PHN Technical 1iotes 'RES 5. THE WORLD BANK DETERMINANTS OF RECENT FERTILITY DECLINE 1N SRI LANKA lyl July 1983 Population, Health and Nutrition Department This paper is one of a series issuedby the Population, *Health and Nutrition Department for the information and * gu'idance of Bank staff working in these sectors. The * views and opinions expressed in"this paper'do not neces-* *sarily reflect those of the Bank'.* This report is ab tof fertility decline in Sri,: Lanka ini the post-1960 period.---,The paper concludes that a combination of historical ci'rcumstances and congeries of social, economic, and-cultural 'factors were respon nbsible 'f or, the initiation' of the decline in Sri Lanka. The fertility decline was sustained by,'rapid decreases-in infant and. child mortality, rapid increases in literacy, especially female literacy,. lessening of gender inequalities and elevation of women's status, lessening of inequalities through direct and indirect taxes, and the family p'lanning program. The Sri Lankan example gives added support to the point that a'. family planning program can have a noticeable'impact,'on' the fertility level of a,population even without,urbanization, industrialization, and other hallmarks of the Western industrial complex. -- Prepared'by N -. Krishnan Namboodiri, .AA.S. Peiri.s', H.R. Gunasekara, Soma Perara, Sulekha Patel, and Kathie -. Ladd Population, Health and Nutriti'on Department July 1983 .,'. . ".',. DETERMINANTS OF RECENT FERTILITY DECLINE IN SRI LANKA by N. Krishnan Namboodiri W.A.A.S. Peir.s H.R. Gunasekara Soma Perara Sulekha Patel Kathie W. Ladd Report No. 3 from RPO 671-70: Case Studies of the Determinants of Fertility Decline in South India and Sri Lanka PREFACE This report is one of four on the World Bank's research project: "Case Studies on the Determinants of Fertility Decline in Sri Lanka and South India" (RPO 671-70). The other three reports in the series are: Anomaly of the Fertility Decline in Kerala (by K.C. Zachariah, 1983) Determinants of Fertility Decline in Rural Karnataka (by N. Baskara Rao, P. M. Kulkarni, and P. Hanumantha Rayappa, 1983). A Comparative Analysis of the Determinants of Fertility Decline in Sri Lanka and South India (by K. C. Zachariah and B. Newlon, 1983). The research project was sponsored by the World Bank, the United Nation's Fund for Population Activities and the Governments of India and Sri Lanka. The project component in Kerala was carried out by the Bureau of Economics and Statistics, Trivandrum, and in Karnataka by the Institute for Social and Economic Change, Bangalore. In Sri Lanka, the Bank collaborated with the Department of Census and Statistics, Colombo. The costs of the study were met by the World Bank, the UNFPA, and by the collaborating insti- tutions. On behalf of the Bank and as the principal researcher of the project, I wish to acknowledge my thanks to the governments and organizations which helped carry out this project--to the UNFPA for providing partial financial support; to the Governments of India and Sri Lanka for collaborating with the Bank on the project; and to the Bureau of Economics and Statistics in Trivandrum, the Institute for Social and Economic Change in Bangalore, and the Department of Census and Statistics in Colombo for carrying out the studies in their respective areas. * - ii - This report was prepared mostly by Dr. Krishnan Namboodiri. Besides writing Chapters III, V, VI, VIII and the summary and conclusions, he put together the contribution by the other authors. Chapters I, II and IV were written at the Department of Census and Statistics, Colombo by W.A.A.S. Peiris (Chapter I), H.R. Gunasekara (Chapter II) and by Soma Perara (Chapter IV). Chapter VII was written by Sulekha Patel. Kathie Ladd co-authored Chapter VI with Mr. Namboodiri. The project could not have been carried out without the whole hearted support of Dr. Wickrema Weerasooria, Secretary, Ministry of Plan Implementation, and Mr. W.A.A.S. Peiris, Director, Department of Census and Statistics. Mr. Peiris organized the field work very efficiently and transferred the survey data on to computer tapes in a short period of time. The success of the project owes a great deal on the support received from th.ese two persons. Mr. Francis Perera managed the administrative and financial matters of the project in Colombo. This project was designed with considerable assistance from Dr. S. A. Meegama. His experience in the Sri Lanka Fertility Survey (WFS) of which he was the local director was of considerable help in the development of the questionnaire and designing of the sampling scheme. Judith Kovenock and Kathie Ladd provided invaluable assistance in data processing. The arduous task of editing and typing of this report was done by Bruce R. Geyer. I am thankful to all these people for their help in carrying out the project in Sri Lanka and bringing this out as a Bank report. K. C. Zachariah Population, Health and Nutrition Dept. TABLE OF CONTENTS Page Preface ..................... * .. * * * * * * *..... ii List of Maps ............... sO............ *.............. ..zQ......... . v List of Figures ............... e@@s@@@@e*e***n****** .................................... . . vi Li st o f Tab l es. . .. .. .. . .. . .. e . . o o. . . .o . . o. . . . o.... e o o. . . . . . . . . vii i Summary and Conclusion...-..... .o. .... o... o....... . .... A xviii CHAPTER I: DEMOGRAPHIC BACKGROUND AND SURVEY DESIGN- ............ 1 CHAPTER II: DATA ............ ..... .. ..... o ............ ........... 20 CHAPTER III: AGE AT MARRIAGE ............... .... o ... ......... - 33 Appendix to Chapter III.. ... ............. 72 CHAPTER IV: FERTILITY TRENDS AND.DIFFERENTIALS ...... .. - ......... .... 97 Appendix to Chapter IV. ......- ... .. ...... 0... 139 CHAPTER V: FERTILITY PREFERENCES, INTENTIONS AND BEHAVIOR .............. 143 CHAPTER VI: STERILIZATION-. -o... ... ...... .. ......... . 196 CHAPTER VTII: CURRENT USE OF CONTRACEPTION: PATTERNS AND DETERMINANTS..... ........ .. .... ............ ..e 246 Appendix to Chapter VII ....................... 277 CHAPTER VIII: SUMMARY, INTERPRETATION, AND CONCLUSION.... .............. 293 Bibliography.............. .......... . .... .. ............... e. 0 351 LIST OF MAPS Page Map 1.1: Sri Lanka (IBRD 3839R3) ...................................... 8 Map 2.1: Sri Lanka: World Fertility and Survey Zones (IBRD 17151) ................................................. 21 Map 4.1: Sri Lanka: The Six Zones--A General Ethnic and Socio-Economic Breakdown ................................... 106 - vi - LIST OF FIGURES Page Figure 1.1: Age Specific Fertility Rates, 1953/63/70/74 ......... 6 Figure 1.2: Sri Lanka's Social Indicators in Relation to Expected Values for Its Per Capita Income (S130) in 1975 16 Figure 3.1 5- and M- Curves Relating Consumption to Daughter's Marriage .......... ............................... 42 Figure 4.1 Parity Progression Ratios, 1975 and 1979: Confined to Ever-Married Women Age 45-49 at the Survey Dates 102 Figure 5.1 Path Diagram of Factors Affecting Fertility in the Interim (1975-79) ............ ................. 158 Figure 5.2 Recursive Model Treating Duration of Marriage (D) and Wife's Education (E) as Exogenous Variables, and Parity as of Time 1 (P), Number of Sons Living as of Time 1 (S), Number of Additional Children Wanted as of Time 1 (I), Contraceptive Use Status as of Time 1 (U), and Number of Live Births in the Follow-up Period (F) as Endogenous Variables .......................... 163 Figure 5.3a Overall-Sample Model ................. ............ 184 Figure 5.3b Sinhalese Model ................. .......... 185 Figure 5.3c Sri Lankan Tamil Model .......................... 186 F.igure 5.3d Sri Lankan Moor Model ..............4 187 Figure 6.1 A Pictorial Representation of a Hypothetical Cohort Whose Size Gets Diminished as its Members Undergo Sterilization or Are Censored ....................... 210 Figure 6.2 Cumulative Hazard Function Graph: Sterilizations for Ever-Married and Married Fecund Women, by: a(i),(ii) Ethnicity ... ....... ..... ... 216 b(i),(ii) Religion .. ............... ............ 218 c(i),(ii) Residence .............................. 221 d(i),(ii) Education .......... ... .........................223 e(i),(ii) Literacy .......................................... 225 f(i),(ii) Work Pattern ............... ............. I .......... 227 g(i),(ii) Excess Fertility ............................. 229 - vii - Figure 6.3 Cumulative Hazard Function Graph: Factors Within Ethnicity--Education: a(i),(ii) Sinhalese ... ........... .................. ..a.......a. 233 b(i),(ii) Sri Lankan Tamils ............a...*.*.**.a.... ....ae . 235 c Indian Tamils ................ a..aa.aaa... aa...a..aa a 237 d Sri Lankan Moors *........................... ..... 238 Figure 6.4 Cumulative Hazard Function Graph: Showing Difference in Popularity of Sterilization Between 1969-74 and 1974-79, by a Ever-Married Women ....... ..................a........ 240 b Married Fecund Women Who Want No More Children ...... 241 - viii - LIST OF TABLES Page Table 1.1 Population Growth, 1871-1971 ....................... 3 Table 1.2 Singulate Mean Age at Marriage, 1946/53/63/71 ...... 5 Table 1.3 Expectation of Life at Birth, Selected Years 1945-71 .............. . ...................... . 10 Table 1.4 Selected Social Indicators ...........15 Table 2.1 Selected Census Blocks by Zones and Strata .......... 23 Table 2.2 Staff Deployment for Field Work ....... ....... 26 Table 2.3 Response Rates by Zone ...... ...... 29 Table 2.4 Percentage Distribution of the Sri Lanka Population by Single Years in the Age Group 0-4 and Sex Ratios for the 1981 Census and the Household Schedule Data from the Survey ... . . ............ ......... ......... 30 Table 2.5 Percentage Distribution of the Sri Lanka Population, by Five-Year Age Groups and Sex .................... 31 Table 2.6 Blended Percentages of Myers' Index for the 1971 and 1981 Censuses and Household Schedule Data from the Survey ...................................... 32 Table 3.1 Age Pattern of Never-Married Males and Females, 1946/ 53/63/71/75 ........................................ 37 Table 3.2 Never-Married Persons in the Household Population, by Age and Sex (1975 Survey) ........39 Table 3.3 Percent Distribution of Successive Marriage Cohorts, by Age at Marriage (1975 & 1979 Surveys) ....... 52 Table 3.4 Mean and Standard Deviation of Age at Marriage of Ever-Married Women Age 25-49 and 25-29 and Married Before Reaching Age 25, by Selected Characteristics 54 - ix - Table 3.5 Incremental R2 Attributable to Individual Regressors [(i) when introduced in the regression in the sequence indicated (A1R2), and (ii) when introduced last (A2R2)]; Multivariate Analysis of Age at iarriage--Ever-Married Women of Selected Age Groups Married Before Reaching Age 25 .............. 59 Table 3.6 Parameter Estimates; Multiple Regression Analysis of Age at Marriage--Ever-Married Women of Selected Age Groups, Married Before Reaching Age 25 ......... 62 Table 3.7 Incremental R2 Attributable to Individual Regressors [(i) when introduced in the regression in the sequence indicated (AlR2) and (ii) when introduced last CA2R2)]; Multivariate Analysis of Age at Marriage--Ever-Married Women of Age 25 Years or Over arnd Married at Age Below 25, by Ethnic Group 63 Table 3.8 Parameter Estimates; Multiple Regression Analysis of Age at Marriage--Ever-Married Women, of Age 25-49 and Married at Age Below 25, in Different Ethnic Groups ...................... ..................... 64 Table 3.9 Ideal Age for Brides and Grooms--Ever-Married Woman's Responses ............ 0............ 0*****@@@*s .....0 .............. 67 Table 3.10 Mean Age at First Marriage of Wives; the Corresponding Figure for Husbands; and Mean Bride's and Groom's Ages Considered Ideal by.the Wives, by Selected Characteristics .............. 68 Table 3.11 Mean Age at First Marriage of Wives; the Corresponding Figure for Husbands, and Mean Bride's and Groom's Ages Considered Ideal by the Wives by Duration of Marriage (Below 25 Years) and Wife's Age at Marriage ........ ....69 Table 3A.1 Mean and Standard Deviation Age at Marriage of Husbands of Age 30-39, Married Only Once and Living with Spouse .... ......... *...................... 76 Table 3A.2 Multivariate Analysis of Husband's Age at First Marriage; Incremental R2 Attributable to Regressors (A1R2, the incremental R2 attributable to a regressor if it is entered-in the sequence shown and A2R2 the corresponding incremental R2 if the regressor is entered last)--Husbands, Married Only Once and Living with Spouse ........ 77 Table 3A.3 Multivariate Analysis of Husband's Age at Marriage (A1R2, the incremental R2 attributable to a regressor if it is entered in the sequence shown and A2R2 the corresponding incremental R2 if the regressor is entered last)--Husbands, Married Only Once and Living with Spouse ................. .......... 78 Table 3A.4 Multivariate Analysis of Husband's Age at Marriage; Parameter Estimates; Regression I Including Year Married among the Regressors and Regression II Excluding It--Husband- of Age 30-39, Married Only Once and Living with Spouse .........................79 Table 3B.1 Mean Expected Age at Marriage by Selected Characteristics and the Effect of the Characteristics on the Expected Age at Marriage When Current Age is Taken into Account--Never-Married Females and Males 86 Table 3B.2a Traits Desired in Prospective Husbands--Never-Married Females ........** ***** **................. 89 Table 33.2b Traits Desired in Prospective Husbands--Never-Married Males . . ....... ..... ......................... 90 Table 3B.3a Association Between Self's Educational Attainment and Desired Level of Education for Prospective Spouse--Never-Married Females ........................ 91 Table 3B.3b Association Between Self's Educational Attainment and Desired level of Education for Prospective Spouse--Never-Married Males ......................... 92 Table 3B.4a Distribution of Never-Married Females of Different Age Groups, by Desired Age of their Prospective Husbands . ....... .................... ............... 93 Table 3B.4b Distribution of Never-Married Males of Different Age Groups, by Desired Age of their Prospective Wives ..... . . . . . . . . . .. . . . . . . . . . . 94 Table 3C.1 Estimated Impact of Changes in Age Pattern of Nuptiality on Fertility in Sri Lanka from 1965-1971 and from 1971-1974 .................................. 95 Table 4.1 The Percent Distribution of Ever-Married and Currently-Married Women of Age 45-49 Years, According to Parity (WFS and WBFS) ....................... 100 - xi - Table 4.2 The Percent Distribution of Women Age 45-49 Years, According to Number of Children Ever-Born (New and Re-interviewed Respondents) ................ 104 Table 4.3 Association Between Background Variables Shown by the Percentage Distribution of the Sample by Pairs of Background Variables ........ .................. 107 Table 4.4 Differentials in Completed Fertility--Unadjusted and Adjusted for other Variables ........ ......... 115 Table 4.5 Mean Parity at the Time of the WFS and 1979 Survey, by Age at that Time ................*....... 124 Table 4.6 Mean Parity by Age--All New Respondents, Re-interviewed Respondents and Respondents Who Got Married after WFS . ............................... 126 Table 4.7 Mean Parity by Years Since First Marriage--Ever- Married Women (WFS and WBFS) ..................... 126 Table 4.8- Differentials in Mean Fertility of the 10-19 Year Marriage Cohort--Unadjusted and Adjusted for other Variables ... . . .. o.. . . . . . . . . . . . 129 Table 4.9 Percentage of Women Reporting a Current Pregnancy, by Age (WFS and WBFS) ........................... 134 Table 4.10 Age Specific Fertility Rates and Marital Fertility Rates for 1963 and 1973 (WFS and WBFS) ........... 136 Table 4A Current Marital Status by Age--Women in Household Population of Age 12-49 Years ................. 140 Table 4B Births in the Calendar Year 1978 by Age of Mother at the Time of the Birth ....................... 141 Table 4C Sample Weights by Zone ........................ 142 Table 5.1 Regression Coefficients (Dependent Variable-- Fertility Preference) .........*............. 145 Table 5.2 Multiple R2 of Different Regressions (Dependent Variable--Fertility Preference) ... .............. 145 Table 5.3 Multiple R2 of Different Regressions (Dependent Variable--Fertility Preference of Re-interviewed Subsample) .......................... ...... *s*@ §*e¢*@ 146 - xii - Table 5.4 Distribution of Women in the Sample According to Whether they Wanted Another Child Sometime and the Number of Live Births in the Interim (Two-Wave Data, 1975-1979) ..... e@................ 9 ......@ 148 Table 5.5 Comparison of Inconsistency Measures of Sri Lanka with those of Taiwan and Korea ............. 150 Table 5.6 Correlates of Inconsistency (Two-Wave Data, 1975-79) 152 Table 5.7 Individual Inconsistency (Percent Inconsistent) by Selected Characteristics According to Whether the Wife Wanted More Children and Duration of Marriage 156 Table 5.8 Standardized Regression Coefficientsi Equations of a Recursive Model--A Comparative Study Involving Taiwan, Korea, and Sri Lanka ..................... 160 Table 5.9 Standardized Regression Coefficients; Equations of a Recursive Model--The Sample as a Whole and Different Ethnic Groups ........................... 162 Table 5.10 A Recursive Model for Fertility Dynamics-- Standardized Regression Coefficients in Various Regression Equations and R2 for Each Regression ... 169 Table 5.11 Proportion of Variance Accounted for by Regressors When Entered in the Sequence Shown ................ 189 Table 6.1 .Proportion Sterilized among Women in Various Subclasses ................................. ... 201 Table 6.2 Logistic Regression--Propensity to Elect Sterilization among different Ethnic Groups ....... 206 Table 6.3 Ordinary Least-Squares Regression--Propensity to Elect Sterilization among Different Ethnic Groups. 206 Table 6.4 Sterilization Table for Ever-Married Women .......212 Table 6.5 Sterilization Function--Ever-Married Women and Currently-Married, Fecund Women Who Want No More Children, by Ethnic Group, Religion, and Residence 214 Table 6.6 Sterilization Function--Ever-Married Women and Currently-Married, Fecund Women Who Want No More Children, in Education Subclasses of Ethnic Groups 232 - xiii - Table 6.7 Sterilization Functions Based on Sterilizations Reported to Have Occurred During 1969-74 and During 1974-79--Ever-Married Women and Currently- Married, Fecund Women Who Want No More Children *.. 242 Table 6.8 Mean Number of Children Ever Born to Ever-Married Women by Current Age and Duration of Marriage for Sterilized and Non-Sterilized Couples ....... ...... 244 Table 7.1 Re-interviewed WFS and New Respondents by Socio-Economic and Demographic Characteristics 248 Table 7.2 Percent Age Distribution of WFS and New Respondents 249 Table 7.3 New Acceptors of Family Planning ................. 251 Table 7.4 Demographic Characteristics of New Acceptors ...... 252 Table 7.5 Proportion of Currently-Married Women Who Are Sterilized by Age of Respondent . .................. 254 Table 7.6 Proportion of Currently-Married Women Who Are Using a Method of Contraception, by Method ......255 Table 7.7 Percent of Women Currently Using a Method of Contraception ..................................... 258 Table 7.8 Percent of Women Currently Using a Method of Contraception, by Parity ................ ....... 258 Table 7.9 Percent of Women Currently Using a Method of Contraception, by Family Formation Status ......... 259 Table 7.10 Percent of Women Currently Using a Method of Contraception, by Religion ........................ 259 Table 7.11 Percent of Women Currently Using a Method of -Contraception, by Area of Residence .. ......... 260 Table 7.12 Percent of Women Currently Using a Method of Contraception, by Education . ....... ....* 260 Table 7.13 Percent of Women Currently Using a Method of Contraception, by Household Expenditure o. ....... 263 Table 7.14 Percent of Women Currently Using a Method of Contraception, by Employment Status .1 ........... 263 Table 7.15 Perlcent of Women Currently using a Method of Contraception, by Education of Husband ........... 264 - xiv - Table 7.16 Description of Variables Included in the Analysis 266 Table 7.17a Standardized and Metric Canonical Discriminant Function Coefficients--WFS Respondents ............ 269 Table 7.17b Standardized and Metric Canonical Discriminant Function Coefficients--New Respondents ............ 270 Table 7.18 Eigenvector Summary and Canonical Correlation for the Derived Function Differentiating Between Users and Non-Users ............ ............. r *........ 274 Table 7.19 Distribution of Observed and Predicted Group Membership of Users and Non-Users ... .......... 275 Table 7A.1 Percent Distribution of Women by Number of Family Planning Methods Known Before Probing (All Respondents) . ......... * *0 ...................... 279 Table 7A.2 Percent Distribution of Ever-Married Women by Methods Known and Age (All Respondents) ........... 280 Table 7A.3 Percent Distribution of Ever-Married Women by Methods Known and Education (All Respondents) ..... 280 Table 7A.4 Percent Distribution of Ever-Married Women by Methods Known and Parity (All Respondents) ....... 282 Table 7A.5 Distribution of Ever-Users by Methods and Age (All Respondents) . . ............... .... .......... 282 Table 7A.6 Distribution of Ever-Users by Methods and Education (All Respondents) ............... .................. 283 Table 7A.7 Distrilution of Ever-Users by Methods and Parity (All Respondents) ........... . .................. 283 Table 7B.1 Description of Variables Included in the Analysis 284 Table 7B.2 Descriptive Statistics (Means) of Selected Variables Included in the Analysis ................ 285 Table 7B.3a Standardized and Metric Canonical Discriminant Function Coefficients (WFS Respondents) ...... ...... 286 Table 7B.3b Standardized and Metric Canonicaal Discriminant Function Coefficients (New Respondents) o........ 287 -xv- Table 7B.4 Eigenvector Summary and Canonical Correlation for the Derived Function Differentiating Between Users and Non-Users .e.e. ......e....e..e..o .*.*..... .... G 291 Table 7B.5 Distribution of Observed and Predicted Group Membership of Users and Non-Users .... ............. 291 Table 8.1 Educational Attainment of Recent Birth Cohorts *... 301 Table 8.2 Industrial Composition of the Labor Force, 1953 and 1973 ......... a.......... **... *O. * ... . ......... 303 Table 8.3 New Acceptors of the Pill, Loops or Sterilization 312 Table 8.4 Comparison of Kurunegala and Anuradhapura Districts 318 Table 8.5 Comparison of Kegalla and Ratnapura ............... 319 Table 8.6 Comparison of Kandy with Nuwara Eliya .......... 320 Table 8.7 Comparison of Puttalam with Trincomalee .......... 321 Table 8.8 Mean Age at Marriage by Educational Attairment 323 Table 8.9 Mean Age at Marriage by Educational Attainment-- Ethnic Groups .......... ....... . * 324 Table 8.10 Percent Change in Duration-Specific Marital Fertility Rates (DSMFR)--1960-65 to 1965-70 and 1965-70 to 1970-75 ................................ 327 Table 8.11 Percent Change in DSMFR by Place of Residence-- 1960-65 to 1965-70 and 1965-70 to 1970-75 ......... 329 Table 8.12 Percent Change in DSMFR (Cumulated to 15 and 20 Years of Marital Duration) by Religio-Ethnic Grouping, Years of Education, and Husband's Occupation--1960-65 to 1965-70 and 1965-70 to 1970-75 ....e..e ooo ... .........eee... o.. 330 Table 8.13 Mean Score of Contraceptive Use by Place of Residence, Wife's Work Experience, and Husband's Occupation, for Women in different Marriage Duration Categories, Adjusted for a Number of Covariates .....* * *. ............... . . .. 334 - xvi - Table 8.14 Comparison of Observed TFRs and the Corresponding Values on the Trend Line Extrapolated from the Observed Values in the Third Quarter of the 1960s 336 Table 8.15 Inconsistency Between Intention Declared at the First Interview and Subsequent Behavior and that Between Intention Declared at the First Interview and Behavior as of that time ........................ 341 Table 8.16 A Recursive Model--Standardized Regression Coefficients 8 . e . . . e e . . . e . e * . . v v v v v e . . . e o . . . . . ... 342 Table 8.17 Median Time to Move from n Parity to (n+1) Parity, by the Survival Status of the nth Born . .......e 344 Table 8.18 Median Time to Move from n.Parity to (n+1) Parity, by the Survival Status of the (n-l)th Born, Given the Survival of the nth Born ....................... 345 DETERMINANTS OF RECENT FERTILITY DECLINE IN SRI LANKA by N. Krishnan Namboodiri W.A.A.S. Peiris H.R. Gunasekara Soma Perara Sulekha Patel Kathie W. Ladd - xviii - Summary and Conclusion Sustained fertility decline in Sri Lanka is a post-1960 phenomenon. Initia1ly, it was the rising age at marriage that caused the Total Fertility Rate (TFR) to fall. Later on, however, declines in marital fertility became an important component of the change. All social and economic strata of the population experienced fertility decline, more or less to the same degree and alr.: st synchronously. This pattern contrasts sharply with that of the Western fertility transition, which started with the upper classes and the urban sector from where it diffused to the lower classes and the rural hinterland--creating, expanding, and finally contracting, in the process, class differentials and rural-urban differences in fertility. The rising trend in age at marriage in Sri Lanka stalled in the latter half of the 1970s, and at about the same time the Total Fertility Rate (TFR) levelled off and even began to show signs of going up. A combination of historical circumstances and congeries of social, economic, and cultural factors was responsible for the initiation of fertility decline in Sri Lanka and for setting the pace thereof in the 1960s and early part of the 1970s. It is not possible to give a complete characterization of the combination of causal factors that lies behind the Sri Lankan fertility experience. Therefore, it is not possible to answer the question whether the experience can be duplicated elsewhere via deliberate intervention. Nevertheless, some of the factors involved are worth special mention. That there are some minimal social and economic changes which serve - xix - as a precondition for the onset of a sustained fertility decline in any society is nowadays accepted by most of those writing on the subject. Unfortunately, no one has been able to identify what these minimal changes are. The speculation is that there may be more than one combination of changes which will be sufficient. Sri Lanka, it may be noted in this connection, is still an agrarian, low-income country (albout 50 percent of its labor force is engaged in agriculture, and the GNP per capita as of 1981, at constant, 1959, prices was only Rs 1,014); Malthusian pressure on land is evident, particularly in the thickly-populated wet zone; unemployment is high--unusually so among the educated; the country is proud of its well-organized system of education (Sri Lankans can rightfully claim that they are better educated than most LDC populations); Sri Lanka is linked to other countries far and near through communication and transportation networks; Western ideas and life styles have influenced the island for centuries (Western attire, soda pop, and the like are usual in all population centers); direct and indirect taxes and other income transfer policies have served to redistribute income; the nation-state has stepped in to insu're against penury and to support the ill and aged--roles traditionally filled by one's children; relatively better health and longer life are characteristic of all population segments; and, finally, gender inequalities have lessened over the years. Also, it may be noted that family planning services have been made available to Sri Lankans by the Family Planning Association since its inception in 1953; the government declared in 1965 that thenceforth its * xx - policy would be to provide the masses with family planning services, information, and supplies; the legitimacy of contraception within marrirage is now widely accepted in the country; and, perhaps due to the efforts of the official family planning program, perhaps due to some other factors as well, the traditional differentials in contraceptive use are either absent or reversed in Sri Lanka (e.g., farmers have higher current use rate than non-farmers). Now, to comment on a few individual factors: sustained mortality reduction is regarded as a possible precondition for sustained fertility reduction. The assumption is that people are guided by the realizations that under reduced mortality conditions fewer births are needed for the survival of any desired number of children. In Sri Lanka, the effect on fertility of reduced infant mortality has been a physiological phenomenon--working through the prolongation of lactation--rather than a behavioral one stemming from a planned, downward adjustment of birthse It is unlikely that we have underestimated the behavioral impact, given the limitations of the data used in the analysis. The Sri Lankan experience also indicates that high literacy and higher educational levels when combined with longer waiting periods for employment cause age at marriage to rise. Although when viewed from this angle, the improvement of educational opportunities may not.appear to be an attractive policy intervention strategy, from another angle it might. Thus one of the consequences of high literacy and higher levels of education is increased exposure to non-indigenous ideas and lifestyles. From this - xxi - viewpoint, the improvement of educational opporttrnities is an attractive policy tool, in that it creates a fertile grouud for the world network of communication and interdependence to exert its full power of persuasion in motivating young people to adopt new life styles or a combination of the old and the new. Another policy intervention strategy is the lessening of class inequalities through direct and indirect taxes. The effect of such welfare polQicy measures on fertility is unclear. If fertility rises as income increases among lower-income strata and falls with income among higher-income strata, then a transfer of income from the upper to the lower classes raises fertility. If the income-fertility relationship is linear and positive, then redistribution of income leaves average fertility unaffected. On the other hand, if the basic relationship between income and fertiliiy is negative in the lower-income strata and positive in the higher-income strata, then a lessening of class inequality via redistribution of income depresses fertility. The available data do not permit a clear-cut characterization of the basic relationship between income and fertility in Sri Lanka. Moreover, the income effect' is confounded with the effect of other factors. There is some speculation that in Sri Lanka, at least for a period, the program of colonization of the dry zone promoted high fertility by giving preference to landless persons with large families. Yet another policy intervention strategy is the lessening of gender inequalities. The elevation of women's status, mainly as a result of increased parity in educational opportunities within a cultural context in - xxii - which females are accorded relatively higher status than is the rule in South Asia, is regarded as a major factor responsible for the reduction in fertility in Sri Lanka. If a lessening of gender inequality results in more autonomy for women over their lives or in having them participate in extra-familial activities which are incompatible with familial activities, then opportunity costs of children will rise and this will have a depressing effect on fertility. The available data do not shed adequate light on this relationship. Finally, there is the family planning program. The available evidence supports the contention that the family planning program's efforts to provide family planning information, supplies, and services facilitated fertility reduction. The Sri Lankan case illustrates the point that such efforts can have noticeable impact on the fertility level of a population even without urbanization, industrialization and other hallmarks of the Western industrial complex. What is not clear, however, is whether some as yet undetermined changes in the social and economic fabric are a precondition for a family planning program to have a significant impact. CHAPTER I DEMOGRAPHIC BACKGROUND AND SURVEY DESIGN Demographic History The demography of Sri Lanka can be traced, on fairly systematic data, as far back as the last quarter of the nineteenth century, when the first modern census was taken in 1871 and the Government Vital Registration System was established in 1868. For periods before that, little is known about the population dynamics, although interesting references to estimates of population size are found in ancient chronicles. These estimates range from 10 to 40 million. There are also records of two partial censuses taken in 1789 and 1814 which are believed to have suffered serious underenumeration. During this century, Sri Lanka, like several other developing countries, progressed from the first.to the third stage of the demographic transition. The first stage of high birth rates,, fluctuating around 40 per thousand population, and high death rates of over 20 per thousand population, ended in the mid-1940s, when mortality levels began to fall rapidly. During the second stage of falling mortality, the crude death rate fell at an unprecedented annual average rate of about 1.5 per thousand, reaching the level of 12 deaths per thousand in 1950 and thereafter descending at a progressively slower rate. It appeared to stabilize at around 6 deaths per thousand in the late 19708. -2 After a time lag of about 15 years following the initiation of the second stage, the third stage, fertility transition, began. The crude birth rate which stood at 37 in 1960 declined gradually to today's level of 28 births per thousand population. These changing vital rates in the transitional period promoted a rapid population growth in the country. The nature of that population growth and its components are outlined below. Population Growth The population of Sri Lanka enumerated at the Census of 1871 was 2.4 million. Growth during the succeeding decades brought the total population to 12.8 million in 1971. Table 1.1 shows the components of this growth--the natural increase of excess of births over deaths and the migration increase. The growth of population during these 110 years was not uniform. As shown in column 3 of Table 1.1, until 1946 the average intercensal rate of growth never exceeded 2 percent but fluctuated at levels below 1.7 percent. The period that followed was one of rapid population growth. The growth rate shot up to 2.8 during 1946-53 and remained as high as 2.7 in the 1953-63 intercensal period. Thereafter, the rate of increase diminished somewhat, averaging 2.2 percent in the 1963-71 period. During the period 1871 to 1900, on the average, the increase in population due to immigration exceeded that due to natural increase. The immigrants were almost entirely South Indian laborers brought to the country to work on tea plantations. In subsequent years with the imposition of strict immigration regulations and the acceptance of bilateral agreements to absorb Indian migrant population by both countries, the volume of immigrants declined and dwindled to insignificant proportions. In fact, after 1953 net Table 1.1: Population Growth, 1871 - 1971. Average Migration Increase Population Annual Rate Natural. Migration as a Percentage of Census Year (millions) of Growth Increase Increase Intercensal Increase (1) (2) (3) (4) (5) (6) 1871 2.4 -- 119,792 239,566 66.7 1881 2.8 1.4 144,260 103,791 41.8 1891 3.0 0.9 225,406 332,759 59.6 1901 3.6 1.7 356,147 184,249 34.1 1911 4.1 1.4 319,410 72,845 18.6 1921 4.5 0.9 656,950 151,316 18.7 1931 5.3 1.7 1,280,916 69,552 5.2 1946 6.7 1.5 1,352,606 64,796 4.5 1953 8.1 2.8 2,604,953 124,103 -- 1963 10.6 2.7 2,288,626 120,092 1971 12.7 2.2 4 migration was negative. In more recent years, as of the 19709, Sri Lanka has experienced another form of international migration--emigration of middle-level skilled and unskilled workers to the West Asian countries. These are, in all probability, short-term contract migrants whose impact on population growth rema.ins to be assessed. The rapid growth of population in the post-war years, then, was due entirely to natural increase--that is, an increase resulting from a phenomenal drop in the death rate from about 20 in 1945 to about 8 in 1971 and a continuation of birth rates on a high plateau of around 35 until the early 1960s. Since then, the crude birth rate has begun to decline and with it the natural increase. Nuptialitv Marriage in Sri Lanka is monogamous and it is the institution within which all fertility occurs. Therefore, levels and trends of nuptiality are important issues in the study of population. Nuptiality in Sri Lanka is characterized by two main features: (1) a steadily rising age at marriage of both females and males, and (2) a high degree of stability, with divorce and widowhood being not only of low prevalence during childbearing years but also largely compensated for by remarriage. The trends in age at marriage can be seen from the singulate mean ages at marriage computed at the census years shown in Table 1.2. The rise of almost 3 years in the mean age at marriage of women within the 25 year period is very impressive. The rise in male age at marriage has been less rapid, and consequently the sex differential has diminished from over 6 years in 1945 to 4 1/2 years in 1971. -5- Table 1.2: Singulate Mean Age at Marriage, 1946/53/63/71. Singulate Mean Age at Marriage Census Year Male Female Difference 1946 27.0 20.7 6.3 1953 27.2 20.9 6.3 1963 27.9 22.1 5.8 1971 28.0 23.5 4.5 Source: Department of Census and Statistics (1978), Census of Population 1971. Sri Lanka General Report, Table 7.1. Fertility Estimates of levels and trends of fertility have been made from vital registration and census data and, more recently, from survey data, particularly the World Fertility Survey data. Figure 1.1 illustrates the gradual decline of age specific marital fertility rates since 1953. The level of fertility remained high until the late 1950s. The first decline, albeit modest, in total fertility began in the late 1950s, almost entirely due to changes in marital composition brought about by rising age at marriage. After the mid-1960s, marital fertility also began to fall with accelerating rapidity--at an average rate of about 3.9 percent per annum during the period 1963-74, and a higher rate of 4.5 percent for the period 1971-74. Rising age at marriage contributed nearly 60 percent to the decline in 1963-74, but only 46 percent to that in 1971-74. -6- Figure 1.1 SRI LANKA Age-Specific Fertility Rates 1953, 1963, 1970, and 1974 ASFR Per 1000 300 250 200 1 970/ > 200 500 15-19 20-24 25-29 30-34 35-39 40-44 45-49 Age Wodd Bank-24871 -7 Thus, the initial contributory factor was almost exclusively delayed marriage; thereafter, its contribution declined gradually, and fall in marital fertilitey became the predominant contributor. Sri Lanka, therefore, has followed the pattern of fertility decline commonly experienced by other Asian countries. Secondary analysis of WFS data (Alam and Cleland, 1981) has also shown that the decline in marital fertility has been relatively even and synchronous across educational, occupational and urban rural categories of the sample. Consequently, fertility differentials did not widen among these subgroups of the population, in contrast to the experience of the European countries, where the decline in fertility among better educated, professional, and urban classes long preceded the occurrence of similar declines among poorly educated, manual, and rural groups. However, among subgroups defined by ethnicity and region, there was a time lag in the onset of fertility decline. Moors, Sri Lankan Tamils, and residents of the North Central (Anuradhapura and Polonnaruwa Districts) and Central Provinces (Matale, Kandy and Nuwara Eliya Districts) [see Map 1.1 of Sri'Lankan Administrative Districts] have achieved much more modest reductions in fertility compared to other racial, religious and residence groups. It has been demonstrated using multivariate analysis of WFS data (Little and Perera, 1981) that regional differentials can be attributed to variations in ethnic, educational and occupational differentials, but ethnic and religious differentials remain, irrespective of compositional variations of education or occupation. Thus, religion and ethnicity appear to be the most important determinants of fertility trends in Sri Lanka. -8- IBRD 3839R3 80- 81' 8(2'NOVEMBER 1982 Kankesanturai SRI LANKA .National roads ; . .. - a, Railways 0 Selected towns and villages ®. National capitao Disirict boundaries Provinciot boundaries .- ,) - Agro-climatic zonal boundaries F ~'A '] --Rivers A F r Elevation (in feet): nfl 0-100 -r Lj " 2 ) ' 100 40 0 6 Tolaimannar I \ -m 3000- 5000 1'> i\ L' O er 5000 lavura-i:!ta/ /' is KILOMETERS \' <' ('4 '" j \ OltoMILES .1 'a,,unli) , T- P n t F Tl;.tFl ; Trincomalee o - / ieda.ca/cro,a ,, rv V .- i2 f-/ P.1 _7h1 Jr t1 '| : "// ;1 B' /~ 0 ).LAt.j , .it j j,, y~ *- *y1"t'---:'~ r Kaaodid) rorn\./ C r I u'qiaI t nott,Y r ta o v )-tticlof 1 rh I jI S Ta 4 D RY< ' doutdaiot hoowy on thisnnpo\ &filntpr o ,,,;Sf ttto#ssJS WonddBtttk todIttillMt.o, tty\ I' jtdomtntott to. t otto of anetoty torrtantty evdormnotttttttt) .. ", r^ l INDIA \ Aata/EyTd I A- Z6' COOB N "il . , a amatt W,l B.W, -f .. ;~iI tf Z ! \ i, - ,hh t h. .." ,., SRI LANKA ,T. d.-.... -d-~Ih. -. 81's -9- Mortality The most significant feature of the mortality trends in Sri Lanka is the unusually rapid fall in death rates in the immediate post-war years. Between 1946 and 1949, the crude death rate fell from 19.8 to 12.4, the infant mortality rate from 141 to 87, and the maternal mortality rate from 15.5 to 6.5. These unprecedented advances have been attributed mainly to eradication of malaria, extension of health services in the rural areas and improved nutrition. The mortality levels continued to fall, and today the crude death rate and maternal mortality rate are as low as observed in most developed countries. Yet, infant mortality still remains high--about 37 per thousand births. Another feature of mortality in Sri Lanka is its sex differential. Until the early 1960s, Sri Lanka was one of the few countries where life expectancy at birth was higher for males than for females; subsequently, as shown in Table 1.3, further improvements in mortality turned the advantage to females, creating a difference of about 3 years in 1971. The decade of the 1970s saw very slow progress in decline of mortality levels. This was to be expected given the already low levels at which they had stabilized. Nevertheless, these levels are still below those achieved in developed countries. It has been suggested (Meegama, 1980) that worsening economic conditions and food shortages in the late 1960s and early 1970s and increased fatality from diseases requiring specialized medical care have held back the progress of mortality decline in Sri Lanka. These same factors have decelerated the mortality decline in many other developing countries (Gwatkin, 1980). -10- Table 1.3: Expectation of Life at Birth, Selected Years 1945-71. Period Male Female 1945-47 46.8 44.7 1962-64 63.3 63.7 1971- 64.2 67.1 Economic Development The economy of Sri Lanka for the century prior to gaiuing independence (1948) has been described as a dual economy, export-import economy or dependent economy. These descriptions take into account the existence of a modern sector which was based on the large commercial plantations--tea, rubber, coffee, and coconut--in the Wet Zone (see Map) started by the British, and the traditional sector, consisting of cultivation of paddy and other highland food crops by the peasant. Side by side with the organized modern sector based on foreign capital and Indian immigrant labor, there existed the traditional sector where the peasants carried on a subsistence form of agriculture and had no economic relationship with the estate sector (Corea, 1975; Snodgrass, 1966). In a century's time, the foreign owned plantation sector had developed into a highly productive and remunerative sector of the country's economy. The large profits yielded by these plantations supported the commercial and financial institutions located in Colombo and provided the revenue required by the British Government to administer the colony. Up to the 1930s, the non-estate rural areas which covered the traditional sector were relatively neglected. However, with the coming of universal adult suffrage (1932) and independence (1948), all succeeding governments sought to compensate for the past neglect of the rural areas by introducing increased welfare measures together with income transfers in the form of food subsidies and extended social services. Along with these measures, the infrastructure in the rural sector of the country was developed by the provision of roads, schools, hospitals, irrigation facilities and cheap transport services. This was seen as a concerted attempt by all post-independence governments to bridge the gap between the modern plantation sector and the traditional rural sector based on subsistence agriculture. At the same time, a systematic and increasing effort was made 'by the government to develop the Dry Zone (See Map) of the country by means of the restoration of ancient irrigation tanks and channels. Peasants were to be resettled in colonies where the restored tanks now made rice cultivation possible. During this period, the economy was relatively 'open' with very few restrictions on trade and foreign payments. The early 1960s, however, saw the end of the period of 'open' economy with the imposition of exchange control measures, individual licensing of imports and a complete banning of luxury goods. The policy of extending the area under rice production by restoration of ancient irrigation works in the Dry Zone continued through the '60s and '70s. In addition, the growth of subsidiary food crops was also encouraged. During the period 1970-77, this policy was continued with a greater emphasis on increasing public sector control, particularly in agriculture and trade. The Land Reforms of 1971-75 imposed constraints on private ownership of land and brought much of the plantation sector under state control. The subsequent neglect of the plantations and drop in production of tea, rubber 12 - and coconut--the main export commodities for over a century--naturally resulted in a drastic loss of foreign exchange. With the formation of a new government in 1977, a new era in the history of the economic development of the island was inaugurated. In a complete departure from the earlier economic policy of rigid government controls, market forces and the private sector began to play an increasing role, encouraged by reforms which included unification of the exchange rate at a depreciated level and the adoption of a flexible exchange rate policy, import liberalization, decontrol of most prices, and a diversion of resources from consumption to investment. As a result of these measures, in the 1978-81 period the government was successful in stimulating a massive increase in investment. Contributing to the economic uplift was a substantial flow of foreign aid and an increase in national savings which enabled the economy to double its investment during the period as compared to the preceding 1970-77 period. Social sectors, however, were not neglected by the new policy of economic growth. The government was, and continues to be, mindful of the poorer segment of its people, retaining welfare programs and consumer subsidies for the needy. Although it is true that during this period rising prices reduced the real value of welfare assistance, it is also true that the increased economic activity boosted employment and wages among these households. Apart from consumer subsidies, the government has not eliminated other free and subsidized services in the fields of education, health and sanitation. Family Planning Program Sri Lanka was one of the first countries in Asia to recognize the 13 - effect of rapid population growth on national development and the need for policies and programs to reduce the high rate of growth of its population. It may be stated that all post-independence governments have recognized this fact and have accepted the need for family planning--as well as the difficulties attendant upon its implementation in a multiracial society. The Family Planning Association (FPA) in Ceylon was established in 1953, although family planning activities started earlier. The existing government recognized immediately the work of this voluntary association by awarding a grant in 1954, which was increased substantially in 1958. At that times the government further demonstrated its (and the country's) acceptance of family planning by entering into an agreement with the Royal Government of Sweden for a pilot project in community planning. The year 1965 was a landmark in family planning in that not only was this component accepted as an integral part of maternal and child health services, but the government had actually adopted a population policy to achieve a low rate of population growth. The Family Planning Bureau was established by the government in 1968 to supervise, coordinate, monitor and evaluate the family planning programs in the country. During this period, the major sources of external assistance to family planning in Sri Lanka were the Swedish International Development Agency (SIDA), the Ford Foundation, and the United Nations Fund for Population Activities (UNFPA). The next major advance in family planning occurred in 1977. At that time the government in its Throne Speech expressed total commitment to r a vigorous policy in family planning, stating: 1. The Government is concerned with the rate of popula.tion growth and its policy is to take all meaningful steps to curb unplanned growth of population. 14 - 2. Enhanced family planning services will be provided by the State and financial incentives with a view to controlling the population explosion will be given to individuals who practise family planning. 3. In the field of family planning emphasis of the Government will be in the field of service oriented programmes to enable moti- vated couples and individuals to receive family planning ser- vices and to undergo sterilization voluntarily. In 1978, the subject of family health was entrusted to the Ministry of Colombo Hospitals which is also responsible for directing family planning programs in the country. In March 1980, a seminar was held by the Ministry of Plan Implementation on the subject of Population and Development. An important outcome of this seminar was the promotion of family planning programs above party politics by the major political parties in the country. Although it is clear that the family planning program facilitated the reduction of marital fertility, the decline in marital fertility began before it became really active in 1968. Between 1968 and 1971 the number of acceptors per annum was low, about 50,000. It rose to about 70,000 in 1972 and continued to rise thereafter. The acceptance of modern contraception is a relatively recent development. This is evidenced, by the observation that about 40 percent of ever-users of the WFS Sample were using traditional methods, and that the use of a modern method started between 1971 and 1975 for about two-thirds of women who had ever used such a method. Of the modern methods, female sterilization has been by far the most popular. Comparison of Sri Lanka with Other Countries In terms of the relationship of social indicators to income, Sri Lanka occupies an exceptional position among other developing countries. During the last four decades, despite modest economic growth, the country has made impressive progress in adult literacy, school enrollment, life - 15 - expectancy at birth, death rate, infant mortality, birth rate, and population growth, as indicated in Table 1.4. A study (Issanman, 1980) comparing Sri Lanka's social indicators to those of 59 other countries at different income levels in the 1970s shows that during this period Sri Lanka's social indicators relative to its income were the best. Figure 1.2 illustrates the relative advantage that this country enjoyed. Table 1.4: Selected Social Indicators. Indicators 1946 1953 1963 1973 Adult literacy (%) 58.0 65.0 72.0 78.0 School enrollment (% ages 5-14) 41.0 58.1 65.0 86.0 Life expectancy (yrs.) 43.0 56.0 63.0 66.0 Infant mortality (per 1000) 141.0 71.0 56.0 46.0 Death rate (per 1000) 19.8 10.7 8.6 7.7 Birth rate (per 1000) 37.4 38.7 34.3 27.9 Natural population growth rate (%) 1.8 2.8 2.6 2.0 Population growth rate, including migration (x) 2.3 3.3 2.5 1.6 It is commonly argued that the roots of Sri Lanka's social development lie in her people's strong commitment to welfare-oriented policies and human capital development. This country has accomplished more in meeting the basic needs of its people than most countries far ahead of it in per capita income. By way of illustration, the governments during the last 15 years have spent nearly as much as half its current expenditure on food subsidy, health and educational programs. Sri Lanka's educational' system is remarkably well-developed in comparison to other Asian countries. More importantly, it has been equally accessible to men and women. In the 1950s adult literacy and in particular female literacy was among the highest Figure 1.2 Sri Lanka's Social Indicators in Relation to Expected Values for Its Per Capita Income ($130) in 1975 (The Curves Show Expected Values at Any Given Income Level) LIFE EXPECTANCY AND INCOME FERTILITY AND INCOME 750 2200 700 2000 - Sri Lankan Expected Vclue 1800 650 - + Sri Lankan Actual Value 1600 - 600 - 20 4 .t 61400 0 11200 8500 - 800 140 + r aknAts au Sri Lankan Actual Value c 800 ~ 400Sri Lankan Expected Value 450 300 - 400 -600 - ,350 I I I I I i I 1 1 1 1 1 I 400 . 1 I I I I I I I 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 70C Per Capita Income, U.S. $ Per Capita Income, U.S. $ INFANT MORTALITY AND INCOME ADULT LITERACY AND INCOME 1800 T 1000 1600 - + 800- 1400-SrLaknAtaVau Sri Lankan Actual Value 700- 120060 1000 500 a+ 800 400 - Sri' Lankan Expected Value 600. 300- + Sri Lankan Expected Value 200- 400 100 200- 0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700 Per Capita Income, U.S. $ Per Capita Income. U.S. S Source: This figure is reproduced from Isenman (1 980). Basic Needs: The Case of Sri Lanka. WNorla Development Vol. 8. - 16 - Wanld Bank-24872 - 17 - in Asia, although Sri Lanka's relative advantage has diminished somewhat as other nations make progress in this area. This widespread, near-universal literacy rate was probably a key factor in initiating and sustaining the fertility decline in Sri Lanka. "High education and social welfare standards and concomitant aspirations for a better material life in conjunction with a deteriorating economic situation and increasing unemployment especially among the better educated provide a plausible explanation for the dramatic fertility decline in the early 19708" (Alam and Cleland, 1980). The island nature of Sri Lanka, its cultural and racial heterogeneity, as well as its exposure to imposed foreign influence by successive invaders set the country apart from most other expansive, homogeneous and landlocked countries. These unique circumstances, combined with an efficient network of transportation and communication, make for easy diffusion of social development efforts and concomitant norms and aspirations. Objectives of the Present Study The Sri Lanka World Bank Fertility Survey (SLWBFS) was the second in a series of surveys conducted by the Department of Census and Statistics to provide the government with data required for population policies and for effectively planning, monitoring and evaluating- the existing family planning program. Undoubtedly, the demographic data derived from this survey will not only provide internationally comparable data on fertility, but will also serve to augment the data base of the existing social and demographic statistics. Further, this survey, like its predecessor the WFS, has served the purpose of establishing a cadre of well-trained personnel in the - 18 Department of Census and Statistics to carry out future surveys of a similar nature. The immediate and most specific objectives of this survey are: 1. To identify the factors responsible for the rise in the mean age at marriage--the principal mechanism of fertility decline and differentials recorded in the World Fertility Survey (1975) and other studies. The mean age at marriage for both males and females has shown a dramatic increase in recent years. In order to assess the determinants of age at marriage and also of fertility, this survey collected more socio-cultural information, such as preferred traits of prospective spouses, prevalence and nature of the dowry system, patterns of marriage arrangement, family obligations, ideal ages at marriage, and importance of the availability of housing and regular income. 2. To provide a longitudinal data set to measure the changes in fertility levels of women interviewed at the Sri Lanka World Fertility Survey for the period 1975-1979 and to identify the factors that have contributed to the change in fertility levels, if ary. 3. To undertake a comparative study of the fertility declines of Sri Lanka, Kerala and Karnataka. Kerala, a state in India is known to be socioeconomically similar to Sri Lanka and to have experienced a parallel fertility decline in recent decades. Karnataka, on the other hand, is another Indian state where fertility has remained high. Thus, the study was designed to provide situations for comparison and contrast in understanding the fertility transition in developing countries. Finally, the ultimate purpose of this survey is to draw upon the experience of these countries in order to formulate pragmatic guidelines for designing population policies. Organization of the Report The remainder of this report is organized into several chapters, beginning with Chapter II which is a description of data. This chapter presents an overview of the study design, field work, matching problems and data quality. Chapter III is an analysis of the trends and determinants of age at marriage, while Chapter IV is a discussion of fertility levels and - 19 - differentials. Chapter V examines fertility preferences. Chapter VI is an analysis of sterilization, including an examination of the covariates and timing of this most widely used method of contraception. Chapter VII discusses the current use of contraception with particular attention to covariates of current use. The report concludes with Chapter VIII, which summarizes the findings of the survey on determinants of fertility in Sri Lanka and delineates what remains unknown and ambiguous. 20 - CHAPTER II DATA The Sample This study is based on a subsample of the World Fertility Survey sample,, In the WFS, the country was first divided into six zones on the basis of regional socioeconomic characteristics as follows: Zone I: Metropolitan Colombo; Zone II: Southwestern lowlands consisting of Kalutara, Galle, Matara and Colombo (excluding Metropolitan Colombo) districts; Zone III: Part of Dry Zone consisting of Anuradhapura, Polonnaruwa, Moneragala, Hambantota, Puttalam, and a part of Kurunegala (Kurunegala North) and Amparai (Sinhala Amparai) districts; Zone IV: Part of the Dry Zone composed of Trincomalee, Batticaloa and part of Amparai (Muslim Amparai) districts. This zone has the highest concentration of the Moor population and a high proportion of Sri Lankan Tamils; Zone V: The Northern part of the Dry Zone consisting of Jaffna, Mannar and Vavuniya districts. This area is the traditional homeland of the Sri Lankan Tamils; Zone VI: The South Central Hill country consisting of Kandy, Matale, Nuwara Eliya, Badulla, Kegalle, Ratnapura and a part of Kurunegala (Kurunegala South). [More than 85 percent of the Indian Tamils in Sri Lanka are estate residents in this area.] Map 2.1 shows the districts and six zones of Sri Lanka. The six zones were further subdivided into seventeen strata for sampling purposes. Zone 1 was divided into four strata, three within Colombo city and the other LU 0 NJ LUL L) . I-V c 0 0 c'z 4....K.. ... .. ..... .... - 22 - consisting of an area outside the city of Colombo. Zones II, III, and VI were divided into three strata each--Urban, Rural and Estate. Zones IV and V were divided into two strata each--Urban and Rural. WFS was based on a two-stage design. In the first stage a systematic sample of 750 census blocks was drawn--200 from Urban, 500 from Rural and 50 from Estates. In the second stage a sample of housing units was drawn within each selected block from a list of all housing units. Finally, within each selected housing unit, all households were included in the sample, and all ever-married women aged 12-49 were interviewed in detail. In this survey a subsample of 550 census blocks from the 750 blocks of WFS was selected. We kept the number of Estate sector blocks intact, as the Estate population has interesting demographic and socioeconomic characteristics. The total number of Urban and Rural blocks was reduced to 500. This was divided between Urban and Rural sectors using the same proportion of Urban to Rural blocks as in WFS. Thus, 143 Urban blocks and 357 Rural blocks were selected. These, in turn, were apportioned among the zones within Urban and Rural sectors using the same proportions as used in WFS. The selection of the sample census blocks was made according to probability proportional to size. The number of sample blocks selected for each stratum are given in Table 2.1. From each of these selected sample blocks the same WFS household addresses were drawn into the sample, and all the WES respondents and other residents who met the eligibility criteria at these addresses were interviewed. Table 2.1: Selected Census Blocks by Zones and Strata. Total Urban Stratum Rural Stratum Estate Stratum No. No. Selected No. No. Total Selected in Total No. Selected Total No. Selected T'otal No. Selected No.of in WFS Survey No.of Selected in No.of Selected in No.of Selected in Zone C. B. 1979 1979 C. B. in WFS Survey C. B. in WFS Survey C. B. in WFS Survey 1 2,358 120 86 2,358 129 86 -- -- -- -- -- II 14,780 130 95 2,346 28 20 11,679 94 67 755 8 8 III 9,255 140 102 455 8 6 8,602 128 92 198 4 4 IV 2,784 80 57 486 l4 10 2,298 66 47 -- -- V 3,866 80 57 811 1.8 13 3,055 62 44 -- -- -- VI 22,304 200 153 1,113 12 8 16,957 150 107 4,234 38 38 Total 55,347 750 550 7,569 200 143 42,591 500 357 5,187 50 50 4+ - 24 - Questionnaires Four questionnaires were used in the survey. These were: 1. Individual Questionnaire-W for ever-married women aged 12-49 years; 2. Individual Questionnaire-M for never-married males aged 28-34 years; 3. Individual Questionnaire-F for never-married females aged 25-34 years; 4e Individual Questionnaire-H for husbands aged 30-39 years. Individual Questionnaire-W: This consisted of Household Schedule, Data Sheet from WFS, and 8 sections on detailed information. The Household Schedule was primarily an instrument for listing the household members, for purposes of identifying respondents who would be eligible for individual interviews. Eligibility for the individual interview depended on two criteria. First, the woman had to be between 12-49 years of age. Second, she had to be ever-married. These were applied to all those who were interviewed in WFS and were living at the household and all new members of the household who satisfied the above eligibility criteria. The Data Sheet attached to Individual Questionnaire-W consisted of basic information with respect to WFS respondents, such as number of live births, date and name of last live birth, and whether pregnant at the time of the WFS interview. This was intended primarily for cross-reference purposes at the time of the interview. Individual Questionnaire-W consisted of the following 8 sections: 1. Respondent's Background and Maternity History 2. Marriage 3. Contraceptive Knowledge and Use - 25 - 4. Fertility Regulation 5. Work History 6. Husband's Background 7. Children's Employment 8. Family Assets The questionnaire was designed and arranged in such a way as to make the interviewer's job less difficult and to make it easy for the respondent to supply the required information. Within the sections mentioned above the individual questions were arranged in a logical sequence. Individual Questionnaires-M, F, and H: Individual Questionnaires-M, F, and H consisted of only one section each, containing questions about marriage or marriage plans. These short questionnaires were used to collect information from never-married males aged 28-34 years, never-married females aged 25-34 years and husbands aged 30-39 years of currently-married women who were eligible for interview with the Individual Questionnaire-W. Field Work General Organization of the Survey: The survey was conducted by the Department of Census and Statistics of the Ministry of Plan Implementation, in association with the World Bank and UNFPA. The Director of Census and Statistics functioned as the National Director and had ultimate responsibility for the survey. The actual implementation of the survey was assigned to the Demography Unit of the Department with an Assistant Director in charge, assisted by three Statisticians and the necessary technical and administrative support staff. - 26 World Bank provided the services of Professor N. K. Namboodiri for technical aspects such as questionnaire design and training of field staff. Organization of Field Work: Based on the experience of WFS, it was decided that one coordinator should be appointed to each of Sri Lanka's nine provinces. Either one or two female supervisors, each in charge of one team of interviewers, were assigned to each coordinator. Each team generally consisted of 5 female interviewers and a male interviewer to interview the husbands. The staff assignments for each province is shown in Table 2.2. Table 2.2: Staff Deployment for Field Work. Number of Number of Province Coordinators Supervisors Teams Interviewers Western 1 2 2 10 Central 1 2 2 11 Southern 1 2 2 11 Northern 1 2 2 11 Eastern 1 2 2 10 Northwestern 1 2 2 11 North-central 1 1 1 6 Uva 1 1 1 6 Sabaragamuwa 2* 2 2 8 Total 10 16 16 84 *Included additional coordinator. The coordinators were to play two major roles in the survey: (1) administrative and organizational and (2) supervisory. The responsibilities of coordinators included arranging accommodation and food for the teams, arranging transportation whenever necessary, briefing interviewers on field conditions, returning completed questionnaires to headquarters, and other - 27 supervisory duties such as spot-checking, scrutinizing questionnaires on a sample basis at the end of each day, and dealing with non-response. One car or jeep was made available to each coordinator by the Department of Census and Statistics for the movement of interviewers during the field operations. If required, another vehicle was supplied by the goverument agent of the administrative district. A supervisor's responsibility Was generally to guide the interviewers and their teams and, more specifically, to: - assign the day-to-day work to the interviewers in consultation with the co-ordinator; - ensure correct identification of sample census blocks and housing units; - carry out spot checks to ensure that interviews were of the desired quality; - scrutinize the completed questionnaires as far as possible at the end of each day (supervisors were asked to strictly adhere to the instructions given in checking the responses for internal consistency); and - ensure that all control documents were correctly filled in. Field operations commenced on 26 June 1979 and were completed by 5 August 1979. Wide publicity was given for the survey through the national newspapers and radio before the commencement of field operations. Field Staff: Recruitment and Training: Altogether, 84 interviewers were selected, in most cases from WFS participants. In addition to the statistical investigators attached to the Department of Census and Statistics, 14 temporary interviewers who were interviewers in WFS were recruited after an in-person assessment of their aptitude for the type of task they were expected to perform. Nine interviewers were males and were appointed specifically to interview male respondents. The interviewers' work was supervised by 16 supervisors, most - 28 - of whom had served as junior coordinators in WFS. A list of interviewers, supervisors, and coordinators appears in the Appendix. The supervisors and interviewers were trained by Professor N. K. Namboodiri of the University of North Carolina-Chapel Hill, whose services were provided by the World Bank. The supervisor training program lasted one week and was followed by a field demonstration. During this training, the following were carefully discussed: the Supervisors' Manual, the Interviewers' Instructions, the control sheets, scrutinizing questionnaires, field work, and the allocation of duties to interviewers. The interviewer training program extended over a 2-week period. It dealt thoroughly with survey objectives, featured question-by-question explanation of the questionnaires (with special attention to "skip" instructions), stressed detailed familiarity with the questionnaires themselves, and included tape recorded interviews for demonstration purposes. A series of written tests were administered, and it was determined that the performances of the trainees were up to a reasonable standard. After the completion of classroom studies, interviewers were required to conduct a few interviews in Colombo. The training provided was very intensive, and every aspect of the Interviewer's Training Manual was adequately dealt with. Non-Response: Response rates by zone are given in Table 2.3. - 29 - Table 2.3: Response Rates by Zone. Eligible Respondents Eligible Respondents Interviewed Response Rate (Z) Zone W F M H W F M H W F M H I 646 95 73 190 611 80 53 149 94.6 84.2 72.6 78.4 II 803 138 96 208 755 109 62 140 94.0 79.0 64.6 67.3 III 755 109 62 140 607 36 23 147 98.1 90.0 74.2 89.6 IV 619 40 31 164 811 25 14 186 99.6 96.2 82.4 90.3 v 814 26 17 206 570 62 32 130 97.6 95.4 100.0 92.9 VI 584 65 32 140 1475 185 91 348 96.7 88.9 78.4 84.3 All 4991 572 363 1321 4829 497 275 1100 97.4 86.9 75.8 83.3 The response rate of ever-married women aged 15-41, was fairly high, the highest level occurring in Zone IV. The response rates of unmarried males, females and husbands were relatively low, probably due to the high mobility of unmarried males, females and husbands. Matching Problems: As indicated in Table 2.3, altogether 4,829 interviews with ever-married women were completed in the survey. Of these, 1,425 women were declared new respondents based on interviewers' declarations regarding the matter, or because matching identification could not be located in the WFS file. If there was at least one live birth recorded at the WFS, an attempt was made to match the sex and birth date of the last birth recorded at the WFS with the corresponding information for the first birth recorded at this survey. A difference in the birth date exceeding 6 months was considered unacceptable. Any discrepancy with respect to information on sex was also- considered unacceptable. Based on these criteria, a total of 415 cases were - 30 declared not matched. This left 2,989 cases to be treated as re-interviews. Preliminary Evaluation of the Quality of Data It is common to find a gross underenumeration of children in the age group 0-4, with corresponding overenumeration of those in the age group 5-9. Digit preferences frequently exist between the ages 10-60; at ages 70 or more, overreporting of age is the main problem. The percentage distribution of children, ages 0, 1, 2, 3, and 4 to the total of 0-4 group for both the survey and the 1981 Census are shown in Table 2.4. Table 2.4: Percentage Distribution of the Sri Lanka Population by Single Years in the Age Group 0-4 and Sex Ratios for the 1981 Census and the Household Schedule Data from the Survey. Age 0 1 2 3 4 0-4 Survey 1979 Males 20.7 17.9 20.2 17.3 -23.9 100.0 Females 18.0 20.2 19.0 18.5 24.3 100.0 Sex-Ratio 119.3 92.5 110.9 97.4 102.7 104.2 Census 1981 Males 21.8 19.1 20.4 19.4 19.3 100.0 Females 21.9 18.9 20.3 19.3 19.6 100.0 Sex-ratio 104.3 105.5 105.0 104.9 102.9 104.5 Source: 1981 data are from 10% sample tabulation of Census of Population and Housing, 1981. This gives an idea of the relative qua'lity of the age group 0-4. One would expect a sex-ratio of about 105-106 at age 0, gradually declining to 103-104 in the age group 3-4, and an age distribution of smoothly 31 decreasing percentages as age advances. The 1981 Census data conforms to this pattern reasonably well. In the survey data, exceptionally high values of sex-ratios are recorded at ages 0 and 2. This fact and the low values of sex-ratios at neighboring ages suggest age misreporting. Table 2e5 shows the age-sex composition of the population recorded in the survey along with that of 1981 Census. Comparison of the two age-sex compositions reveals that there is not much difference between the two, with the exception of the 0-4 and 15-19 age groups. Table 2.5: Percentage Distribution of the Sri Lanka Population by Five-Year Age Groups and Sex (1981 Census and the Household Schedule Data from the Survey). Survey 1979 Census 1981 Age ...._,.:_._ Group Males Females Total Males Females Total 0-4 9.9 9.6 9.7 12.5 12.4 12.5 5-9 11.6 11.0 11.3 11.3 11,4 11.4 10-14 12.0 11.7 11.9 11.4 11.4 11.4 15 19 11.6 12.2 11.9 10.8 109 10.8 20124 10.3 110 10.o7 10.0 104 10.2 25-29 9.1 9.5 9,3 8.4 8e7 8o6 30-34 6.8 7o4 7.1 7.5 7o6 7o6 35-39 5.2 5.6 5.4 5.6 5.7 5.6 40-44 4.8 404 4.6 4.8 4.6 4.7 45-49 4.0 3.9 3.9 4.1 4.1 4.1 50-54 3.5 3.2 3.4 3.8 3.5 3.7 55-59 3.2 3.5 3.3 2.9 2.8 2.8 60-64 2.6 2.4 2.5 2.4 2.2 2.3 65-69 1.9 1.8 1.9 1.8 1.7 1.7 70-74 1.4 1.2 1.3 1.3 1.1 1.2 75+ 1.9 1.5 1.7 1.4 1.4 1.4 32 - Table 2.6: Blended Percentages of Myers' Index for the 1971 and 1981 Censuses and Household Sctedule Data from the Survey. Census 1971 Su-lvey 1979 Census 1981 Digit Males Females Males Females Males Females 0 12.94 14.31 9.66 8.97 12.42 12.85 1 8.73 8.10 8.66 8.84 9.23 8.84 2 10.30 9.81 10.89 11.02 10.01 9.70 3 9.26 8.88 8.33 8.46 9.98 10.03 4 8.82 8.37 12.33 11.83 9.35 9.12 5 12.84 13.76 8.91 9.11 10.90 11.21 6 9.20 9.10 9.92 9.53 9.44 9.42 7 8.38 8.09 10.07 10.65 8.79 8.69 8 11.17 11.53 8.41 8.45 11.07 11.41 9 8.38 8.07 12.82 13.15 8.81 8.73 Index 14.48 19.18 12.22 13.29 8.80 11.00 Myer's Index for the censuses 1971, 1981 and for the survey are shown in Table 2.6. The indices for the survey show that age reporting in the survey is satisfactory. The indices for both males and females are less than the 1971 Census, indicating an improvement of age reporting in the survey. But quality of age reporting in the 1981 Census is better than that of the survey. Also note that in the survey the digits 0 and 5 have been avoided for both males and females. - 33 - CHAPTER III AGE AT MARRIAGE Introduction It is well known thz& the transition from high to low fertility occurred in Europe in association with changes in nuptiality patterns (Hajnal, 1965). During the pretransition stagre, the most common pattern was one of early marriage and high marital fertility. Late marriages came into vogue in the fifteenth and sixteenth centuries, as a result of which moderate fertility levels of about 35 births per thousand population became common. It was not until the 1870s that marital fertility began to be controlled deliberately on a widespread scale in any population. Fertility reduction in today's less developed countries has also been associated with a shift from early marriage to late marriage. This was true, for example, in Taiwan, Korea, and West Malaysia in the 1960s (Cho and Retherford, 1973), and in Sri Lanka during approximately the same period (Fernando, 1976). Despite the recognized importance of shifts in nuptiality patterns in the causal nexus of fertility declines, the study of nuptiality as such has seldom been given any serious attention in fertility investigations. The two-wave study in Sri Lanka, on which this report is primarily based, is not a major exception to this general pattern, but relatively greater attention has been given in the study to factors affecting age at marriage. Reported below are the results of a preliminary analysis of the - 34 - factors affecting age at marriage of women in Sri Lanka. Attention is focused primarily on the information collected in the 1979 interview, although some use is made of the data from the 1975 interview also. Trends as well as differentials in age at marriage are analyzed. The next section outlines the nature of the data. This is followed by a brief examination of the trend iV age at marriage. The theoretical considerations behind the analysis of differentials are then outlined, followed by the presentation of the results of the analyses. The Data In the 1975 as well as in the 1979 interview, the marital status of members of the household was ascertained with the completion of the household schedule. This information was used in determining the eligibility of the household members for personal interviews. As already mentioned, individual questionnaires were used in 1979 to interview ever-married women of age 12-49, and husbands and never-married males and females of specified ages in random subsamples of addresses included in the World Fertility Survey, Sri Lanka, 1975. In the marriage-history section of the individtial questionnaire for ever-married women, each currently-married woman was asked in 1975 and 1979 the month and year in which she started to live with her husband, and all women were asked to give the month and year of the beginning and end of each of their previous marriages, if any. In addition, a number of questions were asked covering background factors such as religion and ethnic group, and such attributes as the following: 1. whether the woman's first marriage was arranged by her parents or other relatives; 2. whether she participated in the labor force before her first - 35 - marriage; 3. whether she was in school during the 12 months preceding her first marriage; 4. whether family obligations intervened to affect the timing of her first marriage; 5. the financial status of her husband at the time of the marriage decision; and 6. the place where she and her first husband lived during the first 12 months of her marriage. Husbands were asKed questions parallel to those asked of ever-married women, while never-married men and women were asked about their marital intentions, their educational attainment, their work experience, their family obligation, etc. In this chapter, attention is focused on the information collected from ever-married women (including those re-interviewed and those interviewed for the first time in 1979). See, however, Appendixes 3A and 3B. Analytic Strategy Survey data on nuptiality can be analyzed with an emphasis on the percentage of persons who remain single or with a focus on the age at marriage of those who marry. In the first approach, a birth cohort (actual or hypothetical) is followed up, noting at each age the proportion of those who remain single (not married) From these figures one estimates the cumulative proportion of persons who remain single by any given age. These figures correspond to the survival function of the mortality table. From the survival function, a number of summary measures can be calculated, e.g., the median age, or the age by which 50 percent of the starting cohort enter ever-married state. The technique of life table analysis with covariates -36 - can be applied to examine the impacts of various demographic, social and economic factors on the survival function. In this chapter, however, we shall be using a different approach, which focuses on the age at marriage of the ever-married persons. Before examining the data from the survey, it may be useful to review some data from sources other than the survey, to get an idea about the trend in age at marriage. Trend in Ag°- at Marriage As already mentioned, the mean age at marriage has shown a dramatic increase in recent years in Sri Lanka. Some of the details of this development are reviewed in this section. Table 3.1 presents the changes over the period 1946-1975 in the propensity to remain single up to a given age for males and females. The changes during the period have been much more dramatic for females than for males, Thus the percentage of females never married in the age group 20-24 steadily increased from 29.4 to 60.6 over the period (1946-1975), while no such drastic change occurred in the case of males in any age group. The singulate mean age at marriage calculated from the percent of males and females single in successive age groups shows the following pattern: 1946 1953 1963 1971 1975 Females 20.7 20.8 22.1 23.6 24.8 Males 26e8 27.0 28.0 27.9 28.4 - 37 - Table 3.1: Age Pattern of Never-Married Males and Females, Sri Lanka, 1946, 1953, 1963, 1971 and 1975. 1946 1953 1963 1971 1975 Age % Never-married Females 15-19 75.3 75.7 85.0 89.5 93.2 20-24 29.4 32.5 41.3 53.1 60.6 25-29 11.8 12.8 17.1 24.6 31.9 30-34 6.6 7.5 8.3 10.9 13.7 35-39 4.3 5.4 4.8 5.6 5.8 40-44 4.1 5.0 4.3 4.3 4.6 45-49 3.4 4.4 3.9 3.6 2.1 Singulate Mean Age at marriage* 20.66 20.81 22.10 23.55 24.81 Age % Never-married Males 15-19 98.7 98.77 99.0 99.4 99.7 20-24 80.5 83.5 84.7 86. 3 88.5 25-29 43.4 45.4 50.5 52.6 57.5 30-34 22.4 21.7 26.1 25.5 26.8 35-39 12.5 11.8 13.1 13.7 12.2 40-44 9.3 8.7 10.4 9.4 6.7 45-49 7.6 7.6 7.2 7.9 7.1 Singulate Mean Age at marriage* 26.80 27.03 27.66 27.86 28.43 Source The % never-married figures are from Fernando (1975) for 1946, 1953, 1963, and 1971 and those for 1975 are from the WFS First Report Sri Lanka for 1975. *Singulate mean age at marriage for those married before age 45. Percent single at age 45 czl.-culated by averaging the figures for 40-44 and 45-49. - 38 The consistent increase in the figures for females is in sharp contrast to the near stasis of the figures for males. A consequence of this time trend of the mean age at marriage has been a narrowing of the gap between the mean ages of the bride and the groom. Assuming that the male age at marriage will continue to remain more or less the same as it has been over the past several decades, and given the universal preference for the bridegroom' s age to be higher than that of the bride, it seems unlikely that the female age at marriage will be pushed up much further. Other analysts differ in their predictions, however. For instance, Fernando (1976) believes that a further dramatic increase in female age at marriage is in the offing. Fernando's prediction is based on an evaluation of trends in age-sex ratios of the form nMx+t/nFx, where the denominator stands for the number of females in the age group x to x+n and the numerator for the males in the age group x+t to x+t+n. To appreciate the problem in using such ratios to speculate about the age at marriage, let us examine the age distribution of males and females never married as of 1975 (Table 3.2). The figures in Table 3.2 show more never-married males than never-married feiaales (15,543 versus 13,835), indicating that in an open marriage market there is no shortage of males. If attention is confined to persons of age 15 years and older, the picture is not any different (6,660 males for 5,048 females). However, if no female of age x is willing to marry a male of age less than x+5, then there is a shortage of males; - 39 - Table 3.2. Never-Married Persons in the Household Population by Age and Sex (World Fertility Survey, Sri Lanka, 1975). Age Group Male Female 0-4 2,827 2,731 5-9 3,045 2,961 10-14 3,011 3,140 15-19 2,752 2,517 20-24 2,065 1,465 25-29 1,041 627 30-34 376 199 35-39 161 76 40-44 69 49 45-49 74 22 50-54 37 16 55-59 31 20 60-64 12 12 65-69 14 14 70-74 11 7 75+ 17 24 Total 15,543 13,835 e.g., Table 3.2 shows only 3,908 males of age 20 and older as opposed to 5,048 females of age 15 and older. The restriction that the bridegroom must be at least 5 years older than the bride implies then that 1,140 of the never-married females of age 15 and older (as of 1975) will have to remain lifetime spinsters, unless non-monogamous marriages become popular. Ruling out such an eventuality, demographic response to the situation just described is to relax the restriction concerning the age gap between the bride and the groom. Thus, if slightly over 20 percent of the 5,048 females of age 15 and older are willing to marry males who are senior to them by less than five years, the market could accommodate all females. It should be emphasized that such relaxation does not necessarily imply an upturn or downturn of mean age at marriage. Thus, a woman who is currently of age 15 - 40 - may marry a man who is currently of age 18-not now, but four years from now. The point, then, is that the sex imbalance of the kind implied in age-sex ratios of the form nMx+t/nFx does not by itself imply an upward push or downward pull of the mean age at marriage, unless it affects the propensity to remain lifetime spinsters (bachelors). There is no indication that any drastic change has occurred in Sri Lanka over the recent decades in the propensity of females to remain spinsters until the end of the reproductive age span (exami,ne the trend in the percent of females never married in the age groups 40-44 and 45-49 in Table 3.1). Therefore, the age-sex ratios of the form nMx+t/nFx are not reliable indicators of future trends in age at marriage. To summarize, the relevant facts are: (a) the average age at marriage of females has been rising for some time now; (b) the average age at marriage of males has remained more or less constant for the past several decades; and (c) only a very small percentage of females remain spinsters beyond age 50. From these observations, one may speculate the following: 1. The average age of the bridegroom is unlikely to increase dramatically in the coming years. 2. Unless a drastic revision occurs in the preference regarding the gap between the ages of the bride and the groom, the near-stasis in the mean age of the groom is likely to block further sharp increases in the mean age of the bride. 3. Any further upward shift that might occur in the mean age of the bride is therefore less likely to result from disproportionately more women marrying at advanced ages than from disproportionately fewer women marrying at very young (less than 16) ages. If these speculations prove true, then the fertility impact of future changes in nuptiality pattern is likely to be rather modest. The data collected in the 1979 survey do shed some light on the - 41 - speculations just mentioned. But before examining the data, it may be useful to discuss some of the theoretical considerations which guided the collection of the data. Theoretical Considerations It has become fashionable when analyzing the timing of marriage to start with the notion that persons marry when the utility expected from marriage exceeds that expected from remaining single (Becker, 1974). A slightly technical rendition of this idea follows. Assume that consumers combine goods and services bought in the market with non-market time and other resources (e.g., approval by relevant others) to produce various commodities for consumption. An example of such a commodity is one's standing in the community. The production and consumption of such commodities shape one's life style. Now, assume that each person has a utility function of the form U(Zl, - esZn), where zj's are commodities of the kind just mentioned, and that each commodity has associated with it a production function which gives the quantity of the commodity that can be produced with given market goods and services and non-market time and resources. While each member of a family has his/her own utility function, seldom do the family members carry out their lives as independent individuals. They work out among themselves a system of transfer of resources so that the fortunes and misfortunes of each member are shared with each other. This makes it possible to imagine that the family acts as a single unit and that, as an abstraction, the utility function of the family head may be taken as a family utility 42 Figure 3.1: S- and NI- Curves Relating Consumpt-ion to Daughter's Mlarriage. utility S t t Time - 43 - function. The behavior of the family can then be studied in terms of the head's utility function, the corresponding production functions, and the associated resource constraints. A postulate is now added stating that the family head chooses that life style (i.e., the production and consumption of that stream of commodity bundles) which gives maximum expected utility within the constraints of the available resources. With risk-neutrality, this behavioral criterion simplifies to the maximization of the present value of the stream of commodities expected to be consumed over time. The application of this idea to the timing of marriage is illustrated in Figure 3.1. The S-curve corresponds to the life plan according to which the family head's daughter (focusing on one daughter for the moment) remains a lifetime spinster; the other curve corresponds to a life plan which includes marrying her off at time t. The M-curve is below the S-curve until time t because, according to the life plan corresponding to M, less consumption and more saving is called for until the daughter is married off. The M-curve is shown above S, after t, because through the marriage alliance the family gains much more than the costs it has had to bear in bringing about the marriage, and also because the continued presence of an unmarried daughter of marriageable age at home is a cause of embarrassment the avoidance of which is certainly a desideratum. Keeping in the background the general ideas outlined above, an attempt will now be made to identify a number of factors each of which has a potential bearing on the timing of marriage of the family head's daughter/son. It is important to recognize at the outset that in Asian countries, - 44 arranged marriages are the rule rather than the exception. This does not mean that the principals (the prospective bride and groom) are given no say in mate selection or other matters. In fact, nowadays virtually no family will finalize a marriage for a daughter/son without consulting her/him. It is also true that the principals are increasingly demanding and are receiving more say in mate selection and related matters. Furthermore, 'love' marriages (as opposed to arranged marriages) are becoming slightly more common than in the past. Despite all these developments, the family continues to be centerstage in every scene, and the senior members of the family, particularly the head and his/her spouse, make all the important decisions affecting the lives of the family members. The crucial role of the family in marriage negotiations can be appreciated if it is realized that when a girl/boy is married off what takes place is transferring one person from the family of origin to another existing family, rather than the creation of a new family unit as is usually the case in the West. The centrality of the family in marriage negotiations and the post-marital life of the couple makes it mandatory that when one searches for factors affecting the timing of marriage, the cost-benefit implications of each factor for the family and its members be kept in mind. Preferred Attributes of the Spouse: The first factor we shall consider is the marriage market itself. Whether the prevailing market offers a suitable match for the individual involved is obviously a crucial factor affecting the probability and timing of marriage, Of course, the availability of a suitable match is a function of what the searcher is looking for, but commonly, the girl's family looks - 45 - for a boy who is of the proper exogamousl and endogamous2 group and whose family is one of good standing. As for personal attributes, the girl's family looks for such things as the means to support a new wife and children (when they are born); an educated girl's family prefers an educated boy to be her husband; and so on. The boy's family looks for counterpart attributes in the girl and considers her demeanor, health, looks, etc. It is not unusual to keep the acceptable level of various attributes high at first, to be lowered if the market conditions dictate. Also, it is not unusual for each searcher to be flexible enough to accept limited inadequacies in certain respects, provided that they are compensated for by better-than-adequate ratings in other respects; for example a low level of education might be compensated for by a large dowry. The mutually satisfactory outcome of any marriage negotiation is, of course, that both families (the girl's and the boy's) should feel that they have gained. In reference to Figure 3.1, the M-curve should be perceived as going above the S-curve by both families. This is not as impractical as it might first appear. Thus, if one family has wealth but low status in the caste, an alliance with another family which needs the wealth and can impart higher status could usually make both families feel that they are better off with the alliance than without it. lExogamous rules require marriage outside a group. Thus, village exogamy requires one to marry outside one's village. 2Endogamous rules require marriage within a group, e.g., within one's own caste. - 46 - Whether one gets what one wants from the marriage negotiations is therefore a function of the diligence with which one makes the search and the adeptness with which one carries out the negotiations (playing up the strengths and playing down the deficiencies, and so on). The fact that nearly everyone gets married seems to indicate that the market is working. But the empirical question remains whether there is any discernible tendency among any population groups to hold on to unrealistic standards with respect to social, economic, or demographic attributes of a prospective bride or groom and whether such unrealistic positions affect the probability or timing of marriage. Dowry Payment: The second factor we shall consider is related to the role of the dowry payments. Providing a suitable dowry to a marriageable daughter is commonly a difficult problem. Even so, virtually no family skimps on the dowry for fear of jeopardizing their child's future. Although no hard data exist, it is generally believed that for generations the dowry payments have wiped out in many cases the family savings, forced families to sell or mortgage part or all of the family land holdings, and pushed many a family to the brink of bankruptcy. Yet the practice goes on, and today it is not uncommon for upper-class families to demand, in return for a well-educated, well-placed bridegroom, steep cash payments, supplemented with gold and jewelry and consumer durables such as refrigerators and automobiles; lower-class families also expect cash payments and consumer durables, albeit of a more modest description. Anyone who reads the matrimonial advertisements in the local newspapers in Sri Lanka gets the unmistakable impression that the practice of dowry payment is alive and thriving. The - 47 - empirical question is whether dowry-related problems affect the probability or timing of marriage. Prerequisites of the Marriage Principals: The next factor worth considering is prerequisites, if any, to which the principals subscribe and their possible effect on the probability or timing of maxriage. As already mentioned, the principals are nowadays demanding and are receiving more say in matters related to mate selection. It is conceivable that the same applies to the timing of marriage. The greater independence and autonomy of the younger generation implied by these developments may result in the principals setting preconditions for their marriage. Thus, a girl might set as a precondition for her marriage the arrangement of a neolocal residence3 for her to move into after marriage. A boy who works in a distant town might insist that his marriage be postponed until he finds a suitable place of residence near his work site so that he and his wife can live together after marriage rather than be separated.4 Other preconditions might include such things as the completion of higher education or the securing of a suitable job. The pdint is that honoring such conditions often means delaying the timing of the marriage. Not honoring them, on the other hand, entails certain social costs. 3Neolocal here means separate from both the bride's and the groom's kin. 4It has not been unusual in Asian countries for the new husband to leave his wife behind and go back to work in a city, if financial or other factors dictate. - 48 - Thus, if a daughter's marriage is arranged without regard for her legitimate requests, the result may be for her M-curve (Figure 3.1) to fall below the S-curve, which is not desirable from anybody's standpoint. The empirical question is to what extent considerations such as the ones mentioned above are given importance in connection with the timing of marriage and the degree to which such considerations affect the decision. Delayed Marriage: Next, we consider the social cost of not marrying off a daughter. The presence in the household of a girl of marriageable age, unwed and unspoken for, is embarrassing to the members of the household because her continued presence in the household is most likely to be interpreted by the relevant others as indicative of failure on the part of the family in its duty or, worse, of some serious defect afflicting the girl. Such an image entails a very high social cost to the family and the girl. Certainly the family is under tremendous pressure to marry her off as soon as she reaches marriageable age, unless there is some defensible reason for putting off the marriage. One way some families handle the situation is to have the girl continue in school when marriage negotiations are underway, or, if she has finished schooling, to have her enter the labor force. Both these strategies are bound to win social approval, given the importance society today attaches to higher education and work participation. Those who use continued education as a stop-gap arrangement are likely to discontinue schooling when the date of marriage nears. The chances are that in such cases the age at marriage will be lower than usual. Those who pursue higher education for the parpose of human-capital - 49 - development or who participate in the labor force for the purpose of career development are likely to be self-selected with respect to independence and autonomy and are likely to be prone to seek individualistic rewards for themselves. Such individuals are likely to marry relatively late, as are those who seek such persons as their spouse. Parenthetically, two points are worth noting in this connection: (1) A family might find it advantageous to delay the marriage of a working daughter because she is an asset to the family rather than a liability. Bear in mind, the ready-cash part of the dowry required to marry off an educated, working female is relatively low, since her lifetime income is counted as part of the dowry. Thus, the average family has an incentive to encourage female education and pre-marital female labor-force participation, both of which tend to delay marriage. (2) Also note that female education might affect the timing of marriage through its effect on the marriage market. If there are a good many eligible bachelors in the marriage markets education can be an asset to a marriageable female; but if educated, eligible bachelors are scarce, it may be difficult to find a suitable match for an educated female. In other words, sometimes there is the danger of educating oneself out of the marriage market. Group-Specific Norms: The next factor to be mentioned is the social context of the family. The social cost of not marrying off a daughter is no doubt a function of the cultural groups in which the family holds membership. Norms regarding the timing of marriage and the sanctions for deviating from the norms are also group specific. Furthermore, the practice of giving dowry, the customs regarding post-marital residence, etc., also vary from one - 50 - subgroup (e.g., ethnic group) to another. Family Structure and Sibling Obligations: Finally, the structure of the family unit itself may have implications for the timing of marriage of a family member. The more children a family has, the greater the likelihood that some of the children will marry late. This is primarily because each has to wait for his/her turn in terms of seniority, particularly among females. Also, it is conceivable that an older sibling may postpone his/her marriage if he/she feels obligated to help younger brothers or sisters and believes that getting married might stand in the way of meeting those obligations. To summarize, the factors identified as affecting the timing of marriage are: 1. preferred attributes of the spouse, and the implications thereof with respect to finding a suitable match; 2. problems related to dowry payment; 3. problems related to meeting requirements, such as the arrangement of a neolocal residence for the newlyweds; 4. non-marriageability/delayed marriage: education as a stop-gap arrangement or for human capital development, and work participation before marriage; 5. religious affiliation and ethnic group identification; 6. family structure, particularly with regard to the number of brothers and sisters and sibling obligations. All but the first factor listed above are given attention in the analysis reported in the next section. Data pertaining to Factor 1 were collected only from never-married males and females. The results of the analysis of that data are shown in the Appendix. Results of Empirical Analyses The main focus of this section is on the analysis of differential,s - 51 - in age at marriage. But first we briefly examine the trends revealed by the data on hand. Trends: Despite the fact that samples selected for fertility surveys with the usual eligibility restrictions (e.g., ever-married and of age 12-50) suffer from truncation bias with respect to age at marriage, one can get a rough idea about recent changes in the age pattern of marriage by comparing the percent distributions by age at marriage of recent marriage cohorts (Table 3.3). The coefficient of dissimilarity.5 between the distributions of women who were married within 5 years of and 5-10 years before the 1975 survey is 15.70; the corresponding figure comparing women who were married 5-10 years before the 1975 survey with those who were married 10-15 years before the survey is 11.07. These figures indicate a tendency for the distribution of marriages by age at marriage to become increasingly dissimilar. The 1979 data, however, show a reversal of this tendency. The coefficient of dissimilarity between the distributions of those who were married 5-10 years and 10-15 years before the 1979 survey is 16.00 (which is of the same order of magnitude as 15.70, the coefficient of dissimilarity between the distributions of those who were married within 5 years of and 5-10 years before the 1975 survey). A comparison of the distributions of 5jf Pil and Pi2, i = 1, 2, ..., n, are two percent distributions with 'Pil = ZPi2 = 100, the coefficient of dissimilarity is equal to one-half of the sum of the absolute values of the differences Pli - P2i, i = 1, 2, .... n. This figure represents the percent of individuals who should be redistributed in order to make the two distributions similar. 52 - Teble 3.3: Percent Distribution of Successive Marriage Cohorts by Age at Marriage (1975 and 1979 surveys, Sri Lanka). Marriage duration Age at marriage in completed years <15 15 to 17 16 to 21 21 to 26 25+ All (i) 1975 survey* < 5 years 0.25 1lc72 25.61 25.31 27.11 100.00 5 to 9 years 7.59 20.00 25.69 25.38 21.34 100.00 10 to 16 years 13.64 22.82 27.90 19.1 n6 16.49 100.00 15 to 19 years 12.10 29.01 25.94 19.69 13.26 100.00 20 to 24 years 16.46 30.00 25.87 18.81 8.85 100.00 (ii) 1979 survey** < 5 years 1.18 17.19 29.22 29.32 23.09 100.00 5 to 9 years 2.40 18.61 28.88 29.68 20.43 100.00 10 to 14 years 7.02 28.99 26.31 22.61 15.07 100.00 15 to 19 years 14.04 27.67 28.21 19.16 10.93 100.00 20 to 24 years 18.15 28.74 26.93 19.52 6.66 100.00 * Source WFS Sri Lanka First Report Table 1. 1. 2. ** The Figures reported in this table are based on the full sample, which includes reinterviewed as well as first-time interviewed ever-married women. - 53 those who were married within 5 years of and between 5 and 10 years before the 1979 survey, however, gives a coefficient of dissimilarity of only 3.00. Thus, there is some indication that the tendency for the age pattern of marriage to change has slowed down. Has any social or demographic group participated more than others in this development? The analysis of the differentials reported below sheds some light on this question. Differentials: Table 3.4 shows the variation of mean age at marriage by each of a number of selected characteristics, taken one at.a time. Women of age 25-49 years and those of age 25-29 are separately considered in Table 3.4. In both groups only those who were first married before age 25 are included. This last restriction and the minimum age restriction together make the statistical risk of first marriage the same for the women compared. The differences in the patterns for the two age groups indicate recent changes. The highlights of the patterns discernible in Table 3.4 include the following: 1. The mean age at marriage is about the same for women in the age groups 35-39, 40-44, and 45-49. But the mean is higher for women of age 30-34, and still higher for women of age 25-29. This reflects the continued increase in the mean age at marriage over time. 2. The differentials by pre-marital work experience show that work participation and delayed marriage are associated with each other. This association was not observed in the 1975 data (see WFS First Report, p. 63). 3. A positive association is discernible between education and the age at marriage. The relationship is clearly non-linear for the younger cohort, an outcome of a dramatic increase in the mean age at marriage of the uneducated. As already mentioned, education may affect the age at marriage in various ways, and it may very well be that the relationship between the two variables runs in both directions, the timing of marriage - 54 - Table 3.4: Mean and Standard Deviation of Age at Marriage of Ever-Married Women Age 25-49 and 25-29 and Married Before Reaching Age 25, by Selected Characteristics (Sri Lanka, 1979). Age 25 to 49 Age 25 to 29 Characteristics N Mean S.D. N Mean S.D. Age 25 to 29 784 18.7 4.07 784 18e7 4.07 30 to 34 777 18.3 4.00 35 to 39 710 17.5 4.40 40 to 44 614 17.6 3.50 45 to 49 560 17.6 3.39 Pre-marital work No work 2411 17.6 4.07 536 18.3 4.40 Some work 1035 18.8 3.52 248 19.6 3.08 Education (wife) No education 752 16.8 3.65 107 18.0 3.37 Grades, 1 to 5 1409 17.3 4.31 301 17.6 5.05 Grades, 6 to 9 925 18.8 3.23 254 19.0 3.07 Higher 360 21.1 2.41 122 21,2 2.25 In school during 12 months before marriage Yes 278 18.6 3.16 70 18.1 3.09 No 3168 17.9 4.01 714 18.7 4.15 Religion Budhist 1738 18.5 3.95 372 19.3 4.00 Hindu 961 17.6 3.59 210 18.4 3.14 Muslim 438 16.4 4.26 122 17.0 5,58 Christian 302 18.6 3.84 79 19.2 2.97 Ethnic group Sinhalese 1880 18.5 3.98 403 19.2 3.93 Sri Lanka Tamil 825 17.7 3.64 192 18,7 3.10 Indian Tamil 291 17.6 3,38 63 18.7 3.25 Sri Lanka Moor 421 16.3 4.25 118 17.0 5,65 Mode of mate-selection "Love" marriage 815 18.6 3.63 244 18.9 4.46 Arranged 2383 17.7 4,04 483 18.6 3.98 Mixture 248 18.5 3.83 57 18.8 2.97 No. of proposals considered Two or more 761 18.8 3.51 157 19.4 3.13 Only one 2119 17,6 4.07 .463 18e5 3,99 Engagement period < 12 months 2614 17.7 4.13 571 18,5 4.36 > 12 months 832 18.8 3.20 213 19.2 3,13 Residence after marriage Wife's parents 1189 17.5 4.40 288 18.4 4.48 Husband's parents 1366 18.3 3,66 294 19.3 3.01 Neolocal 891 18.1 3,67 202 18.2 4.66 Financial shape of bridegroom Regular income 2809 18.0 3.82 629 18,8 3.78 No regular income 637 17.7 4.48 155 18.2 5i06 - 55 - Table 3.4 (cont'd) Age 25 to 49 Age 25 to 29 Characteristics N Mean S.D. N Mean S.D. Number of older brothers 0 1580 17.9 3.80 320 18.4 3.10 1 1006 18.0 3.81 249 18.8 4.45 2 510 18.1 4.86 121 18.9 5.81 3 222 18.4 3.48 61 19.0 3.26 4 or more 128 18.5 3.68 33 19.1 2.96 Older brother's obligation He postponed his mar. 346 18.3 3.37 82 19.1 3.07 No postponement 1520 18.1 4.22 382 18.8 4.89 Number of older sisters 0 1755 17.9 3.98 350 18.1 4.99 1 989 18.0 3.31 243 19.0 2.91 2 439 18.2 5.26 113 19.3 3.17 3 151 18.4 3.51 48 19.1 3.42 4 or more 112 18.6 3.31 30 19.8 3.38 Older sister's obligation She postponed her mar. 38 18.5 2.98 11 18.4 2.69 No postponement 1653 18.1 3.94 423 19.2 3.08 Obligation to older sister R's marriage postponed 33 19.6 2.49 11 19.6 1.97 No postponement 1657 18.1 3.94 422 19.1 3.10 Number of younger brothers 0 1219 17.8 3.92 227 18.7 3.31 1 888 17.9 3.69 218 18.8 3.23 2 609 18.3 3.83 156 18.5 5.17 3 375 17.9 5.17 97 18.7 5.83 4 or more 355 18.4 3.33 86 18.6 3.10 Obligation to younger brother R postponed her mar. 41 19.3 3.44 10 19.9 3.25 No postponement 2186 18.1 3.97 547 18.7 4.36 Number of younger sisters 0 1259 17.8 3.68 248 18.9 3.26 1 902 18.1 4.59 199 18.7 6.03 2 608 17.9 3.47 142 18.3 3.23 3 359 18.3 4.51 104 18.8 3.08 4 or more 318 18.2 3.12 91 18.5 2.75 Obligation to younger sister R postponed her mar. 41 18.8 3.37 16 19.1 2.47 No postponement 2142 18.1 4.11 520 18.6 4.44 - 56 - affecting the educational attainment for some women and the latter affecting the former for others. In the 1979 survey, respondents were asked whether they were enrolled in school any time during the 12 months preceding their wedding. Among women of age 25-49 years, those who said they were had an average age at marriage of 18.6 years, whereas those who said they were not had an average age of 17.9 years. The pattern is reversed, however, for women of age 25-29 years. Of course, for some women who stay in school until marriage, school enrollment is a stop-gap arrangement while marriage negotiations are underway; for others, the timing of marriage may be adjusted to the attainment of a pre-planned level of education (e.g., graduation from college). It is possible that the age at marriage is lower on the average for the former group than for the latter, but the available data do not permit separating the two groups. 4. Religious differentials show that the Buddhists and Christians have the highest, and the Muslims the lowest mean age at marriage. The ethnic group differentials parallel the religious differentials. 5. Arranged marriages tend to occur at a relatively younger age than do 'love' marriages. The difference is not clear-cut in the younger age group, however. Part of the relationship observed may be attributable to a common antecedent, namely, education. If this is the case, when education is taken into account the relationship between the mode of mate selection and the age at marriage may differ from what Table 3.4 reveals it to be. 6. Respondents who reported 'love' marriages were asked whether parents or other relatives took any steps in arranging marriage for the respondents. The mean age at marriage for those who said 'yes' and 'no' are 19.0 and 18.4, respectively, for those of age 25-49, and 19.8 and 18.5, respectively, for those of age 25-29. This pattern is consistent with the notion that some women take matters into their own hands after waiting for a time for their parents to arrange marriages for them. 7. Respondents who reported that their relatives were involved in one way or another in arranging marriage were asked whether two or more proposals were considered. Not surprisingly, the mean age at marriage of those who said 'yes' was higher than that of those who said 'no' (18.8 years versus 17.6 years for women of age 25-49, and 19.4 versus 18.5 for women of age 25-29). 8. Those who said that two or more proposals were considered were asked whether any proposal was dropped because of dowry-related problems. Those who answered 'yes' did not differ from the rest with respect to mean age at marriage, indicating that, - 57 - contrary to our expectation, there is probably no dowry effect on the age at marriage. 9. In an effort to test whether problems of finding a suitable neolocal place of residence affect the age at marriage, respondents were asked where they resided during the first 12 months after their marriage. It was assumed that finding a neolocal residence is probably difficult and that consequently those who succeeded in securing such a residential arrangement might have had to postpone their marriage. The figures in Table 3.4, however, do not support this hypothesis. (Women who reported neolocal residence were asked whether their marriage was delayed because of any problems in finding a suitable residence. Those who answered the question in the negative do not differ from those who said 'yes' in mean age at marriage.) 10. There is a slight tendency for the financial security of the groom--whether he had a regular source of inccme at the time of marriage decision--to influence the age at marriage, particularly among the younger cohorts. This relationship is probably attributable to a common antecedent, namely, the groom's age. (Women whose husbands were reportedly financially secure were asked whether the marriage decision would have been any different if their husbands had not been so at the time of the decision. A substantial minority said that the decision would have been delayed or the prpposal dropped if the groom's financial position had not been satisfactory. Apparently, the groom's financial security is being given increasingly more importance; thus, when times are bad marriages are likely to be delayed more than when times are good.) 11. A positive association is discernible between the number of older brothers and sisters and the age at marriage, this being true for the older as well as the younger cohorts. No corresponding association involving younger brothers and sisters is discernible, however, for the younger cohort. 12. There is a tendency, although not widely prevalent, for a woman's marriage to be delayed because of the delay in marrying off one of her older sisters (see the pattern under 'obligation to older sister'). But this obligation-to-wait effect is not discernible for the younger cohort. The tendency to postpone self's marriage because of the obligation to younger brothers and sisters is discernible for the younger cohort also (see the patterns under obligation to younger brother and obligation to younger sister). Multivariate Analysis: From univariate analysis we now turn to multivariate analysis. Table 3.5 presents incremental R2's attributable to each of a number of. - 58 - variables. Three separate analyses are presented in the table, one for women of age 25-49, one for women of age 25-29 and one for women of age 35-39--in all cases confining attention to women whose age at first marriage is less than 25 years. Two sets of incremental R2ts are presented in the table, one set labeled A1R2 (the incremental R2.attributable to each variable when it is entered in the order indicated) and the other set labeled A2R2 (the incremental R2 attributable to each variable when it is entered last). First, considering women of age 25-49, we note that the following variables exhibit significant net relationship (A2R2) to the age at marriage. 1. Pre-marital work 2. Education 3. Year married6 4. Mode of mate-selection (whether the marriage was arranged by parents, by couple themselves, or a mixture of the two modes) 5. Residence after marriage (with the wife's parents, with the husband's parents, or neolocal) 6. Economic security of the husband at the time of marriage decision (whether the husband had a regular source of income at that time) 7. Staying in school until marriage (whether the wife was enrolled in sghool any time during the 12 months preceding the marriage) 8. Obligation to sibling (whether marriage was postponed on account of a sibling or whether a sibling--older brother or sister--postponed his or her marriage on account of the respondent) 6Appendix 3A examines the problem of interpreting the effect of this factor. - 59 - Table 3.5: Incremental R2 Attributable to Individual Regressors (i) When Introduced in the Regression in the Sequence Indicated (A1R2), and (ii) When Introduced Last (A2R2); Multivariate Analysis of Age at Marriage; Ever-Married Women of Selected Age Groups, Married Before Reaching Age 25 (Sri Lanka, 1979). Age (25, 49) Age (25,29) Age (35-39) Regressors Age at mar. <25 Age at mar. <25 Age at mar. <25 A R2 AR ARR AR2 ARR 2 1 2 1 2 AR 3 3 3 13 Pre-marital work .020 .006 .024 .001 013 .001 3 3 3 3 Religion .036 .000 .048 .003 .017 .001 2 3 3 Ethnic Group .002 .001 .004 .004 .002 .001 3 3 3 3 3 3 Education .108 .040 080 .010 .084 .020 3 3 3 3 3 3 Year married .143 .1433 .643 .588 295 .292 Engagement period .000 .000 .000 .000 .004 .0052 3 3 2 1 Mate-selection mode .005 .004 .002 .001 .001 .001 Value of Dowry e000 .001 .000 .000 .001 .001 3 3 2 2 2 2 Residence after marriage .002 .002 .002 .002 .007 .007 Econ. security of husband 2 2. . at nmarriage .001 .001 000 .000 .002 .003 3 3 In school till marriage .005 .004 .000 .001 .000 .000 Obligation (old brother's) .000 .000 .000 .000 .002 .001 2 2 Obligation (old sister's) .000 .000 .000 .000 .006 .006 Obligation (to old sister) .000 .000 .000 .000 .000 .001 Obligation (to young brother) ;O0202 .0011 .001 .001 .001 .003 Obligation (to young sister) .000 .000 .000 .000 .004 .004 .326 .806 .439 N 3037 732 630 2 3 prob. bet-ween .05 and .1; prob. between .01 and .05; p-rob. b-elow. .01. - 60 - For religion and ethnic group, only AIR2 is significant. Ethnic group has some effect independent of religi. i (AlR2 is significant for Ethnic Group when entered in the regression after entering Religion), but when all variables are simultaneously entered, Ethnic group is left with no net effect. The multiple R2 is .326, indicating that two-thirds of the variation in age at marriage remains unaccounted for. Now comparing the younger cohort (of age 25-29) with its older counterpart (of age 35-39), the following similarities and differences are discernible. Pre-marital work has a significant gross effect--A1R2 is significant--in both groups. But the net effect (see A2R2) is considerably weaker for the younger cohort and negligible for the older one. Both religion and ethnic group show significant A1R2 as well as A2R2 for the younger cohort, but for the older one only A1R2 is significant. Why the net effect is significant for the younger cohort but not for the older one is not clear. Both Education and Residence after marriage have significant net effects, for the younger as well as the older cohorts. - The length of the engagement period (whether under one year or more) shows a significant effect only for the older cohort. - The mode of mate selection has a significant effect only for the younger cohort. - Economic security of the bridegroom shows significant net effect only for the older cohort. - For the older cohorts there 'is a significant obligation-to-wait effect; but that effect is absent for the younger cohort. Obligation toward younger brothers or sisters has no effect for younger or older cohorts. - The value of dowry shows no significant effect for any of the age groups. It is tempting to speculate that families view it as a violation of the norm to permit dowry-related problems to block marriage arrangements. It seeps that even if there are difficulties in raising the money needed to give the usual dowry, the family will raise it somehow because of their conviction that it is their duty to do so. - 61 - Turning to the parameter estimates (Table 3.6), the discernible patterns include the following: - Women with pre-marital work experience tend to marry late. (The effect is significant only for women of age group 25-49. When examined separately for women of age group 25-29 and 35-39, the effect, although showing the same sign as for those in the age group 25-49, is not statistically significant.) - Education has a curvilinear effect for young women, but more or less a linear effect for older women. - The year-of-marriage effect is positive.7 - Shorter engagement period on the; average signifies marriages at yc:nger age. But this effect is significant only for the older cohort. - 'Love' marriages occur at relatively younger ages (significant for women of age 25-49). This is a reversal of the crude relationship revealed in Table 3.4. - Bridegroom's economic security signifies relatively older age for the bride. This effect does not persist when examined separately for birth cohorts. - Staying in school until marriage signifies relatively younger age for the bride. This effect also does not persist when examined separately for birth cohorts. - The (positive) obligation-to-wait effect is present for the older cohort but absent for the younger cohort. The (positive) obligation-to-younger-brother effect is present for women of ages 25-49 but is not statistically significant when estimated separately for birth cohorts. 7See Appendix 3A. 62 - Table 3.6: Parameter Estimates, Multiple Regression Analysis of Age at Marriage--Ever-Married Women of Selected Age Groups, Married Before Reaching Age 25 (Sri Lsnka, 1979). Age 25-49 Age 25-29 Age 25-29 Age at mar. <25 Age at mar. <25 Age at mar. <25 Regressors 0 (t) B (t) B (t) Pre-marital work No work -.686 (5.02) -.272 (1.82) -.357 (1.03) Some work (ref.) Religion Budhist .057 ( .21) .494 (1.66) .701 ( .83) Hindu -.218 ( .76) -.220 ( .76) .605 ( .78) Muslim -. 61 ( .58) -3.736 (3.07) .733 ( .71) Christian (ref.) Ethnic group Sinhalese .959 ( .88) -3.804 (3.02) -.458 ( .38) Sri Lanka Tamil .371 ( .35) -3.300 (2.73) .575 ( .93) Indian Tamil .412 ( .39) -3.922 (3.23) .000 ( - ) Sri Lanka Moor (ref.) Wife's education None -2.834 (11.39) -1.001 (3.73) -2.656 (4.15) Grades 1 to 5 -2.860 (13.06) -1.280 (5.92) -2.158 (3.76) Grades 6 to 9 -1.843 ( 8.72) - .875 (4.39) -1.246 (2,14) Higher (ref.) Year married .187 (25.26) .786 (46.13) .501 (17.71) Engagement period !12 months - .113 ( .80) - .116 ( .75) - .771 ( 2.24) >12 months (ref.) Mode of mate selection "Love" marriage - .723 ( 2.97) - .240 ( .91) - .554 ( .90) Arranged - .102 ( .46) .125 ( .51) - .192 ( .35) Mixture (ref.) Value of dowry .002 ( 1.50) .001 ( 1.26) .004 ( 1.12) Residence after Mar. With wife's parents-.273 ( 1.78) - .007 ( .04) - .915 ( 2.34) With husb.'s par. .172 ( 1.18) .376 ( 2.29) - .048 ( .13) Neolocal (ref.) Economic shape husband was in at the time of marriage decision Regular income .349 ( 2.35) .202 ( 1.24) .671 ( 1.79) No reg. income(ref.) Wife in school till marriage Yes -1.02 ( 4.46) - .341 ( 1.43) - .453 ( .73) No (ref.) Older brother None .001 ( .01) - .043 t .30) - .148 ( .47) He postponed his marriage .180 ( .91) .118 ( .55) .381 .79) No postponement (ref.) Older sister None .273 ( .09) .578 ( .34) .273 ( .93) She postponed her marriage .407 ( .78) .238 ( .46) 3.785 ( 2.58) No postponement(ref.) Older sister None .367 ( .12) - .657 ( .39) .000 ( - ) R's mar. postponed .536 .92) .380 ( .66) - .916 ( .76) No postponement(ref.) Younger brother None .084 ( .65) - .180 ( 1.17) - .577 ( 1.76) R's mar. postponed 1.314 t 2.23 .652 t 1.10) .276 ( .16) No postponement(ref.) Younger sister none .012 ( .09) - .094 ( .63) .623 ( 1.92) A's mar. postponed - .002 ( .00) .105 ( .23) 1.096 ( .56) No postponement(ref.) - 63 - Table 3.7: Incremental R2 Attributable to Individual Regrei;sors [(i) when introduced in the regression in the sequence indicated (A1R2) and (ii) when introduced last (A2R2)] multivariate Analysis of Age at Narriage--Ever-Married Women of Age 25 Years or Over and married at Age Below 25, by-Ethnic Group (Sri Lanka, 1979)1. Regressors Sinhalese Sri Lanka Tamil Indian Tamil Sri Lanka Moor AR2 AR2 A I2 AR2 AR2 AR2 AR2 AR2 AR R A P R A1R ARR A1R A2R 1 2l 2212 2 3 3 3 3 3 3 2 Pre-marital work .020 006 .018 .006 .036 .020 .007 .002 Education .1223 0533 .169 055 .013 .001 .070 .028 3 3 3 3 3 3* 3 3 Year married .108 109 .177 165 .246 248 250 240 Engagementperiod .000 .000 .000 .000 .000 .000 .000 .000 3 3 2 2 Mate-selection mode .008 .006 .000 .000 .023 .021 006 .007 Value of dowry .000 .000 .001 .001 .001 .000 .000 .000 3 3 Res. after mar. .005 .005 .004 .004 .002 .001 .002 .002 3 3 Econ. sec. of hus. .000 .000 .009 .009 .003 .003 .004 .005 at marriage 3 3 2 2 2 2 In school till mar. .006 .006 .004 .004 .003 .002 .008 .006 Obligation 2000 .000 * °000 .003 .004 007 .0122 (older brother's) Obligation 3 3 (older sister's) .002 .001 .003 .002 .027 019 008 .001 Obligation (to older sister) .000 .000 .000 .000 7- .001 .001 Obligation 2 2 (to younger brother) .001 .001 .006 .005 .008 .007 .004 .001 Obligation (to younger sister) -o00 .000 .001 .001 .003 ..003 .005 .005 R .274 .393 .369 .372 Mean age at marriage 18.66 17.73 17.74 16.46 *No variation. - 64 - Table 3,8: Parameter Estimates; Multiple Regression Analysis of Age at Marriage-Ever-Harried Women, of Age 25-49 and Harried at Age Below 25, in Different Ethnic Groups (Sri Lanka, 1979)o Sinhalese Sri Lanka Indian Sri Lanka Tamil Tamil Moor Regressors 8 (t) 8 (t) 8 (t) B (t) Pre-marital work No work -.643 (3.86) -.830 (2.69) -1X362 (2.74) -.545 ( .98) Some work (ref.) Wife's education None -3.013 (9.35) -3.145 (6.94) -.260 ( .06) -1.938 (1v62' Grades 1 to 5 -2.920 (10.68) -2.817 (7.31) -.235 ( .05) -2.971 (2.59) Grades 6 to 9 -1.883 (7.37) -1.777 (4.67) -1.083 ( .24) -1.320 (1.12) Higher (ref.) Year married .160 (15.76) .193 (13.82) .218 (9.70) .286 (11.48) Engagement period < 12 months -.118 ( .70) -.218 ( .71) -.145 ( .16) -.221 ( .39) > 12 months Mode of mate selection "Love" marriage -.896 (2.63) -.138 ( .31) -.736 (1.00) -1.590 (1.21) Arranged -.271 ( .82) .057 .16) .735 (1.30) -1.536 (1.97) Mixture ('ef.) Value of dowry .002 ( .71) .002 (1.18) -.002 ( .02) .004 ( .27) Residence after mar. With wife's parents -.425 (1.98) -.492 (1.78) .136 ( .26) .070 ( .14) With husband's parents .221 (1.19) -.550 (1.83) .286 ( .66) .534 ( .81) Neolocal (ref.) Economic shape husband was in at time of marriage decision Regular income .015 ( .08) 1.040 (3.22) .824 (1.03) .780 (1.62) No regular income (ref.) Wife in school till marriage Yes -.922 (3.55) -1.224 (2.25) -2.556 ( .87) -2.102 (1.83) No (ref.) Older brother None -.097 ( .58) -.038 ( .16) -.185 ( .48) .974 (2.29) He postponed his mar. .163 ( .62) -.134 ( .33) .931 (1.27) 1.209 (2.02) No postponement (ref.) Older sister None -.500 ( .16) -.199 ( .85) +.634 (1.68) -.690 (1.74) She postponed her mar. .901 (1.49) -1.829 (1.59) 5.919 (2.70) -3.458 ( .90) Older sister None .778 ( .25) -- -- -- R's mar. postponed .332 ( .49) .008 ( .01) -- 1.045 .57) No postponement (ref.) Younger brother None -.043 ( .25) .438 (1.82) -.452 (1.11) .229 ( .51) R's mar, postponed .954 (1.35) 2.879 (1.77) 1.916 (1.12) -1.304 ( .47) < No postponement (ref.) Younger sister None .067 ( .38) -.235 ( .98) .091 ( .22) .751 (1.65) R's mar. postponed .377 ( .53) -.457 ( .28) -1.685 ( .99) 1.330 ( .48) no cases - 65 - Now, since ethnic group differences are very pronounced with respect to age at marriage and since in many respects each ethnic group may be regarded as a population in its own right, there is some merit in excamining factors affecting age at marriage separately for the major ethnic groups in the country. Table 3.7 presents the incremental R2's attributable to selected individual regressors. The estimated regression coefficients are shown in Table 3.8.. There are clear-cut differences between the ethnic groups with respect to which factors exhibit a significent effect. 1. Pre-marital work experience has no significant effect for the Sri Lankan Moors, while it has a significant net effect for all the other ethnic groups (Sinhalese, Sri Lankan Tamils, and Indian Tamils). 2. Education has no significant effect for the Indian Tamils (neither A1R2 nor A2R2 is significant, probably because of low variation in education for the group), but it has a highly significant effect for all the other ethnic groups. 3. The mode of mate selection is not significant for the Sri Lankan Tamils; but it is for the Sinhalese and Indian Tamils. For the Sri Lankan Moors neither A1R2 nor A2R2 is significant, but the regression coefficient attached to arranged marriage is significantly different from that attached to the mode of mate selection in which the couple as well as their parents are involved (see Table 3.8). 4. Residence after marriage is significant for the Sinhalese only. 5. Economic security of the bridegroom does not seem to matter for any ethnic group other than the Sri Lankan Tamils. 6. The timing of marriage does not seem to be significantly associated with staying in school until marriage as far as the Indian Tamils are concerned, but that is not the case for any of the other three ethnic groups. 7. Family obligation does not seem to be a significant factor for the Sinhalerse, but it is for all the others. 8. From Table 3.8 one notices that the effect of education is curvilinear, more obviously so for the Sri Lankan Moors. It is one thing to invite attention to such ethnic group - 66 differences, quite another to interpret them. Only those who are intimate with the cultural peculiarities of the various ethnic groups can explain why some factors are significant for some groups but not for others. The results presented so far in this section permit only conditional (i.e., if-x-then-y type) speculations concerning the future trends in age at marriage. Thus, one might say that if higher education were to become more widespread or female pre-marital work participation more common, the age at marriage would be pushed further upward; or if arranged marriages became unpopular, the age at marriage would be pulled downward; and so on. But we have no way of predicting how soon higher education or pre-marital work participation would become widespread or how rapidly the practice of arranging marriages would give way to 'love' marriages. Hence, conditional prognoses of the type just mentioned are not helpful in guessing the likely immediate future trend in the age at marriage. The preferences people currently hold regarding ages of the bride and groom are probably more helpful in speculating whether dramatic changes in the age at first marriage are in the offing. Preferences Regarding Age at Marriage Direct data on preferences are difficult to obtain. However, one can get a rough idea by examining what people regard as ideal ages for boys and girls to enter married life. Table 3.9 shows that for females neither very early nor very late marriage is preferred by the respondents. Virtually none prefers marrying off a girl at an age below 15, and less than 5 percent consider first - 67 - Table 3.9: Ideal Age for Brides and Grooms--Ever-Married Women's Responses (Sri Lanka, 1979). Ideal Age Bride Groom 15 years 0.2 0.0 15-16 years 12.8 0.0 17-18 years 12.8 0.5 19-21 years 36.1 4.2 22-25 years 43.5 44.2 26-30 years 4.5 46.0 31-35 years 0.0 4.9 36 years 0.0 0.2 Total 99.9 100.0 N 4. - 68 - Table 3.10: Mean Age at First Marriage of Wives; the Corresponding Figure for Husbands; and Mean for Bride's and Groom's Ages Considered Ideal by the Wives, by Selected Characteristics (Sri Lanka, 1979). Mean age at first marriage Ideal age for Characteristics Wife Husband Bride Groom Age of wife 25 to 29 18.7 25.3 21.5 26.6 30 to 34 18.3 25.7 21.6 26.8 35 to 39 17.5 25.5 21.2 26.7 40 to 44 17.6 26.1 21.4 26.5 45 to 49 17.6 26.2 21.3 26.6 Education (wife) None 16.8 25.4 20.5 26.1 Grades 1 to 5 17.3 25.3 21.0 26.3 Grades 6 to 9 18.8 25.9 22.1 27.1 Higher 21.1 27.3 23.0 28.0 Religion Buddhist 18.5 26.1 22.0 26.8 Hindu 17.6 25.1 21.1 26.7 Muslim 16.4 25.7 19.6 26.0 Christian 18.6 25.6 21.6 26.7 Ethnic Group Sinhalese 18.5 26.0 22.0 26.7 Sri Lanka Tamil 17.7 25.2 21.4 27.2 Indian Tamil 17.6 25.0 20.3 25.4 Sri Lanka Moor 16.3 25.6 19.6 25.9 NOTE: The means for husband's age are based on slightly fewer cases than the other means. Data were collected for the current husband only. Hence it was necessary to restrict the calculations to currently married women who have not been married more than once. Hence the smaller N's. - 69 - Table 3,11: Mean Age at First Marriage of Wives; the Corresponding Figure for Husbands; and Mean Bride's and Groom's Ages Considered Ideal by the Wives by Duration of Marriage (Below 25 Years) and Wife's Age at Marriage (Sri Lanka, 1979). Duration of marriage Age at first marriage (wife) and the variable of 15 15 to 17 18 to 28 21 to 24 25+ interest Under 5 years Age at marriage: Wife * 16.2 18.9 22.2 28.2 Husband * 23.2 24.8 26.8 31.0 Ideal age for: Bride * 20.0 21.1 22.2 23.1 Groom * 24.9 26.2 27.1 28.2 5 to 9 years Age at marriage: Wife * 16.2 19.0 22.3 28.0 Husband * 23.8 25.3 27.0 32.0 Ideal age for: Bride * 21.1 21.3 22.5 23.1 Groom * 26.2 26.2 27.3 27.9 10 to 14 years Age at marriage: Wife 13.1 16.2 19.0 22.3 27.5 Husband 21.9 24z4 25.7 28.2 31.2 Ideal age for: Bride 20.4 21.2 21.7 22.2 22.5 Groom 26.6 26.6 27.0 27.6 27.4 15 to 19 years Age at marriage: ,Wife 12.9 16.1 18.9 22.3 27.5 Husband 24.7 24.7 26.2 28.0 32,4 Ideal age for: Bride 20.4 21.1 21.3 22.0 22.6 Groom 26.2 26.7 26.3 26.9 27.3 20 to 24 years Age at marriage: Wife 13.0 16.0 18.9 22.2 25.7 Husband 23.8 24.1 26.2 28.5 31.2 Ideal age for: Bride 20.0 21.0 21.5 21.6 22.0 Groom 25.4 26.6 26.5 26.6 26.7 Under 5 years Number of cases Wives 11 160 272 273 215 Husbands 11 150 261 262 205 5 to 9 years Wives 21 163 253 260 179 Husbands 19 149 236 245 165 10 to 14 years Wives 55 227 206 177 118 Husbands 51 216 196 166 111 5 to 19 years Wives 104 205 209 142 81 Husbands 100 183 196 135 70 20 to 24 years Wives 120 190 178 129 44 Husbands 104 167 159 109 31 NOTE: Women of all ages are included in this table. *Base less than 25 - 70 - marriage at ages 26 or above as ideal. The average ideal age for brides falls between 19.5 and 23.5 when we compare respondents in the various demographic and socio-economic groups (Table 3.10). By and large, those cultural groups characterized by low actual age at marriage for women prefer early first marriage for girls. An example is the Sri Lankan Moors. However, there is no evidence that the women consider the age at which they themselves entered first marriage as necessarily ideal under the current circumstances. This is clear from the fact that in Table 3.11 the mean ideal age at marriage for girls is higher than the mean actual age at marriage for those who married very young and lower for those who married older. Comparing the mean ideal age for brides and grooms, one notices that in all subgroups the preference is for the groom to be, on the average, 5-6 years older than the bride. Also, very few respondents prefer the groom to be above 30 years of age (see Table 3.9). The range within which the subgroup means of ideal age for groom fall is 24.0-28.5 when respondents of the various demographic and socio-economic groups are compared. It is particularly noteworthy that the mean ideal age for the bride is 22 or below for women married for 20-24 years (see Table 3.10). These women are likely to have their daughters entering the marriage market now; hence, their preferences regarding ideal age for brides presumably reflects a desire to see their daughters married in their early 20s. This seems to indicate that changes in age at marriage that might occur in the near future are likely to result from a shift towards moderately higher ages at marriage (e.g., early 20's) than from a dramatic increase in the proportion of women marrying in their late 20's or at older agese Put differently, marriage - 71 - below 17 years of age is soon likely to be a phenomenon of the past; on the other hand, few women will postpone marriage beyond age 25. Bride's age, therefore, may be concentrated in a short range centering on 22. Such developments will have very little effect on the fertility level of the population. (See Appendix 3C for figures regarding the impact of the marital factor on fertility during the period 1965-1971 and 1971-1974.) - 72 - APPENDIX TO CHAPTER III 3A. Husbands Table 3A.1 - 3A.4 3B. Expected Age at Marriage and Timing of Marriage of Spinsters and Bachelors Table 3B.1 - 3B.4b 3C. Impact of Changes in Age Pattern of Nuptiality on Fertility Table 3C.1 - 73 - APPENDIX 3A Husbands In the 1979 survey, husbands of age 30-39, currently married and living with their spouses were interviewed, with a view to identifying the factors that influenced their age at marriage. Reported herein are some of the results of the analysis based on the data thus collected. Attention is focused on husbands who have been married only once. Table 3A.1 shows the mean and standard deviation of the age at marriage of husbands, by selected social and demographic characteristics. No consistent pattern in age at marriage is discernible by current age. As for religious differentials, the Hindus, Muslims, and Christians have approximately the same mean age at marriage, while the Buddhists have a higher mean. Among the ethnic groups, the Sri Lankan Tamils and Sri Lankan Moors have a lower age at marriage. The education effect appears to be U-shaped. Family-obligations seem to have an effect on the age at marriage--some males apparently postpone their marriages on accouvt of their obligation to their siblings. Table 3A.2 shows the results of a multivariate analysis, in which all characteristics shown in Table 3A.1 (except age) are used, along with year of marriage, as explanatory factors. Religion shows a gross effect, but no significant net effect. Ethnic Group retains its effect even when all other variables are taken into account. In view of the high association between Religion and Ethnic Group, the estimated net effects of these variables should be interpreted with caution. Surprisingly, Education shows no significant effect. Among the obligation effects, only the obligation - 74 - toward a younger sister is statistically significant. A cautionary note is in order with respect to using year of marriage as an explanatory variabl'e when age variation in the data is small. Conceptually, year of marriage is supposed to capture the period effect (i.e., the effect of the socio-economic conditions characteristic of the time when the marriage took place). But, obviously, if all respondents are of the same age, their age at marriage will be the same as their year of marriage, except for an additive constant. (For age at marriage = date of marriage - date of birth = date of marriage - date of survey + age = date of marriage + a constant if the survey date and age are constants for all the respondents.) The implication of this relationship is that when age variation is small in the data, year of marriage accounts for a large proportion of the variance in age at marriage, even when there is no period effect. In the present case, age of husband is constrained to the range 30-39. How much overlap this causes between age at marriage and year of marriage is difficult to say. But it certainly is not safe to interpret the large proportion of variance attributable to year cf marriage as exclusively reflecting the period effect. One other point worth nioting is that, because of the overlap between year married and age at marriage, the use of the former as an independent variable when the latter is the dependent variable in a regression analysis amounts to controlling for the dependent variable, thereby introducing bias in the estimates of the net effects of the other independent variables. A comparison of the results presented in Table 3A.2 with those presented in Table 3A.3 is of some interest in this connection. - 75 - Note that the education effect is significant when year of marriage is ignored (Table 3A.3), but not so when year of marriage is one of the independent variables (Table 3A.2). Also note that the net effect of obligation toward younger sisters is highly significant when year of marriage is ignored (3A.2) but only barely so when year of marriage is in the regression (Table 3A.3). The parameter estimates presented in Table 3A.4 tell the same story. These differences in the results may be partly due to the bias mentioned above, and partly due to the association between the time of marriage and the other independent variables involved. - 76 - Table 3A.1: Mean and Standard Deviation Age at Marriage of Husbands of Age 30-39, Married Onlv Once and Living with Spouse, 1979. Characteristics N Mean S.D. Age 30 91 24.6 4.33 31 82 24.5 4.17 32 85 26.2 5.89 33 80 24.8 3.85 34 96 26.2 5.18 35 101 25.8 3.97 36 98 25.7 4.45 37 121 25.1 4.27 38 78 25.9 4.59 39 84 25.3 4.30 Religion Buddhist 460 26.0 4. 39 Hindu 258 24.9 4.85 Muslim 113 24.7 4.60 Christian 86 24.8 3.90 Ethnic Group Sinhalese 506 25.8 4.39 Sri Lanka Tamil 219 24.5 4.14 Indian Tamil 87 26.0 5.65 Sri Lanka Moor 107 24.6 4.67 Education None 41 25.6 4.67 Grades 1 to 5 349 24.2 4.97 Grades 6 to 9 294 25.3 4.14 Higher 241 27.3 3.64 Marriage delayed on account of older sister Inapplicable* 439 25.4 4.22 Delayed 21 27.5 4.39 Not delayed 465 25.4 4.82 Marriage delayed on * account of younger brother Inapplicable* 328 25.6 4.50 Delayed 64 26.6 3.92 Not delayed 533 25.2 4.62 Marriage delayed on account of younger sister Inapplicable* 310 25.6 4.59 Delayed . 131 26.8 4.63 Not.delayed 484 24.9 4.41 *No siblings of the specified type - 77 - Multivariate inalysis of Husband's Age at Firstt Marriage; Incremental R Attributable to Regressors (A R2 signifying Table 3A.2: Incremental R Attributable t2 a regressor i} it is entered ii the sequence shown and A2R the corresponding incremental R if the regressor is entered last)--Husbands, Married Only Once and Currently Living with Wives, 1979. Regressors AlR2 A2R2 Religion .0163 .002 Ethnic Group °0073 .OO83 Year Married .5173 .4243 Education .001 .001 Obligation to older sister .002 .001 Obligation to younger brother .001 .001 Obligation to younger sister .0021 .0021 R2 .547 1 probability between .05 and .1 2 probability between .01 and .05 3 probability below .01 - 78 - Multivariate Analysis of HusDand's Age at Marriage (A1R , the incremental R attributable to a 5egressor where it 's entered Table 3A.3 in the sequence indicated and A2R the incremental R attribut- able to the regressor when it is entered last)--Husbands Married Only Once and Currently Livi:ag with Wives, 1979. Regressors A1R2 A2R2 Religion e0163 .004 Ethnic Group .0071 .0163 Education .0753 .0713 Obligation to older sister .0062 .003 Obligation to younger brother .0072 .002 Obligation to younger sister .0103 .0lO3 1 probability between .05 and .1 2 probability between .01 and .05 3 probability below .01 - 79 - Table 3A.4:: Multivariate Analysis of Husband's Age at Marriage; Regression I Including Year Married among the Regressors and Regression II Excluding It-Husbands of Age 30-39, Married Only Once and Living with Spouse, 1979 Regressors Regression I Regression II B (t) (t) Religion Buddhist .684 (1.36) 1.298 (1.86) Hindu .474 ( .93) .431 ( .61) Muslim 2.565 (1.62) 1.571 ( .72) Christian 0.000 (.ee) 0.000 (eee) Ethnic Group Sinhalese 2.524 (1.57) 1.685 ( .76) Sri Lanka Tamil 1.911 (1.19) 1.204 ( .54) Indian Tamil 3.431 (2.08) 3.411 (1.49) Sri Landa Moor 0.000 ( ... ) 0.000 ( O . ) Year MarrIed .678 (28.95) - Education None -.044 ( .09) -2.076 (2. 93) Grz'.des 1 to 5 -.330 (1.15) -3.194 (8.51) Grades 6 to 9 -.351 (1.25) -2.022 (5.30) Higher 0.000 (.e. ) 0.000 ( e. ) Obligation to older sister No older sister .250 (1.18) -.165 ( .56) Respondent's mar. delayed .800 (1.14) 1.599 (1.64) No delay 0.000 C...) 0.000 (.) Obligation to younger brother No younger borther .159 ( .69) .342 (1.06) Respondent's mar. delayed -.352 ( .78) .538 ( .86) No delay 0.000 (ooe ) 0.000 (oeX ) Obligation to younger sister No younger sister .279 (1.16) .535 (1.60) Respondent's mar. delayed .710 (2.07) 1.508 (3.18) No postponement 0.000 ( ... ) 0.000 ( ... ) Reference category - Not included in the regression - 80 - APPENDIX 3B Expected Age at Marriage and Timing of Marriage of Spinsters and Bachelors When studying age at marriage, it is logical to assume that for every bachelor and spinster in a population there is an expected age at marriage. This assumption leads to the question whether the covariates of the expected age at marriage of the never-married are the same as those of the actual age at marriage of the ever-married. If the answer to the question is yes, then information collected from ever-married women would be sufficient to shed light on the covariates of age at marriage in the population as a whole. To see whether the answer is yes, a small sample of spinsters and a smaller sample of bachelors were interviewed in the 1979 survey, along with ever-married men and women. This note is based on the data for the bachelors and spinsters interviewed. Altogether, 186 bachelors of age 27-34 and 399 spinsters of age 24-34 were interviewed. The smallness of the samples makes detailed analyses of the data problematic. Hence the focus herein is on simple analyses. The main interest pursued below is the respondent's intention to get married and the expected timing of marriage. Some attention is also given to the issue of whether the preferences held by the never-married regarding the various traits in their prospective spouses are such that the supply of suitable matches is extremely low in the marriage market. Yet another matter examined is the prevalence of homogamy, i.e., similarity between self and prospective spouse with respect to one or more traits (e.g., education). As far as marriage intentions are concerned, none of the respondents intends to remain a lifetime bachelor or spitster. As for the - 81 - timing of marriage, 21 males and 101 females reported that they intend to get married as soon as possible. Since such a response does not yield a unique estimate of the expected age at marriage, respondents who gave such responses are excluded from the analysis of expected age at marriage. Now, since the expected age at marriage cannot be lower than the current age, it is possible that differentials of expected age at marriage by socio-economic factors might be partly due to corresponding differences in age composition. With this in mind, two sets of differentials are prssented in Table 3B.1, one set ignoring the age effect, and one adjusting for it.1 For never-married females, the following patterns are discernible in Table 3B.1: 1. The education effect seems to be curvilinear (U-shaped), but the effect is not statistically significant (examine the t-values). We return to the education effect later. 2. As for religion, the Buddhists and Hindus have significantly higher expected age at marriage than the Christians; but the Muslims do not differ from the Christians in this respect (the t value for the Muslim vs. Christian contrast is only 1.29). 3. An overall statistical test shows that the ethnic group effect is significant at the 10% level, but the contrasts in Table 3B.1 are not statistically significant. 4. Those who think that their marriages will be arranged by their parents or other relatives expect to get married at a relatively younger age, compared to those who are not sure whether their marriage will be of the arranged type. 1To adjust for current age, a two-factor analysis of variance was performed, each analysis using age and a particular factor of interest (e.g., religion). To avoid too many cells with zero frequencies, age groups rather than individual years of age were used. - 82 - 5. Those who are seeking a mate expect to get married at a relatively younger age than those whose search has not yet begun. 6. Dowry-related problems do not seem to matter much. (A substantial proportion of the female respondents reported that no dowry payment is required to marry them off. Their expected age at marriage is, on the average, not significantly different from that of those for whom dowry payment is required. Also, difficulties, if any, in raising the funds needed to pay the dowry do not seem to affect the expected timing of marriage.) 7. Those who believe that their marriage will be delayed if no neolocal2 residence can be arranged beforehand expect to get married at a relatively older age. 8. The obligation to wait for one's turn (among sisters) tends to delay marriage by a significant length of time. As for bachelors, very few of the patterns in Table 3B.1 are statistically significant. One interesting exception is the delay in marriage attributable to the absence of ohl!ar sisters.- This pattern is opposite to the wait-for-one's-turn effect mentioned above for spinsters. It may be that the family has an incentive to arrange the marriage of a son sooner if he has older sisters than otherwise, because the financial burden associated with marrying off one of the so8s's older sisters could be eased somewhat using the dowry the son's wife brings with her. We now turn to the data in Tables 3B.2a and 3B.2b concerning the traits respondents consider important in their prospective spouses. The issue of interest is whether there is any tendency to hold on to unusually high standards, thereby making it difficult to find a suitable match. 2The term neolocal refers to the practice by which newly-married couples establish their own residence separate from both the bride s and the groom's kin. - 83 - Table 3B.2a shows that never-married females want their husbands to be of "good" character. As for educational attainment, 42 percent want their spouses to have "a little" or "some" education, 14 percent require "enough education to have a job," 27 percent want 0-level (high school) graduates, while 6 percent are looking for college graduates. A regular source of income for their husbands is desired by most. As far as physical characteristics are concerned, 61 percent are satisfied by "ordinary" features, 19 percent want "handsome" men, while 12 percent do not care about it at all. Finally, caste homogamy3 is desired by 85 percent of the respondents. Turning to Table 3B.2b, we notice that bachelors also look for spouses of "good" ("unblemished," "excellent") character. Education is not a major consideration in the minds of a significant proportion of the bachelors interviewed, although some do prefer educated women for their spouses (more about this later). Slightly more than 33 percent of the respondents prefer not to marry a working woman, while almost 20 percent prefer just the opposite. As for physical characteristics, most bachelors are satisfied with "ordinary" features. Interestingly, caste homogamy is. not a consideration fzr almost 30 percent of the bachelors, which is in contrast with the preference structure of spinsters referred to above. 3Homogamy refers to the tendency of most individuals with given characteristics to select, as a mate, a person with similar characteristics, 84 - With respect to homogamous preference, one matter of particular interest is whether better-educated persons prefer to marry better-educated persons. Tables 3B.3a and 3B.3b show that there is such a tendency among bachelors as well as spinsters. As far as the desired age of the spouse is concerned, there is some difference between bachelors and spinsters, as Tables 3B.4a and 3B.4b show. Spinsters of age 24-26 want their mates to be of age 30, on the average, the corresponding figures for those of age 27-28, 29-31, and 32-34 being, respectively, 32, 34, and 36. There is thus a tendency for the gap between self's age and the preferred age of the prospective husband to narrow as one becomes older. The opposite, however, seems to be true for bachelors. Older women are not preferred by older men. This implies that finding a suitable match for older spinsters is particularly difficult. To summarize, there is little evidence in the data that unusual demands are being put forward by never-married men or women as far as the traits in their prospective mates are concerned. Of course, the principal's preferences in these respects are of less consequence than those of the elders in the family who are responsible for arranging marriages. Unfortunately, we do not have any data on the preference structure of the elders in the family. But it seems safe to assume that if a son or a daughter is not inclined to put forward unusual demands, neither will be the elders in the family. One may, therefore, conclude that the timing of marriage is not significantly affected by any drastic disequilibrium between the trait compositions of the supply and the demand in the marriage market. Finally, the covariates of the expected age at marriage identified in this note do not seem to be much different from those of the actual age - 85 - at marriage identified in the text. But this inference should be received with caution, in view of the peculiarities of the samples of never-married persons used in the analyses (small sample size, and narrow age r,estriction). - 86 - Table 3B.1: Mean Expected Age at Marriage by Selected Characteristics and the Effect of the Characteristic on the Expected Age at Marriage When Current Age Is Taken into Account--- Never-Married Females and Males, 1979. Characteristics Never-married females Never-married males N Mean 6 (t) N Mean B (t) Education Under grade 6 48 30.2 o08(.19) 27 32.2 -.26(.60) Grades 6 to 9 82 29.7 -.33(.85) 63 32.6 -.03(.08) Grade 10 124 29.6 -.27(.73) 52 32.7 70.00( - ) Higher 34 29.3 0.00( - ) 14 l Religion Buddhist 204 29.8 .73(2030) 101 32.9 .44(1.39) Hindu 42 ! 29.5 .87(2.24) 23 Muslim 12 .72(129) 20 *- 0.00 ( - ) Christian 30 29.2 0.00(- ) 11 Ethnic Group Sinhalese 220 29.8 o02(e05) 105 32.9 .39(1.21) Sri Landa Tamil 47 28.9 o.34(.60) 25 32.5 7 Indian Tamil 9 | 1.21(1.62) 6 | * 0.00( - ) Sri Lanka Moor 10 0.00 ( - ) 19 ... J Work (Past 12 months) , Worked 102 30.1 1 .18 (.90) 119 32.4 -.48(1. 36) Did not work 186 29.5 0.00 ( - ) 37 33.1 0.00( - ) Mate selection Arranged 180 29.6 -.76(2.95) 87 32.6 .00(.00) "Love" 58 29e8 -.52(1.67) 34 32.4 -.11(.24) Either 12 29.5 ? 0.00( -) 18 00 - Undecided 38 30.0 J 17 j Searched/match found not yet searched 129 29.8 .80(3.77) 97 33.0 1. 19(2.75) match found 56 29.2 -.05( .20) 38 31.8 .33( .67) match not found 103 29.7 0.00( - ) 21 .eo 0.00 ( - ) - 87 - Table 3B.1 (cont'd) Characteristics Never-married females Never-married males N Mean B (t) N Mean 8 (t) Delay in self's marriage due to delay in an older sister's marriage Not applicable 93 29.6 .03( .15) 57 33.0 67(2.10) Yes 32 1 30.3 1.03(3.33) 13 *.72(1.30) No 163 29.6 0.00(- ) 86 32.2 i0.00( ) Delay in self's i marriage due to responsibility toward a younger l brother Not applicable I 94 29.9 .16( .77) 48 32.9 j .57(1.66) Yes 34 29.5 .21( .69) 30 32.4 .20( .72) No 160 29.6 0.00( - ) 78 32.4 |0.00( - ) Delay in self's marriage due to responsibility toward a younger sister Not applicable* 101 | 29.9 .22(1.08) 47 32.6 .17( .44) Yes 29 29.9 .57(1.74) 58 32.9 .58(1.63) No 158 29.5 0.00( - ) 51 32.2 0.00( - ) * No siblings of the kind in question Sample size too small - Reference category - 88 Table 3B.1 (cont'd) Characteristics Never-married females Never-married males N Mean ( Ct) N Mean S (t) Dowry required? Yes 142 29.7 -.21(l109) No 122 29.7 00 Depends 23 29.5 7 O00( - ) Dowry payment difficult/easy I I Not required 15 29.7 -.48(1.05) Easy 28 30.1 -.97(1.84) Not-so-easy 34 29.4 -.54(1.05) Difficult 67 29.4 -.81(l.71) Very difficult 14 0o00( - ) Whether marriage will be delayed if neolocal residence not arranged Yes 90 29.9 .37(1.81) 51 32.9 .79(2.51) No 198 29.6 0O00( ° ) 105 32.4 0.00( - ) Number of older brothers 0 117 1 29.8 -.10(.32) 55 32.3 -.42(1.14) 1 87 29.6 .02(.07) 54 32.6 -.51(1.37) 2 42 | 29.6 -.22(.63) 31 32e8 3 27 29.8 9 -) ) 00.00( Number of older sisters 0 93 29.6 .01( e03) 57 33.0 .61(1.69) 1 84 29.7 .06(.19) 49 32.3 .08(. .22) 2 61 30.2 .41(1.32) 25 31.8 3 31 29.0 2000( 21 0.00( ) 4 or more 19 29.2 .- 4 - 89 - Table 3B.2a: Traits Desired in Prospective Husbanas -Never-Married Females, Sri Lanka, 1979. Trait Percent desiring the trait Character: "Good" 83 Non-smoking; non-drinking 8 Other; not stated 9 Education: "A little" or "some" 42 Educated enough to have a job 14 0-level (Grade 10) 27 College degree 6 Other; not stated 11 Income: "Average" of "some" 15 "Sufficient to live" 31 Under Rs.500 26 Rs. 500 or higher 21 Not a consideration; not stated 7 Job: "Ordinary" 30 Government job 28 Other jobs (non-government) 30 Other specification; Don't care; not stated 12 Physical "Ordinary" 61 traits: Handsome 19 Not a consideration 12 Other specifications; not stated 8 Caste: Same 85 Not a consideraton 13 Other; not stated 2 - 90 - Table 3B.2b: Traits Desired in Prospective Husbands --Never-Married Males, 1979. Trait Percent desiring the trait Character: ."Excellent"; "Unblemished" 72 "Ordinary"; "Good" 17 "Simple"; "Humble" 7 Other; Not a consideration; not stated 4 Education: "A little" or "Some" 41 "Well educated" 15 0-level or higher 27 Not a consideration 13 Other; not stated 4 Income: Should have an income 34 Not a consideration 51 Other; not stated 15 Job: Should have a job 19 Should not be employed 36 Not a consideration 25 Other; not stated 20 Physical Very prettv 27 traits: "Ordinary" 57 Fair, tall, slim, etc. 6 Other; not stated 10 Caste: Same 65 Not a consideration 30 Other; not stated 5 - 91 - Table 3B.3a: Association Between Self's Educational Attainment and Desired Level of Education for Prospective Spouse--Never-Married Females, Sri Lanka, 1979. Desired level of education for prospective spouse Self's Little Enough College Not a education or some to have 0-level degree consid- Other Total a job eration Grade 5 or 68.5 1.4 13.7 0.0 9.6 6.8 100.0 below Grades 6 44.5 16e4 27.3 1.6 3.1 7.0 99e9 to 9 Grade 10 34.1 15.9 33.5 7.1 0.0 9.4 100.0 Higher 7.1 21.4 17.9 39.2 0.0 14.3 99.9 - 92 - Table 3B.3b: Association Between Self's Educational Attainment and Desired Level of Education for Prospective Spouse-- Never-Married Males, Sri Lanka, 1979. Desired level of education for prospec.i>ve spouse Self's Little Well 0-level Not con- Other Total education or some educated or degree sidered Grade 5 54.8 25.8 12.9 6.5 0.0 100.0 or below' Grac>s-6 50.0 16.2 18.9 10.8 4.1 100.0 to 9 Grade 10 (0-level) 27.2 9.9 40.7 17.3 4.9 100.0 or higher - 93 - Table 3B.4a Distribution of Never-Married Females of Different Age Groups by Desired Age of Their Prospective Husbands, Sri Lanka, 1979. Desired age Current Age for husband 24 to 26 27 to 28 29 to 31 32 to 34 All < 23 0.7 1.0 O.C 0.0 0.5 24 to 26 5.5 1.0 1.1 0.0 2.6 27 to 28 22.1 1.0 2.2 0.0 9.2 29 to 31 51.7 42.1 16.3 2.1 34.6 32 to 33 12.4 33.7 27.2 2.1 20.1 34 to 36 7.6 20.0 46.7 68.1 27.7 37 to 38 0.0 lo1 3.3 19.1 3.4 > 39 0.0 0.0 3.3 - 8.5 1.8 Total 99.9 99.9 100.1 99.9 99.9 Mean 30 32 33 36 32 N 145 95 92 47 379 - 94 - Table 3B.4b ; Distribution of Never-Married Males of Different Age Groups By Desired Age of Their Prospective Wives, Sri Lanka, 1979. Current Age Desired age for wife 27 to 28 29 to 31 32 to 34 All <18 5.1 1.0 3.2 2.3 19 to 21 5.1 8.9 3.2 7.0 22 to 23 17.9 11.9 3.2 11.7 24 to 26 43.6 44.6 22.6 40.4 27 to 28 20.5 20.8 22.6 21.1 29 to 31 7.7 10.9 35.5 14.6 32 to 33 0.0 2.0 9. 7 2.9 Total 99.9 100.1 100.0 100.0 Mean 25 26 28 26 N 39 101 31 171 - 95 - APPENDIX 3C Impact of Changes in Age Pattern of Nuptiality on Fertility Given the age-specific fertility rates and proportion married in various age groups, it is possible to estimate the impact of shifts in nuptiality pattern on fertility, using a simple standardization procedure. Table 3C.1 gives the details of the calculations. From Table 3C.1, one notes that if during 1965-1971 there had been no change in marital fertility, then the Total Fertility Rate in 1971 would have been 4.44, instead of the observed figure, 4.16. Hence, 56 percent of the decline from 4.80 in 1965 to 4.16 in 1971 is attributable to the marital factor. Similarly, if there had been no decline in marital fertility between 1971 and 1974, the Total Fertility Rate in 1974 would have been 3.89 instead of the observed 3.36. The marital factor thus accounted for 34 percent of the decline from 4.16 to 3.36 in Total Fertility Rate during 1971-1974. 96 Table 3C1: Estimated Impact of Changes in Aige Pattern of Nuptiality on Fertility in Sri Lanka from 1965-1971 and from 1971-1974. Age-specific fertility rate Proportion married2 Expected age-specifi* Age group fertility rate 1965 1971 1974 19653 1971 1974- 19714 19745 15 to 19 50.0 39.8 31.0 .139 .106 .078 38.1 29.3 20 to 24 218.9 184.2 146.0 .557 .468 .412 183.9 162.2 25 to 29 269.2 231.9 161.0 .810 .754 .699 250.6 215.0 30 to 34 220.7 199.1 158.0 .910 .891 .870 216.1 194.4 35 to 39 153.6 131.0 126.0 .950 .942 .942 152.3 131.0 40 to 44 41.9 39.6 43.0 .956 .953 .954 41.8 39.6 45 to 49 5,9 5.6 6.0 .960 .959 .974 5.9 5.7 Total fert. rate 4.80 4.16 3.36 4.44 3.89 1 Source: Population Reference Bureau. 1981. World Fertility: A Chart. The 1965 and 1971 figures are based on registration data, while the 1974 figures are based on the 1975 Sri Lanka World Fertility Survey. Note that First Report of the Sri Lanka World Fertility Survey shows age-specific fertility rates as whole numbers per 1000 women; the entry zero after the first decimal shown herein has been inserted for the sake of uniformity; it does not represent reality. 2 Source: First Report of the 1975 Sri Lanka World Fertility Survey, Table 4.5. 3 These figures were obtained by linear interpolation. 4 ASFR (Proportion married1971/Proportion married1965) 5 ASFR1971 (Proportion married 974/Proportion married 197) - 97 CHAPTER IV FERTILITY TRENDS AND DIFFERENTIALS Introduction This chapter presents levels, differentials and trends of fertility. Being a second wave, the survey collected fertility information similar to that collected in the first wave. The fertility section of the questionnaire was designed to elicit total numbe- of live births, the birth history, and the history of pregnancies terminating in non-live births. The total number of live births was obtained from all who reported ever having given birth to a child, by asking questions about: 1. the number of sons living with the respondent; 2. the number of daughters living with the respondent; 3. th2 number of sons who do not live with the respondent; 4. the number of daughters who do not live with the respondent; and 5. the number of children who have died. The interviewer was instructed to add these numbers and verify with the respondent that the total was the number of children she had ever given birth to. Any discrepancies led to probing and clarification. The reason for adopting this procedure of asking a sequence of questions rather than a single direct question on the total number of live births was to minimize the omission of dead children and children living away from home and to maintain comparability of data between the two surveys. 98 - Birth history information was recorded only for the births beginning with the last live birth recorded at the WFS for those respondents who were interviewed at WFS. The entire birth history was recorded for the new respondents. The names of all children were first recorded, and then referring to each child by name, the interviewer ascertained the child's year and month of birth, sex, whether the child was still living, and, if dead, the year and month of death or, if the death occurred within the first month of life, the number of days the child survived. Subsequent to obtaining the birth history, each respondent was asked whether she was currently pregnant; if she was, the duration of that pregnancy was recorded. This was followed by a number of questions on each pregnancy that did not terminate in live births: its date of termination, and, if the duration was longer than seven,months, whether the baby showed any signs of life. These questions served to stimulate the recollection of any children who may have died shortly after birth and were not reported as live births. Pregnancy history was recorded for the period since the time of the last live birth recorded in WFS for the re-interviewed respondents, but for the entire marital duration for the new respondents. The measure of fertility used in this chapter is mostly current parity--the total number of live births up to the time of the survey-and is hence unaffected by errors, if any, in the reporting of dates of birth. Current fertility measures are based on births in the calendar year preceding the survey, and proportion pregnant as of the survey date. Partly to enable a comparison with the WFS data and partly for purposes of evaluation of the data, this chapter is organiied in the same manner as the parallel chapter in the WFS First Country Report. The rest of the - 99 - discussion is, therefore, organized into four parts: (1) Completed Fertility, (2) Cumulative Fertility, (3) Current Fertility, and (4) Fertility Trends. The analysis is broadly similar to that presented in the WFS First Report with the addition of a multivariate analysis within some sections. Completed Fertility A woman's completed fertility refers to the total number of live births she has had at the end of her reproductive life span, that is--for all practical purposes--on reaching the age range 45-49 years. The mean completed fertility of the 45,49 year-old cohort of women in a population represents the level of completed fertility of that cohort. This section discusses the levels of completed fertility and also differences in that respect between subgroups of the population. Levels of Completed Fertility The survey population of ever-married women of age 45-49 years had an average of 5.6 children per mother. The completed parity distribution of this cohort is shown in the first column of Table 4.1. This distribution shows that only about 25% of all women have had a number of children within a range of one child from the overall mean (i.e., between 4.6 and 6.6 children). Of the remainder, a little over 50% had an average number of children below that range. Only about 16% of all women had below replacement level fertility of 2 or fewer children, but the proportion with extremely large families of 9 or more children was also equally small. Table 4.1: The Percent Distribution of Ever-Married and Currently Married Women 45-49 Years, According to Parity (WFS and WBFS). Ever-Married Currently Married WFS WFS -WBFS WBFS WFS WFS WBFS WBFS Absolute Cumulative Absolute Cumulative Absolute Cumulative Absolute Cumulative Parity Frequency Frequency Itrequency Frequency Frequency Frequency Frequency Frequency (1) (2) (3) (4) (5) (6) (7) (8) 0 3.2 3.2 3.8 3.8 2.3 2.3 3.0 3.0 1 5.4 8.6 5.1 8.9 3.6 5.9 4.5 7.5 2 7.8 16.4 7.2 16.1 6.4 12.3 6.4 13.9 3 7.5 23.9 9.7 25.8 6.9 19.2 8.5 22.4 4 8.8 32.7 11.8 37.6 8.5 27.7 119 34.3 5 10.4 43.1 10.5 48.1 11.2 38.9 10.3 44.6 6 13.3 56.4 12.5 60.6 14.4 53.3 13.8 58.' 7 11.6 68.0 14.8 75.4 11.0 64.3 15.7 74.1 8 9.3 77.3 7.4. 82.8 9.7 74.0 6.7 80.8 9+ 22.8 100.1 17.3 100.1 26.0 100.0 19.0 99.8 Total 100.1 100.1 100.0 99.8 No. of Women 995 741 817 577 Mean Parity 5.96 5.60 6.30 5.90 Source: WFS data are from the World Fertility Survey, Sri Lanka, 1975. First Country Report (Table 5.2). - 101 - Comparison with the parity distribution observed at WFS for the women 45-49 years old at the time of that survey (columns 1 and 3 of Table 4.1) shows that the two distributions are roughly similar--with the difference that the dispersion has somewhat lessened and the mean lowered (from 6.0 to 5.6) between the two points in time (1975 and 1979). Completed fertility of ever-married women can also be examined in terms of their parity progression ratios. These ratios represent the proportion of women of a given parity who ever had a child of the next higher order. For example, the parity progression ratio for parity three is the ratio of the number of women who had four or more children to the number of women who ever had a third child. Figure 4.1 illustrates graphically the variation of parity progression ratios according to parity at the WFS and the 1979 survey. The WFS curve relates to the cohort of women aged 45-49 in 1975 and the 1979 curve relates to a younger cohort who reached that age group four years later (in 1979). The WFS curve shows a smooth gradual decline from one parity to the next, reflecting the absence of effective contraception among women in this cohort. On the other hand, the 1979 curve shows a less gradual variation; there is a steeper decline from parity 2 through 4 and a much more abrupt decline from parity 6 to 7. Also, the ratio is smaller at parities above 2 in the 1979 curve. The smaller parity progression ratios and their rather abrupt plunges suggest that, in the younger cohort, women have adopted effective contraception to some extent after their second child. Figure 4.1 SRI LANKA Parity Progression Ratios, 1975 and 1979: Confined to Ever-Married Women Aged 45-49 at the Survey Dates 1.00 .90 2 0 CE = .80 X v WSFSWFS .70 .60 - 01 I I I ! I - 1 J 1 2 3 4 5 6 7 8 Panty kNortd Baink-24870 - 103 - The parity distributions of currently-married women is shown in column 7 of Table 4.1. The mean parity of this group (5.9 children) is slightly higher than that of ever-married women because of their longer exposure to child bearing. The proportion of nulliparous currently-married women--about 3 percent (this was 2.3 percent at the WFS)--is evidence of the usually observed low level of primary sterility among Sri Lankan women. It is of some interest also to examine separately women of age 45-49 years who were interviewed for the first time in the current survey and those who were re-interviewed. The parity distributions of the re-interviewed and the new respondents are shown in Table 4.2. Unfortunately, because of the small numbers at each parity (generally less than 50) of the new respondents and the currently-married re-interviewed respondents, their parity distributions are not very reliable. However, the re-interviewed ever-married women are a large enough group to allow a detailed examination. In examining the parity distribution of this group, it must be remembered that these women were approximately 41-44 years old at the WFS--about four years younger than the 45-49 year old cohort in the WFS sample. The completed mean parity of 5.8 of this younger cohort is slightly less than the average parity achieved by their older counterparts who completed their families in 1975. The parity distribution of the younger cohort also shows a lesser dispersion about the mean; smaller proportions are in high parity groups of parities of more than 7. These differences indicate a declining trend in completed fertility between the two cohorts: those who were 45-49 years old in 1975 and those about four years younger. - 104 Table 4.2: The Percent Distribution of Women Age 45-49 Years According to a Number of Children Ever-Born (New Respondents and Re-Interviewed Respondents). Re-Interviewed Respondents New Respondents Number of Currently Currently Children Ever-Married Married Ever-Married Married Ever-Born Women Women Women Women 0 1.2 0.7 2.5 0O0 1 500 4.6 7.2 5.4 2 6.9 6.2 11.8 10.6 3 904 7.8 14.5 19.5 4 12.1 12.3 12.8 10.2 5 11.4 10.9 5.0 7.7 6 13.2 14.5 9.3 10.8 7 16.1 16.7 6.7 10.3 8 7o0 6.9 12,8 7.1 9+ 17.8 19e6 17.4 18.3 Total 100.0 100.0 100.0 100.0 No. of women 651 517 71 46 No. of children ever-born 5,8 6.0 5.4 5.6 Differentials in Completed Fertility That in Sri Lanka, ethnic, regional and certain socio-economic differentials in fertility have existed to varying degrees and that these differentials have tended to converge has been established in the preliminary and more detailed analysis of WFS data. With a view to assessing the more recent magnitudes and directions in these differentials, we examine in this section the differentials in complete fertility between subgroups defined by a number of variables. The variables considered are - 105 - zone of residence, race/religion group, urban/rural place of residence, level of education, pattern of work, husband's education and husband's occupation. Zone: Zones, the stratifying variables in the sample design, are relatively large areas comprising a number of administrative districts, with the exception of Zone 1 which is the city of Colombo. Map 4.1 shows the positions of the six zones. The zones represent relatively homogeneous socio-economic strata of the population. A more detailed description of the ethnic, religious, educational and occupational profiles of the zones is given in the next section as summarized by the sample distribution of the respondents' characteristics shown in Table 4.3. Type of Place of Residence: Type of place of residence identifies the urban, rural and estate sectors of the country. The urban areas are those defined as such for administrative purposes. The estates are plantations of tea and rubber, confined mainly to the south central parts of the country. The remaining areas are rural. Race-Religion: There are four ethnic groups in the country: Sinhalese, Sri Lankan Tamils, Indian Tamils and Moors; and four religious groups: Buddhists, Hindus, Christians, and Muslims. Because of the almost one-to-one correspondence between ethnicity and religion (see Table 4.3), in this analysis these two variables are combined to form a joint variable with the following categories: Sinhala Buddhists, Sinhala Christians, Tamil Hindus, Tamil Christians and Moor Muslims. - 106 - Map 4.1: The Six Zones of Sri Lanka; A General Ethnic and Socio-Economic Breakdown. ZONE 5 PREDOM I NAN'TLY HINDU rJ ( ~ZONE 3\ IRRIGATED AGRICULTURAL LANDS MTOSTLY BUDDHISTS ZONE 4 LEAST DEVELOPED ZONE 6 HIGIH PROPORTION Or SRI LANHA TAMILS ZONE 1 ESTATE AREA AND MOORS CITY OF COLOMIBO ( RESIDENCE OF INDIAN TAIILS MSOST DEVELOPED RACIALLY HETEROGENEOUS - ~~~ZONYE: RELATIVEL BETTER DEV ELOPED .PREDOLINANTLY BUDDHIST T~i. . )1)ob...... 6c6kro- V-r)obl.. Shoo by h.o P .. Ihlo 111 6oib)o of rh. S..p(6 by P.11. of 6o-6iroo-d lokro0Vrob. opr)OI oorlg ) (2) (3) (4) (51 (6) (1) (21 (5) (1) (2) (3) (4) (S)) () (2) (3) (4.) (5)l (0) (1) (2) (3) (1) (2) (3) (41 (0) (3 (2 () 3 (2) (3) (4) (5) (6) 1 302 100.0 (04.0 01.0 0.0 bb61: 12.6 1:Z) 16.4 3.3 59.2 9.0 17.2 13.6 1.0 I. 26.0 39.4 26.5 57.8 13.3 LL.0 ((.0 12.3 10.3 39.6 29.8 12.0 46.6 0.8 0.0 25.2 15.8 (2) Zo 2 1340 M0.0 . . . . 20.5 71.6 5.6 9).) 2. 1.5 3.7 0.4 02.0 3.8 3.8 8.7 0.8 .9 33. 38.0 20.8 63.) (6.6 13.0 21.2 15.5 26.6 317.1 20.8 6.1 29.9 81.3 14.9 26.9 139 (37 Zo.-3 729 um- . . . . . . (.2 00. 0.3 53.5 1. 0.5I 13.7 (.2 69.2 1.0 15.2 187 0.0 183 63.0 27.3 11.6 56.2 8.9 18.6 16.3 1,9.3 35.3 32.6 (2.9 4.0 21.2 34.3 ?.a 10.6 13.6 (4) Zo 422 120 . . . . 81 719 0.0 (4.0 02 .6 34.0 0.6 13.9 43. 35.3 7.3 0.0 37.3 44'.8 71. 64 72.7 0. :10 7.9 25. 40.8, 21.7; 12.0 6.2 20. 20. 16.8 19.o (6. () oo5 336 100.0. . . . . 26 3. . . 0.3 58 60 02 18 7. 6.0 18. 0.0 .6 37.7 31.t 22.6 84.2 32 7. . 35 30. 2.1 27. 9.3 23.7 24. 2. 21.0 93 (6) Zo (04 (0. . . . . . 43 656 3. 6. . 2.:. 03 6. 27.1 4.0 29 9.) 2).1 39. 21.5 11.5 30.3 96 (42 357.31 2).65 40.8 24.7 (3.1 4.5 13.8 17.2 35.0 15.6 5.7 III O,b. (95 (00.0 34.3 30.7 9.1 0.5 8. 8.0 . b):7 18.:4 1.4 15.0 3.7 4,7.5 14. 13.7 (9. " 8.3 29.0 39.1 23.) 59.5 11.) 22.3 16.3 13.9 30.2 36.5 29.4 12.3 41.3 3.6 4.) 25.0 13. 3 (2 MO. 398 10. 0.0 29.9 18.6 5.9 7. 36. . 17 105 05 30 0.3 77.6 9.2 7.3 5.9 0.0 (58.5 38.8 30.1 12.7 54.0 12.5 14.8 is. (8.5 34. 5316.4 is.6 48 22.5 23.9 13.3 22.2 11.I (3) Emoo936 100.0 0.0 12.3 0. .0 00 07.3 . . . 11.4. 22.50 64.85 1.8 0.0 ((0 8.5 16 57 0.8 42.9 45.3 7.6 3.2 6.3 3.2 9. 80.7 25. 49.6 16.6 8.2 1. 03 . 20 . . (1) SIoblh or 133 100.0 6.2 36.9 183 1.2 0.1 37.2 16.6 8.2 2.2. . . . . 1.7 0.2 0.1 7.3 0.4 13.) 34.6 33.6 18.3 40.1 (4.2 15.7 21.9 17.3 31.2 33.8 17.7 5.8 24.2 20.3 12.6 24.2 (. (2) r L.rr _,_ ooJ 65 I100 5.9 6. t.3 20.1 66.1 9.8 25.3 53.2 221.5. . . . . . 80. 7 0.3 17. 0.1 I17.3 39.2 26.81 106.2 63.6 5.7 9.1 21.61.6 33 2.62. 76 2.9 57 2.8 63 99 (3) Irrd).o ToLlS 441 1000 0.8 4. . .0 1.3 91. 2. 36 93.5. . . . 0. 4.8 0. 4.3 0.0 67.0 50.0 2.4 06 S.9 3.) 8.5 81.4 27.0 54.6 14.8 3.2 3.4 7.2 0.8 82.9 4.7 3.o (41 K.., 38'. 100.0 13.1 12. 259 2.6 5.2 (8.2 36.8 60.3 2.9. . . . . 0.0 0.1 99.5 0.3 0.1 24.9 50.8 18.3 6.0 71.0 4.0 10.6 8.6 19.0 34.4 28.7 17.9 6.7 44.8 10.1 9.1 13.5 15.0 (( .5(6 3055 (00.0 5.9 36.3 16.5 1.2 0.2 39.0 13.9 63.8, 2.3 997 03 0. 00 . . . . . . 4.5 35.5 32.4 17.5 68.4 14.1 15.8 21.7 (7.8 32.1 53.1 17.0 5.3 23.8 21.4 12.8 24.0 17.0 (2 (lo.4. 95 0. . . . 26 2. 2. 35 3. 65 06 5. 39 0.8 0.2. . . . 32.1 63.8 16.1 6.1 3Y7. 3.9 0.1 50.3 22.6 44.1 19.9 13.3 3.1 15.7 9.4 32.4 10.2 7.3 ()8 ).44 100.0 13.0 12.7 27.3 23.7 5.0 18.3 3 .0 59.2 2.8 04 0. 0.9 94.5 3.0. . . . 23.7 50.3 19.9 6.1 73.0 4.5 10.0 8.4 20.1 33.2 28.6 19.1 6.5 45.9 9.7 9.4 14.1 14.4 (41) rh.Io6(oo 4 03 100.0 10.3 29.1 26.5 6.9 15.8 13 .4 42.8 48.1 9. 33 28.3 47 0.3 3.2. . . . . .0 29.0 30.3 26.0 49.8 (2.1 14.5 23.6 14.3 23.8 35.61 26.2 10.0 29.2 10.3 14.3 21.3 12.9 (0) W. SobooILog (114 102.0 2.6~ 12. 154 .6 3.3 57.8 06 59.1 32.3 5 1. 13.0 24.0 11.1 0.0 591.3 35.3 11.1 2.3 0.0 . . . 34.6 5.6 13.4 62.5 34.2 50. 13.0 2.1 0.9 10.7 (9.5 44.8 13.2 10.9 1( oor rd.15 1826 100.0 4.5 24.4 17.2 6.7 6.704 1. 0.0 15.0 62.8 14.1 12.1 30.711. 94 7. 11.1 6.4 0.2 . .6 . . 247.8 9.8 (5.6 26.8 21.1 44. 30.7 6.2 1.6 17.5 19.6 26.6 1.2 15. (2 Srodoy 1 -9 39 10. . 36. 14. 4. . 06 2. 14 35 8. 26 08 51 32 7. 10 50 1. . 46 167 0. 3. 1.5 2.1 423 10 .4 306 16590. 4. 1. (31 1g6- (0or 9 100.0 10.9 37.3 (1.1 2.3 10.2 20.4 28. 09 27 (. 41 . . . 15 1.2 33 1. . 47.2 10. 10.5 31.5 5.9 7.6 27.3 5. 20.1 42.6 9.8 6.3 18.6 4.5 III 0.-or rord 2354 10.0 35 24. 17.6 8. 12.0 30.0 22.6 75.3 1.3 67.9 (7.6b . 12.6 0.9 62.0 15.2 13.2 8.5 0.2 12.7137.3 35.2 12.0 . . . . 14.4 301.6 35.2 19.8 5.6 37.6 15.1 8.5 1~7.3 0. (2 olr. ro orr (00 7. 12 12.0 4. 30 33.0 19.4 76.8 3.0 87.8 6.b . .:. 00 . . . . 9 3.2 43.0 14.9 2.. 3.1 3'.8 37.0 16. 0.3 (9.8 33.6 17.5 31.8 13.4 - (3) 1,Arr o hbo., 661 (00.0 5.13 263 2U.5 .5 3.11 39.3 (.8 73. 9.4 9.0 a.9 7. 6.0 0.3 73.00 11.7 6.1 8. 0.3I 22.7 43.) 22.2 11.9 . . . . 31.4 27.7 21.3 9.7 6.3 22.4 23.2 19.5 19.0 11.6 (4 1 o.roo r 1274 100. 4. 28.5 9.3 1.7 1.3 54.8 11.5 48.3 40.3 53.1 ((.0 28.9 2.6 0.5 52. 3.6 2. 75 0.3 2.8. 38.4 114.3 (8.5 . . . . 22.4 37.5 22.4 (6.7 7.1 21 .2 9.6 41.6 12.2 9.1I H0..W,.6o LJ-1-. I.1 (0o)-H 400110 I9 (0..0 4.2 23.:1 I15.6 3.1 5.1 44.3 13.8 67.9 18.3 64.0 13.5 13.5 61.1 0.) 6(.6 24.0 9.0 6.4 0.0 32.8 43.0 19.3 4.9 37.0 73 2. 10 . . . . 0. . 90 4. 18 1. 11) l11.r.~o. -5 163 001 .0 2.I1. 6.8 6.3 45.9 11.1 69.7 19.3 63.4 (3.7 14.7 8.) 0.2 60.0 25. 8.2 5.9 0.2 26.0t 49.2 20.5 3.5 44. 13. 152242 . . . 06 19 2.3 5. 3.2 42 (2) So.C.=dsy 6-9 1466b (000 8.3 35. 16. 4.0 6.4 391.2 22.3 305 7.2 76.6, 10.9 4.4 7.5 0.5 69.0 12. 7.9 9.808.3 7.7 20.3 40.1 13.9 56. 13.6 9. 20. . . 63 17.3 311.67 20.0 13.4 (31 0(1 10 or 830 (00.0 1.0 33.6 ((3 3.9 1. 291 37 6.0 .3 07 1.6 .7 .3 .3 62.0 152 8.8 2.7 0. 2.2 9.2 35.2 53.4 56.3 104 7.7 25.6 a 24 4. . . 11.5 2. U1) Pr-foo.(..I.. 237 1021.0 (3.6 29.2 10.6 5.0 10.7 30.1 40.3 5~7.1 7.6 60.3 18.1 2.1 9.3 1.9 50.2 17.4 9.3 14.7 0.4 2.5 10.7 27.5 59.3 49.6 8.7 8.7 32.9 0.7 4.) 10.8 76.4 12I(rook.0 005 100.0 12.3 33.7 13.0 4.6 7.1 29.3 31.3 63.7 5.5 61.1 13.4 2.6 14.-5 I.5 6,1.4 12.6 15.5 10.1 0.3 7.3 26.5 37.0 2. 75 96 (. 21 232. 78 3. (37.1-o(yd((. 00 100.0 0.2 13.4 30.4 61.7 10.6 38.6 4.3 95. 0.2 "Z. (2.7 0.5 47 0,.1 792 (.2. . . 904. 86 9.9 60.5 9.5 16.0 14.3 6.3 26.8 36.4 30.1 (4 .oo) ~oe 4 0. . 00 57 45 46 6. . 43 5 .9:2. 38. 35. 3. . 95 5. .8 5.9 0.4 36. 44.6 15,.6 3.69 27.3 10.0 13.2 51.5 1. 60.7 39. 3. (51 SkIo 840 ( 00. 7. 36.3 13.9 5. 7.5 29.2 22.9 74.3 2. 81.4 10.9 2. 5.3 0.4 74.2 9.9 5.1 9.7 0.5 10.7 30.3 35.6 15.3 55.3 18.3 10.9 13.2 4.0 36.9 49.1 ((.0 (6) 00.611II.d 487 100'D.0 a. 32.4 1'7.2 3.8' 5.7 281.1 21.1 36.8 2.1' 76.42 11. 5 2. 3' 10.1 0.0 69.4 11.7 10.1 9.2 0.4 15. 48.0 30.2 6.4 57.5 13.3 11.5 17.5 8.6 43.0 39.8 6.3 361i. -...1oY 10016.0 0 1 . .ol ob.r of 0oopOOd... of orb. -r r .ood Z-pr1o bo.. Th.ir, d).frlbuhloo by -h-0r-o06bl.. 10 001 b -o o.oo. of h6 .7 -y 3 o1 oo BEST COPY AVAILAB LE - 108 Level of Education: Level of education of both the husband and wife identifies five categories: no schooling; primary (grades 1-5); secondary (grades 6-9); high (grades 10-12); and university. Pattern of Work: Pattern of work distinguishes between four work statuses, based on their timing with respect to marriage. These categories are: (1) never worked, (2) worked before marriage only, (3) worked after marriage, (4) worked before and after marriage. A woman is considered to have worked if she has done any work away from home for pay or profit. If she has done any such work before marriage, she is considered to have worked before marriage irrespective of the length of period for which such work was done. Work after marriage is similarly defined. Therefore, women in one category of work could include those with a wide range of experience, from manual work to professional or from specialized occupations and those with very short spells of work to others with long years of continuous experience. Husband's Occupation: The occupational classification of husbands excludes the unemployed, armed service men and undefined or unstated occupations, and identifies five categories: (1) agricultural occupations, (2) professional and clerical workers, (3) sales and service workers, (4) skilled workers, and (5) unskilled workers. As in any classification which condenses occupations into a small number of groups, this classification includes many heterogeneous work categories, thereby.making detailed substantive interpretations of occupational differentials difficult. - 109 - Association Between Background Variables The observed fertility differentials are difficult to interpret if the variables are associated. Fertility differentials observed across categories of a given variable may not be entirely attributable to differences between these categories if the variables are inter-correlated. It is therefore important to evaluate the inter-correlations among the variables. Clearly, the nature of these variables suggests that they are not independent of each other. Table 4.3 shows the extent of association between eight background variables in terms of the percentage distribution of respondents in each category of a variable by all other variables. Thus, row 1 of the table shows that in Zone 1 all respondents lived entirely in urban areas; 67 percent were Sinhalese, 13 percent Sri Lankan Tamils, 1 percent Indian Tamils, and 16 percent Moors, etc. Some of the more pronounced associations between variables from Table 4.3 may be summarized as follows. Zone of residence is highly associated with all other variables. Zone 1 is all urban, is racially heterogeneous. with a relatively large proportion of Moors and Christians and has a better-than-average educational and occupational-status profile. The other zones have particularly higher-than-average concentrations of particular ethnic groups--Zones 2 and 3 are predominantly Buddhist, Zone 4 consists largely of the Sri Lankan Tamils and Moors, Zone 5 is predominantly Hindu, and Zone 6 carries almost all of the Indian Tamils. Urban areas have distinctly higher proportions of better-educated people; estate areas have an exceptionally high proportion (80 percent) of women who worked before and after marriage. (Almost all women living on tea and rubber estates are - 110 employees--mostly manual workers--of the estates.) Ethnic groupti also show some strong associations with other variables. For example, the Sinhalese population live mainly in rural areas and are almost exclusively Buddhists; the Sri Lankan Tamils have a higher concentration in urban and estate areas than in rural areas, and are mostly Hindus. The Indian Tamils are almost entirely Hindus: half the population has never been to school and lives exclusively on estates; 80 percent has worked before and after marriage. The Moors are exclusively Muslims and have a proportion of urban dwellers twice as high as the overall samrle proportion of 18 percent. The Moor women, but not their husbands, have less than average proportions of higher levels of education, and over 75 percent of them have never worked at any time. Educational levels of women and of husbands bear an almost direct relationship. In the classification of the sample by these two variables (see Table 4.3), the largest percentages are on the diagonal cells, except when the wife has had no schooling, in which case the sum of the proportions on the diagonal and above it is larger thau that below it for each level of the wife's education. Thus, men's educational attainment is in general either equal to or higher than the educational level of their wives. Level of education is also related to the pattern of work. The significant feature is that nearly half of women with no schooling have worked both before and after marriage. These are largely the estate women, as can be seen by their disproportionately large concentration (33 percent) on the estates (in the total sample, the proportion of women on estates is only 13 percent). The other educational levels do not display any significant deviations from the overall pattern, except that a somewhat large proportion of primary-educated women are in the never worked category and a slightly high proportion of better-educated women are among those whose work continued uninterrupted by marriage. Husband's level of education, as expected, is related to his occupation, with large proportions of better-educated men being in professional and clerical occupations. Thus there is a clear correlation between zone and place of residence, ethnicity and zone, race and religion, education and occupation, and educational attainment and work pattern. These associations make all variables interdependent and make any analysis of the effects of one variable on the fertility differentials a difficult task. Multivariate Analysis of Differentials in Completed Fertility The Method: The method of examining fertility differentials here is to examine the differences in mean levels of fertility between categories of each variable. The effects of other variables on these differences are identified by controlling for their compositional variations through step-wise multiple regression. In the regression equations each category of categorical variables is treated as a dummy variable. An important consideration in controlling for other variables is to establish a causal ordering among them. A unique and valid causal ordering is impossible to establish because of the possible two-way causal relationships between certain variables and the absence of theoretical or empirical evidence of the direction of causality between others. This analysis does not specify a causally ordered unique set of controls. Instead, a sequence of variables is specified in which unadjusted group - 112 - means are adjusted by entering them in that sequence into the regression equation. In this sequence, the first variable to enter the equation is the background variable across which differentials are to be measured. Then the two demographic controls, duration since first marriage and age at first marriage, are entered, followed by the remaining controls in the following pre-determined order: - Type of place of residence - Race/religion - Level of education - Pattern of work - Husband's education - Husband's occupation For example, in examining the differentials by pattern of work, the variables enter the regression equation in the following sequence: pattern of work, duration since first marriage, age at first marriage, zone of residence, race/religion, type of place of residence, level of education, pattern of work, husband's education, husband's occupation. In this hierarchical order, ascribed demographic variables precede the acquired ones in a broad sense. In order to complete mean fertility levels for categories of any variable--say, zone--a regression is first performed of children ever-born (CEB) on zone, with -variables measured about the sample means and with Zone 1 as the reference category. the resulting regression equation is: CEB = CEB + b2 (ZONE 2 - ZONE 2) + b3 (ZONE 3 - ZONE 3) + b4 (ZONE 4 ZONE 4) + b5 (ZONE 5 - ZONET5) + b6 (ZONE 6 -ZONE6), where Zone 1 equals 1 if the respondent is in Zone 1 and 0 otherwise, etc., - 113 - and the bar above the variable indicates that it is the sample mean of that variable. For example, CEB is the sample mean of the variable children ever born, and ZONE 2 is the sample mean of the dummy variable ZONE 2, i.e., the proportion of women living in Zone 2. The mean number of children ever born for the reference category, Zone 1, is then calculated by substituting ZONE 2 = ZONE 3 = ZONE 4 = ZONE 5 ZONE 6 = 0 and CEB = the overall mean number of children ever born in the sample. Thus, CEB = CEB b2 (ZON`F 2) - b3 (ZONE 3) - b4 (ZONE 4) - b5 (ZONE 5) - b6 (ZONE 6) = bl, say. The mean fertility levels for other zones are calculated from the relations CEB/ ZONE 2 = bl + b2 ZONE 2, CEB/ZONE 3 = bl + b3 ZONE 3, CEB/ ZONE 4 bl + b4 ZONE 4, CEB! ZONE 5 =bl + b5 ZONE 5, CEB/ZONE 6 = bl + b6 ZONE 6. To obtain the means for zones adjusted for other variables, these variables are introduced into the regression equation one by one, and at each step the above procedure is repeated using the regression coefficients b2 to b6 from the equation at that step. In order to produce unbiased estimates of the population means from this regression analysis, each individual is assigned a weight equal to the sample design weight, thus correcting for the differences in probabilities of selection among individuals. The weights are assigned by zone, the stratifying variable in the sample design. All individuals in any one zone are assigned a weight equal to the inverse of the average of the probability of selection from the various strata in that zone, the weights.being scaled - 114 - to equalize the total numbers of weighted and unweighted observations (see Appendix for a description of weights). The issue of weighting in regression analysis is a complex one of varied opinion among statisticians and demographers. It is argued that when the stratifying variable is included in the regression weights should not be used. A simple approach is taken in this analysis, which is believed to provide reliable enough estimates for assessing differences between mean levels of fertility of population subgroups. The procedure adopted here is as follows. When zone is not included among the variables, the regression coefficients are taken from the weighted regressions, and when zone is included the coefficients are taken from the unweighted regressions. The means, however, come from the weighted regressions. The unadjusted and adjusted means for categories of each variable obtained from the regression analysis are shown in Table 4.4. A substantive discussion of the differentials follows. Differentials by Zone of Residence Completed fertility, as observed, shows a certain pattern of variation between zones. It is highest--6.6 children per woman--in Zone 4, the economically least developed, dry, eastern coastal belt of the country. It is lowest--4.8 children per woman--in Zone 2, the most developed southwestern region surrounding the capital city of Colombo. It is also quite low in Zone 1, the city of-Colombo. Completed fertility levels of the remaining 3 zones are close to one another and within .2 of the overall mean of 5.8. These differences between zones are largely attributable to variations in age at first marriage. When adjusted for marital duration - 115 - Table 4.4: Differentials in Completed Fertility: Unadjusted and Adjusted for Other Variables Duration Place Hus- Hus- of and of Race/ Wife's band's band's Unad- Age at Resi- Reli- Educa- Wife's Educa- Occupa- Subgroup justed Marriage dence gion tion Work tion tion Zone 1 5.37 5.73 6.04 5.66 5.73 5.68 5.73 5.79 Zone 2 4.85 5.44 5.62 5.35 5.44 5.43 5.45 5.45 Zone 3 6.00 5.64 5.78 5.64 5.50 5.53 5.41 5.45 Zone 4 6e62 6.04 6.21 6.00 5.79 5.81 6.00 5.98 Zone 5 5.88 5.53 5e70 5.89 5.93 5.99 6.02 5.97 Zone 6 6.07 5.76 6.02 5.79 5.77 5.75 5.75 5.74 Rural 5.76 5.74 5.73 5.64 5.66 5.64 5.64 5.64 Urban 5.13 5.39 5.53 5.61 5.76 5.76 5.79 5.80 Estate 5.63 5.27 5.11 5.57 5.17 5.29 5.27 5.27 Sinhalese: Buddhist 5.58 5.73 5.76 5.75 5.72 5.72 5.72 6.33 Christian 4.77 4.75 4.70 4.71 4.99 4.97 5.08 5.66 Tamils Hindu 5.77 5.39 5.14 5.17 5.24 5.25 5.26 5.91 Christian 4.91 4e75 4.92 4.95 5.24 5.15 5.14 5.75 Moors 7.18 6.48 6.72 6.73 6.58 6.56 6.49 7.15 Wife: Worked Never 5.64 5.61 5.63 5.61 5.59 5.57 5.52 5.52 Before only 5.78 5.17 6.10 6.07 6.16 6.17 6.08 6.04 After only 6.34 5.85 5.93 5.90 5.84 5.80 5.91 5.91 Before & after 5e08 5.38 5.31 5.37 5.42 5.49 5.53 5.53 Education None 6.73 6.20 6.23 6.32 6,20 6.20 6.27 6.27 Primary 5.76 5.57 5.64 5.63 5.65 5.64 5.59 5.57 Secondary 5.00 5.23 5.33 5.26 5.32 5.32 5.34 5.36 High School 3.51 4.93 4.61 4.53 4.67 4.68 4.65 4.72 University 1.91 4.25 4.26 4.17 4.42 4.48 4.50 4.57 Husband: Education None 5.27 5e07 5.11 5.22 5.13 5.04 5.02 4.98 Primary 6.36 6.06 6.08 6.19 6.07 5.98 5.98 5.92 Secondary 5.71 5.90 5.93 6.01 5.88 5.96 5.97 5.99 Higher 4.00 5.31 5.02 5.10 5.11 5.56 5.59 5.90 Occupativn Agricultural 5.75 5.56 5.53 5.53 5.54 5.48 5.48 5.68 Prof./Clerical 4.30 5.29 5.11 5.11 5.14 5.50 5.51 5.15 Sales/Services 5.58 5.67 5.60 5.61 5.54 5.60 5e60 5.35 Skilled 6.24 6.28 6.43 6.41 6.41 6.41 6.39 6.10 Unskilled 5.60 5.46 5c88 5.85 5.87 5.75 5.76 5.47 116 - and age at first marriage the means of the low fertility zones rise and that of the hi.gher fertility zones lower considerably, with the result that the range of variation narrows to .6 children from the unadjusted range of 1.2 children. The effect of controlling for place of residence has little substantive significance because of the interrelation between the two variables: Zone 1 is entirely urban; estate population is largely confined to Zone 6; the other zones are largely and almost equally rural. Both urban and estate places have lower fertility levels than rural places. Thus, controlling for place of residence has raised the means of Zones 1 and 6 about .3 and of other zones just slightly. Adjustment for racial -composition lowers the mean of Zone 4, which is largely a Muslim area, and of Zone 6. Adjustment for other variables-education, pattern of work, husband's education and husband's occupation-have very little effect on the mean fertility levels of individual zones. When the effects of the compositional variations in all the variables are removed--with the exception of Zones 2 and 3, where fertility level is about .4 births below the overall mean--all other zones have mean fertility levels within .2 of the overall mean. This differs from the WFS data which reported that the only significant departure when adjusted for all controls was seen in the high fertility of Zone 4. These residual interzonal differences are not statistically significant. Thus it is mainly.the zonal variations in age at marriage and, to a much lesser extent, urbanity and racial composition to which the zonal differentials in completed fertility can be attributed. Differentials by Place of Residence As expected, on.the basis of the marginal distinction beween rural and urban places in Sri Lanka, fertility differentials by place of residence - 117 - are very marginal. Rural residents have a mean fertility of 5.8 children, equal to the overall mean level. Estate fertility level of 5.6 is close to that of the rural sector, but the urban fertility is about .7 children lower. Adjusting for other variables has very little effect on the average fertility level of the rural sector which represents about 80 percent of the total sample, but it has a substantial effect on the urban and estate fertility levels. Among the variables to which the low fertility of urban areas can be attributed are the age at marriage, racial composition and education. When these variables are adjusted for, the urban fertility level rises by .7 to 5.8. High estate fertility is attributable in large measure to low age at marriage and low levels of education, as seen by a drop of .4 births from the steps previous to the introduction of these variables. Interestingly, controlling for race raises the mean estate fertility adjusted for variables up to zofle by about .4 births, implying that the racial composition of estate population--almost exclusively Indian Tamil--is favorable to a low fertility level. Their pattern of work--working both before and after marriage-has a similar effect. Thus the mean fertility of the estate women is conditioned by age at marriage and zonal distribution, promoting high fertility and a racial composition and work pattern favoring low fertility. The final effect of adjustments for all variables is an interesting one: relative positions of the types of places change; urban fertility rises by .7 to the highest position, reaching 5.85; rural fertility changes little and remains just .2 births below the urban level; but estate fertility emerges as the lowest--.6 births below the average urban level. - 118 - Differentials by Race-Religion Ethnic and religious differentials in completed fertility are remarkably high, the difference between the highest and lowest levels being over two children. The small minority, the Moors, have the highest mean fertility level of 7.2 children. The minority of Christians of both Sinhala and Tamil races have the lowest fertility levels, almost one child less than the overall average of 5.8 children. The majority of Sinhalese whose religion is Buddhism have a mean fertility slightly below this overall mean level, and the majority of Tamils whose religion is Hinduism has the same mean fertility as the overall average. Adjusting the observed means for the demographic controls related to marital age raises the mean level of late-marrying Sinhala Buddhists and lowers the means of early-marrying Christians and, very particularly, of the Moors. Adjustments for other variables show: (1) that the high fertility of the Moors is attributable to their lower levels of education and pattern of work despite a favorable distribution of their community in developed regions and places of residence and husband's occupation; and (2) that the low fertility of the Tamil Christians is attributable in some measure to their residence in developed parts of the country, their better-than-average education, and their occupational distribution. These same factors, except for the residence variable, are partially responsible for producing the low fertility level of the Sinhala Christians. The significant feature of ethnic-religious differentials in completed fertility is that they still persist even after being adjusted for the demographic and background variables: a range of variations of nearly 1.5 children between the Moors and Christians still pers.ists. Differentials by Education A highly pertinent issue in examining fertility differentials is the relative importance of women's and husband's educational attainments as determinants of fertility. However, in this analysis their relative importance cannot be determined quantitatively because it does not take into account the interaction effects arising from the high degree of correlation between the two variables. It is, however, possible to make a close comparison of the effect that one variable has on the other, and, in order to allow for interactions to some extent and eliminate biases due to small sample size of university women of age 45-49, the extremely deviant and very small categories of university educated women and wives of university educated men are combined into one category where either spouse has had university education. Educational differentials are among the strongest. The range of variation between the two extreme levels-i.e., where the wife has never been to school and where either spouse has had university education--is over 4.5 children. The relationship between completed fertility and level of education is negative and curvilinear: the decrease in fertility from one level to the next increases with increasing education. Adjustments for other factors show that the single most critical factor to which these differences can be attributed is, of course, age at marriage. Adjusting for age at marriage shrinks the differentials to less than two children. The other background variables have only marginal effects on these differentials by wife's education, with the exception that race-religion tends to lower the mean of the unschooled category and raise the mean of the university group. Particularly significant is the absence of any effects by husband's - 120 - education and occupation. In contrast to this, the effect of wife's education on differentials by husband's education is considerable because it raises the mean for the high school category. The mean for this category is further raised by the two variables, pattern of work and husband's occupation. One other significant difference between differentials by wife's and husband's education is that among men it is the primary level rather than the no schooling level that starts off with the highest mean fertility. Even after adjusting for all variables, educational differentials still persist. Across husband's education levels the only deviant group is the no schooling category, which has a mean of almost 1 child below the three higher levels whose adjusted means are virtually the overall mean of 5.8 children. Across wife's educational levels the range is wider, 1.7 children; thus, the monotonically decreasing relation is still retained--the mean levels change from 6.3 for unschooled women to 4.6 for university educated women or wives of university educated men. Differentials by Pattern of Workt Considerable differentials in completed fertility exist by pattern of work. Women who worked both before and after marriage have the lowest fertility level, 5.1 children; curiously, women who worked only after marriage have the highest fertility, 6.3 children. The other two groups of women who worked only before marriage and those who never worked have means of 5e8 and 5.6, respectively. Again, this pattern is different from that observed in WFS, where it was the women who never worked that had the highest fertility. Never worked category is very nearly half the total sample, and their mean level is little affected by compositional variations in the other variables, even-age at marriage. On the other hand, the low 121 - fertility of women who have worked both before and after marriage is attributable mainly to their late marriage and to a favorable distribution of almost all other factors. Adjustment for these variables raises the mean fertility level of this category by .5 children to 5.5 children. Women who gave up work after marriage have a lower-than-average age at marriage, as reflected by the lowering of their mean fertility level when controlled for age at marriage; but their residential distribution in low fertility zones is shown by an increase of over one child in the adjusted mean when controlled for zone. Early age at marriage is chiefly responsible for the high fertility of women who took up work after marriage. Fertility differentials among women identified by different patterns of work almost disappear when adjustments are made for the variables considered; the adjusted differentials lie within a narrow range of less than .5 children. Differentials by Husband's Occupation Differentials exist by husband's occupation as by most other variables observed earlier. The most deviant group are the wives of men in professional and clerical occupations, who have the lowest fertility level of 4g3 children--far below the overall mean of 5.8 children. Wives of men in all other professional groups have fertility levels lying within a narrow range of the overall mean, with skilled occupations at the highest level, 6.2 children. Adjustment for age at marriage raises the mean fertility level of the professional group by almost one child but affects only slightly the fertility levels of other groups. This reduces the range of variations between skilled workers and professionals to one child from the unadjusted range of 2.5 children. Adjustment for other variables affects the various occupational means to varying degrees, depending on the nature - 122 - the various occupational means to varying degrees, depending on the nature of the obviously high correlation between these variables and occupation. Among the pronounced effects are that: controlling for zone raises the means of both skilled and unskilled workers; controlling for educational level of the wife raises the mean of the professional/clerical and sales/services groups and lowers the level of unskilled workers; and, most pronounced of all, controlling for husband's educational level sharply raises the mean of agricultural workers but reduces the levels of the other groups by about .3. This strong influence of husband's education on occupational differentials in fertility is clearly due to the strong correlation between the two variables. Thus, the higher-than-average fertility of agricultural workers is largely caused by their below-average ages at marriage. The high fertility of skilled workers is attributable to the low educational level of the men. The low fertility of the professional and clerical group cannot be attributed to late marriage, but is mainly the result of the higher-than-average educational attainments of the wife. Some occupational differences in completed fertility remain even after adjustments are made for all other controls: the professional and clerical group remains at a low level of 5.2, and skilled workers remain at a high level of 6.1. Agricultural workers have a mean of 5.7-close to the overall mean--but the two groups, sales and service and unskilled workers, are about .3 children below that mean. Cumulative Fertility Parity by Age: Cumulative fertility--the mean parity up to the time of the survey--of successive age cohorts describes the cross-sectional pattern of - 123 - childbearing in the population. The mean parities of five-year age cohorts of ever-married and currently-married women are shown in columns 2 and 4 of Table 4.5. On the average, women who had ever been married and who were currently married had had an equal number of children, 3.7. Within each of these groups, each age cohort had a slightly higher parity than the preceding cohort because of the longer marital durations of older age cohorts. Within each age group, currently-married women, in general, had a slightly higher level of mean parity--again, because of their longer exposure to childbearing. These age patterns of mean parities of ever-married and currently-married women can be compared with those observed at WFS shown in columns 1 and 3 of Table 4.5. The consistency of the two distributions lends credence to the quality of data. The difference between main parity of ever-married women between the two survey dates is a measure of the overall change in marital fertility. This mean has declined from 3.94 to 3.75, a reduction from 4.8 percent in 4 years. Among currently-married women, the overall mean declined from 3.96 to 3.74, a reduction of 5.5 percent over the same period. All age groups had contributed to this decline in marital fertility. The decline was most substantial in the young age range of 20-39, particularly in the 20-24 group where it was nearly 20 percent. Thus, the comparison of patterns of childbearing among age cohorts at the two surveys indicates a declining trend of fertility mainly in the peak childbearing years. This change is significant in that there was no such clear declining trend in mean parities of ever-married or currently-married women--i.e., fertility within marriage--between the previous four-year period 1971-1975 as reported in the WFS First Report. 124 - During that period, the change in fertility was shown to be largely due to changes in marital composition, not marital fertility. Table 4.5: Mean Parity at the Time of the WFS and 1979 Survey by Age at That Time. Ever-Married Women Currently-Married Women WFS 1979 WFS 1979 Age (1) (2) (3) (4) 15-19 0.75 0.71 0.76 0.68 20-24 1.57 1.27 1.58 1.27 25-29 2.59 2.29 2.62 2.31 30-34 3.86 3.30 3.90 3.35 35-39 4.89 4.42 5.00 4.48 40-44 5,54 5.49 5.67 5.49 45-49 5e94 5.63 6.30 5.85 All 3.94 3.75 3.96 3.74 Source: WFS Data are from the WFS Sri Lanka First Report, 1978. Having discussed the mean parity levels of all ever-married and currently-married women, we turn now to a comparison of new respondents with the re-interviewed respondents. Mean parities by age of new respondents, re-interviewed repondents and those who married after WFS are shown in Table 4.5. Considering the reinterviewed respondents first, these are a subset of the ever-married women interviewed at WFS, and it is therefore possible to broadly consider the parity distributions observed at WFS as pertaining to this group when they were in the immediately younger five-year age group. Comparison of their parity distribution in 1975 given in column 1 of Table 4.5 with that in 1979 given in column 1 of Table 4.6 reveals a broad conformity between the two surveys: at all ages there is an increase in the - 125 - mean parities, reflecting the births added during the 4 interim years. For example, women 15-19 in 1975 had an average of .75 births; when these women were approximately 20-24 in 1979, their average parity had increased to 2.45, reflecting an addition of 1.7 births during the 4-year interval. With the exception of the oldest age group, the increments decreased with increasing age; e.g., it is only one child for those between 30-39 years in 1975 because of the slow rate of childbearing in advanced years. The average mean parity for the entire group had increased by .7 births from 3.94 to 4.68. Women who married after WFS have had an average of 1.2 births. Their parity distribution is shown in column 2 of Table 4.6.' Unfortunately because of the very small frequency of first marriages after age 35, the estimates for these ages are based on cell sizes too small to be reliable. Women 25-34 years of age who can be considered to have had approximately the full 4-year interim period of exposure have had an average of 1.5 children, the same number as the additional children born to women of the same age who were already married and who had borne some children. The very small parities observed for the 15-24 age groups may be because a large proportion of these women got married towards the end of the interval and therefore had a small average duration of exposure. Adolescent subfecundity could also have been a cause of the small number of births to the very young women of 15-29 years. - 126 - Table 4.6: Mean Parity by Age: All New Respondents, Re-interviewed Respondents, and Respondents Who Got Married After WFS. Respondents Who Re-interviewed Married After All New Respondents WFS Respondents Age (1) (2) (3) 15-19 2.25* 0.61 0.97 20-24 2.45 0.99 1.21 25-29 3.04 1.53 1.86 30-34 3.94 1e55 2.53 35-39 4.86 1.87* 4.35 40-44 5.84 2.33* 4*53 45-49 5.96 175* 5.76 All 4.68 1.21 2.39 Sample size 3,303 957 1,277 *Base'd on sample size less than 15. Table 4.7: Mean Parity by Years Since First Marriage, for Ever-Married Women (WFS and WBFS). Years Since Ever-Married Women First Marriage WFS WBFS 0-4 0.9 0.9 5-9 2.5 2.4 10-14 3.7 3.5 15-19 4.9 4.6 20-24 5.7 5e6 25-29 6.5 6.3 30-34 6.9 7.0 35+ 6.7* 6.8 All 3.9 3.8 *For these women there is internal evidence of underreporting of- early births, particularly of female births, It is suggested that their mean parity be taken to be tlhat of the immediately subsequent marriage cohort. Source: WFS data are from the World Fertility Survey, Sri Lanka, 1975 First Report. - 127 - Parity By Marital Duration: Cumulative patterns of childbearing can also be examined by marriage duration as by age. Mean parities for women of five-year marriage duration groups are shown in column 2 of Table 4.7. The cumulative number of births rises steadily with increasing duration but with progressively diminishing increments as it does with age. Both the level and the pattern of distribution are almost identical to those observed at WFS, shown in column 1. The mean parity of 6.8 children of women who have completed 35 or more years of marriage is too high an estimate of the completed fertility when compared to 5.6 observed for the cohort of women of age 45-49 years. The higher parity of the marriage cohort is due, as explained in the First Report of WFS, to the selection criteria of the survey limiting the upper age of the respondents to 49 years: women in the survey who were of marriage duration 35 or more years are a select group who married very young and were especially fertile, while women of equal duration but older in age who married not so youug and therefore not so fertile were excluded from the survey. This selection bias which allows only early-marrying women to be included in the long marital duration cohorts biases positively the mean parity of these women. Multivariate Analysis of Differentials in Cumulative Fertility: The Method. Differentials in cumulative fertility are discussed for women who have been married for 10-19 years rather than for the entire sample of women. This selection is made because these women, while representing a broad range of ages from mid-20's to mid-40's, have already had their first few births. The method of analyzing differentials is the same as in the - 128 - previous section on cumulative fertility except that the regressions are now performed on the 10-19 year marriage duration cohort. Table 4.8 gives the unadjusted and adjusted mean cumulative fertility levels by each of the background variables. A brief discussion of the fertility differentials based on these tables follows. Differentials by Zone of Residence. Cumulative fertility of the 10-19 year marriage cohort shows some degree of variation by zone of residence. Largely responsible for this variation are the high fertility level of 4.77 of Zone 4 and the low fertility level of 3.6 of Zone 1. The other zones have fertility levels lying within close range of the overall mean of 3.9 children. The most conspicuous feature of the zonal differentials in cumulative fertility is the remarkable absence of any effects on these differentials by other variables, not even by age at marriage. Each successive adjustment for other variables leaves unchanged the mean fertility of all zones except Zone 4. For Zone 4, the mean level is raised by about .1 when adjusted for age at marriage but brought down by about .2 when controlled for racial composition; other controls have no effect. Thus, the differentials in cumulative fertility by zone of residence of this marriage cohort cannot be attributed to any of the variables considered. Differentials by Place of Residence. Cumulative fertility does not show any significant variation by residence. The mean levels are 4.1 in urban, 3e9 in rural and 4.1 in estate areas. Nor are the individual levels changed by the demographic or background variables. Differentials by Race-Religion. Some difference in cumulative fertility can be observed among different race-religion groups. - 129 - Table 4.8: Differentials in Mean Fertility of the 10-19 Year Marriage Cohort (Unadjusted and Adjusted). Duration Place Hus- Hus- of and of Race/ Wife's band's band's Unad- Age at Resi- Reli- Educa- Wife's Educa- Occupa- Subgroup justed Marriage dence gion tion Work tion tion Zone 1 3.63 3.68 3.70 3.67 3.66 3.66 3.62 3.63 Zone 2 3.82 3.80 3.80 3.84 3.86 3.86 3.87 3.85 Zone 3 4.02 4.02 4.01 3.99 3.93 3.95 3.94 3.96 Zone 4 4.67 4.74 4e74 4.52 4.47 4.47 4.47 4.50 Zor.e 5 4.17 4.24 4.24 4.14 4.23 4.21 4.21 4.20 Zone 6 3.84 3.82 3.82 3.85 3.85 3.85 3.86 3.87 Rural 3.94 3.91 3.92 3.94 3.93 3.92 3.92 3.92 Urban 4.15 4.13 4.14 4.15 4.15 4.14 4.13 4.14 Estate 4.07 4.05 4.06 4.07 4.07 4.06 4.05 4.06 Sinhalese: Buddhist 3.79 3.78 3.86 3.84 3.86 3.86 3.86 3.86 Christian 3.91 3.90 3.78 3.79 3.89 3.81 3.93 3.99 Tamil: Hindu 3,95 3.96 3.84 3.82 3.83 3.81 3.81 3.81 Christian 4.51 4.55 4,45 4.48 4.50 4.51 4.52 4.49 Moors 4.65 4,65 4.48 4.49 4.38 4.37 4.36 4.32 Wife Worked Never 3.97 4.01 4.00 4.01 3.99 3.99 3.98 3.97 Before only 3.95 4.03 4e06 4.07 4.11 4.07 4.02 4.03 After only 3.81 3.70 3.64 3.64 3.64 3.63 3.68 3.68 Before & after 3.86 3.81 3.84 3.82 3o83 3.86 3.87 3.89 Education None 4.32 4.23 4,13 4e16 4.12 4.13 4.16 4.16 Primary 4.07 4.05 4.07 4.07 4.06 4.06 4.02 4.01 Secondary 3.81 3,83 3.86 3.83 3.87 3.84 3.83 3.82 High School 3.43 3.54 3.52 3.49 3.53 3.55 3.63 3.67 University 2.41 2,46 2.60 2.56 2.59 2.62 2.73 2.90 Husband Education None 3.55 3.43 3.39 3.39 3e39 3.33 3.36 3.40 Primary 4.30 4.27 4.24 4.24 4.24 4.17 4.17 4.14 Secondary 3.90 3.93 4.00 4.00 4.00 3.99 3.98 3.93 Higher 3.51 3.61 3.60 3.59 3.59 3.80 3.78 3.91 Occupation Agricultural 3.84 3.84 3.78 3.78 3.79 3.76 3.76 3.85 Prof./Clerical 3.36 3.37 3.42 3.43 3.44 3.67 3.63 3.59 Sales/Services 4.24 4.24 4.31 4.31 4.23 4.26 4.24 4.17 Skilled 4.11 4.11 4.16 4.17 4.20 4.18 4.15 4.07 Unskilled 4.10 4.10 4.09 4.09 4.08 4.03 4.03 3.95 - 130 - The differences lie in the high fertility levels of the Moors, 4.6, and Tamil Christians, 4.5. The. Sinhala Buddhists, Sinhala Christians and Tamil Hindus, with levels of 3.8, 3.9 and 4.0, ripectively, have nearly equal . fertility levels. Adjustments for other variables have only marginal effects on the fertility levels of various race-religion groups except of the Moors. The high fertility level of the Moors is gradually brought down to 4.3 when zone, education, work and occupation variables are adjusted. Significantly, the fertility of the Moors in this marriage cohort is not attributable to early marriage. The adjusted fertility level is highest for the Tamil Christians, 4.5; the Moors follow closely, 4.3 , but the two Sinhalese groups and the Tamil Hindus remain close together near the overall mean of 3.9. Thus the Sinhalese (Buddhists and Christians) and the Tamil Hindus, together comprising over 90 percent of the population, are at one level of childbearing; only the very small minority of Tamil Christians and Moors have a higher level of fertility. There seem to be no important racial differences, then, in the reproductive behavior of these women who were completing their families in the second decade of marriage. This contrasts with the strong racial differences observed for completed fertility earlier. Differentials by Education. As in the discussion of cumulative fertility, differentials by wife's and husband's education will be examined together. Education is the variable by which the strongest differentials exist in cumulative fertility. The unadjusted means decrease monotonically with the rising level of wife's education. The greatest fall is from high school to university (which category also includes wives of university educated men), spanning a range of over 2 children. 131 - Adjustments for other variables have only marginal effects on primary and secondary levels. The low fertility of the university group is attributable in a very small measure to their concentration in low fertility zones and husband's occupation, but not to their age at marriage distribution. Even after adjustments are made, the mean fertility of this group remains as low as 2.9, one child below the overall mean of 3.9 children. The mean of ehe high school category is also reduced to a very small extent, but the means of primary and secondary levels are left virtually unchanged by each of the other variables. Thus the educational differentials persist to a large extent eves after all adjustments. Fertility differentials also exist by husband's education, but the relation between fertility and husband's education is no longer monotonic. Mean fertility level rises from 3.5 for no schooling to 4.3 for primary, and thereafter falls to 3.9 for secondary and 3.5 for high school levels. Thus, there is no difference between no schooling and high school levels. These educational differentials are not significantly affected by adjustments for variations by age at marriage, although they lower the mean of the primary group by .1 and raise that of the high school group by .1. The high fertility level of wives of primary-educated men is partially attributable to their early marriage and their lower-than-average educational attainment. The low fertility of wives of high school educated men is entirely attributable to their late marriage, their higher education and the higher-status-occupation profile of their husbands. When adjusted for the other variables, educational differentials disappear among all levels, except that the no schooling category remains about .5 below the oerall mean of 3.9. - 132 - Differentials by Pattern of Work. Womens' pattern of work does not bring about any significant fertility differentials. The mean level of fertility for all four categories is almost the same as the overall mean of 3.9 children. These means are not affected in a significant way by the compositional variations of any of the variables. This absence of differentials in cumulative fertility by work status contrasts with the strong unadjusted differentials observed for completed fertility in this sample and for completed and late fertility (i.e., fertility in the second decade of marriage) in the WFS sample. Differentials by Husband's Occupation. Husband's occupation is one of the variables by which cumulative fertility differentials exist. As expected, unadjusted fertility is lowest, 3.5, among professional and clerical occupations. Agricultural workers come next with 3.8 children. The two groups skilled and unskilled workers have a higher fertility level of 4.1, and the sales and service workers have the highest level of 4.2. The low fertility of the professional and clerical group is mainly attributable to the better-than-average educational attainments of the wives. The fertility levels of the other occupational groups are only marginally affected by adjustments for other variables. The result is that, while professional,and clerical occupations having an adjusted mean of 3.6 retain their ranking at the lowest level, all other occupations have mean levels within the range of 3.8 to 4.2, around the overall mean of 3.9. Current Fertility The indicator of fertility discussed in this section is the percentage of women reporting a current pregnancy. Information on current pregnancy was collected using two questions: "Are you pregnant now?" and if - 133 - so, "What is the duration of pregnancy?" The responses to the first question were classified into three groups: Yes, No, Don't Know. The responses to the second were recorded in months, ranging from 0 to 8 or more. Responses on pregnancy status are usually subject to two sources of bias: (1) negative bias resulting from not reporting a current pregnancy because of uncertainty, unawareness or other reason, and (2) positive bias resulting from non-live-birth terminations of some pregnancies. Although these biases are compensatory in direction they are of uncertain magnitudes. The age distribution of percentages reporting a current pregnancy are shown in Table 4.9, along with comparable data from WFS. The overall proportion of pregnant women is 9.4 percent, which compares well with the, 10.1 percent observed at WFS. Among women below 35 years, with the exception of those 15-19 years, there had been a small increase in reported pregnancies between the two surveys; on the other hand, among older women 35 to 49 years, slightly lower proportions were reported pregnant. It is likely that this increase was due to improved reporting rather than a real increase of pregnancies. WFS First Report states that reported pregnancies in the age groups 15-29, 30-39 and 40-49 were about 90 percent, 60 percent and 50 percent, respectively, of the estimates expected on the basis of the actual birth rate of the previous year. The levels and age pattern of pregnancies are in close conformity between the two surveys. The age pattern of pregnancy for re-interviewed and new respondents shown in columns 3 and 4 of the table shows a large proportion of 18 percent were pregnant among new respondents, while the proportion was only 6 percent among re-interviewed rspondents. Apparently, the differences are mainly among the young women below 35 years. This higher incidence of pregnancies 134 - among new respondents is expected because over 50 percent of them were married within the previous four years and were therefore expecting their first or second births, for which the rates are usually much higher than for higher order births. Table 4.9: Percentage of Women Reporting a Current Pregnancy, by Age (WFS and WBFS). WFS WBFS Current Age (1) (2) (3) (4) 15-19 25.2 23.2 36.8 24.0* 20-24 22.6 24.5 28.7 25-29 15.1 16e5 16e5 19.3 30-34 9.0 9.4 8.3 14.2 35-39 6.3 4.4 4.8 2.9 4044 1,7 1.6 1e5 2e5 45-49 0.3 0.2 00 2.0 All 10.1 9.4 6.4 18.7 *The two age groups are combined because of the small number of cases in the 15-9 age group. Source: WFS data are from WFS First Report. Fertility Trends A preliminary assessment of fertility trends can be made by comparing the age specific fertility rates and age specific marital fertility rates for 1978 computed from the survey data with those for previous years. Age specific fertility rate is defined as the ratio of the number of births in a given period of time occurring to women in a specified age group to the number of woman years of exposure in that age group in the same period irrespective of the marital status of the women. Total Fertility - 135 - Rate (TFR), defined as the sum of the age specific fertility rates for age 15-49, represents the average number of births a woman would have at the end of her reproductive span if she experienced the specified age specific fertility rates. Marital fertility rates are defined similarly with the exception that only births within marriage are included in the numerator, and only currently-married women in the denominator of the ratio. Table 4.10 presents the age specific fertility rates and age specific marital fertility rates for 1978 computed from the survey data for 1974 from the WFS survey data, and for 1963 and 1970 from census data. The Total Fertility Rate does not show the continuation of decline between 1974 and 1978 that occurred between 1963 and 1974. TFR fell precipitously from 5.04 in 1963 to 4.2 in 1970, and to 3.5 in 1974, but thereafter it steadied itself, remaining at 3.4 in 1978. This is due to a small increase in the age specific fertility rates it the 25-34 age range which has counterbalanced the declines in the other age groups. The greatest declines have occurred in the 20-24 and 35-39 age groups. Within marriage, too, the fertility change has shown a similar pattern. Up to age 25 and between ages 35-39, marital fertility has declined, but between 25-34 it has shown an increase. It is noteworthy that between 1970-1974 it is for this particular age range that the sharpest fall (26 percent) in fertility rates was observed, although it is more common internationally for marital fertility to begin declining at older ages. These observations based on age specific fertility rates are only a broad indication of recent fertility trends; the changes observed may be Table 4.10 Age Specific Fertility Rates and Marital Fertility Rates for 1963, 1970 {Registration Data) (1975, WFS; 1979, WBFS). Age Specific Fertility Rates Age Specific Marital Fertility Rates Age Group 1963 1970 1974 1978 1963 1970 1974 1978 15 - 19 52 38 31 38 354 449 339 310 20 - 24 228 172 146 129 396 408 357 316 25 - 29 278 238 161 186 344 323 240 272 30 - 34 240 219 158 164 270 253 189 201 35 - 39 157 134 126 105 175 1.51 139 124 40 - 44 46 38 43 55 53 42 53 66 45 - 49 7 6 6 9 8 7 7 12 TFR 5.04 4.22 3.35 3.43 - Source: For 1963, 1970 and 1974 rates, see Sri Lanka Department of Census and Statistics. World Fertility Survey 1975 First Country Report. ,Table SB. 1978 Rates are computed from Appendix Tables C 1 and C 2. I - 137 - real changes or errors in the data or a composite of both. A more comprehensive estimation of trends awaits a more detailed analysis of birth history data subject to an evaluation of their quality. Conclusion This chapter examined the levels and differentials in completed and cumulative fertility and broadly as^essed the current fertility levels and recent fertility trends. The level of completed fertility of the survey population 45-49 years was 5.6 children, a little less than the 6.0 observed at the WFS. Considerable differentials in completed fertility were found to exist by zone of residence, race, education, and husband's occupation. These differences could be attributed partly to variations in age at first marriage and, to smaller but varying degrees, to the various background variables. Net of the effects of other variables some differentials remained. It is noteworthy that these residual differentials were largely confined to one or two deviant groups, while all other groups stayed close together in a narrow range about the overall mean. Among zones, the high fertility residents of Zone 4; among ethnic-religion groups, the low fertility Christians and high fertility Moors; and among the educational levels, the low fertility university group were the subgroups that deviated from the norm in completed fertility. In contrast, all occupational classes differed considerably from one another, spanning a wide range, although there was one 3roup--the low fertility professional and clerical class-that could be identified as the most deviant group. Interestingly, pattern of work brought about only moderate differences in fertility levels, and these were entirely attributable to other factors. Between urban, rural and 138 - estate areas there were very little differences in completed fertility. The one noteworthy observation was that adjusted fertility level was the lowest in estates, although the unadjusted level was close to rural and higher than urban levels, in contrast to the observation made at WFS that estate fertility was the lowest despite a low age at marriage. Turning to cumulative fertility, the mean parity of all ever-married women in the sample was found to be 3.7, which represented a reduction of 4.8 percent since the WFSO The mean parity of currently--married women also registered a 5.5 percentage reduction from the WFS level. 20-24 years was the age group that most contributed to this reduction in fertility within marriage. Differentials in cumulative fertility of the 10-19 year marriage cohort was found to exist for the same variables as in completed fertility. The groups that differed considerably from the norm were: high fertility Zone 4 and low fertility Zone 1, high fertility Moors and Tamil Christians, the,low fertility university group, and low fertility professional and clerical classes. The most significant feature of differentials in cumulative fertility, in contrast to completed fertility, was that they were not attributable to the variations in age at marriage. This is also in strong contrast to the observation in the WFS sample that age at marriage is the most important factor to which fertility differentials of ten-year marriage conorts can be attributed. Age specific fertility and marital fertility rates indicate that in the four-year period between 1974 and 1978 there was some decline in the fertility levels in the young age groups but a slight increase in the 20-34 year range. - 139 - APPENDIX TO CHAPTER IV 4A. Current Marital Status by Age, for Women in Household Population Ages 12-49 Years. 4B. Births in the Calendar Year 1978, by Age of Mother at the Time of the Birth. 4C. Sample Weights by Zone. - 140 - APPENDIX 4A Current Marital Status by Age--Women in Household Population of Age 12-49 Years Age Single Married Widowed Divorced Separate Total 12 418 0 0 0 0 418 13 415 0 0 0 0 415 14 451 3 0 0 0 454 15 395 4 0 0 0 399 16 413 14 0 0 0 427 17 366 22 0 0 1 389 18 343 50 0 0 3 396 19 314 65 0 0 4 383 20 251 88 0 2 2 343 21 213 95 0 1 3 312 22 228 113 0 1 5 347 23 179 117 0 1 5 302 24 153 148 4 1 10 316 25 103 163 3 2 5 276 26 106 149 2 1 6 264 27 83 195 1 2 5 286 28. 66 172 2 2 4 246 29 71 182 3 0 7 263 30 43 208 5 0 7 263 31 34 187 5 0 8 234 32 39 216 0 0 9 264 33 18 141 8 0 4 171 34 21 178 11 3 6 219 35 15 162 10 2 7 196 36 21 155 6 0 3 185 37 16 188 16 1 3 224 38 11 116 6 0 4 137 39 15 193 12 3 8 231 40 6 112 9 0 2 129 41 3 130 10 0 0 143 42 10 140 18 3 6 177 43 4 114 14 1 4 137 44 8 161 13 1 8 191 45 5 95 19 0 3 122 46 3 106 22 0 9 140 47 3 135 28 1 11 178 48 2 61 24 0 4 91 49 3 128 28 1 9 169 50 1 71 11 0 1 84 - 141 APPENDIX 4B Births in the Calendar Year 1978 by Age of Mother at the Time of the Birth Age No. of Births 12 00 13 00 14 00 15 01 16 05 17 12 18 18 19 38 20 33 21 43 22 28 23 45 24 52 25 63 26 54 27 42 28 49 29 38 30 50 31 42 32 36 33 29 34 21 35 27 36 25 37 15 38 18 39 10 40 10 41 13 42 08 43 09 44 02 45 02 46 02 47 02 48 00 49 00 - 142 - APPENDIX 4C The sample of this survey was exactly the same sample of households selected in the WFS except for non-response. The sample weights for zones, therefore, are computed by adjusting the corresponding WFS sample weights for non-response; that is, the sample weight for each zone is obtained by multiplying the weight assigned to that zone at the WFS by the inverse of the non-response rate. These weights were normalized to make the weighted and unweighted sample sizes equal. The following table shows the computations. Sample Weights by Zone No. of Response WBFS WFS Respondents Rates Weights Norm Weights WFS WBFS (3) (2) (1) x 1/(4) Weights Zone (1) (2) (3) (4) (5) (6) 1 0.4647 927 611 0.6591 0.7050 0.5018 2 1.7376 1083 755 0.6971 2.4925 1.7743 3 0.8216 1246 607 0.4872 1.6865 1.2006 4 0.4435 861 811 0e9419 0.4708 0.3351 5 0.6090 775 570 0.7355 0.8280 0.5894 6 1.3503 1920 1475 0.7682 1.7577 1.2512 -143- CHAPTER V FERTILITY PREFERENCES, INTENTIONS AND BEHAVIOR Some couples are prone to run their lives in accordance with detailed, pre-specified plans, while some are likely to be at the other extreme, leaving virtually everything to factors outside their control. The distribution of couples may be skewed to one of these extremes in some populations, while it may be skewed to the other extreme in some others. When two-wave data are available, as they are for Sri Lanka, it is possible to measure the degree to which people's reproductive behavior conforms to or deviates from their earlier plans. In this chapter we focus on questions such as the following: 1. What factors determine couples' fertility behavior between two time points? In particular, do expressed intentions predict subsequent fertility? 2. What are the covariates of inconsistency between expressed intentions and subsequent behavior? 3. How are intentions and fertility preference at one time point related to those at an earlier time point? 4. What are the covariates (determinants) of fertility preference at a given time point? Determinants of Fertility Preference: A Preliminary Analysis We examine the last question first. The reason for doing so is that this question has been examined using the 1975 (WFS) data (Pullum, 1980); hence, by repeating some of the analyses it should be possible to assess the quality of the material included in this chapter as well as tb implications of selectivity, if any, in the re-intervitewed subsample. - 144 - The measure we shall use for fertility preference is the response to the interview question: "If you could choose to have exactly the number of children to have in your whole life, how many would that be?" We suspect that this measure is strongly influenced by the tendency to rationalize, i.e., the tendency to revise the response over time to be consistent with the actual family size attained. On this basis, the actual family size is expected to be a strong predictor of fertility preference. Representing family size (= number of children living) as a class (categorical) variable, taking values 0, 1, ...e 10+, the regression of fertility preference at time 1 (1975, WFS) on actual family size as of time 1 yields an R2 of 0.553 for the re-interviewed subsample. The corresponding figure for the WFS full sample is 0.560 (Pullum, 1980), When three age-related controls are introduQed (along with family size), R2 increases to 0.559. The estimated regression coefficients and the associated standard errors of these control variables are shown in Table 5.1. (See the section on "A Modified, Extended Model," below, for a discussion of these age-related controls and their relationships to fertility preference. See also Pullum, 1980.) As these regression coefficients indicate, given the number of children living as of time 1 (1975, WFS), the higher the age at first marriage or the longer the interval since last live birth, the smaller the preferred family size-the effect of the interval between first marriage and last live birth being insignificant. Note that in all these respects the WFS full sample and the reinterviewed subsample look similar, indicating that the re-interviewed subsample is not biased with respect to the covariates of fertility preference. - 145 - Table 5.1: Regression Coefficients (Dependent Variable--Fertility Preference). 1975 WFS 1979 Re- Regressors* Full Sample Interviewed Sample Age at first marriage -.0014 -.0197 (.0003) (.0055) Interval between first marriage and last live birth (WFS) .0006 .0007 (.0004) (.0005) Interval since last live birth (WFS) -.0012 -.0019 (.0002) (.0004) *Includes number of children living, not shown in the table. To proceed with the analysis, adding selected regressors one at a time, each as a class variable, yields the R2's shown in Table 5.2. Table 5.2: Multiple R2 from Different Regressions (Dependent Variable--Fertility Preference). WFS 1979 Re- Regressor Full Sample* interviewed Subsample Region .569 .563 Type of place of residence .565 .561 Ethnic group .566 .564 Religioia .566 .563 Wife's education .565 .561 Literacy (wife) .565 .561 Pattern of wife's work .565 .560 Occupation of husband .565 .561 *The WFS full sample figures are from Pullum (1980). Each regressor accounts for a statistically significant fraction of the variance of fa&uily-size preference left unaccounted for by the - 146 - combination of current family size and the age-related controls. (The variable "pattern of wife's work" used by Pullum in his analysis of the WFS full sample is somewhat different from the one used in the analysis of the re-interviewed subsample. But it is unlikely that the results would be much different if Pullum's categories had been used instead of those available in the WFS standard tape, the ones we used in the analysis of the re-interviewed subsample.) The 1975 versus 1979 comparison, based on the two-wave data shown in Table 5.3, indicates very little change in the effects of the various factors on the family size preference. Subtracting the R2 for equation 2 from that for equation 3, we notice that region of residence accounts for 0Q4 percent of the variance at time 1 and 0.9 percent at time 2; the corresponding figures for type of place of residence being 0.2 and 0.2, respectively; and so on. Table 5.3: Multiple R2 of Different Regressions (Dependent Variable--Fertility Preference Reinterviewed Subsample). 1975 (WFS) 1979 Regressors (time 1) (time 2) 1. Children living at t .553 .502 2. Age-related controls and children living .559 .506 3. Those in 2 and region of residence .563 .515 4. Those in 2 and type of place of residence .561 .508 5. Those in 2 and ethnic group .564 .512 6. Those in 2 and wife's education .561 .510 7. Those in 2 and wife's literacy .561 .507 8. Those in 2 and wife's religion .563 .513 9. Those in 2 and wife's pattern of work .560 .507 - 147 - Of these results, the following two points are noteworthy: (1) The analysis of the 1975 data for the re-interviewed subsample gives results very similar to those obtained by the corresponding analysis of the 1975 full sample, indicating that the re-interviewed subsample is not a biased subsample from the point of view of the relationship between the various factors examined and the fertility preference. (2) The most powerful predictor of fertility preference is the current family size (number of children living), but other variables also exhibit statistically significant predictive power. We shall return to the examination of the determinants of ferility preference in the context of a recursive model. The primary objective of the results presented in Tables 5.1-5.3 has been to determine whether the re-interviewed subsample is biased. Although one cannot be certain about this, it is safe to assume that for the analysis of fertility preferences and its correlates we may treat our re-interviewed subsample as a random subsample of the original WFS (1975) full sample. Declared Intention and Subsequent Behavior We now turn our attention to an analysis of the inconsistency between declared intention and subsequent fertility behavior. When expressed desire for additional children is compared with subsequent fertility behavior, it is possible to classify a respondent as consistent or inconsistent, depending upon whether the behavior conforms to the expressed intention. Thus, a respondent is inconsistent if she wanted no more children but had one or more births in the follow-up period. Examination of such inconsistencies is useful in fertility studies insofar as it sheds light on the degree to which people plan their reproduction and - 148 - successfully implement that plan. The (1975, 1979) two-wave data permit a limited examination of the inconsistency just referred to. The available data do have a major limitation, however. The period covered by the behavior measure does not coincide with the period specified in the question on intentions. The behavior measure is for the four-year period, 1975-1979, whereas the intention (desire) measure refers to the entire reproductive period extending from the first interview date onward, the wording of the interview question being, "Do you want another child sometime?" Table 5.4: Distribution of Women in the Sample According to Whether They Wanted Another Child Sometime and the Number of Live Births in the Interim (Two-Wave Data, 1975-1979). Whether the Woman Live Births in the Interim Wanted Another Child Sometime None One or more Total 1,031 634 1,665 Row %=61.9 Row %=38.1 Row %=100.0 No Cell A Cell B Col.%= 67.2 248 563 811 Row %=30.6 Row %=69.4 Row %=100.0 Yes Cell D Cell C Col.%= 32.8 1,279 1,197 2,476 Total Row %=51.7 Row %=48.3 .4 The incompleteness of the available data, from the point of view of the information on the inconsistency in question, can be better appreciated by examining the distribution shown in Table 5.4. Of the 2,476 women for whom the relevant information is available, 67.2 percent (1,665) said they did not want any more children, while 32.8 percent (811) said they - 149 - wanted at least one more child. Of the 1,665 in the former subgroup, 1,031 (see Cell A) had no births in the interim, while 634 (see Cell B) had one or more; and of those (811) who wanted another child, 248 (see Cell D) had none, while 563 had one or more. Obviously, over the course of time some women in Cell A will shift to Cell B as they bear additional children. Similarly, transfers from Cell D to Cell C are also likely. Any transfer from Cell A to Cell B would increase the number of inconsistents from the level shown in Table 5.4, and any transfer from Cell D to Cell C would decrease it. It follows, therefore, that the picture of inconsistency revealed by the figures in Table 5.4 is incomplete because of the truncation of the follow-up period. It seems safe to assume, however, that the transfers from Cell A to Cell B in the future will be proportionately as well as in absolute numbers larger than the transfers from Cell D to Cell C, the reason being that Cell D contains relatively fewer women, a large number of whom have already had long open intervals by the second interview date (e.g., 122 had an open interval of 4 years or more). Hence, it is reasonable to assume that the number of inconsistents would increase rather than decrease over time from the level shown in Table 5.4. The comparative figures for Sri Lanka, Taiwan, and the Republic of Korea in Table 5.5 reveal that the inconsistency is considerably larger in Sri Lanka than in the other two countries. Proportionately fewer Sri Lankan women are able to prevent unwanted births compared to their Taiwanese counterparts, while proportionately more are able to do so compared to Korean women. (See percent of inconsistent women among those who wanted no more.) In comparison with the Taiwanese and Korean samples, the Sri Lankan sample shows a stronger tendency to set goals which are not likely to be - 150 - Table 5.5: Comparison of Inconsistency Measures from Sri Lanka with Those from Taiwan and Korea. Aggregate-level Individucal-level Measure of Inconsistency Measure of Inconsistency % Inconsist. % Inconsist. % who % who among those among those wanted had who wanted who wanted Whole Population more more C more: no more: group and period A B =(B-A)100/A Pi PI Pi -_w w a Taiwan 1967-70 45.7 41.6 -9.0 25.0 13.5 18.8 1970-74 31.6 32.4 2.5 27.9 14.1 18.5 1967-74 47.1 51.8 9.9 14.6 21.9 18.5 1967-72 45.1 47.0 4.2 17.0 17.5 17.3 Korea 1971-76 67.5 77.5 14.8 12.2 55.8 8.4 Sri Lanka 1975-79 32.8 48.3 47.3 30.6 38.1 35e6 Per annum basis* Taiwan 1967-74 1.4 2.1 3.1 Korea 1971-76 3.0 2.4 11.2 Sri Lanka 1975-79 11.8 7.7 9.5 *Obtained by dividing the measure for the follow-up period by the length of the period. Sources: Taiwan figures are from Hermalin et al., 1979; Korean figures are from Foreit and Suh (1980); and Sri Lanka figures are from Table 4. - 151 - attained. (See percent of inconsistent women among those who wanted more.) The occurrence of "unwanted" births may reflect, in part, less efficient or less frequent use of contraception, as the failure to attain targets of additional children may be explained by the onset of subfecundity or effective intervention of the family planning program. Also, part of the explanation for the observed inconsistency among those who wanted more as well as among those who wanted no more may lie in genuine plan changes during the follow-up period. The available data do not permit identifying who, if any, changed their plans (desires, intentions) during the interim; however, data are available on the respondent's fertility intentions (desires) as of the second interview. The relationship between the 1975 (time 1) and 1979 (time 2) intentions will be examined later on. For now, we confine attention to the observed inconsistency between expressed intentions (desires) and behavior in the interim. Correlates of Inconsistency Table 5.6 shows that the percentage of inconsistents among women who wanted more children increases rapidly with marriage duration, with wife's age, and with number of live births. At the same time, the corresponding percentage of inconsistents among women who wanted no more children decreases. Furthermore, the percentage of women who wanted more children decreases with age, with marriage duration, and with number of live births. - 152 - Table 5.6: Correlates of Inconsistency, Two-Wave Data (Sri Lanka, 1975-79). Percent Inconsistent Among Those Those Percent Percent who who Number wanted had wanted wanted of Characteristics more more more no more All women Total 32.8 48.3 38.1 30.6 35.6 2,746 Duration of marriage <5 years 79.4 83.2 15.6 79.2 28.5 474 5-9 years 41.8 65.6 30.5 62.9 49.4 541 10-14 years 21.3 45.3 43.9 42.5 42.8 535 15+ years 10.0 22.0 75.3 21.7 27.1 926 W4ife's age <20 years 81.4 90.0 8.8 84.6 22.9 70 20-24 years 62.9 79.5 17.1 74.1 38.0 389 25-29 years 45.3 69.2 23.1 62.8 44.8 592 30-34 years 23.6 43.3 47.8 40.6 42.5 568 35-39 years 15.1 30.5 61.1 28.8 33.7 475 40-44 years 9.9 7.2 93.3 7.3 15.8 304 Number of live births 0 91.2 66.2 34.3 73.3 37e4 190 1 78.3 72.7 25.5 66.2 34.3 341 2 48.0 58.9 29.9 48.6 39.7 348 3 24.0 49.3 32,2 43.5 40.8 363 4 17.4 41.0 36.2 36.0 36.0 333 5 9.2 38,5 40.0 36.4 36.8 272 6+ 5.1 31.6 34.4 29.6 29.9 629 Number of living sons 0-1 53.6 59e3 30.4 47.3 38.2 1,253 2 15.4 42.6 25.3 36.7 34.9 538 3+ 7.9 32.9 40.7 30.6 31.4 685 Open interval 12 months* 39.6 79.4 14.6 68.9 47.4 846 12-23 39.6 57e6 21.2 43.6 34.8 417 24-35 28.6 52.1 24.1 42.5 37.2 290 36-47 28.6 36.2 39.3 26.4 30.2 196 48+ 23.7 13.3 70.9 8.5 23,2 727 Wife's education none 23.5 47.2 30.0 40.0 37.6 425 grades 1-5 29.2 47.0 31.6 38.1 36.2 1,003 grades 6-9 34.8 49.4 29.1 37.9 34.9 700 grade 10 or higher 49.7 51.7 31.2 34.9 33.0 348 *Includes women who:were currentiy pregnant at the first interview; their open interval being taken as 0. -153- Table 5.6 (cont'd) Percent Inconsistent Among Those Those Percent Percent who who Number wanted had wanted wanted of Characteristics, more more more no more All women I. Wife's work pattern never worked 34.2 50.4 28.7 39.4 35.7 1,371 worked only before marriage 35.4 50.8 30.4 40.7 37.1 259 worked after marriage: away from home 27.0 42.0 39.0 34.8 35.9 540 at home 33.7 48.7 27.2 36.4 33.3 306 Ethnic group Sinhalese 30.1 43.2 32.3 32.5 32.5 1,435 Sri Lankan Tanril 36.4 54.2 28.6 44.6 38.7 537 Indian Tanril 34.4 50.2 31.3 40.6 37.4 195 Sri Lanka Moor 37.5 62.1 27.3 55.5 44.9 292 Type of place of residence Urban 31.7 44.0 35.0 34.2 34.5 618 Rural 32.7 49.7 29.1 39.3 36.0 1,657 35.3 50.8 29.6 40.0 36.3 201 Religion Budhist 30.5 43.6 32.0 32.7 32.5 1,329 Hindu 36.3 53.6 29.6 44.2 38.9 633 Muslim 37.6 61.4 26.3 53.7 43.4 302 Christian 28.9 43.6 32.8 34.2 33.8 210 Zone (region) 1 29.7 37.3 36.3 26.0 29.1 306 2 26.3 41.9 39.6 34.8 36.1 382 3 35.4 50.0 27.0 38.5 35.1 322 4 33.1 62.8 22.0 55.7 44.5 382 5 43.2 52.8 31.7 41.0 37.0 322 6 31.3 46.3 29.3 35.0 33.2 762 - 154 - Now, if PIw = percent of inconsistents among women who wanted more, PIw- = percent of inconsistents among women who wanted no more, and X = percent of women who wanted more, then 100PIa =XPIw + (100 - X) PIW, where PIa is the percent of inconsistents among all women (those who wanted more plus those who did not want any more). At the early stages of the life course (shorter marriage duration, younger age, or lower parity), X, the percent of women who want more children, is large. Moreover, most of those who want more children have them, as a consequence of which PIw is small. A large proportion of the very small group who want no more children also have more, making for a large Pl-w. But, in the calculation of PIa, it is the small PIw that gets larger weight (x), and the result is a smaller PIa" At the later stages of the life course, X, the percent of women who want more, becomes small, while PIw becomes large and PI-W small. Again, the larger weight, 100-X, goes to the small quantity, PI:W in the present case, yielding a smaller PIa' At the intermediate stages of the life course, however, X, PIW, and PIW all remain moderately large, yielding a moderately large PIa- As for the other correlates of inconsistency, we note from Table 5.6 that inconsistency decreases with the open interval and decreases somewhat with the wife's education. The association of inconsistency with the pattern of the wife's work is not clear. - 155 - Among the ethnic groups, the Sinhalese are the least and the Sri Lankan Moors the most inconsistent. Likewise, the Buddhists are the least and the Muslims the most inconsistent, the Hindus remaining in the middle, and the Christians closely resembling the Sinhalese. Residence also differentiates the inconsistency somewhat: urban residents, predictably, are less inconsistent than rural or estate residents; and residents of Zones 1 and 2 are less inconsistent than residents of the other zones, particularly of Zone 4. As Table 5.7 shows, the overall patterns of variation of inconsistency persist when duration of marriage is controlled for. In particular, the better-educated are more successful in preventing unwanted births. Among those who want more births, however, the better-educated are relatively more inconsistent, indicating that they less frequently have the births they want. Clearly, unwanted births as well as failure to have births which are wanted occur with some frequency in all strata of Sri Lankan population. In all socio-economic strata, the incidence of unwanted births predominates over the prevalence of failure to have wanted births. We do not have any data on what explanations the people themselves have for these phenomena. (Part of the explanation for unwanted births lies in the inefficacious use or nonuse of contraception, as part of the explanation for not having the births which are wanted lies perhaps in the prevalence of subfecundity or the effective intervention of the family planning program.) Whatever the .explanation, the high incidence of inconsistency between expressed intentions and subsequent fertility makes the former a poor predictor of the latter. The next section examines how poor a predictor the expressed - 156 - intention is of subsequent fertility. Table 5.7: Individual Inconsistency (Percent Inconsistent) by Selected Characteristics According to Whether the Wife Wanted More Children and Duration of Marriage (* Indicates Very Small Base). Those Who Want More Those Who Want No More 0-4 5-9 10+ Total 0-4 5-9 10+ Total Number of live births 0-1 14.4 47.1 86.3 29.0 92.6 * 4,5 67.4 2 17.2 23.0 65.7 29.9 60.6 64.0 21.0 48.6 3+ * 17.9 43.8 34.7 * 62.0 29.6 34e8 Age of Wife Under 25 11.3 27.4 * 15.5 78.2 75.6 * 75.0 25-29 17.1 19.6 39.3 23.0 81.5 66.7 56.3 62.8 30-34 29.4 50.0 56.5 47.8 * 50.0 37.4 40e6 35+ * * 77.6 71e4 * 28.6 17.9 18.4 Wife's education None 14e8 11.1 50.0 30.0 * 64.1 35.9 40e0 1-5 12.6 32.2 56,8 31.6 83.9 66.1 30.2 38.1 6-9 13.2 28.2 67.3 29.1 74.4 65.1 23.7 37.9 10+ 21e6 43.9 57.1 31.2 81.0 51.7 14.6 34.9 A Simple Recursive Model of Fertilitv in the Follow-Up Period We start with a simple model of fertility in the interim (between 1975 and 1979). This model is the same as the one proposed by Hermalin et al. (1979) for Taiwan for the period 1967-1974 (also see Westoff and Ryder, 1977). In the model, fertility over the period 1975-1979 is viewed as determined by: 1. Desire for additional children, expressed at the first interview; 2. Contraceptive use status as of that time (defined as a dichotomy: current user or not; and - 157 - 3. Certain social and demographic background variables: - number of children ever born before the first interview, - number of sons living then, - marriage duration as of the first interview (excluding periods outside married state), and - wife's education. It is assumed that contraceptive use status is influenced by desire for additional children and not vice versa. Thus, if a woman decided at an earlier date that she did not want any more children, she may have then proceeded to implement that decision through measures such as sterilization. The other causal sequences postulated in Figure 5.1 are more or less straightfoward: parity (as of 1975) is determined by marriage duration and wife's education; the number of sons living is determined solely by parity (1975); and the desire for additional children is determined by marriage duration, education, parity (1975), and the number of living sons (1975). As is well known, in the absence of reciprocal causation and under the assumption that the residuals are uncorrelated with each other or with the exogenous variables, the different equations in the sy3tem shown in Figure 5.1 can be estimated one by one using ordinary least squares. The path coefficients shown in Figure 5.1 are the standardized regression coefficients thus obtained. The analysis shows that marriage duration has the strongest direct effect (-.62) on fertility in the follow-up period; the effects of parity and contraceptive use status as of 1975 (.23 and -.25, respectively) are considerably smaller than that of marriage duration; the effect of education is modest (-.07), as is that of desire for additional children expressed in 1975 (.05); and the number of living sons shows no significant effect at all. - 158 - DURATION OF MARRIAGE, 1975 WIFE'S EDUCATION, 1975 0.73 -0.05 -0.24 -0.04 PARITY, 0.76 NUMBER OF -.09 1975 SONS LIVING ____1975 IN~TENTIONS .3100 -0.32 1975 0.23 1-0.07 0.12.0 -0.62 -0.21 0.05 CONTRACEPTIVE USE STATUS, 1975 0.23 -0.25 FERTILITY IN THE FOLLOW-UP PERIOD, 1975-1979 Fig. 5.1- Path Diagram of Factors Affecting Fertility in the Interim (1975-1979). -4 - 159 - These results closely correspond to those for Taiwan and the Republic of Korea (see Table 5.8), the major exception being that in Taiwan and Korea the expressed desire to have additional children has a stronger effect on fertility in the interim than is the case in Sri Lanka. The wife's education and her expressed desire for additional children are more important as determinants of contraceptive use status than are parity and the number of sons living. In Taiwan, parity has a negative effect, while the number of sons living has a positive effect. In Sri Lanka, both, these variables have positive effects: the more children a woman h 3--irrespective of their sex composition--the greater the likelihood that she is a contraceptive user. Among the determinants of expressed desire for additional children, parity and marriage duration are the strongest ones (with path coefficients -.32 and -.24, respectively), wife's education and the number of living sons having only modest effects (-.09 and -.04). In general, all the path coefficients have the expected signs: - The longer the marriage duration or the lower the wife's education, the higher the parity. - The higher the parity, the larger the number of sons living. The longer the marriage duration or the higher the parity, the lower the likelihood of desiring additional children. - The higher the wife's education, or the higher her parity, or the lower her desire for additional children, the higher the likelihood that she is a contraceptive user. Parity and expressed desire for additional children both have a positive effect on the likelihood of having a birth in the interim, while all the other variables considered have negative direct effects. Table 5.8: Standardized Regression Coefficients; Equations of a Recursive Model--A Comparative Study Involving Taiwan, Korea, and Sri Lanka Equation for Equation for Equation for Equation for Equation for Parity: Time No. of sons Additional children Contraceptive fertility One Living, wanted or not user or not in the interim Independent variables T K SL T K SL T K SL T K SL T K SL Duration of marriage .79 .57 .73 35 -.14 -.24 .16 .12a .49 .18 .62 Wife's education -.08 .03 -.05 -.08 -.12 -.04 .15 .02 .23 -.08 -.14 -.07 Parity: Time one .66 .13 .76 -.21 -.08 -.32 -.13 -.09 .12 .19 .14 .23 Number of sons living -.24 -.34 -.09 .09 .15 .06 -.02a -.20 .00a I Additional children wanted or not -.42 -.16 -.21 .36 .30 .05 Contraceptive user or not -.15 -.12 -.25 R .66 .32 .56 .44 .02 .57 .49 .08 .33 .26 .08 .13 .52 .28 .33 aNot significant. - 161 - Ethnic Group Differences In many respects, the Sinhalese, Sri Lankan Tamils, Indian Tamils, and Sri Lankan Moors are four different subpopulations in the country. Therefore, it is important that we examine whether these ethnic groups differ among themselves with respect to the patterns just described. In Table 5.9, the standardized regression coefficients for the recursive model described above (fitted separately for the four subpopulations) are shown along with the corresponding figures for the sample as a whole. That the pattern for the overail sample is virtually identical to the pattern for the Sinhalese--judging from the magnitudes (and signs) of the standardized regression coefficients--is not surprising, since the Sinhalese constitute almost 60 percent of the sample. The major differences between the four subpopulations with respect to the recursive model under consideration are: 1. Wife's education is a significant factor affecting parity at time 1 (1975) among the Sinhalese, but no so among any of the other groups. 2. As far as the predictors of the additional number of children wanted are concerned, only duration of marriage and parity are significant for the Timil groups and the Sri Lankan Moors, while wife's education and the.number of living sons are also significant for the Sinhalese. 3. With respect to contraceptive use status (as of time 1), wife's education and the number of additional children wanted are the only significant predictors for the Indian Tamils and the Sri Lankan Moors, while parity (as of time 1) is also significant for the Sri Lankan Tamils, and all factors except duration of marriage are significant for the Sinhalese. - 162 - Table 5.9: Standardized Regression Coefficients; Eqqations of a Recursive Model--The Sample as a Whole and Different Ethnic Groups. Independent Variables Addi- Contra- Duration Parity Number tional ceptive Equation of Wife's as of of Sons Children use 2 for Group Marriage Education Time (1) Living Wanted Status R All* 0.73 -.05 .56 Parity Sinhalese 0.74 -.08 .60 as of S.L. Tamil 0.73 -.03a . 55 Time (1) Indian Tamil 0.65 -.oia .42 S.L. Moor 0.73 -.02a .52 All* .76 .57 Number Sinhalese .76 .58 of Sons S. L. Tat;nil .76 .58 Living Indian Tamil .77 .59 S.L. Moor .74 .54 No. of All* -.24 -.04 -.32 -.09 .33 Additional Sinhalese -.25 -.07 -.28 -.13 .32 Children S.L. Tamil -.18 -.02a -.42 -.04a .36 Wanted (as Indian Tamil -.36 -. 01a -.30 -. 07a 43 a oftime 1) S.L. Moor -.17 -.06 -.40 _.01a .30 ,a Cota All* .01a .23 .12 .06 -.21 .13 Coptiae Sinhalese -.08 .27 .14 a13 -:20 :16 ustie aa a use Indian Tamil 094a .08a -.26 e11 Status S.L. Moor .03a .23 .07 -.01 -.21 o10 a All* -.62 -.07 .23 .00 05 -.25 .33 Fertility Sinhalese -.60 -.08 .19 -.01a .07 -.23 .34 in the S.L. Tamil -.68 -.08 .29 -.03a .03 -.20 .32 Interim Indian Tamil -.55 -.04 .01a .13a -.03a .11 .25 S.L. Moor -.66 _.00a .31 .04a .01a -.31 .34 *See Table 5.8. Not statistically significant. - 163 - Fizure 5.2- P (a) Sinhalese (b) Sri Lankan Tamil P -S p > S E . U F F (c) Indian Tamil (d) Sri Lankan Moor Figure 5.2: Recursive Model Treating Duration of Marriage (D) and Wife's Education (E) as Exogenous Variables, and Parity as of Time 1 (P), Number of Sons Living as of Time 1 (5), Number of Additional Children Wanted as of Time 1 (I), Contraceptive Use Status as of Time l (U), and Number of Live Births in the Follow-Up Period (F) as Endogenous Variables. - 164 - 4. Finally, the significant determinants of fertility in the interim are: duration of marriage and contraceptive use status for the Indian Tamils; these and parity (as of time 1) for the Sri Lankan Moors; these and wife's education for the Sri Lankan Tamils; and these and the number of additional children wanted for the Sinhalese. Thus, the Sinhalese stand apart from the Sri Lankan Tamils, and both these groups stand apart from the Indian Tamils and the Sri Lankan Moors. The last two groups resemble each other, except for the determinants of the interim fertility. A Modified. Extended Model We now consider a model which is a modification and an extension of the one presented in Figure 5.1 and Table 5.8. The modification involves: (1) ignoring the number of living sons as a relevant factor; (2) replacing parity (1975) by the number of children living (1975); and (3) treating the substitute of parity just mentioned as an exogenous variable. The results presented in Figure 5.1 and Table 5.8 suggest removing from consideration as a relevant factor the number of sons living. The second modification, replacing parity with the number of children living, is suggested by the consideration that a woman's fertility intentions and her subsequent behavior are influenced by the actual family size attained rather than by the number of live births she has had. The third modification is introduced simply for analytic convenience (to reduce the number of equations to be estimated) and has no impact on the results presented below, since we shall be using a recursive model. In the present modelf we shall introduce a number of exogenous variables not considered in the simple model of Figure 5.1. Moreover, this new model goes beyond fertility in the interim to the fertility preference - 165 - and intentions and contraceptive'use status as of the end of the follow-up period. This model postulates the following causal sequence (each variable being an antecedent to each of those appearing later in the list). 1. Fertility preference (time 1) 2. Demand for additional children (time 1) 3. Contraceptive use status (time 1) 4. Fertility behavior in the interim 5. Fertility preference (time 2) 6. Demand for additional children (time 2) 7. Contraceptive use status (time 2) As a measure of the fertility preference, we use the response to the interview question, "If you could choose exactly the number of children to have in your whole life, how many would that be?" The demand for additional children is measured by the response to the question, "How many more children do you want to have?" Contraceptive use status is measured by a dichotomous variable indicating whether the respondent is a contraceptive user or not (sterilized women and wives of sterilized husbands being regarded as users with no desire for additional children). Fertility behavior in the follow-up period is measured by the number of live births in the period. As for the exogenous variables, a few words are in order regarding their linkages to the fertility preference function. Exogenous Variables: Current Family Size. It is postulated that a very important - 166 - determinant of a woman's family size preference is her current family size (= the number of children living). This postulate is based on the notion that women tend to revise their family size preference upward to match the actual family size attained. No doubt there are women whose reproduction behavior is a consequence of, rather than an antecedent to, their fertility preference; but in countries like Sri Lanka, rationalization is probably stronger than implementation. Age Related Factors. Age-related factors are also important determinants of fertility preference. Three such factors may be distinguished: (1) age at marriage, (2) the interval between marriage and the last birth, and (3) the open interval or the interval since last birth. It is quite possible that those who marry late are self-selected for small-family mindedness, for commitment to extra-familial involvement, for birth control proneness, and so on, the opposite being true of those who marry at younger ages. Also, in populations such as Sri Lanka, where pre=marital pregnancies are rare and the loss of exposure (to the risk of becoming pregnant) due to marital dissolution insignificant (see, e.g., Caldwell, et al., 1980), age at marriage mostly determines a woman's total reproductive period. Assuming that exposure to pregnancy risk starts at marriage, the second factor mentioned above, namely the interval between marriage and last birth, represents the length of the woman's reproductive period used up in producing the number of children already born. The third factor, the open interval, represents the age of the youngest child, a factor which is of considerable significance to women in connection with their planning of inter-birth intervals. In the absence of contraceptive use, the open interval can also be treated as a measure of subfecundity, - 167 - which is knowu to thwart a couple's plan-implementationuefforts. It may be noted that the sum of the three factors just mentioned is equal to the woman's current age and that the sum of the last two factors is equal to the woman's duration of marriage. Those who have been married long or are of older age belong to older birth cohorts and are also likely to be at advanced stages of their life course. Older birth cohorts may differ from younger ones in family-size orientation. Similarly, a woman's propensity to cease childbearing is influenced by the life-course stage she is in. Thus, if she is a grandmother she may not wish to have any more births, in deference to the norm proscribing grandmothers from having babies of their own. Background Variables. It is logical to include among the determinants of fertility preference background variables such as type of place of residence, region of residence, and religion, and socio-economic status indicators such as wife's education, husband's occupation, housing conveniences such as toilet facilities, water supply and light source, and ownership of modern articles such as radio and appliances. (We treat religion as a dichotomy for the moment [Muslim vs. others], as we do region of residence (Southwest, including Colombo vs. Other) and type of place of residence (Urban vs. Other). It is recognized that dichotomies do not always capture the full flavor of polytomies; but it is believed that the basic picture will not be drastically altered by the use of dichotomies.) Ideal Number of Children. It is assumed that the number of children a woman considers ideal for a girl from a family like that of self is a logical exogenous variable. This variable may be thought of as a rough measure of the social norm regarding family size. (Note that this measure - 168 - is unaffected by factors such as rationalization.) Unfortunately, this measure is available only for the end of the follow-up period. But we are treating it as exogenous, assuming that it is a lagged reading of the 1975 norm. Wife's Work Pattern. Lastly, wife's work pattern as determined from the 1975 survey is a logical exogenous variable. Involvement in non-child-centered activities is a possible determinant of a woman's fertility preference. The standardized regression coefficients in the model fitted for the overall sample are presented in Table 5.10. The highlights of the Sinhalese, Sri Lankan Tamil, and Sri Lankan Moor models are indicated in the text as the individual equations are discussed. The Indian Tamil model is described in the Appendix. While religion is included in the overall model, it is excluded from all the other models in view of its overlap with ethnic group membership. The number of children living (as of time 1) is the most important determinant of this variable. Its path coefficient in the overall model is 0.71. The corresponding ethnic-group specific figures are 0.70, 0.71, and 0,69, respectively, for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors. Thusi the larger the current family size, the higher the preferred number of children per family. This probably reflects, as suggested earlier, a tendency to rationalize (i.e., to make one's preference consistent with what one actually has). Table 5.10: A RecursiveModel- for Fertility Dynamics--Standardized Regression Coefficients in Various Regression lquations and R2 for Each Regression. Equation for 14 15 16 17 18 19 20 1. No. of children living: time 1 .71* -.77* .48* ,c08* .60* -.36* .23* 2. Age at first marriage .00 .02 -.01 -.08* o04* -.01 -.03 3. Interval: Marriage to last birth: time 1 -.01 -.22* -.22* -.35* -.02 -.13* -.16* 4. Interval: Since last birth: time 1 -.05* -.02 .13* --o35* .01 -.06* .09* 5. Wife's education -.07* -.03 .08* .04 -.05* -.03 .07* 6. Husband's occupation .01 -.03 .02 -.03 .o0 -.01 -.00 7. No. of modern articles owned .01 .00 .09* -.05* -.02 .06* .05 8. Housing convenience .03 o04 .03 -.01 .02 -.02 .01 9. Type of place of residence (urban) -.05* .01 .00 -.02 -.02 .00 -.01 H C,' 10. Region of residence (Zones 1, 2) -.03 -.08* .08* -.04 .00 -.04 .08* 11. Religion (Muslim) .05* .05* -.11* .02 -.01 .00 -.05* 12. Ideal family size for a girl .11* .04* -.01 .05* .23* .08* .03 13. Work pattern: Wlife -.01 .01 .01 .00 -.00 .01 -.02 14. Fertility preference: tinie 1 ,55* -.09* .14* .14* -.03 .05 15. Demand for additional children: time 1 -.16* -.02 .05* .41* -.08* 16. Contraceptive use status: time 1 -.22* -.01 -.07* .32* 17. Fertility in the interim .18* -.26* -.01 18. Fertility preference: time 2 .28* -.08* 19. Demand for additional children: time 2 -.12* R. G561 .476 .226 .344 .547 .439 .270 * Statistically significant - 170 Next in importance to the current family size is the number of children considered ideal for a girl from a family like that of the respondent. The path coefficient of this variable in the overall model is 0.11, the corresponding ethnic group-specific figures being 0.14 for the Sinhalese and 0.08 for the Sri Lankan Moors. (This factor does not have a significant effect in the Sri Lankan Tamil model.) There is some difficulty in interpreting the relationship of this variable to fertility preference, because the observed relationship between the two variables may be due to a common antecedent. This could be the case, if, for example, the two variables were indicators of a common unmeasured variable, family size norm. Thus, one could very well specify the relationship involved as: Propensity to rationalize - --- - Fertility preference Current family size Family size norm i Ideal family size for a girl from a family like that of the respondent In this specification, the propensity to rationalize and family size norm are both unmeasured variables, the latter regarded as a common antecedent of fertility preference and the number of children considered ideal for a girl from a family like that of the respondent. In the present analysis, however, a direct causal relationship between the last variable and fertility preference is postulated. This oversimplification and the resulting possible misspecification should be kept in mind when interpreting the patterns presented here. Apart from the two factors mentioned above (current family size and - 171 - ideal family size for a girl from a family like that of the respondent), the following factors (shown with path coefficients) significantly affect the fertility preference at time 1: - wife's education (-0.07); - interval since last birth (-0.05); - type of place of residence (-0.05); - religion (-0.05). As already mentioned, religion was excluded from the ethnic group-specific models, because of its overlap with ethnic group membership. Among the other variables just mentioned, only wife's education and type of place of residence have significant path coefficients in the Sinhalese model (-0.05 and -0.05). The latter is the only variable with a significant path coefficient in the Sri Lankan Tamil model (-0.06), while the open interval is the only variable significant in the Sri Lankan Moor model, with path coefficients -0.10. The significant path coefficients have the expected signs. Thus, wife's education has a depressing effect on fertility preference; and those who reside in urban areas prefer smaller families than do their counterparts residing elsewhere. The regressors together account for 56 percent of the variance in fertility preference in the overall sample, the corresponding figures for the different ethnic groups being 58 percent for the Sinhalese, 55 percent for the Sri Lankan Tamils, and 61 percent for the Sri Lankan Moors. Demand for Additional Children at Time 1: The number of children living as of time 1 is the most important predictor of this variable also, the path coefficient being -0.77 in the overall model and -0.76, -0.86, and -0.71, respectively, in the Sinhalese, 172 - Sri Lankan Tamil, and Sri Lankan Moor models. The negative sign of this path coefficient is what one would expect: the larger the current family size, the fewer the additions one desires (wants). The current fertility preference, with a path coefficient in the overall model of 0.55, is the next important determinant of the number of additional children wanted. The corresponding ethnic group-specific figures are 0.54 for the Sinhalese, 0.50 for the Sri Lankan Tamils, and 0.68 for the Sri Lankan Moors. Other determinants with significant path coefficients in the overall model are: interval between marriage and last live birth as of time 1 (-0.22); region of residence, i.e., Colombo and Other Southwest vs. Other (-0.08); religion (0.05); and ideal family size for a girl from a family like that of the respondent (0.04). For the interval between marriage and last live birth, the Sinhalese, the Sri Lankan Tamil, and the Sri Lankan Moor models have path coefficients of -0.26, -0.15, and -0.25, respectively. Region of residence (Colombo and Other Southwest vs. the remainder) is significant only for the Sri Lankan Moors (-0.14), and ideal family size for a girl from a family like that of the respondent is significant only for the Sinhalese (0.08). The R2 for the overall sample is 0.48, the corresponding ethnic group-specific figures being 0050, 0.51, and 0.39, respectively, for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors. Contraceptive Use Status as of Time 1: The number of children living (as of time 1) is the most important determinant of this variable also, its path coefficient being 0.48 in the overall model. The corresponding figures for the ethnic group-specific - 173 - models are 0.45, 0.54, and 0.46 for the Sinhalese, Sri Lankan Tamils and Sri Lankan Moors, respectively. The positive sign of this path coefficient reflects the relatively higher tendency for people with large families to use contraception, particularly sterilization. Next in importance is the interval between first marriage and last live birth as of time 1. The path coefficient for this variable in the overall model is -0.22, while the corresponding figures in the ethnic group-specific models are -0.20, -0.28, and -0.29 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. In interpreting these coefficients, it should be noted that, for all practical purposes, the interval between marriage and last live birth as of time 1 represents the duration of marriage, and that in the WFS (1975) an inverse pattern was observed for the percent users of contraception by marital duration (see, e.g., WFS First Report, Table 5.2.1A). The lower use rate at longer marital durations is to be expected, given the fact that the national family planning program got underway only a few years before the 1975 survey, and that users, especially those who chose sterilization, were naturally recruited disproportionately from those who were not near the end of their reproductive period. The demand for additional children is another important determinant of contraceptive use status as of time 1. The path coefficient for this variable in the overall model is -0.16, while the corresponding figures in the ethnic group-specific models are -0.18, -0.11, -0.17, for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. The interval between last live birth and the date of the first interview (the open interval as of time 1) is also a significant determinant 174 - of the contraceptive use status. The path coefficient of this variable in the overall sample is 0.13, while the corresponding figures for the ethnic group-specific models are 0.15, 0.09, and 0.12 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. The positive sign of these figures reflects the fact that contraceptive use history (especially the adoption of sterilization) prior to time 1 is a common antecedent of and positively associated with both the open interval and the contraceptive use status of time 1. Fertility preference as of time 1 is another significant determinant oi the contraceptive use status as of time 1. The path coefficient of this variable in the overall sample is -0.09. This variable is not statistically significant in any of the ethnic group-specific models, although the magnitudes and the sign are consistent with the figure for the overall sample (-0.07, -0.09, -0.09 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively). Among the background variables, religion (Muslim vs. others) is significant in the overall sample, as are the number of modern articles owned and the region of residence (Southwest including.Colombo vs. Other). In the ethnic group-specific models, religion was not included for reasons mentioned earlier. The number of modern articles owned has a path coefficient of 0.08 in the Sinhalese model and 0.20 in the Sri Lankan Moor model, the corresponding figure in the overall sample,being 0.09. In the Sri Lankan Tamil model this variable is not significant. As for region of residence, only the Sri Lankan Tomilts and the Sri Lankan Moors show significant effects, the path coefficients--0.08 and 0.14, respectively--corresponding to 0.08 in the overall sample. - 175 - The R2 for the overall sample is 0.23, the correspondiLg ethnic group specific figures being 0.20, 0.22, and 0.23 for the Sinhalese, the Sri Lankan Tamils, and the Sri Lankan Moors, respectively. Fertility in the Interim: The two components of marriage duration (the interval between marriage and last birth and the open interval, both as of time 1) are by far the most important predictors of this variable. Listed below are the path coefficients of these predictors: Interval Between Marriage and Last Birth Open Interval Overall sample -0.35 -0.35 Sinhalese -0.32 -0.33 Sri Lankan Tamils -0.40 -0.38 Sri Lankan Moors -0.39 -0.36 Note: The negative sign reflects the effect of subfecundity. Next in importance is contraceptive use status as of time 1. In the overall model, the path coefficient of this variable is -0.22, while the corresponding figures in the ethnic group-specific models are -0.22, -0.18, and -0.27 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. Fertility preference as of time 1 is also an important predictor of the number of children born in the follow-up period, the path coefficients of this variable being: - Overall sample 0.14; - Sinhalese 0.15; - Sri Lankan Tamils 0.16; - Sri Lankan Moors 0.13 (not significant). The number of children living as of time 1 is also a significant - 176 - predictor of fertility in the interim. The path coefficients are -0.08 and -0.13 for the overall model and the Sinhalese, respectively. In the other models, the corresponding figures are not statistically significant. Age at marriage is significantly related to fertility in the interim, each path coefficient equalling -0.08 in the overall and the Sinhalese models. The corresponding coefficients in the models for the other ethnic groups are not statistically significant. In the ozverall sample, the number of children considered ideal for a girl from a family like that of the respondent has a significant path coefficient (0.05); of the ethnic group-specific models, only that of the Sri.Lankan Moors shows significant effect for this variable (0.11). Among the background variables, the number of modern articles owned has a significant path coefficient in the overall sample (-0.05). The corresponding figures in the ethnic group-specific models are not , significant, except in the Sri Lankan Tamil model in which the variable has a path coefficient -0.13. Wife's work pattern, with a path coefficient of 0.08, significantly affects fertility in the interim in the case of the Sri Lankan Tamils, indicating that those who worked had more children in the follow-up period. This relationship is difficult to interpret in view of the fact that neither the type of work (except whether it is at home or away from home and before and/or after marriage) nor duration of work is taken into account in the definition of the variable. A word is in order about the effect of the demand for additional children at time 1 on fertility in the follow-up period. This effect is not statistically significant in any of the models except the one for the Sri 177 - Lankan Tamils, in whose case the path coefficient is -0.11. The negative sign of this coefficient indicates that: actualization of expressed intention has fallen behind. To a certain degreej this casts some doubt on the validity of the respondent's expressed intentions. Fertilitly Preference at Time 2: The number of childreu living at time 1, fertility preference at time 1, and fertility in the inzerim are by far the most important determinants of fertility preference at time 2. The respective path coefficients of these predictors in the overall sample! and the ethnic groups are as shown below (superscript a means 'not significant'). Sri Lanka Sri Lanka All Sinhalese Tamils Moors Number of children living, time 1 0.60 0.53 0.49 0.68 Fertility preference, time 1 0.23 0.31 oo9a -0.1oa Fertility in the interim 0.18 0.22 0.17 0.14 The number of children considered ideal for a girl from a family like that of the respondent is also a strong predictor of fertility preference at time 2. The path coefficient of this predictor in the overall model is 0.14, while the corresponding figures in the ethnic group-specific models are'0.17, 0.25, and 0.32, for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. The strength of this; relationship probably is partly an artifact caused by the close proximity in the questionnaire of questions asking for fertility preference at time 2 and the ideal family size for a girl from a family like that of the responadent. ° 178 - Another significant predictor is wifels education, although this variable has significant path coefficients only in the overall sample (-0.05) and one ethnic group, the Sinhalese (-0.05). With a path coefficient of 0.05 in the overall sample, demand for additional children as of time 1 is another predictor; however, the factor is not significant in the ethnic group-specific models. Similarly, age at marriage has a significant path coefficient in the overall sample, but not in any of the ethnic group-specific models. The R2 for the overall sample is 0.55, the corresponding ethnic group-specific figures being 0.68, 0.42, and 0.50 for the Sinhalese, Sri Lankan Tamil, and Sri Lankan Moor models, respectively. Demand for Additional Children at Time 2: By far the most important predictors of this variable are its own lagged value at time 1, the number of children living at time 1, fertility preference at time 2, and fertility in the interim. The respective path coefficients in the various models are shown below: Sri Lankan Sri Lankan All Sinhalese Tamil Moors Demand for additional children, time 1 0.41 0.54 0.35 0.24 Number of children living, time 1 -0.36 -0.39 -0.33 -0.34 Fertility preference, time 2 0.28 0.39 0.16 0.26 Fertility in the interim -0.26 -0.30 -0.26 -0.20 The interval between first marriage and last birth, as of time 1, - 179 - is also a significant factor affecting the demand for additional children as of time 2. The path coefficients are -0.13, -0.19, -0.26, respectively, in the overall sample, the Sri Lankan Tamil, and Sri Lankan Moor models. The Sinhalese model shows no significant effect for this factor. The number of children considered ideal for a girl from a family like that of the respondent is another relatively strong predictor of the number of additional clildren wanted at time 2. The path coefficients for this factor are 0.08 and 0.19 in the overall and the Sri Lankan Moor models, while the corresponding figures for the other models are not statistically significant. Contraceptive use status is also a relatively strong predictor of the number o;uf additional children wanted at time 2. The path coefficients are -0.07 and -0.08, for the overall and the Sinhalese models. This factor does not have a significant impact on the variable in question in the models for the other ethnic groups. The open interval at time 1 is a significant predictor of the demand for additional children in the overall sample, in the Sinhalese, and in the Sri Lankan Moor models, their respective path coefficients being -0.07, -0.06, and -0.10. Among the background variables, the number of modern articles owned shows a significant impact on the demand for additional children at time 2 for the overall sample and the Sri Lankan Moor model, their respective path coefficients being 0.06 and 0.12. The R2 for the overall sample is 0.27, the corresponding ethnic group-specific figures being 0.50, 0.44, and 0.41 in the Sinhalese, Sri Lankan Tamil, and Sri Lankan Moor models, respectiiTely. 180 - Contraceptive Use Status, Time 2: The most important predictor of this variable is its own lagged value (at time 1). The path coefficient of this predictor is 0.32 in the overall model, while the corresponding figures in the ethnic group-specific models are 0.27, 0.38, and 0.21, in the Sinhalese, Sri Lankan Tamils and Sri Lankan Moors, respectively Next in importance is the number of children living as of time 1, with a path coefficient of 0.23 in the overall model. The corresponding figures in the ethnic group-specific models are 0.18, 0.36, and 0.37 in the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. Duration of marriage is also a strong predictor of the contraceptive use status as of time 2. The path coefficients of the two components of this factor are as shown below (superscript a means "not significant"): Interval Between Open Marriage and Interval, Last Birth, Time 1 Time .' Sinhalese -0.09a -0.O05a Sri Lankan Tamils -0.27 -0.11 Sri Lankan Moors -0.20 0.08a All -0.16 -0.09 The negative sign of the path coefficient of the interval between marriage and last live birth as of time 1 partly reflects a subfecundity and partly a cohort effect. Longer-married women are less likely to be users partly because they believe that they are infecund and partly because they are traditionally oriented toward non-contraception. The negative sign of the path coefficients under Open Interial,'Time 1 signifies that users of - 181 - contraception as of time 2 were not disproportionately recruited from those with relatively long open intervals as of time 1. (Those who got sterilized prior to time 1, 1975, did so in most cases within one or two years of time 1, following the popularization of this method in 1973.) The number of additional children wanted as of time 2 is another significant predictor of the contraceptive use status as of time 2, its path coefficient in the overall sample being -0.12. The corresponding figures in the ethnic group-specific models are -0.14, -0.03, and -0.10 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. The figure for the Sri Lankan Tamils is not statistically significant. The lagged value of the variable just mentioned, i.e., the number of additional children wanted as of time 1, is also a significant factor affecting the contraceptive use status as of time 2. The path coefficient in the overall model is -0.08, while the corresponding ethnic group-specific figures are -0.10, -0.09, and -0.05 in the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. Fertility preference as of time 2 is anucher factor significantly affecting the contraceptive use status as of time 2. Its path coefficient is -0.08 in the overall model, whi;e the corresponding ethnic group-specific figures are -0.14, -0.13, and -0.23; in the Sinhalese, Sri Lankan Tamils, and the Sri Lankan Moors, respectively. Among the background variables, region of residence (Southwest, including Colombo vs. Other) and wife's education significantly affect the contraceptive use status as of time 2. The path coefficients for the former are 0.08 in the overall model and 0.09 in the Sri Lankan Tamil model. The corresponding figures are not statistically significant in the other models. - 182 - As for wife's educati2n; its path coefficient in the overall model is 0.07, the corresponding figures in the ethnic group-specific models being 0.06, 0.13, and 0.16 for the Sinhalese, Sri Lankan Tamils, and Sri Lankan Moors, respectively. Wife's work pattern affects the contraceptive use status as of time 2 in the case of the Sinhalese and Sri Lankan Tamils, with path coefficients of -0.07 and 0.07, respectively. In the case of the Sinhalese, the propensity to be a user of contraception is higher among never worked, while the opposite is true among the Sri Lankan Tamils. With the limited amount of information available for the work variable, it is not possible to interpret this ethnic-group difference in the associai ion in question. J4 - 183 - LEGEND, FIGURE 5.3 X1 Number of children living as of time 1 X2 Age at first marriage X3 Interval from marriage to last birth, as of time 1 X4 Open interval, as of time 1 X5 Wife's education X6 Husband's occupation (1 upper class, 0 middle class, -1 lower class) X7 Number of modern articles owned X8 Housing convenience X9 Type of place of residence (urban vs. other) XIORegion of residence (Colombo and southwest vs. other) XIlReligion (Muslim vs. other) X12Ideal number of children for a girl from a family like that of the respondent Xl3Wife's work pattern (1 never worked; 2 worked before only; 3 worked after, away from home; 4 worked after) Pl Fertility preference as of time 1 D1 Demand for additional children as of time 1 C1 Coutraceptive use status as of time 1 F Fertility in the interim P2 Fertility preference as of time 2 D2 Demand for additional children as of time 2 C2 Contraceptive use status as of time 2 184 - Lisgure 5.3a: Overall-Sample Model D p c1 P1 L:C F C22 Exogenous variables significantly affecting the endogenous variables shown in the figure: P1: X1,X4,X5,X9,Xll,Xl2 DI: Xl1X31Xl0,Xll,X12 Cl: X1,X3,X4,X5,X7,X 10 F: Xl,X2,X3,X4,X7,X12 P2: X1,X2,X5,X12 D2: X1,X3,X41X7X12 C2: X1lX3,X4,X5,X10X11 185 a Figure 5.3b: Sinhalese D - _ _ _ _ 1~ C p12 Exogenous variables significantly affecting the endogenous variables shown in the figure: Pl: X1,X5,X9,X12 D1: Xl,X3,X5,X8,Xj2 C1: X1,X3,X4,X5,X7 F: Xl,X2,X3,X4,X6 P2: X1,X5,XI0,X12 D2:, X1 X4,XIO C2: 1Xl,4,X5,X7,X13 - 186 - Figure 5.3c: Sri Lankan Tamils P1 \ / F C2 Exogenous variables significantly affecting the endogenous variables shown in the figure: P1: . 1,X9 .DI: X1 ,X3 Cl: X1,X3 X41X5,Xl0 F: X2,X3,X4,X7,Xl3 P2: X19XI2 D2: X1,X3 C2: Xl,X3X41X5,X6,XI,X,X13 - 187 - Figure 5.3d: Sri Lankan Moors D C p11 F f 2 Exogenous variables significantly affecting the endogenous variables shown in the figure: P1: XI,X4,X12 D1: X1,X3-XI0 C1: X1,X3,X4,X7,X10 F: X3,X4,X9,Xl2 P2: X1,X12 D1: Xl,X3,X4,X7,X8,Xj2 C2: XljX2,X3,X5 - 188 - The significant paths described above are shown in Figures 5.3. Clearly the figure for the Sinhalese and that for the overall sample resemble each other to a large degree, while those for the Sri Lankan Tamils and Sri Lankan Moors stand apart in many respects from each other and are quite different from the others. Relatively more of the hypothesized relationships are statistically significant in the case of the Sinhalese than is true for the other ethnic groups. This reflects, in part, the corresponding differences in sample sizes (n is over 1000 in the Sinhalese, about 500 in the Sri Lankan Tamils, and slightly less than half that in the Sri Lankan Moors). Nonetheless, some of the differences and similarities may be real, notably: 1. The number of children living as of time 1 is a strong factor affecting all dimensions of subsequent fertility behavior (contraception, fertility preference, demand for additional children, and actual fertility). 2. The number of additional children wanted as of time 1 (D1 in Figure 5.3) does not significantly affect fertility in the interim, except in the Sri Lankan Tamils. 3. Fertility preference at time 1 significantly affects fertility in t1.e follow-up period, except in the Sri Lankan Moors. 4. Contraceptive use status as of time 1 significantly affects fertility in the interim in all three ethnic groups. 5. The! effect of wife's work pattern is not consistent across the different ethnic groups. 6. By and large, status indicators such as housing convenience do not. seem to be significant. 7. Wife's education significantly affects the various dimensions of reproduction examined, although there are clear differences between the ethnic groups in this respect. Table 5.11: Proportion of Variance Accounted for by Regressors When Entered in the Sequence Shown Dependent Variable Y1 Y2 Y3 Y4 Y5 Y6 Y7 Demographic Controls No. of children living: time 1 .547 3 .298 .080 3 .0883 .4223 .2053 .06 3 Age at first marriage .001 .000 .019 3 .0043 .000 .000 .015 3 Interval: Marriage to last birth .000 .01 3 .0143 .0153 .000 .0043 .0lO3 Interval: Since last birth .003 .0023 .026 .168 .007 .000 .000 Background Variables Religion .004 .0133 .0383 .0l13 .0063 .0043 .0423 Ethnic Group .0023 .0053 .002 .002 .001 .0021 .003 2 Type of place of residence .0053 .005 3 .0073 .0073 .0073 .0053 .0093 Region of residence .001 .0073 .0113 .003 .009 3 .009 .3 0 Socio-Economic Status 3 1 3 3 3 23 Literacy: Wife .002 .001 .016 .003 .007 .002 .0123 Wife's education .001 .002 .007 .000 .002 3 .002 2 .009 3 Number of modern items owned .000 .001 .006 3 .004 .001 2 .003 3 .004 3 H 1 2 c Housing convenience .001 .001 .000 .000 .000 .000 .000 Social norm Ideal family size for a girl .0103 .0093 .001 .005 .0633 .0283 .000 Extra-familial activity 2 Work pattern: Wife .000 .002 e002 L000 L000 .002 .003 Endogenous Variables Yl: Fertility preference: Time 1 .136 .0l1l .012J .0193 .0203 .0022 Y2: Demand for additional children: Time 1 .010 .000 .0023 .0983 .0143 Y3: Contraceptive use status: Time 1 .0313 .0023 .001 -0723 Y4: Fertility in the interim .0243 .026 3 O.G Y5: Fertility preference: Time 2 0035 .o0093 Y6: Demand for additional children: Time 2 .0073 Y7: Contraceptive use status: Time 2 R2 .577 .492 .250 .350 .573 .446 .292 1 Probability between .05 and .1; 2 probability between .01 and .05; and. 3 probability less than .01 - 190 - Another Look at the Relationships In order to make the data amenable to path analysis, in the preceding section several of the class variables were reduced to dichotomies. We now take another look at the various regression relationships, retaining the polytomous character of the various factors, such as region, religion, and type of place of residence. Table 5.11 shows the contributions to R2 from the different variables when they are entered in the respective regressions in the order shown. Thus for the equation for fertility preference, the figure, .547, in the first row represents the proportion of variance in fertility preference accounted for by the current family size (number of children living); the next figure, .001, is the incremental R2 attributable to age at marriage, given the R2 of (the proportion of variance accounted for by) current family size; and so on. The sum of the incremental R2's attributable to the various regressors should equal (within rounding errors) the multiple R2 for the equation (shown at the bottom row of Table 5.11). Obviously, this type of analysis requires a sensible ordering of the variables. We have chosen the following ordering: Demographic controls Background variables Socio-economic status indicators Social norm (ideal family size for a girl from a family like that of the respondent) Extra-familial activity of the wife Endogenous variables Fertility preference: time 1 Demand for additional children: time 1 - 191 - Contraceptive status: time 1 Fertility in the interim Fertility preference: time 2 Demand for additional children: time 2 Contraceptive use status: time 2 The figures presented in Table 5.11 tell the following stol:ye Fertility Preference at Time 1: The number of children living (current family size) at time I accounts for almost 55 percent of the-variance. All the other variables together account for only an additional 3 percent of the variance. Given current family size, the increment in R2 attributable to age at first marriage is only .001; that attributable to the interval between first marriage and last live birth, given current family size and age at marriage, is almost nothing; and that attributable to the open interval, given the three demographic control variables just mentioned is only .003. The demographic control variables together account for 55.1 percent of the total variance of fertility preference at time 1. Given the demographic control variables, the background variables (religion, ethnic group, type of place of residence, i.e., urban/rural/estate, and region of residence) together account for an additional 1.2 percent of the variance of fertility preference at time 1. The incremental R2 attributable to the socio-economic status variables, given the demographic controls and the background variables, is only .005. Social norm concerning family size (ideal family size for a girl from a family like that of the respondent) accounts for another 1 percent of the total variance, and the wife's work participation adds nothing at all. - 192 - Demand for Additional Children at Time 1: The demographic control variables together account for 31.1 percent of the total variance of demand for additional children as of time 1. Age at marriage supplies no predictive power as far as this variable (demand for additional children: time 1) is concerned. Given the demographic controls, the background variables all show significant incremental R2ts, altogether accounting for an additional 3 percent of the total variance. Given the demographic controls and the background variables, the socio-economic status indicators add very little predictive power, their combined incremental R2 being only .005. Social norm (ideal family size for a girl from a family like that of the respondent) has an incremental R2 of .009, while the figure for wife's work pattern is .002. Given all the exogenous variables, fertility preference at time 1 accounts for an additional 13.6 percent of the variance of demand for additional children as of time 1. Altogether, the variables considered account for 49.2 percent of the total variance. Contraceptive Use Status as of Time 1: Almost all variables contribute significant incremental R2'8, the exceptions being ethnic group, housing convenience, social norm, and wife's work pattern. Fertility in the Interim: All demographic controls contribute significant incremental R29s, as do the background variables, with the exception of ethnic group and region of residence. Only wife's literacy contributes a significant incremental R2 among the socio-economic status variables. Social norm contributes significantly, but wife's work pattern does not. As in Table 5.10, additional number of children wanted at time 1 does not contribute a - 193 - significant incremental R2. Fertility Preference as of Time 2: The relationships between fertility preference as of time 2, on the one hand, and the demographic controls, the background variables, and the socio-economic status indicators on the other, are similar to the corresponding relationships to fertility preference at time 1. All the endogenous variables included in the regression contribute significant incremental R2's. The combined explanatory power of all regressors is 57.3 percent. Demand for Additional Children as of Time 2: The results are more or less similar to those for demand for additional children as of time 1 with respect to the incremental R2's of demographic controls, background variables, socio-economic status indicators, social norm, and wife's work pattern. All the endogenous variables included in the regression also contribute significant incremental R28s. Contraceptive Use Status as of Time 2: The results are similar to those for the corresponding variable at time 1, with all endogenous variables in the regression, except fertility in the interim, contributing significant R2es. In general, the relationships among the endogenous variables revealed in Table 5.11 are the same as the corresponding relationships revealed in Table 5.10. On the other hand, the relationships involving demographic controls, background variables, etc., are not the same in the two tables. This is to be expected, since Table 5.10 presents net regression (standardized) coefficients, whereas Table 5.11 presents. - 194 - incremental R2's, given the order of the regressors shown. It is significant that, irrespective of the way one looks at the data, socio-economic status indicators and wife's work pattern have little impact on fertility in the interim. Of the explained variance of fertility in the follow-up period (3"' 35 percent of the total variance), 78.6 percent (= .275 of .35) is accounted for by demographic controls, and an additional 6.6 percent (= .023 of .35) by background factors; only 2 percent (= .007 of .35) is accounted for by socio-economic indicators, while 12.3 percent is attributable to fertility preference and contraceptive use status. From a policy standpoint, among the determinants mentioned above, the following are of particular interest: - contraceptive use status; - fertility preference; - socio-economic status; and - age at marriage (one of the demographic controls). Contraceptive use status presumably can be manipulated by family planning action programs; fertility preference by communication programs; socio-economic status by improving opportunities for higher education, by providing better housing conditions, by encouraging consumption of modern articles, and so on; and age at marriage by legislation, or perhaps more effectively by changing the social and economic living conditions. Concluding Remarks Fertility behavior bears little correspondence to previously stated intentions: unwanted births are frequent, as is failure to have wanted births. Family size preferences and contraceptive use status as of time 1 - 195 - 9 are good indicators of the number of children born in the follow-up period; wife's education is also a good indicator in this respect; status indicators are not; nor is wife's work pattern. On the other hand, wife's education is of some help in predicting fertility in the interim. By far the strongest predictors of subsequent fertility are cumulative fertility to date and life-cycle stage in general. The lack of correspondence between stated intentions and subsequent behavior may be attributed to a combination of several factors. It may very well be that the data collection procedure is faulty., (The interview question, "How many more children do you want?" may not be a suitable device to obtain the required intention measure. Also, partly to blame is the fact that the behavior and intention measures correspond to different time intervals in most cases: the truncated follow-up period, in the former, and the entire reproductive period lying beyond the first interview date, in the latter.) In addition, there is the possibility that the family planning program effectively intervened, in some cases, to cause the actual fertility in the follow-up period to fall short of the previously declared intentions. Finally, some couples may have had a genuine change of mind during the follow-up period, causing some revision in their plans. With the available data, it is not possible to estimate the separate effects of these factors on the discrepancy between stated intentions and subsequent behavior. - 196 - CHAPTER VI STERILIZATION Although family planning has been promoted in Sri Lanka since the early 1950s by social workers and non-governmental agencies, it was only in 1965 that a national family planning program became an integral part of the government's activities. Many details of the early family planning movemernt in the country are available in Abhayaratne and Jayawardene (1968). The contraceptives first made available were foam tablets, condoms and diaphragms. The government's initial emphasis was on intrauterine devices (IUDs). A shift toward the pill (oral contraceptive) occurred later on and to sterilization still later. This chapter is concerned with (a) subclass variations in proportion sterilized, (b) subclass-specific probabilities of electing sterilization, estimated by the life table method, (c) changes in the popularity of sterilization over time, and (d) the effect of sterilization on fertility. The analysis presented below is based on the re-interviewed subsample. For our purposes, no distinction is made between sterilizations with and those without contraceptive intent. (The relevant information needed to make this distinction was not collected in the 1979 survey.) Also, no distinction is made between sterilized women and wives of sterilized husbands. There is no reason to believe that the patterns described below would be any different if these distinctions had been made. - 197 - Subclass Variations in 2roportion Sterilized First we will examine subclass variations in proportion sterilized, defining subclasses in terms of a wide range of demographic and social variables. Data will be presented separately for: - Ever-married women, and - Currently-married, fecund women who do not want any more children. Sterilized women are considered fecund at the time of sterilization. They are also treated as women who want no more children. Table 6.1 presents the proportions of women in different subclasses who are sterilized, The following patterns are discernible: 1. Among ever-married women, the proportion sterilized rises sharply with age, peaks, and then declines. Among currently-married, fecund women who do not want any more children, however, the proportion sterilized rises at first, but flattens out after about age 30. This suggests that of the inclusive group (ever-married women), it is the infecund, and the widowed or separated women who are responsible for the decline in the proportion sterilized at older ages. 2. The pattern shown by the proportion sterilized by duration of marriage is similar to the one described for age. 3. Age at marriage shows an inverted-U relationship with sterilization for the more inclusive group (ever-married women) as well as for the smaller sample (consisting of currently-married, fec-und women who want no more children). This pattern is partly due to the linkage between age at marriage and duration of marriage and the association between the latter and the propensity to want no more children. Because age at marriage has been rising in the country for some time now, those with longer marriage - 198 - duration have married at relatively younger ages, and, having reached the end of childbearing, have a relatively greater propensity to want no more children. Conversely, the older the age at marriage, the greater the likelihood that the duration of marriage will be shorter and that the propensity to want no more children will be lower. 4. The relationship between parity (number of children ever born) and proportion sterilized is similar to that involving duration of marriage, particularly among the smaller group (currently-married, fecund women who want no more children). There does not seem to be the tendency, as in some other countries (e.g., Panama--see Westoff, 1979), for the. proportion sterilized to decline as parity progresses beyond the middle level. 5. A relatively higher proportion of women in urban than in rural areas elect sterilization. 6. Among the religious groups, the Buddhists show above average propensity to elect sterilization, whereas the Hindus and the Muslims remain below average in this respect. 7. Among the ethnic groups, the proportion sterilized is above average for the Sinhalese and below average for the others. 8. Literacy is positively associated with the tendency to elect sterilization, but the proportion sterilized shows an inverted-U pattern with the level of educational attainment. 9. Those with recent, home-centered work experience seem to have a relatively higher tendency to elect sterilization. The proportion sterilized is not higher among those with post-marital, non-home-centered work experience than among those with no work experience. 10. The pattern of age at first birth resembles that of both age at 199 - marriage and current age; also the pattern of length of interval between marriage and last birth corresponds to that of duration of marriage. 11. Those with unwanted births (the excess-fertility group) have a higher propensity to elect sterilization compared to those with no unwanted births. Reaults from Logistic-Regression Analyses: Treating sterilization as a binary response variable (1 for sterilized ; 0 for not sterilized), a logistic regression was fitted separately for the four ethnic groups using age, parity, education, (age)2, (parity)2, residence, and excess fertility as regressors. The objective was to examine the net effect of the regressors on the tendency to elect sterilization. For this analysis, age was entered in completed years, parity with no grouping, education in terms of the numbers of years of schooling completed, residence as a dummy variable (1=urban; O=other), and excess fertility also as a dummy [1=(desired > actual); O=other]. The logistic model expresses the logit transform of the probability that a woman in a subclass elects sterilization as a linear function (linear in parameters) of the regressor values that define the subclass. If P is the probability that a woman elects sterilizatioa, the logit transform of P is y = In ( -) - 200 which implies and is implied by eY p5 1+ eY where e is the base of the natural logarithm. Since P/(1-P) denotes the odds on sterilization, another name for logit is log odds. Setting y=l for those sterilized and y=O for those not, the logistic model can be written as exp(0O+0jxI+. .kxk) Pr (y=1,given Xl,X2,...Xk) = 1 + exp(UO+klx1+...+kxk) or equivalently logit[Pr(y=I,given Xl, ...xk)] = %o+f1x1+.--.+ xk In this form, % represents the log odds on sterilization for a woman with a standard set of x values (xl=x2=...=xk=0), while exp(%j) represents the fraction by which the odds are increased (or decreased) for every unit change in xj. The results of the logistic regression fitted are shown in Table 6.2. For the Sinhalese and the Sri Lankan Tamils, all the regressors show significant relationship to the odds on electing sterilization. But for the Indian Tamils none of the regressors seem to be significantly related to the odds in question. As for the Sri Lankan Moors, parity and excess fertility do not seem to make any difference; while age, education and residence do. Results from Ordinary Least Squares Regression: The corresponding results from a multiple regression E(y) = 60 + 3lxl+..m+$kxk where y=l if sterilized and =0 otherwise, fitted by ordinary least squares - 20* - Table 6.1: Proportion Sterilized Among Wlomen in Various Subclasses. Currently-Married WES Full Samplea Ever-Married Fecund Women Who Ever-Married Women Women Want No More Children Proportion Proportion Proportion Sterilized Subclass Sterilized N Sterilized N as of WFS '75 N Total .18 2,989 .30 1,667 .09 6,810 Current age: 20-24 .05 139 .12 58 .02 912 25-24 .09 478 .18 245 .08 1,995 30-34 .24 633 .34 442 .15 1,221 35-39 .22 661 .31 457 .14 1,203 40-44 .18 552 .32 319 .11 968 45-49 .10 522 .37 145 .05 1,035 Duration of marriage (yrs.): <5 .02 94 .08 24 .01 1,280 5-9 .07 560 .15 269 .07 1,231 10-14 .22 576 .32 379 .16 51,118 15-19 .26 583 .36 427 .15 1,057 20-24 .20 534 .32 342 .14 893 25-29 .13 409 .30 171 .06 1,231 30+ .09 231 .37 54 AgD at first marriage: <15 .12 428 .21 245 .09 984 15-17 .20 820 .33 484 .12 1,777 18-19 .21 586- .36 340 .09 1,214 20-21 .16 428 .30 232 .08 977 22-24 .16 397 .28 217 .09 955 25-29 .11 261 .23 122 .08 709 30+ .09 67 .23 26 .03 194 - 202 - 'Table 6.1: -Proportion Sterilized Among Women in Various Subclasses (cont'd). Currently-Married, WFfS Full Samp:.e, Ever-MIarried Fecund Women Who Ever-Married Women Women W4ant No More Children Proportion Proportion Proportion Sterilized Subclass Sterilized N Sterilized N as of WFS '75 N Number of live births: 0-1 .01 271 .05 40 .002 1P572 2 .05 412 .13 145 .05 1,874 3 .14 486 .22 309 4 .18 462 .28 302 5 .27 378 .38 263 .14 1,474. 6 .24 301 .36 1962 7-8 .23 A411 .36 261 .16 1,891 9+ .22 268 .40 151 Type of place of residence^b Urban .22 729 .36 449 .08 4,781 Rural .15 2,017 .27 1,080 .13 1,236 Estate .17 243 .30 138 .10 793 Religion:c Buddhist .20 1,590 .36 856 .10 4,521 Hindu .13 776 .23 442 .08 1,295 Muslim .10 358 .17 213 .04 467 Christian .17 259 .30 150 .09 515 Ethnicity: Sinhalese .20 1,713 .36 935 Sri Lankan Tamil .13 688 .22 390 Indian Tamil .16 214 .29 120 Sri Lankan Moor .10 349 .17 209 - 203 Table 6.1: Proportion Sterilized Among Women in Various Subclasses (cont'd) a Currentlv-Married lT;S Full Sample Ever-Married Fecund Women Who Ever-Married Women Women Want No More Children Proportion Proportion Proportion Sterilized Subclass Sterilized N Sterilized N as of WFS '79 N Literacy: Literate .19 2,198 .32 1,282 .10 ',,953 Not literate .11 791 .23 385 .08 1,855 W,"ife's education: None .12 602 .24 301 .06 1,512 Grades 1-5 .16 1,199 .30 651 .09 2,686 Grades 6-9 .21 812 .34 508 .10 1,704 Higher .16 376 .29 207 .10 908 d Work pattern: Never worked .17 1,622 .29 930 .10 3,253 Before only .18 278 .30 161 .11 679 After-away .15 758 .28 392 .08 1,678 After-home .20 329 .37 183 .08 1,205 Age at first birth: <15 .14 357 .25 209 16-17 .21 439 .34 273 18-19 .19 570 .32 331 20-24 .19 1,043 .32 612 25-29 .12 377 .24 189 30+ .04 203 .15 53 - 204 - Table 6.1: Proportion Sterilized APmong Women in Various Subclasses (cont'd)r Currently-Married, %TES Full Sample Ever-Married Fecund Women Who Ever-Miarried Women Women Want No Mlore Children Proportion Proportion Proportion Sterilized Subclass Sterilized N Sterilized N as of WFS '79 N Age at last birth: <17 .01 115 e 12 18-19 .10 42 e 14 20-24 .10 437 .20 210 25-29 .20 814 .33 492 30-34 .21 865 .32 545 35+ .16 716 .29 394 Interval between marriage and last birth (yrs.): <5 .05 519 .18 142 5-9 .19 856 .30 525 10-14 .23 746 .34 500 15-19 .18 537 .29 329 20+ .15 331 .28 171 Excess fertility: Desired < living .25 548 .37 375 Desired = living .20 1,545 .29 1,078 Desired > living .05 884 .22 212 Unless otherwise stated the figures for the WFS full sample are from Immerwahr (1981, Table 12, pp. 24-25). bThe figures for the WEFS full sample are obtained by combining Immerwahr's figures (1981, Table 12, p. 25) for childhood places of residence. Also, Immerwahr's "other" category has been taken as equivalent to "Estate" herein. cThe figures for the IUFS full sample are from WFPS Sri Lanka, 1975, First Report, p. 543. 4d The figures for the PFS full sample are from UPS Sri Lanka, 1975, First Report, p. 544. "IVrked Away Home Only Bef. Marr." and "Worked Home Only Before Marriage" have been combined to get the figures for "Before only." eNot clacuiated because of very small base. i 205 are shown in Table 6.3. The two analyses (the logistic and the ordinary least squares regressions) lead to more or less parallel inferences regarding the effects of various factors. Remarks: A number of major determinants of the propensity to elect sterilization remain to be identified. One might speculate that among these are the strength of sterilization campaigns carried out by the family planning program; the availability of sterilization services at the government aud non-government medical centers; group pressures on women in reproductive age groups for and against electing sterilization; and so on. A major limitation of the approaches (logistic and OLS regression) just presented is that they treat the classifications "sterilized" and "not. sterilized" as though they were final, when in fact they are not; it is likely that in time a number of women in the "not sterilized" category will transfer themselves to the other category. Any procedure that ignores this inherent incompleteness of information, caused by the truncation of the observation period, is likely to give biased inferences about factors affecting the propensity to elect sterilization. Life Table Analysis Introduction: The propensity to elect sterilization can be analyzed using what has come to be known as lifetime distribution methodology or simply life table methods. By lifetime is meant the time interval between a date of entry and the date of occurrence of an event of interest. ("Survival time" and "failure time" are often used as synonyms for lifetime.) In the usual mortality analysis, the date of entry is the date of birth and the event of - 206 - Table 6.2: Logistic Regression- Propensity to Elect Sterilization ,Among Different Ethnic Groups. Sri Lankan Indian Sri Lankan Sinhalese Tamil Tamil Moor Factors (s.e.) (s.e.) (s.e.) (s.e.) Age .3893 (.111) .7543 (.249) .7271 (.416) .968 (.2411? Parity .9323 (.133) .7953 (.279) .199 (.341) .271 (.325) Education .054 (.021) .133 (.042) -.104 (.106) .207 (.075) (Age)2 -.0063 (.002) -.0103 (.003) -.0101 (.006) -.013 (.006) 2 3 2 (Parity) -.057 (.011) -.054 (.022) .009 (.026) -.003 (.022) Excess a 3 3 1 fertility .812 (.194) 1.528 (.559) 1.525 (.853) .619 (.650) Residenceb .579 (.155) .758 (.254) .974 (.719) .954 (.402) Excess fertility=0 or 1 according as the desired number of children< or > actual number of children. Residence=l or 0 according as the current residence is in urban area or elsewhere. 1Significant at .10. 2Significant at .05. 3Significant at .01. - 2Q7 Table 6.3: Ordinary Least-Squares Regression-- Propensity to Elect Sterilization Among Different Ethnic Groups. Sri Lankan Indian Sri Lankan Sinhalese Tamil Tamil Moor Factors (s.e.) (s.e.) (s.e.) (s.ce.) Age .054 (.014) .049 (.016) .062 (.037) .046 (.021) Parity .084 (.013) .041 (.018) .001 (.034) .001 (.022) Education .006 (.003) .012 (.004) -.010 (.012) .015 (.006) 2 3 3 1 2 (Age) -.001 (.0002) -.001 (.0002) -.001 (.001) -.001 (.0002) (Parity)2 -.005 (.001) -.003 (.001) .004 (.003) .002 (.002) Excess 3 3 fertility .088 (.023) .094 (.034) .103 (.066) .036 (.041) Residence .086 (.023) .079 (.028) .166 (.110) .079 (.033) R2 .106 .102 .127 .103 N 1708 683 214 347 1Significant at .10. Significant at .05. 3Significant at .01. - 208 - interest is death. For our purposes, however, the date of entry is the date of marriage and the event of interest is sterilization. [Another possible date of entry is the date of birth of the last wanted child (see Westoff, et al., 1979). We found it difficult to determine satisfactorily the last wanted births from the available data; hence the date of marriage will be considered the date of entry.] Basic concepts of lifetime distributions include the following: If T (usually assumed continuous) represents the lifetime of an individual in a population, the probability that T > t is called the survivor function at t; its complement (= the probability that T t) is called the distribution function; and the instantaneous rate of experiencing the event of interest at t, given that the event has not been experienced until then, is called the hazard function. The last function is also known by other names, inciuding force of mortality, hazard rate, and age (duration)-specific rate of experiencing the event (failure rate). Lifetime data often come with a feature that creates problems in their analysis. This feature, known as censoring, broadly defined, occurs when exact lifetimes are known only for some of the individuals under study, while for the remaining individuals all that is known is that their lifetimes exceed certain values. In the present case only a fraction of the individuals surveyed had undergone sterilization before the survey date; for the remaining women in the sample we know only that their dates of sterilization are beyond the survey date. [Note that we do not know the implications of implicitly assuming, as is done here, that eventually all women will be sterile, either through surgery or as a natural consequence of menopause.] - 209 Censoring is incorporated into our analysis in the following fashion. For each woman, a termination event and an exposure period are defined. For sterilized women, the termination event is sterilization, and the exposure period for each is the time interval between her marriage and the date of sterilization. For women not sterilized as of the survey date, the termination event is the survey interview, and the exposure period is the interval between the date of marriage and the date of interview. The exposure periods and the termination events of different women are then put together so as to provide the event history of a hypothetical cohort, whose size diminishes as its members undergo sterilization or are censored (see Figure 6.1). All censoring is assumed to occur at random, in the sense that the times to censoring are stochastically independent of each other and of the times to sterilization. Putting together the experiences of different individuals in the fashion shown in Figure 6.1 enables us to estimate the number of individuals (in the hypothetical cohort) who are at risk of electing sterilization at any given time point and the number of persons who actually elect sterilization during any given time interval. From these we can estimate the survivor function, using methods developed, for example, by Kaplan and Meier (1958). Subsequently, from the survivor function all other functions of the life table can be estimated. - 210 - x Sterilized -x Censored x Censored -x Sterilized Entry Date Time since entry date Figure 6.1: A Pictorial Representation of a Hypothetical Cohort Whose Size Gets Diminished as its Members Undergo Sterilization or Are Censored. (The entry dates are not chronologically the same for all women observed, but when they are put together to form a hypothetical cohort we assume that all of them start from a common entry date.) 211 Table 6.4 illustrates the output one can obtain by using the life table (survival) program of SPSS. The data used for constructing Table 6.4 are for ever-married women taken as a group. The column headings describe the entries. By comparing tables of the kind shown in Table 6.4 constructed for separate strata of a population one can determine whether the population is heterogeneous. Variables affecting the tendency to elect sterilization can be identified in a similar fachion, using the categories of the variables of interest to define strata. Factors Considered One at a Time: Sterilization tables such as the one shown in Table 6.4 were constructed separately for each category of the following variables (taken one at a time rather than in combination): ethnicity, religion, type of place of residence, education, literacy, work pattern, and excess fertility. Since complete tables contain too many details, only summary measures obtained from the detailed tables are presented below. These are supplemented with plots showing the cumulative hazard function for each table. The summary measures used below are 1. Values of the sterilization function at t t 5, 10, 15, 20, and 25 (where t stands for duration since marriage, in years); 2. Inter-quartile range obtained from the sterilization function standardized so that the value of the function is unity at t = 25; and 3. Tukey's Trimean (- a weighted average of the first, second, and third quartiles of the standardized sterilization function, with weights 1, 2, and 1, respectively, for the first, second, and third quartiles). The interquartile range is a measure of variance and the trimean a measure of average. - 212 - Table 6.4: Sterilization Table for Ever-'Married Women.- Interval No. Cum. Cum. Since Enter- Prop. Cum. Prop. Ha- Starting ing No. No. No. Prop. Not Ha- Prop. Not zard Time Inter- With- Ex- Steri- Steri- Steri- zard Steri- Steri- Func- (Months) val drawn posed lized lized lized Raite lized lized tion 0 2981 0 2981.0 0 .0000 1.0000 .00(00 .0000 1.0000 .0000 12 2981 0 2981.0 3 .0010 .9990 .0001 .0010 .9990 .0010 24 2978 3 2976.5 2 .0007 .9 993 .0001 .0017 .9983 .0017 36 2973 10 2968.0 10 .0034 .9966 .0003 .0050 .9950 .0050 48 2953 79 2913.5 7 .0024 .9976 .0002 .0074 .9926 .0074 60 2867 100 2817.0 17 .0060 .9940 .0005 .0134 .9866 .0135 72 2750 99 2700.5 28 .0104 .9896 .0009 .0237 .9763 .0239 84 2623 112 2567.0 31 .0120 .9879 .0010 .0354 .9646 .0361 96 2480 83 2438e5 34 .0139 .9861 .0012 .0489 .9511 .0501 108 2363 125 2300.5 35 .0152 .9848. .0013 .0634 .9366 .0654 120 2203 108 2149.0 30 .0140 .9860 .0012 .0764 .9236 .0795 132 2065 87 2021.5 37 .0183 .9817 .0015 .0933 .9067 .0980 144 1941 92 1895.0 39 .0206 .9794 .0017 .1120 .8880 .1188 156 1810 81 1769.5 38 .0215 .9785 .0018 .1311 .8689 .1405 168 1691 81 1650.5 35 .0212 .9788 .0018 .1495 .8505 .1619 180 1575 83 1533.5 20 .0130 .9870 .0011 .1606 .8394 .1750 192 1472 87 1428.5 28 .0196 .9804 .0016 .1770 .8230 .1948 204 1357 85 1314.5 26 .0198 .9802 .0017 .1933 .8067 .2148 216 1246 83 1204.5 14 .0116 .9884 .0010 .2027 .7973 .2265 228 1149 87 1105.5 15 .0136 .9864 .0011 .2135 .7865 .2402 240 1047 113 990.5 15 .0151 .9849 .0013 .2254 .7746 .2554 252 919 75 881.5 11 .0125 .9875 .0010 .2351 .7649 .2680 264 833 77 794.5 14 .0176 .9824 .0015 .2486 .7514 .2858 276 742 83 700.5 1 .0014 .9986 .0001 .2496 .7504 .2872 288 658 76 620.0 6 .0097 .9903 .0008 *.2569 .7431 .2969 W 300 576 86 533.0 4 .0075 .9925 .0006 .2625 .7375 .3044 312 486 83 444.5 4 .0090 .9910 .0008 .2691 .7309 .3135 324 399 66 366.0 3 .0082 .9918 .0007 .2751 .7249 .3217 s 336 330 67 296.5 0 '.0000 1.0000 .0000 .2751 .7249 .3217 348 263 52 237.0 0 .0000 1.0000 .0000 .2751 .7249 .3217 360+ 211 210 106.0 1 .0094 .9906 -- .2819 .7181 -- - 213 - Ethnicity. Comparing the values of the sterilization function across the ethnic groups (Table 6.5), we notice that the Sinhalese have the highest tendency to elect sterilization, the Sri Lankan Moors the lowest, and the Indian and Sri Lankan Tamils hold an intermediate position. The plots of the cumulative hazard function [Figs. 6.2a ti) & (ii)] tell the same story. (Note that the higher the value of the cumulative hazard function at a given time (marriage duration), the smaller the fraction remaining non-sterilized at that time.] Religion. Parallel to the ethnic group differences, the Buddhists have relatively higher cumulative proportion sterilized at any given marital duration. The graph of the cumulative hazard function [Figs. 6.2b (i) & (ii)] for the Buddhists remains higher than the corresponding graphs of the other religious groups, indicating that, with marriage duration, a cumulatively smaller proportion remain not sterilized among the Buddhists than among the other major religious groups. Residence. Urban residents have higher cumulative proportion sterilized at each marital duration than do rural and estate residents IFigs. 6.2c (i) & (ii)]e Education. Women with no formal schooling have the smallest cumulative proportion sterilized by any given marital duration; those with 1-5 years of formal education show markedly higher tendency to elect sterilization; and those with 6 or more years of schooling have a still higher propensity to do so [Figs. 6e2d (i) & (ii)]. Table 6.5: Sterilization Function for Ever-Married Women and Currently-Miarried, FecuLnd Women W4ho Want No More Children, by Ethnic Group, Religion and Residence Ever-Married Women Currently Married, Fecund Women Who Want- No Mor Clhi I irpn Cumulative Proportion Sterilized by Cumulative Proportion Sterilized by Subclass t=5 t=10 t=15 t=20 t=25 IQR TM t=5 t=10 L=15 t=20 t=25 IQR TM Ethnic Group: Sinhalese .0112 .0844 .1870 .2508 .2907 8.04 13.25 .0202 .1472 .3256 .4455 .5372 8.36 13.67 Sri Lankan Tamil .0015 .0422 .1004 .1569 .2073 8.37 15.32 .0026 .0672 .1635 .2557 .3560 8.69 15.79 Indian Tamil .0000 .0214 .0982 .1882 .2115 5.63 15.72 .0000 .0349 .1540 .2964 .3353 5.70 15.76 Sri Lankan Moor .0058 .0254 .0787 .1331 .1952 10.63 17.08 .0097 .0407 .1211 .0278 .3283 10.86 17.61 Religious Group: Buddhist .0108 .0826 .1841 .2513 .2938 8.02 13.48 .0198 .1462 .3261 .4538 .5521 8.60 13.96 Hindu .0026 .0352 .0995 .1643 .2004 6.53 14.97 .0045 .0561 .1604 .2669 .3322 6.71 15.21 Mluslim .0057 .0248 .0816 .1358 .1969 10.49 17.01 .0095 .0398 .1256 .2124 .3322 10.72 17.56 Christian .0039 .0816 .1724 .2266 .2738 8.68 13.09 .0066 .1286 .2676 .3580 .4642 9.21 13.74 Residence: Urban .0154 .1028 .2248 .3036 .3541 7.83 13.12 .0244 .1550 .3355 .4558 .5562 8.39 13.50 Rural .0055 .0501 .1253 .1844 .2301 8.49 14.64 .0102 .0887 .2222 .3339 .4358 8.75 15.11 Estate .0000 .0563 .1310 .1964 .2057 6.99 13.16 .0000 .0916 .2055 .3103 .3263 7.21 13.18 Table 6.5: Sterilization Function for Ever-Married Women and Currently-Married, Fecund Women Who Want No More Children, by Ethnic Group, Religion and Residence (cont'd). Ever-Married Women Currently-Married, Fecund Women Vnio Want to More Children Cumulative Proportion Steriliz<-d by Cumulative r2portir n Sterilized hy Stubclass t=5 t=10 t=15 t=20 t=25 IQR ni t=5 t=10 t=15 t=20 t-15 IQR TM Iducation: None .0167 .0248 .0584 .1191 .1506 6.71 16.53 .0033 .0451 .1114 .2390 .3151 6.62 16.91 Grades 1-5 .0093 .0466 .1364 .2003 .2430 6.62 14.33 .0169 .0823 .2371 .3534 .4477 7.21 14.75 Grades 6-9 .0062 .0988 .2205 .2891 .3647 8.79 13.72 .0098 .1464 .3242 .4276 .5561 9.51 14.06 Higher .0136 .1200 .2417 .3153 .3153 7.07 11.32 .0242 .1929 .3964 .4770 .4770 7.19 11.12 Literacy: Literate .0097 .0790 .1855 .2545 .3110 7.83 13.77 .0164 .1268 .2931 .4068 .5184 8.63 14.26 U' Not literate .0013 .0233 .0652 .1190 .1452 7.05 15.95 .0026 .0437 .1269 .2392 .3010 7.11 16.29 Wife's Work: Never worked .0062 .0596 .1436 .2073 .2632 8.35 14.54 .0108 .0980 .2347 .3433 .4554 9.21 15.04 Before only .0218 .1139 .2304 .2804 .3310 6.96 12.29 .0368 .1818 .3605 .4537 .5892 10.20 13.66 After-Away .0066 .0571 .1360 .1828 .2033 7.77 13.21 .0126 .1019 .2436 .3322 .3790 8.14 13.46 After-Home .0031 .0580 .1510 .2633 .2991 6.21 14.75 .0055 .1010 .2629 .4655 .5433 6.18 15.11 Excess Fertility: Desired < living .0018 .0412 .1511 .2515 .3088 6.20 15.22 .0027 .0573 .2215 .3820 .4962 7.00 16.04 Desired = living .0017 .0909 .1810 .2334 .2747 8.39 12.87 .0167 .1299 .2655 .3524 .4332 8.90 13.47 Desired > livingw .0035 .0195 .0638 .1120 .1346 7.22 15.92 .0141 .0733 .2135 .3469 44218 7.42 15.15 - .216 - Figure 6.2a(i) >1:.>PiE- TO STER !L 1, T I N EVER M2Ri.'ED WOMEN E',l TRC GROUP CUMULVIFE HfiZPlRD FUNCTION 0.50 5 C / M 0.30 . L T I 9.02-- v/ H OCOI / // 7 -- FR 0. 0- ' 3. i 2 ; 1t' 2 2 2 2 2 2 2 2 3 n2 3 ,, 6 -, 8- a I 2 5 e 7 K G TIM ?E SINCE MARRIMGE J7 IR; tt,LESE DT9M0ND=SIRI LRNKIKR TfMIL D . 11 D 7 h. N T./l 11 - X --C I . 1N M 'sET /PY AVAILA LE Figure 6.2a(ii) MARRIAGE TO STERILIZATION: MARRIED FEUUNO WOMEN WHO WANT NO MORE CHILDREN ETHNIC GROUP UUMULATIV'E H9.A9RD FUNCTION 1. zJ - 0.g M 0.6 'UX L T R A 0.4t D 0.2 11 1 1 1 1 1 1 i1 2 2 2 2 2 2 2 2 2 2 3 2 3 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 X 5 6 7 8 9 0 TIME SINCE MARRIAGE 9TAR=SINH9LESE DIAMOND=SRI LANKA TAMIL HASH INDIAN T9MIL X=SRI LANKA MOOR - 217 - 218 - Figure 6.2b(i) MARRTGE- TO STE.RILIZ7TTIflN'. EVER MARRIED WOMEN REU IGI'ON CUIJUtLATIVE- HRZARD FUNCTJl0N -3 0.. |. DrD. .3 C 11 03. -, T I0-2 S l 2 3 . i , t 1, 2 2 ] 2, 2 2, 2, 2 2, 2 :a T'IME SIlNCE MA iRRI'A G S lAR=BUD3HJ'ST' DI1AMNDatHJ MDU HRSHMU~SLI,M - X=CHRJIST I RN BEST COYAVAILABLE Figure 6. 2b(ii) MARR IGE TO STERILIZATTON: MARRIED FECUND WOMEN WHO WANT NO MORE CHILDREN RELIGTON uUMULATI'/E HAZARD FUNCTION M 9.6- A T E H D rJ3- 0.2 1 11 1i 11 11 1 2 2 2 2 2 2 2 2' 2 2 3 2 3 5 6 78 9 0 1 2 3 5 6 7 8 9 0 1 2 3 5 6 7 8 9 0 TIME SIN8CE MARRIA&E 9TAR=BUDHiST DIAMOND=HINDU HASH=MUSLIM X=CHRISTIAN - 219 - BEST COPY AVAILABLE - 220 - Literacy. As one would expect from the educational differentials just described, literacy is positively associated with the tendency to elect sterilization. The graph of the cumulative hazard function for the literates is much higher than that for the illiterates [Figs. 6.2e (i) & (ii)]. Work Pattern. Women who work only before marriage have the highet tendency to elect sterilization by any given marital duration. There are no clear-cut contrasts among the other work-pattern groups, within the first 15 years of marriage. Beyond 15 years of marriage, however, women who work at home have a higher propensity to elect sterilization than do women who never work, whereas this latter group shows a higher tendency to elect sterilization than do women who work away from home after marriage [Figs. 6.2f Ci) & (ii)]. It is impossible to make sense of such patterns without taking into account other variables (e.g., education). Excess Fertility. At longer marital durations, the cumulative proportion sterilized are relatively higher for excess-fertility women (i.e., those who desire fewer children than they actually have). Women whose current family size is equal to their desired family size do not seem to be markedly different from the excess-fertility group. Women with deficit fertility (those with actual family size < desired family size) show a relatively lower tendency to elect sterilization. The differences among the three groups are much less pronounced in the smaller sample (currently married, fecund women who do not want any more children). [See Figs. 2g (i) & (ii).] This is not surprising, since these women are relatively homogeneous in their desire to prevent additional births. .-221- Figure 6.2c(i) MRRRIAGE-TO STERITIZRTION' EVER. MARRIED' WOMEN TY.PE- OF- lPLfCE .OF- RESIDENCE CAUMULATTIVE HRZARD FUNCTIDON. a.qo. a. 3 - U M a. 30. U- L T- I Q. 25 E ' a. . L, , 1, 1, 2 I 2 , 2 ,2 I. 2Q 25 4. / . 1 .5 6 . 0 ,2 f .6 7 // TIME- S'INCE-.MRRRI'fGE. S'T-A,-RIRB RN' DIRM,SND- RURti HASH- ESTRTE BEST COPY AVAILADLE Figure 6.2c(ii) !MARRIAGE TO STERILIZATION: MARRIED FECUND WoMEN WHO WANT NO MORE CHILDREN TYPE OF PLACE OF RE3IDENCE CUMULATIVE HAZRD FUNCTION 0.7- m 9. t u L T E S 9 - Z- Ar 0. 22 - 9.0 'I..I, . I- I'7 7i 7 I7 l 7I ... I I I' 12345 67 8 90 12 3 5 67 8 90 1 23 45 67 89 0 TIME SINCE MARRIAGE STAR=URBAN DTAMOND=RURAL - HASH=ES TATYE -222 EST Co"O PY AVVA I L AB'L E Figure 6.2d(i): MARRIAGiE TO STERILIZATION: EVER MARRIED WOMEN EDUCATION CUMULATIVE HAZARD FUNCTION 0.50- M 0.30 U L T I 0.25 V E H A 0.20 z R D 0.10 0. 05 TIME SINCE MARRIAGE TFAR=NONE DIAMOND=G3RADES 1-5 HASH=GRADES 6-9 X=GRADES 10+ -223 Figure 6.2d(ij.): NARRI9GE TO STERTLIZ9TION: MARRIED FECUND WOMEN 4HO WANT NO MORE CHILDREN EDUCATTON CUMULATIVE HAZARD FUNCTION 09. m 9. .9 - tt 0.7 D U; 1 1 1 1 1 1 1 1 i 1 2 2 2 2 2 2 2 2 2 2S i 2 3 56, 8 9 0 IL 2 3 4 5 6 7 8 9 0 1 2 3 't 5 6 7 8 9 0 TIME SINCE MARRIAGE ST,9R=NONE DIAMOND=GRADES 1-5 HW,93H=GRADES 6-9 X=GRRDES 10+ -224 - BE$T CC0PY AV AILA-St E Figure 6e2e(i): MARRIAGE TO STERILIZATION: FVER MARRIED WOMEN L.ITERACY CUMULATIVE HAZARD FUNCTION 0.50- 0. 5 uI H 0 . 3 0 / M 0.30 U T E A 0.20 R 0.15 0.10 0.05 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 TIME SINCE MARRIAGE STAR=LITERATE DIAMOND=ILLITERATE - 225 - Figure 6.2e(ii): MARRTAO,GE TO STERILIZ,ATION: MARRIED FECUND WOMEN WHO wANT NO MORE WHILDREN L-ITERACY LAUMULATIVE HA7ARD FUNCTION 0.9 M . 5 -0 I 0.5CXor V E H, R D 0.2 ,, , .. ,;. I ....I , , ,, , ., 7 .. .7 ,I I ,i,-7 I...................... 1111111 11 1 2 2 2 2 2 2 2 2 2 2 3 2 3 5 6 7 8 9 0 1 2 3 q 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 TIME SINCE MARRIAGE STAR=LITERATE * DIAMOND=ILLITERATE - 226 - Figure 6. 2f (i): MRRRIAGE- TO STERILI2OTION,. EVER MARRIED WOMEN WORK. PIAT'TERN CUMULATIVE- HfZRRD FUNCT'I'ON Q.5 -i 0-35) 0 Q. 35 Q 1 ..3Q' L T D.. 227 Q.. is 1. 2.,3 U , 5 6 I. R. 9.,O 0.-1 2, 3 5'- S~ 7 a R,. 0. 3. 2 3 q ~ 7T1ME- S~INCE- M1ARRIfRGE STRR=NCVER-14ORKED flIAMOND-WOR?KED BEF ORE ONLY HASH-WORKEn AFTER AWRY- X=WORKE]2 RFTER R7,HOME -227 - Figure 6.2f(ii): MARR19GE TO STERILIZATION: MARRIED FE-iUND WOMEN 1HO WANT NO MORE CHILOREN 4OR.K PATTERN CUMULATIVE HAZARD FUNCTION i 9. 6 -2 - 9L A 9. . e .- D U- 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 i2 3 $ 5 6 78 9 0 1 2 3 q 7 6 7 'k 9 0 1 2 3 4 5 6 7 8 9 0 TIME SINCE MARRIAGE STAR=NE'VER WORKED DIAMOND=WORKED BEFORE ONLY HASH=WORKED AFTER RWR ' X=WORKED AFTER AT HOME M 228 Figure 6.2g(i): RRIAGEf TO STERIL IZRTTONW EVER.-MYRRIED WOMEN EXCESS- f ER :I ITiY (AUMULATIVE .HRZARD FUNCTION. 0.O. .......,.. '' O>S- 01.~ 0. 0.- 0. 35 - C M . 3 0 U L T R 0..20.- z R 0 0 I0..I iIflI I I I -1 11 71 7 1 1 1 1 1. 1 1 1. 1. 1, 2, 2 2 2, 2, 2, 2, 2 2, 3 2 3 5 6 7 a 9, 0. 1, 2. 3 5 78 9, 0. 1. 2, 3 4 6 7-8 9' o. T'ITME-.S'INCE MRRIRRGE. SPTIR[DESI'RED LES'i THPfN' 4IV-TND. DIRtMDN0D ESJIRED EO.VWl TO UVING. HASH=0ESJIRED GREATER-.THAN' LIV'ING, - 229 - Figure 6.2g(ii):MARRIAuE TO STERILIZATION: MARRIED FECUND WOMEN WHO WANT NO MORE CHILDREN EXCESS FERTILITY UUMULATIVE HAZARD FUNCTION 0.8 0.7 C L A T I 0.5 E H A R- D 0.32 0. 0 7- 'I' 7,7 - 1 1111 71-111122222222223 I19 ..11-11- 1 1 1 11 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 1 2 3 4 5 6 7 8 9 0 1 2 3 q 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 TIME SINCE MARRIAGE STARRDESIRED LESS THAN LIVING DIAMOND=DESIRED EQUAL TO LIVING HASH=DESIRED GREATER THAN LIVTNG - 230 - BEffbT COP"T AVAILAMULE - 231 Examination of Factors Within Each Ethnic Group: We now examine each ethnic group to see whether the differences described above hold when ethnicity is controlled, and also whether there is any ethnic-group diversity in the differentials. Education. For the Sinhalese, the major difference is between (i) those with no formal schooling, (ii) those with 1-5 years of formal education, and (iii) those with 6 or more years of education [Table 6.6 and Figs. 6.3a (i) & (ii)]. The differences by education are less pronounced for the smaller sample (currently-married, fecund women who want no more children). The patterns for the Sri Lankan Tamils are not much different from those observed for the Sinhalese. [See Table 6.6 and Figs. 6.3b (i) & (ii). Because of small numbers of cases, only the lower two educational categories, no education and 1-5 grades of schooling, are considered for currently-married, fecund women who want no more children among Sri Lankan Tamils.] For the Indian Tamils, the smallness of the sample size permits only one contrast, namely that between women with no formal education and those with 1-5 years of schooling. Those with some schooling show a greater propensity to elect sterilization [Table 6.6 and Fig. 6.3c]. A similar pattern prevails for the Sri Lankan Moors also [Table 6.6 and Fig. 6.3d]. Other Factors. Only education is discussed here because, for other factors, the patterns described earlier for all ethnic groups combined prevail within each ethnic group. - 232- Table 6.6: Steriliz-ation Function--Ever-Married Women and Currently-Married, Fecund Women Who Want No More Chi.ldren in Education Subclasses of Ethnic Groups. Cumulative Proportion S,terilized by 'arriage Duration (Years) Education t=5 t=10 t=15 t=20 t=25 IQR TM Ever-Married Women: Sinhalese None .0037 .0310 .0739 .1439 .1820 6.59 16.65 Grades 1-5 .0139 .0649 .1631 .2267 .2778 7.43 14.03 Grades 6-9 .0094 .1211 .2591 .3243 .3572 7.20 12.47 Higher .0156 .1280 .2679 .3226 .3226 7.01 10.94 Currentlv-Married, Fecund W%Iomen XTho Want No More Children: Sinhalese None .0084 .0703 .1686 .3367 .4388 6.61 11.58 Grades 1-5 .0273 .1233 .3068 .4348 .5660 8.86 14.69 Grades 6-9 .0146 .1769 .3782 .4793 .5359 7.94 15.15 Higher .0269 .2038 .4052 .4836 .4836 7.09 16.36 Ever-Married Women: Sri Lanka Tamil None .0202 .0422 .0563 .0872 .0872 10.45 11.23 Grades 1-5 .0079 .0554 .1355 .1517 .1799 4.65 11.61 Grades 6-9 .0000 .0618 .1355 .1997 .3334 9.42 16.47 Higher .0106 .1162 .2051 .2820 .2820 7.60 11.84 Currently-Married, Fecund Women Who Want No More Children: Sri Lankan Tamil None .0000 .0208 .1298 .2312 .2884 4.19 18.21 Grades 1-5 .0000 .0962 .2041 .3082 .5279 9.53 17.44 Ever-Married Women: Indian Tamil None .0000 .0300 .0738 .1411 .1752 6.91 16.02 Grades 1-5 .0000 .0120 .1447 .2810 .2810 4.17 14.86 Ever-Married Women: Sri Lankan Moor None .0000 .0000 .0211 .0768 .0971 2.74 17.77 Grades 1-5 .0012 .0457 .1142 .1416 .1982 12.12 14.72 - 233 - Figure 6. 3a(i): .pMRiGGE TO STERILUfTI.ON: EVE-R MFRRIED WO.IEN SSINHALESE CDUCflT1TN CUIJMJLfTIVE fi-ZflRD fUNCTIDN 0.. -. 0 3.S U - Z L Fl 0.20 O.0i 0.10 0.. 0.5 1 2 5 3 '$78S. 0 a2 3 1$ 6 7 8 oa2 :3 4 $ 7 8 9 0 TIME SINCE- 11aR.[TGE -STfU~=N.NE D1ORND=GJnfiDf.S- 0-s HFl.~#G~lDESEs-SY::GRffES~ 0-1I-r Figure 6.3a(ii)^ARRIAGE To STERILIZATION: MARRIED FECUND WOMEN WHO WANT NO MORE CHILDREN SINHALESE CUMULATIVE HAZARD FUNCTION EDUCATION 0.9 ru 0.8- M 0.6 U L T I i 0. 1 11 111 1 i 2 2 2 2 2 2 2 2 2 2 3 i 56789 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 TIME SINCE MARRIAGE STAR=NONE DIRMOND=GRRDES 1-5 HRSH=GRROES 6-9 X=GRRDES 10+ -234- - 235..- Figure 6.3b(i): .MRRIAGE TO STERILIZRTION: EVE:R MARRIED 'WOMIN SRI LfANKEx TPMILS EDUCRTION CUMULfiTIVE HflZFRD FUNCTION u 0.30 L T I 0.25 E H Pi 0.20 z 0.15 0. 10 TIME SINCE MfIRRRIGE STRR=NDNE DlIMON0=GR5f-DS 0-5 HPISH=GRADES 6-9 X=GRPrDES 104 - 236 -- Figure 6.3b(ii): 1RRF1Rf TO SKIRLU7fIlRDN: MRKIf.D ffEC1JND WOIMEN MN WPLNT N0 MIR1- CflLDRf.N .SRI LfNKft TIMIL-S DTJC.RTITON CUIJMULJ2VE I-lflZRfl FIJNCU7DN M 0.6 L ft -T a 0..5 fi -. . At O: 2 0.2 I I I I I 1 1 1 a2Q 2 2 2 2 2 2 2 2 23 1 2 3 5 16 -7 8 9 0 1 2 13 T S i 6 .9 0 123 2 I 5 fi 1 8 0 T2M£ $SINC C MfRRhP f- STfIR=GRRDf-S 1-5 DIfMOND-6RflD-S 6-.9 237 - Figure 6.3c: MPURRlf. TO S7fhRIUTTiO7DN: £VftR M1lRRRlfD W1OMEN 2NDIRN 'TMLLS f-D1JC.RT20N CUNULA7TVE iffl7fRD FVUNCT7-ON O..5 - C 1 'U M1 0.30- .t 'U L -I. 1 0.25- V R D / 15i r a a z 1: 1 1 2 Q '2 Q '2 2 '2 72 2 72 3 l 2 73 e 5 -6 7 8 9 0 1 2 3 7 5 -6 7 .9 0 1 2 3 q .5 6 7) 8. .9 0 TIME S2NC{f MRRRhG-E STrRR=NflNE DfIflMND= GRfIDV D-.5 - 238 - Figure 6.3d: M1R2Rfl90- O SEJCLZfl7,J-N- fVER-A RRIF-D WOMEN -SR: LRANK9 QORS f DUCfl7TDN CU.MULA7JVE. -HflZfRD FUNCT7O.N o.S'Q ----- - -. - l o-- c 11 V n1 D 3D 'U L I D- 25- v R is D 15 0- os 00 0 O ~~~ ~ J QD J ; J. .j;J,.; ....; . . 1 1 1 1 2 2 2 2 22 2 2 23 1 '2 3 1. 5-6 7 e. 9 B 1 2<3 i -5 6 7 8 9 5 12 S5 t i. o llflE DMND !CfiRRD&I-9S S7flR-JNXM DIRMOND= GIUD9E-S 3-5 - 239 - Popularity of Sterilization: By sorting reported sterilizations into two categories according to those reported to have taken place during the 60 months and' during the 60-120 months prior to the 1979 interview, a comparison may be made to determine whether the popularity of sterilization as a means of birth control changed from 1969-74 to 1974-79. There is, however, one serious limitation to this approach, i.e., that the reinterviewed respondents in the 1979 survey are not equally representative of the population in both periods, 1969-74 and 1974-79, because of aging. Table 6.7 and Figures 6.4a and b portray selected features of life tables constructed from the two categories of sterilizations described above. For the life table based on sterilizations reported to have occurred between 60-120 months prior to the 1979 interview, all sterilizations occurring outside that period were treated as censored. A similar step was adopted with respect to the life table based on sterilizations during the 60 months prior to the 1979 interview. Table 6.7 shows that the sterilization function for the 1974-79 period is steeper than that for the earlier 1969-74 period, and that the cumulative hazard function for 1974-79 lies correspondingly far above that for 1969-74 (see Figs. 6.4a and b). This signifies that there was a noticeable increase between 1969-74 and 1974-79 in the duration-specific propensities to elect sterilization as a means of birth control. Figure 6.4a: MARRIAGE TO STERILIZATION: EVER MARRIED WOME.N POPULARITY CUMULATIVE HAZARD FUNCTION 0.50 - 0. i- 0 0.35 - C. U M 0.30 U L T I 0. 25 - V E H R 0.20 R 0.15 - 0.05 7 7 - ,7 .1,7 1 I.. - . 1 . i2 3 4 5 6 78 9 0 1 2 3 4 5 6 7 8 9 0 2 2 3 4 2 6 7 2 2 ' TIME SINCE MARRIAGE STAR=1969-1974 DIRMOND=1974-1979 - 240 - Figure 6.4b: MARRIAGE TO STERILIZRTION: MARRIED FECUND WOMEN W1H WANT NO MORE CHILDREN POPULARITY CU?JLRTIVE HAZRRD FUNCTION 1.0 - 0.9 0.8 0.7 C M 0.6 U L A T I 0.5 V H n O. z R R D Oe3 0.2- 1111 1 1 1 1 22 2 2 2 2 2 2 22 3 1 2 3 f 5 6 7 8 9 0 1 2 3 i 5 6 7- 8 9 0 1 2 3 4 5 6 7 8 9 O TIME SINCE MARRIRGE STAR-1969-197i DIRMOND-197f-1979 - 241 - - 242 - Table 6.7: Sterilization Functions Based on Sterilizations Reported to Have Occurred During 1969-74 and During 1974-79--Ever-Married Women and Currently-Married, Fecund Women Who Want No More Children. Currently-Married Fecund Women Who Ever-Married Women Want No More Children Measures 1969-74 1974-79 1969-74 1974-79 Sterilization function for duration (yrse) 5 .0020 .0034 .0036 .0060 10 .0190 .0361 .0323 .0616 15 .0508 .0877 .0888 .1512 20 .0762 .1261 .1391 .2246 25 .0923 .1564 .1809 .2949 IQR 7.49 8.24 8.10 8.71 TM 14.51 14.31 15.39 15.08 -I. z - 243 - Effect of Sterilization on Fertility: The effect of sterilization on fertility depends on several factors, e.g., the proportioni of women who undergo sterilization, the interval between sterilization and the end of the reproductive age span for those electing sterilization, and the fertility of the sterilized couples if they have not been .sterilized. With regard to this last factor, in most Asian countries the couples who elect sterilization are self-selected for higher fertility, and Sri Lanka is no exception. At each duration of marriage and at each age, sterilized women in Sri Lanka had higher cumulative fertility than their non-sterilized counterparts (Table 6.8). This does not necessarily imply, however, that sterilized couples would have continued to experience higher fertility if they had not been sterilized. It is possible, for example, that those factors which motivated people to elect sterilization might have prompted them to adopt some other effective method of contraception if the sterilization option had not been available. A compromise assumption is that the sterilized couples would have experienced the same marriage duration-specific fertility as observed for nonsterilized couples. Westoff et al. (1980) estimate, from the 1975 data, that the total marital fertility rates for the five-year period, 1970-75, would have been approximately 3-7 percent higher in the absence of sterilization. The lower figure (3 percent) is based on the assumption that the sterilized couples would have experienced the unwanted birth rates of those who wanted no more births, while the higher figure (7 percent) is - 244 - Table 6.8: Mean Number of Children Ever Born to Ever-Married Women, by Current Age and Duration of Marriage for Sterilized and Non-Sterilized Couples. Factor Sterilized Non-Sterilized Difference Current Age 25-29 4,22 2.38 1.84 30-34 5.10 3.58 1.52 35-39 5.86 4.74 1.12 40-44 6.92 5.35 1.57 45-49 7.78. 5.85 1.93 Duration of marriage (yrs.) 5-9 3.55 2.35 1.20 10-14 4.61 3.59 1.02 15-19 6.03 4e63 1.40 20-24 7.03 5059 1.44 25-29 8.17 6.34 1.83 30+ 8.36 6.90 1.46 Source: C. F. Westoff, N. Goldman, and M. K. Choe, "Prevalence and Demographic Significance of Contraceptive Sterilization in Fiji, the Republic of Korea, and Sri Lanka." Papers of the East-West Population Institute, No. 66, April 1980. - 245 - based on the assumption that the sterilized couples would have had the same marriage duration-specific fertility rates as those of the nonsterilized couplese Another estimate of the impact of sterilization on fertility available is that of Immerwahr (1981). Based on the 1975 data, Immerwahr (1981) estimates that approximately 185,000 couples in the country were sterilized prior to 1976. [This figure is obtained by inflating Immerwahr's estimates of sterilization in 1975 by a factor of 4/3 and then adding it to the number of sterilizations prior to 1975. The reason for inflating the 1975 estimate is that the given estimate corresponds to roughly the first three quarters of 1975, the survey having been completed in the months of August to October.] Also, Immerwahr (1981:36) gives 68,000 as a lower estimate of the births averted by sterilization prior to 1976. Distributing this figure over the years in proportion to the number of sterilization, one finds that approximately 19,000 births were averted in 1975 due to sterilization. This is 5 percent of the 1975 registered number of births. Hence, according to Immerwahr's (1981) figures, the crude birth rate would have been 5 percent higher than in 1975 if there had been no sterilization. The official statistics (Family Health Bureau, 1982) show that the annual number of sterilizations declined somewhat during the latter half of the 1970s from the peak levels of 1974 and 1975. However, the total number of sterilizations during 1975-79 was almost double that during 1970-74 (151,000 vs. 81,000). Hence, the impact of sterilization on fertility mitst have been higher in the 1975-79 period than during the early 1970s. No attempt has been made in this report to present a quantitative assessment of the relative impacts of sterilization on fertility during the two halves of 1970-79. - 246 - CHAPTER VII CURRENT USE OF CONTRACEPTION: PATTERNS AND DETERMINANTS Introduction The central concern of this chapter is to examine the patterns of contraceptive usage in Sri Lanka. The focus of the analysis is on the current use of conventional methods of contraception,l a measure that is appropriate for describing the contraceptive practice prevailing in a population at any given time. Ever-use of contraception, although important in its own right, lacks some of the characteristics of good measurement associated with current use. First, it does not reflect a woman's current status; second, and more fundamentally, it does not measure exclusively the probability of use by women who feel they need contraception (Ryder and Westoff, 1971:114). For comparative purposes, however, data on contraceptive knowledge and ever-use are also analyzed and the results are presented in Appendix 7A. Information on current use was obtained by routing the responses of currently-married women exposed to the risk of pregnancy (non-sterilized, fecund women) through the following questio,as: a. Are you or your husband using a method to keep you from getting pregnant? If the response was affirmative, they were further asked: b. What method are you using? The respondents in the study sample can be categorized into two 1Use of sterilization for contraceptive purposes is analyzed separately. - 247 - subgroups. The first group consists of women who had participated in the World Fertility Survey conducted in 1975 and were declared eligible for the present survey by meeting the age criteria. The second group consists of women who became eligible for interview either (1) by meeting the age and marital status criteria in the interim period, or (2) by change of residence to units that were targeted for interview in the World Fertility Survey. The two sets of respondents differ systematically from each other in their socio-economic and demographic characteristics, as well as in their orientation toward family size (Table 7.1). The newer respondents are younger (see also Table 7.2), married at older ages, desire smaller families, and are more educated than their reinterviewed WFS counterparts. However, fewer of them are currently in the labor force, and fewer have ever worked, either before or after marriage. In view of these differences, it is not unreasonable to assume that the two groups of respondents are likely to differ in their contraceptive behavior also. Further, iS can be assumed that the contraceptive behavior of the new respondents is a better indicator of the more recent pattern of contraceptive usage. The two groups therefore are treated separately in the analysis and, for the sake of convenience, are hereafter referred to as the WFS respondents and the new respondents (NR). The next section briefly examines the Family Planning Program, followed by a discussion of the differentials in current use of contraception by socio-economic and demographic variables identified by other studies as being relevant for the analysis of contraceptive use. Finally, the relative importance of factors that differentiate between users and non-users of conventional methods of contraception are identified within a multivariate framework. - 248 - Table 7.1: Re-interviewed WFS Respondents and New Respondents by Socio-Economic and Demographic Characteristics. Respondents WFS New Mean age (years) 34.4 27.9 Mean age at marriage (years) 19.8 21.2 Desired family size (mean) 3.9 21.2 Children ever born (mean) 4.2 2.3 Children dead (mean) 0.4 0.2 Wife's education (years) 5.1 5.6 Husband's education (years) 5.8 6.7 Avg. household expenditures (Rs) 504.4 526.7 % currently in labor force 25.9 22.4 % ever worked 41.0 40.1 249 - Table 7.2: Percent Age Distribution of the WES and New Respondents, Sri Lanka. Respondents Age Group WFS New 15-19 0.2 7.0 20-24 5.2 29.9 25-29 20.5 28.1 30-34 25.1 18.8 35-39 25.6 8.7 40-44 16.1 5.7 45-49 7.4 1.8 Total 100.0 100.0 (N) (1765) (789) - 250 - The-National Family Planning Program As already indicated, family planning was initially made available in Sri Lanka by the Family Planning Association (FPA) as early as 1953 for protecting the health of mothers and children. In 1965, by a Cabinet decision the Government of Sri Lanka (GOSL) accepted family planning as an integral part of the Maternal and Child Health Services, and the Family Planning Bureau, now known as the Family Health Bureau was established within the Ministry of Health (now the Ministry of Family Health) with the responsibility of implementing family planning activities. The FPA continues to function, but mostly in research and promotional activities, directing its efforts to areas and groups outside the scope of the official program.. Table 7.3 illustrates the pattern of family planning acceptance in Sri Lanka during the period 1968-1980. There was an increase in the number of acceptors from 1968 to 1975 and substantial increases in female sterilization during 1972-74, a period distinguished by heightened family planning activity on the one hand, and by the lack of any policy movement towards anti-natal legislation on the other. This was followed by a decline in total acceptors in 1976-77. Thereafter, an increase was again observed, and in 1980 there were 153,533 new acceptors of family planning. Roughly, this means that in 1980 approximately 42 out of every 1000 women of reproductive age were new acceptors of family planning, an increase of 8 new acceptors per 1000 since 1975.2 2Computed using acceptor statistics from Table 7.3 and estimates of women in reproductive ages. - 251 - Table 7.3: New Acceptors of Family Planning. Sterilization Year Male Female Total Orals IUD Condom Total 1968 - - 1,011 16,014 20,615 4,375 43,964 1969 2,947 25,284 19,537 5,207 54,534 1970 -- - 4,971 26,889 15,799 6,416 55,279 1971 245 4,090 4,335 25,828 11,446 6,945 49,323 1972 498 9,078 9,576 32,300 18,599 9,290 71,043 1973 1,850 18,398 20,248 34,214 27,528 12,963 95,931 1974 7,292 34,942 42,234 35,924 29,693 NA 107,851 1975 6,034 33,130 39,164 37,720 32,755 NA 109,639 1976 2,924 32,664 35,588 25,597 27,030 NA 88,215 1977 1,302 17,752 19,055 27,514 21,321 NA 67,890 1978 2,325 19,624 21,949 31,146 23S085 NA 76,180 1979 5,586 27,740 33,326 29,022 18,878 NA 85,804 1980a 45,967 55,215 101,182 26,867 17,055 NA 153,553 aProvisional totals. Source: Reports on New Acceptors of Family Planning, Evaluation Unit, Family Health Bureau, Sri Lanka. Table 7.4 gives information on the demographic characteristics of new acceptors. Among acceptors of sterilization, there is a decrease in both the age and parity at which sterilization is performed. The increase in the number of vasectomy acceptors in 1980, if maintained, appears to bode well for the future of fertility decline in Sri Lanka, since the wives of vasectomy acceptors are at lower parities than women who themselves get sterilized. The pill and the IUD are favored by younger women at lower parities, mainly for spacing purposes. - 252 - Table 7.4: Demographic Characteristics of New Acceptors, 1976-79. Vasectomy Tubectomy IUD Pill 1976 Mean age of mother 32.3 31.5 27.0 27.5 Mean parity 3.8 4.8 2.8 2.8 1977 Mean age of mother 30.5 31.2 27.0 27.5 Mean parity 3.5 4.5 2.6 2.7 1978 Mean age of mother 30.0 31.1 27.4 27.4 Mean parity 3.5 4.5 2.7 2.7 1979 Mean age of mother NA 31.1 27.0 NA Mean parity 3.5 4.4 NA NA NA - Not available. Source: Service Statistics, 1976-1979. Much of this upsurge in contraceptive acceptance since 1978 can be attributed to initiatives in family planning that were implemented by the GOSL with a view to controlling population growth. Bonuses in excess of daily wages lost and out-of-pocket expenses incurred, as well as extra leave days, were granted to any employed person voluntarily undergoing sterilization. The success of the scheme, which has been in operation since May 1979 is reflected in the number of sterilization acceptors, both male and female, in 1980. It is the hope of the GOSL to extend a similar system - 253 - of sterilization bonuses to the unemployed.3 Another move that is hoped will prove beneficial to the family planning program is that program administration and direction are decentralized, and have become the responsibility of the 24 districts. Such organizational decentralization is expected to ensure that motivation, provision of services and referral in each district are responsive to local conditions.4 In recent years, attention has been focused on analyzing the impact of the national family planning program on the decline in the crude birth rate. It is generally agreed that up to 1970 the majority of the decline is attributable to an increase in the age at marriage (Fernando, 1976). Since 1970, however, the effect of declining marital fertility has increased and, more recently, even matched the effect of increasing age at marriage (Alam and Cleland, 1981). Although there is no direct evidence, the coincidence in the timing of the shift in the relative importance of the contributions of the two factors5 to fertility decline and the increased emphasis accorded the family planning program in general, and the promotion of services in particular, suggest a causal connection. 3Many public and private corporations, including the tea industry, which is Sri Lanka's largest single employer, already pay sterilization bonuses which are set to compete with generous maternity benefits. 4"UNFPA Assists Sri Lanka in its Drive to Lower Birth Rates," in Population, UNFPA Newsletter, Vol. 6, No. 1, January 1980. 5Age structure is the third factor that affects the crude birth rate. In developing countries, however, the age structure is generally conducive to a higher fertility because of past high fertility and improving health conditions which ensure a large proportion of the female population surviving to reproductive ages. - 254 - Current Use of Contraception Among the women interviewed for the survey, sterilization is not only the most popularly known, but also the most frequently used method of contraception. Of all (4360) currently-married women 12-49 years of age, 14.3 percent, or 624 women, were protected against future pregnancies because either they, or their husbands, were sterilized. Table 7.5 shows that the proportion of women sterilized increases with age up to age group 35-39 years; thereafter, the proportion sterilized declines steadily. This variation in the prevalence of surgical sterilization by age is not unusudl for a terminal method of birth control (see chapter VI). Use of sterilization is low at younger ages because these women are still in the process of family formation; use is low among older women either because of biological reasons (they have become sub-fecund or biologically sterile), or because they are less aware of and receptive to innovation than their younger counterparts. Table 7.5: Proportion of Currently-Married Women Who Are Sterilized by Age of Respondent. Age Group Percent Sterilizeda 15-19 20-24 1.4 25-29 6.3 30-34 20.4 35-39 22.3 40-44 21.0 45-49 12.6 Total 14.3 .(N) (624) aBased on 4,360 currently-married women in the sample. - 255 - Of the remaining 3,736 currently-married women, 2,762 women at risk of pregnancy are the focus of this analysis; 974 currently-married women were declared ineligible because they reported fecundity impairments or considered themselves physically unable to have more children. Among those at risk, 35.1 percent of the WFS respondents and 30.2 percent of the new respondents were using a conventional method of family planning at the time of the survey (Table 7.6). Table 7.6: Proportion Currently Using a Method of Contraception, by Method. Respondents Method Used All WFS New Pill 4.0 4.1 4.6 IUD 5.1 6.2 3.9 Condom 1.6 1.8 1.3 Subtotal 10.7 12.1 9.8 Douche 3.7 3.8 4.2 Withdrawal 9.9 10.9 9.5 Rhythm 5.0 5e9 4.4 Abstinence 2.2 2.4 2.3 Subtotal 20.8 23.0 20.4 Percent using 31.5 35.1 30.2 Percent not using 68.5 64.9 69.8 Total 100.0 100.0 100.0 (N) (2,762) (1,765)a (789)a -s 44 aDo not add up to 2,762 because 208 eligible respondents were not classified as either WFS or new respondents. - 256 - Overall, the proportion of women at risk using the more efficient conventional methods, namely the pill, IUD, or condom, ie low, both for the WFS and the new respondents. Only 12 percent of all WFS respondents, or 34 percent of current users of any method, were using an efficient method of birth control. The proportion among new respondents is even lower: approximately 10 percent, or 32 percent of current users. Withdrawal is the most popular among the conventional methods, used by 10.9 and 9.5 percent of the WFS and new respondents, respectively. The next in popularity is the IUD for WFS respondents and the pill for new respondents. The condom, used by 1.8 and 1.3 percent, respectively, of the WFS and new respondents, is the least popular method. Tables 7.7 through 7.15 give details on current use of contraception by selected socio-economic and demographic characteristics. The relationships between use of contraception with age and parity (Tables 7.7 and 7.8) take a typical inverted U shape, with relatively lower proportions of women in younger and older ages, and at lower and higher parities using a method of family planning. The pattern is not pronounced among the new respondents who are younger, and therefore concentrated at low parities. The fact that a higher proportion of the new respondents compared to the WFS respondents in the 20-24 year age group and among women of low parity are current users suggests a significant emphasis on child spacing on the part of these new respondents. However, when we examine the respondents on the basis of their family formation status (Table 7.9) more WFS than new respondents who have current parities less than desired parities are using a method of contraception, thus indicating an interest in chile spacing on the part of WFS respondents also. Among the respondents who had attained their - 257 - desired family size, slightly more new respondents (41 percent) than WFS respondents (39 percent) were currently using a birth control technique. Current use of contraception by religious denomination (Table 7.10) has the same distribution among both the WFS and the uew respondents. Buddhists have the highest proportion using a method (46 and 39 percent, respectively, of the WFS and new respondents), followed by Christians (43 and 29 percent, respectively), Muslims (27 and 28 percent), and Hindus (18 and 16 percent), in that order, Use of family planning is lowest among the Hindus (18 and 16 percent, respectively). Rural-urban differentials (Table. 7.11) are in the expected divection among both sets of respondents. Use of contraception is most frequent in urban areas and least frequent among women on the estates. Similarly, educational differentials in use (Table 7.12) are also in the expected direction, i.e., lowest among women with no education (23 and 19 percent, respectively) and highest among women with 10 or more years of education (51 and 42 percent, respectively), implying that for women, the opportunity cost of children is positively related to education. - 258 - Table 7.7: Percent of Women Currently Using a Method of Contraception, by Age. Respondents Age Group All WFS New 15-19 10.3 -- (18.2) 20-24 24.3 21.7 28.2 25-29 30.8 34.9 28.8 30-34 36.1 38.7 35.1 35-39 36.6 37.9 38.2 40-44 35.1 36.4 (33.3) 45-49 25.5 25.4 (35.7) Total 31.7 35.4 30.3 (N) (872) (623) (238) Note: () indicates fewer than 20 respondents. Table 7.8: Percent of Women Currently Using a Method of Contraception,. by Parity. Respondents Parity All WFS New 0 4.9 -- (4.0) 1 24.7 24.6 24.8 2 36.0 35.6 36.5 3 38.8 38.4 40.2 4 40.8 41.9 (33.3) 5 41.5 42.0 (38.7) 6 35.8 35.0 (42.1) 7 28.0 29.4 (15.4) 8 <;$. 7 31.7 (8.3) 9 + 18.8 19.6 (33.3). Total 31.7 35.4 30.3 (N) (872) (623) (238) Note: () indicates fewer than 20 respondents - 259 - Table 7.9: Percent of Women Currently Using a Method, by Family Formation Status. Respondents Fertility Status* All WFS New P less than D 22.9 27.2 25.3 P = D 39.4 39.0 41.1 P greater than D 39.3 40.6 (30.3) Total 31.7 35.4 30.3 (N) (872) (623) (238) *P = parity D = desired family size Note: () indicates fewer than 20 respondents. Table 7.10: Percent of Women Currently Using a Method of Contraception, by Religion. Respondents Religion All WFS New Buddhist 41.4 46.4 39.4 Hindu 15.8 17.8 15.8 Muslim 2.4.8 26.6 27.9 Christian 36.7 43.4 29.0 Others 37.5 25.0 (50.0) Total 31.7 35.4 30,3 (N) (872) (623) (238) Note: C) indicates fewer than 20 respondents. - 260 - Table 7.11: Percent of Women Currently Using a Method of Contraception, by Area of Residence. Respondents Place of Residence All WFS New Urban 37.4 39.4 40.5 Rural 32.2 36.4 29.4 Estate 13.8 15.9 13.6 Total 31.7 3504 30.3 (N) (872) (623) (238) Table 7.12: Percent of Women Currently Using a Method of Contraception, by Education. Respondents Level of Education All WFS Nev None 21.2 23.3 (19.4) 1-5 26.6 29.9 24.5 6-9 36.7 41.7 32.3 10+ 42.2 50.7 41A9 Total 31.7 35.4 30.3 (N) (872) (623) (238) Note: () indicates fewer than 20 respondents. - 261 - Contreceptive use is aiso positively associated with household expenditure (Table 7.13)06 As household expenditure increases, the proportion of women using a method also increases, from a low of 28 percent and 20 percent, respectively, for WFS and new respondents with a monthly expenditure less than 250 rupees, to a high of 44 percent and 41 percent, respectively, among women with a monthly expenditura of over 750 rupees. This pattern of acceptance seems to indicate that as household expenditure increases, children compete with other consumption goods for a share of the total expenditure. Current use of contraception by respondents' current employment status (Table 7.14) follows the expected pattern for the WFS respondents, but not for their new counterpa.ts. Among the former, 38 percent of the women who are currently warking were using a method, compared to 25 percent of those not working. Amoz-g the latter, more non-working women (31 percent) compared to working women (27 percent) were currently using a method of birth control. Table 7.16 shows that. contrary to the usual pattern contraceptive use is higher among farm workers than among non-farm workers and higher among those who work at (or near) home than among those who work away from home (see Birdsall, 1980:44-47). One possible explanation for this pattern of diffusion could be the family planning program itself. Until 1975, in spite of heightened family planning activity between 1972 and 1975, political support for the program was weak and acceptance of contraception was largely determined by the 6Information on HR was obtained by asking respondents "How much money per month does it cost your family to live including the cash value of any food you raise yourself?" - 262 - t socio-economic characteristics of the women. Since 1977, there has been a steady policy movement on the part of the new government (elected in 1977) v4 towards recognizing family planning as an essential component of population and health policies. Because enhanced family planning services and financial incentives have been provided by the state, sub-groups previously not disposed to fertility control now (1979) use a method of contraception. However, such an explanation leaves unanswered the question of why a similar pattern of diffusion did not take place ationg estate women as well as among respondents with no education. Besides the characteristics of the wives, husbands' education (Table 7.15) also has a positive effect on contraception. For both sets of respondents, the proportion using a method consistently increases with husband's education (from 27 to 48 percent for the WFS respondents and from 20 to 37 percent for the new respondents). It is interesting to note that current use of contraception is much higher among WFS respondents whose husbands have no education than among WFS respondents who themselves have no education. - 263 - Table-7.13: Percent of Women Currently Using a Method of Contraception, by Household Expenditure. Respondents HHE All WFS New 250 or less 23.5 7.8 20.4 250-349 27.5 31.6 24.6 350-449 29.1 32.7 27.5 450-549 32.5 36.3 30.5 550-749 34.8 37.6 35.2 750-4,500 40.0 43.6 41.1 Total 31.4 35.1 41.1 (N) (845) (604) (230) Table 7.14: Percent of Women Using a Method of Contraception, by Employment Status. Respondents Employment Status All WFS New Currently employed 32.5 37.6 27.3 Currently not employed 31.4 24.6 31.3 Total 31.7 35.4 30.3 (N) (872) (623) (238) Farm 38.2 40.9 (27.8) Off-farm 31.0 35.8 26.3 Total 31.7 35.4 30.3 (N) (872) (623) (238) Off-farm: Home 41.7 46.7 31.3 Off-farm: Away 28.4 32.5 25.7 Total 31.7 35.4 30.3 (N) (872) (623) (238) Note: For Tables 7.13 and 7.14, C) indicates fewer than 20 respondents. - 264 - Table 7.15: Percent of Women Currently Using a Method of Contraception, by Education of Husband. Respondents Level of Education All WFS New None 24.2 27.1 (20.3) 1-5 26.5 30.3 22.2 6-9 34.0 37.0 35.3 10+ 39.5 47.6 36.5 Total 31.7 35.4 30.3 (N) (872) (623) (238) Note: O) indicates fewer than 20 respondents. Broad conclusions drawn from the data presented above are summarized below: 1. The popular use of traditional, inefficient methods completely overshadows the use of efficient methods, even among the younger, more educated, new respondents. 2. The stronger and more consistent relationships between socio-economic variables (such as female education and household expenditure) and contraceptive use continue to hold for both sets of respondents. The relationship between current employment and use of contraception, namely, that use is higher among working women, does not hold for new respondents. Additionally, the data also reveal a reversal in the pattern of contraceptive use by nature and location of employment for all respondents. As noted above, while it can be hypothesized that this pattern of diffusion may be due largely to a change in the nature of the family planning - 265 - program and its service delivery, the selective pattern of the diffusion certainly warrants a further investigation. 3. Both WFS and new respondents appear to place emphasis on spacing of births. Among the latter, this desire for spacing is indicated by higher proportions of respondents using contraception at younger ages (20-24) and lower parities (2-3 children); among the former it is implied by a higher proportion of women with a current parity less than their desired family size using a method of birth control. The only difference between the WFS and new respondents is that, of those respondents who had attained the family size desired, slightly more new respondents than WFS respondents were using a method of birth control. Multivariate Analysis In this section, data are analyzed within a discriminant function procedure, which weights and linearly combines the variables on which the groups are expected to differ. Hence, it is not only possible to determine the extent to which the groups differ from one another, but also to estimate the discriminating power of each variable. The variables included in the analysis are defined in Table 7.16. Current use of contraception, because of its greater precision and more direct implications for fertility, is the dependent variable. In view of the cross-tabular results which showed that a larger proportion of women who work on farms or in the house (but no housework) are currently using a method of contraception, economic activity for this equation is represented by classifying respondents into two broad categories (those who have ever worked; those who have never worked). Excluded from the explanatory . ... .......... .... ..... 266 - Table 7.16: Description of Variables Included in the Analysis. Variable Code Description Dependent 1. Current use of contraception (1) Currently using (0) Currently not using Independent 2. Age In single years 3. Age at marriage In single years, 4. Children ever born Total number of children both dead and alive 5. Family formation status' (1) Parity less than desired (2) Parity greater than desired 6. Education of respondent Highest grade achieved 7. Labor force participation (1) Ever worked (0) Never worked 8. Monthly household expenditure In rupees 9. Education of husband Highest grade achieved 10. Religion2 (1) Hindus (2) Muslims (3) Christians 11. Area of residence3 (1) Urban (2) Estates Omitted Categories Parity equal to desired 2Buddhists 3Rural - 267 - variables are program-specific variables such as access to and availability of supplies and services of family planning, and the costs of contraception, for which no information was available. The omission of these variables subject the model to a large specificatLon error. The data were also analyzed separately to examine the relative importance of these s-rme explanatory variables in differentiating between users of more efficient and less efficient methods of contraception. Included in the former were couples using the pill, IUD or condom, while the latter encompassed couples using rhythm, withdrawal, abstinence, etc. No discernible pattern was obtained in the discriminant analysis, thus precluding a more substantive interpretation of the results. These are presented in Appendix 7B. The standardized and metric coefficients associated with each of the variables in the canonical discriminant function are presented in Table 7.17a,b. The former reflect the relative importance of each variable to the derived function; the latter facilitate the interpretation of the nominally scaled variables. The standardized coefficients are analogous to the partial betas in multiple regression analysis, lending themselves to similar interpretation. In Table 7.17a,b, a negative sign denotes a pull towards contraceptive use; a positive sign reflects a push towards non-use of contraception. The metric coefficients reported for nominally scaled - 268 - variables7 show the net effect relative to the grand mean of being in a particular category of a nominally-scaled variable. The metric coefficient for a continuous variable denotes the change in the probability of being a current user, attributable to a unit change in the explanatory variable. Based on the values and signs associated with the standardized and metric coefficients, the following conclusions emerge: 1. Religious affiliation is the most important predictor of contraceptive use among both the WFS and new respondents. Buddhists are the most likely and Hindus the least likely to use a method of contraception. Among new respondents, however, the difference between Muslims and Christians is considerably less than that among their WFS counterparts. 2. For both the WFS and new respondents, family formation status emerges as another important predictor of whether or not a couple is currently using a method of birth control. For both sets of respondents, those who had attained or exceeded their desired family size were more likely to be currently using a method of 7These were derived using the following formula: K = -[bipi] where K = deviation of omitted group from the grand mean. bi,= unstandardized beta coefficient for variable 1. Pi = proportion of cases in group i. The estimated intercept for the omitted category is: (bo + 7) where bo is the value of the intercept. For the ith group, the estimated intercept is: bo + (bi + K) - 269 - Table 7.17a: Standardized and Metric Canonical Discriminant Function Coefficients--WFS Respondents. Coefficient Variables1 Standardized Metric 1. Religion Hindu 0.720 -0.165 Muslim 0.356 -0.840 Christian 0.149 -1.299 Buddhists2 -- -1.831 2. Family formation status Parity greater than desired -0.247 -2.157 Parity less than desired 0.477 -0.396 Parity equal to desired3 -- 1.421 3. Respondent's education 0.349 -0.104 4. Age 0.236 0.036 5. Labor force participation -0.180 -- Ever worked -1.391 Never worked -- -1.035 6. Husband's education -0.120 -0.037 7. Children ever born 0.086 0.037 8. Monthly expenditure -me.084 -0.003 9. Area of residence Urban 0.005 -1.198 Estate 0.074 -0.950 Rural4 -- -1.211 10. Age at marriage -0.020 -- lVariables are given in order of importance. Omitted categories: ZBuddhists 3Parity = desired 4Rural - 270 - Table 7.17b: Standardized and Metric Canonical Discriminant Function Coefficients--New Respondents. Coefficient Variables' Standardized Metric 1. Religion Hindu 0.588 1.032 Muslim 0.240 0.470 Christian 0.255 0.472 Buddhist2 -- -0.299 2. Family formation status Parity greater than desired 0.122 0.214 Parity less than desired 0.443 0.557 Parity equal to desired3 - -0.382 3. Monthly expenditure -0.370 -0.001 4. Area of residence Urban -0.241 -0.179 Estate 0.107 0.673 Rural4 -- 0.364 5. Respondent's education -0.153 -0.044 6. Children ever born -0.152 -0.080 7. Husband's education -0.094 -0.016 8. Age -0.031 -0.005 9. Labor force participation 0.031 Ever worked -- 0.295 Never worked -- 0.232 10. Age at marriage -0.029 lVariables are given in order of importance. Omitted Categories: 2Buddhist 3Parity = Desired 4Rural Note: Only non-sterilized, fecund women were used in the denominator. - 271 - contraception than their counterparts who were still in the process of family formation. It is, however, interesting to note that whereas among the WFS respondents, those who had exceeded their desired parity were more likely to be currently usi-ng a method of contraception than those whose parity equalled their desired family size, among the new respondents the opposite pattern prevails. 3. As far as the relative importance of the remaining variables is concerned, there is less in common between the WFS and the new respondents. Among the former, the traditionally strong and more consistent influences on fertility regulation still exist. Respondent's education, age, and labor force participation have an appreciable effect in differentiating between use and non-use of contraception. The more educated a woman is, the more likely she is to use contraception, since education improves the likelihood that a woman has knowledge of and can use modern contraception; it also influences the speed with which a new method of birth control is accepted. For the WFS respondents, use of contraception decreases with age. This is not surprising since older women are less prone to use family planning. Interestingly enough, among the new respondents, although the relative importance of age in predicting contraceptive use or non-use is considerably less, the use of contraception increases with age. Finally, among the WFS respondents, women who have ever participated in the labor force are more likely to use contraception. Such women have probably developed extra-familial interests which tend to moderate their fertility goals. Among new respondents, not only is labor force - 272 - participation relatively unimportant in predicting contraceptive use, but the relationship is not in the expected direction. Consistent with the cross-tabular results, women who have never worked are more likely to be using a method of contraception. It can therefore be concluded that the variables which traditionally influenced fertility and contraceptive use in the absence of strong family planning programs and convenient delivery systems are becoming less salient for the new respondents, among whom income-related variables and economic calculations appear to be assuming a greater significance. 4. For the new respondents, monthly household expenditure is positively associated with contraceptive use, and emerges as the third most important predictor of contraceptive use. For these respondents, it seems that since parity is also a relatively important predictor of use of contraception (the number of children ever born is positively associated with contraceptive use) that children compete with other consumer goods. 5. Area of residence also emerges as an important predictor of contraceptive use, for the new respondents. Urban residents are more likely to use contraception, and estate residents the least likely. Area of residence may influence contraceptive use in one of two ways: (1) since family planning services are both easily available and more accessible in urban areas, the financial and psychological costs of using contraception are smaller; (2) having and raising children may be more costly in urban than in rural areas. These costs may range from purchasing health and - 273 - educational services to purchasing child care if both parents work. For the WFS respondents, area of residence is the least important predictor of contraceptive use; among them, rural women are the most likely to use contraception and estate women the least likely. 6. It is instructive to examine the signs associated with some of the other variables in the analysis which have relatively little predictive power. For the WFS respondents, the number of children is negatively associated with contraceptive use. This can only be explained by the fact that parity increases with age and older women are less likely to use family planning. Also, for both the WFS and the new respondents, husband's education and age at marriage are positively associated with contraceptive use. Selected statistical measures associated with discriminant function analysis are shown in Table 7.18. The value of the Wilk's lambda (.878 for the WFS respondents and .884 for the new respondents) and the associated chi square statistics indicat hat the discriminating power in the variables examined is small although statistically significant. The probability of obtaining chi square values of 223.0E (for WFS respondents) and 92.86 (for new respondents) with 14 degrees of freedom by chance is one in 10,000. The canonical correlation is a measure of association between the derived discriminant function and the dependent variable. The coefficient, when squared, can be interpreted as the proportion of variance in the response groups explained by the discriminating variables. The data show that 11.83 percent (for WFS respondents) and 11.45 percent (for new respondents) of the variance between users and non-users is explained by the predictors in the analysis. - 274 - Table 7.18: Eigenvector Summary and Canonical Correlation for the Derived Function Differentiating Between Users and Non-Users. Function 0 Function 1 WFS Respondents Eigenvalue .140 Canonical correlation -- .350 Wilk's lambda .878 -- Chi square* 223.08 -- New Respondents Eigenvalue -- .131 Canonical correlation -- .341 Wilk's lambda .884 -- Chi square* *92.86 -- *Degrees of Freedom = 14. One of the purposes to which the derived canonical discriminant function can be put is to classify ex post facto the study sample based solely on values of the discriminating variables. Table 7.19 examines the observed and predicted group membership of users and non-users of contraception. For both sets of respondents, the percent of couples accurately classified as users and non-users was approximately the same: 64.43 percent for the WFS and 65.78 percent for the new respondents. - 275 - Table 7.19: Distribution of Observed and Predicted Group Membership of Users and Non-Users. Predicted Observed Non-Users Users Total WFS Respondents Non-users 62.3 37.7 100.0 (708) (429) (1137) Users 31.6 68.4 100.0 (197) (426) (623) % of total correctly classified 64.43 New Respondents Non-users 66.1 33.9 100.0 (362) (186) (548) Users 34.9 65.1 100.0 (83) (155) (238) % of total correctly classified 65.78 Conclusion As has been noted, there have been changes in the provision of family planning services since 1975. The impact of these changes generally has been to make family planning both more attractive and available to groups previously not disposed to using contraception, thereby making less salient the socio-economic factors that traditionally influenced fertility and contraceptive use. According to the results of the multivariate analysis, among WFS respondents, rural women were now most likely to be -276- using a method of contraception.8 This is in contrast to the findings of the Sri Lanka Fertility Survey, which showed that urban women, or urban women who migrated to rural areas, were the most likely to use contraception. In addition, the results of the cross-tabular analysis showed that for all respondents, those working on farms or at home were more likely to be using a method of contraception. Such a pattern of diffusion, which is likely to have a beneficial effect on future fertility, may be attributed in large measure to the program effect. 8The cross-tabular results (Table 7.11) show slightly more urban women currently using a method of contraception. This discrepancy between the bivariate and multivariate results is due to the fact that, by definition, the former do not control for the effects of other variables. - 277 - APPENDIX TO CHAPTER VII 7A. Knowledge and Ever-Use of Contraception. Tables 7A.1 - 7A.7 7B. Multivariate Analysis of Type of Contraceptive Method Used. Tables 7B.1 - 7B.5 - 278 - APPENDIX 7A Knowledge and Ever-Use of Contraception Information on knowledge and ever-use of contraception was obtained by asking the following questions of all ever-married women: 1. Do yi-u know any methods which prevent or delay pregnancies? If the answer was affirmative, two further questions were asked: 2. What methods do you know? 3. Have you ever used (method)? Knowledge of Family Planning Methods: Eighty-five percent, or 4,091 ever-married women in the sample, without any prompting by the interviewer, knew of one or more methods of birth control, and 15 percent, or 738 women did not know of any method. After probing and describing the methods not mentioned, the number of women who knew, or had heard of, at least one method of birth control increased to 4,669 or 97 percent of the sample. This indicates a six percentage point increase in knowledge over 1975, when 91 percent of all ever-married women knew of one or more methods of birth control. On the average, an ever-married woman knew of 2.7 methods of birth control (Table 7A.l). Female sterilization was the best known method, with 92 percent of the ev'er-married women claiming to know or to have heard of it. This was followed by the pill, known to 88 percent of the respondents. The condom and male sterilization, mentioned by 54 percent of the respondents, were the least known scientific methods. Among traditional non-scientific methods, the best known, mentioned by 36 percent of ever-married women, was rhythm. This proportion indicates a decline over 1975, when 44 percent of the - 279 - Table 7A.1: Percent Distribution of Women by Number of Family Planning Methods Known Before Probing (All Respondents). No. of Methods Known Percent of Women 0 15.3 1 10.0 2 18.7 3 24.7 4 18.1 5+ 13.2 Total 100.0 (N) (4892) Average 2.7 ever-married women knew, or had heard of, the method. Abstinence, known or recognized by 21 percent of women, was the next best known of the non-scientific methods. Knowledge about family planning methods is highest among women 30-34 years, at which age over 98 percent of the ever-married women knew about family planning; it is lowest at the two extremes. The overall pattern of knowledge by age is thus in the shape of an inverted U (Table 7A.2). Years of schooling is positively associated with knowledge of family planning (Table 7A.3). As the level of education increases, there is an increase in the percent of women who know about family planning. Whereas 7 percent of the ever-married women with no education had never heard of any family planning methods, less than one percent with 10 or more years of schooling had never heard of family planning. - 280 - Table 7A.2: Percent Distribution of Ever-Married Women by Methods Known and Age (All Respondents). Method Known Never Sterilization Heard Age All of Any Group Pill IUD Condom Female Male Methods Method Total 15-49 77.2 44.1 40.7 79.5 30.1 88.1 11.6 145 20-24 83.6 57.9 53.1 88.4 48.2 94.5 5.5 542 25-29 89.1 69.1 64.1 93.2 56.3 97.9 2.1 871 30-34 91.7 75.5 62.1 94.8 61.6 98.4 1.6 960 35-39 89.1 73.0 54.9 95.3 58.5 97.8 2.2 81 40-44 88.8 68.7 49.0 91.2 54.5 96.2 3.8 731 45-49 85.1 56.8 39.1 88.8 44.8 95.3 4.7 699 Total 88.1 67.2 54.2 92.0 54.1 96.7 3.3 100 (N) (4,253) (3,247) (2,616) (4,444) (2,612) (4,669) (160) (4,829) Table 7A.3: Percent Distribution of Ever-Married Women by Method Known and Education (All Respondents). Method Known Never Sterilization Heard Educa- - All of Any tion Pill IUD Condom Female Male Methods Method Total None 75.2 43.9 28.8 87.8 39.0 92.9 7.1 884 1-5 87.5 63.3 46.4 90.6 45.1 96.5 3.5 1,843 6-9 92.0 78.1 67.6 94.9 61.9 97.9 2.1 1,363 10+ 97.3 85.0 79.2 96.2 80.1 99.5 0.5 739 Total 88.1 67.2 54.2 92.0 54.1 96.7 3.3 100.0 (N) (4,253) (3,247) (2,616) (4,444) (2,612) (4,669) (160) (4,829) - 281 In contrast to the other variables, variation of knowledge by parity has no consistent pattern (Table 7A.4). Knowledge of family planning is highest among women with 4 and 8 children, where less than one percent of the women did not know of even one method of birth control. Except for women with no children, among whom 13 percent did not know of family planning, over 95 percent of the women at all other parities knew about contraception. Ever-Use of Contraception: Thirty-six percent of the ever-married women stated that they had ever-used a method of contraception. Rhythm, used by 18 percent of the women was the most popular methcd, followed by sterilization and abstinence, both used by 13 percent of the women. Among the conventional efficient methods, the pill, used by 9 percent of ever-married women, is the most popular. Notwithstanding increases in knowledge of the more efficient methods and decreases in knowledge of the more efficient methods and decreases in knowledge of the less efficient traditional methods, the latter continue to play a more significant role in the attempts of the Sri Lankan women to regulate their fertility. Contraceptive use by age (Table 7A.5) exhibits the same pattern as knowledge of contraceptive methods by age: experience of use was lowest among the youngest age group (12 percent), relatively low among the oldest age group (26 percent) and highest among women aged 30-34 years (43 percent). Women below age 30 favored the conventional methods of family planning, of which the pill was the most popular: among women 30 years of age and older, on the other hand, the terminal method of sterilization dominated. - 282 - Table 7A.4: Percent Distribuition of Ever-Married Women by Method Known and Parity (All Respondents). Method Known Never Sterilization Heard All of Any Parity Pill IUD Condom Female Male Methods Method Total 0 74.1 40.7 45.2 79.4 40.7 86.9 13.1 398 1 85.7 64.4 57.5 90.2 53.2 95.5 4.5 671 2 90.9 73.3 63.1 94.5 61.9 98.4 1.6 745 3 90.2 73.0 61.8 92.8 62.5 96.8 3.2 696 4 92.3 75.0 59.9 93.8 57.7 99.2 0.8 613 5 90.0 72.5 51.2 94.4 54.0 98.0 2.0 498 6 88.7 71.9 48.8 92.6 52.3 97.5 2.5 363 7 84.0 62.7 41.5 91.8 48.4 95.8 4.2 306 8 91.5 63.5 51.2 96.2 48.3 99.1 0.9 211 9+ 89.0 58.2 35.4 93.9 40.9 98.5 1.5 328 Total 88.1 67.2 54.2 92.0 54.1 96.7 3.3 100 (N) (4,253) (3,247) (2,616) (4,444) (2,612) (4,669) (160) (4,829) Table 7A.5: Distribution of Ever-Users by Methods and Age (All Respondents). Method Ever Used Sterilization * All Never Age Pill IUD Condom Male & Female Methods Used Total 15-19 2.1 0.0 2.7 0.0 12.3 87.7 145 20-24 6.8 4.1 5.0 1.3 27.7 72.3 542 25-2'? 10.3 6.5 6.7 6.2 36.3 63.7 871 30-34 14.0 8.3 8.6 19.3 43.4 56.6 960 35-39 10.4 9.9 5.3 20.7 42.2 57.8 881 40-44 8.6 9.0 3.7 19.0 40.3 59.7 730 45-49 3.4 3.1 1.7 10.0 26.3 73.7 699 Total 9.2 6.9 5.3 13.2 36.3 63.7 4,829 (N) (443) (334) (258) (637) (1,751) (3,078) -- - 283 - Table 7A.6: Distribution of Ever-Users by Methods and Education (All Respondents). Method Ever Used Level of Sterilization Educa- All Never tion Pill IUD Condom Male & Female Methods Used Total None 4.9 5.0 1.1 10.4 23.4 76.6 884 1-5 8.0 7.4 3.8 12.8 31.0 69.0 1,843 6-9 11.7 7.6 6.8 16.5 43..5 56.5 1,363 10+ 12.6 6.8 11.5 11.5 51.4 48.6 739 Total 9.2 6.9 5.3 13.2 36.3 63.7 4,829 (N) (443) (334) (258) (637) (1,751) (3,078) -- Table 7A.7: Distribution of Ever-Users by Methods and Parity (All Respondents). Method Ever Used Sterilization All Never Parity Pill IUD Condom Male & Female Methods Used Total 0 1.0 0.0 1.8 0.0 5.8 94.2 398 1 4.3 1.9 5.7 0.1 28.9 71.1 671 2 11.4 7.7 7.9 3.2 43.1 56.9 745 3 10.8 9.1 8.0 13.4 46.6 53.4 696 4 12.7 10.3 6.9 19.7 45.4 54.6 613 5 11.6 9.2 4.6 25.3 40.2 59.8 498 6 10.2 8.3 3.6 24.5 38.0 62.0 363 7 10.1 10.5 3.3 23.2 33.7 66.3 3.06 8 9.5 7.1 1.9 21.3 33.2 66.8 211 9+ 7.9 4.6 1.8 20.4 30.5 69.5 328 Total 9.2 6.9 5.3 13.2 36.3 63.7 4,829 (N) (443) (334) (258) (637) (1.751) (3,078) -- - 284 - APPENDIX 7B Multivariate Analysis of Type of Contraceptive Method Used The results of the analysis focusing on factors differentiating the use of efficient (pill, IUD, and condom) and inefficient (douch, withdrawal, etc.) methods of contraception are presented in this appendix. The variables included in the analysis are defined in Table 7B.1. Table 7B.1: Description of Variables Included in the Analysis. Variable Code Description Dependent 1. Type of method used (1) Efficient method (0) Inefficient method Independent 2. Age In single years 3. Age at marriage In single years 4. Children ever born Total number of children, both dead and alive 5. Family formation statusl (1) Parity less than desired (2) Parity greater than desired 6. Education of respondent Highest grade passed 7. Labor force participation (1) Ever worked (0) Never worked 8. Monthly household expenditure In rupees 9. Education of husband Highest grade passed 10 Religion2 (1) Hindu (2) Muslim (3) Christian 11. Area of residence3 (1) Urban (2) Estates Omitted Categ-ories: Parity equal to desired 2Buddhist 3Rural 285 To put the analysis in perspective, the data are examined briefly to determine whether couples using efficient and inefficient methods differ with respect to any of these variables. The comparison of univariate differences is shown in Table 7B.2. In general, among WFS respondents, those using efficient methods are slightly younger (33.5 vs. 34.8 years) and have a smaller monthly household expenditure (Rs. 503 vs. Rs. 565) than their counterparts using inefficient methods. By contrast, among the new respondents, users of efficient methods are slightly older (29 vs. 28.8 years) and have higher monthly expenditures (Rs. 669 vs. Rs. 606). However, they have more children (2.7) than respondents using inefficient methods (2.3), and both the wives and their husbands are less educated than their counterparts using inefficient methods. Table 7B.2: Descriptive Statistics (Means) of Selected Variables Included in the Analysis. WFS Respondents New Respondents Variables Inefficient Efficient Inefficient Efficient Age (years) 34.8 33.5 28.8 29.0 Age at marriage (years) 20.4 20.1 21.9 20.8 Children ever born (mean) 4.1 4.0 2.3 2.7 Respondent's education (years) 6.0 6.0 7.2 6.2 Monthly expenditure (rupees) 565.0 503.5 606.1 669.3 Husband's education (years) 6.4 6.4 8e0 6.9 - 286 - Table 7B.3a: Standardized and Metric Canonical Discriminant Function Coefficients (WFS Respondents). Coefficients Variables' Standardized Metric 1. Monthly expenditure 0.636 0.002 2. Age 0.588 0.098 3. Religion Hindu -0.058 -4.236 Muslim 0.197 -3.457 Christian -0.340 -5.478 Buddhist2 -- -4.067 4. Family formation status Parity greater than desired 0.330 -3.356 Parity less than desired 0.017 -4.246 Parity equal to desired3 -- -4.285 5. Area of residence Urban -0.035 -4.261 Estate 0.317 2.563 Rural4 -- -4.182 6. Husband's education -0.240 -0.074 7. Children ever born -0.216 -0.106 8. Age at marriage 0.121 0.028 9. Labor force participation -0.091 -- Ever worked -- -4.060 Never worked -- -4.240 10. Respondent's education 0.011 0.003 lVariables are given in order of importance. Omitted Categories: 2Buddhist 3Parity = Desired 4Rural Note: Only non-sterilized, fecund women currently using a method of contraception were used in the denominator. - 287 - Table 7B.3b: Standardized and Metric Canonical Discriminant Function Coefficients (New Respondents). Coefficients Variables' Standardized Metric 1. Area of residence Urban 0.593 -1.758 Estate 0.199 -2.161 Rural2 - -2.987 2. Monthly expenditure -0.580 -0.001 3. Family formation status Parity greater than desired -0.157 -3.822 Parity less than desired 0.552 -1.931 Parity equal to desired3 -- -3.054 4. Age at marriage 0.392 0.087 5. Respondent's education 0.323 0.096 6. Religion Hindu 0.204 -1.828 Muslim -0.294 -3.380 Christian -0.146 -2.843 Buddhist4 -- -2.405 7. Labor force participation -0.233 Ever worked - -2.775 Never worked -2.295 8. Husband's education 0.226 0.033 9. Children ever born 0.178 0.990 10. Age -0.144 -0.021 lVariables are given in order of importance. Omitted Categories: Rural 3Parity= Desired 4Buddhist Note: Only non-sterilized, fecund women currently using a method of contraception were used in the denominator. - 288 - The results of the discriminant function analysis are presented in Table 7B.3. Based on the values and signs of the standardized coefficientsl, the following comments can be made: A. WFS Respondents: 1. Monthly household expenditure and age are the most important predictors of whether a woman is a user of an efficient or inefficient method of contraception. The data indicate that older women and couples in households with higher expenditures are more likely to use inefficient methods. Such a finding is not unusual since it is quite likely that older couples are more established and affluent, and older women are less receptive to technological advances and hence more prone to using inefficient methods. Among the new respondents, younger women are more likely to use efficient methods; however, the importance of age in influencing the kind of contraceptive used is negligible. 2. Contrary to the findings of the analysis on use/non-use of contraception, not only do respondent's education and labor force participation not have an appreciable effect among WFS couples on the type of method used, but more educated women are more likely to use inefficient methods of contraception. Of the variables that do make an appreciable contribution to discriminating users of efficient from users of inefficient methods, only area of residence shows the expected pattern. Among the WFS respondents, urban 1A negative sign indicates a pull towards use of an efficient method of contraception; a positive sign denotes a push towards an inefficient method. - 289 - residents are more likely than rural or estate residents to use an efficient method of contraception. 3. Religion and family formation status are the other variables that are important in differentiating the type of method used by the WFS respondents. Christians and Hindus are more likely than Buddhists to use efficient methods, whereas Muslims are least likely to use efficient methods. Also, women who have more children than they desire are least likely to use efficient methods of contraception. Such women are generally older and, as noted above, are less likely to use efficient methods. B. New Respondents: 1. Area of residence is the most important predictor of the type of method used among new respondents. Unlike the WFS respondents, women living in urban areas or estates are less likely than rural residents to use an efficient method of family planning. Like their WFS counterparts,. monthly household expenditure, family formation status and religion are important predictors of the type of method used, although the direction of the relationships differ. Among these new respondents, couples with higher household expenditures are more likely to use efficient methods of birth control. Women whose parity exceeds their desired family size are most likely to use an efficient method of family limitation, whereas women who have not yet achieved their desired family size are the least likely to use an efficient method of contraception. Such a pattern of use indicates a more consistent effort on the part of the new respondents to limit their fertility effectively. - 290 - Finally, Muslims and Christians are more likely than Buddhists to use efficient methods, and Hindus are most likely to use inefficient methods of birth control. 2. Contrary to the findings for the WFS respondents, age at marriage and respondent's education are important predictors of type of method used for the new respondents. However, like their WFS counterparts, the relationships are not in the expected direction: more educated women and those married at older ages are more likely to use an inefficient method of birth control. Statistics summarizing the results of the analysis are presented in Table 7B.4. For the WFS respondents, the value of the Wilk's lambda (.974) and the associated chi square statistics (15.56) indicate that the discriminating power of the variables examined is small and statistically insignificant. For the new respondents, the discriminating power of the variables is smaller than that for explaining use of contraception but is statistically significant at the .05 level. The data show that 10.56 percent of the variance between couples who use efficient and inefficient methods of birth control are explained by the variables in the analysis. Table 7B.5 examines the observed and predicted group membership of users of efficient and inefficient methods of contraception. Not unexpectedly, fewer WFS respondents (55 percent) than new respondents (64.71 percent) were classified accurately. - 291 - Table 7B.4: Eigenvector Summary and Canonical Correlation for the Derived Function Differentiating Between Users and Non-Users. Function 0 Function 1 WFS Respondents Eigenvalue - .027 Canonical correlation -- .161 Wilk's lambda .974 -- Chi square* 15.56 -- New Respondents Eigenvalue -- .118 Canonical correlation -- .325 Wilk's lambda .895 - Chi square* 24.48 *Degrees of freedom = 14. Table 7B.5: Distribution of Observed and Predicted Group Membership of Users and Non-Users. Observed Non-Users Users Total WFS Respondents Non-users 53.3 46.7 100l 0 (217) (190) (407) Users 41.8 58.2 100.0 (89) (124) (213) % of total correctly classified 55.00 New Respondents Non-users 66.5 33.5 100l 0 (107) (54) (161) Users 39.0 61.0 100.0 (30) (47) (77) % of total correctly classified 64.71 - 292 - Conclusion Although factors that made an appreciable contribution in predicting type of contraceptive used are the same for both the WFS and new respondents, the pattern is inconsistent and therefore not amenable to a more substantive interpretation. Future analysis focusing on type of method used would gain considerably by incorporating other factors such as program variables, actual costs incurred in obtaining various methods, and access to and availability of such services. - 293 - a VIII. SUMMARY, INTERPRETATION, AND CONCLUSION The Bare Facts In Sri Lanka, the demographic transition has advanced steadily, though unaccompanied by such hallmarks of the Western industrial complex as urbanization and high standard of li-,ing. Migration: Between 1871 and 1911, about 50 percent of the population growth was due to net migration. For the following two decades, this component accounted for less than 20 percent of the population growth. Since 1931, however, migration has not been a major factor in population change: between 1931 and 1953, about 5 percent of the population increase was due to net migration, and since 1953, the country experienced net outmigration, but never to any marked degree. Mortality: The expectation of life at birth was 32.7 years for males and 30.7 years for females in 1921. By 1971, these figures had climbed to 64 and 67 years, respectively. For females, the average gains in life expectancy at birth were 2.0, 0.7, and 0.5 years, respectively, during 1946-53, 1953-63, and 1963-71. The corresponding figures for males were 2.0, 0.5, and 0.15, respectively. Infant mortality declined from 141 per 1000 live births in 1946, to 101 by 1947, to 82 by 1950, and 48 by 1970. Other indices of mortality have shown similar dramatic declines. - 294 - Fertility: Up until the 1950s, no consistent decline in fertility seems to have occurred: Year 1900 1910 1920 1930 1940 1950 Crude birth rate: 39 39 37 39 36 40 The late 1950s, however, marked.a turning point, as can be seen from the trend in the total fertility rate (TFR)1: Year TFR Year TFR Year TFR 1952 5.2 1961 4.9 1971 4.2 1953 5.1 1962 4.9 1972 4.1 1954 4.7 1963 4.8 1973 3.9 1955 4.8 1964 5.0 1974 3e8 1965 4.8 1975 3.8 1956 5.0 1966 4.7 1976 3.8 1957 5.0 1967 4.6 1977 3.8 1958 4.9 1968 4.6 1978 3.9 1959 5.1 1969 4.4 1960 5.1 1970 4.3 Interestingly, after a period of consistent decline, TFR levelled off--in fact, reversed its trend slightly--in the latter half of the 1970s. Thus, two questions invite attention: (1) What accounted for the consistent decline in fertility in the 1960s and the first half of the 1970s? (2) Why did the pattern stall in the latter half of the 1970s? This A chapter deals chiefly with the first question. 'These figures have been calculated from the age-specific fertility rates presented in Department of Census and Statistics, 1982, Contraceptive Prevalence Survey, Table 5. - 295 - Conceptual Framework To interpret Sri Lanka's demographic experience, one may start with any one of a number of hypotheses recently proposed in the literature. To mention a few, Ronald Freedman (1979), along with many others, hypothesizes that marital fertility decline is a response to changing costs and benefits of children. He agrees with Kingsley Davis (1963) that changes in age at marriage and proportion marrying, limitation of fertility within marriage, and migration are among the possible responses to changing cost-benefit ratio of children, and points out that nuptiality changes have played a significant part in the fertility declines of the less-developed countries. Focusing on marital fertility decline, he hypothesizes that if people are exposed (through mass media or other channels) to ideas and models of ways of life prevailing elsewhere, motivation for lower fertility develops, because such exposure changes people's perception as to what is desirable and possible. Another situation which generates motivation for lower fertility is extreme Malthusian pressure on land, coupled with rising consumer and educational aspirations. Furthermore, Freedman believes that there are small subsets of developmental changes, any one of which can generate motivation for small families (although, what these subsets are nobody precisely knows).2 Freedman believes also that the dissemination of the mere idea that contraception within marriage is legitimate has a causal force of its own, determining when motivations for lower fertility are realized and at what rate. Put differently, it is possible for a 2"We don't know just how much change in which subset of conditions is sufficient to motivate fertility declines. Probably more than one combination will turn out to be sufficient" (Freedman, 1979:3). - 296 - noLcontracepting population to transform itself into a contracepting one and experience fertility decline at a rate that outpaces the rate of economic development--legitimation of the concept of contraception within mnrriage being an innovation in itself and, like other innovations, having its own dynamics of diffusion. Parallel ideas have been suggested by others. Thus, Caldwell (1976) hypothesizes that fertility decline starts if, and only if, the intergenerational "wealth" flow reverses its traditional direction, which is from the younger to the older generation, implying that if, in the perceptions of parents, the cost of children outweighs the corresponding expected benefits, fertility begins to decline. Caldwell believes that Westernization (i.e., Western domination over schools and mass media) causes parents to spend more on their children than they ever expect to receive in return. He emphasizes that non-Western societies cannot escape WesternizationD, because the industrialized West is so powerful that it can create demand for whatever it exports, and the habit of making children a net liability is one such social export from the West. A closely related notion is that when the nation-state assumes many of the basic welfare and protection functions traditionally fulfilled by children, a larger family becomes unnecessary, and people, therefore, adopt a small-family orientation. Similarly, Mead Cain (1981) hypothesizes that people in less-developed countries have children because children are an insurance against possible cut-off or drastic curtailment of sustenance flow and that when children lose this insurance value--as, for example, when the nation-state begins providing the needed insurance--a small-family pattern develops, depending on the balance of other factors. - 297 - Yet another notion is that only when class and gender inequalities lessen will a significant fertility decline occur (Safilios-Rothschild, 1982). Clues regarding the emergence of factors such as those mentioned above may be found in Sri Lanka's history. The next section treats long-term developmental changes,and the section following it the relatively recent implementation of family planning programs. Developmental Change in Sri Lanka The story of Sri Lankan development is one of nearly 450 years of foreign tutelage, followed by independence in 1948 and the fostering, in the post-independence period, of a social welfare policy aimed at improving the lot of the poor and underprivileged. The first foreign power to dominate Sri Lanka were the Portuguese, who arrived in 1505, determined to capture the lucrative cinnamon trade. Soon, rivalries between local rulers drew them into internal politics, and by the end of the sixteenth century, the Portuguese ruled the entire island, with the exception of the Kingdom of Jaffna in the north and the Kandyan Kingdom in the interior highlands. Jaffna fell to the Portuguese in 1619, but the Kandyan rulers maintained the independence of their kingdom until the arrival of the British. The Portuguese were succeeded in 1658 by the Dutch. The descendants of these two waves bf early European settlers are the island's Burghers, a small but important minority. The period of British involvement in Sri Lankan history began with the capitulation of the Dutch settlements to the British East India Company in 1796. The Kandyan Kingdom, which had remained independent throughout Portuguese and Dutch rule finally acceded to Western colonial rule in 1815. - 298 - The period that followed was one of consolidation of British rule, as British economic and cultural influence penetrated the length and breadth of the island. The island's long exposure to foreign rule had the important consequence that Buddhism, the traditional dominant religion of Sri Lankans, lost the state patronage it had enjoyed for centuries. Under the Portuguese and the Dutch the state machinery was actively used against Buddhism in support of Christianity (Malalgoda, 1976:28); and although the British subscribed in theory to the notion of separation of religion and state, in practice their sympathy was with the proselytizing Christian missionaries. During the latter half of the nineteenth century, however, there occurred a "Buddhist revival," in which foieign Theosophists played a part by helping Sri Lankan Buddhists combat Christian proselytizing. Subsequently, a militant Buddhism--aptly called "Protestant Buddhism" by Obeyesekere (1970)--took shape and has continued to exist in one form or another to this day. Also of lasting significance during the later period of British rule was the implementation of a far-reaching program of education and literacy. Before the Western incursions, the practice was for village temples to educate children in reading, writing, and religious precepts. The Portuguese and the Dutch set up a few post-elementary seminaries and a number of elementary schools here and there, but it was under British rule that the traditional temple schools atrophied while a new school system flourished in all but a few pockets of cultural resistance (Jayaweera, 1979). Established and operated for much of the nineteenth century by - 299 - missionaries, the British system became highly centralized, and was geared to creating a Westernized "elite" with which to fill local administrative jobs. Few in number, initially the schools were attended exclusively by children from upper-caste families. In due course, however, they enhanced social and economic mobility as a result of the linkage established between educational attainment and employment opportunities.3 But, while the system helped members of the lower castes4 to rise in power and prestige (Fernando, 1979:32), it created and perpetuated a two-class structure, consisting of a privileged minority and an underprivileged majority (Jayaweera, 1979:133-34). A number of changes were introduced in the colonial education system during the second quarter of the twentieth century. The introduction of universal suffrage and a semi-representative form of government early in the 1930s made these changes inevitable. Greater accese to secondary education was created in 1940 with the opening of 44 new secondary schools in rural areas, and in the late 40s with the introduction of a policy of 3Until British rule, the social status of a person in Sri Lanka was determined almost exclusively by caste and kinship (Fernaando, 1979:24). Historically, the distinguishing feature of caste in Sri Lanka was its linkage with Sinhalese feudalism. A caste was an occupation with feudal obligations to the king and the feudal lords. The traditional elite were from the Goyigama caste. Initially, it was the Goyigama families who had access to Western education. But in due course, an increasing number of non-Goyigama children began to attend English schoole. Many members of the lower castes rose in affluence as time went by (Ludowyk, 1966:27). 4English being the language of the administration, the restriction of the availability of an English education meant creating a privileged class (those who possessed the scarce background). Thus, the English educated became the new elite, their power and prestige being determined by achieved characteristics (education, mainly) rather than ascribed ones (caste and kinship). - 300 - free education.5 The number of schools providing primary and secondary level education increased substantially in the 1950s. In 1960 there were close to 7,000 schools in the country, a substantial increase from the figure of less than 5,000 on the eve of independence. The medium of instruction in secondary schools had already reverted to the mother tongue by the 1950s. Beginning in 1959, English was abandoned as the medium of instruction at the university level, and the university entrance examination has been administered thenceforth in the indigenous languages. One of the important results of such changes was a rapid increase in school enrollment. In 1947, almost 60 percent of children in the age group 5-14 years were enrolled in schools; the corresponding figure in 1953 was 71 percent, and in 1963, 75 percent. In 1975, 77 pe-cent of children in the age group 5-9 years and 65 percent of children in the age group 10-14 years were enrolled in schools (Jayaweera, 1979:139). Thus, formal education had come within reach of the masses, and most children availed themselves of the opportunity. The educational attainment statistics tell the story (Table 8.1). 5Those English secondary schools which had levied fees were given the option of either receiving government grants and abolishing fees (except a small facilities fee) or continuing as unaided fee-levying schools. A vast majority opted to join the "free education" scheme. Its significance was perhaps more in its implications for the future than in its being a revolutionary measure at the time it was introduced. (Later on, fees at the University of Ceylon [established in 1962] were also abolished.) - 301 - Table 8.1: Educational Attainment of Recent Birth Cohorts. 1961-1966 1956-1961 1951-1956 Birth Cohort Birth Cohort Birth Cohort (Age 15-19 (Age 20-24 (Age 25-29 Level of in 1981) in 1981) in 1981) Education Completed Male Female Male Female Male Female None 7.1 8.5 6.0 8.5 5.4 8.9 Grades 1-4 11.0 9.6 10.9 9.1 10.5 9.2 Grades 5-9 62.3 58.3 54.8 51.2 55.6 52.5 Higher 18.8 23.0 27.8 30.6 28.1 28.4 Not known 0.8 0.6 0.5 0.6 0.4 1.0 Source: Puvanarajan (1982). There is also abundant evidence to indicate that the educational gap between males and females and between rural and urban residents has been fast closing (Jayaweera, 1979:142) and that economic and rural background no longer prevent anyone from obtaining a college education. (A survey of the origins of students who entered college in 1976 showed that 40 percent came from a low-income, working-class background, and 45 percent from rural areas [Gunaratne et al., 1977])). Like education, Sri Lanka's economic history is inseparable from its colonial heritage. At the time of independence (1948), the country inherited from Britain a dual economy in which plantation agriculture (tea, rubber, coconut) provided most of the country's export income, while a semi-subsistence rural economy based on traditional peasant agriculture existed virtually in isolation from the plantation sector. In the - 302 - post-independence period, several development policies were adopted to cope with the undiversified export economy and the heavy dependence on food import necessitated by decades of neglect of the domestic peasant agriculture. Unfortunately, adverse trends in external trade, among other things, seriously undermined the country's developmental efforts. Between 1960 and 1970, Sri Lanka's Gross National Product (GNP) at constant (1959) prices increased at an average annual rate of 4.4 percent, while the per capita GNP only increased at an average annual rate of 2 percent. Economic growth slowed down further during the first three quarters of the 1970s: between 1970 and 1977, GNP increased at an average annual rate of only 3.0 percent, while the corresponding growth rate in the per capita GNP was only 1.5 percent. The picture brightened after 1977: between 1977 and 1981, GNP at constant (1959) prices increased at an average annual rate of 5.9 percent, and per capita GNP, at the rate of 4.1 percent (Central Bank of Ceylon figures reproduced in Abeysekera, 1982). The relative share of GNP contributed by agriculture, forestry, and fishing has declined over the years, though together they still constitute the largest sector of the economy, accounting for a third of the country's GNP. Also, agriculture provides employment for 50 percent of the country's labor force, a figure which has remained virtually unchanged over the years (Balakrishnan, 1979), as can be seen from the figures in Table 8.2. - 303 - Table 8.2: Industrial Composition of the Labor Force, 1953 and 1973. Percent of the Labor Force Industry 1953 1971 Agriculture 52.9 50.4 Plantations 28,6 20.1 Other 24.3 30.3 Industry 12.1 13.4 Manufacturing 9.6 9.6 Services 28.3 28.0 Other 6.6 8.2 Source: Department of Census and Statistics, 1976, Population of Sri Lanka (Colombo). Since independence, the domestic farm sector has been given a great deal of attention in the country's developmental policies and programs. Between 1946 and 1962 (two consecutive agricultural census years), the land under cultivation increased by 400,000 acres (Department of Census and Statistics, Census of Agriculture, 1962, vol. 1, p. 29). The acreage under paddy, more than a third of the total area under cultivation today, has increased considerably in the post-independence period--from 964,484 acres to 1,448,403 acres between 1952 and 1972 (Balakrishnan, 1979:111). In addition to making new lands available for cultivation by means of irrigation, the government stepped in to provide subsidies for various inputs such as fertilizer and seeds, to arrange credit and marketing, and to provide price support (Balakrishnan, 1979:112). It should be noted that the opening of new lands for cultivation was chiefly confined to the dry zone, there being little new land available for cultivation in the wet zone which - 304 - accommodates the plantation crops and most of the urban centers in the southwest portion of the island. Mention must be made, in this connection, of two land reform legislations. The Paddy Land Act of 1958 and subsequent amendments to it were aimed at ameliorating the plight of the tenant cultivatorsG The act declared that tenancy rights were to be permanent, transferable and heritable. It also fixed the maximum payable rent, and put a ceiling on the interest (rent) chargeable to monetary (non-monetary) loans accorded tenant cultivators. Opinions vary as to whether this act had its desired impact. Evictions became common in the wake of the act: between 1958 and 1967 the number of eviction complaints filed by tenants approached 23,000; in the next five years, 12,400 more complaints were registered (Wickremeratne, 1977:249-50). Most tenants, however, continued to pay 50 percent or more of their produce ac rent (Agrarian Research Training Institute, 1974), perhaps owing, in part, to the fact that in many cases the tenants were friends or relatives of the landlords, and were not inclined to invoke the law, and in part to the perception that the government was lethargic in implementing the act, The government did not seek to remedy the tenurial problems by vigorously enforcing the principle of "land to the tiller." Instead, emphasis was laid on providing credit facilities, subsidizing fertilizers, arranging for marketing the produce, and similar measures. The second land reform legislation consisted of the Land Reform Law of 1972, which put a ceiling on private ownership of land, and the Land Reform (Amendment) Law of 1975, which resulted in the nationalization of the estates. In the ensuing alienation of lands under these laws, considerable preference was given to the ownership and management by state agencies and 305 - cooperative units, in conformity with the basic policy of promoting collective ownership and control. The 1972 legislation did not affect paddy landholdings, most of which were much below the ceiling etablished by the law. In fact, according to one estimate, 85 percent of all paddy landholdings are smaller than two acres--60 percent smaller than one acre (Wilson, 1979, citing Report of Land Utilisation Committee, paragraph 52). Many such small holdings are probably non-viable. The practice of distributing the land equally among several sons might have been responsible for increasing the number of small, non-viable holdings. In the traditional agricultural setting, such fragmentation would have caused few problems, as the sons would have continued to cultivate their lands collectively. But with increasing acceptance of the concept of individual property ownership and rights, backed by institutional arrangements which define and safeguard such rights, the practice of brothers jointly cultivating their lands has gone out of fashion, and consequently the adverse effect of fragmentation has begun to assert itself (Wickremeratne, 1977:252). Another way of saying the same thing is that the population pressure on land has become increasingly heavy in recent years, even with the net increase in the area of land under cultivation. For, while the area of cultivated land increased by 0.7 percent per year between 1946 and 1969, the rural population increased at the rate of 2.5 percent per year, thereby increasing the density of population per acre of cultivated land steadily from 1.34 in 1946 to 2.02 by 1969 (Jones and Selvaratnam, 1972:193). A parallel feature has been the aggravation of the unemployment problem in the country as a whole. Different sample surveys in the 1960s - 306 - put the unemployment rate between 10 and 14 percent of the labor force. The Socio-Economic Survey of 1969-70 (Department of Census and Statistics, 1970) reported an estimated unemployment rate of 14 percent. A more recent survey put it at 17 percent (Central Bank of Ceylon, 1975:143), while a subsequent survey found that the rate had gone up to 20 percent by the mid-1970s (Central Bank of Ceylon, Bulletin, April 1976:270). According to one estimate, the unemployment rate in 1973 was 24 percent (Abeysekera, 1982). Most of the unemployed are educated young men and women. Thus, according to the 1969-70 Soc'io-Economic Survey, 85 percent of the unemployed were of age 15-24 years, and of these about 75 percent had completed at least middle school. It seems that "the more a young person had been educated, the greater the likelihood that he or she would be unemployed" (International Labor Office, 1971:21). The census data show the following trend in the number of persons employed and the number of persons in the age grcup 15-24 years (the latter representing roughly the number entering employment age): Total in the Age Group Total Employed 15-24 Years Index Index Number 1946 Number 1946 Year ('000) -1000 ('000) =1000 1946 2$611 1,000 1,322 1,000 1953 2,993 1,123 1,471 1,113 1963 3,200 1,226 1,907 1,443 1971 3,622 1,387 2,607 1,972 These figures indicate that total employment has not been increasing at the same rate as the new entrants to employment age. - 307 - From the census data we can discern also the following pattern for the ratio of the number of persons of age 15-24 years to the number of persons of age 25-64 years: Year (15-24/25-64) 1946 0.50 1953 0.47 1963 0.50 1971 0.57 These figures indicate that in the 1960s as compared to the 1950s each young adult was sharing the family resources with relatively more brothers and sisters. In all discussions of the developmental efforts in Sri Lanka,one finds a great deal of attention paid to the social welfare strategy adopted by successive governments. Between 30 and 40 percent of government expenditures and a considerable sum of foreign exchange are spent on providing subsidized food. This involves a guaranteed price for paddies harvested domestically and a weekly ration of rice, which, though varying in price and quantity from time to time, has generally been made available at a cost well below that on the open market. Thus, during the first half of the 1960s, the weekly ration per adult person was four pounds of rice at half the prevailing world market price (Wilson, 1979:55).6 6This welfare measure "absorbed nearly one-fourth of all revenue, blocked public savings for public investment, increased the use of foreign exchange for imports of rice and discouraged domestic production of substitutes for rice because rice was available at cheaper rates than those other substitutes" (Wilson, 1979:55). - 308 - Medical services were imparted free of charge (in government medical institutions) until 1971, since which time a very nominal charge has been imposed. Since 1945, three years before the country attained independence, Sri Lanka has had a universal free education policy. As already mentioned, formal education has come increasingly within reach of every boy and girl. According to one estimate, social benefits received via education, heaith, and the food subsidy as a percentage of taxes paid was, in the mid-1960s, 19.4 percent for the poorest and below 10 percent for the highest income groups (Taxation Inquiry Commission, 1968). The government's land development program, aimed at transfering population from the overpopulated regions of the country to newly opened lands, also come within the framework of government welfare policies, because of heavy subsidies given to colonists and of the public expenditure involved in developing such lands. Close to 650,000 allotments have been made, totaling about 1.75 million acres of land, under such programs. Again, opinions vary as to whether these measures have had the desired results (Wilson, 1979:66-72). One other factor worth mentioning is that the nation succeeded in achieving greater equality in income distribution between 1953 and 1973. The Gini concentration ratio, which is a commonly used measure of income inequality, declined from 0.50 in 1953 to 0.45 in 1963 and to 0.35 in 1973. However, the figure has since climbed to 0.49 (as of the late 1970s). Finally, there is evidence indicating that gender inequality is decreasing. Sex differentials in educational attainment and school enrollment are very narrow or are narrowing fast. Thus, of the medical - 309 - students enrolled in 1976, 48.5 percent were females, a marked improvement from 29.3 percent in 1966 (Jayaweera, 1979:172). A 1973 study of vocational preferences of secondary school female students showed that they aspired to be doctors, teachers, lawyers, accountants, nurses, clerks, air hostesses, and engineers (Jayaweera, 1973). While such data on preferences may be of questionable validity, they indicate that young women are at least aware of vocations other than homemaking. In fact, the labor-force participation rate of women in the age group 20-44 years, which covers most of the reproductive age span, increased from 27.1 percent in 1963 to 37.1 percent in 1971 (cited in Puvanarajan, 1982:9). In summary, Sri Lanka is an agrarian, low-income country (about 50 percent of the labor force is engaged in agriculture, and the GNP per capita as of 1981, at constant, 1959 prices was only Rs 1,014); Malthusian pressure on land is evident, particularly in the wet zone; unemployment is very high, even among the educated; the country has a well-organized system of education, in fact, Sri Lankans are better educated than most LDC populations; the country is linked to others far and near through communication and transportation networks; Western ideas and life styles have influenced the island for centuries (it is not unusual to see Western attire, soda pop, radios, and the like in all population centers); direct and indirect taxes and other income transfer policies have served to redistribute income; the nation-state serves as a source of insurance against penury, and as a source of support for the ill and the aged--roles traditionally filled by one's children; and, finally, gender and class inequalities have lessened over the years. 310 - Family Planning Programs Although family planning services had been made available in Sri Lanka through the Family Planning Association (affiliated with the 74 International Planned Parenthood) since its inception in 19535 it was not until 1965 that the government declared family planning an integral part of the Department of Health Services. A Family Planning Bureau (now known as the Family Health Bureau) was created in 1968 within the Ministry of Health (now the Ministry of Family Health) to coordinate and evaluate the program. While the initial emphasis was on intra-uterine devices (IUDs), a shift toward the pill (oral contraceptives) occurred later, after which sterilization became the favored method. Family planning activities received external assistance first, beginning in 1968 from Sweden, and later, beginning in 1973 from the United Nations Fund for Population Activities (UNFPA).8 7Early history of the family planning movement in the couatry is available in Abhayaratne and Jayewardene (1968). 8The UNFPA support covered many activities, all having to do with the spread of family planning in the country; these involved (Immerwahr, 1981): i) a program for strengthening the nursing and midwifery education in family planning services; ii) programs for populations and family planning education through unions and employers, in urban areas and estates; iii) population education in schools; iv) teaching of family planning in medical schools; v) a communication strategy program to promote family plaanning through *the media; vi) provision for surgical facilities for sterilization. - 311 - The Family Health Bureau's (1982) statistics (see Table 8.3) indicate that the number of sterilizations plus new acceptors of the pill and IUD increased between 1968 and 1975, declined thereafter, reaching a low in 1977, picked up again--reaching an unprecedented high in 1980--then dropped again the following year. The accuracy of the new acceptor counts published by the Bureau is questionable. Thus, Immerwahr (1981:21) points out that the number of sterilizations are underreported, while the number of new acceptors of the pill and IUD are overreported.9 If we assume that the time trend in such errors has remained more or less constant, then we may infer from the figures in Table 8.3 that there was indeed a trough in governmental family planning activities in the third quarter of the 1970s. (Interestingly, this coincided with the levelling off of the total fertility rate.) The new government, which came to power in 1977, has given high priority to population activities and accordingly reactivated the family planning program. Incentives for new acceptors of sterilization were substantially increased, as were emoluments to-members of the medical team who performed the operations. The upturn in the figures in Table 8.3 during the fourth quarter of the 1970s reflects this renewed activity. 9Immerwahr (1981) found that the number of sterilizations, estimated from the World Fertility Survey (Sri Lanka), 1975, data far exceeded the corresponding figures available in published official statistics. He mentions also the possibility that it was quite likely for a woman to be counted more than once as a new acceptor if she discontinued, say, the pill for a while and then started on it again, particularly so if there was a new midwife at the clinic (a common occurrence) when use of the pill was resumed. - 312 - Table 8.3: New Acceptors of the Pill, Loops or Sterilization. Year The Pill Loops Sterilizations 1966 1,000 10,000 3,000 1967 8,982 18,506 3,616 1968 16,014 20,615 1,011 1969 25,284 19,537 2,947 1970 26,889 15,799 4,971 1971 25,828 11,446 4,335 1972 32,300 18,599 9,576 1973 34,214 27,528 20,248 1974 35,924 29,693 42,234 1975 37,720 32,755 39,164 1976 25,597 27,030 35,588 1977 27,514 21,321 19,056 1978 31,146 23,085 21,949 1979 30,394 20,187 35,643 1980 29,296 19,232 112,926 1981 21,478 13,582 65,490 Source: Family Health Bureau, 1982. The legitimacy of contraception within marriage is now widely accepted by Sri Lankans. In the 1975 Surve- (WFS), of the 1,245 women who said they wanted no more children but had never used contraception, only 263 (21.1 percent) said that they or their husbands disapproved ef contraception within marriage. In the 1982 Contraceptive Prevalence Survey, of the 1,418 currently married women who were not pregnant at the time of the survey but who were not using any contraception, only 2 percent mentioned "disapproval of family planning" as their reason for not contracepting (Department of Census and Statistics, 1262: Table 5.15). In an earlier nationwide survey (conducted in the mid-1960s), Abhayaratne and Jayewardene (1967) found that - 313 - 53 percent of the respondents did not know any family planning method and that, of those who knew at least one method, 29.2 percent disapproved family planning--mostly on religious grounds (Table 52, page 281). This earlier study is not strictly comparable to the 1975 WFS or the 1982 Contraceptive Prevalence Survey. Nevertheless, one can easily infer from the three surveys, that since the mid-1960s considerable changes have taken place in Sri Lankans' attitudes toward contraceptive practice within marriage. An Interpretation We are now ready to suggest an interpretation for the fertility change in Sri Lanka in the 1960s and 1970s. Mortality started to drop precipitously in the mid-1940s, resulting in a sudden improvement in the rate of survival of infants to adulthood. School enrollment also improved beginning in the mid-1940s. A combined result of these developments was that in the early 1960s there were relatively more young people, many with some formal education, sharing their family resources and seeking work in the job market. Unfortunately, jobs that matched the aspirations of the educated were not easy to find, as the economy became more or less stagnant in the late 1950s.. Moreover, land had already become scarce in the wet zone, and government's attempts to absorb young people into the land development programs in the dry zone were not succeeding.'0 100ne such plan was put into operation in 1965. This involved alloting land in the dry zone to young people who had completed high school or finished their training in the farm school. The plan did not succeed because of a lack of "planning, supervision, and financial inducement" (Wilson, 1979:68). - 314 - As a result of decreasing employment opportunities in the non-farm sector and a low rate of absorption in the farm sector, the age at marriage of males went up because families were presumably guided by the rule that they should not let a son get married before he is assured of a more or less permanent means of support. The singulate mean age at marriage for males, which was constant in the 1940s and early 1950s (about 27 years), rose to 28 years by 1963. The female age at marriage followed suit, the gap between the bride's and the groom's age being relatively more resistant to change. The singulate mean age at marriage of females, more or less constant in the 1940s and early 1950s at 21 years, climbed to 22.1 years by 1963, almost parallel to the rise in age at marriage of males. Meanwhile, increasingly larger proportions of young women began attending school and entering the labor-force (Puvanarajan,1982:9). One of the results of these changes has been that, between 1963 and 1975, the singulate mean age at marriage of females rose from 22.1 years to 24.8 years (while the corresponding figures for males changed relatively little): 1963 1971 1975 Males 28.0 27.9 28.4 Females 22.1 23.6 24.8 Thus, in the 1960s and early 1970s, the dependency period of children became protracted: (1) children were increasingly enrolled in school well into their teens; (2) the waiting period for employment after completion of school was lengthening; and (3) female's age at marriage was on the rise. A response to this changing situation was fertility limitation within marriage. Parents were presumably guided by the necessity of adjusting reproduction to, among other things, the net cost of children as - 315 - reflected in their duration of dependency,11 i.e., the length of time children were a drain on the family resources on a net basis. Obviously, the rate at which marital fertility declined depended largely on the family planning efforts expended in the country. Marital fertility fell at a steeper rate during periods of intense family planning program activities and at a slower pace when there was a lull in those activities. The Evidence As already mentioned, the total fertility rate (TFR) remained more or less constant until the beginning of the 1960s, declining consistently until the middle of the 1970s, and thereafter levelling off and even turning slightly upward. Many scholars have examined the demographic components of these changes. Thus, Fernando (1972), using data from the 1963 census, the 1969-70 Socio-Economic Survey, and the registration data, concludes that almost all of the changes in fertility between 1963 and 1969 were due to rising age at marriage. In a 1976 study, however, he expresses the opinion that there were indeed shifts in marital fertility rates in the 1960s. Alam and Cleland (1981), using data from WFS, 1975, estimate that 59 percent of the TFR decline between 1963 and 1971 was due to nuptiality changes--the rest to marital fertility changes. From the official statistics (Census and Registration data), the corresponding estimates are 54 and 45 percent, respectively, as can be verified using the figures reproduced in the First Report of WFS, 1975 (Department of Census and Statistics, 1978). '1The duration of dependency of a child, that is, the length of time the child is a dependent, on a net basis, on the family resources, is as good an index as any of the net cost of the child, when education, medical services, etc., are free (for all practical purposes),as has been 'the case in Sri Lanka. - 316 - As for the period 1971-75, Alam and Cleland (1981) estimate that only 46 percent of the decline in TFR during this period was due to the nuptiality factor and 54 percent due to marital. fertility. As they put it, "Sri Lanka has followed the typical Asian pattern of fertility decline. Initially, changes in total fertility were caused largely or exclusively by rising age at marriage; this was followed by a period in which the effects of nuptiality and marital fertility were about equal, and finally the contribution of marital fertility became more important" (Alam and Cleland, 1981 :42). The question naturally arises as to why some persons marry at a younger age than others. First, one should recognize that arranged marriages are the rule rather than the exception in Asian countries. It is true that in recent times the prospective bride and groom have been granted some say in mate selection and related matters. Nevertheless, arranged marriage is the prototype for any discussion of age at marriage in Asian countries. We shall assume that it is the parents who arrange the marriages for their sons and daughters, although in particular instances, depending upon the circumstances, other relatives may substitute for the parents. The rule generally followed by parents is: marry off a daughter as soon as she reaches marriageable age, unless there is a socially-acceptable reason for a delay; as for a son, arrange a marriage for him, but not before he is assured of a means of livelihood. To elaborate, the family is under-tremendous social pressure to marry off a daughter as soon as she is of marriageable age, because the continued presence of her in the household, unwed and unspoken for, is likely to be interpreted as a breech of responsibility on the part of the - 317 - family members--particularly the parents--or as an indication of an affliction of some sort in the girl. The situation is less likely to be misinterpreted by society at large if the girl is pursuing higher education or is in the labor force. Moreover, when formal education is perceived as a step toward securing a job, parents usually find it advantageous to send their daughter to school, especially if the cost involved is negligible. The advantage is that the lifetime income accruing from a regular job can be counted as part of the dowry, thereby reducing the heavy burden on the parents. As for a son, from the parents' point of view, it makes little sense to arrange a marriage for him when he is still a dependent, for to do so amounts to bringing home yet another dependent (the daughter-in-law). Of course, parents may view the situation differently if they are very wealthy or if there are prospects for upward mobility resulting from an alliance. (These comments apply to the arrangement by which the wife becomes a member of the husband's family after marriage, the predominant pattern in Sri Lanka.) Apart from the general rule mentioned above, there may be cultural factors that determine when a man or woman attains marriageable age. Religious and ethnic background represent such cultural factors. Age at Marriage: With these general ideas in mind, let us compare selected districts in Sri Lanka to see how particular characteristics of territorial communities affect age at marriage. We shall use for this purpose the singulate mean age at marriage (SMAM) for females for the years 1963, 1971, and 1981, calculated by Nadarajah (1982) and district characteristics reflected in census and other published statistics. - 318 Our first comparison is between Kurunegala and Anuradhapura districts (Table 8.4). Table 8.4: Comparison of Kurunegala and Anuradhapura Districts. Characteristics Year Kurunegala Anuradhapura Population density/sq. mi. 1953 493 57 1971 557 141 Percent urban 1953 3.2 8.0 Net migration rate 1946-1953 4.4 27.3 1953-1963 -2e 16.6 1963-1971 -2e4 7.5 Percent Sinhalese 1953 91.9 86.9 Unemployment rate: males 25-29 1971 11.5 4.3 Percent literate 1953 67.5 63.5 SMAM 1963 20.9 19.1 1971 22.6 21.0 1981 24.0 21.9 It is clear that the two districts were quite similar as of 1953, particularly with respect to ethnic composition and literacy level. But they differ in many respects. Kurunegala cuts across the the wet and dry zones, whereas Anuradhapura is entirely within the dry zone. By and large, Kurunegala has been more densely settled and less able to provide economic opportunities for new entrants to the labor force (see migration and unemployment rates). Consistent with these characteristics, we find that SMAM has remained lower in Anuradhapura than in Kurunegala. Another itrteresting comparison is between Kegalla and Ratnapura - 319 - districts (see Table 8.5). While both lie partly in the central hig4lands, they also border the most urbanized southwest coastal districts of the country. In terms of economic opportunities, Kegalla has been consistently worse off than Ratnapura throughout 1953-1971 (see net migration and unemployment rates). The Indian Tamils, who tend to marry at a relatively young age, constitute a higher proportion of the population of Ratnapura Table 8.5: Comparison of Kegalla and Ratnapura. Characteristics Year Kegalla Ratnapura Population density/sq. mi. 1953 735 337 1971 1,020 529 Percent urban 1953 1.2 4.5 Net migration rate 1946-53 -3.1 -0.2 1953-63 -5.2 -1.3 1963-71 -3.9 -0.1 Percent Sinhalese 1953 82.0 75.6 Percent Indian Tamil 1953 12.4 20.7 Unemployment rate: males 25-29 1971 20.4 12.7 Percent literate 1953 62.4 55.1 SMAM 1963 22.5 22.0 1971 24.3 23.6 1981 25.5 24.1 320 - than of Kegalla; moreover, Kegalla is much more densely populated than is Ratnapura. And, as we would expect, the SMAM has been higher in Kegalla than in Ratuapura. Kandy and Nuwara Eliya of the central highland regions (see Table 8.6) differ economically as well as ethnically, the Indian Tamils constituting a much smaller proportion of the population in Kandy than in Nuwara Eliya. While having the higher literacy rate in 1953, Kandy seems to have had less capacity for population absorption than Nuwara Eliya from 1953 to 1971. Consistent with these characteristics, SMAM has been higher in Kandy than in Nuwara Eliya. Table 8.6: Comparison of Kandy with Nuwara Eliya. Characteristics Year Kandy Nuwara Eliya Population density/sq. mi. 1953 920 686 1971 1,9301 950 Net migration rate 1946-53 -3.1 -1.0 1953-63 -6.1 -6.9 1963-71 -6.6 -5.3 Percent Sinhalese 1953 58.0 35.8 Percent Indian Tamil 1953 30.5 59.2 Unemployment rate: males 25-29 1971 - 13.5 7.3 Percent literate 1953 57.3 46e7 SMAA4 1963 22.0 21.6 1971 24.0 23.4 1981 25.4 24.3 - 321 - Yet another interesting comparison is between Puttalam on the west coast and Trincomalee on the east coast (see Table 8.7). The main difference between these districts is in their ethnic compositions: in 1953, the Sinhalese were in the majority among the residents of Puttalam, while they were a small minority in Trincomalee. In terms of population absorption capacity, Puttalam seems to have fared worse than Trincomalee. The lower SMAM in the latter is consistent with these characteristics. (Why the 1981 SMAM figures are almost the same for the two districts is not clear.) Table 8.7: Comparison of Puttalam with Trincomalee. Characteristics Year Puttalam Trincomalee Population density/sq. mi. 1953 65 80 1971 329 186 Percent urban 1953 17.4 31.4 Net migration rate 1946-53 13.5 -5.8 1953-63 -0.1 14.9 1963-71 0.8 3.9 Percent Sinhalese 1953 53.7 18.0 Percent Sri Lankan Tamil 1953 13.5 40.6 Percent Sri Lankan Menor 1953 29.0 33.1 Unemployment rate: males 25-29 1971 8.9 7.8 Percent literate 1953 67.5 65.1 SMAM 1963 20.7 18.1 1971 21.9 20.2 1981 22.1 22.0 - 322 - The purpose of these comparisons has been to illustrate that ethnic composition and population absorption capacity are probably important factors affecting the nuptiality patterns in different territorial populations.12 From these macro-level comparisons, let us turn to micro-level patterns. From the general ideas outlined earlier,a positive association between educational attainment of the wife and her age at marriage can be deduced. For males as well as females, in the Sri Lankan context, there is a positive association between educational attainment and the waiting period for employment after the completion of schooling.13 For males, there is a strong association between this waiting period and the gap between the completion of schooling and marriage. For females, this last association may be weaker. Nonetheless, these two associations (between educational 12A11 possible pair-wise comparisons of districts are not informative, because in some pairs forces operating in opposite directions prevail. The comparisons presented here seem to involve more or less unambiguous patterns. 13According to the 1969-70 Socio-Economic Survey, percent unemployed in Sri Lanka was much higher for the better-educated than for the less-educated: No schooling 8 Some primary 15 Some secondary 39 Passed GCE 63 As already mentioned, the ILO Mission, which examined the employment situation in Sri Lanka in the 1960s came to the conclusion that "the more a young person had been educated, the greater the likelihood that he or she will be unemployed" (ILO, 1971:28). According to one interpretation, this association (between educational attainment and the likelihood of being unemployed) is attributable to the tendency for the educated to have higher job ambitions and the inability of the economy to create enough jobs to match the aspirations kindled by higher education (Wilson, 1979:56-57). - 323 - attainment and the waiting period for employment, on the one hand, and the latter and the interval between completion of schooling and marriage, on the other) combine to produce a positive relationship between educational attainment and age at marriage for the wife as well as for the husband. The relationship for each spouse indirectly influences the relationship for the other because (owing to homogamy) age, age at marriage, and educational attainment of the spouses are all associated. (For example, if a college-educated man has to wait five years after graduation for employment, not only is his age at marriage increased by that waiting period, but that of his spouse as well, who in all likelihood is also college-educated.) The data from the 1979 Survey show the pattern, revealed in Table 8.8, for women of age 25-49 years who married before turning 25 years of age: Table 8.8: Mean Age at Marriage by Educational Attainment. Wife's Wife's Mean Age at Marriage Educational Attainment Unadjusted Adjusted14 None 16.8 15.1 Grades 1-5 17.3 15.1 Grades 6-9 18.8 16.1 Grades 10 or higher 21.1 17.9 14Adjusted for premarital work experience, religion, ethnicity, year married, length of engagement period, mode of mate selection, value of dowry, place of residence during the first 12 months after marriage, whether the respondent was in school during the 12 months immediately before the wedding, obligations to older sisters and younger brothers and sisters, and older brothers' and older sisters' obligations. - 324 - Separate analyses for the Sinhalese, Sri Lankan Tamils, Indian Tamils and Sri Lankan Moors reveal the patterns shown in Table 8.9. Table 8.9: Mean Age at Marriage by Educational Attainment--Ethnic Groups. Wife's Mean Age at Marriage (Adjusted)15 Wife's - Educational Sri Lankan Indian Sri Lankan Att.ainment Sinhalese Tamils Tamils Moors None 15.6 14.6 17.5 14.5 Grades 1-5 15.7 14.9 17.5 13.5 Grades 6-9 16.8 16.0 16.7 15.1 Grades 10 or higher 18.1 17.7 17.7 16.5 The patterns are statistically significant for all but the Indian Tamils. The major distinction seems to be between those with less than sixth grade education and those with sixth grade or higher level education. The educational effect is stronger for the Sinhalese than for the others. Another factor affecting age at marriage, according to the general ideas outlined above, is wife's work experience before marriage. In the 1979 survey it was found that among women of age 25-49 who got married before age 25, the mean age at marriage was 18.8 years for those who worked 15Adjusted-for premarital work experience, year married, length of the engagement period, mode of mate selection, value of dowry, place of residence during the 12 months immediately following marriage, economic security of the husband at the time of marriage, whether the respondent was in school during the 12 months immediately before marriage, obligations to older sisters and to younger brothers and sisters, and older brothers' and sisters' obligations. - 325 - before marriage and 17.6 years for those who did not--a difference of 1.2 years. The corresponding difference when other variables (see f.n. 14) were controlled for was 0.68 years (a statistically significant difference). Separate analyses for the major ethnic groups showed that pre-marital work experience had a significant effect on age at marriage for all but the Sri Lankan Moors. Although some attempt was made in the 1979 Survey to collect data on the relationship between financial security of the husband at time of marriage and age at marriage, the information collected did not shed sufficient empirical light on the relationship. However, the data showed slightly higher age at marriage for women whose husbands had a regular income at the time of marriage than for those whose husbands had no regular income at the time of marriage. This relationship was statistically significant only among the Sri Lankan Tamils. Respondents who reported that their husbands had regular income before marriage were asked whether the marriage decision would have been any different if the husband did not have a regular income at the time of marriage. A significant minority reported that the marriage negotiations would have been postponed or dropped altogether if such was the case. A problem with this type of data is that they rely on the impressions of the wife, whose part in the marriage negotiations was probably modest. What is needed is information about the strategies the actual decision makers adopted. Unfortunately, this type of information was not collected in the 1979 survey. Marital Fertility: With regard to marital fertility, it seems reasonable to assume that married couples are guided by the following rule: adjust the number of - 326 - children to the chances of their survival to adulthood and the expected length of their dependency period.16 If so, under sustained mortality reduction, people are likely to shift from higher to lower fertility, the familiar argument being that when mortality is low it is not necessary to have as many births to ensure the survival of a given desired number to adulthood. While outmigration opportunities will probably dampen the rate of shift just referred to (Davis, 1963), in the case of Sri Lanka, as in most of the less-developed countries, this factor is of little importance (Freedman, 1979). Circumstances that prolong the dependency period of children, such as widespread educational opportunities and prolonged waiting period for jobs, are likely to accelerate the transition from high to low fertility. Obviously, marital fertility can be reduced by marrying late and/or by using contraception within marriage. The latter reduces the monthly probability of conception, while late marriage reduces fertility by curtailing the total exposure period (assuming no pre-marital conception) or by inviting the operation of fecundity impairment for women marrying in their late 20s or thereafter. The particular combination of these strategies (late marriage versus contraception within marriage) actually adopted by population groups may be partly determined by cultural factors and the nature of family planning programs. From Table 8.10, a summary picture of the changes in 16The emphasis on the length of the dependency period rather than on the dependency burden itself is predicated on the assumption that the dependency burden is primarily a positive function of the length of the dependency period. In Sri Lanka this latter assumption is apparently a tenable one, in view of the availability of free education, free health care, etc. - 327 - Table 8.10.: Percent Change in Duraticn-Specific Marital Fertility RLates (DSMFR)--1960-65 to 1965-70 and 1965-70 to 1970-75. Marital Percent Change in DSMFR Duration From 1960-65 From 1965-70 at Birth Sample To 1965-70 To 1970-75 0 to 4 years 1. All women -0.6 -4.2 *2. Women married before age 30 0.0 -3.6 5 to 9 years 1. All women -8.6 -13.2 2. WorL.n married before age 25 -5.7 -11.2 10 to 14 years 1. All women -11.3 -21.7 2. Women married before age 20 -3.1 -19.1 15 to 19 years 1. All women -15.6 -27.2 2. Women married before age 15 +2.0 . -27.2 Cumulative: Up to 15 years 1. All women -6.2 -12.0 2. Women married before specified ages -2.8 -10.6 Up to 20 years 1. All women -8.0 -14.8 2. Women married before specified ages -1.8 -14.3 Source: Alam and Cleland (1981); p. 30. - 328 - duration-specific marital fertility since 1960-65 can be obtained. Two sets of figures are shown, one for 'all women' and the other for women first married before reaching specified ages. The figures for 'all women' are affected by truncation bias, resulting from the fact that those for longer durations are progressively restricted to women marrying young, as women older than 50 years were excluded from the survey. The figures for women first married before reaching specified ages are affected by selectivity bias, i.e., because of the rising age at marriage, as in Sri Lanka recently, many of the early-marrying members of the more recent marriage cohorts may have been 'selected' for high fertility, while many of the late-marrying members of the older marriage cohorts may have been 'selected' for low fertility. The actual declines probably lie between the two sets of estimates. The decline in Duration-Specific Marital Fertility Rate (DSMFR) was negligible for marriage duration under 5 years during 1960-65 to 1965-70 and very modest (about 4 percent) during 1960-70 to 1970-75. The decline was progressively higher at longer durations, particularly in the more recent period (1965-70 to 1970-75). In terms of cumulative figures, a fall in fertility of 3-6 percent occurred during the first period and 10-12 percent during the second (more recent) period in the first 15 years of marriage--the corresponding figures for the first 20 years of marriage being 2-8 percent during 1960-65 to 1965-70 and 14-15 percent during 1965-70 to 1970-75. Thus, the pace of DSMFR decline more than doubled toward the end of the 1960-75 period. To what extent did different subpopulations participate in the declines? Tables 8.11 and,8.12 reveal that all major population segments - 329 - Table 8.11: Percent Change in DSMFR by Place of Residence--1960-65 to 1965-70 and 1965-70 to 1970-75. Marital Percent Change in DSMFR Place of Duration From 1960-1965 From 1965-1970 Residence at Birth To 1965-1970 To 1970-1975 Urban 0-6 +7.0 -6.6 5-9 -13.7 -15.9 10-16 -6.4 -25.1 15-29 -22.3 -36.9 Cumulative: 0-15 -3.4 -13.7 0-20 -6.8 -17.1 Rural 0-6 -2.9 -0.8 5-9 -7.9 -13.1 10-16 -12.0 -22.3 15-19 -13.6 -21.1 Cumulative: 0-15 -7.2 -10.8 0-20 -8.4 -12.6 Estate 0-6 -5.5 -13.9 5-9 +3.5 -12.5 10-14 -11.2 -6.8 15-19 -8.7 -42.9 Cumulative: 0-15 -4.0 -11.6 0-20 -4.9 -17.6 Source: Alam and Cleland, Table 16, p. 36. - 330 - Table 8.12: Percent Change in DSMFR (Cumulated to 15 and 20 Years of Marital Duration) by Religio-Ethnic Grouping, Years of Education, and Husband's Occupation-1960-65 to 1965-70 and 1965-70 to 1970-75. Percent Change in DSMFR Cumulated to 15 Years of Duration 20 Years of Duration During During 1960-65 1965-70 1960-65 1965-70 to to to to Subpopulations 1965-70 1970-75 1965-70 1970-75 Religio-Ethnic Grouping: Sinhalese Buddhists -8.2 -12.9 -9.7 -15.4 Tamil Hindus +2.0 -9.7 +0.2 -7.8 Moor Muslims +0.8 -6.1 -3.2 -7.3 Years Educated: Husband Wife <6 <6 -7.5 -8.9 -8.1 -11.0 <6 6+ -5.9 -12.6 -7.7 -11.3 6+ <6 +3.8 -8.4 -18, rl0,8 6+ 6+ -12.0 -12.6 -13.9 -16.4 Husband's Occupation: White collar -13.1 -16e3 -17.2 -19.2 Sales & services -3e8 -10.9 -4.6 -14.7 Agriculture, self-employed -4.6 -7.6 -4.1 -9.0 Agricultural labor -5.2 -10.5 -5.6 -10.9 Skilled, manual -3.2 -13.3 -8.9 -15.8 Unskilled, manual and household work -7.0 -9.4 -9.6 -10.2 Source: Computed from Alam and Cleland, 1981, pp. 37, 39, and 40. - 331 - participated in the declines more or less equally, and almost synchronously. Table 8.11 shows that, during 1960-65 to 1965-70, cumulative DSMFR over 15 years of marriage declined about 3 percent in the urban sector, about 7 percent in rural areas and about 4 percent in the estates, the corresponding figures for the more recent period (1965-70 to 1970-75) being 14, 11, and 12 percent, in the urban, rural, and estate sectors, respectively. The declines were large at longer durations and small at short durations in the urban and rural areas, while there seem to have been sharp declines even at shorter durations in the estate sector. Table 8.12, from which estate residents are excluded, shows the following patterns: 1. The Tamil Hindus and Moor Muslims experienced negligible declines during 1960-65 to 1965-70, while the Sinhalese Buddhists experienced nearly 10 percent decline in DSNFR. During 1965-70 to 1970-75, however, all three communities participated in the decline, with the Hindu Tamils outpacing the Moor Muslims and the Sinhalese Buddhists outpacing the Hindu Tamils. 2. The declines were steeper during the more rci.ent period compared to the earlier period in each educational stratrum, at each period increasing with the level of education. 3. All occupational strata participated in the declines and in each stratum the declines were steeper during 1965-70 to 1970-75 than during 1960-65 to 1965-70. In summary, the rate of marital fertility decline was much more pronounced during 1965-70 to 1970-75 than during 1960-65 to 1965-70. All strata participated in the process. The estimated declines shown above may have been inflated somewhat in some strata (e.g., the white-collar and the better-educated strata), due to late age at marriage. Allowing for such -J inflations, the picture that emerges is one of uniform participation by almost all strata in the decline process. This is contrary to what one would expect under the diffusion model of family limitation practices, 332 - according to which family limitation spreads from the urban center to the rural hinterland and from the upper to the lower classes. The fact that all strata participated more or less to the same degree and more or less synchronously in the process gives some credence to the hypothesis that the family planning program had an independent impact on the process. Contraceptive Practice: It remains now to examine whether the spread of use of modern contraceptives coincided with the introduction and expansion of the national family planning program, whether subpopulations in which use rates were relatively high would have been so had there been no family planning program, and whether the popularity of methods used corresponds to those emphasized by the program. There is ample evidence indicating that widespread use of modern contraception in Sri Lanka was uncommon until recently. Among women who were sterilized or had ever used the pill, loop, or condom as of 1975, 66 percent had become f.irst acceptors of these methods between 1970 and 1975 (Immerwahr, 1981:317). Sterilization, to which the national family planning program shifted emphasis early in the 1970s, was, as of 1975, the most popular method used, adopters of sterilization accounting for almost 50 percent of the users of modern methods (the pill, loop, condom, sterilization, injectables, other scientific methods such as foam tablets, and the diaphragm) and 29 percent of ever users of any method (Immerwahr, 1981:8). A great impetus for its popularity occurred in 1973 (see Table 8.3) when the United Nations Fund for Population Activities (UNFPA) began supporting the push for national sterilization with special provisions for more effective surgical facilities. In the third quarter of the 1970s, the - 333 - family planning program seems to have weakened in its efforts, judging from the new acceptor statistics (see Table 8.3), and during that period TFR levelled off. Also pertinent are differentials in contraceptive use, of which Table 8.13 shows some relevant figures, borrowed from Immerwahr (1981). These are the mean scores of contraceptive use calculated by assigning a score of 0 for non-users, 1 for i1sers of traditional methods, 2 for users of the pill, condom, or the injectables, 3 for users of IUD, and 4 for the sterilized. The mean scores shown are adjusted via regression for a number of demographic and socio-economic variables, the regressors being months since first marriage, age at marriage, number of births in the 5 years following first marriage, region (zone) of residence, race (ethnicity) and religion, present and childhood place of residence, wife's work pattern, husband's occupation, wife's education, and a standard-of-living score based on housing convenience and modern consumption items owned. The following patterns are worth noting: 1. Contrary to what one would expect based on the Western experience, urban residents brought up in urban areas do not have, consistently across marriage durations, hither mean contraceptive use scores compared with other segments. 7 17Ever-use, however, has consistently been higher in the urban sector than in the rural sector. Thus, for those brought up in rural areas, Immerwahr (1981) reports (from the 1975 Survey, WFS) the following ever-use rates: Rural 43.1 percent; Urban 61.4 percent--the corresponding figures for those brought up in Urban areas being: Rural 50.1 percent and Urban 55.4 percent. In the 1982 Contraceptive Prevalence Survey, the ever-use rates observed were: Rural 65 percent, Urban 68 percent (Department of Census and Statistics, 1982). 334 - Table 8.13: Mean Score o:E Contraceptive Use by Place of Residence, Wife's Work Experienace, and Husband's Occupation, for Women in Different Marriage-Duration Categories, Adjusted for a Number of Covariates, Duration of Marriage Population Segments 0 to 4 5 to 9 10 to 19 20+ Type of place of residence: Childhood Current Rural Rural .437 1.003 1.163 1.109 Rural Urban .577 .754 1.514 1.105 Urban Rural .469 1.199 1.703 .945 Urban Urban .647 .825 1.518 1.273 Wife's work pattern: Never worked .450 .946 1.241 1.153 Worked only before marriage .424 1.273 1.470 1.107 Worked after marriage Away from home .587 1.062 1.279 1.116 At home .573 .707 1.209 1.028 Husband's occupation: Farming .331 .744 1.241 .996 Professional & clerical .701 1.058 1.076 1.139 Agricultural laborer .449 .786 1.392 1.013 Sales & services .578 1.056 1.287 1.413 Other .418 1.095 1.325 1.134 Source: Immerwahr (1981:28-31). - 335 - 2. A similar finding is that women with non-home-centered work experience differ little from women with no work experience in adjusted mean score of contraceptive use.18 3. Farmers and farm laborers have relatively lower mean scores during the first 10 years of marriage; but at 10-19 years of marital duration, their mean adjusted scores exceed that of the professional and clerical stratum.19 Such differentials as these seem partly to reflect the activities of the family planning program, which presumably was organized to reach particularly those population segments otherwise less prone, given the costs involved, to adopt family limitation practices within marriage. While these patterns support the contention that the national family planning program played a part in reducing fertility, it should be remembered that the downward course of fertility had already begun by the time the government program came into existence in the late 1960s. In fact, if we were to extend the downward trend which became discernable during 1965-67, we would be able to reproduce the total fertility rates for all years from 1965 to 1975, except for 1973 and 1974 (see Table 8.14). The deficit in the observed figures for 1973 and 1974 can be attributed to the national family planning program, particularly the UNFPA-assisted efforts. 18The ever use rates observed in the 1982 Contraceptive Prevalence Survey were: Not Working 66.9 percent, Working 68.8 percent; the corresponding age-standardized figures being 67.6 and 65.4 percent, respectively. 19The ever-use rates observed in the 1975 Survey (WFS) were: Farmers 37.7 percent Professional & Clerical 58.6 percent Agricultural laborers 27.1 percent Sales & Service workers 53.1 percent - 336 - Table 8.14: Comparison of Observed TFRs and the Corresponding Values on the Trend Line Extrapolated from the Observed Values in the Third Quarter of the 1960s. Year Observed TFR Estimated TFR 1965 4.8 4.8 1966 4.7 4.7 1967 4.6 4.6 1968 4.6 4.5 1969 4.4 4.4 1970 4.3 4e3 1971 4e2 4.2 1972 4.1 4.1 1973 3.9 e 1974 3.8 3.9 1975 3.8 3.8 1976 3.8 3.7 1977 3.8 3.6 1978 3.9 3.5 - 337 - Between 1972 and 1974 the Total Fertility Rate would have declined from 4.1 to 3.9 if the 1965-67 trend had continued. The actual decline was sharper--from 4.1 to 3.8. Hence, one-third of the decline in TFR during this period may be attributed to the national family planning program. This method credits the program with no part of the decline in TFR between 1965 and 1975, except for the short period in the second quarter of the 1970s.20 20Nimali Kannangara (1982) claims that a minimum of 5,500 births were averted by the family planning program in 1973, the corresponding figures for 1974 and 1975 being 17,700 and 18,600, respectively. She refers to Siva Obeysekira, who has claimed that about 117,000 births were averted by the family planning program during 1968-72. From these figures, one could calculate what would have been the Crude Birth Rates in the country had there been no family planning program. For 1968-72 period, one thus gets an estimate of 32.3 births per 1,000 population, if there were no family planning program. This figure is consistent with the reported CBR of 32.3 for 1966 in the official statistics. For 1975, the observed and estimated CBRs are 27.8 and 29.1, respectively, the latter being the level one would have expected had there been no family planning program. Hence, between 1966 and 1975 (29.1 - 27.8)/(32.3 - 27.8), or 28.9 percent of the decline in CBR is attributable to the family planning program, according to the estimated numbers of births averted, given by Kannangara (1982). Immerwahr (1981) gives 68,000 as a lower estimate of the number of births averted prior to 1976 by sterilization. Immerwahr's own estimate of the number of sterilizations prior to 1976 is 185,000 (after inflating his estimates for the first three quarters of 1975 by 4/3), of which 52,400 (or 4/3 of his figure) were performed in 1975 alone. Distributing the estimated, number of births averted prior to 1976 by sterilization in proportion to the number of sterilizations, we get 19,000 as a lower estimate of the number of births averted in 1975 due to sterilizaion, which implies that the Crude Birth Rate would have been 29.2 in 1975 in the absence of sterilization. A lower estimate of the decline between 1966 and 1975 in CBR attributable to the family planning program is, therefore, (29.2 - 27.8)/(32.3 - 27.8), or 31 percent, which is comparable to the estimate obtained above, using Kannangara's (1982) figures. 338 - Another point worth noting in this connection is that there is a great deal of inconsistency between people's declared intentions and their subsequent behavior, as the two-wave data (consisting of the 1975 interview and the 1979 re-interview) reveals. Information is available for 2,476 women regarding whether, as of the first interview date (1975), they wanted "another child sometime" and their actual fertility behavior in the interim. Of these women, 1,665 wanted no more and had none in the interim; 811 wanted no more, but had one or more; 248 wanted one or more, but did not have any; and 563 wanted one or more, and had one or more. A limitation of this data, of course, is that while the expressed intentions pertain to the full remaining reproductive period as of the first interview date, the actual behavioral information available is confined to the truncated follow-up period, i.e., the interval between the two interviews (1975 and 1979). There is reason to believe, however, that the number of inconsistents is underestimated rather than overestimated by the truncation of the follow-up period. Those who wanted more children, as of the first interview date, but had none in the interim are few in number (248), and many of them already had long open intervals by the second interview date. Therefore very few of them are likely to move into the category "wanted more and had more." The transfer of women, subsequent to the follow-up period, from "wanted no more and had none" to "wanted no more but had one or more" is likely to be proportionately and in absolute numbers much higher than the transfer from "wanted more, but had none" to "wanted more and had more." Put differently, if the follow-up period was not truncated, more inconsistents would have been encountered. The propensity to have unwanted births is high in Sri Lanka as - 339 - compared to, say, Taiwan: 38.6 percent as opposed to 14.1 percent (the Taiwan figure is from Hermalin et al., 1979, and pertains to the period 1970-76, while the Sri Lankan figure is from the two-wave data covering the period 1975-79). Moreover, unwanted births occur with considerable frequency in all strata in Sri Lanka. When we examine the patterns of incidence of unwanted births by background variables, we get a picture similar to the corresponding patterns involving non-use of contraception at the time of the first interview (see Table 8.1-5). While there is little doubt that part of the discrepancy between declared intention and subsequent behavior is due to the non-use of contraception, there is no satisfactory explanation available for non-use. When asked at the first interview, 10 percent of the never-users who wanted no more children mentioned lack of knowledge of contraception as their reason for non-use. To an equal proportion, disapproval of family planning was the explanation. About 9 percent mentioned that their husbands were opposed to family planning, and 6 percent did not know how to obtain or where to go for supplies or services. But many women were unable to give any satisfactory explanation for not using a contraceptive method (Immerwahr, 1981:23). There is no doubt that the high incidence of inconsistency between declared intentions and subsequent behavior makes the former a poor predictor of the latter, whatever the explanation for the inconsistency. The following section makes a quantitative assessment of this matter. A Recursive Model: Let us assume a causal sequence in which variables 1,2,...,k are antecedents to variable k+1, for k = 1,2,and 3, where - 340 Variable 1 stands for fertility preference functions at time 1 (the response to the question: "If you could choose exactly the number of children to have in your whole life, how many would that be?"); - Variable 2 stands for demand for additional children at time 1 (the response to the question: "How many more children do you want to have?"); - Variable 3 stands for contraceptive use status (user vs. non-user) as of time 1 (sterilized women and wives of sterilized husbands being regarded as users with no desire for additional children); and - Variable 4 stands for fertility behavior in the interim (the number of live births between the two interviews). Let x8s also enter the following variables with no presumed causal ordering among them as predictors of the ones already listed: - Number of children living (family size) as of time 1; - Age at first marriage; - Interval between marriage and last birth as of time 1; - Open interval (i.e., the interval since last birth) as of time 1; - Wife's education; - Husband's occupation (coded 1 0 -1, respectively, for the upper, middle, and lower statuls occupations; - Number of 'modern' articles owned; - Housing convenience score; - Type of place of residence (urban=l; other=O); - Region of residence (1 for zones 1 and 2; 0 for other zones); - Religion (1 for Muslims and 0 for others); - Ideal family size for a girl from a family like that of the respondent (as ascertained at the secoTad interview, but treated as a lagged reading of the social norm concerning family size as of time 1); and - Wife's work pattern, time 1. In this model, there are four simultaneous equations and each equation can be estimated separately using ordinary least squares, under the usual assumptions regarding the residuals. See Table 8.16 for the results. - 341 - Table 8.15: Inconsistency Between Intention Declared at the First Interview and Subsequent Behavior and that Between Intention Declared at the First Interview and Behavior as of that Time. Percent who had Percent Non-Users One or More Births Among Exposed Women Characteristics But Wanted Nonel Who Wanted No More2 Region of residence: Zone 1 26 54 2 26 67 3 38 71 4 56 82 5 41 76 6 35 63 Type of place of residence: Urban 34 60 Rural 39 67 Estate 40 73 Tevel of education. None 40 74 1-5 years 38 68 6-9 years 38 61 10+ years 35 58 Religion Buddhist 33 63 Hindu 44 73 Muslim 54 73 Christian 34 70 1Based on the re-interviewed subsample. 2Based on the sample interviewed in 1975; source: World Fertility Survey, Sri Lanka, 1975, First Report, Table 8.3, p. 146. 342 Table 8.16: A Recursive Model--Standardized Regression Coefficients. Fertility Demand for Contracep- Fertility Pr.eference: additional tive use in the ReEressors Time 1 Children: Status: interim Time 1 Time 1 No. of children living: time 1 0.71* -077* 0.48* .0.08* i Age at marriage 0.00 0.02 -0.01 -0.08* ITnterval: Marriage to last blrth -C.01 -0.22* -0.22* -0,35* Interval: Since last birth: time 1 -0.05* -0.02 0.13* -0,35* Wife's education 0,07* -0.03 0,08* 004 Husband's occupation 0.01 -0.03 0.02 -0 03 No. of modern articles owned 0.01 0.00 0.09* -0.05* Housing convenience 0.03 0.04 0.03 -0001 Type of place of residence (Urban) -0,05* 0.01 0.00 -002 Region of residence (Zones 1,2) -0.03 -0e08* 0.08* -0 e04 Religion (Muslim) 0,05* 0o05* 0,o11* 0.02 Ideal family size for a girl 0e11* 0.04* -0.01 0.05* Work pattern: wife -0D01 0.01 0e01 0.00 Fertility preference: time 1 0.55* -0,09* 0.14* Demand for additional -0o16* -0.02 children: time 1 Contraceptive use status: time 1 -0.22* Source: Table 5.10, Chapter V. - 343 - Fertility preference as of time 1 is significantly affected by the current family size and the family size norm for the respondent's stratum. Longer open intervals depress it, as.do higher education and urban background. Religious preference also has a significant impact. The regressors together account for 56 percent of the variance. The demand for additional children at time 1 is significantly affected by the current family size and fertility preference at time 1, duration of marriage, region of residence, religion, and social norm concerning family size. Almost 50 percent of the variance is accounted for by the regressors taken together. The contraceptive use status at time 1 (user or not) is significantly affected by the current family size, marital duration, the current demand for additional children, the open interval, religion, social norms concerning family size, the number of modern articles owned, and the region of residence. The nature of the relationship involving the open interval is ambiguous. Some women with long open intervals, despite non-use of contraception, might have interpreted their experience to mean that they were not exposed to the risk of pregnancy, and their decision to continue to be a non-user logically followed from that interpretation. For others, contraception (e.g., sterilization) might have been responsible for long open intervals. Fertility in the follow-up period is significantly affected by duration of marriage, age at marriage and the open interval as of time 1, fertility preference at time 1, social norm concerning family size, and the number of articles owned. Note that the demand for additional children as of time 1 has no significant impact on fertility in the interim. When - 344 - declared family-size goals do not tell anything about subsequent fertility behavior, it is logical to suspect that the declared goals do not truly reflect deliberate choices or that the pursuit of the goals was abandoned for one reason or another (e.g., unanticipated heavy costs) or that the goals themselves were subsequently changed for some reason. The Effect of Infant Deaths: If, as suggested earlier, people adjust their reproductive behavior to their beliefs concerning their infants' chances of survival to adulthood, it is reasonable to expect an association between infant deaths experienced by a mother and her subsequent fertility behavior. The data from WFS, 1975, show that the death of the nth born tends to hasten the birth of the (n+l)th child (see Table 8.17). Table 8.17: Median Times to Move from n Parity to (n+l) Parity, by the Survival Status of the nth Born. Survival Status of the nth Born n n+l Surviving Not Surviving 1 2 27.72 21.30 2 3 28.38 20.91 3 4 28.85 21.63 4 5 29.14 22.36 5 6 30.13 21.49 Source: Suchindran and Adlaka, 1981. To see whether this relationship was primarily due to the shortening of the post-partum amenorrhea period resulting from the termination of lactation in the event of death of the infant, Suchindran and Adlaka (1981) examined the association between the time to move from - 345 - n-parity to (n+l) parity and the survival status of the (n-l)th born, among women with their nth borns surviving. Their results are shown in Table 8.18. Table 8.18: Median Time to Move from n Parity to (n+l) Parity, by the Survival Status of the (n-l)th Born, Given the Survival of the nth Born. Survival Status of the (n-1)th Born n n+l Surviving Not Surviving 2 3 28.36 28.64 3 4 28.85 28.78 4 5 29.12 29.53 5 6 30.07 31.67 Source: Suchindran and Adlaka (1981). These results indicate that the association between infant deaths experienced and subsequent fertility is primarily due to a physiological effect. (When the variation in the physiological effect is controlled for, the association vanishes.) Suchindran and Adlaka (1981) examined the impact of infant death experience on the instantaneous rate of having additional births, given current *parity, controlling for a number of demographic and socio-economic background factors. They found that for 2-parity women the instantaneous rate of having additional children among women whose second born had died was 1.29 times that of women whose second born had not died, the corresponding results for 3-, 4-, and 5-parity women being 1.29, 1.18, and 1.43, respectively. The number of infant deaths among the first (n-i) children had no significant impact on the instantaneous rate of having - 346 - additional children among n-parity women. The indications are, therefore, that the link between infant deaths and subsequent fertility is primarily physiological. If there is a behavioral effect (stemming from planned behavior on the part of couples to "replace" infant losses already incurred or to insure against future losses) it is a minor one at best. Summary Gathering the main threads of the material reviewed so far we find that: - Sustained fertility decline in Sri Lanka is a post-1960 phenomenon. - Initially it was mainly rising age at Trarriage that caused the drop in the Total Fertility Rate (TFR); later on, however, declines in marital fertility became an increasingly important contributor. - Marital fertility declined much faster during the period 1965-70 to 1970-75 than during 1960-65 to 1965-70. - In the 1960s, the use of modern contraception was not common. As the national family program became a reality in the late 1960s, however, the popularity of modern contraceptive methods began to increase. - All strata participated in the marital fertility decline during 1960-75, more or less to the same degree and almost synchronously. This is in sharp contrast to the patterns that characterized the Western fertility transition, starting with the upper class and the urban sector and moving down to the lower class and to the rural hinterlands, creating, expanding, and finally contracting in the process class differentials and rural-urban differences. - The rising trend in age at marriage stalled in the latter half of the 1970s. The singulate mean age at marriage for females was 23.6 in 1971, 24.8 in 1975 and 24.4 in 1981 (the 1981 figure is from s Nadarajah, 1982:8). - The national family planning program slackened its efforts somewhat during the third quarter of the 1970s. - The Total Fertility Rate levelled off and even showed signs of an upturn during 1975-1980. - Unwanted births occur frequently in all strata, and declared - 347 - intentions with respect to additional children do not help predict subsequent fertility behavior.' This seems to indicate one or more of the following: the declared goals do not reflect deliberate choices; implementation of plans is handicapped by problems related to contraceptive instrumentalities; goals are unstable, reflecting indecision on the part of some and genuine attempts to adapt to new circumstances on the part of others. In what combination these prevail in Sri Lanka is unknown. Implications Raising age at marriage is obviously one way to bring down fertility levels. In Sri Lanka, female education and labor-force participation have been strongly associated with rising age at marriage. But quite possibly it was increasing aspirations (created by formal education and exposure to the expanding world warehouse of ideas and new life styles) frustrated by lack of opportunity that played a major part in raising the age at marriage of males and, consequently, that of females. The recent levelling off, if not lowering, of the singulate mean age at marriage is perhaps partly due to the improvement in Sri Lanka's economy. A combination of a number of social, economic, and cultural factors was responsible for initiating and setting the pace of the fertility decline beginning in the early 1960s. Among these factors are: (1) sustained mortality reduction, mainly brought about by planned intervention, dating back to the 1940s (e.g., malaria eradication program); (2) high literacy and education--also a result of deliberate planning; (3) income transfers and other measures resulting in a lessening of class inequality; (4) elevation of women's status, mainly as a result of increased parity in educational opportunities within a cultural context in which females are accorded relatively higher status than is usually the case in South'Asia; and (5) governmental efforts to provide family planning information, supplies and services. 348 - The effect of infant mortality on fertility seems to have been primarily a physiological phenomenon--working through the termination of lactation on the death of the infant--rather than a behavioral one stemming from a planned increase in births to "replace" lost infants or to insure against potential losses. Thus, in the Sri Lankan context, reduction in infant mortality depressed fertility primarily through the prolongation of lactation. The Sri Lankan experience indicates that high literacy and education when coupled with longer waiting periods for employment cause age at marriage to rise. Obviously, this particular combination is not an attractive policy tool. On the other hand, high literacy and education affect age at marriage and marital fertility by creating extra-familial interests and exposing young men and women to ideas and life styles not indigenous to the Sri Lankan culture. This situation lends itself easily to policy intervention in that high literacy and education create a fertile ground for the world network of communication and interdependence to exert its full power and persuasiveness in motivating people to adopt new ways of life or some combination of the old and the new. The impact of social welfare measures which lessen class inequality is unclear. If fertility rises as income increases among lower-income strata and falls with income among higher-income strata, then a transfer of income from the upper to the lower classes raises fertility. If the income-fertility relationship is linear and positive, then redistribution of income leaves average fertility unaffected. If, on the other hand, the basic relationship between income and fertility is negative in the lower-income strata, fertility is U-shaped, and a lessening of class 349 - inequality via redistribution of income depresses fertility. There is also another aspect: if particular welfare measures such as free food or food subsidies amount to a transfer of an insurance function from children to the nation-state or another agency--i.e., insurance against sudden discontinuance of or drastic reduction in sustenance flow over time--such welfare measures might depress fertility, other things being equal. It is thus unclear whether the lessening of class inequalities has a clearly identifiable, universal impact on the fertility level.21 The lessening of gender inequality depresses fertility insofar as it increases the direct opportunity costs of children. That is, the direct cost of children is influenced by equalizing educational opportunities for men and women, if education is costly or--even when it is not--there is a relatively long waiting period after the completion of schooling to obtain suitable employment. Simply put, if less gender inequality results in more autonomy for women over their own lives or engages them in extra-familial activities (which are also incompatible with familial activities), then opportunity costs of children will increase. The reduction of gender inequality is thus another attractive policy tool. The family planning program is also a factor. There is sufficient ZlThere are certain Central Asian republics in the Soviet Union where high fertility has continued to prevail long after substantial land reform (collectivism) has been introduced (Coale, 1978:411). Iraq and Kuwait have lessened class inequalities through deliberate intervention by the nation state, but fertility levels there have not declined (Safilios-Rothschild, 1982:195). In Sri Lanka, at least for a period, the colonization program seems to have given preference to landless persons with large families thereby promoting, unintentionally, higher fertility (Abhayaratne and Jayewardene, 1968:24). 350 - evidence to indicate that government efforts to provide family planning information, supplies, and services facilitated fertility reduction in Sri Lanka. It is clear that such efforts can have noticeable impacts on the fertility level of a population even without urbanization, industrialization, and other hallmarks of the Western industrial complex. What is not clear, however, is whether some as yet undetermined changes in the social and economic fabric are necessary before a family planning program can be expected to have a dramatic impact on fertility levels. As the Sri Lankan case amply demonstrates, fertility decline is possible even without the high standard of living and other characteristic features of Western-style industrial modernization. Among the factors responsible for the Sri Lankan fertility decline are: sustained mortality decline; higher levels of literacy and education, with no discrimination along gender lines; linkage with other populations far and near, via a worldwide network of communication and interdependence; government efforts to provide family planning information, supplies, and services; and welfare policies which create substitutes for traditional support arrangements expected of children. I4 - 351 - BIBLIOGRAPHY Abeyesekere, G. 1982. "Socio-Economic Background to Fertility Changes in the Post-War Era of Sri Lanka. Paper presented at the National Workshop on Socio-Economic Development and Fertility Decline in Sri Lanka, 24-26 August. Colombo Abhayaratne, O.E.R. and C.H.S. Jayawardene. 1968. Family Planning in Ceylon. Colombo: Apothecaris Press. . 1967. Fertility Trends in Ceylon. Colombo: Apothecaris Press. Agrarian Research Training Institute. 1974. The Agrarian Situation Relating to Paddy Cultivation in Five Selected Districts of Sri Lanka. Colombo. Alam, Iqbal. 1980. An Analysis of Fertility Levels and Trends from the Sri Lanka Fertility Survey. WFS Scientific Reports-Series. Alam, Iqbal and John Cleland. 1981. "Illustrative Analyses: Recent Fertility Trends in Sri Lanka." World Fertility Survey Scientific Report No. 25 (November). Balakrishnan, N. 1979. "A Review of the Economy." In T. Fernando and R. N. Kearney (eds.), Modern Sri Lanka: A Society in Transition. Syracuse: Foreign and Comparative Studies. Syracuse University. Becker, G. S. 1974. "A Theory of Marriage, Pt.. II.` Journal of Political Economy 82, 2:11-26. Birdsall, Nancy. 1980. "Population and Poverty in the Developing World." World Bank Staff Working Paper No. 404. Washington, D.C. (July) :44-47. Cain, Mead. 1981. "Risk and Fertility in India and Bangladesh." Population and Development Review 7:435-74. Caldwell, John C. 1976. "Toward a Restatement of Demographic Transition Theory." Population and Development Review 2:321-66. Caldwell, J. C., P. F. McDonald, and L. T. Puzicka. 1980. "Interrelationships between Nuptiality and Fertility: The Evidence from the World Fertility Survey." Paper presented at the World Fertility Survey Conference, London (July). Central Bank of Ceylon. 1976. Bulletin (April). - 352 - _ 1975. Annual Report, 1974. Colombo. Cho, L. J. and R. D. Retherford. 1973. "Comparative Analysis of Recent Fertility Trends in East Asia. IUSSP International Conference, Liege. Vol. II:163-81. Coale, A.J. 1978. "Reply." Foreign Affairs 57:410-11. Corea, Gamini. 1975. The Instability of an Export Economy. Colombo, Sri Lanka: Marga Institute. Davis, Kingsley. 1963. "The Theory of Change and Response in Modern Demographic History." Population Index 29:345-66. ESCAP. 1976. Population of Sri Lanka Country Monograph Series No. 4. Fernando, D.F.S. 1976. "Fertility Trends in Sri Lanka and Future. Prospects." Journal of Biosocial Science 8 (January). v 1972. "Recent Fertility Decline in Ceylon." Population Studies 26:445-53. Fernando, Tissa. 1979. "Aspects of Social Stratification." In T. Fernando and R. N. Kearney (eds.), Modern Sri Lanka: A Society in Transition. Syracuse: Foreign and Comparative Studies, Syracuse University. Freedman, Ronald. 1979. "Theories of Fertility Decline: A Reappraisal." Social Forces 58:1-17. Gunaratne, D. et al. 1977. "University Admission 1976: Background Characteristics of the Entrant Population." Unpublished report. Colombo: Ministry of Education. Cited in Jayaweera (1979). Gwatkin, D. R. 1980. "Indications of Change in Developing Country Mortality Trends: The End of an Era?" Population and Development Review 6:615-44. Hajnal, J. 1965. "European Marriage Patterns in Perspective." Pp. 101-43 in D. V. Glass and D.E.C. Eversley (eds.), Population in History. London: Arnold. Hermalin, A., R. Freedman, T. Sun, and M. Chang. 1979. "Do Intentions Predict Fertility? The Experience in Taiwan, 1967-74." Studies in Family Planning 10:75-95. Immerwahr, George. 1981. "Contraceptive Use in Sri Lanka: World Fertility Survey." Scientific Reports No. 18 (January). International Labor Office. 1971. Matching Employment Opportunities and Expectations: A Programme of Action for Ceylon. Geneva: International Labor Organization. - 353 - Issanman, Paul. 1980. "Basic Needs: The Case of Sri Lanka." World Development 8:237-58. Jayaweera, S. 1979. "Education." In T. Fernando and R. N. Kearney (eds.), Modern Sri Lanka: A Society in Transition. Syracuse: Foreign and Comparative Studies Program, Syracuse University. . 1973. "A Study of Educational Opportunities and Employment Opportunities Open to Women in Sri Lanka." Mimeographed. Colombo & Paris: UNESCO. Cited in Jayaweera (1979). Jones, G. W. and S. Selvaratnam. 1972. Population Growth and Economic Development in Ceylon. Colombo: Hansa Publishers. Kannangara, Nimali. 1982. "The National Family Planning Programme and Its Effect on Fertility in Sri Lanka in the Last Three Decades." Paper presented at the National Workshop on Socio-Economic Development and Fertility Decline in Sri Lanka. Colombo. Kaplan, E. L. and P. Meier, 1958. "Nonparametric Estimation from Incomplete Observations." Journal of the American-Statistical Association 5, 3:457-81. Klecka, William R. 1975. "Discriminant Analyses." Statistical Package for the Social Sciences. 2nd ed. McGraw-Hill Book Company. Kleinbaum, David and Lawrence Kupper. 1978. Applied Regression Analyses and other Multivariable Methods. North Scituate, MA: Duxbury Press. Ludowyk, E,F.C. 1966. Modern History of Ceylon. London: Weidenfeld and Nicolson. Malalgoda, K. 1976. Buddhism in Sinhalese Society, 1750-1900. Berkeley: University of California Press. Meegama, A. N. 1980. Socio-Economic Determinants of Infant and Child Mortality in Sri Lanka: An Analysis of Post-War Experience. WFS Scientific Reports Series,'No. 8. Melicl,er, E. 1965. 'Least Square Analyses of Economic Survey Data." Paper presented at the Annual Meeting of the American Statistical Association, Philadelphia. Nadarajah, T. 1982. "Some Aspects of Nuptiality Trends." Paper presented at the National Workshop on Socio-Economic Development and Fertility Decline in Sri Lanka. Colombo Obeyesekere, G, 1970. "Religious Symbolism and Political Change in Ceylon." Modern Ceylon Studies I:43-63. Pullum, T. W. 1980. "Illustrative Analysis: Fertility Preference in Sri Lanka-. WFS Scientific Reports No. 9. 354 - Puvanarajan, P. 1982. "Social Change and Fertility in Sri Lanka." Paper presented at the National Workshop in Socio-Economic Development and Fertility Decline in Sri Lanka, 2w-6 August. Colombo. Ryder, Norman and Charles Westoff. Reproduction in the United States, 1965. Princeton, NJ: Princeton University Press, 1971. Safilios-Rothschild. 1982. "A Class and Sex Stratification Theoretical Model and Its Relevance for Fertility Trends in the Developing World." In C. Hohn and R. Mackensen (eds.), Determinants of Fertility Trends: Maior Theories and New Directions for Research. IUSSP. Government of Sri Lanka. 1982. Department of Census and Statistics. Contraceptive Prevalance Survey. 1981. Colombo. _ 1978. Department of Census and Statistics. World Fertility Survey, Sri Lanka, 1975: First Report. Colombo: Ministry of Plan Implementation. -__-_. 1978. Department of Census and Statistics. Census of Population. 1971. General Report. Colombo. e 1976. Department of Census and Statistics. Population of Sri Lanka. Colombo. _* 1970. Department of Census and Statistics. Preliminary Report of the Socio-Economic Survey, 1969-70. Colombo. =_,_-___* 1962. Department of Census and Statistics. Census of Agriculture, Vol. 1. Colombo. * 1982. Family Health Bureau. _ 1980. Family Health Bureau. Eva'luation Unit, Reports on New Acceptors of Family Planning. Snodgrass, D. R. 1966. An Export Economy in Transition. Homewood, IL: Richard D. Irwin, Inc. Suchindran, C. M. and A. Adlaka. 1981. "Infant Mortalitty and Its Effect on Fertility: The Experience of Sri Lanka." Paper presented at the Annual Meeting of the Population Association of America. Washington, D.C. Taxation Inquiry Commission. 1968. Report of the Taxation Inquiry Commission. UNFPA. 1980. "UNFPA Assists Sri Lanka in its Drive to Lower Birth Rates." Population, UNFPA Newsletter Vol. 6, No. 1 (January). - 355 - Westoff, C. F., N. Goldman, M. K. Choe. 1980. "Prevalance and Demographic Significance of Contraceptive Sterilization in Fiji, The Republic of Korea, and Sri Lanka." Papers of the East-West Population Institute, No. 66 (April). Westoff, C. F., J. McCarthy, N. Goldman, and M. Mascarin. 1979. "Illustrative Analysis: Contraceptive Sterilization and Birth Averted in Panama." WFS Scientific Reports No. 4. Westoff, C. F. and N. B. Ryder. 1977. "The Predictive Validity of Reproductive Intentions." Demography 14:431-50. Wickremeratne, L. A. 1977. "Peasant Agriculture." In K. M. DeSilva (ed.), Sri Lanka: A Survey. Honolulu: The University Press of Hawaii. Wilson, A. J. 1979. Politics in Sri ,anka, 1947-1979. London: Macmillan.