See Urban sector, towns ponent and Training of skilled workers, 296-97 Women, 67 nl3, 276; estimates for earnings Transfer of income, 333, 334, direct, 281-85; function and education of, 247, 253-54; Sen's poverty measure and, 122, 123, salary discrimination and, 261; urban pov- "weak" principle of, 81 n21 erty and, 186 Two-digit industrial and occupational level World Bank, 38 n20, 157, 292 n37 (poverty): fishermen and, 165-66; la- Yields; paddy, 156-57; rubber. 162 borers on rubber estates and smallholders and, 163-64; other subgroups in agricul- Za'ba (literary figure). 111- 12 turearid, 167;paddyagricultureand, 157- Zakat, 286, 288 The full range of World Bank publications, both free and for sale, is described in the Catalog of World Bank Publications; the continuing research program is outlined in World Bank Research Program: Abstracts of Current Studies. Both booklets are updated annually; the most recent edition of each is available without charge from the Publications Unit, Dept. B, World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A. Sudhir Anand is fellow and tutor in economics at St. Catherine's College, Oxford, and formerly an economist at the Development Research Center of the World Bank. 0195201531