isms &L,1 /00 Living Standards Measurement Study Working Paper No. 100 Income Gains for the Poor from Public Works Employment Evidence from Two Indian Villages .~~~~~~~ .-A ge . |- I 1 1 ,. ; ., . {~~~~I " 11 LSMS Working Papers No. 28 Analysis of Househlold Expenditures No. 29 7The Distribuitioni of Welfre in Cte d'l1voire in 1985 No. 30 Qzuality, Qua7ntitt, and Spatial Variation of Price: Estimating Price Elasticities fronm Cross-Sectional Da ta NO. 31 Fiinancinig t1e Health Sector in Peru No. 32 Informal Sector, Labor Markets, anid Returns to Education itn Per No. 33 Wage Determinants in CMIc d'Ivoire No. 34 GuidelinesforAdaptingl the LSMS Living Stanidards Questionnaires to Local Conditions No. 35 Thle Demand for Medical Care in Developing Couintries: Quanittity Rationinig in Rutral Citc d'Ivoire No. 36 Labor Market Activity in CeIc dIlvoire anitd Peru No. 37 Health Care Finanlcinlg an thlc Demand for Medical Care No. 38 Wage Dctermiinants and Schlool Attainment among Men in Peru No. 39 Thle Allocationz of Goodls Tithin the Household: Adilts, Clhildren, anid Genider No. 40 7The Effects of Household anttd Community Characteristics on the Nuitritioni of Preschool Clhildrcin: Evidence from Ruiral C6te d'lItoire No. 41 Public-Private Sector Wage Differentials in Peru, 1985-86 No. 42 Tle Distributionl of W elfare in Peru in 1985-86 No. 43 Profits frott Self-Em ployment: A Case Study of Chte dlcoire No. 44 Tlh Liv ing Standards Surveyt/ an7d Price Policy RefJrml: A Study of Cocoan anid Coff'e Production in C6te d'1voire No. 45 Alcasuring the Willinigness to Pay for Social Services in Developing Couintrics No. 46 Nonagricultural Familh Enterp7rises in C(itc d'lvoirc: A Descriptivme A.nalhsis No. 47 The Poor dutrintg Adjustment: A Case Study of Citc dItvoire No. 48 Confronlting PoV0erty inl Developing Countries: Definitions, Informatiol, anid Policies No. 49 Sample Dcsigns for the Living Stanidards Surveyls in Glinina antd Mauritaniia/Planis de sondagc pour les en qetes sur le nilocaili dc v'ie ani Ghana et en MA'larit7nic No. 50 Food Subsidics: A Case Studjy of Pricc Reform in Morocco (also in French, 50F) No. 51 Child Arthropometrni in Cotc divoire: Estimates fromn Two Surceys, 1985 a7td 1986 No. 52 Public-Private Sector Wage Comparisons and Moonlighting in Develop7ing Coun tries: ELidecce from CCite d'Ivoirc andi Peru No. 53 Socioeconom71ic Determinan7ts o7f Fertility il Ctte d'lvoire No. 54 The Willinigntess to PnyJfor Education in Developing Countries: Evidence from Ruarl Pe rul No. 55 Rigiditr des salaires: Donnecs microheonomiqites et macreconmomiques simr l'ajustemnlit dim niarclht diu travail dans Ic secteur niodernIe (in French only) No. 56 Tlie Poor in Latiii Amierica during Adjustmenit: A Case Stitdy of Peru No. 57 The Substitutabilityi of Pi blic 1a7id Prianate H-letli Care for tlic Treatment of Cliildlrcn in Pakistan No. 58 Idcetifying the Poor: Is "Headship" a Llsefiul Concept? No. 59 Labor Market Performlan1Ce as a Determinant of Migration No. 60 The Rclativ Eff ctivnicss of Priz:itc anid Public Schools: Ev,ideiuct' from T T o Developiig Coun1ltr ics No. 61 Larget Sample Distribittion of Several neqitalilty McAisitres: Withi Ap?plication to Chte d1voirc No. 62 Testing for Significance of Pozvrty Differenices: With Appl1ication to Chte d'lvoirc No. 63 Poverty antd Econom17lic Gro0Tthi: Withi Ap1plicationi to CoIc itdvoire (List continues on the inside back cover) Income Gains for the Poor from Public Works Employment Evidence from Two Indian Villages The Living Standards Measurement Study The Living Standards Measurement Study (LSMS) was established by the World Bank in 1980 to explore ways of improving the type and quality of house- hold data collected by statistical offices in developing countries. Its goal is to foster increased use of household data as a basis for policy decisionmaking. Specifically, the LSMS is working to develop new methods to monitor progress in raising levels of living, to identify the consequences for households of past and proposed gov- ernment policies, and to improve communications between survey statisticians, an- alysts, and policymakers. The LSMS Working Paper series was started to disseminate intermediate prod- ucts from the LSMS. Publications in the series include critical surveys covering dif- ferent aspects of the LSMS data collection program and reports on improved methodologies for using Living Standards Survey (LSS) data. More recent publica- tions recommend specific survey, questionnaire, and data processing designs, and demonstrate the breadth of policy analysis that can be carried out using LSS data. LSMS Working Paper Number 100 Income Gains for the Poor from Public Works Employment Evidence from Two Indian Villages Gaurav Datt Martin Ravallion The World Bank Washington, D.C. Copyright X) 1994 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 Manufactured in the United States of America First printing March 1994 To present the results of the Living Standards Measurement Study with the least possible delay, the typescript of this paper has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. Some sources cited in this paper may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. Requests for permission to reproduce portions of it should be sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permnission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copyright Clearance Center, Inc., Suite 910,222 Rosewood Drive, Danvers, Massachusetts 01923, U.S.A. Tlhe complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list (with full ordering information) and indexes of subjects, authors, and countries and regions. The latest edition is available free of charge from the Distribution Unit, Office of the Publisher, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66, avenue d'Iena, 75116 Paris, France. ISSN: 0253-4517 In the Poverty and Human Resources Division of the Policy Research Department at the World Bank, Gaurav Datt is an economist and Martin Ravallion is principal economist. Library of Congress Cataloging-in-Publication Data Datt, Gaurav. Income gains for the poor from public works employment: evidence from two Indian villages / Gaurav Datt, Martin Ravallion. p. cm. - (LSMS working paper, ISSN 0253-4517 ; no. 100) Includes bibliographical references. ISBN 0-8213-2724-0 1. Public works-India-Maharashtra-Employees. 2. Poor- Employment-India-Maharashtra. 3. Income-India-Maharashtra. 4. Manpower policy, Rural-India-Maharashtra-Case studies. I. Ravallion, Martin. II. Title. m. Series. HD5713.6.142M33 1994 362.5'84-dc20 9343861 CIP Contents 1 Introduction ........................................... 1 2 The Time Allocation Model ..................................... 4 3 Setting and Data ........................................... 6 4 Estimating Foregone Income: A Non-Parametric Approach .................. 8 5 The Econometric Model and Estimation ............................. 10 6 Discussion of the Results ....................................... 14 7 Specification Tests ......................................... . 17 8 Implications for the Transfer Benefits from Public Works ....... . . . . . . . . . . . 20 9 Impact on Poverty ......................................... . 23 10 Conclusions .26 Appendix 1: A Consistent Estimator and Exogeneity Test ..27 Appendix 2: Asymptotic Covariance Matrix for the CTAM Estimator . .32 References ..35 Tables 1 Time allocation and household attributes by days of public-works participation in Shirapur .38 2 Time allocation and household attributes by days of public-works participation in Kanzara .39 3 Non-parametric estimates of average number of days displaced and average foregone incomes attributed to public works ..................... 40 4 Explanatory variables in the time allocation model ......................... 41 5 Estimates of the public-works employment equations ....................... 42 6 Estimates of the conditional time allocation model for males in Shirapur .... ....... 43 7 Estimates of the conditional time allocation model for females in Shirapur .... ...... 44 8 Estimates of the conditional time allocation model for Kanzara ................. 45 9 Specification tests for public employment equations ........................ 46 10 Alternative parameter estimates for public-works employment in the time v allocation model ............. .............................. 47 11 Effects of an extra unit of public-works employment on time allocation of males and females in Shirapur ................. 48 12 Effects of an extra unit of public-works employment on time allocation of males and females in Kanzara ................ 49 13 Average number of days displaced and average foregone incomes attributed to public works ............................................ 49 Figures 1 Moment residual plots ........................................... 19 2 Income distributions with and without earnings from public employment projects ... ... 25 vi Acknowledgments The staff of the International Crops Research Institute for the Semi-Arid Tropics, India, provided the raw data used here, and have helped us greatly in our work on those data. Financial support from the World Bank's Research Committee (RPO 675-96) is also gratefully acknowledged. Jere Behrman, Tim Besley, Richard Blundell, Emmanuel Jimenez, John Newman, Lyn Squire, and Tom Walker provided helpful comments. vii Foreword In efforts to better target public outlays for poverty reduction, "workfare" schemes have been a popular alternative to cash or kind handouts. Workfare participants are offered unskilled jobs at relatively low wages. Yet we know very little about one of the key determinants of the cost-effectiveness of such schemes in reducing poverty: the behavioral responses through time allocation of the participants and their families. Those responses will determine, in part, the foregone incomes of the participants, and (hence) the net transfer benefits. This paper provides estimates of how time allocation within sampled households responded to new rural employment opportunities provided under the "Employment Guarantee Scheme' of the State of Maharashtra in India. The paper is one of a series documenting results of a research program in the Poverty and Human Resources Division of the Policy Research Department. The program has aimed to improve our knowledge about the impacts on poverty of some of the targeted poverty alleviation schemes found in developing countries. It is hoped that intensive analysis of a few selected schemes will also yield lessons for the design and assessment of anti-poverty schemes in a wider range of settings. Lyn Squire Director Policy Research Departznert Abstract Current knowledge provides little guidance on one of the key issues in evaluating workfare schemes: What are the net income gains to participants? This paper offers an answer for rural public employment in the state of Maharashtra in India. An econometric model of intra-household time allocation is proposed, and the paper offers a consistent estimator, recognizing that the model entails that both regressands and endogenous regressors will be censored. The empirical implementation indicates that workfare projects induce significant behavioral responses, though the predominant time displacement is such that the net income gains remain large. Employment on the projects led to a reduction in poverty, of almost the same magnitude as a uniform and undistorting allocation of the same gross budget. ix 1 Introduction Public works projects have been a popular policy instrument for poverty alleviation in developing countries'. The net income gains to participating workers will depend in part on how time allocation across activities and persons responds to the new employment opportunities. Impacts on intra-household activity patterns are of interest for other reasons; for example, it has often been asked: when a woman takes up new employment on a development project does this augment, or displace, her domestic labor? Quite generally, the opportunity cost of the labor employed in a new project will depend on behavioral responses through time allocation; that will depend in turn on preferences, and on the way markets and institutions work in determining the constraints facing individuals. Thus the social evaluation of a new project will require evidence on, or assumptions about, behavioral responses through time re-allocation. One can distinguish two quite different ways of addressing these issues in past analyses:2 i) It is often argued that there is substantial "disguised unemployment" in underdeveloped rural economies, such that the opportunity cost of the labor employed in a new project will be low. This approach typically emphasizes quantity constraints on behavioral responses, such as due to the existence of wage inflexibility, or social customs in the gender division of labor. ii) Alternatively, one can appeal to the neoclassical model of time allocation, in which the assumption of wage and price flexibility guarantees that any allocation consistent with time and other resource endowments is feasible. This model gives less straightforward behavioral predictions, though the opportunity cost of labor will almost certainly not be zero, and may well be quite high. For example, some past estimates of the net income gains from direct poverty reduction schemes have used prevailing market wage rates for similar work.3 In each case, the outcome will also depend on whether one is concerned with the opportunity cost in terms of forgone utility (the "welfarist" approach), or that in terms of forgone incomes (a common I On the arguments for and against this type of anti-poverty policy in developing countries see Dreze and Sen (1989), Draze (1990), World Bank (1990, Chapter 6), Ravallion (199la), Besley and Coate (1992), Dev (1992) and Lipton and Ravallion (1993). For a more general discussion of targeted anti-poverty schemes see Besley and Kanbur (1993). 2 For surveys of the arguments and approaches to measurement see McDiarmid (1977) and Rosenzweig (1988). 3 See, for example, Hossain's (1988) evaluation of the famous Grameen Bank in Bangladesh, suggesting low net income gains. I "non-welfarist" approach).4 In principle, this distinction is independent of that between the "disguised unemployment" approach and the neoclassical approach. In practice, however, it is more common to find that those who follow the former approach tend to emphasize income gains and losses, while those following the latter adopt the welfarist approach by which the choice maximand - utility - is the sole metric of well-being. This paper takes a fresh look at behavioral responses to the new employment provided by public- works projects, and their implications for assessing transfer benefits. The essential idea is to model the intra-household allocation of time conditional on existing public-works employment; we call this the conditional time allocation model (CTAM). The theoretical formulation of CTAM is general enough to encompass both approaches described above. The quantification and valuation of displaced time relies largely on the empirical evidence provided by the estimated CTAM. Though our approach could be used to help inform either "welfarist" or "non-welfarist" evaluations, we will apply it here solely to the issue of assessing the net income gains from public employment projects. This will throw empirical light on the arguments that have been made about the efficiency of rural development projects as instruments for the alleviation of income poverty. In the process, we shall also examine the question of whether households appear to be rationed in their access to public employment; the ability of public-works schemes to provide work to whoever wants it at the going wage has long been recognized as a key determinant of their success in alleviating poverty.' We estimate CTAM on household data for two villages in the state of Maharashtra in India, where rural public-works employment is available under that state's "Employment Guarantee Scheme" (EGS).6 An explicit aim of this scheme is to alleviate rural poverty through the income gains to the participating workers. Unskilled manual labor is provided at low wages (on a par with the agricultural wage). The work is mainly small scale rural public works projects, such as roads, irrigation facilities, and re-forestation. EGS was introduced as a statutory program in the mid 1970s and expanded rapidly to reach average annual attendances of about 100 million person days in recent years. It is financed almost entirely out of taxes on the urban sector of the State of Maharashtra, including an income tax levy. Our data are from a longitudinal household level survey done over many years by well-trained resident 4 On the distinction between "welfarist" and "non-welfarist" approaches, see Sen (1979). 5 For further discussion on this point see Ravallion (1991 a,b). 6 There is a large literature on this particular scheme, including Dandekar (1983), Acharya (1990), Dev (1992), Bhende et al. (1992), and Ravallion et al. (1993). For other references see Ravallion (1991a). 2 investigators for the International Crops Research Institute of the Semi-Arid Tropics (ICRISAT), India. The data are widely considered to be of high quality, and they appear to provide a unique opportunity for evaluating household behavioral responses to the EGS. The following section outlines our model of time allocation in the abstract, and the properties of our estimator. Section 3 introduces the setting and the data. Section 4 presents the non-parametric estimates of foregone incomes, and notes some of their limitations. In section 5, we discuss the parametric CTAM and its estimation. We then go on to present and discuss the results in sections 6 and 7, while section 8 and 9 examine their implications for evaluating the net transfer benefits from the projects, and the impacts on poverty respectively. Our conclusions are summarized in section 10. 3 2 The Time Allocation Model In this setting an econometric model of time allocation should be at least consistent with the existence of involuntary unemployment of labor. This rules out formulations which assume that any time allocation consistent with the overall time endowment is feasible. We believe that one cannot dismiss either the casual evidence, or that from the survey data we are using, suggesting that quantity constraints on time allocation are common in this setting. For example, Walker and Ryan (1990) quote estimates of the rate of involuntary unemployment in the ICRISAT villages in 1975-76 of 19 percent for men and 23 percent for women.7 This aggregate figure hides a good deal of variation; the rate of unemployment in slack periods is far higher, being 39 percent for men and 50 percent for women. Insufficient wage employment is clearly the main factor here. But a number of constraints on time allocation (both internal and external to the household) appear to underlie such figures. For example, there appear to be binding constraints on what sort of farm work different genders can do, as a result of which men will often be idle in one season, while women are clearly over-worked. Our model should be consistent with a potentially complex web of constraints on time allocation. The model predicts time devoted to an activity (wage labor, own farm work and other self- employed work, leisure and domestic work, and unemployment) as a function of exogenous variables and the amount of time spent on public-works projects. Consider the household's choice problem of allocating total available time L across n activities, where the last activity is public-works. Income is y = F(L1,..,L.,x) (1) where x is a vector of relevant variables (prices, fixed factors in production and taste parameters). The vector x can be taken to include L. The chosen allocation maximizes U(Y,L1,..,L.,X) (2) 7 These are on a daily status basis, giving the number of days in which work was desired but not obtained as a proportion of the total number of days of work desired. 4 In addition to the time constraint, ML, = L, an optimum must satisfy a set of rationing constraints Li e [O, Li= (3) where it is assumed that the maximum time that can be devoted to activity i is itself a function of the household or individual characteristics, as Li x= (4) This last assumption can be justified in a number of ways. For example, it may reflect work-related disability. Or it may reflect social determinants of how employment is rationed, possibly to preserve some tacit collusion to keep wages above market clearing levels.! The optimal allocation of time across the n activities then maximizes U1F(L . ,L^,x),L1.. ,L,,,x] + ,.(L - EL) + EIX(L" - L) where the A1 (i=l,..,n) are non-negative Lagrange multipliers. We can write the solution as Li= LAx, L) i =1,..n-l L= L.(x) in which we make explicit the conditionality of L, (for ion) on L,. We refer to the functionsL4, L,) as the "conditional time allocation model (CTAM)". Note that one can always write solutions in this form, whether or not the allocation to any activity i is rationed. To do so, one solves the first-order and complementary slackness conditions for activities L= l,..,n-I conditional on the (rationed or un-rationed) solution for L.. This does not assume that L. is exogenous. ' On such models of wage determination in this setting see Datt (1989) and Osmani (1991). 5 3 Setting and Data Our data are for households surveyed over six years (1979-80 to 1984-85) in two Maharashtra villages, Shirapur and Kanzara.9 The villages differ in a number of respects. Kanzara is relatively a richer village, has more assured rainfall, is better irrigated and has less participation in the EGS. The villages also differ in terms of the cropping pattern, the occupational structure, and the distribution of land holdings. These differences have implications for household time allocation pattern which we discuss later. The data allow us to distinguish the time allocation of each person across five activities: 1: wage labor other than on the public works projects (mainly agricultural labor) 2: own farm labor and labor on handicrafts/trade activities 3: unemployment (number of days for which work was sought but could not be found) 4: leisure/domestic work 5: wage labor on public works (mainly EGS) These categories are self-explanatory, though we would make two observations: i) Unfortunately, the data do not allow leisure to be differentiated from domestic work. For women, a large proportion is probably domestic labor, which may also include some farm activities. ii) The fact that positive unemployment is reported (as noted in section 2) suggests that one or more activities must be rationed. It also indicates that the EGS is falling short of its aim of providing work to whoever wants it.'0 Tables I and 2 give cross-tabulations of days of participation against various variables, including time allocation by gender, for Shirapur and Kanzara respectively. Some points worth noting about the results in Tables 1 and 2 include: i) There is no sign of a displacement of wage labor time for females in either village, though there is such displacement for males, particularly in Kanzara. However, the propensity to do wage work of all forms will tend to be correlated with other variables such as landlessness and caste; later we will see whether any displacement is evident when one controls for these other variables. 9 For details on the agro-economic profile of these villages, see Bhende (1983) and Kshirsagar (1983). Also see Walker and Ryan (1990) for further information on the socio-economic conditions in these and other ICRISAT villages. 10 There is supportive evidence for that conjecture from an independent source; see Ravallion et al. (1993). 6 ii) For both villages and genders, there is a stronger suggestion of displacement in own-farm labor, though it should also be noted those who are employed most in public works tend to own the least land to farm; this is evident in the (rapid) decline of both LANDI and LANDU as public-works participation increases (bottom panels of Tables 1 and 2). We will see if the effect survives when one controls for land-holding. iii) Where we do see stronger, and more convincing, signs of displacement is in leisure/domestic work, which declines steadily as public-works employment increases. 7 4 Estimating Foregone Income: A Non-Parametric Approach As a first approximation, we used a simple non-parametric approach. The time allocation of non- participants in EGS is used to predict what the time allocation of participants would have been had they not worked on EGS. A simple comparison of the average time allocation of all non-participants with that of all participants does not control for other differences in the characteristics of participating and non- participating households which also influence their time allocations (Tables 1 and 2). A better approach would be to make this comparison conditional on some important determinants of household time allocation. For example, we may compare the time allocation of land-poor participating households with that of non-participating households who are also land-poor. Since such comparisons involve partitioning the sample by the conditioning variables, there are obvious limits to how far this approach can be taken given the overall sample size. The sample of 33 households in each village is first partitioned into three equal-sized groups on the basis of their average real asset holdings over the 6-year period, which is possibly the single most important determinant of household time allocation." Each asset group is further partitioned into participating and non-participating sub-groups. Let s' denote the average share of activity k in total available time for adult male members of all non-participating households in asset group g.l2 The female share sfk is defined analogously. Then, for a participating household i in year r, the time displaced in different activities may be estimated as Dk. = Sk '7 -Li (for = mj) where 74' is the total available time for adult members of genderj in household i in year t.13 Displaced time is estimated separately for each village. For Kanzara, male and female displaced time were not computed separately owing to the limited public-works participation of women. " See Tables 2 and 3. Other variables - such as land owned - matter, but are correlated with real assets. 12 The average share is defined as the mean, over all non-participating household years in any asset group, of the ratio (La,"/T,m), where the T,, is the total available days for adult male members in household i in year , of which Ld,, are devoted to activity k. 13 Note that, by construction, displaced time in all activities will add up to the time allocated to public works. 8 We do not expect this procedure to yield credible estimates of displaced time for every participating household-year; the variation in time allocation amongst participants is too great to allow that. However, average displacement may be better tracked. Thus, in Table 3 we report the average displaced time in different activities over all participating household years (across all asset groups). Table 3 also gives our valuations of the average displaced time. The general principle is to value displaced time at household-year specific prices. With the non- parametric estimates, since we will be concerned only with average displaced time (rather than its distribution), the valuation is done at average "prices", where the average is constructed over all household-years with participation. The displaced wage labor time is valued at household- and gender- specific average daily wage rates for any given year. For the valuation of time displaced in own farm and other self-employed activities, we use the normalized quadratic profit function estimated for these villages in Datt (1989). This profit function is estimated at the village level using data for all the ten villages, including Shirapur and Kanzara, surveyed by ICRISAT. The arguments of the profit function include gross cropped un-irrigated area, gross cropped irrigated area, family labor, owned bullock labor, the daily wage rate for hired labor, rental rate for hired bullock labor, and a composite price for seeds, fertilizers, Desticides, manures and machine hours. The profit function is used to derive a village and year specific estimate of the marginal profit per acre with respect to family labor time, which is then assumed to apply to both own farm and other self-employment activities. The foregone income in these activities is then calculated by multiplying this marginal profit per acre by the household's gross cropped area and the time displaced in these activities. The important point is to use a marginal and not average valuation. The prevailing agricultural wage rates may grossly overstate the value of marginal time displaced in own-farm and other self-employment activities. Time displaced from unemployment and leisure/domestic work is assumed to entail zero foregone income. The estimates in Table 3 suggest that foregone income represented 30 percent of earnings from public works in Shirapur (43 percent for males, 10 percent for females) and 4 percent in Kanzara. Some aspects of the estimates in Table 3 are, however, rather implausible. The estimates for males in Shirapur indicate that, as against their mean public-works employment of 58 days, 80 days of leisure/domestic work are displaced, and their unemployment would have been significantly higher in the absence of public works. Clearly, what is producing these results is that the households participating in public works are also the ones with high rates of unemployment and labor-force participation; partitioning by real assets is not a adequate control. The problem seems less serious for females in Shirapur and for Kanzara. But, probably in all cases, displaced time in leisure/domestic work is over-stated and that in unemployment is under-stated. 9 5 The Econometric Model and Estimation The problem with the above approach is insufficient control for the determinants of time allocation. A parametric model holds greater promise from this point of view. Following the discussion in section 2, the general (gender-differentiated) CTAM can be written as follows. Let LJ denote time allocation to activity k by persons of gender j(-m,) in household i in time period t . The equations for time allocation across all other activities are: X. k ~~~~~~~~~~~~~~~(7) LLS, Pk -r x+y'L,,, + y'L + ua,* ( Lb', -k X. + Yk L, + YkL.& + uL(8) (for k=l,n-1; i=l,H; t=1,T) where x,, is a vector of explanatory variables, and r;,; . Li,* Li,; >O L4 = 0 otherwise for j =m,f. The dependent variables are of course censored at the lower bound of zero. The time allocation equations are specified for three of the four non-EGS activities: wage labor, own farm labor (including other self-employment), and unemployment. Leisure/domestic work is determined residually, though we discuss the implications of relaxing this. However, since the days worked on any of these activities are censored variables, there is no obvious way of imposing additivity in the form of cross-equation parameter restrictions (Pudney, 1989). Estimation of the CTAM poses a difficult econometric problem, in that the Oimited dependent variable) model contains a censored right hand side variable (L.) which may well be endogenous. In Appendix 1, we derive a consistent estimator for the CTAM, and an exogeneity test for the censored right hand side variable. This estimator is a generalization of that proposed by Smith and Blundell (1986) who allow only continuous endogenous variables in the limited dependent variable model. Following that approach, our estimation strategy is outdined below. First, the model (7)-(8) is written conditional on error processes v, and vf as L, - Pk +yk' L; + y' + VS + Vft + e;;,(91 I10 Laf = Zx* + YkL,, + YkLL + WkV, + WVff + CL (10) where the equations for EGS employment are LX = is x + v,* L,,* = rg X. + vf(11) where L} = L. if L} > O, L} = 0 otherwise Model (9)-(10) is estimated by the Tobit maximum likelihood (ML) estimator after replacing V, andvf by their consistent estimates; the latter are obtained as residuals from equations (11) for male and female employment on EGS respectively. Exogeneity of male and female public employment to the time allocation for any activity is then tested by the significance of the corresponding a parameter. For any time allocation equation, exogeneity is tested sequentially for male and female public employment, beginning with the a parameter with the lower t-ratio. If the hypothesis of exogeneity of male or female employment on the public works projects is found to be statistically acceptable (we use a conservative significance level of 10 percent), then the time allocation model is re-estimated assuming exogeneity. If both male and female public employment variables are found to be exogenous, we are left with the standard tobit model. In this case, we further prune down the model to exclude either of the public employment variables if they turn out to be highly insignificant (with t-ratios less then unity). Next, the time displaced in any activity k due to male and female participation in public works (denoted DA,, and D./ is estimated as the difference between the expected days of work in activity k with no participation in public works and the expected days of work in that activity conditional on the household's current level of participation. Thus D= E[L'I Ix&,L; -0,L =O,0ZOS,i=o] (12) - E[LJ Ixi,L L, L!, lo;,9 O] (for j=mf) The model is estimated using longitudinal data of six years duration for the two villages with 33 households each."' For Shirapur, we shall estimate the CTAM separately for male and female 14 Despite the panel structure of the data set, we were unable to exploit it for estimating a fixed (or random) effects time allocation model because of the censored nature of the dependent variables. Apart from the inconsistency of a fixed effects Tobit estimator for a short panel, to estimate a fixed effects tobit model we would have had to exclude all households with zero participation in any activity for all six years in the panel (Heckman 11 household members, allowing for gender differences in time allocation, with cross-effects across genders (so, for example, the female time spent working on the own-farm may increase when the male joins the public works project.) A gender disaggregation was not feasible for Kanzara owing to the very limited participation of women in the EGS there, resulting in very few non-limit observations (see Table 2). This made both the parameter estimates of the female public employment equation and its residuals (included amongst the regressor variables of the CTAM) sensitive to minor changes in the specification. We therefore opted for aggregating male and female time allocations for Kanzara. For similar reasons, the model is not separately estimated for children. Given their extremely low participation in the labor force and still lower participation in the projects, we assume that their public-works employment comes entirely out of the time they were not in the labor force. The x, vector of explanatory variables in the time allocation model includes the variables listed in Table 4. Apart from the year dummies, five sets of variables are included, viz. variables relating to (i) household size and composition, (ii) value and composition of household assets, (iii) caste and educational status, (iv) incidence of some form of work disability, and (v) total available time (days per year) for adult males and females. The last set of variables allow for the overall time constraint (section 2), and are constructed as the total reporting days (the number of days for which the respondent provided time allocation information) minus days of sickness or non-residence in the village. The wage rates for both public and private work are deliberately excluded because of their potential endogeneity. EGS employment is remunerated on a piece-rate basis, and thus the time wage rate is not independent of the level of employment. The wage rate for private employment also has an endogeneity problem since the average wage received by a household member is an employment-weighted average of wage rates for different agricultural and non-agricultural operations performed by him or her over the year. Note that the xi vector of explanatory variables used in the public-works employment equations is taken to be the same as that in the conditional time allocation model. Thus there is a potential identification problem. The problem is less of a concern for non-linear models, of which the CTAM is an example. In particular, the usual exclusion restrictions are typically not required for identifying non- linear simultaneous models (Amemiya, 1985). For the CTAM, identification is possible by exploiting and MaCurdy 1980). For our data set, this would have meant throwing out more than half the observations in many cases. A further potential problem would be that the effective sample would have varied enormously across activities and gender. However, we do attempt to capture household-specific effects on time allocation by including in the set of regressors variables which are invariant over time for a household, for example, the caste ranking of the household, 6-year average of real assets of the household. 12 the non-linearity of the Tobit predictions from the public employment equations. That is our approach here, although we did test the robustness of our results to this choice by also estimating a model where the xi vector in the public employment equations had two additional variables, quadratic terms in CASTE and SCHYRH (see Table 4). The results for the latter (not reported, but available from the authors) were very similar to the case with identical set of exogenous variables in the public employment and conditional time allocation equations. For Shirapur, the estimates of time displaced by EGS employment in all activities except unemployment were identical in the two cases; for Kanzara too, the estimates of displaced time were quite similar in the two cases. 13 6 Discussion of the Results Table 5 gives the estimated parameters of the models determining employment on the public works projects. The salient features of these results are noted below. i) Male participation in the projects (in Shirapur, where the gender break-down is feasible) fluctuates far more over time than does that of females, as is evident from the year dummy variables in Table 5. The year 1982-83 was the year with highest village mean income in Shirapur, and this may well explain the drop in male employment in 1982-83, with workers being attracted into other activities. The drop in the following year is not explicable the same way; that year was no better than average. In Kanzara, we also see a sharp fall in employment on EGS in 1983-84, which is also the year in which average income peaked in that village, though the coefficient is barely significant at the 10 percent level. Of course, one must also consider the possibility that we are observing the effects of some form of rationing of EGS employment, such as through the opening and closing of works in progress; this appears to be an important factor in the EGS, at least over recent years (Ravallion, Datt and Chaudhuri, 1993). ii) The effect of long-term wealth (MRAST being the six-year mean of real wealth) on EGS employment differs between genders and villages. At the mean point, wealthier households tend to work less on these projects (holding the other variables constant) in Shirapur, though the relationship has different curvature for males and females (convex in the former case, concave in the latter). The effect is also concave in Kanzara, though the coefficients are not significantly different from zero. iii) Ceteris paribus, higher caste status is also associated with reduced public works employment in Shirapur, though not Kanzara. For example, in Shirapur, low wealth but high caste households do less of that work than do otherwise identical low caste households. Thus, some form of caste-related social stigma seems to be in operation here. iv) There is little to suggest that literacy or education affects employment on the projects, except in Kanzara, where SCHYRH is significant and positive. Again it should be noted that this effect holds constant other variables, including wealth. So the positive sign in Kanzara may be picking up a direct effect of education on the likelihood of participating in public employment, through (for example) knowledge of one's rights under the EGS. v) The effect of land owned on EGS employment is also quite different between genders and villages. In Kanzara, households with more un-irrigated land tended to work less on the projects, but there is no significant effect of irrigated land. In Shirapur, household land-holding has no significant effect on EGS employment of females, while (at the mean points) the effect is positive on male 14 employment. Of course, since wealth is held constant, it is not clear that the latter result implies poor targeting to landless households. It is also worth noting that the turning point in the relationship between male EGS employment in Shirapur and un-irrigated land owned is quite close to the mean, so that the relationship is positively sloped amongst those with relatively low land, and negatively sloped amongst larger holdings. vi) The only significant effects of livestock ownership are in the female equation for Shirapur, and only for livestock other than bullocks. The effect is negative at the mean. vii) The demographic variables come out quite strongly in Shirapur, but not Kanzara. As one would expect, larger households in Shirapur tend to have larger employment in the projects, ceteris paribus. viii) Work disability tends to reduce EGS employment, though the effect is only strongly significant amongst females in Shirapur. ix) Total available days (excluding days sick, or absent from the village) have the expected positive sign in each gender's equation in Shirapur, though there is no indication of significant gender cross-effects. The effect is not evident in Kanzara, though possibly the need to combine genders is confounding it there. x) As a general comment though, we do note that the equations for Shirapur appear to be better estimated than that for Kanzara; the possibility that we may not have a particularly good set of instrumental variables for public employment in Kanzara should be kept in mind in interpreting the rest of our results. Of greater interest are the estimates of the CTAM which are given in Tables 6 to 8. The salient features are as follows: i) The tests for exogeneity of EGS employment are indicated by the estimated ca-parameters and their t-ratios. We find that for both genders and both villages, exogeneity of EGS employment is accepted for both wage labor (other than public works) and own farm labor (including trade and handicraft activities), but not unemployment. Thus only for the unemployment equation do we retain the residuals from the first stage Tobit models for public-works employment in Table 5. The endogeneity of EGS employment in the unemployment model, but not other equations, could reflect the fact that employment on EGS requires a minimum spell of work (typically two weeks). Those experiencing greater unemployment are then more likely to obtain whatever EGS employment is available. ii) The conclusion that the public works employment is exogenous to other activities suggests that such employment is rationed. This is consistent with the reported unemployment, and also with the 15 findings (using aggregate time series data) of Ravallion, Datt and Chaudhuri (1992).1s Failure to obtain this work whenever needed will tend to undermine the social insurance function of public-works schemes, and it may also facilitate the possibilities for corruption. iii) For males in Shirapur, higher wealth is (ceteris paribus) associated with higher wage labor supply and lower unemployment at the mean point, though there is no significant effect on own farm activities. For females, higher wealth implies lower wage labor supply as well as lower unemployment, while there is a positive effect on own farm labor time, though it could barely be considered significant statistically. In Kanzara, wealth has a negative effect on wage labor supply, but no significant effects on other activities. iv) Ceteris paribus, high caste households are less likely to do wage labor for both genders and villages, though the effect is only strongly significant for males in Shirapur. High caste is associated with lower male unemployment in Shirapur. There are no significant effects of caste on own farm labor time in Shirapur, but some sign of a positive effect in Kanzara. v) Literacy tends to have a positive impact on recorded unemployment in Shirapur for both males and (though less significant) females; there is little to suggest effects on other activities. The sign of the effect of literacy on unemployment is reversed in Kanzara, where there are also indications of quite pronounced effects on other activities, with a shift out of wage labor time into own farm and other activities as the number of literate persons in the household increases. However, more years of schooling for the household head is associated with the reverse switch across activities in Kanzara. We have no explanation for this pattern. vi) The effects of land ownership on time allocation are pretty much as one would expect for males in Shirapur (with a switch out of wage labor time into own-farm time), but the same impacts are not evident on female labor time. Livestock assets follow a similar pattern, though positive effects on female own farm time become evident. vii) Male disability significantly curtails male wage labor time in Shirapur. Female disability does not do the same to female wage labor time. Furthermore, male disability results in higher female wage labor time, ceteris paribus. '5 Bhende, Walker, Lieberman and Venketaram (1992) also report that many of the chronically unemployed in these villages, who were not accommodated on EGS sites because of an excess turnout of laborers, became discouraged and did not return to work sites. 16 7 Specification Tests We conducted specification tests for non-normality and heteroskedasticity in the public employment equations, using the test proposed by Chesher and Irish (1987).16 The tests are based on second and higher moment residuals, where the rth moment residual is defined (suppressing sub-scripts) as c(t) = d._t(l'ldd=1) + (l-d).t(r'Id=O) - .0 where d is an indicator function with value 1 for non-censored observations, rn is the standardized error (vNa), the operator ^ indicates that the expectations are evaluated at the maximum likelihood estimates, and p(r) is the rth moment of the standard normal distribution. The non-normality test is motivated by noting that expected value of the third and fourth moment residuals is zero under the maintained hypothesis of normality; the heteroskedasticity test is motivated by noting that the covariance between second moment residuals and a set of exogenous variables is zero under the maintained hypothesis of homoskedasticity. Since we have a large number of exogenous variables in the public employment equations, we instead base the heteroskedasticity test on the squared predicted values of the regression function. The test results are given in the top panel of Table 9. For Shirapur, the assumption of normality is strongly rejected although in the case of males, this is entirely on account of the rejection of the fourth moment (kurtosis) condition. For Kanzara, while the normality assumption is acceptable at the 5 percent per cent level of significance, both the third and fourth moment conditions are individually rejected. Except for females in Shirapur, the tests also indicate significant heteroskedasticity. To examine further the sources of misspecification, the second, third and fourth moment residuals are plotted in Figures l(a), (b), and (c). Under the maintained hypothesis of spherical errors, the expected value of all moment residuals is zero. Figure 1 suggests that the moment residuals depart significantly from zero only for a limited number of observations'7: a single household for Shirapur males and for Kanzara, and two households in case of Shirapur females. We thus re-estimated the model deleting these households; three in case of Shirapur and one for Kanzara. The specification tests 16 These tests are also discussed by Gourieroux et. al. (1987) and Pagan and Vella (1989). '7 The observations in Figure 1 are sorted by ascending values of 6-year mean real assets of households, and year. 17 subsequent to the deletion are reported in the bottom panel of Table 9. The test statistics indicate the assumption of normality is now acceptable in all cases; heteroskedasticity is also attenuated in all cases, though still significant for Shirapur males and for Kanzara. Given our main interest in deriving estimates of displaced time (and hence foregone incomes) associated with public-works participation, we further look at the effect of deleting the 'problem' households on the estimates of the key parameters determining time displacement. The results are given in Table 10, which shows the estimated parameters of public-works employment in the time allocation equations, both for the full and the pruned sample. Two observations can be made. First, it turns out that the results for the exogeneity of public-works employment are unaffected by pruning; as before (see section 6), exogeneity is rejected only for the time spent in unemployment in either village. Second, the parameter estimates are quite similar in the two cases, with only one notable exception for Kanzara, where the pruned sample indicates a significantly lower displacement of wage labor time. In the light of these results, we chose to use the entire sample, recognizing that this may yield an over-estimation of foregone incomes. 18 Figure 1: Moment Residual Plots eV2), e(3), .(4X e(2), e(3), 6OM 100 1 00 50 Fig. l(a): Shirapur male Fig. I(c) Kanzara VD2, .23), 2(4 to - 0 ___ o lb 12 14 16 18 20 22 24 26 20 30 32 H"Iftld rJO6i l Fig. 1(b): Shirapur female 19 8 Implications for the Transfer Benefits from Public-Works Projects We shall only discuss here the implications for the average transfer-benefits; elsewhere we examine the distribution of those benefits, and policy implications (Ravallion and Datt, 1993). Table 11 gives a summary of the implied displacement of labor time in Shirapur for each activity associated with extra time working on EGS sites. The corresponding results for Kanzara are given in Table 12. The following points are of interest: i) For males in Shirapur, there is a strong negative displacement of time unemployed by extra EGS employment. When averaged over all participants in the EGS, the probability of the non-limit observation of unemployment for maJes in Shirapur is 0.77 (so that only 23 percent of pooled male household-years are predicted to have zero unemployment). Thus about 80 percent of an extra week of EGS employment came out of unemployment for males in Shirapur. The rest is made up largely of a displacement of own farm labor (Table 1I l). ii) For females in Shirapur, there is little sign of significant displacement effects in these three activities, so that extra time in EGS employment comes almost entirely out of leisure or domestic work (Table I1). We also ran the leisure/domestic labor equation separately (rather than retrieving it residually); the estimated marginal displacement was -1.1, close to the figure in Table 11. iii) There is no displacement of own wage labor for both males and females in Shirapur. All of the displacement of wage labor time in this village occurs through the cross-effects. iv) There is an indication of quite strong cross-effects of female EGS employment on male time allocation, though the reverse effect (of male public employment on female time allocation) is less evident. Male time allocation to own farm labor increases when female employment on EGS increases, while male time allocation to other wage labor falls. These effects are more difficult to interpret. Possibly some of what is being classified as "domestic work" by females in the data set is actually an own farm activity, and this is what men are taking up when the female joins the EGS project. The extra male own farm labor appears to be coming out of other wage labor and leisure. Is We also directly estimated the displacement effecIs using a separate model for leisure/domestic work (i.e., without imposing additivity). We found that there is a larger displacement of leisure/domestic work for males than implied by the other equations of CTAM. However, the key conclusion on foregone incomes is unaffected, given that marginal foregone income from leisure/domestic work is assumed to be zero. 20 v) The overall displacement to time allocation from public employment is quite different in Kanzara. The bulk of the time is coming out of wage labor and own farm activities (Table 12). The foregone income is then going to be larger, as we discuss below. Table 13 gives our estimates of the average number of days in the various activities displaced by public works, and the value of the corresponding income losses. The average is for all participating household-years, defined as those with non-zero adult male or female participation. The basis of valuation is the same as for the non-parametric estimates in section 4, except for one difference, viz., the valuation uses household-year-specific, rather than average, prices. It turns out that foregone income per displaced day in own farm activities is much lower than that in wage labor. Three factors may be at work here. First, the marginal contribution to farm profits of displaced days in own farm activities is found to be much less than the average profit per day of own farm labor. Second, the marginal contribution of family labor is an increasing function of gross cropped area (Datt, 1989), and thus tends to be low for the bulk of public-works participants who are small operators. Third, the extra time in own farm work may often be spent in tasks, such as soil conservation, whose contribution to farm output and profits, though probably small, is difficult to detect empirically. If true, the last explanation implies some under-estimation of foregone incomes. For Shirapur, the main activity displaced is unemployment for males and leisure/domestic work (which, unfortunately, cannot be split from the survey data) for females. In the aggregate, slightly over 40 percent of the time spent on the EGS site came out of leisure/domestic work in Shirapur, while one third came out of unemployment. Only one fifth involved a sacrifice of other wage labor time. The pattern is rather different in Kanzara, where nearly a third came out of other wage labor time, and a quarter was from own farm activities. Foregone incomes are estimated at 21 percent of gross wage earnings from public works in Shirapur, and 32 percent in Kanzara. The overall level of EGS employment was lower in Kanzara, though (given the pattern of displacement), its pecuniary opportunity cost was higher."9 This is intuitive; the same factors which drive up participation would presumably also reduce foregone income. Net transfer benefits from public works generated on average (for participating household-years) a 10 percent increase in pre-transfer earnings in Shirapur, a 7 per cent increase in Kanzara. Finally, we ask how the assessment of transfer benefits would change if the prevailing market wage is used as the opportunity cost of public-works employment. Using household-year specific (also '9 This result supports the similar conjecture in Bhende, Walker, Lieberman and Venketram (1992). 21 gender-specific in case of Shirapur) average daily agricultural wage rates, it turns out that foregone incomes would be 93 percent of gross earnings from public works in Shirapur (102 percent for males; 86 percent for females), and 77 percent in Kanzara. The differences amongst our various estimators are small compared to this difference; we conclude that the prevailing market wage greatly overestimates the foregone income of participating workers. 22 9 Impact on Poverty To assess the distributional impact of net earnings from public employment, Figure 2 gives the actual ("post-intervention") cumulative distribution of six-year mean income and that of the corresponding "pre-intervention" income (actual income minus gross workfare earnings plus foregone income).' To help assess the role played by foregone incomes, we also give the "pre-intervention" distribution that one would have obtained if foregone incomes were assumed to be zero (so this is simply actual income minus gross earnings from EGS). Two points are notable from Figure 2: i) Allowing for foregone incomes, earnings from public employment unambiguously reduce poverty; this would hold for any poverty line, and any poverty measure within a broad class.21 Of course, this does not mean that poverty is lower than it would have been under some alternative method of disbursing the same gross budget; we pursue this question somewhat further below. ii) The greater effects of allowing for foregone incomes (as measured by the vertical distances between the "pre-intervention" and "zero foregone income" lines in Figure 2) tend to be found at the lower end of the distribution. For example, ignoring foregone incomes one would have concluded that the proportion of the population with an income less than Rs 500 per person per month fell by about seven percentage points (from about 10 percent of the sample population to under 3 percent) as a result of earnings from EGS. However, the impact is a far more modest two percentage points when one allows for foregone incomes. It is beyond the scope of this paper to properly assess the performance of workfare in reducing poverty relative to alternative policies with the same aim. However, we can preform one simple test, in which the counter-factual allocation is a uniform transfer of the same gross budget across all households in both villages (whether poor or not).2 We assume that this can be done at negligible administrative cost, and with negligible impact on the pre-intervention income distribution (such as through income effects on time allocation or household dissolution). Neither assumption is realistic, and they will I The cumulative distribution functions (CDF) in Figure 2 have taken into account the stratified random nature of the sample; the densities at any income level have thus been obtained by using the relevant inverse sampling rates. For clarity in Figure 2 we have truncated the distributions at the top, though this does not alter any of the conclusions that follow. 21 This follows from the first-order dominance tests for poverty comparisons; see Atkinson (1987). 2 Note that uniformity across households does not require information on household size. 23 probably lead us to over-estimate the impact on poverty of uniform transfers.' We are probably also under-estimating the impact of the EGS on poverty since we do not consider induced income effects such as through asset creation and the effect on agricultural wages. We also assume that the EGS wage bill accounts for two-thirds of the gross budget; this figure is the estimate obtained by Ravallion, Datt and Chaudhuri (1993) from their analysis of the accounts of the Maharashtra Employment Guarantee Scheme. Figure 2 also gives the distribution of post-transfer incomes implied by uniform transfers under these assumptions. Though it can be seen that uniform transfers do not first-order dominate the distribution achieved by earnings from public employment, there is only a very narrow range in which the actual distribution function lies below that implied by uniform transfers. And uniform transfers second-order dominate the actual distribution, implying that a wide range of poverty measures indicate a lower poverty under uniform transfers for all poverty lines (Atkinson, 1987). This last calculation should not be taken to imply that the poor would be better off if one abandoned the rural public works projects in favor of uniform transfers; there are a number of other factors that would need to be considered, as noted above, including the results of specification testing (section 7) which suggest some under-estimation of net income gains from workfare. However, Figure 2 does lead us to question any presumption that the costs of this form of targeting - namely the foregone incomes and the non-wage costs - are of negligible consequence to an assessment of the policy's cost- effectiveness in reducing poverty. 23 Elsewhere we examine how large these costs would need to be to reverse the conclusions we come to below; we find that they would need to be equivalent to about 40% of gross disbursement (Ravallion and Datt, 1993). This would seem to be a high figure, leading us to conjecture that our qualitative conclusion could well be robust to a complete accounting of these costs. 24 Figure 2: Income distributions with and without earnings from public employment projects Cumulative percent of population 90 80 70 60 50- 40 30 -_ P re -intervention 20 - Actual Zero forgone income 10 - ----- Uniform transfer 0 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 Six-year mean income per person (Rs/person/year) 10 Conclusions We have proposed an empirical approach to estimating the impact on intra-household time allocation of employment on public-works projects. The model explains time allocation conditional on public-works employment, which is allowed to be endogenous, when suitable tests reject exogeneity. The statistical implementation is for two villages in the state of Maharashtra in India. Behavioral responses differ markedly between the villages. In Shirapur, wealthier households participate less in the projects, though there are signs that social stigmas and work disabilities dilute targeting performance somewhat. There is little sign of such effects in Kanzara. In both villages, employment in the projects is generally exogenous to time allocation, suggesting that the ideal embodied in the EGS of providing such work on demand is not being met. This confirms time-series evidence in Ravallion et al. (1993). The one exception to the exogeneity finding is for unemployment. This is consistent with the rationing of available public employment according to (in part) unemployment in other activities. In Shirapur, where a gender disaggregation of the model is feasible, there are signs of significant gender cross-effects in time allocation, such as through men taking up more own-farm work when women join the project sites. The projects also displace different activities for different genders: unemployment for men, leisure/domestic work for women. Our results suggest that the pecuniary opportunity cost to public-works participants is low, though this is more true of Shirapur, where participation is also higher. Overall, the projects do appear to generate sizable net income gains to participants, certainly far greater than implied by using market wages rates for similar work to value the foregone income. The transfer benefits alone led to a reduction in poverty, of almost the same magnitude as a uniform and undistorting allocation of the same gross budget. 26 Appendix 1: A Consistent Estimator and Exogeneity Test It is crucial to our analysis that we can obtain consistent estimates of the parameters of the tim allocation model. We have estimated CTAM using a limited information simultaneous tobit model where the endogenous variable appearing in the structural tobit model is also censored. As the properies of this estimator have not - to our knowledge - been discussed in the literature we shall outline them in detail here. To simplify the exposition, consider the following two equation simultaneous tobit model, a linearization of (6) for n=2 and t= 1,..,T: L-x, + yL2, + u, (Al) L2, = X,7' + V (A2) in which [] - N [0, 2] (A3) where - [all 012" - 12 @22) and where x,' is a row vector of K weakly exogenous variables. On defining d - I if L; > (A0) = O if L; S. O for =,I the observation mechanism for the model (A1)-(A2) can be written as: d L - ;dL (AS) Before we describe our estimator, it should be emphasized that model (Al) - (AS) is the structural-form model for the time allocation problem. In particular, the workfare employment equation (8) is derived as the solution of the household's optimization problem (discussed in section 2), and represents a structural equation of the CTAM. Also note that, while equations (Al) and (A2) imply a 27 triangular matrix of the coefficients of endogenous variables, they still constitute a simultaneous system because the covariance (a,2) between u, and v, cannot be assumed to vanish.2' Allowing forQm2 P 0 is appropriate insofar as (Al) and (A2) are derived from the same household optimization problem. Our estimator is a generalization of that proposed by Smith and Blundell (1986) who allow only continuous endogenous variables in the structural limited dependent variable model. However, our generalization of the Smith-Blundell estimator (and the exogeneity test) is limited in scope. Specifically, the generalized estimator proposed below is consistent for a simultaneous tobit model with censored endogenous variables and a triangular matrix of structural coefficients. While that is appropriate here, we do not offer a consistent estimator of the general simultaneous tobit model where no restrictions are placed on the structural coefficients.' Letting av, + e, be the value of u, conditional on v, the structural equation for L1, is -x,' + Y'2g + aV, + (E (As =w8 + E where E,-N(O,0112), =olI%, o112-a11-r4io22, w'=(xL2,v) and 8'=(P',y,a) We shall refer to the model (A2), (A4)-(A6) as the "expanded CTAM", recognizing that the initial CTAM ((A1)-(A5)) has been augmented by the inclusion of v, as an extra variable in (A6). Notefirst that the joint likelihood function for (L,I..) can be decomposed into a conditional likelihood function for Li, given Z.2, and a marginal likelihood function for L2. Then L,*(LIL.2 - Q1ZlA,8I2)-fi2(L2;e/ 24 If the covariance matrix was a diagonal matrix, the L., would cease to be endogenous in (Al), and a simple tobit ML estimator would yield consistent estimates. I Amemiya (1974) proposed an estimator for such a model. But the estimation procedure only uses the subset of observations for which all the endogenous variables attain their non-limit values. For our data, this would have meant throwing out most of the sample. 28 for the initial CTAM, and L(LI,L) = 01(L1;X',O' IL2)AQ2(L2;O') for the expanded CrAM, where r = (p',y,uI,,cr2), A' = (P',y,a,a1l12), 0' = (n',a22). It is easly sbown tht; - n, and hence L = L. 26 Let us additionally define Ad = (P',y,a,I). Then, following Engle, Hendry and Richard (1983),L2 is defined to be exogenous for Ad' if and only if L'(L,L,) = a;(L1;A'jL1).2(L2;O') where Ad' and 0' are not subject to any cross restrictions.' A sufficient condition for L'(L1,L) to be factorizable as above is that ac2 = 0. Given a one-tone correspondence between (alP, 0,2) and ( a), a2 ' 0 is equivalent to a =0. When a =0, c ll.=2=, c Q(L;A)LJ) = Ql(L,;AouIL2) and L(L1,L can be factorized as L(L1,L2) = fl(L1;XA' L2).CQ2%L;O) Testing the exogeneity of L2, for Ad' in equation (Al) is equivalent to testing for a = 0. Following Smith and Blundell (1986), our estimation procedure involves replacing v, in the expanded CTAM with a consistent estimate P,, which yields C, = x,j + yL2,f, ai7, (A7) =w,8 + et 2' This is proved in an addendum available from the authors. 2 Since we have a static model, weak exogeneity coincides with strong exogeneity (Engle, Hendry and Richard 1983). We shall thus stick to the term 'exogeneity". 29 In the Smith-Blundell model, I, is estimated by the OLS residuals from the second (uncensored) equation. Here we propose that v, is obtained as the residual from the tobit model for L>, which is a consistent estdmator of v, given by: vt= 4 - A - (AS) where F- 1 fzJ$x exp [-n2/(2a.)Yn fA' 1= exp [_(xtC)2/(2C22)] We then estimate (A7) as a standard tobit. To derive the key properties of this estimator, we write the log-likelihood of the expanded CTAM as follows '-' K ~~~dl,h d__2 In L ( -d,1)In(l -F1,) - 2 n11.2 e2t(A9) + (1 -d4)Jn(l -F2) - ~la2 -. d2 2 da22 where F,t=~~ fwa exp [-82/(2oj,.2)]dn fi,= ~ exp |('6/(2all2] Now cosider the disrutn ofthe ML estimatr of' = (', u112) given th the parametso' = ', C22) in the log-likelihood function have been replaced by their consistent estimates. Then, following Amemiya 30 (1979, equation 3.27), we can write 82ln -1 ,L+ B,2lnL I -1a -E 1 +E-(I-0e) (AIO) axal8J A aAO/ (where a denotes that each side has the same asymptotic distribution.) Given that the expectations are taken conditional on v,, it follows from Amemiya (1973) that -1 -21A pIim (AA) = [E dAaA, [P AA dA + ] (All) since plim (8inL4ia) =0 and plim(b -0)=0. Thus the estimator A is consistent. Next, we derive the asymptotic covariance matrix of A. From (A10), -2,t 1 ' 21L O2l 0bnL B2rz - 2InL 11InLV__ E0n OIL V(A) = E + [E E V(E)E_ (A12) Lala8A/ | a AAd axao' aeax' LBAxaBx In Appendix 2 we show that this can be written as: V01) = (WAM-' + d21(HAM-'{WAt[OJBXV(6)xB1[p O1 W}(YLAI)-l (A13) where and the first term on the right-hand-side of (A13) is the standard covariance matrix of the (tobit) ML estimator i (see Appendix 2 for details on the definition of matrices W; A, B, X). It is obvious from (A13) that under the null hypothesis of exogeneity, i.e. a =0, the covariance matrix of I collapses to the covariance matrix of the tobit ML estimator A. Standard t -ratios for a from a ML estimate of (A7) can thus be used to test for the exogeneity of the censored regressor La. Even when exogeneity is rejected, the ML estimator of (A7) is consistent, with its covariance matrix then given by (A13). To simplify the exposition, we have only considered a single censored (potentially) endogenous regressor. The results are easily generalized for a vector of censored endogenous regressors. 31 Appendix 2: Asymptotic Covariance Matrix for the CTAM Estimator Equation (A12) is obtained from (A10) after dropping the terms involving the covariance ( ), since E{anAL (o -8)'] = E( AnAL j( -O)e = 0, and using the result from Amemiya (1973) that E[anL.nL] =-E aLnL. Notice that in (A12), the first term is the covariance matrix for the ax1 a7xT axax' standard Tobit ML estimator, which following Amemiya (1973) can be written as - E 0"nL = (]eA.W)-l (A14) W= O I (A15) and W = (wl, w2,...wT)' with dimensions Tx(K+2), Q is a Tx(K+2) matrix of zeros, 0 and I are column vectors of zeros and ones, All A2 |(A16) Al= LA12 A22 and A. is a RT diagonal matrix with diagonal elements a(t) given as in Amemiya (1979; (3.17), (3.18), (3.19)). In particular: adt)= 1 |(w1),- F_ l , (A17) a2(t) = 2 [(w8)2fu, + au12 - I 112(Wlf] (A18) 32 To evaluate E irn' E 882% 2LnL (A19) E a hL E 82hL 8aul,2an' aO"II22] the following results are used: a:, = 8a(W,') =-xF2, an asl act a(w"8) 2(2fJ2 aa22 au 2 a, = _ (wI)fA __* __f (w,8)f,x, 88 011.2 afit = |(W,'8) - 0112]f c3al2 2 2c1, .2 aa22 2all2 afl, (wa'8) xf, a/i, _ w a,F2 aJ' ll2 ' =- a (WI8)x,ff,f2) aa22 Cyl11l3 33 From the log-likelihood function (A9), we have a_ y [e(d _w) (1 -dbl) t Oita (1 pI)AIWS 21 MaL 1 l(W:8)1- + dot aall.2 2112 (1-P) U11.2 Thus, using (A17) and (A18), we can write E -hL = -a E ajj(t)F2w,x, = -aW'A,,CX E aS = - a E2aI2(F2x,/ = - al AI2CX E a1a * - aeEa,)(If2/2)w, -- -eW'A11DI _ - -sEa12(t)(f/2) = -al'AI2DI .I.2aa22 I whore C and D are diagonal matrices with F2 and 12/2 as their t-th diagonals respectively. Further defining E m [C D] and X [= o], we can write E eL = -aEA[ ]BX and so V(i) -_ WAjD-1 + 2(]?AjD-y1{O[]BX KA) X'B'[I O]A W} A-1 V(s) = -1,3 _n_ a claimed in Appendix 1. 34 References Acharya, Sarthi. (1990), "The Maharashtra Employment Guarantee Scheme: A Study of Labour Market Intervention", ARTEP Working Paper, International Labor Organization, World Employment Programme, New Delhi, India. Amemiya, Takeshi. (1973), "Regression Analysis when the Dependent Variable is Truncated Normal". Econometrica 41:997-1016. (1974), "Multivariate Regression and Simultaneous Equation Models when the Dependent Variables are Truncated Normal". Econometrica 42:999-1012. (1979), "The Estimation of a Simultaneous-Equation Tobit Model". International Economic Review 20:169-181. (1985), Advanced Econometrics. Oxford: Basil Blackwell. Atkinson, Anthony B. (1970) "On the Measurement of Poverty", Econometrica 55:749-64. Besley, Timothy, and Steven Coate. (1992), "Workfare vs. Welfare: Incentive Arguments for Work Requirements in Poverty Alleviation Programs". American Economic Review 82:249-261. Besley, Timothy and Ravi Kanbur. (1993), "Principles of Targeting", in M. Lipton and Jacques van der Gaag (eds), Including the Poor, Washington DC: The World Bank. Bhende, M. J., T. S. Walker, S. S. Lieberman, and J. V. Venkataram. (1992), "EGS and the Poor: Evidence from Longitudinal Village Studies". Economic and Political Weekly 27:A19-A22. Chesher, Andrew, and Margaret Irish. (1987), "Residual Analysis in the Grouped and Censored Normal Linear Model". Journal of Econometrics 34: 33-61. Datt, Gaurav. (1989), "Wage and Employment Determination in Agricultural Labour Markets in India", Doctoral Dissertation, Australian National University. Dandekar, Kumudini. (1983), Employment Guarantee Scheme. An Employment Opportunity for Women. Bombay: Orient Longman. Dev, S. Mahendra. (1992), "Poverty Alleviation Programmes: A Case Study of Maharashtra with Emphasis on Employment Guarantee Scheme", Discussion Paper 77, Indira Gandhi Institute of Development Research, Bombay, India. Dreze, Jean. (1990), "Famine Prevention in India", in Jean Dreze and Amartya Sen (eds), The Political Economy of Hunger: Volume 2. Oxford: Oxford University Press. 35 Drbze, Jean, and Amartya Sen. (1989), Hunger and Public Action. Oxford: Oxford University Press. Engle, Robert F., David F. Hendry, and Jean-Francois Richard. (1983), "Exogeneity". Econometrica 51: 277-304. Gourieroux, Christian, Alain Montfort, Eric Renault, and Alain Trognon. (1987), "Generalized Residuals". Journal of Econometrics 34: 5-32. Heckman, James J., and Thomas E. MaCurdy. (1980), "A Life Cycle Model of Female Labor Supply". Review of Economic Studies 47:47-74. Hossain, Mahabub. (1988), Credit for the Alleviation of Rural Poverty: the Grameen Bank in Bangladesh, Research Report no. 65, Washington D.C.: International Food Policy Research Institute. Kshirsagar, K.G. (1983), "Agro-economic Profiles of Selected Villages in Akola Region of Maharashtra State in India". Economics Program - VLS Series, International Crops Research Institute for the Semi-Arid Tropics, Hyderabad. Lipton, Michael, and Martin Ravallion. (1993), "Poverty and Policy", in Jere Behrman and T.N. Srinivasan (eds), Handbook of Development Economics Volume 3. Amsterdam: North-Holland (forthcoming). McDiarmid, Orville John. (1977), Unskilled Labor for Development. Its Economic Cost. Baltimore: Johns Hopkins University Press for the World Bank. Osmani, S.R. (1991), "Wage Determination in Rural Labour Markets. The Theory of Implicit Co-Operation". Journal of Development Economics 34: 3-23. Pagan, Adrian, and Frank Vella. (1989), "Diagnostic Tests for Models Based on Individual Data: A Survey". Journal of Applied Econometrics 4: S29-S59. Pudney, Stephen. (1989), Modelling Individual Choice. The Econometrics of Corners, Kinks and Holes. Oxford: Blackwell. Ravallion, Martin. (199la), "Reaching the Rural Poor through Public Employment: Arguments, Evidence and Lessons from South Asia". World Bank Research Observer 6: 153-175. . (1991b), "On the Coverage of Public Employment Schemes for Poverty Alleviation". Journal of Development Economics 34: 57-79. Ravallion, Martin, Gaurav Datt, and Shubham Chaudhuri. (1993), "Does Maharashtra's 'Employment Guarantee Scheme' Guarantee Employment?". EconomicDevelopment and Cultural Change 41: 251-276. 36 Ravallion, Martin, and Gaurav Datt. (1993), "Is Targeting Through a Work Requirement Efficient? Some Evidence for Rural India", in D. van de Walle and K. Nead (eds) Public Spending and the Poor: Incidence and Targeting, World Bank, Washington DC, forthcoming. Rosenzweig, Mark R. (1988), "Labor Markets in Low-Income Countries", in H. Chenery and T.N. Srinivasan (eds) Handbook of Development Economics: Volume 1. Amsterdam: North-Holland. Sen, Amartya. (1979), "Personal Utilities and Public Judgements: Or What's Wrong with Welfare Economics?". The Economic Journal 89: 537-558. Smith, Richard J., and Richard W. Blundell. (1986), "An Exogeneity Test for a Simultaneous Equation Tobit Model with Application to Labor Supply", Econometrica 54: 679-686. Walker, T.S., and J.G. Ryan. (1990), Village and Household Economies in India's Semi-Arid Tropics. Baltimore: Johns Hopkins University Press. World Bank. (1990), World Development Report 1990. Poverty. Washington DC: World Bank. 37 Table 1: Time allocation and household attributes by days of public-works participation in Shirapur Days of participation per year per adult household member 0 < 30 30-60 > 60 Number of household years 93 67 22 16 Average days per year per adult: Public works 0 11.92 41.65 89.34 Wage labor 31.18 41.20 34.48 40.39 Own farm labor 49.87 62.44 48.00 39.38 Unemployment 6.82 16.64 26.57 23.94 Leisure/domestic work 212.24 181.86 160.56 140.05 Total available days 300.12 314.07 311.26 333.11 Average days per year per adult male: Public works 0 14.71 46.37 80.24 Wage labor 52.45 54.71 44.43 50.23 Own farm labor 65.68 80.62 68.51 58.36 Unemployment 7.72 19.72 32.89 27.07 Leisure/domestic work 162.21 136.36 128.04 117.50 Total available days 288.06 306.12 320.24 333.40 Average days per year per adult female: Public works 0 9.01 37.26 99.58 Wage labor 11.60 27.12 25.25 29.32 Own farm labor 35.32 43.49 28.99 18.03 Unemployment 6.00 13.44 20.71 20.42 Leisure/domestic work 258.29 229.29 190.71 165.42 Total available days 311.21 322.36 302.93 332.77 Household attributes:' MRAST 60.95 47.77 36.20 17.33 CASTE 2.08 1.81 1.68 2.31 MLITM 1.33 1.21 0.89 0.90 MLITF 0.98 0.43 0.20 0.10 SCHYRH 2.23 2.82 2.23 1.25 LANDU 4.93 3.69 1.87 1.37 LANDI 0.43 0.43 0.26 0.07 BULOK 0.72 0.81 0.55 0.50 OLVSTK 2.85 2.34 1.86 0.94 ADULTM 1.99 2.18 1.73 1.69 ADULTF 2.16 2.09 1.86 1.50 CHILD 2.43 1.78 1.68 2.19 HHSIZE 6.58 6.04 5.27 5.38 ' See Table 4 for definitions of the following variables. 38 Table 2: Time allocation and household attributes by days of public-works participation in Kanzra Days of participation per year per adult household member 0 < 30 > 30 Number of household years 137 48 13 Average days per year per adult: Public works 0 8.19 77.26 Wage labor 31.94 72.34 63.02 Own farm labor 61.52 42.03 12.27 Unemployment 8.52 17.36 32.38 Leisure/domestic work 230.15 184.20 150.49 Total available days 332.14 324.12 335.42 Average days per year per adult mak: Public works 0 13.10 118.90 Wage labor 36.22 71.48 44.50 Own farm labor 109.79 65.09 15.12 Unemployment 7.69 16.03 29.25 Leisure/domestic work 192.15 166.81 151.41 Total available days 345.86 332.50 359.19 Average days per year per aduli female: Public works 0 3.29 33.42 Wage labor 28.42 73.19 82.51 Own farm labor 21.77 18.97 9.26 Unemployment 9.21 18.70 35.69 Leisure/domestic work 261.44 201.59 149.52 Total available days 320.84 315.74 310.39 Household attributes: MRAST 64.99 15.57 12.72 CASTE 2.06 2.58 3.00 MLITM 1.37 1.22 0.90 MLITF 1.04 0.38 0.24 SCHYRH 4.30 3.17 4.69 LANDU 5.38 1.39 1.04 LANDI 0.70 0.09 0.00 BULOK 2.04 0.71 0.15 OLVSTK 2.92 1.38 1.38 ADULTM 1.74 2.31 1.54 ADULTF 2.11 2.31 1.46 CHILD 2.59 2.00 1.23 HHSIZE 6.44 6.63 4.23 * See Table 4 for definitions of the following variables. 39 Table 3: Non-parametric estimates of average nunber of days displaced and average foregone incomes attributed to public works. Shirapur Kanzara Male Female Totar Total' Activity Days Value Days Value Days Value Days Value Wage labor 24.02 197.78 5.32 22.97 29.34 220.75 3.05 17.56 Own farm/ -18.09 -4.00 0.26 0.06 -17.83 -3.94 45.11 4.12 other activities Unemployment -27.44 0 -7.20 0 -34.64 0 -4.21 0 Leisure/ 79.66 0 49.99 0 129.65 0 35.26 0 domestic work Total: 58.15 193.78 48.36 23.03 112.86 216.81 84.39 21.68 Public works 58.15 454.23 48.36 233.82 112.86 717.95 84.39 571.97 Note: Values are in 1983-84 rupees; all figures are annual, averaged over all household-years with participation. 'Totals include children. 40 Table 4: Explanatory variables in the time allocation model Shim=gu KICo Variable Description Mean Std. Dev. Mean Std. Dov. INT intercept D80 - D84 dummy variables for the years 0.167 0.374 0.167 0.374 1980-81 to 1984-85 MRAST average real assets of the household 50.335 40.596 49.575 66.392 (6-year average: Rs. '000 at 1983 prices) MRASTSQ square of MRAST x 10.2 41.734 57.258 68.433 153.79 CASTE caste ranking of the household 1.965 1.151 2.248 0.984 MLITM average number of literate adult nules 1.202 1.022 1.303 0.894 in the household (6-year average). MLITF average number of literate adult females 0.636 1.216 0.828 0.954 in the household (6-year average) MLITSQ square of the average number of literate 6.865 13.056 6.963 8.981 adults in the household (6-year average) SCHYRH years of schooling of the head 2.364 3.532 4.051 4.011 of the household LANDU owned un-irrigated land (hectares) 3.885 3.768 4.128 5.740 LANDI owned irrigated land (hectares) 0.384 0.816 0.503 1.235 LANDSQ square of total owned land 33.736 50.934 63.590 166.52 BULOK number of owned bullocks 0.707 0.990 1.591 2.244 CLVSTK number of owned other livestock 2.424 2.320 2.444 2.597 (other than bullocks) LVSTKSQ square of owned total livestock 17.232 26.552 34.742 61.632 ADULTM number of adult males in the household 2.000 1.008 1.864 1.245 ADULTF number of adult females in the household 2.051 1.107 2.116 1.279 CHILD number of children in the household 2.101 1.414 2.359 1.852 HHSZSQ square of household size 43.232 35.096 52.641 75.764 DISABM1 number of men (women or children) 0.505 0.682 0.268 0.498 DISABM2 reporting some work disability. 0.020 0.141 0.051 0.242 DISABF1 Type I is when the person cannot do 0.242 0.453 0.278 0.603 DISABF2 more than light farm or domestic work. 0.020 0.141 0.131 0.381 DISABC1 Type 2 is when the person 0.611 0.858 0.212 0.519 DISABC2 is completely disabled. 1.434 1.219 2.126 1.742 TM total available adult male days x 10' 60.119 31.062 63.840 42.463 (including leisure and domestic work but excluding sick and non-resident days) TF total available adult female days x 10l' 64.704 37.083 67.509 42.137 (including leisure and domestic work but excluding sick and non-resident days) 41 Table 5: Estimates of the public-works employment equations Shirapur Kanzara Male Female Male and Female Explanatory Parameter t-ratio Parameter t-ratio Parameter t-ratio variable estimate estimate estimate INT -239.12 -2.615 -736.26 -3.308 -133.28 -1.565 D80 -20.141 -0.805 17.063 0.543 16.733 0.421 D81 -26.289 -1.029 -38.308 -1.137 90.469 2.227 D82 -101.12 -3.511 -58.189 -1.636 44.475 1.049 D83 -105.12 -3.568 -16.409 -0.464 -76.737 -1.584 D84 -52.352 -1.865 -23.527 -0.666 -30.006 -0.616 MRAST -4.6309 -2.596 7.9138 2.07 3.3018 0.955 MRASTSQ 3.5226 3.23 -13.887 -3.015 -2.6025 -0.855 CASTE -27.18 -3.105 -51.278 -3.803 26.961 1.925 MLITM 19.039 0.871 38.907 0.911 -53.03 -1.227 MLITF 30.522 1.036 83.945 1.09 -68.726 -1.448 MLITSQ -3.3167 -0.825 -35.748 -1.981 4.9971 0.454 SCHYRH 4.475 1.499 -0.0527 -0.012 18.979 3.838 LANDU 49.426 2.928 17.413 0.533 -73.487 -2.618 LANDI 49.899 2.415 11.888 0.28 -51.11 -0.741 LANDSQ -5.0235 -3.726 -2.8734 -0.734 3.3419 1.439 BULOK 6.612 0.492 -6.2632 -0.324 -13.377 -0.524 OLVSTK -6.5347 -0.785 -61.012 -3.867 0.50741 0.027 LVSTKSQ 0.0092 0.011 5.7802 3.643 0.007817 0.002 ADULTM 91.18 2.258 213.8 2.413 -17.223 -0.377 ADULTF 97.368 2.35 279.73 3.246 32.562 0.827 CHILD 126.46 2.73 298.02 3.656 -281.17 -0.076 HHSZSQ -8.393 -3.615 -19.712 -3.155 0.58997 0.543 DISABMI -13.952 -0.757 -49.5 -1.62 29.93 0.905 DISABM2 -72.772 -1.337 -145.83 -0.033 37.351 0.45 DISABF1 -46.173 -1.978 -125.27 -3.387 59.487 1.534 DISABF2 128.59 2.135 52.102 0.523 -65.058 -1.389 DISABC1 -87.171 -1.957 -100.28 -1.69 288.13 0.078 DISABC2 -36.01 -0.939 -40.941 -0.853 239.6 0.065 TM 1.8286 2.863 0.37994 0.397 1.8972 1.465 TF 0.8942 1.166 3.2874 2.908 -0.44814 -0.414 SIGMA 81.514 12.207 83.556 9.715 101.24 10.577 Log Likelihood -541.654 -366.731 -398.818 Correlation between actual and predicted values 0.60 0.68 0.70 42 Table 6: Estimates of the conditional time allocation model for males in Shirapur Wage labor Own Farm labor Unemplovment Explanatory Parameter Parameter Parameter variable estimate t-ratio estimate t-ratio estimate t-ratio INT -99.108 -0.938 -323.82 -3.621 -204.96 -3.638 D80 25.256 0.782 42.66 1.485 71.128 4.338 D81 24.627 0.738 81.303 2.746 102.21 6.136 D82 79.347 2.342 130.15 4.288 53.967 2.247 D83 24.945 0.705 99.986 3.158 -23.942 -0.862 D84 12.858 0.361 49.966 1.583 -19.525 -0.915 MRAST 3.7335 2.054 -1.1821 -0.708 -7.0378 4.668 MRASTSQ -2.507 -2.38 1.2065 1.24 5.0658 4.913 CASTE -34.398 -3.384 3.2682 0.358 46.441 -6.239 MLITM 31.71 1.344 -36.346 -1.914 39.972 2.749 MLITF 10.304 0.312 2.5935 0.09 38.069 2.016 MLITSQ -5.4919 -1.301 0.63487 0.19 -5.0361 -1.949 SCHYRH -1.4867 -0.401 -3.5673 -1.208 2.9862 1.673 LANDU -52.107 -3.524 38.891 2.506 66.419 4.303 LANDI -70.803 -3.195 66.298 3.149 48.337 2.813 LANDSQ 3.303 3.246 -3.1457 -2.629 -6.3599 4.589 BULOK -64.74 4.342 52.133 3.895 21.095 2.534 OLVSTK -24.507 -2.428 21.542 2.494 -3.6086 -0.661 LVSTKSQ 3.3202 3.812 -2.2711 -2.995 -0.48747 -0.896 ADULTM 90.434 2.035 51.241 1.327 81.207 2.998 ADULTF 17.307 0.354 70.91 1.657 85.698 2.618 CHILD -56.122 -0.976 98.232 1.889 177.26 4.697 HHSZSQ -4.1224 -1.653 -3.3555 -1.611 -7.3587 -3.581 DISABMI -69.02 -3.076 -6.0312 -0.311 -20.228 -1.923 DISABM2 -139.97 -1.805 49.672 0.776 -104.11 -2.749 DISABF1 35.01 1.197 -94.424 -3.584 -57.159 -3.692 DISABF2 -64.349 -0.811 212.9 3.132 83.116 1.614 DISABC1 121.72 2.238 -109.02 -2.231 -127.45 -4.083 DISABC2 80.553 1.666 -54.8 -1.308 -92.082 -3.8 TM 0.22798 2.95 0.21471 3.153 0.22699 4.4 TF 0.13406 1.403 0.057071 0.753 0.043303 0.953 L5M -0.40238 -2.46 -1.0604 -2.78 L5F -0.43057 -2.31 0.59017 3.467 0.06896 0.798 ofM 1.3029 3.303 aF SIGMA 117.77 16.532 105.41 17.86 48.308 13.894 Log Likelihood -936.548 -1026.48 -571.629 43 Table 7: Estimates of the conditional time aOlocation model for females in Shirapur Wa2e labor Own farm labor Unemplovment Explanatory Parameter Parameter Parameter variable estimate t-ratio estimate t-ratio estimate t-ratio INT 48.639 0.667 -152.94 -2.879 35.612 0.492 D80 -27.156 -1.366 17.758 1.055 57.163 2.95 D81 -1.4954 -0.073 35.757 2.053 57.774 2.905 D82 -14.044 -0.666 79.561 4.545 27.773 1.322 D83 2.2664 0.104 63.156 3.474 -21.633 -0.93 D84 2.2438 0.106 29.649 1.582 -21.935 -0.969 MRAST -3.2763 -2.432 1.459 1.608 -2.9829 -2.238 MRASTSQ 1.257 1.225 -0.36452 -0.71 1.6043 1.847 CASTE -8.5702 -1.385 -4.68 -0.863 -12.962 -1.885 MLITM 6.5638 0.432 -2.0677 -0.186 37.922 2.362 MLITF 17.199 0.79 -12.548 -0.769 40.885 1.8 ML1TSQ -3.2192 -1.093 1.858 0.949 -7.3481 -2.488 SCHYRH 4.8335 2.118 -0.10614 -0.061 3.8099 1.474 LANDU 10.317 0.784 -4.1356 -0.554 1.6115 0.141 LANDI 8.1079 0.445 20.626 1.88 0.29169 0.017 LANDSQ -1.0804 -0.888 -0.68056 -1.311 -0.6115 -0.625 BULOK -17.103 -1.248 25.975 3.419 -4.6001 -0.442 OLVSTK -12.951 -1.291 22.963 4.543 -17.418 -2.895 LVSTKSQ -0.10812 -0.068 -1.5353 -3.463 1.1604 2.248 ADULTM -17.499 -0.559 -15.783 -0.695 43.409 -1.363 ADULTF 28.147 0.895 9.939 0.4 2.17 0.063 CHILD 39.211 1.142 31.533 1.014 9.8912 0.291 HHSZSQ -1.1988 -0.653 -0.77427 -0.655 0.72736 0.381 DISABMI 63.228 4.436 2.7525 0.24 59.893 4.058 DISABM2 -39.336 -0.715 9.9192 0.267 -197.7 -0.064 DISABF1 2.5687 0.141 -24.919 -1.666 46.458 2.417 DISABF2 20.53 0.367 62.115 1.596 -220.91 -0.072 DISABC1 -0.5155 -0.016 -39.015 -1.337 2.1516 0.067 DISABC2 -4.3616 -0.155 -9.4643 -0.368 -1.0876 -0.041 TM 0.013409 0.272 0.10494 2.649 0.027786 0.59 TF 0.037756 0.65 0.10673 2.38 0.012941 0.22 L5M -0.1734 -1.592 -0.1391 -1.304 LSF 0.029972 0.154 C(F 0.36158 1.655 SIGMA 61.253 14.363 62.014 18.114 54.023 11.694 Log Likelihood -628.216 -939.085 -444.925 44 Table 8: Estimates of the conditional time allocation model for Kanzara Waee labor Own farm labor UneMplovment Explanatory Parameter Parameter Parameter variable estimate t-ratio estimate t-ratio estimate t-ratio INT 290.14 4.758 -298.37 -4.363 120.63 4.126 D80 6.8942 0.216 32.641 0.944 -9.113 -0.762 D81 15.824 0.471 70.5 1.946 -33.204 -2.433 D82 -7.6023 -0.219 108.28 2.913 -89.108 -6.628 D83 -5.9376 -0.169 146.16 3.88 -101.33 -7.096 D84 -51.733 -1.416 155.42 4.041 -79.739 -5.54 MRAST -13.477 -6.481 3.3019 1.566 -1.5976 -1.274 MRASTSQ 5.8455 5.769 -0.84208 -0.869 -2.9243 -1.255 CASTE -23.6 -1.919 26.064 1.811 -3.8407 -0.73 MLITM -224.73 -6.874 120.23 3.373 -47.43 -2.968 ML1TF -158.2 -4.417 87.346 2.204 -33.453 -1.676 MLITSQ 32.95 4.235 -21.348 -2.685 7.5653 1.345 SCHYRH 15.171 3.832 -12.426 -2.793 5.2555 2.192 LANDU 33.01 1.989 -18.057 -1.055 -3.1977 -0.355 LANDI 31.956 1.429 2.3592 0.113 35.143 1.34 LANDSQ -2.9565 -3.573 -0.04002 -0.053 -0.98957 -1.115 BULOK -1.838 -0.135 58.534 4.366 12.405 1.584 OLVSTK 7.085 0.735 15.727 1.703 -8.4409 -1.21 LVSTKSQ -0.55587 -0.611 -1.0241 -1.428 -0.24668 -0.19 ADULTM 49.876 1.298 6.5344 0.151 6.5612 0.448 ADULTF 18.288 0.616 -52.195 -1.656 13.536 1.017 CHILD -71.001 -0.981 9.1682 0.129 20.253 0.751 HHSZSQ -2.3236 -3.352 0.49372 0.685 -0.30667 -0.842 DISABMI -11.9 -0.431 -10.806 -0.35 -9.8495 -0.92 DISABM2 -26.549 -0.483 61.503 1.007 1.4514 0.046 DISABF1 7.1723 0.281 40.108 1.418 22.336 1.844 DISABF2 18.133 0.548 30.349 0.909 -7.3069 -0.488 DISABC1 167.77 2.242 14.797 0.198 6.6022 0.239 DISABC2 98.965 1.376 -3.1805 -0.045 -29.059 -1.095 TM 0.20401 1.91 0.097874 0.815 0.067902 1.471 TF 0.18921 2.284 0.27814 3.182 0.047734 1.421 L5M + L5F -0.42628 -3.041 -0.42949 -2.306 -0.21102 -1.215 a 0.44076 2.248 SIGMA 111.44 17.295 129.52 18.856 39.078 16.423 Log Likelihood -952.533 -1143.01 -698.648 45 Table 9: Specification tests for public employment equations Test (d.f.) Shirapur: male Shirapur: female Kanzara Full sample Skewness (1) 0.86 19.57 4.88 Kurtosis (1) 13.69 12.38 4.35 Non-normality (2) 18.46 28.42 4.95 Heteroskedasticity (1) 10.42 3.06 16.62 Joint (3) 18.46 28.54 17.66 Excluding three households for Shirapur and one household for Kanzara Skewness (1) 1.71 3.90 2.71 Kurtosis (1) 0.09 0.06 3.63 Non-normality (2) 2.58 4.59 3.65 Heteroskedasticity (1) 9.05 0.75 6.35 Joint (3) 11.74 4.73 10.76 Note: The heteroakeduticity test is a test for zero covariance between the second-moment residuals and squared predicted values of the regression function. All tests are asymptotically central X. The 95 percent critical values for 1, 2, and 3 degrees of freedom are 3.84, 5.99, and 7.82; the corresponding 99 percent critical values are 6.64, 9.21, and 11.34 respectively. 46 Table 10: Alternative parameter estimates for public-works employment in the conditional time allocation model Parameter estimate (t-ratio) Equation Right hand side variable Full sample Pnmed sample Shirapur: males Wage labor Male public works 0 0 Female public works -0.431 (-2.31) -0.394 (-1.533) Own farm labor Male public works -0.402 (-2.46) -0.332 (-1.583) Female public works 0.590 ( 3.467) 0.714 ( 3.077) Unemployment Male public works -1.06 (-2.78) -1.024 (-2.595) Male public works residual 1.303 ( 3.303) 1.312 (3.152) Female public works 0.069 ( 0.798) 0.045 (0.409) Shirapur: females Wage labor Male public works -0.173 (-1.592) -0.091 (-0.761) Female public works 0 0 Own farm labor Male public works 0 0 Female public works 0 0 Unemployment Male public works -0.139 (-1.304) -0.074 (-0.542) Female public works 0.03 ( 0.154) -0.032 (-0.142) Female public works residual 0.362 ( 1.655) 0.66 ( 2.279) Kanzara: Wage labor Public works -0.426 (-3.041) -0.195 (-1.185) Own farm labor Public works -0.429 (-2.306) -0.394 (-1.995) Unemployment Public works -0.211 (-1.215) -0.236 (-1.165) Public works residual 0.441 ( 2.248) 0.388 ( 1.695) Excluding three households for Shirapur, one household for Kanzara. 47 Table 11: Effects of an extra unit of public-works employment on time allocation of males and females in Shirapur. Extra EGS employment of: Males Females Tobit Prob. Impact Tobit Prob. Impact coeff- non- on time coeff- non- on time icient limit allocation icient limit allocation Males: Wage labor 0.00 0.77 0.00 -0.43 0.77 -0.33 Own farm -0.40 0.71 -0.29 0.59 0.71 0.42 Unemploy- -1.06 0.77 -0.82 0.07 0.77 0.05 ment Leisure/ 0.10 -0.14 dom.work -1.00 0.00 Females: Wage labor -0.17 0.61 -0.11 0.00 0.61 0.00 Own farm 0.00 - 0.00 0.00 - 0.00 Unemploy- -0.14 0.46 -0.06 0.03 0.46 0.01 ment Leisure/ 0.17 -1.01 dom.work 0.00 -1.00 48 Table 12: Effects of an extra unit of public-works employment on time allocation in Kanzara. Tobit Prob. Impact coeff- non- on time icient limit allocation Wage labor -0.43 0.98 -0.42 Own farm -0.43 0.79 -0.34 Unemploy- -0.21 0.92 -0.19 ment Leisure/ -0.05 dom.work -1.00 Table 13: Average number of days displaced and average foregone incomes attributed to public works. Shirapur Kanzara Male Female Total, Total Activity Days Value Days Value Days Value Days Value Wage labor 15.55 127.96 5.30 22.85 20.85 150.81 32.96 182.31 Own farml -1.62 1.29 0 0 -1.62 1.29 23.04 2.01 other activities Unemployment 36.28 0 0.19 0 36.47 0 11.70 0 Leisure/ 7.94 0 42.87 0 50.81 0 11.51 0 domestic work Total: 58.15 129.25 48.36 22.85 112.86 152.11 84.39 184.32 Public works 58.15 454.23 48.36 233.82 112.86 717.95 84.39 571.97 Note: Values are in 1983-84 rupees; all figures are annual, averaged over all household-years with pafticipation. ' Totals include children. 49 Distributors of World Bank Publications ARGENTINA TheMiddleEast Oberver ITALY PORTUGAL Carlos Hirsh, SRL 41, Sherif Street Lkoa Commnionaria Sanami SPA Livranria Portugal Galera Guemes Cairo Via Duca Dl Calabra. 1/1 Rua Do Cartno 70-74 Florida 165'4th Floor-Ok 453/465 Caila Porstle 552 1200 Lisbon 1333 Buenoa Aires FINLAND 50125 Flrenze Akateeminen Kirrakauppa SAUDI ARABIA, QATAR AUSTRALIA, PAPUA NEW GUINEA, P.O Box 128 JAPAN larir Book Str FIJI, SOLOMON ISLANDS, SF-00101 Helsinki 10 Eastern BookService P.O. Box 3196 VANUATU, AND WESTERN SAMOA Hongo 3-Chome, Bunkyoku 113 Riyadh 11471 D.A Information Services FRANCE Tokyo 648 Whitehorse Road World Bank PubLications SINGAPORE, TAIWAN, Mitcham 3132 66, avenue dlen KENYA MYANMAR,BRUNEI Victoria 75116 Paris Africa Book Service MMA Ltd. Cower Asia Padfic Pta Ltd. Quaran House, Mfangano Strwt Golden Wheel Budlding AUSTRIA GERMANY P.O. Box 45245 41, Kallang Pudding 004-03 Gerold and Co. UNO-VeriaSg Nairobi Sinppor 13.34 Graben 31 Poppelaiorfer Allee 55 A-1011 Wien D-S300 Bonn 1 KOREA. REPU8UC OF SOUTH AFRICA, DOTSWANA Pan Korea Book Corporation For singLe tlew BANGLADESH HONG KONG, MACAO P.O. Box 101, Kwangwhamun Oxford University Pri Micro Indusiies Development Asia 2000 Ltd. Seoul Southern Africa Assistance Society (MIDAS) 46-48 Wyndham Street P.O. Box 1141 HouseS, Road 16 Winning Centtw Korean Stock Book Centre Cape Town 8000 Dhanmondi R/Area 2nd Floor P.O. Box 34 Dhaka 1209 Central Hong KorLg Yeoeido For subscrWliordersm Seoul International SubKsipt1on Servic Branrh offlkc HUNGARY P.O. Box 41095 Pine View, 1st Floor Foundation for Market Economy MALAYSIA Craighall 100 Agrabad Comrcial Area Dombovari Ut 17-19 University of Malaya Coopratiwv Johannesburg 2024 Chittagong 4100 H-1117 Budapest Bookhop, Lurited P.O. Box 1127. Jalan Pantai Baru SPAIN BELGIUM INDIA 59700 Kuala Lumpur Mundi-Prensa Lbron, S.A Jean De Lannoy Allied Publishers Private Ltd. Castello 37 Av. du Roi 202 751 Mount Road MEXICO 28001 Madrid 1060 Brussels Madras- 600 002 INFOTEC Apartado Postal 22-60 Librerla Internadonal AEDXS CANADA Bnmr chffei: 14060 Tlalpan, Meico DF. Consell de Cant 391 Le Diffuseur 15 I.N Heredia Marg 08009 Barcelona ISIA Boul. de Mortagne Ballard Estate NETHERLANDS Boucherville, Quebec Bombay - 400 038 De Lindeboom/InOr-Publikatl_ SRI LANKA AND THE MALDIVES J4B SE6 P.O. Box 202 Lake House Booksbhop 13/14 Asaf All Road 7480 AE Haakabergen P.O. Box 244 Renouf Publishing Co. New Delhi -110 C02 100, Sir Chittampalam A. 1294 Algoma Road NEW ZEALAND Gardiner Mawatha Ottawa, Ontario 17 Chittaranjan Avenue E5CO NZ Ltd. Colombo 2 KIB 3W8 Calcutta - 700 072 Private Mail Bag 99914 New Market SWEDEN CHILE Jayadeva Hostel Building Auckland ForsingisetitLst Invertec ICT S.A 5th Main Road, Gandhin Frtzes Fackbok fore1635 Av. Santa MaNria 6400 BaAplore - 560 009 NICERIA Regerinsgat n 12, Box 16356 Edificio ITEC. Of. 201 5University Pren Limited 5-103 27 Stockholm Santiago 3-5-1129 Kachiguda Three Crowns Building Jericho Cross Road Private Mail Bag S509 For siunriptin orders: C5HNA Hyderab -500027 IbdaA Wennergren-Wiliams AB China Finandal & Economic Pa -O box 1305 Publishing House Prarthara Flats, 2nd Floor NORWAY 5-171 25 Solna 83 Da Fo Si Dong Jie Near Thakore Baug&Navrangpurm Narvesen Information Center Beijing Ahmedabad -380009 Book Department SWITZERLAND P.O. Box 6125 Etterstd For single tickLe COLOMBIA Patiala House N-0602 Oslo 6 Lbrairie Payot Infoenilace Ltds. 16-A Ashok Marg Case postaie 3212 ApartadoAereo34270 Lucknow-226001 PAKISTAN CH01002 Lausnnr Bogota D.E Mirza Book Agency Central Bazaar Road 65, Shahrh_-e-Quaid-e-Azm For subssrnti onards: COTE DIVOIRE 60 Bajsj Nagar P.. Boa No. 729 Llbrairi Payot Centre d'Edition et de Diffusion Nagpur 440 010 Lahore 54000 Service des Abonnaments Africaines (CEDA) Case postile 3312 04 B.P. 541 INDONESIA PERU CH 1002 Lausanne Abidjan 04 Plateau Pt. Indira Limited Editorial Desarraoo SA Jalan Borobudur20 Apartado3824 THAILAND CYPRUS P.O. Box 181 Lrisa I Central Department Store Center of Applied Research Jakarta 10320 306 Silom Road Cyprus College PHLIPPINES Bangkok 6, Diogenes Street. Engormi IRAN nternational Book Center P.O. Box 2006 Kowkab Publishers Suite 1703 Cltyiland 10 TRINIDAD Ac TOBAGO, ANTIGUA Nicosia P.O. Box 19575-511 Condomainium Tower I BARBUDA. BARBADOS, Tehran Ayala Avenue, H.V. delx DOMINICA, GRENADA. GUYANA, DENMARK Cosht Extension JAMAICA, MONTSERRAT, ST. SamfundsUtteratur IRELAND Makatil Metro ManIa KITTS & NEVIS, ST. LUCA, Rosenoerns AlUe 11 Government SuppUes Agency ST. VINCENT & GRENADINEY DK-1970 FrederisbergC 4-5 Harcourt Road POLAND Systematics Studie Unit Dublin 2 International Publshing Service 9 Watt Strtee DOMINICAN REPUBUC UL Plckna 31/37 Curepe Editors Taller, C. por A. ISRAEL 00-677 Warzawa Trinidad, West Indies Restxuracion a Isabel la Cat6lica 3D9 Yozmot Uterature Ltd. Aparthdo de Corne 2190 Z-1 P.O. Box 56055 For sulwrrtwn ordr: UNITED KINGDOM Santo Domingo Tel Aviv 61560 IPS Journals Microinfo Ltd. UL Okrezra 3 P.O. Box 3 EGYPT, ARAB REPUBUC OF 02-916 Warszswa Alton, Hampahire GU34 2PC Al Ahrm England Al Galas Street Cairo LSMS Working Papers (continued) No. 64 Eduication and Earnings in Peru's Informial Nonfarm Family Enterprises No. 65 Forman and dInformal Sector Wage Determination in Urba.an Low-0Incorme Neighborhoods in Pakistan No. 66 Testhng for Labor Market Dutality: The Private Wage Sector in C6te d'lvoire No. 67 Does Eduication Pay int the Labonr Mark-et? Thte L.albar Force Participation, Occutpation, and Earninlgs of Peruvian Women No. 68 Thze Comtipositiont and Distriibuttioni of Incormie in Cdte d'lvoire No. 69 Price Elasticitiesfrom Suirvey Data: Extensionzs and Indonesian Resuilts No. 70 Efficient Allocation of Transfers to t17e Poor: The Problem of Unobserved Household Income No. 71 Investigating the Determinants of Household Welfare in Cote d'lvoire No. 72 The Selectivity of Fertility and thze Determinanats of Hmanai Capital Investments: Parametric anid Semiparametric Estimates No. 73 Shadow Wages anid Peasant Family Labor Suypply: Anz Ecoznometric Application to the Penrvian Sierra No. 74 The Action of Hiumizan? Resozurces and Poverty on One Another: What We Have Yet to Learn No.75 The Distribittion of Welfare in Ghana, 1987-88 No. 76 SCh7ooling. Skills, and the Returns to Government Invz7estmenit in Education: Ani Exploration Using Data from Ghana No. 77 Workers' Benifitsfrom Bolivia's Emergency Social Fund No. 78 Diual Selection Criteria zwithi Mlulltiple Alternatives: Migration. Work Statuts, and Wages No. 79 Genider Differences in Houisehlold Resozurce Allocations No. 80 Tle Houisehold Surey nasa Toolfor Policy Clhange: Lessons fromrr tlhe Jamaican Surzey of Living Co ndition s No. 81 Patterns of Aging in Thzailand anid Cote d'lzoire No. 82 Does Llndernuiitritioni Respond to Incomes and Prices? Dominance Tests for Indonesia No. 83 Growthl and Redistribution Components of Clhanges in Poverty Measure: A Decompositionz with Applications to Brazil antd Inidia in thze 1980s No. 84 Measurinig Incomnefrom Family Enterprises with Houtsehold Suirveys No. 85 Demand Analysis anid Tax Reform in Pakistan No. 86 Poverty anld Inequiality dutrinlg Unorthiodox Adiuistmtienit: Tlhe Case of Peru, 1985-90 No. 87 Family Productivity, Labor Suipply, and Welfare in a Low-Income Country No. 88 Poverty Comparisons: A Guide to Concepts anzd Methods No. 89 Puiblic Policy anld Anthropoinetric Ouitcomes in Cote d'lvoire No. 90 Measuiring tec Impact of Fatal 4dult Illness in Sub-Saha ran Africa: An Annotated Hozuselhold Qn's tionnuire No. 91 Estimating tlhc Determinants of Cognitive Achievement in Low-Income Countries: The Case of Ghana No. 92 Economic Aspects of Chiild Fostering in Cdte d'Iz'oire No. 93 l1lVestmlent InZ Humninii Capital: SchIoolinIg SIpply ConIstrainits in Rutral Gliana No. 94 Willingness to Payffor the Quality and Intensity of Medical Care: Low-Incomtie Houtselholds in Ghana No. 95 Measuiremenet of Retutrns to Aduilt Health: Morbidity Effects onz Wage Rates in C6te d'Ivoire and Glzania No. 96 Welfare Implications of Female Headslhip in Jataican Houiselholds No. 97 Houselhold Sizc in C6te d'lvoire: Sampling Bias in the CILSS No. 98 Delayed Primary Sclhool Enrollment antd Clhildhlood Malnutrition in Ghana: Ani Economic Atnalysis No. 99 Poverty Redutctionz throuigh Geographic Targeting: Ho.w Well Does It Work? The World Bank Headquarters European Office Tokyo Office 1818 H Street, N.W. 66, avenue d'lena Kokusai Building Washington, D.C. 20433, U.S.A. 75116 Paris, France 1-1, Marunouchi 3-chome Chiyoda-ku, Tokyo 100, Japan Telephone: (202) 477-1234 Telephone: (1) 40.69.30.00 Facsimile: (202) 477-6391 Facsimile: (1) 40.69.30.66 Telephone: (3) 3214-5001 Telex: wuI 64145 WORLDBANK Telex: 640651 Facsimile: (3) 3214-3657 RCA 248423 WORI-DBK Telex: 26838 Cable Address: INTBAFRAD WVASHrN(;TONDC kw ~ ~ ~ ~ V V-.AAtAA., t~~~p .V A' AA A ,,. : " 0 4 ' |'~~~4 . ''' tAt ' K~ c X ,,..4 S".g6yt :~~~~~~~~~~~~~~~~~~~~~~~~ ..A, ," A"F; V 4~~~~~~~~A. ~ ~ ~ ~ ~ ~ A _ 0 -A.A , , A ,A0 C;7 0&;Ft0D 0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~A A 9 A - AAt 0S ':,'>. A/A . ' ..AA ;~A. ,, 4..;. Ar .A..tffAA~~~~~~~~~~~~~~#~7 .~ ' 'cr; ~ ~ ~ ,t. 4~A.4 4 AN .AA ,. A... A,3 A A.? ..~~~~~~~~~~~~~~~v 4's. -- A~i -i4 .Z - AA(A4 ~ ~ ~ ~ ~ N A A A , S ..~~~~~~~~M. A 7 A9