Labor force participation in a developing metropolis : does sex matter?

SWP-749 Labor Force Participation in a Developing Metropolis Does Sex Matter? Rakesh Mohan WORLD BANK STAFF WORKING PAPERS Number 749 SECTORAL LIBRARY INTERNATIONAL BANK FOR RECONSTRUCTION AND DEVELOPMENT SEPA 1 81988-4 HD * HD5757.A6 R35 1985 c.2 5757 Labor force participation in a developing metropolis does .AG R35 SLCO2820 1985 SLC028620 J, __ _ __ _  WORLD BANK STAFF WORKING PAPERS Number 749 Labor Force Participation in a Developing Metropolis Does Sex Matter? Rakesh Mohan The World Bank Washington, D.C., U.S.A. Propcrty of Projects Policy Department Library Copyright @ 1985 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W Washington, D.C. 20433, U.SA All rights reserved Manufactured in the United States of America First printing August 1985 This is a working document published informally by the World Bank. To present the results of research with the least possible delay, the typescript has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. The publication is supplied at a token charge to defray part of the cost of manufacture and distribution. The World Bank does not accept responsibility for the views expressed herein, which are those of the authors and should not be attributed to the World Bank or to its affiliated organizations. The findings, interpretations, and conclusions are the results of research supported by the Bank- they do not necessarily represent official policy of the Bank. The designations employed, the presentation of materiaL and any maps used in this document are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the World Bank or its affiliates concerning the legal status of any country, territory, city, area, or of its authorities, or concerning the delimitation of its boundaries, or national affiliation. The most recent World Bank publications are described in the annual spring and fall lists; the continuing research program is described in the annual Abstracts of Current Studies. The latest edition of each is available free of charge from the Publications Sales Unit, Department T, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from the European Office of the Bank, 66 avenue d'I6na, 75116 Paris, France. Rakesh Mohan is an economist in the Country Programs Department of the World Bank's East Asia and Pacific Regional Office. Library of Congress Cataloging-in-Publication Data Rakesh, Mohan, 1948- Labor force participation in a developing metropolis. (World Bank staff working papers ; no. 749) Bibliography: p. 1. Labor supply--Colombia. 2. Labor supply-- Colombia--Bogotd. I. Title. II. Series. HD5757.A6R35 1985 331.11'09861'4 85-16929 ISBN 0-8213-0591-3 ABSTRACT After sustained economic growth in Colombia over two decades, the labor market in Bogota exhibited considerable tightening up in the late 1970s. Unemployment rates were seen to have declined and the rate of participation of women to have increased perceptibly over the decade. This study uses data from a 1978 household survey to examine the determinants of labor force participation. The sample is divided into seven age/sex groups to capture the life-cycle effects on participation. Different estimation methods (ordinary least squares, Probit, and Tobit) are used to estimate the elasticity of labor supply with respect to our earnings, wealth, household income, and education, along with other variables. Particular attention is paid to estimation technique to eliminate sample censorship bias, selectivity bias, and variable censorship bias. The estimates show that the main component of the labor force that can be expected to expand rapidly with rising real wages is that of married women. As access to secondary and higher education improves further, more young men and women will stay longer in school before entering the labor force. Hence, fewer young men and women will enter the labor force at a young age. The same forces that induce greater participation of married women will serve to reduce the entry of young women into the labor force at a young age. These economic inducements are essentially the likelihood of receiving relatively higher wages with higher level education in relation to the opportunity cost of nonmarket work. Women have been catching up with men rapidly in average educational attainment. This suggests an increasing desire of women to enter the labor market and this can be expected to continue for some time. Once these results are coupled with the association of labor force participation with lower fertility, the future expansion of women in the labor force can be appreciated further. At the same time, the late 1970s showed the increased participation of poor women in response to the real wage increases observed over that period. Among nonmarried women (widows, divorcees, never married), there is a clear element of necessity in their labor market participation, especially of those who are household heads. They simply have to work at whatever the existing wage level; and, hence, their labor force participation rates are not as responsive to wage changes as those of married women. Prime age men are similar since almost all of them are in the labor force anyway. These results have implications for urban labor markets in developing countries in general. It may be the case that the remarkable increase in female labor participation that has taken place in developed countries may occur in developing countries earlier than might be expected at rather lower absolute income levels. PREFACE A rather large amount of data handling, documentation and preparation has gone into the work leading to this paper. I would like to thank Sungyong Kang and M. Wilhelm Wagner for making the 1978 City Study-DANE Household Survey ready for use. Yoon Joo Lee was responsible for the early estimations for this paper and the original literature search. A particular word of appreciation goes to him for his help over a protracted length of time. John Roseman and Sungyong Kang did the final computations under considerable time pressure. A special word of appreciation goes to the staff of DANE for having painstakingly conducted the survey. The high quality of the 1978 City Study- DANE Household survey owes much to the diligence of Roberto Pinilla and Maria- Christina Jimenez of DANE and Gary Losee, Alfredo Aliaga, Alvaro Pachon, Jose Fernando Pineda and Jairo Arias for CCRP-World Bank. I would also like to thank Michael Hartley for introducing me to the intricate world of limited dependent variables and the pitfalls in their estimation, Surjit Bhalla for useful discussion, and Dennis de Tray, Manny Jimenez and Nancy Birdsall for helpful comments on an earlier draft. This paper is part of a program of research which has been conducted by the World Bank on Bogota and Cali, Colombia. The goal of the program, entitled The City Study, is to increase our understanding of the working of five major urban sectors--housing, transport, employment location, labor markets and the public sector--in order that the impact of policies and projects can be assessed more accurately. This paper is part of the labor market and income distribution portion of the study which is coordinated by Rakesh Mohan. Other papers in this series are: Gary Fields, "How Segmented in the Bogota Labour Market", World Bank Staff Working Paper No. 434, Washington D.C., October 1980. Rakesh Mohan "The People of Bogota: Who They Are, What They Earn, Where They Live", World Bank Staff Working Paper No. 390, Washington D.C., May 1980. , "The Determinants of Labour Earnings in Developing Metropolises: Estimates from Bogota and Cali, Colombia", World Bank Staff Working Paper No.498, Washington, D.C., 1981. , M. Wilhelm Wagner and Jorge Garcia, "Measuring Urban Malnutrition and Poverty: A Case Study of Bogota and Cali, Colombia", World Bank Staff Working Paper No. 447, Washington D.C., April 1981. and Nancy Hartline, "The Poor of Bogota: Who They Are, What They Do, Where They Live", World Bank Staff Working Paper No. 635, Washington D.C. 1984. _ "An Anatomy of the Distribution of Urban Income: A Tale of Two Cities in Colombia", World Bank Staff Working Paper No. 651, Washington, D.C., 1984. TABLE OF CONTENTS I. Introduction: Evolution of the Urban Labor Market in Colombia.........................................1 II. Estimating Labor Supply: Theoretical Considerations.......6 2.1 Estimation Technique...... .... .... . .... ........ .8 2.2 Final Empirical Specification and Variables Used...16 2.3 Dependent Variables........... . .......... ...... 24 III. Estimates of the Determinants of Labor Supply..............25 3.1 Introduction......................................25 3.2 The Labor Force Participation Decision.............26 3.21 Comparing OLS and Probit Estimates........ e......26 3.22 Young Men and Young Women...........................29 3.23 Prime Age Unmarried Women and Married Women........33 3.3 The Total Supplyof Labor: Hours Worked in the Labor Market......................********36 3.31 Interpreting Tobit and Heckman Coefficients........36 3.32 The Tobit and Heckman Estimates...................37 3.4 Comparison of Estimates Across Models..............44 3.41 Young Men and Young Women..........................44 3.42 Prime AGe Unmarried Women (UF) and Married Women (MF).......................................45 IV. Urban Labor Supply in a Developing Metropolis: What have we learned?.. ... . ..... .. ....... ........48 4.1 Supply of Labor in Bogota............................48 4.2 Econometric Considerations...... ................... .51 Appendix I: Wage Function Estimates............................53 Appendix II: Derivation of the Labor Supply Function and Estimation Strategy......................56 Bibliography ............********** ****6 List of Tables 1 Bogota Participation Rates by Age Group..*.....o..........3 2 Bogota Participation Rates by Education..................5 3 Specification of Labor Supply Equations.................20 4a The Participation Decision ............................ 28 4b The Participation Decision.............................. ** 34 5a The Supply of Labor Schedule: Young Men................40 5b The Supply of Labor Schedule: Young Women..............41 5c The Supply of Labor Schedule: Prime Single Women........42 5d The Supply of Labor Schedule: Married Women.............43 6 Elasticities of Participation and of Supply of Labor....47 A.1 Wage Functions . o ....................................... .55 Maps.....*.*** ***.....*.*****..... ... ...* ** ** ......... .. .23 I. INTRODUCTION: EVOLUTION OF THE URBAN LABOR MARKET IN COLOMBIA There is now widespread agreement that the Colombian urban economy underwent a remarkable series of changes during the decade of the seventies which has led inevitably to a number of perceptible changes in the urban labor market.1/ With a sustained overall real growth rate in CNP of about 6 percent a year annually over 2 decades from 1960 to 1980, per capita incomes in Colombia doubled over that period. There has been rapid structural change in the economy so that agriculture and industry now each contribute about a quarter of gross national product with the tertiary sector accounting for the rest. The level of urbanization rose drastically from about 38 percent in 1951 to 52 percent in 1964, 60 percent in 1973 and, at a somewhat slower rate to about 65 percent (estimated) in 1980-82. All of these changes resulted in a perceptible tightening in the labor market in the late seventies. A dramatic expansion in employment took place in Colombia in the late seventies. The growth in overall employment was about 6.2 percent a year between 1973 and 1978, with the manufacturing and tertiary sector employment expanding at over 8 percent annually (World Bank, 1983). This occurred in a period when overall expansion in the labor force was at a rate of 3.4 percent a year. As a result, the labor market visibly tightened up and urban unemployment rates were observed to fall along with increased participation rates. The key component of these higher participation rates was the increased participation of women in the labor force. That this occurred is shown in Table I which shows the labor force participation of men and women 1/ These are well summarized by Miguel Urrutia (1985) in his careful examination of the changing income distribution in Colombia in the 1970s. - 2 - from 1964 to 1978. In the earlier years, the expansion of education served to reduce the participation rate of young men and women. This expansion in education was substantially completed by the mid seventies and the participation rates of the young stabilized or increased slightly. Among men, the participation rate of the prime working age group (25-54) has remained roughly constant at between 93 and 97 percent. Among women, however, the same age group has clearly increased its participation and specifically in the late seventies, after a possible decline earlier. There is some evidence that the participation rates of poor women particularly increased in these years. Table 2 gives indirect indication of this by giving the labor force participation (LFP) rates broken down by educational attainments of men and women. The LFP rates of women with primary education increased significantly toward the latter part of the decade. A striking feature of Table 2 is the comparatively higher LFP rates of higher educated women--but they account for only 9 percent of all women. The most significant change in the labor markets of the developed countries in the post-war period has also been a secularly increasing proportion of women participating in the labor force and considerable analytical work has therefore been devoted to explaining this phenomenon. The maintained hypothesis is essentially that LFP can be expected to increase when the expected wage in market work is greater than the perceived opportunity cost of alternative activities. In the case of married women, this is the perceived returns from non-market or house work. As the economy expands and - 3 - Table 1: BOGOTA PARTICIPATION RATES BY AGE GROUP Age group 1964 1973 1975 1977 1978 Men 15-24 66.7 61.2 52.2 50.6 52.2 25-34 94.9 92.8 94.3 94.3 92.7 35-44 97.2 93.7 96.2 96.2 96.0 45-54 94.4 88.2 91.9 94.9 87.3 55-64 81.8 68.8 73.8 72.0 68.9 65+ 41.6 36.5 42.1 33.7 37.5 Total 15+ 75.1 70.2 66.9 65.9 65.4 Women 15-19 40.4 ) 36.2 36.8 38.8 37.5 20-24 47.5 ) 25-34 37.4 37.0 42.2 46.1 45.1 35-44 34.0 34.4 33.9 41.2 37.9 45-54 27.3 26.9 26.3 31.0 35.6 55-64 18.0 17.2 14.6 17.5 19.6 65+ 9.3 13.9 6.7 7.2 6.6 Total 15+ 33.5 30.7 30.7 33.8 33.4 Source: City Study estimates from 1964, 1973 Population Censuses and 1975, 1977 and 1978 DANE Household Surveys. the stock of educated women deepens, it can be expected that the market work opportunity costs increase disproportionately to the returns from house-work, which would lead to increased participation of women. This is an important issue for the fast growing "Newly Industrializing Countries" from both the supply and the demand side of the labor market. From the supply side, as educational opportunities expand, it may be expected that more and more women would expect to be in the market labor force. Thus, policies for educational expansion need to take this phenomenon into account, both in terms of quality of education supplied and the types of education supplied. Similarly, from the demand side, in conditions of rapid economic expansion, the supply of labor could become a constraint once the traditional unlimited supply from the countryside slows down. The importance of increasing female LFP in the income distribution has also been indicated. This study has been conducted in the context of a wider study of Bogota in an attempt to understand the implications of a rapidly growing city on the labor market--but in the context of wider economy wide concerns. The overall tightening of the labor market and falling urban unemployment which occurred in the late seventies in Colombia is a novel phenomenon for developing countries and deserves greater analysis than it has received. The events in recent years have, of course, reversed these trends but the importance of the events of the late seventies remain, since they suggest conditions in which rapid urbanization is not necessarily accompanied by secularly rising urban unemployment as is often suggested. - 5 - Table 2: BOGOTA PARTICIPATION RATES BY EDUCATION (Percentage of persons in labor force above 15 years) Education 1973 1975 1977 1978 None 69.4 64.3 63.2 60.2 Primary 74.3 74.6 71.8 73.0 Secondary 63.7 58.3 58.5 57.3 Higher 71.7 72.0 72.5 70.6 Total 70.2 67.1 65.9 65.4 Women None 30.0 27.9 34.0 31.3 Primary 30.2 32.3 36.2 35.5 Secondary 31.1 26.7 29.0 27.9 Higher 42.5 48.6 49.4 53.0 Total 30.7 30.8 33.8 33.4 Source; City Study estimates from 1973 Population Census. 1975, 1977, 1978 DANE Household Survey (EH8E, EH15 AND EH21). - 6 - II. ESTIMATING LABOR SUPPLY: THEORETICAL CONSIDERATIONS Most work on labor supply is of relatively recent origin and has largely been conducted since the early sixties. This is one branch of economics where theoretical and empirical work has progressed hand in hand. The availability of new kinds of data has spurred the development of new extensions to theory, while at the same time, theoretical developments have been closely linked with new estimation techniques. An excellent extensive and almost exhaustive review is available in Killingsworth (1983) so only a brief review of theory relevant for this paper will be made here. The explosion of research on labor supply has to a large extent been stimulated by the observation of dramatically increasing labor force participation (henceforth LFP) rate of women in western countries in the post- second World War period. There had been little interest prior to that since there is little historical variance in the LFP of prime age men, being universally near 100 percent in the urban labor market. The interest in prime age men is much more in cyclical changes in the labor market, that is, in changes in the unemployment rate. Significant changes have also taken place in the LFP of other groups of secondary workers. Thus the expansion of education has typically decreased the LFP of young men and women, while the provision of better social security has affected the LFP of old men and women. There are, however, other considerations as well. The improvements in life expectancy and general health levels have on the other hand, tended to increase the LFP of older people. The declines in fertility, availability of home appliances, etc. have tended to increase the LFP of women at all age - 7 - levels. The explanation of LFP is therefore not a simple matter and earlier ad hoc approaches have gradually led to a unified theoretical framework. Earlier work on LFP had to rely on variance in LFP rates between and among different sex and age groups in different cities and regions and over different periods of time to derive inferences on the determinants of LFP. 1/ Thus the average participation rate of a group was typically taken as the dependent variable to be explained. The implications of such an approach are several. First, the average LFP rate for the group is implicitly interpreted as the probability of participation of each individual. Second, the participation decision of the individual, as revealed by the group average, is abstracted from his position in the household. Again, it was the focus on the LFP of married women that forced the consideration of labor supply in the household context (Mincer (1962) and Becker (1965), Ashenfelter and Heckman (1973)). It was the increased availability of household level data which made it feasible to examine the individual LFP decision within the household context. The basic theory is derived from the traditional utility maximization framework. Although the theory is relatively straightforward and available in other sources (e.g. Heckman (1974), Nakamura and others (1978) and Killingsworth (1983)) it is presented here in brief in Appendix II because of its close relationship with the estimation technique. As reviewed in Killingsworth (1983), the estimation of labor supply can be done in a bewildering number of ways. Consequently, a number of different techniques 1/ An exhaustive work of this kind was Bowen and Finnegan (1969) for the U.S. - 8 - have been employed in this study in order to (a) test for the robustness of the estimates and (b) to test for the need for the more sophisticated methods. 2.1 Estimation Technique As derived in Appendix II the basic labor market equations are Appendix equations (A17), (A18) and (A30). wi* = 0 +1h + a2i + a3A + a4Z i + u. = a h. + ZMa + u. (1) 1 1 1 1 w.= 8~ + 6 S. B2E+ 83X.+ e. Wi o 1 Si + 02 E 1 3 X 1 S X8 + e. (2) 1 1 and P = f(W. - w*) (3) 1 1 i h.= 0 1 = f(X? - Zft + e. - u.) 1 1 1 1 where wi* = shadow price of the individual's time, W. = market wage rate, and P = 1 when h > 0 S 0 when h = 0; b is the number of hours of market work, y is the household labor income, Ai is household non-labor income, i.e. return from assets, Z is a vector of other individual and household characteristics arrived at through earlier decisions, Si is years of schooling, E is years of market experience, I is a vector of other individual characteristics relevant for market work, - 9- ui and ei are random disturbances, Ui is - N(O, a1) e is - N(O, 02) and may be correlated with each other. From (1), the labor supply equation can be derived, 1 h = - {w* - (aO 2 i+ a3A + a4 Z + u)} (4) 1 0 2 3i 4i i There are five basic problems in the estimation of (2), (3) and (4): (i) Censored Dependent Variable: The labor supply equation (4) and the participation equation (3), both have dependent variables which are censored since h. is observed only when h. > 0 and P. is a binary variable. (ii) Censored Sample: In the cases of both equation (2) for wi and equation (4) for hi (if estimated for workers only), the sample itself is censored since it is composed of workers only and not the whole population. As shown in the appendix, ui and ei are both correlated with each other and are truncated as well. (iii) Selectivity Bias: Again, in the cases of equations (2) and (A), the available sample of workers is one resulting from systematic decision making in the past (the decision on schooling, age at marriage, number of children, etc.) the sample is therefore subject to selectivity bias. Naturally, this is more important for groups of people among whom levels of participation are low, that is, different types of secondary workers. It is, for example, not important for prime age men, almost all of whom work. - 10 - (iv) Errors in Measurement: This is the most serious for hours of work and wages. Annual or monthly income is the sum of the product of hours and wages and bonus payments, etc. Thus, invariably, the "true" wage has to be calculated by dividing some index of total labor income by the number of hours worked. Moreover, income is notoriously subject to reporting error. (v) Costs of Entry into the Labor Market: The derivation of the labor supply function has characterized the function to be continuous in the positive quadrant. This assumes zero costs of entry. In practice there are considerable search costs and the hours of work are also not entirely flexible so that the decision to work can be dependent on the minimum hours of work feasible. The different estimations account for these problems in varying degrees. The four methods that have been used are: Method I: This is a one stage procedure which estimates (3) and (4) in reduced form. This procedure uses the whole sample so problems (ii) and (iii) do not arise. Problem (iv) is also muted since wi is not used. Problem (i) is solved by using Tobit for estimation of (4) and Probit or logit for - equation (3). The main drawback in this procedure is that it is difficult to interpret the coefficients derived since the supply and demand sides are mixed up. Moreover, elasticities of interest, e.g. the own wage elasticity of labor supply has to be calculated indirectly--which itself gives rise to selectivity bias. This procedure also ignores the differences between workers and non- - 11 - workers, 1/ in terms of their decision making. Thus method I can be written as: h. = -1 (X*8 - Z.a + e. - u.) i = 1...n (4a) 1 0 1 1 1 1 P. = f(X.8 - Z.a + e. - u.) i = 1...n (3a) 1 1 1 1 1 Method IA. OLS. Method IB. Tobit for equation (4a) Probit or logit for equation (3a). Note also, that if (3) is linearized, the coefficients from equation (4a) and (3a) should differ by only a factor of proportionality. Method II: 2 Stage Procedure The first stage is to estimate wi by equation (2) over a sample of workers and then to use w. as an instrumental variable to estimate equations (3) and (4) over the whole sample. The assumption here is that sample selectivity bias does not exist and that the wage estimated from the market demand equation for workers applies equally well to non-workers. Both of these can be estimated by OLS (admitting problem (i)) or by logit or probit for equation (3) and tobit for equation (4). Thus Method II can be written-as W= X.8 + e. i =...m (2b) 1 ^ * h. - {w. - (Z.a+u.)} i = 1...n (4b) 1 al 1 1 1 1/ Heckman (1974) formed a likelihood function of workers'and non workers and then solved the resulting simultaneous equation system by Full Information Maximum Likelihood procedures. This is also a "one stage" method but too costly to implement. - 12 - P. = f(w. - (Z*a + u.)} = 1...n (3b) 1 1 1 1 where there are m workers and n-m non-workers. Method IIA OLS for all equations. Method IIB OLS for equation (2b). Probit or Logit for equation (3b) Tobit for equation (4b) Method III 3 Stage Procedure The first stage is to estimate P (reduced form) by probit to calculate the probability of selection as a worker and thereby to account for sample selectivity bias in equations (2) and (4). The second stage is to estimate equation (2) admitting selectivity bias by the inclusion of ~I) as an explanatory variable. The third stage is to estimate equation (4) by using w. (corrected for selectivity bias) and X. as explanatory 1 1 variables. Here equation (3) is estimated over the whole sample and equations (2) and (4) over workers only. Equation (3) is estimated by probit and equation (2) and (3) by OLS. Here problem (i) is accounted for in equation (3) and does not arise in equation (4) since the sample itself is censored but 1/ See appendix. - 13 - selectivity bias (problem (iii)) is accounted for. Thus Method III can be written as: P. = f(X.8 - Z.ca + e. - u.) Probit i =1...n. (3c) 1 1 1 1 1 w iB + k 2 OLS i1...m. (2c) 1 ^ Z* h. - (. - Z.a + akX. - V. ) OLS i = 1...m (4c) 1 al 1 1 i i1 Method IV 3 Stage This is the same as method II except that (4) is estimated over the whole sample by Tobit. The assumption is that w. having accounted for 1 sample selectivity bias is a good predictor of potential market earnings for non-workers as well. Thus Method IV can be written as - P. = f(X. - Z. + e. - u.) Probit i = 1...n (3d) 1 1 1 1 1 w. = X.8 + 8 + V i2 OLS i = l...m (2d) - 1 ^ * h. = L (w. - Z.Q + - u.) Tobit i = 1...n (4d) 1 a 1 1 1 - 14 - No comments have been made above on problems (iv) and (v). The reason for using w. as opposed to observed wi ( for workers) is to eliminate measurement error and transitory variations in earnings. Measurement error in hi remains. The variation in experience due to costs of entry are accounted for, in part, by the inclusion of family non-labor income and total family labor income in equation (4) as explanatory variables. Presumably, people with high family income are better able to search. That the supply of labor function is not continuous is suggested in our data by the fact that, of those who work, very few work less than 15 hours a week. It is useful at this point to summarize what has been derived so far. The individual labor supply schedule has been derived in the Appendix from the optimization of the household utility function. Subsequently all the discussion has been in terms of the individual and the household has been lost sight of to some extent. It should be noted that, in terms of the derivation, the main result of the household optimization is that family labor income as well as non-labor income are arguments in the supply function. Second, 'Z' contains family characteristics which have been achieved by earlier household decisions. These characteristics essentially include variables which affect the opportunity cost of the individual's time--such as number of children present, size of house, etc. The discussion so far has also been "system- free" or "culture-free" with the implicit argument that, at this level of generality, the specification of the labor supply function is equally valid for developing as well as developed countries. It is in the specification of the two vectors, Z on the labor supply side and X on the market demand side, that the differences between different situations can be brought in. Similarly, the discussion so far has not distinguished between the male labor - 15 - supply and the female labor supply function. Again, at this level of generality, it is argued that the framework is a unified one and that the differences between men and women would be in the specifications of Z and X, to the extent that different kinds of variables affect the opportunity cost of time and of market demand for men and women. Lastly, the optimization made has been a static one with no dynamic elements to capture life cycle effects. Life Cycle Effects In principle, a dynamic household utility function could be optimized to obtain a lifetime labor supply function, as has been done for women by Lehrer and Nerlove (1980). Here, as in most other studies, the life cycle effects are capture by dividing the sample into groups at different stages of the life cycle. Two key results from Lehrer and Nerlove are worth noting. First, a woman may supply labor to the market at a given time even if her shadow price of time is higher than the market wage at that time, if this results in an increase in her life time earnings which is more than compensating over her life cycle. Second, husband's (or other family) income has a bigger effect on the wife's labor supply after the arrival of children than before. The first result means that we should observe higher participation of women who expect higher incomes later, at all stages of their life cycle. Thus, women with higher education who expect an increasing age- earnings profile should have higher participation rates at early ages as well. This is clearly true in the sample. But this result also implies that a lower return to education should be observed for women. Although these results are not directly captured by the procedures used in this study, the results are quite consistent. However, the knowledge of these results helps - 16 - to interpret the coefficients estimated, and to delineate the variables that should be included in Z. The different groups used in this study are: Group Age group Other comments I. Young men (YM) 15-24 II. Prime age men (PM) 25-54 III. Old men (OM) 55+ IV. Young women (YF) 15-24 V. Married women (MF) All ages With husband present VI. Prime age unmarried women (UF) 25-54 All prime age other than MF VII. Old women (OF) 55+ While estimations have been made for all the categories for the sake of completeness, the group of greatest interest is that of married women, and then UF, YM and YF--since they are the main secondary worker group whose participation is seen to vary more as a result of different economic conditions. Prime age men are of less interest since almost all work. We do not have enough information on OM and OF to derive any results of interest. For the old people, it would be useful if there was better information on their social security provisions and on their other living arrangements in lieu of social security. In the case of old women, the working sample is often too small for good estimation. Hence, although estimations were made for these two groups, the results have not been reported here. 2.2 Final Empirical Specification and Variables Used The estimation equations for each group have a common structure. The vectors Zi and Xi in equations (4) and (2) can be divided into subvectors Z.. and Xki for conveniece. Hence: 1* h = L {wi - (c+ + Ci A + a Z + i Z + C1 Z + a Z + u) 1 a1 1 0 3i 4i 52i 63i 74i i - 17 - + a*w + S + SA + Sz +.Sz + Sz + Sz + U (5) 60 1 i 2yi 3Ai 4 i 5 2i 6 3i 7 4i + and wi 0 + 8 + 82E. + a3 X1 + 8 4X2i +5 X3i + e (6) where Zii = quality of leisure variables Z2i = life cycle variables Z3i background variables Z4i= home activity variables, and among the Xs', X = region of origin X2i = location of residence X = characteristic of employment. The earnings function--equation (6)--has been investigated in detail elsewhere (Mohan, 1981) and is therefore not discussed here. The only concern here is in the magnitude of selectivity bias and in the use of equation (6) in predicting w for use in equation (5). Moreover, since x3i, the characteristics of employment are available, by definition, for workers only, they are not used in these estimations. A detailed description of the variables used follows: (i) yi: Household labor income: This has been redefined to be "other family labor income" (OFLY). This has been done to purge it of wi, the individual's own contribution to yi. OFLY: Total monthly labor income of all other household members (in '000 pesos). The coefficient of OFLY is expected to be negative, if leisure is taken as a normal goal. (ii) Ai: Household non-labor income: FNLY: Monthly non-labor income of the household. ('000 pesos). This variable includes - 18 - interest/dividend earnings, rental income, pensions, gifts and other transfers. (iii) Zli Quality of Leisure Variables: Only one variable has been used here, YRSEDU: The number of years of education. It is expected that education pushes up the value of a person's time since he can use leisure time more productivity. Hence it should have a negative effect on labor supply (purged of its effect of increasing expected market wages). (iv) Z2i Life Cycle Variables: AGE: years AGESQ: (Age)2, (years2) MARRIED: = 1 if currently married 0 otherwise WIDOW = 1 if widowed (and not remarried) = 0 otherwise HHHEAD = 1 if individual is identified as the head of household 0 otherwise. FREUNI = 1 if individual is living with a spouse but not formally married. 0 otherwise These variables are included to account for life cycle effects. The last four variables measure some sense of "need to work". It would be expected that being married or being household head would increase the responsibility of the individual, thus making it important that they work. Naturally, these variables are of more importance for young men and women. FREUNI may capture - 19 - the lack of security for women, if the lack of formal marriage implies lack of a permanent commitment. It may be expected to increase their labor supply. Being a widow, among currently unmarried women, may increase their compulsion to work. (v) ZI; Background variables: MIG3 = 1 if the individual is a recent migrant and has been in Bogota for less then 3 years. = 0 otherwise. HUSBLU = 1 if the head of the individual's household is a blue collar worker. 0 otherwise. It may be expected that recent migrants have a lower value placed on their leisure time and hence are more likely to participate and work more. The variable HUSBLU could capture some social background effects. If it is expected that blue collar workers are more socially conservative, they would have a negative effect on the labor supply of women. (vi) Z4i Home Activity Variables: CHIL6 : The number of children in the household under 6 CHlL612 : The number of children in the household between 6 and 12 years in age. MREL = 1 if there is a maid or adult relative in the household. 0 otherwise. NROOMS : The number of rooms in the house. - 20 - Table 3: SPECIFICATION OF LABOR SUPPLY EQUATIONS Young Prime age Young Married Prime age Variable men men women women unmarried women 1. OFLY x x x x x 2. FNLY x x x x x 3. zli YRSEDU x x x x x 4. z2i AGE x x x x x AGESO x x x x x MARRIED x x FREUNI x WIDOW x HHHEAD x x x x 5. z3i MIG3 x x x x x HUSBLU x 6. z4i CHIL6 x x CHIL612 x x MREL x x NROOMS x - 21 - These variables are expected to capture the opportunity cost of womens' time in terms of the degree of home activity. The signs of their expected effects are the obvious ones. The data can only identify the presence of children in the household and their relationship to the household head. Hence the number of children are not necessarily those of the women in question. It has been assumed (a) that in almost all cases of married women (with husband present), they would be their children; and, (b) that for other women, the presence of young children in the household raise the demand for home-work for the women, even if they are not her own. Table 3 gives the specification of equation (5) used in the different labor supply equations. The variables used in equation (6) are: (i) Si: YRSEDU (ii) E i: Work experience EXPER = Age - YRSEDU -6 (years) EXPSQ = (EXPER)2 (years) The data do not give a direct measure of experience, except for a measure of the number of years spent by a worker in the same firm or in the same occupation. As demonstrated in Mohan (1981), the indirect measure performed better than these limited direct measures. This is therefore regarded as an accurate enough measure for men. It is clearly inadequate for married women who may have interrupted their careers. Experiments with constructing a better measure by using the number of children to impute number of years interrupted did not succeed either. Thus this inadequate measure has been used for women as well. - 22 - (iii) X1i: Region of Origin There are a set of dummy variables designed to account for the origin of the individual. DCITY = 1 if migrant was born in the three next largest cities 1/ 0 otherwise. DBOG = 1 if the individual was born in Bogota or migrated before the age of 10. O Otherwise. DTOWN = 1 if migrant was born in towns over 100,000 people. 2/ = 0 otherwise. DURB = 1 if migrant was born in other urban places. The excluded category is migrants from rural areas. (This set of variables has been used in some estimations only but not reported here). (iv) X2i The Location of Residence Mohan (1981) showed that the current location of worker's residence in the city was a good proxy for other background variables such as quality of schooling, ability, etc. Map 1 gives the division of Bogota into radial sectors and rings. The location of residence of each worker can be classified according to which sector he lives ino The residential sectors are controlled for by a set of seven dummy variables with sector 2 (the poorest sector located in the south of the city) acting as the reference sector. The seven 1/ Medellin, Cali, Barranquilla. 2/ Bucaramanga, Cartagena, Cucuta, Manizales, Pereira, Ibague, Armenia, Palmira, Pasto, Buenaventura, Neiva, Santa Marta. 拼一 、～·～衅～~ - 24 - dummy variables are: RSECT19 RSECT39 RSECT49 RSECT5v RSECT6p RSETC7 and RSECT8. Each takes the value 1 if the worker lives in that sector, 0 otherwise. The specification of equation (6) is same for all the groups so all the variables are used for estimating the earnings function for each. Where reduced form estimation is done, all the variables speficied are used as appropriate for each group. 2.3 Dependent Variables There are three different dependent variables: M pi = 1 if the person is in the labor force or not, 0 otherwise. A person is in the labor force if he (she) is 15 years old or more, is defined to be working or unemployed during the week previous to the survey (live-in domestic servants have been excluded from the sample). (ii) hi = the number of weekly working hours as reported by the workers. (iii) wi = log of hourly wage. Workers were asked the periodicity of their wage payments along with the unit wage. A second question was asked to elicit the same information: their total labor earnings in the previous month. The monthly earnings is the derived monthly earnings for employees who report the periodicity of wage payment and unit wage. For others, it is the monthly earnings reported for the previous month. The hourly wage variable is a constructed variable from the monthly earnings and hours of work information. - 25 - The Data All the estimations are made using the 1978 City Study-DANE Household survey conducted in Bogota in December 1978. The survey sampled about 3,000 households in Bogota and the quality of the data is regarded as good, since it covered over 90 percent of labor income. Earnings questions were asked of each individual in the household alongwith other information on their work status and place of worker other details on the survey are available in Mohan (1981) and Mohan (1984). The main exclusions made for the purposes of this study are of all live-in domestic servants who are almost all female. The reason is that since they are listed with the households for whom they work there is no background information on their own families. A small number of other workers also had to be excluded becausel of no information on their hours of work. III. ESTIMATES OF THE DETERMINANTS OF LABOR SUPPLY 3.1 Introduction This section reports the empirical results obtained from the different methods outlined in the last section. The aim in this section is both to obtain robust parameter estimates for labor supply in Bogota and also to make judgements on the estimation strategy used. It is cumbersome to report the results of all the 4 models seriatem. Moreover, the methods overlap and the aim is partially to compare results. Hence, first all the results for the participation equation (3) are reported and subsequently all the results for the hours worked equation (4). - 26 - 3.2 The Labor Force Participation Decision 3.21 Comparing OLS and Probit Estimates There are essentially two participation equations: equation (3a) and (3b) (since equation (3c) and (3d) are identical to (3a)). Tables (4a) and (4b) report the parameter estimates along with the means of each variable used. The first two estimates reported are of the reduced form equation (OLS and Probit A), corresponding to Model Ia and Ib respectively and the third equation includes estimated own wage as determinant 1/ (Model II). As explained in the last section, the OLS estimates are biased becuase the dependent variable is censored, and is in fact a dichotomous variable. It is, however, useful to see the magnitude and direction of the bias. Overall, it is observed that for each of the groups, the Probit coefficients are larger in absolute magnitude than the OLS coefficients in almost all cases. The effect is clearer when the coefficient is significant, and in each of these cases the t statistic is higher for the probit coefficient. This is as should be expected, but it is worth noting that the OLS coefficients are all of the right sign and of a similar order of magnitude. Chi-square statistics with the appropriate degrees of freedom are reported in the tables. The chi-square statistics exceeds the critical value of X2 0.995 for all the regressions. Thus the null hypothesis that all slope parameters are equal to zero is rejected. 1/ See Appendix I for estimation of the wage equation. - 27 - The probit coefficients reported are transformed since the probit model is: P(I.) - et2 dt i = 1...n (equation A.30) i /2-f CO ( where P(Ii) is the probability that a person participates and X.8 - Z.ca I. = i ) 1 3 (see Appendix II for derivation). Hence the probit coefficients 8 and a are not comparable to OLS coefficients and are difficult to interpret intuitively. However, our interest is in the marginal effect of each variable on participation, i.e. in aj- which is analogous to OLS coefficients. It can be shown that ax.. ij BE(PM (I).S (5) axij evaluated at (Ii), where 0(I) is the standard normal distribution function. At the mean, *(I.), the cumulative normal distribution, is merely 1 the proportion of people participating, hence 0(1i) can also be found at the mean. The coefficients reported are those given by (5) evaluated at the mean. The linear approximation suggested by Amemiya (1981) for the constant term is I.) 8 + 0.5 and this is reported in Tables 4a and 4b. The approximate similarity of the OLS and probit coefficients is then only at the mean, where the OLS coefficients are downward biased in absolute terms. The slopes would be quite different the further away one is from the mean. - 28 - Table 4a: 1TIE PARrICIPATION DECISI(N T Ihotmwus dependent variable) Young Rn Young amen Variable Wan 01S Probit A2 Probit B2 W-an OIS Probit A2 Prbit B2 ONTMAT -2.769 -1.839 (3.29) AGE 19.26 0.304 0.376 0.412 19.35 0.169 0.349 0.347 (3.42) (4.25) (4.67) (2.03) (5.02) (5.02) AGESQ 379.16 -0.006 -0.007 -0.007 382.86 -0.003 -0.007 -0.007 (2.46) (2.94) (2.93) (1.20) (3.93) (3.92) YRSED 8.24 -0.045 -0.064 8.11 -0.008 -0.008 (7.69) (10.87) (1.39) (1.99) OFLY 11.95 -0.001 -0.002 -0.002 11.89 -0.001 -0.002 -0.002 (0.81) (1.27) (1.48) (1.13) (1.93) (2.09) FNLY 2.11 -0.012 -0.023 -0.024 1.78 -0.007 -0.009 -0.010 (3.48) (5.29) (5.58) (1.95) (2.94) (3.19) HIHIAD 0.116 0.129 0.472 0.499 0.023 0.156 0.128 0.137 (1.79) (4.22) (4.50) (1.51) (1.74) (1.87) MIG3 0.121 0.069 0.105 0.112 0.134 -0.033 -0.033 -0.031 (1.40) (2.20) (2.40) (0.73) (0.96) (0.93) SECH 0.015 -0.054 -0.148 0.011 -0.126 -0.227 (0.40) (1.12) (0.84) (1.47) SEC3 0.278 -0.055 -0.093 0.283 -0.014 -0.027 (1.15) (1.96) (0.30) (0.79) SECM4 0.082 -0.119 -0.172 0.078 -0.060 -0.077 (1.76) (2.60) (0.91) (1.50) SECI5 0.072 -0.066 -0.124 0.066 0.039 40.042 (0.95) (1.79) (0.55) (0.82) SECI6 0.194 -0.095 -0.134 0.211 -0.001 -0.010 (1.80) (2.62) (0.02) (0.27) SECI7 0.119 -0.054 -0.081 0.111 -0.008 -0.017 (0.89) (1.37) (0.13) (0.38) SECIB 0.058 -0.139 -0.237 0.068 -0.029 -0.035 (1.70) (2.84) (0.38) (0.63) CHI6 0.474 -0.069 -0.078 -0.076 (3.07) (4.49) (4.39) 0111612 0.515 -0.005 -0.006 -0.005 (0.25) (0.40) (0.35) M 0.352 0.048 0.052 0.055 (1.43) (2.12) (2.24) MARRIED 0.108 0.062 0.278 0.254 0.263 -0.188 -0.188 -0.185 (0.85) (2.70) (2.48) (4.32) (5.94) (5.88) WqHAT 2.964 -0.663 3.050 -0.062 (11.97) (1.79) R2 0.391 0.393 0.385 0.177 0.176 0.172 Ug likelihood -727.4 -738.9 -904.4 -909.1 x2 statistic 820 797 355 345 No. of observations 1044 1641 1641 1830 1830 1830 Man dependent variable (I ) 0.502 0.278 (I 0.3989 0.3989 0.3352 0.3352 t statistics in parentheses Note: 1. OWYHAT is estimated tn of expected own hourly wage. (See 'able A.1) 2. 1he coefficients reported for the probit estimations are transformed coefficients to give the marginal effect of each variable on P i.e. = (i j).$ where 0(li) is the standard normal distrihition evaluated at the mean of P, and 8 is the probit coefficient. 3. t statistics reported are of the probit coefficients themselves. - 29 - 3.22 Young Men and Young Women It is of interest to compare the results for young men and young women. The mean values of the independent variables for the two groups are quite similar. In particular, it is worth noting that both the groups have mean schooling of just over eight years, with the men having a marginally higher mean. The deepening of education in Colombia in recent years was mentioned earlier in the introduction. This indicates that this deepening has been of almost equal importance for men and women and that the women have now almost caught up. Education: In these groups of young people (15-24), one would expect education to be a negative influence on participation since many people would still be in their schooling years: in fact those in the labor force are almost bound to be with less schooling and will comprise of the early entrants. This effect is quite pronounced for the men but only barely so for the women, despite the fact that about half of the men are participating while only about a quarter of the women are in the labor force. Clearly, the women are still largely secondary workers. Family Income and Assets: The coefficient of OFLY measures the cross substitution effect and that of FNLY the pure income effect. Both have coefficients in the predicted direction, i.e., negative, but have different magnitudes and significance between men and women. The cross substitution effect on women is stronger as might be expected, while the pure income effect is stronger on the 1/ OFLY - other family labor income. FNLY - family nonlabor income. - 30 - men. Again, this points to the behavior of women as secondary workers. As may be expected, men from wealthier familes are more apt to stay in school longer. Life Cycle Effects Life cycle effects are observable for both young men and women, but in different directions. In the case of men, as might be expected, the responsibilities of marriage along with duties as head of household have a strongly positive effect on the probability of participation. For women, the effect is opposite, marriage would seem to take them out of the labor force. In this age group, only about 11 percent of the men are married while about 26 percent of the women are married. The coefficients of age are quite similar and in the right direction (positive) for both the groups. Home Activity: These variables are relevant for the women only and their coefficients are all in the expected direction. The presence of small children inhibits labor force participation while the presence of a maid or adult relative in the household is quite conducive to it. Background: The location of residence has been used as a proxy for background. In previous work on the determinants of earnings (Mohan, 1981), these proxies were found to have quite significant effects on labor earnings. The normalising sector is sector 2, the poorest part of the city. For both men and women, all the remaining sectors have a negative sign, implying that the poor are more likely to leave school and participate early. Note, however, that none of the coefficients for women are significant. These results do not imply that the poor are disadvantaged in terms of participation. They reinforce the earlier result that people from wealthier families tend to stay in school longer and thus have a lower - 31 - probability of participation at a young age. The other background variable tested is for whether a person is a recent migrant. For men, the coefficient is significant and positive implying that, ceteris paribus, young migrant men are more likely to be in the labor force than natives in a city. This is again as might be expected since it may be presumed that recent young male migrants have come to the city in search of jobs. For women, however, the coefficient is not significant. It should be recalled that a large proportion of young women workers are live-in domestic servants many of whom are recent migrants (MOhan, 1980), but they have been excluded from this sample. Own Wage: The second probit estimation (Probit B) includes an estimated value of expected hourly wage. 1/ The key determinants of wage are education and experience with location of residence used as background proxies. The intention in this estimation was partly to obtain a direct estimate of own wage elasticity of participation but also to distinguish education as a quality of leisure variable as distinct from its contribution to expected earnings. It was found, however, that education and estimated own wage were too collinear and so education was dropped from this equation. Standard theory predicts a positive association between LFP and expected wage: the higher the expected wage, the more likely a person is to participate in the labor force. In these samples of young men and women, the main alternative is to invest in education for higher future earnings: hence LFP is found to be negatively associated with higher expected earnings. The effect is much more pronounced for young men than for women. 1/ See Appendix I for the estimation. - 32 - The coefficients of the other variable which are common between Probit A and Probit B are very similar, the confidence in their magnitudes is therefore reinforced. In summary, these estimates indicate that the participation of young women in the labor force is far more dictated by life cycle and home activity considerations than by labor market career aspirations despite a considerable deepening of the education system and almost equal investment in education by men and women. Many more women get married earlier than men and enter the child bearing stage and can therefore not enter the labor market. The presence of household help in the shape of a domestic servant or of a live-in adult relative is quite significant in assisting participation in the labor market.1/ Given these results, the indications are that women still enter the labor market mainly as secondary workers. For the young men, the estimations give largely expected results. Investment in education for later higher earnings takes precedence over current labor force participation. It is clear that family wealth assists in this process: the negative coefficients for the richer residence locations in the city support this conclusion. Life cycle considerations are also evident in that married young men and heads of household are more likely to participate-but there may be some self selection here. 1/ There is a simultaneity problem in the use of this variable. The presence of a domestic servant cannot really be regarded as exogenous. The decision to participate often involve the contingent hiring of domestic help. The causation then runs the opposite way. - 33 - 3.23 Prime Age Unmarried Women and Married Women Table 4b reports the results on LFP for prime age unmarried women (UF) and for married women (MF). These results are of great interest since they pertain to women over similar age groups 1/ with the main difference being their marital status. Unmarried women include widowed, divorced and separated women. Married women include only those with husband present. The two groups are quite similar in characteristics except that the UF are somewhat better educated (average YRSEDU = 6.85) than the MF (average YRSEDU = 6.38). At the same time there is a paradoxical result that average expected earnings are lower for the UF. The UF also have higher family non- labor income (FNLY) but naturally lower other family labor income (0FLY). Presumably this is partially explained by the fact that about 44 percent of the UF are widows who may be expected to have higher shares of non-labor income from bequests, insurance and transfers. The participation rate of the UF (0.69) is more than twice that of the MF (0.28). The striking result for the UF is that most of the coefficients are not significant with the exceptions of widowhood and household headship. Clearly, these women appear to have little choice in that they have to work as breadwinners of their families. Higher education and higher expected earnings do have a small but significant positive effect as might be expected. Family Income and Assets As might be expected, both other family Labor and non-labor income reduce the probability of a married women working. The cross-substitution 1/ The married women include all married women of all ages while UF are in the 25-54 age group. - 34 - Table 4b: TE PARMICIPATION DECISIGN (ehotomous dependent variable) Prinfa SLngle lma arried Women Variable Man OLS Probit A2 Probit B2 lan OLS Probit A2 Probit B2 00NSTAr 0.412 0.475 0.345 -0.190 -0.377 -0.518 (0.74) (1.22) AGE 36.8 0.025 0.022 0.016 35.54 0.024 0.037 0.027 (0.82) (0.96) (0.71) (3.15) (5.61) (4.07) AGEsq 1436.9 -0.0005 -0.0004 -0.0004 1394.4 -0.0003 -0).0003 -.0003 (1.27) (1.63) (1.30) (3.53) (5.92) (4.57) YRSEDI 6.85 0.013 0.014 6.38 0.028 0.031 (1.90) (2.84) (5.84) (8.77) OFLY 8.17 -0.002 -0.002 -0.002 11.06 -0.004 -0.004 -0.004 (0.90) (1.39) (1.23) (3.26) (4.13) (4.11) FNLY 2.23 -0.003 -0.003 -0.002 1.43 -0.005 -0.007 -0.007 (0.55) (0.95) (0.77) (1.66) (2.34) (2.53) HHHEAD 0.51 0.191 0.221 0.223 (2.97) (4.58) (4.71) MG3 0.07 -0.161 -0.171 -0.150 0.088 0.082 -0.092 -0.097 (1.57) (2.31) (2.05) (1.60) (2.40) (2.57) FWEUNI 0.101 0.053 0.049 0.040 (1.07) (1.41) (1.19) HUSBWU 0.211 0.001 -0.004 -0.021 (0.03) (0.16) (0.83) SECrl 0.010 0.046 0.151 0.012 0.050 0.039 (0.18) (0.61) (0.38) (0.43) SECIM 0.275 -0.064 -0.066 0.295 -0.027 -0.027 (0.85) (1.19) (0.64) (0.87) SECMA 0.066 -0.136 -0.159 0.074 0.003 0.014 (1.18) (1.95) (0.05) (0.31) SECI5 0.067 -0.012 -0.008 0.061 -0.084 -0.103 (0.11) (0.09) (1.25) (2.02) SECI6 0.195 -0.078 -0.085 0.195 0.005 0.013 (0.94) (1.42) (0.11) (0.39) SECI7 0.121 0.043 0.060 0.108 0.023 0.022 (0.45) (0.83) (0.41) (0.55) SECIB 0.070 0.038 40.055 0.068 -0.029 -0.011 (0.3)) (0.59) (0.40) (0.21) alh, 0.781 -0.046 -0.053 -0.053 (2.56) (4.05) (4.09) llnbl2 0.710 -0.013 -0.017 -0.020 (0.84) (1.47) (1.77) MR9L 0.293 0.112 0.126 0.125 (3.38) (5.34) (5.35) NFDCM 3.394 -0.009 -0.024 -0.021 (1.54) (3.38) (2.97) WIDow 0.441 -0.079 -0.094 -0.096 (1.26) (2.00) (2.07) Ownw1 2.748 40.096 3.120 0.175 (3.03) (7.73) 0.131 0.128 0.114 0.089 0.093 0.080 4 likelihood -387.17 -392.93 -1180.5 -1195.9 X statistic 96 85 213 182 No. of observations 701 701 701 2169 2169 2169 I'aan of dependent variable C(1 ) 0.688 0.280 0(1 0.3538 0.3538 0.3372 0.3372 t statistics in parentheses. Notes: 1. (NWYHAT is estimated in of expected un hourly age. (See Thble). 2. Ihe coefficients reported for the probit estimations are transformed coefficients to give the marginal effect of each variable on p. i.e. (4 = (i).B ) where 0(i) is the standard norual distribution evaluated at the mnan of P, and ii 6 is the probit coefficient. 3. t statistics reported are of the probit coefficients thenselves. - 35 - effect of other family labor income, mainly husbands earnings, is quite significant in magnitude and more important than non-labor income. The negative pure income effect is quite small. Surprisingly, despite the higher mean value of FNLY for UF, it does not have a significant effect on their participation--reinforcing the conclusion that the majority of UF essentially have to work. Life Cycle Effects Life cycle effects are quite important for the UF in terms of household headship and widowhood as mentioned earlier. For married women, age is an important factor. Older married women are more likely to work, presumably after their child bearing period is over. The probability of their LFP increases up to age 61 according to these estimates. In the married women sample, about 10 percent are in "Free Union" 1/, i.e. not legally married but living together. The hypothesis was that these women would be more likely to participate if this status involves a greater state of uncertainty. This seems to be only partially confirmed: the coefficient is positive but not statistically significant. This result implies that for the majority of women in this category, the non-legality of their "marriage" is not important in terms of stability and they behave pretty much as if they were married. Home Activity These variables are relevant for the married women only. The presence of small children below the age of six inhibits their participation significantly, while the presence of older children is not important in the participation decision. The magnitude of the effect of young children on the 1/ Union Libre. - 36 - probability of LFP is of the same order as for young women. The presence of a domestic servant or adult relative (MREL) makes a highly significant effect on LFP making it more likely. Indeed, the magnitude of the effect just about neutralizes the negative effect of a young child. This positive effect of domestic help is much larger for this group than for young women. For young women, the young child will mostly be the first while this sample includes later born children. This result suggests that domestic help is more likely to be seen as a substitute for the mother's care after the first born. NROOMS is the number of rooms in the house being used as a proxy for the volume of housework seen as a competing activity for the married women. The coefficient has the expected negative sign but this may also be interpreted as a proxy for wealth. Background As for young women, the location of residence does not emerge as a significant proxy for background affecting the decision to participate in the labor force for both the UF and MF. Again, there is little evidence of spatial disadvantage in terms of labor force participation. Unlike the men, however, recent migrants are less likely to participate among both the groups. The implication is that there are few women migrants who come in search of jobs: most probably come as spouses and are less likely to work. Own Wage; Unlike the results for the young men and women, a higher expected wage does increase the probability of women working. The effect is more pronounced for married women than for the unmarried women. Clearly, married women have a greater choice of not working and this is reflected in the estimates made. - 37 - Again, as for young men and women, the estimates in Probit B are quite similar to the estimates in Probit A for both the UF and MF for the variables that are commmon between the two equations. In summary, the estimates for prime age unmarried women indicate that they behave much more like primary workers, somewhat like prime age men, while the married women behave much more like secondary workers. Their propensity to participate is affected much more by greater schooling and expected earnings, while at the same time inhibited by the presence of small children and other home activities. These results indicate that a tightening in the labor market would indeed increase the participation of married women as has been observed in Colombia in the late seventies. Moreover, higher average levels of education among women will also lead to higher participation--or at least availability for participation. The availability of child care facilities undoubtedly facilitates participation but the results for young women indicate that this has its limitations. It is quite likely that women will still opt out of the labor force for the first child but the availability of child care could hasten their return to the labor force. 3.3 The Total Supply of Labor: Hours Worked in the Labor Market 3.31 Interpreting Tobit and Heckman Coefficients This section gives estimates for the full labor supply function seen as a continuous function from zero hours worked to the maximum number of hours supplied. The Appendix gives the derivation of two different methods of estimating these functions. The first method is Heckman's procedure which utilizes information from the reduced form participation equation (Model I) to derive estimates of selectivity bias (X) . These are then used (i) to make 1 estimates of the wage function corrected for selectivity bias and (ii) the - 38 - hours supplied by workers, again corrected for selectivity bias. This is Model III. The second method is a direct Tobit estimation of the censored dependent variable of hours supplied in the market. This itself is done in two ways: first of the reduced form hours equation (Model IB) and second with estimated wage from Model III. This is Model IV. The two Tobit estimations correspond to Probit A and Probit B. The last section in Appendix II explains the interpretations of different coefficients. For convenience all the Tobit coefficients reported here correspond to the marginal effect of each independent variable y on the aE(h.) expected value locus of hours supplied i.e., ay evaluated at the mean. The Tobit coefficients are therefore "scaled down" by @(I.) evaluated at the mean (which is merely the proportion of people working in the sample). The OLS coefficients from Heckman are reported directly. 3.32 The Tobit and Heckman Estimates Tables 5(a,b,c,d) give the estimates for Model I, Model III and Model IV as explained above for young men, young women, prime age unmarried women and married women respectively. The tobit coefficients are as scaled above. Overall, the results of the tobit estimates are similar in direction to the probit results and are therefore not discussed in detail. The tobit coefficient estimates (suitably transformed) are of a similar order of magnitude (evaluated at the mean) and in the same direction as the OLS coefficients. In general, the tobit coefficients give a higher slope than the OLS coefficients and have a higher level of significance. Further, the tobit coefficients from the reduced form equation (Model IB) are similar in magnitude to the structural equation with estimated wage as a determinant (Model IV). - 39 - In contrast to these generally satisfactory results, it is found that the coefficients in the Heckman procedure (Model III) are quite different from those obtained in the tobit analysis. As shown in the Appendix II, the Heckman coefficients should in principle identical to the tobit coefficient, i.e. 6. . In fact, the tables present estimates of (I.) 6. for the tobit regressions while the Heckman coefficients given are 6. . Almost all the Heckman coefficients are small in magnitude and not significantly different from zero statistically. The reason for this feature of the results is that in these samples, among the people who do work, there is relatively little variance in their weekly hours. The standard deviation in the number of hours worked by workers is in the region of .25 to .37 in all the samples. Moreover, for those who do work the mean is relatively high: between 44 and 47 hours in each group. This suggests that there are some costs of entry to the labor market. It would appear that if a person decides to work he is faced with a decision to work a substantial number of hours or not at all. In addition, for married women in particular, the decision to participate also often involves the simultaneous decision to use childcare services. Hence she participates only if the number of hours worked and the wage are high enough to offset these costs. This is merely another way of saying that her reservation wage is high. If these speculations are correct, the Heckman procedure is inappropriate for this problem without adequate corrections for costs of entry. 1/ The correction for sample selectivity is not "enough" to generate the "true" labor supply schedule. There is another problem at issue here. The dependent variable used in this study is weekly hours of work in a reference week as reported in the survey. There is considerable inconclusive 1/ See Cogan (1980). - 40 - Table 5a: THE SUPPLY OF LABOR SCHEDULE: Young Men Model IV Model III Model I (Reduced form) (w/estimated wage) Heckman: worker only Mean OLS Tobit 2,3 Tobit 2,3 OLS Mean CONSTANT -140.7 -194.1 -176.8 96.64 (5.23) (6.92) (6.33) (2.44) AGE 19.26 15.95 19.17 20.64 -2.30 20.61 (5.63) (6.56) (6.95) (0.59) AGESQ 379.2 -0.308 -0.394 -0.397 0.06 431.4 (4.20) (5.30) (5.30) (0.66) YRSEDU 8.24 -2.72 -2.039 (4.68) (11.06) OFLY 11.95 -0.057 -0.121 -0.129 -0.093 9.22 (1.39) (2.63) (2.85) (1.67) FNLY 2.11 -0.560 -0.983 -1.03 -0.285 1.02 (5.24) (6.50) (7.01) (1.31) HHHEAD 0.116 7.91 3.94 4.63 -0.177 0.227 (3.44) (1.55) (1.87) (0.09) MIG3 0.121 4.45 4.16 4.46 1.445 0.154 (2.84) (2.86) (3.17) (1.04) SECTI 0.015 -5.01 03.25 (1.17) (0.92) SECT3 0.278 -3.05 -2.55 (2.00) (1.74) SECT4 0.082 -5.83 -5.21 (2.69) (2.47) SECT5 0.072 -3.95 -3.41 (1.77) (1.52) SECT6 0.194 -4.93 -4.29 (2.94) (2.66) SECT7 0.119 -3.50 -2.43 (1.80) (1.33) SECT8 0.058 -6.86 -7.66 (2.63) (2.75) MARRIED 0.108 4.31 2.54 1.88 0.849 0.210 (1.84) (1.02) (0.74) (0.49) OWNYHAT 1 -21.70 -8.12 2.894 (12.08) (3.10) SIGMA 17.00 17.16 (28.31) (28.69) LAMBDA -4.33 0.724 (1.78) R2 0.402 0.113 Log likelihood -4548.2 -4558.9 No. of Observations 1641 1641 1641 823 Mean Dependent variable 23.97 23.97 23.97 47.74 (I) 0.504 0.504 94 0.37 0.37 t statistics in parenthesis. Notes: 1. OWNYHAT is log of Estimated Expected Hourly Wage. 2. The Tobit coefficients reported are transformed to make them comparable to OLS estimates. The coefficients reported are E(h) 1 i 3. t statistics reported are of original Tobit coefficients. 4. See text. - 41 - Table 5b: THE SUPPLY OF LABOR SCHEDULE: Young Women Model IV Model III Model I (Reduced form) (w/astimated wage) Heckman: worker only Mean OLS Tobit 2,3 Tobit 2,3 OLS Mean CONSTANT 9/.35 -189.2 -183.Z 140.9 (3.92) (6.88) (6.70) (1.1/) AGE 19.35 9.62 16.45 16.50 -4.52 20.91 (3.71) (5.92) (5.92) (0.46) AGESQ 382.9 -0.164 -0.336 -0.334 0.105 443.0 (2.47) (4.84) (4.82) (0.51) YRSEDU 8.11 -0.984 -0.537 (5.47) (3.27) OFLY 11.89 -0.056 -0.072 -0.074 0.089 10.84 (1.44) (1.79) (1.86) (1.16) FNLY 1.78 0.362 -0.444 -0.470 -0.466 1.42 (3.35) (3.38) (3.61) (1.46) HHHEAD 0.02 8.27 3.93 4.24 0.081 0.049 (2.57) (1.49) (1.61) (0.02) MIG3 0.134 -0.060 -0.659 -0.554 4.92 0.118 (0.04) (0.49) (0.41) (2.16) SECTI 0.011 -5.89 -10.47 (1.26) (1.69) SECT3 0.283 -0.88 -1.57 (0.60) (1.13) SECT4 0.078 -2.30 -2.71 (1.11) (1.34) SECT5 0.066 1.34 1.43 (0.61) (0.71) SECT6 0.212 0.26 -0.61 (0.17) (0.41 SECT7 0.112 -1.45 -1.30) (0.77) (0.73) SECT8 0.068 -0.20 -0.97 (0.08) (0.44) CHIL6 0.474 -3.70 -3.45 -3.36 -1.61 0.385 (5.28) (4.88) (4.75) (0.80) CHIL612 0.515 -0.25 -0.10 -0.091 0.44 0.403 (0.42) (0.18) (0.16) (0.46) MREL 0.352 2.16 2.09 2.19 -1.38 0.397 (2.08) (2.15) (2.25) (0.73) MARRIED 0.263 -9.84 -7.45 -7.37 0.141 0.263 (7.23) (5.90) (5.84) (0.03) OWNYHAT 1 3.05 -4.23 -13.28 2.87 (3.10) (5.12) SIGMA 774.8 783.4 (14.63) (14.61) LAMBDA -6.68 1.36 (0.71) R 2 0.193 Log likelihood -3307.4 -3313.1 No. of Observations 1830 1830 1830 509 Mean Dependent variable 12.78 12.78 12.78 45.83 (l ) 0.281 0.281 04 0.26 0.26 t statistics in parenthesis Notes: 1. OWNYHAT is log of Estimated Expected Hourly Wage. 2. The Tobit coefficients reported are transformed to make them comparable to OLS aE(h i) estimates. The coefficients reported are Eyx - 0 (see text). 3. t statistics reported are of original Tobit coefficients. - 42 - Table 5c: THE SUPPLY OF LABOR SCHEDULE: Prime Single Women Model IV Model III Model I (Reduced form) (w/estimated wage) Heckman: worker only Mean OLS Tobit 2,3 Tobit 2,3 OLS Mean CONSTANT 32.19 11.23 10.56 62.75 (1.57) (0.55) (0.51) (2.79) AGE 36.77 0.776 1.11 1.03 0.192 35.32 (0.70) (1.01) (0.95) (0.18) AGESQ 1436.9 -0.020 -0.025 -0.023 -0.008 1321.4 (1.44) (1.80) (1.71) (0.52) YRSEDU 6.85 -0.217 0.028 (0.90) (0.11) OFLY 8.17 -0.108 -0.114 -0.107 -0.024 7.71 (1.28) (1.39) (1.31) (0.27) FNLY 2.23 -0.195 -0.163 -0.151 -0.148 2.12 (1.13) (0.93) (0.90) (0.94) HHHEAD 0.505 10.35 10.52 10.85 3.27 0.531 (4.43) (4.60) (4.79) (0.88) MIG3 0.066 -4.08 -6.27 -5.89 3.33 0.052 (1.09) (1.72) (1.64) (0.76) SECT1 0.010 11.50 8.47 (1.20) (0.77) SECT3 0.275 -1.60 -2.29 (0.58) (0.85) SECT4 0.066 -5.00 -5.54 (1.19) (1.46) SECT5 0.067 2.49 1.78 (0.60) (0.47) SECT6 0.195 -1.67 -2.54 (0.56) (0.87) SECT7 0.121 2.39 2.39 (0.69) (0.60) SECT8 0.070 -1.66 -0.56 (0.37) (0.11) WIDOW 0.441 -2.39 -3.10 -3.50 0.514 0.405 (1.04) (1.39) (0.15) (0.21) OWNYRAT 1 2.75 0.255 -6.91 3.01 (1.59) (3.57) SIGMA 22.69 22.80 (24.46) (24.39) LAMBDA 8.47 - 0.717 (1.15) R2 0.108 0.126 Log likelihood -2576.7 -2580.0 No. of Observations 701 701 701 482 Mean Dependent variable 31.79 31.79 31.79 46.17 0(li) 0.690 0.690 04 0.48 0.48 t statistics in parenthesis. Notes: 1. OWNYRAT is log of Estimated Expected Hourly Wage. 2. The Tobit coefficients reported are transformed to make them comparable to OLS aE(hi) estimates. The coefficients reported are ax =(I1 )0 (see text). 3. t statistics reported are of original Tobit coefficients. 4. See text. - 43 - Table 5d: THE SUPPLY OF LABOR SCHEDULE: Married Women Model IV Model III Model I (Reduced form) (v/estimated wage) Heckman: worker only Mean OLS Tobit 2,3 Tobit 2,3 OLS Mean CONSTANT -2.84 -36.43 -41.87 28.02 (0.56) (6.26) (6.44) (0.90) AGE 35.54 1.01 1.68 1.30 0.888 34.08 (3.99) (5.95) (4.91) (1.05) AGESQ 1394.4 -0.014 -0.023 -0.019 -0.013 1252.5 (4.66) (6.54) (5.64) (1.16) YRSEDU 6.38 0.798 1.105 (5.14) (5.58) OFLY 11.06 -0.187 -0.186 -0.179 -0.104 10.65 (4.22) (4.11) (4.00) (0.85) FNLY 1.43 -0.278 -0.342 -0.358 -0.458 1.11 (2.62) (2.49) (2.64) (1.45) MIG3 0.088 -2.58 -3.55 -3.82 2.82 0.066 (1.54) (2.09) (2.26) (0.67) FREUNI 0.101 1.38 1.53 1.30 -2.69 0.105 (0.86) (0.99) (0.84) (0.83) HUSBLU 0.211 -0.91 -0.668 -1.31 -3.93 0.189 (0.76) (0.58) (1.15) (1.56) SECTI 0.012 0.873 1.00 (0.20) (0.26) SECT3 0.295 -0.486 -0.734 (0.35) (0.54) SECT4 0.074 1.45 1.46 (0.71) (0.76) SECT5 0.061 -3.88 -4.52 (1.76) (1.96) SECT6 0.195 -0.84 0.204 (0.54) (0.14) SECT7 0.108 0.770 0.947 (0.41) (0.51) SECTS 0.068 -0.639 0.174 (0.27) (0.08) CHIL6 0.781 -2.73 -2.68 -2.70 -3.64 0.714 (4.58) (4.29) (4.32) (2.33) CHIL612 0.710 -0.426 -0.69 -0.82 0.465 0.673 (0.82) (1.41) (1.66) (0.42) MREL 0.293 4.88 5.40 5.42 4.08 0.373 (4.47) (4.59) (4.60) (1.31) NROOMS 3.39 -0.448 -1.13 -0.976 -1.05 3.36 (2.43) (3.36) (2.99) (1.33) OWNYHAT 3.12 6.34 -2.45 2.88 (5.08) (0.62) SIGMA 16.36 16.47 (16.86) (16.71) LAMBDA 11.56 1.47 (1.33) R2 0.057 0.115 Log likelihood -4082.7 -4094.6 No. of Observatione 2169 2169 2169 608 Mean Dependent variable 12.64 12.64 12.64 44.91 0(i 0.285 0.285 e 0.26 0.26 t statistics in parenthesis. Notes: 1. OWNYHAT is log of EsLimated Expected hourly wage. 2. The Tobit coefficients reported are transformed to make them comparable to OLS 8E(h ) estimates. The coefficients reported are ax 0 1 (see text). 3. t statistice reported are of original Tobit coefficients. 4. See text. - 44 - debate in the literature on which is the correct dependent variable to use in a labor supply study. Annual hours accounting for vacation time and also other weeks not worked are often suggested as a better measure to use. It is probably the case that the use of this measure would at least inject greater variance in th'e hours variable. In view of all these problems, the specification of the model for using the Heckman procedure is simply inadequate here. As is shown in Appendix II, the tobit coefficients can be used to obtain three types of information: (i) The marginal effect of each explanatory variable on the latent tobit index. This is the estimated coefficient a. (ii) The marginal effect on the observed hours supplied. This is *(I)6. , evaluated at li. In a study of hours supplied this is the marginal effect of interest, i.e., the effect of an explanatory variable on actual hours supplied. (iii) The proportion of the effect that is due to variation in the hours supplied given that a person is participating. This is given by: (I. W(I.)2 1 1 and (1 - 8) would be the effect on the probability of the person participating. The value of 8 is noted in the tables in each case. In principle, the probit coefficients should be equivalent to (1 - e) of the tobit coefficients. 3.4 Comparison of Estimates Across Models Table 6 gives the elasticities of participation and labor supply for each of the models estimated. The elasticities given are for only a few key variables: OWNYHAT (expected own wage), OFLY (other family labor income), FNLY - 45 - (family non-labor income) YRSEDU (years of schooling) and*CHIL6 (number of children less than 6, for women). These are discussed in turn for each group. The elasticity for OWNYHAT measures the gross uncompensated income elasticity; that for OFLY measures the cross substitution effect of (mainly) spouse's income and that for FNLY measures the pure income effect. The elasticity with respect to education is important in a situation where there is continuing educational deepening as is occurring in Colombia. Finally, the elasticity with respect to the presence of young children gives an estimate of the importance of childcare facilities in assisting married women to work. 3.41 Young Men and Young Women The own wage elasticity of participation as well as labor supply is highly negative for young men and women. The negative effect is much greater for young men. It seems that in both cases, for those who do work, it is the poorer who tend to work longer hours since the Heckman elasticities are significant and negative in both cases. The probit and tobit characteristics are roughly comparable for men. For women, the participation elasticity is not statistically significant. These results, though plausible in that young men and women expecting higher income are likely to delay participation, are at odds with most estimates in the U.S. where own wage elasticity is invariably positive. Results are, however, always difficult to compare across studies became of different model formulations and variable definitions. In Jen's (1983) study of urban female labor force participation in Brazil she found the elasticity for young women (daughters) to be about .2. Further disaggregated into age groups of 15-19 and 20-24 the elasticities were about -1.0 and 0.1 respectively--which are consistent with our results. Moreover, her income variable was a monthly income measure which tends to inflate the - 46 - wage elasticities if there is a positive association between hours worked and wage. The estimates of cross substitution elasticity with respect to other family income are generally not significant for these groups. This is somewhat surprising since the family nonlabor income elasticities are significantly negative and are between -0.05 and -0.1 for both the groups. These are quite consistent with most estimates in other studies. As might be expected from the OWNYHAT elasticity estimates, the education elasticities are significantly negative and much more so for the men for all the same reasons. For young women, the presence of young children is a major inhibitor of labor force participation. The magnitudes of the tobit and probit elasticities are very similar. It would seem that the presence of the child is important in the decision to participate but once the decision is made it has little effect on hours supplied. 3.42 Prime Age Unmarried Women (UF) and Married Women (MF) For both these groups, the participation elasticities with respect to own expected wage and to education are positive. For UF the estimate is about 0.14-0.15 while for MF the estimate is about 0.6-0.7. These are quite consistent with other studies although, as documented by Killingsworth (1983), there is still no general agreement on their magnitude. Elasticities as high as 4 to 5 have been calculated by Heckman (1980), but most estimates are in the range of 0.5 to 2.0 (e.g. Layard, Barton and Zabalza (1980) for the U.K., Schultz (1980) for the U.S., Jen (1983) for Brazil). 1/ In both cases, the Heckman elasticities are negative and the tobit elasticities are lower than 1/ See Killingsworth (1983) for a full survey. - 47 - Table 6: ELASTICITIES CF PARTICIPATIC AND CF SUPPLY CF IABOR (All elasticities calculated at the mean) Work hours Workers Only Participation Model I Model I Model IV Model III Variable 01S Proit A Probit B OIS Tobit Tobit OS Young Men WNYHAT 1 -).828 4 -1.321 -0.552 4 -0.905 -0.170 OFLY -0.024* -0.048* 0).048* -0.028* -0.060 -0.064 -0.018* FNLY -0.050 -0.097 -0.101 -0.049 -0.087 -0.091 -0.006* YRSEDU -0.739 -1.051 -0.936 -0.701 Young Women OWNYAT 1 -0.208 4 -0.223* -0.304 4 -0.331 -0.290 (FLY -0.043* -0.086 -0.086 -0.052* -0.067* -0.069* 0.021* FNLY -0.045 -0.058 -0.064 -0.050 -0.062 -0.065 -0.015* YRSEDU -0.233* -0.233 -0.625 -0.341 GIIL6 -0.118 -0.133 -0.129 -0.137 -0.128 -0.125 -0.014* Prine Age Unmrriel Wnen OWNEAT 1 0.155 4 0.140 -0.007 4 0.008* -0.150 (FLY -0.024* -0.024* -0.024* -0.028* -0.029* -0.028* -0.004* 0NLY -).010* -0.010* -0.007* -0.014* -0.011* -0.011* -0.007* )RSEDU 0.129 0.139 -0.047* -0.006* Married Wcmen OWNYHAT 1 0.772 4 0.624 0.610 4 0.501 -0.055* OFLY -0.158 -0.158 -0.158 -0.164 -0.163 -0.157 -0.025* NLY -0.025 -0.036 -0.036 -0.031 -0.039 -0.040 -0.011* YRSEJIU 0.637 0.705 0.402 0.557 HIL6 -0.128 -0.148 -0.148 -0.168 -0.165 -0.167 -0.058 Notes: 1. All elasticities for OWNYHAT are with respected to Estinated Expected Hourly Wage. 2. All Tobit elasticities are for the Expected Value Lowus (not for the Tobit Index). 3. * 11e elasticities umrked with an asterisk are derived frn coefficients not sigifficant at the 5% level. 4. Calculated throigh the coefficient on education in the participation or hours equation and that in the wage equation. - 48 - the probit ones. The implicaton is that people earning lower wages are apt to work longer hours. This has been documented for Bogota workers in an earlier paper (Mohan and Hartline, 1984). This effect is much stronger for the UF as might be expected. Many of them have to work out of necessity (about 40 percent are widows) and could be working long hours at low wages because of earlier low educatonal investments. Jen (1983) reports similar participation elasticities for wives (0.9) and female household heads (0.2) in urban Brazil. The positive participation elasticities are consistent with, indeed explain, the increased participation of women in the late seventies in Bogota documented in section I. The rise in average real wages was also documented. The elasticities for both OFLY and FNLY are not significant for UF, as might be expected. For MF, however, both are significantly negative but unlike the young men and women, OFLY (mainly spouse's income) is much more important than FNLY. The magnitude of the elasticity with respect fo OFLY for MF is about -0.15 to -0.16, while that for FNLY is about -0.03 to -0.04. The latter are quite consistent with Schultz's estimates for the U.S. but he finds a much higher elasticity for spouse's income (about -1.0) than found here. The participation elasticity with respect to CHIL6 for MF is quite comparable (about -0.15) to that for YF. In addition, in this case, the tobit elasticities are greater in magnitude: married women with small children who do work tend to work fewer hours. Once again, the Heckman elasticities are generally not statistically significant. - 49 - IV. LABOR SUPPLY IN A DEVELOPING METROPOLIS: WHAT HAVE WE LEARNED? 4.1 Supply of Labor in Bogota The tightening of the labor market in the late seventies in Colombia has provided an interesting data set for investigating the determinants of changes in urban labor supply in a situation of relatively rapid economic growth. Although changes in the overall labor supply in urban Colombia are not readily discernible because of compensating changes in different components of the labor force the changes that have been occurring have provided important pointers for the future. The estimations made in this paper provide useful guides to understanding the forces underlying the observed changes. The estimates show that the main component of the labor force that can be expected to expand rapidly with rising real wages is that of married women. Until the expansion of education is complete there will be a continuing diminution in the number of young people--both men and women--who will enter the labor force at young ages. The same forces that induce greater participation of married women slow down the entry of young women into the labor force (and vice versa). Rising real wages are associated with increases in marginal returns to human capital which, in turn, is related to overall increases in productivity. The increase in marginal returns to human capital that is observed is really an increase in the average endowment in human capital. In fact, I have shown elsewhere (Mohan 1981), that increases in the stock of human capital in Colombia have tended to decrease the returns to education over time. This is consistent with increasing average real wages if the average stock of human capital is increasing. This seems to have been the case in Colombia in the seventies and was reflected in increases in the - 50 - participation of prime age women. At the same time, women have been catching up with the men rapidly in terms of average educational attainment: the estimates show marked negative association of the probability of participation with expected wage for both young men and women. Clearly, the achievement of high education levels has become important for women. The elasticity estimates found for labor force participation were found to be quite robust across different estimation methods and therefore can be relied on. It would seem that married women are the most wage responsive group, followed by prime age unmarried women. There is clearly an element of necessity in the participation of unmarried women--specially those who are household heads. They are therefore less responsive to wage increases. While estimates for prime age men have not been reported here, it was found that, since almost all are in the labor force anyway, their participation is not wage responsive. One consequence of the results for young men and women is that as these cohorts move up in age we can expect much higher levels of desired labor force participation in the future. Once there results are coupled with the association of LFP with fertility, the future expansion of women in the labor force can be appreciated further. The presence of young children was found to be a key determinant of the participation of married women. As fertility declines, as it has in Colombia, there will be fewer married women of prime age with young children. We can therefore expect the combined effect of higher educational attainment with lower fertility to work itself out over the rest of the decade leading to marked increases in the desire of women to join the labor force. Coupled with this we can also expect a higher demand for childcare services. Both of these issues are important for policy in the coming decades. - 51 - The results on the hours of work decision are less robust than those on the participation decision. The fact that the poor work longer hours in Bogota leads to evidence of little or no elasticity of labor supply with respect to wage. Indeed, the labor supply curve seems to be backward bending, once the decision to participate has been made. This indicates that there is a limitation in choices available to poor women: they need income for survival and are willing to work long hours if a job is at all available. This is quite consistent with the observation that there was a major increase in the participation of poor women in the tightened labor market in urban Colombia in the late seventies. It is possible, however, that somewhat different results might have been obtained had the measure of labor supply been something different from weekly work hours. Migrants do not appear to have any disadvantage in terms of LFP as is often supposed in the literature. Earlier papers have shown (Mohan (1981), Mohan and Hartline (1984)) that migrants are not any worse paid than non- migrants, if anything they are marginally better off. The results for young men show that migrants have, ceteris paribus, higher LFP than others. It must therefore be concluded that migrants are neither disadvantaged in terms of getting a job nor do they earn less than others once they do have a job. Earlier work has found the location of residence of workers to be a good proxy for ability and background in that equivalent people from richer neighborhoods are found to earn significantly more than those from poorer neighborhoods. No such differences are found to suggest any degree of locational disadvantage in labor force participation. In fact, the probability of LFP of equivalent people from poor neighborhoods is found to be marginally higher than those from better off neighborhoods. The poor simply have to work. - 52 - 4.2 Econometric Considerations Considerable attention has been given to econometric considerations in the estimation of both participation and hours worked in this study. I am not aware of other similar work on developing countries, except for Jen (1983) and this did not extent to estimation of hours worked. It is therefore difficult to compare results. Jen's results for urban Brazil are quite similar, however. As household level data sets get more easily available in developing countries we can expect an outpouring of this of work in the future. It is therefore important to give attention to estimation issues so that biased estimates do not lead to wrong understanding. There have been a number of studies using OLS for participation. My results suggest that OLS is a good first approximation and is therefore better than no estimation. Since the direction of bias is known, viz. OLS tends to understate the effect of different variables on LFP, a reasonable understanding of the participation decision can be achieved by merely doing OLS estimates. However, there is no question that the estimates are seriously biased. The probit estimates were found to be quite robust and are therefore recommended as the best estimates. The hours of work estimates from the tobit analysis suffered from relatively little variance in the hours of work of those who do work so little additional information was gained in addition to the probit estimates. Given the significant additional costs and difficulties of tobit estimation the additional information gained from these estimations was probably not worth the time devoted to them. It would be more useful to use tobit analysis if the data do contain somewhat more variance in the hours of work. - 53 - The Heckman method was found to be not too useful in this study although the wage estimates corrected for selectivity bias were mildly different from the uncorrected ones. The hours worked estimation suffered from the same problem mentioned above and hence the Heckman procedure was inappropriate. It is therefore important to specify the model more carefully and to tailor the estimation accordingly. In any case, the utility of the Heckman procedure is limited if a tobit algorithm is easily available. The choice of variables was found to be important. Much early estimation was done using monthly earnings rather than hourly wages. The reason was that this was the primary data available in the survey and with which much earlier earnings function work was conducted. Since monthly income is a product of monthly hours and hourly wages, the results obtained can be quite misleading. If, for example, LFP is positively associated with expected wage, then the own earnings participation elasticity is highly overstated. This was found to be the case and therefore corrected. Results obtained from a typical census when only monthly earnings are available should therefore be treated with caution. The problem with the choice of labor supply variable has been mentioned earlier: it is probably the case that a measure of annual supply is better than weekly hours and more likely to have greater variance. - 54 - APPENDIX I WAGE FUNCTION ESTIMATES Table A.1 gives the estimated wage functions for each of the four groups under consideration. These estimates are corrected for sample selection bias as set up in Model III (see Section II). There is only mild evidence of sample selection bias in these samples since the coefficients of LAMBDA are not statistically significant. The last row in the table gives the schooling (YRSEDU) coefficient if the regressions are run without LAMBDA. These are higher than the corrected regression in all cases: suggesting that women with higher education who do work are those with a higher propensity to work and therefore the returns to schooling are exaggerated. The difference is least among the prime age unmarried women-- those with the least choice in their decision to work or not to work as shown in the main part of this paper. The experience variable in the equations is not very appropriate. There was no direct information in the sample on the number of years spent working by women. Hence, the usual (Age-Schooling-6) approximation was made. This naturally takes no account of the time taken out by women for childbearing, etc. The low experience coefficients provide evidence of this problem. As shown by Heckman (1980) it is women's experience which is much more susceptible to sample selelctivity bias, but this cannot be shown in this estimation. - 55 - Table A.1: WAGE FUNCTIONS (Corrected for Selectivity Bias) (Dependent variable Ln (hourly wage) from Model III) Prime single Variable Young men Young women women Married women 00NSTANT 1.29 1.70 1.56 1.64 (4.03) (3.96) ' (5.40) (3.69) YRSEDU 0.154 0.138 0.131 0.143 (10.63) (6.65) (9.89) (8.46) EXPER 0.057 0.046 0.026 0.033 (1.51) (1.20) (1.41) (2.24) EXPSQ -0.0004 -0.001 -0.0006 -0.0005 (0.22) (0.57) (1.72) (1.82) SECT1 0.560 -0.068 -0.681 -0.031 (2.94) (0.13) (1.73) (0.09) SECT3 0.168 0.070 0.015 0.196 (2.49) (0.71) (0.12) (1.44) SECT4 -0.044 -0.070 0.054 -0.249 (0.39) (0.46) (0.28) (1.34) SECTS 0.014 0.128 0.285 0.222 (0.14) (0.94) (1.55) (0.95) SECT6 0.100 -0.065 0.276 0.290 (1.25) (0.63) (2.01) (2.02) SECT7 0.164 0.179 0.101 0.167 (1.84) (1.45) (0.66) (0.99) SECT8 0.320 0.260 0.592 0.548 (1.85) (1.68) (3.13) (2.57) LAMBDA -0.105 -0.191 0.273 -0.272 (1.15) (1.73) (1.29) (1.43) R2 0.260 0.311 0.385 0.330 No. of observations 823 509 482 608 Mean of dependent variable 2.89 2.87 3.01 2.88 YRSEDU 0.164 0.162 0.133 0.158 (not corrected for sample selectively bias) - 56 - APPENDIX II DERIVATION OF THE LABOR SUPPLY FUNCTION AND ESTIMATION STRATEGY Assume that each individual decides on his work-leisure choice such that the utility of the whole household is optimized.l/ If there are n individuals in the household the utility function U may be written as U ( 1' 92' *** n; x, Z) (Al) where 11 2... n is the time devoted to leisure by each family member i, x is the Hicksian composite good consumed by the household. 2/ and Z is a vector of other characteristics of the household arrived at through earlier decisions. The household maximizes U subject to the budget constraint: n px = Z h.w. + A (A2) i=l 1 1 and O 5 . 5 T i = 1,...n (A3) where hi is hours worked in the market by individual i, for hourly market wage wi p is the price level. A is the non-labor income of the household. (A3) states that leisure time is non-negative and is, moreover, bounded by the total time available T. 1/ This is not universally accepted. For example, Nwaganga Shields (1980) suggested that in many African countries, both husband and wife provide independently for their material needs and therefore optimize individual utility functions. 2/ This assumes that relative prices between goods do not change over the optimization procedure. - 57 - (A3) can be rewritten as: I. + h. = T i = 1,...n (A4) 1 1 where hi can be interpreted as a slack variable in a programming context and h. > 0. 1 Thus the household maximizes U subject to (A2), (A3) and (A4). Assuming the usual conditions on U, that is, U is a twice differentiable quasiconcave function, and has positive first partial derivatives for all its arguments, the Lagrangian may be written as n n L = U(t , I ,.... ; x; Z) + X{ h w + A - px} + E Y (T - Z - h ) (A5) 1 2 n i=1 ii i=1 i i i Now, for simplicity, assume that individual 1 is optimizing E1 given all other ., j = 2,...n . We can then interpret 1 2...9 to be included in 2n Z, i.e. decisions taken prior to the problem of individual i. Now, the Kuhn Tucker conditions are: -- U -p = 0 (A6) ax x 3L =U - : y5 (A7) 1 1 1 I (U - = 0 at1 1 Z1 (A8) (1 0) (A9) - 0 (A2) - 0 (A4) L + Xw 5 0 (AO) 1h - 58 - h (-y + Xw,)hl = 0 (All) 1 (h 1 0) We have five unknowns (X, y1, h1p 1, x) and five equations and given the conditions on U, the system of equations can be solved for these unknowns in terms of p, wl, A and Z. Our objective is to derive the labor supply function. From (A6): U x p and when l > 0 , from (A7) and (A8) aL- - - U -= 0 1 1 or Yl= U 1 i.e. the marginal utility from leisure. when hi > 0, from (A10) and (All) Yl Xw 1 or U Xw 9. 1 1 U or U w I, p 1 U MU i.e. w1 = 1 when h > 0 A12) 1 (U /p) MU 1 (A12) shows that when h1 > 0 , i.e. when a person works, the market wage is equal to the marginal rate of substitution between leisure and other goods, in other words, the shadow price of the individual's time. - 59 - However, if h1 = 0, from (Al0) and (All) 1 XW1 > 0 U - u l7p' W1 > 0 x We can write the first term as wl*, the shadow price of the individual's time, then w* -w1 > 0 if h = 0 (A13) 1 1 where w* = (A14) x Equations (A12), (A13) and (A14) determine the labor supply schedule and, incidentally also suggest the estimation procedure. (A14) holds for both h = 0 and h > 0 and can be written as wl= f 1, x; Z; p) = f (h1Z ih + A; p; Z) (A15) This is the result of the household's optimization process. If f exists and is continuous, the labor supply function for hi also exists in terms of wit and the other variables. The demand side function for the market wage rate wl is well known and is generally expressed in terms of Education (S), Experience (E) and other background variables (X). i.e. w1 = g(S, E; X) (A16) but w1 is only observed if h > 0. - 60 - The problem now is to estimate these functions econometrically. Estimation Strategy For estimation purposes, (A15) and (Al6) can be re-written as linear functions along with a stochastic disturbance term. For convenience, the individual identifier (within the household) subscripts are now omitted for h and w. Hence, wt = a + a h + a + a A + a Z. + u. (Al7) 1 0 1 i 2yi 3 i 4 1 1 w. = 8 + B 1S. + 8 2E. + X. + e (A18) 2 2 where E(u) = 0, E(u ) = 01 , E(u u.) = 0 1 1 1 i1 i*j 2 2 E(e.) = 0, E(e. ) = o , E(e.e.) = 0 1 1 1J3 i *j Assume that ei, ui are jointly normally distributed, and may be correlated i.e. E(ui ei) = poa12 where p is the correlation coefficient between ei and ui. For convenience, rewrite (Al7) as * * w. = a. h + Z a + u. (A20) 1 1i 1 i 1 and (A18) as w. = X.6 + e. (A21) 1 1 1 Now, the problem in the estimation of (A20) and (A21) is that (i) w. is unobservable. - 61 - (ii) hi and wi are observed only when hi > 0. In a population sample, we have 2 groups of people: those who are currently in the labor force and for whom hi and wi are observed; and others who are not, and for whom hi and wi are not observed. However, Xi and Z are available for all. From conditions (A12) and (A13), hi = 0 if *1 W. - w. > 0 1 1 i.e. Pr(h. = 0) = Pr{(w. - w.) > 0} whose complement is 1 1 1 h.=O0} hs opeeti 1 Pr(h. > 0) = Pr{(w. - w.) > 0} 11 i h.=0 1 At h = 0, w. = Z.a + u. 1 1 1 and hence, Pr(h > 0) = Pr[{X.8 - Z.a) - (u. - e.)}> 0] 1 1 1 1 1 * * = Pr{(X.8 - Z.a) > (u - e} (A22) For hi > 0, market wage wi is observed, from (A21) * E(w.Ih. > 0) = X. + E(e. h. > 0) (A23) = X.a + E {e. (X.a - Z.a) > (u. - e.)) 1 1 1 11 1 - 62 - Hence ei is truncated and the conditional expectation of ei * 0. * Similarly, from (A20), for hi > 0, wi = w and hence, h. = -- (X.8 - Z.ia + e. - u.) 1 G 1 1 1 1 and E(h. h. > 0) --(X*8 - Z.a) + --E{(e. - u.)Ih.> 0} 1hh 1 1 -1 1 1h1 = (.8 - Z. a) + ---E{(e. - u.)(X. - Z*a) > (u.- e.)} S 1 1 G 1 1 1 1) (1. e. 1 1 = (8 - Z ) + ---E{(e. -u.)(e - u > (X*O - Z*)} G 1 1 1 1 1 1 1~ 1 1 (A24) Again it is clear that the conditional expectation of the disturbance term is not zero. However, both (A23) and (A24) can be estimated if estimators for E(e.i.) and E(e. - u.l .) can be utilized as R.H.S. variables. Now, u is - N(0, a12 2 and e. is - N(0, 02) 2 2 Hence e. - u is N(0, a1 + a2 - 2 ala 1 1 2 1 12 - NO0, a32 ) - 63 - Then, using the standard result for a truncated normal distribution 1/ 2 E{(e.- u.)1h. > 0} = 03 ) (A25) 1 1 a 13 (. 3 1 where *(I) and *(I) are the standard normal density and cumulative distribution function respectively, and * * X.8 - Z.c0 I= 1 1 1 03 3 Similarly, a I.) E(e.1h.> 0) = . I 10 1 - (. 3 where 13 cov(e., e. - u.) 01 2 Pa12 I.) i.e. E(e.Ih > 0) = 1 3 * 1 I (A26) 3 write X=*I i -MI) Then 1 * * h. = (X - Z.C + X + V. ) (A27) 1 L 1 1 3 1 11 * 013 and w. = X.B + -- X. - V 1 1 a 1 A 2 3 1/ e.g. Maddata (1983), Appendix. - 64 - where E(V.ilh. > 0) = 0 and E(Vi2 Ih > 0) = 0 (A28) Now, if estimates of X are available, then (A27) and (A28) can be estimated from data on workers only. Now, return to (A22), Pr(h. > 0) = Pr{aI. > u. - e.} 1 1 1 1 u. - e. Pr(h. = 0) = Pr{I. < 1 = (-. 1 1 G3 1 u. - e. and Pr(h. > 0) = Pr{I. > 1 1 i) 1 1 G 1 The likelihood function for the participation decision is then m n L = .J *(I.) .H *(-I,) (A29) 1 = 1 1 i=m+1 1 Maximum likelihood techniques yield the probit estimates for the participation decision and hence the values of X. which can be used in (A27) 1 and (A28). The full estimation model is then 2 P(I.) = 1e -t2dt i = 1...n (A30) 1 /2r X.8 Z.Q 1 - 1 (where I. = )___ 03 - 65 - * 013^ w. = X.8+ --- . + V.i=1.m (A8 1 1 0 1 12i 1.m (A8 3 1 * * and h. = -(X.8 - Z.a + a X. + V ) i = 1...m (A27) 1 a 1 1 3 1 1 in a sample of n people of whom m are in the labor force. The procedure of (A30), (A28) and (A27) describes the Heckman procedure which is essentially a two step procedure to estimate an equation in which the left hand variable, in this case hi, is censored and is only observable for positive values. An alternative procedure is to estimate a censored regression model directly, i.e., a tobit model. The Tobit Procedure The tobit procedure merely extends the likelihood function of (A29) utilising the information on observations of hi > 0. For the observations hi that are zero, P (h = 0) = 0(-I.) (as above) (A31) For the observations hi greater than zero 1 -- (I.) P (h.> 0).f(h.lh. > 0) = O(I.) . 3 r 1 3 The likelihood function is therefore m 1 n L =. -- (I. 0*(-I) (A33) 3 Maximum likelihood techniques then yield direct estimates of 8, ct and a3 from (A33). - 66 - Interpreting Tobit Coefficients 1/ To interpret the coefficients derived from the tobit technique and to see their relationship with both the OLS procedure and Heckman procedure it is useful to rewrite the mode as follows: 1 * * h. = -- (X 8 - Z.a + e.- u.) when RHS > 0 (A34) 1 a 1 1 1 1 1 h= 0 otherwise. To simplify notation rewrite (A34) * * h. = Y.6 + U. (A35) 1 1 1 where Y = (X., Z.) 1 =(1 1 1 1 * 1 u = -(e. - u.) - N(0, a2) o 1 a 1 1 3 1 We can now write * * h. = h. when h. > 0 1 1 1 h. = 0 otherwise. h. is a latent variable observable only when positiveo 1 Now, 1/ See Maddala (1983) Ch. 6 for more detailed discussion, and McDonald and Moffitt (1980) for interpretation of tobit coefficients. - 67 - E(h.) = Y.6 since E(u.) = 0. (A36) 1 1 1 E(h* h. > 0 = E(h.Ih.> 0) is the same as equation (A27) hence E(h.jh. >0) = Y.6 + a X (A37) 1 1 1 3 i (A36) gives the expected value of h. , i.e., the latent 1 unrestricted variable.' The tobit coefficients that are obtained are 6 (A37) gives the expected value of h given that hi is positive. Thus the coefficients given by Heckman's procedure are also in principle 6 if X. is a good estimator of X. in that procedure. 1 1 What we now need is the expected value of hi, that is the mean of all observed values of hi, both positive and zero. E(h.) = Pr(h.> 0).E(hl.h.>0) + Pr(h. = 0) E(h.Ih. = 0) = (I) (Yi6 + r3X.) + *(-I.).o *(I.)(Y.6 + a3x .) [= 1 (I.) . E(h.lh.> 0)] (A38) 1 1 31 The expected value of all observations is then merely the expected value conditional upon being above the limit multiplied by the probability of being above the limit. (A38) gives the expected value locus of labor supply. Corresponding to each of these three expected values are their derivatives with respect to each yj (in Yi) - 68 - BE(h.) = 6. from (A36) 1 (A39) Syj aE(h.) from (A38) yh - *(I.).6. (A40) ayj 1 j from (A37) aE(h=i6h. > 0) ? ] (A41) 1 where X. as before. T( 6. is then the slope of the fitted tobit line relating the tobit index or latent variable h to the yi. Since hours of work cannot in principle be negative, this is not the slope of interest in a study of labor supply. The slope that is sought here is the second one 3E(h.) = -*(I.) . 6. By. 1 j the slope of the expected value locus. At the mean, 0(1i) is merely the proportion of observations that are positive. (A41) gives the fraction of total effect of yj on hi due to variation above the limit. Thus (1 - I. X. -?) gives the portion of response of hi given that it is positive. If *(1) is known, 0(1i) and Ii can be calculated and hence X. 1 aE(h.) OLS coefficients are analogous to ay. and are therefore to be compared with *(i.) . 6. . This makes intuitive sense as the marginal effect of yj on hi and is therefore the estimate reported in all tobit tables. - 69 - As mentioned earlier, the Heckman coefficients estimated in (A27) correspond to 6. . If, however, the derived estimates of X. are not "good" estimates the Heckman coefficients may correspond to (A41) since the OLS regression is done on workers only and the variance in hi is only due to effects above the limit, i.e., zero hours. - 70 - BIBLIOGRAPHY Amemiya, Takeshi: "Qualitative Response Models: A Survey", Journal of Economic Literature, Vol XIX, No. 4, 1483-1536, December 1981. : "Tobit Models: A Survey", Journal of Econometrics, Vol. 24, pp. 3- 62, Annals 1984-1, 1984. Ashenfelter, Orley C. and James J. Heckman: "The Estimation of Income and Substitution Effects in a Model of Family Labor Supply", Econometrica, Vol. 42, pp. 493-517, 1965. Becker, Gary S.: "A Theory of the Allocation of Time", Economic Journal, Vol. 75, pp. 493-517, 1965. Ben-Porath Yoram: "Labor Force Participation Rates and the Supply of Labor" Journal of Political Economy, Vol. 81 No. 3. May/June 1973, pp. 697-704. Berry, R. Albert: "Work and Welfare in Colombia", (mimeo). World Bank, Washington, D.C. 1980. Bowen, William G. and T. Aldrich Finnegan: "The Economics of Labor Force Participation, Princeton: Princeton University Press, 1969. Cain, Glen: Married Women in the Labor Force, Chicago: University of Chicago Press, 1960. Cogan, John: "Married Women's Labor Supply: A Comparison of Alternative Estimation Procedures", in James P. Smith, 1980. : "Labor Supply with Costs of Labor Market Entry", in James P. Smith, 1980. Chez, Gilbert and Gary S. Becker: The Allocation of Time and Goods over the Life Cycle, New York: Columbia University Press, 1975. Greenhalgh, Christine: "Participation and Hours of Work for Married Women in Great Britain", Oxford Economic Papers, Vol. 32, No. 2, pp. 276-318, July 1980. Heckman, James: "Shadow Prices, Market Wages, and Labor Supply", Econometrica, Vol. 42, No. 4, 679-694 July 1974. : "Dummy Endogenous Variables in a Simultaneous Equation System, Econometrica Vol. 46, No. 4, pp. 931-959, July 1978. : "Sample Selection Bias as a Specification Error" in James P. Smith, 1980. Killingsworth, Mark R.: Labor Supply, New York: Cambridge University Press, 1983. - 71 - Jen, Li-Ming Caroline: Urban Female Labor Force Participation in Brazil, Unpublished Ph.D. Dissertation, State University of New York at Buffalo, 1983. Layard, R., M. Barton and A. Zabalza: "Married Women's Participation and Hours", Economica, Vol. 47, 51-72, February 1980. Lehrer, Evelyn and Marc Nerlove: "The Impact of the Female Life Cycle: Time Allocation Decisions on Income Distribution Among Families", in Human Resources Employment and Development, Volume 3 of The Problems of Developed Countries and the International Economy edited by Burton A. Weisbrod and Helen Hughes. London: MacMillan Press 1982. : "The Labor Supply and Fertility Behavior of Married Women: A Three Period Model", in Research in Population Economics Volume 4, pp. 217-235, 1982. : "Women's Life Cycle Time Allocation: An Econometric Study" in S. F. Berk (ed.) Women's Policy Studies Annual Series, Volume 5, Beverly Hills, California: Sage Publications, 1980. McDonald, John F. and Robert A. Moffit, "The Uses of Tobit Analysis", Review of Economics and Statistics, Volume 62, pp. 318-321, May 1980. Maddala, G. S.: Limited Dependent and Qualitative Variables in Econometrics, New York: Cambridge University Press, 1983. Mincer, Jacob: "Labor Force Participation of Married Women: A Study of Labour Supply", in H. G. Lewis (ed.) Aspects of Labor Economics, pp. 63- 97, Princeton, New Jersey: Princeton University Press, 1962. Mohan, Rakesh: "The People of Bogota: Who They Are, What They Earn, Where They Live", World Bank Staff Working Paper No. 390, Washington D.C. 1980. , M. Wilhelm Wagner and Jorge Garcia: "Measuring Urban Malnutrition and Poverty: A Case Study of Bogota and Cali, Colombia", World Bank Staff Working Paper No. 447, Washington D.C., 1981. : "The Determinants of Labor, Earnings in Developing Metropoli: Estimates from Bogota and Cali", World Bank Staff Working Paper No. 498, Washington D.C. 1981. , and Nancy Hartline: "The Poor of Bogota: Who They Are, What They Do, and .Where They Live", World Bank Staff Working Paper No. 498, Washington D.C., 1984. : "An Anatomy of the Distribution of Urban Income: A Tale of Two Cities in Colombia", World Bank Staff Working Paper No. 651, Washington D.C. 1984. - 72 - Nakamura, Masao, Alice Nakamura and Dallas Cullen: "Job Opportunities, the Offered Wage, and the Labor Supply of Married Women", American Economic Review, Volume 69, No. 5, 784-805, December 1979. Schultz, T. Paul: "Estimating Labor Supply Functions for Married Women" in James P. Smith, 1980. Shields, Nwanganga: "Women in the Labor Markets of Africa: The Case of Tanzania". Washington, D.C. World Bank Staff Working Paper No. 380. 1980. Smith, James P. (ed.): Female Labor Supply: Theory and Estimation, Princeton, New Jersey: Princeton University Press, 1980. Urrutia, Miguel: Winners and Losers in Colombia's Recent Growth Experience, New York and Oxford: Oxford University Press, 1985. World Bank: Colombia: Economic Development and Policy under Changing Conditions, Washington D.C., 1983. French: Croissance et emplot en Chine. Oxford University Press, 1980. 444 pages World Bank Economica, 980. ISBN 2-778-0282-7, (including Index). Publications Stock No. IB 0548, $795 LC 79-23354. ISBN 0-19-520115-9, hard- Spanish: Crecimiento economico y emnpleo cover; ISBN 0-19-520116-7, paperback. of Related en China. Editorial Tecnos, 1980. ISBN Stock Nos. OX 520215, $27.50 hardcover; Interest 84-309-0871-4, Stock No. IB 0530, 57 95. OX 520126, $12.95 paperback. TheoDesHousehold Income or Inpomen ro ten- n Household Income per Capita E sin Welfare Corarisons Joseph J. Stem and Jeffrey D. Gauta Datta LewkesndJacMohanma Staff WorkIng Paper No. 378 1980. 32 Staff Working Paper No. 419 1980. 70 pages (including 3 appendixes, references). pages (including bibliography, 2 appen- Stock No. WP 0378. b3. dixes) Stock No. WP 049 $3 How Segmented is the Bogota The Determinants of Labor Emly etPlc nLabor Market? Earnings in Developing n Policylin Metropolises: Estimates from Developing Countries: A 19ar2 S. Fields Bogota and Cali, Colombia Survey of Issues and Evidence Staffs Workluing Papeirap44.y98, 9 paesh mudm LynSqfrepenscinlsdngbibiora).) RLStock No. WP 0434 $3. Staff Working Paper No. 498 1981. 135 Low rates of growth in industrial em- pages (including 2 appendixes, bibliog- ployment, high rates of unemploy- Incidence of Poverty and the raphy). ment among new entrants to the ur- Characteristics of the Poor in Stock No. WP 0498. S5. ban labor market, and low levels ofPeislrM aya,17 Stock N. WP 048. $5.labor productivity and remuneration Peninsularialyi,97 Differences in Income, are the three issues addressed in this Pai iai Nutitin, ndPovrtywitinstudy. The author identifies the impor- Staff Working Paper No. 460 2982. 221 Nuriio, ndPoery itintant determinants of labor demand pages (including statistical appendix). Brazil and supply and the extent to which Stock No. WP 0460. $10. Vinod Thomas the growth of labor demand has been Staff Working Paper No. 505 1982. 91 constrained-and labor supply ad- Income Distribution and pages (including references, map). vanced-by inappropriate policies. On Poverty in Mexico Stok N. W 055. 3.the demand side, industrial trade poi- Joel Bergsman Nicy, agricultural growth, and the oper- The Distribution of Income ation of capital markets are discussed; Staff Wor-ing Papern. fon the supply side, attention is fo- Brzlcused on population and education Stock No. WP 0395. $3. Guy P. Pfeffermann and Richard policy. C. Webb Income Distribution Policy in Staff Working Paper No. 356. 1979. 116 Oxforudin Unirsityrs, 1981x. 22pgs the Developing Countries: A ps(includ ng 2ap pndixe)bichdn ilography, index). Lsc W p p0356.e$5 . LC 81-2844. ISBN 0-19-520266-X, hard- Case Study of Korea co.cover; ISBN 0-19-520267-8, paperback. Irma Adelman and Sherman oStock Nos OX 520266, $6.95 hardcover; Robinson EmploX mentGinwChina OX 520267, $7.95 paperback. Seeks to answer the question of how roh ran Et m n Sm. much can actually be done to improve Thomas G. Rawski rwhadEqiyi ei income distribution by means of a dy- Examines how China's agricultural Industrialized Countries namnic general equilibrium model for sector has made significant strides to- Joel Bergsman investigating the potential impact of ward full employment for a labor force Staf Working Paper No. 352. 2979. 215 standard policy instruments and pro- of nearly half a billion persons. pages (including references) grdasonten e ofprv the rel.v Oxford University Press, 1979. 208 Pages Stock No. WP 0351. $5. 1 including maps, bibliography, index). -Oxford University Press, 1978, 364 pages LC 79-19550 ISBN 0-19-520p51-5, hard- Growth with Equity: The (including 6 appendixes, index). cover. ISBN 0-19-520152-3, paperback. Taiwan Case LC 76-4269. SBN 0-19-920084-X, hard- Stock Nos. OX 520151, $18.95 hardcover; John C. H. Fei, Gustav Ranis, and cover. Stock No. OX 920084, $250. OX 520152, $7.95 paperback. Shirley W.Y. Kuo income Inequality and Poverty: introduces a method for tracing the in- Methods of Estimation and equality of family income to the ie- quality of various kinds of factor in- ptices subject to change without notice come and the shares of factor income Naniak C. Kakwani and may vary by country. in family income. Deals with income distribution meth- ods and their economic applications. Labor Productivity: Un Tour A Model for Income Oxford University Press, 1980. 432 pages d'Horizon Distribution, Employment, and (including bibliography, indexes) Susan Horton and Timothy King Growth: A Case Study of LC 80-14229. ISBN 0-29-520227-9, Stock Staff Working Paper No. 497. 1981. 72 Indonesia No. OX 520227, $12.95 paperback. pages (including bibliography). Syamaprasad Gupta Inequality and Poverty inStock No. WP 0497 $3. Presents a quantitative framework to explore the implications of alternative Malaysia: Measurement and policies for investment, employment, Decomposition NEW income distribution, and fiscal activity Sudhir Anand for medium-term and long-term An account of income inequalities and Labor and Poverty growth. poverty in Malaysia. The research is Michael Lipton The Johns Hopkins University Press, 1977. policy oriented and the fndings, to Explores factors that enable poor 138 pages (including 5 appendixes, bibliog- which the author's statistical technique households to transform their capacity raphy). is applied, are thoroughly discussed. to produce income in the labor market. LC 76-53909. ISBN 0-8018-1950-4, Stock A range of issues is covered, from data Discusses effect of dependency on No. JH 1950, $5.50 paperback. problems to conceptual questions aris- wage rates and labor duration on un- ing with respect to measurement. employment rates. Also looks at disad- Patterns in Household Demand Oxford University Press, 1982. 384 pages vantage factors such as age, sex, caste, and Saving (including bibliography, index) and health. Suggests that access to Constantin Lluch, Alan A. Powell, LC 82-14178 ISBN 0-19-520153-1, hard- physical and human assets may need and Ross A. Williams covr.StckNo OX5215, 2750to be redistributed to improve labor Studies the evolution of the structure cover. Stock No. OX 520253, $27.50 mres markets.of consumer preferences as a function International Migrant Workers' Staff Working Paper No. 616. 1983. 160 of economic development. Remittances: Issues and pages. Oxford University Press, 1977. 322 pages Prospects .ISBN 0-8213-0266-3 Stock No. WP 0616 (including bibliography, indexes). Gurushri Swamy LC 77-3442. ISBN 0-19-920097-1, hard- Staff Working Paper No 481. 1981. 64 cover; ISBN 0-19-920100-5, paperback. pages (including 2 appendixes, bibliog- Stock Nos. OX 920097, $22.50 hardcover; raphy). OX 920200, $12.95 paperback. Stock No. WP 0481. $3. Poverty and Growth in Kenya Paul Collier and Deepak Lai NEWI . - Staff Working Paper No. 389. 1980. 76 - pages (including 2 appendixes, bibliog- - -*~'~uraphy). Labor-Based Construction Stock No. WP 0389. $3. Programs: A Practical Guide for Planning and Management Poverty and Unemployment in Basil Coukis and others India: An Analysis of Recent Summarizes the basic technical and or- 6 Evidence ganizational know-how needed to de- Pravi Visaria sign and implement labor-based civil works projects. In this practical, heav- Manpower and International ily illustrated manual, decisionmakers Labor Migration in the Middle pages (including statistical annex). and project managers can examine East and North Africa Stock No. WP 0417. $3. some new ways of thinking about con- Ismail Serageldin, James A. struction and the complexity of these Socknat, Stace Birks, Bob Li, and Public Expenditure in projects. Ten years of research and Clive A. Sinclair Malaysia: Who Benefits and field work in nine countries provide an authoritative base for this guidebook. This study of labor market trends in Why Oxford University Press, New Delh.to 1985 analzes Jacob Meerman 1984. 392 pages. the demand for and supply of man- A methodologically novel study of the 2984 39 paes.power by country, sector, occupation, household distribution of public serv- ISBN 19-561511-5, hardcover; ISBN sex, educational level, and ethnic com- ices in relation to incomes. The studv 19-561512-3, paperback. Stock Nos. OX position. It indicates the sources and examines public spending for educa- 561511, $24.95 hardcover; OX 561512, destinations of workers and examines tion, medical care, agriculture, public $12.95 paperback. the social and economic effects of mi- utilities, and welfare transfers and the gration for both importers and export- use of these services by households. Labor Migration from ers, particularly the implications for Bangladesh to the Middle East government policy and planning. (incldin iiorap, in7. e Syed Ashraf Ali and others(icuigmpbblorhyine) Sye AsrafAllandothrsOxford University Press. 2983. 152 pages LC 79-1122. ISBN 0-19-520096-9, hard- Staff Working Paper No 454. 1981. 396 (including appendixes, index), cover; ISBN 0-19-520097-7, paperback. pages (including appendix). ISBN 0-19-520406-9, hardcover. Stock No. Stock Nos. OX 520096, $27.50 hardcover; Stock No. WP 0454. $15. OX 520406, $27.50. OX 520097, $12.50 paperback. Rural Enterprise and Nonfarm skilled labor by examining market con- Who Benefits from Employment ditions that cause a significant spread Government Expenditure? A Dennis Anderson and Mark between economic and market wages Case Study of Colombia Leiserson The Johns Hopkins University press, 1977 Marcelo Selowsky Examines the importance of rural non- r8pap x Clearly identifies which income groups farm activites as a source of emplov- benefit from public expenditures and ment and income for the poorest LC 76-47398 ISBN 0-8018-1938-5, hard- attempts to separate the factors behind groups of the world's rural population cover, ISBN 0-8018-1949-0, paperback the supplv of and demand for a public A World Bank Paper 1978 87 pages (in- Stock Nos JH 1938, $14 hardcover, JH service as-a means of understanding cluding 25 annexes) 1949, S5 50 paperback the present pattern of consumption. Stock Nos BK 9087 (English), BK 9057 The Urban Labor Market and Oxford University Press, 1979 202 pages (French), BK 9059 (Spanish) 55 Income Distribution: A Study (including statistical appendix, bibliog- __________of Malaysia raphy, index) NEW Dipak Mazumdar LC 79-16384 ISBN 0-19-520098-5, hard- cover, ISBN 0-19-520099-3, paperback An integrated analysis of differences in Stock Nos OX 520098, 520, OX 520099, Saudis in Transition: The income among urban households, the 9 Challenges of a Changing determinants of employment rates of Labor Market household members, and the major Women in the Urban Labor Naiem A Sherbiny, Ismail A factors affecting personal earnings Markets of Africa: The Case of Sirageldin, and M. Ismail Oxford University Press, 1981 392 pages Tanzania Serageldin (including index) Nwanganga Shields Presents the first systematic and com- LC 80-24413 ISBN 0-19-520213-9, hard- Staff Working Paper No 380 1980 136 prehensive examination of a rapidly cover, ISBN 0-19-520214-7, paperback pages (including 5 appendixes) changing labor market in a developing Stock Nos. OX 520213, 526, OX 520214. Stock No WP 0380 55 country that does not tace the typical 59 9 capital constraints Using hitherto un- available data on many labor activities and processes in Saudi Arabia, the au- thors examine the push of modernity and the pull of tradition and measure their effects on skill formation, wage Staff Working Papers are documents Some 200 Staff Working Papers not levels, and geographic and occupa- based on research conducted within available from the Bank are available tional mobility They review the basic the World Bank or by its consultants through the National Technical Infor- conceptual and methodological issues on issues affecting the economic and mation Service (NTIS) involved in the analysis of the Saudi social development of developing labor market and present a detailed ac- countres. The texts are released in an count of the dynamics of labor supply informal manner to make the results of and demand and the charactenstics of this research available quickly On av- both the workers and their employers erage, sixty such working papers are Oxford University Press 1984 About 256 pagesof a year's output of pages StafWorking ISBN 0-19-520457-3 July 1984 522 50 hardcover Size of Land Holding, Living ' , Standards and Employment in Rural Western India, 1972-73 Pravin Visana Staff Working Paper No 459 1981 117 4 pages (incluaing statistical appendix) . 0 r Stock No WP 0459 55 Some Aspects of Relative Poverty in Sri Lanka, 1969-70 A Pravin Visaria Staff Working Paper No 461 1981 254 pages (including 5 annexes) Stock No WP 0461 510 Unskilled Labor for Development: Its Economic Cost Orville John McDiarmid Estimates the economic cost of un-w  The World Bank Publications Order Form SEND TO: YOUR LOCAL DISTRIBUTOR OR TO WORLD BANK PUBLICATIONS (,See the other side o/ thiv formj P.O. BOX 37525 WASHINGTON, D.C. 20013 U S.A Date Name Ship to: (Enter if different from purchaser) Title Name Firm Title Address Firm City State - Postal Code _ Address Country 'Iblephone ( ) City State - Postal Code_ _ Purchaser Reference No. Country 'Ilephone (-) Check your method of payment. Enclosed is my Ol Check O International Money Order O Unesco Coupons O1 International Postal Coupon. Make payable to World Bank Publications for U. S. dollars unless you are ordering from your local distributor Charge my LI VISA l MasterCard l American Express l Choice. (Credit cards accepted only for orders addressed to World Bank Publications) Credit Card Account Number Expiration Date Signature 0 Invoice me and please reference my Purchase Order No. . Please ship me the items listed below. Customer internal Stock Number Author/ Title Routing Code Quantity Unit Price Total Amount S All prices subject to change. Prices may vary by country Allow 6-8 weeks for delivery Subtotal Cost $ Tbtal copies Air mail surcharge if desired ($2.00 each) $ Postage and handling for more than two complimentary items ($2 00 each) 5 Thtal $ IBRD-OU53 Thank you for your order. Distributors of World Bank Publications ARGENTINA GERMANY, FEDERAL REPUBLIC KUWAIT SRI LANKA AND THE MALDIVES Carlos Hirsch SRL OF MEMRB Lake House Bookshop Galena Guemes UNO-Verlag P0 Box 5465 P 0 Box 244 Florida 16 4th Floor-Ofc 453,465 D-5300 Bonn I Kuwait 100. Sir Chittampalam A Gardiner 1333 Buenos Aires Sunrockstrasse 23 MALAYSIA Mawatha Colomnbo 2 ALSTRALIA, PAPUA NEW GREECE Lni%ersitv ot Malava Cooperative Sn Lanka GUINEA, FIJI, SOLOMON MEMRB Bookshop, Lurited ISLANDS, AND VANUATU 24 Ippodamou Street P 0 Box 1127 Jalan Pantai Baru SWEDEN Into-Lmne Athens-11635 Kuala Lumpur For single titles Overseas Document Delivery HONG KONG, MACAU MEXICO ABCE Fritzes KungJ Hovbokhandel 80x 506, GPO Asia 2000 Ltd LNFOTEC Regeringsgatan 12 Box 16336 Sydney, SW 2001 6 Fl 146 Prince Edward Road, W. San Lorenzo 153-i11 Col del Valle 5-103 27 Stockholm Austraha Kowloon Deleg Benito Juarez For subseriion orders BAHRAIN Hong Kong 03100 Mexico City Wennergren-Wilhams AB MEMR INDIA MOROCCO Box 30004 P 0 Box 22103 LBS Publishers Distributors Ltd MEMRB 5-104 25 Stockholt Manama Town 317 Post Box 7015 2 Rue Moliere Racne SWITZERLAND BANGLADESH New Delhi 110002 Casablanca Libramie Pavot Micro Industries Development 10 First Main Road THE NETHERLANDS 6 Rue Grenus Assistance Society Gandhi \agar Medical Books Europe BV (MBE) Case postal 381 GP 0 Box 800 Bangalore 56009 \oorderwal 38, CH 1211 Geneva 11 Dhaka DEaka Apeela Chambers, P0 Box 736 41 BL Lochem TANZANIA BELGIUM 5 Wallace Street NIGERIA Oxford University Press Publcations des Nations Lmes Bombay 400001 University Press Limited P 0 Box 5299 Av dBu Rot 202 Three Crowns Building Jericho Dar es Salaam 1060 Brussels tdBChom6righee LanePvaealBg505TIl.N Calcutta 700016 Private Mail Bag 5M THAILAND BRAZIL Ibadan Central Department Store Publicacoes Tecrucas Internactonais 7188 I(CA), Swarup Nagar NORWAY 306 Sdom Road Ltda Kampu 208002 Tanum Karl Johan. AS Bangkok Rua Peixoto Gomide, 209 Sivaganga Road P 0 Box 1177 TRINIDAD & TOBAGO 01409 San Paulo, SP ungambakkam Sentrum Sstematcs Studies Lrut CANADA Madras 600034 Oslo 1 55 Eastern Main Road Li Difuseur t INDONESIA PAKISTAN Curepe C hP 85, 1501 Qmpere Street Pt Indira Luited Mirza Book Agency Trinidad West Indies Boucherville, Quebec J4B E JI Sam Ratulangi 37 65. Shahrah-e-Quaid-e-Azam TUNISIA CHILE lakarta Pusat P 0 Boxo 29 Societe Tunistenne de Diffusion Editorial Renacumento PO Box 181 Lahore 3 5 Avenue de Carthage Miraflores 3 IRELAND PANAMA Tuis Santiago TDC Publishers Ediciones Librena Cultural TURKEY COSTA RICA 12 North Frederick Street Panamena S A Haset Katapevi A 5 Librena Trelos Dublin I Av 7, Espana 16 469 Istiklai Caddest Cale 11-13 Panama Zone I Aiv Fernander Guell ITALY Bevogli Istanbul San J Gue Licosa Conmissinana Sansoi SPA PERU UGANDA Via Laiarmora 45 Editorial Desarrollo SA Lganda Bookshop CYPRUS 50121 Florence Apartado 3824 Attn Mr Martin-Luther Galiwango MEMRS Lima P0 Box 7145 P 0 Box 2098JA N Nicox Eastern Book Service (EBS) PHILIPPINES Kamoala Nicosta 37-3 Hongo 3-Chome. Bunkvo-ku \ational Book Store UNITED ARAB EMIRATES DENMARK 113 701 Rizal Avenue MEMRB SamfundsLitteratur Tokso Metro Manila P O Box 6097 Rosenoerns Alle 11 JORDAN PORTUGAL Sharjah DK-1970 Copenhagen V MEMR5 Livraina Portugal UNITED KINGDOM AND EGYPT, ARAB REPUBLIC OF P C 3, ,43 Rua Do Carmo 70-74 NORTHERN IRELAND Al Ahram labal 1200 Lisbon Microundo Ltd Galaa Street Amrran SAUDI ARABIA P 0 Box 3 Cairo KEN,A Jar Book Store Alton. Hampshire GL, 34 2PG FINLAND lmaa 3Sco Service E A Ltd P 0 Box 31% LUnited Kingdom Akateeminen Kirakauppa P C' Kx 43:45 Rivadh 11471 VEINEZUELA PS0 0x 128 \a01ri SINGAPORE, TAIWAN, BURMA Librena del Este SF0101 KOREA REPMBLIC OF Information Publications Private, Ltd Apido 00 337 Helsinki 10 Pan Korea Sook Corporation 02-06 1st Fl Pei-Fu Industrial Bldg . Caracas 1060-A FRANCE P 0 So 01 Kwang%namun 4 New Industrial Road WESTERN SAMOA World Bank Publications eoul Singapore 'esley Bookshop 66 Avenue d'iena SPAIN Apia 75116 Ps Mundi-Prensa Libros S A Castello 37 28001 Madind HD5757 .A6 R35 1985 c.2 Rakesh, Mohan, 1948- Labor force participation in a The World Bank Headquarters European Office Tokyo Office 1818 H Street, N W. 66, avenue d'16na Kokusai Building Washington, D C. 20433, U.S.A 75116 Paris, France 1-1 Marunouchi 3-chome Telephone: (202) 477-1234 Telephone: (1) 72354.21 Chiyoda-ku, Tokyo 100, Japan Telex: WUI 64145 WORLDBANK Telex 842-620628 Telephone (03) 214-5001 RCA 248423 WORLDBK Telex. 781-26838 Cable Address: INTBAFRAD WASHINGTONDC ISSN 0253-2115/ISBN 0-8213-0591-3