LSM-87 LISIIIEIhIS SEPT. 1992 Living Standards Measurement Study Working Paper No. 87 Family Productivity, Labor Supply, and Welfare in a Low-Income Country Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized
69
Embed
Family Productivity, Labor Supply, and Welfare in a Low ... · No. 19 The Conceptual Basis of Measures of Household Welfare and Their Implied Survey Data Requirements No.20 Statistical
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
LSM-87LISIIIEIhIS SEPT. 1992
Living StandardsMeasurement StudyWorking Paper No. 87
Family Productivity, Labor Supply, and Welfarein a Low-Income Country
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
LSMS Working Papers
No.18 Time Use Data and the Living Standards Measurement Study
No. 19 The Conceptual Basis of Measures of Household Welfare and Their Implied Survey Data Requirements
No.20 Statistical Experimentation for Household Surveys: Two Case Studies of Hong Kong
No.21 The Collection of Price Data for the Measurement of Living Standards
No.22 Household Expenditure Surveys: Some Methodological Issues
No.23 Collecting Panel Data in Developing Countries: Does It Make Sense?
No.24 Measuring and Analyzing Levels of Living in Developing Countries: An Annotated Questionnaire
No.25 The Demandfor Urban Housing in the Ivory Coast
No.26 The Cote d'Ivoire Living Standards Survey: Design and Implementation
No.27 The Role of Employment and Earnings in Analyzing Levels of Living: A General Methodology withApplications to Malaysia and Thailand
No.28 Analysis of Household Expenditures
No.29 The Distribution of Welfare in C6te d'Ivoire in 1985
No.30 Quality, Quantity, and Spatial Variation of Price: Estimating Price Elasticities from Cross-SectionalData
No.31 Financing the Health Sector in Peru
No.32 Informal Sector, Labor Markets, and Returns to Education in Peru
No.33 Wage Detenninants in Cote d'Ivoire
No.34 Guidelines for Adapting the LSMS Living Standards Questionnaires to Local Conditions
No.35 The Demandfor Medical Care in Developing Countries: Quantity Rationing in Rural Cote d'Ivoire
No.36 Labor Market Activity in COte d'Ivoire and Peru
No.37 Health Care Financing and the Demandfor Medical Care
No.38 Wage Detenninants and School Attainment among Men in Peru
No.39 The Allocation of Goods within the Household: Adults, Children, and Gender
No.40 The Effects of Household and Community Characteristics on the Nutrition of Preschool Children:Evidence from Rural Cdte d'Ivoire
No.41 Public-Private Sector Wage Differentials in Peru, 1985-86
No.42 The Distribution of Welfare in Peru in 1985-86
No.43 Profits from Self-Employment: A Case Study of C6te d'Ivoire
No.44 The Living Standards Survey and Price Policy Reform: A Study of Cocoa and Coffee Production inCote d'Ivoire
No.45 Measuring the Willingness to Payfor Social Services in Developing Countries
No.46 Nonagricultural Family Enterprises in C6te d'Ivoire: A Descriptive Analysis
No.47 The Poor during Adjustment: A Case Study of Cote d'Ivoire
No.48 Confronting Poverty in Developing Countries: Definitions, Information, and Policies
No.49 Sample Designs for the Living Standards Surveys in Ghana and Mauritania/Plans de sondagepour les enquetes sur le niveau de vie au Ghana et en Mauritanie
No.50 Food Subsidies: A Case Study of Price Reform in Morocco (also in French, 50F)
No.51 Child Anthropometry in C6te d'Ivoire: Estimatesfrom Two Surveys, 1985 and 1986
No.52 Public-Private Sector Wage Comparisons and Moonlighting in Developing Countries: Evidencefrom Cote d'Ivoire and Peru
No. 53 Socioeconomic Determinants of Fertility in C6te d'Ivoire
(List continues on the inside back cover)
Family Productivity, Labor Supply, and Welfarein a Low-Income Country
The Living Standards Measurement Study
The Living Standards Measurement Study (LSMS) was established by theWorld Bank in 1980 to explore ways of improving the type and quality of house-hold data coUected by statistical offices in developing countries. Its goal is to fosterincreased use of household data as a basis for policy decisionmaldng. Specificaly,the LSMS is working to develop new methods to monitor progress in raising levelsof living, to identify the consequences for households of past and proposed gov-ernment policies, and to improve communications between survey statisticians, an-alysts, and policymakers.
The LSMS Working Paper series was started to disseminate intermediate prod-ucts from the TSMs. Publications in the series include critical surveys covering dif-ferent aspects of the LSMS data collection program and reports on improvedmethodologies for using Living Standards Survey (LSs) data. More recent publica-tions recommend specific survey, questionnaire, and data processing designs, anddemonstrate the breadth of policy analysis that can be carried out using Tis data.
ISMS Worlkng PaperNumber 87
Family Productivity, Labor Supply, and Welfarein a Low-Income Country
All rights reservedManufactured in the United States of AmericaFirst printing September 1992
j To present the results of the Living Standards Measurement Study with the least possible delay, thetypescript of this paper has not been prepared in accordance with the procedures appropriate to formalprinted texts, and the World Bank accepts no responsibility for errors.
The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s)and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to membersof its Board of Executive Directors or the countries they represent. The World Bank does not guarantee theaccuracy of the data induded in this publication and accepts no responsibility whatsoever for anyconsequence of their use. Any maps that accompany the text have been prepared solely for the convenienceof readers; the designations and presentation of material in them do not imply the expression of anyopinion whatsoever on the part of the World Bank, its affiliates, or its Board or member countriesconcerning the legal status of any country, territory, city, or area or of the authorities thereof or concerningthe delimitation of its boundaries or its national affiliation.
The material in this publication is copyrighted. Requests for permission to reproduce portions of it shouldbe sent to the Office of the Publisher at the address shown in the copyright notice above. The World Bankencourages dissemination of its work and will normally give permission promptly and, when thereproduction is for noncommercial purposes, without asking a fee. Permission to copy portions forclassroom use is granted through the Copyright Clearance Center, 27 Congress Street, Salem, M\assachusetts01970, U.S.A.
The complete backlist of publications from the World Bank is shown in the annual Index of Publications,which contains an alphabetical title list (with full ordering information) and indexes of subjects, authors,and countries and regions. The latest edition is available free of charge from the Distribution Unit, Office ofthe Publisher, Department F, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or fromPublications, The World Bank, 66, avenue d'1ena, 75116 Paris, France.
ISSN: 0253-4517
John L. Newman is economist in the World Bank's Welfare and Human Resources Division of thePopulation and Human Resources Department. Paul J. Gertler is senior economnist for the RAND Corporationand associate director of the Family and Economic Development Center.
Library of Congress Cataloging-in-Publication Data
Newman, John L., 1955-Family productivity, labor supply, and welfare in a low-income
country / John L Newman and Paul J. Gertler.p. cm. - (LSMS worldng paper, ISSN 02534517 ; no. 87)
Table Al. Kolmogorov-Smirnov Statistics for Hypothesis that Residualsare Normally Distributed ............................... 51
Figure Al. Plot of Quantiles of Simulated Residuals Against Quantiles ofNormal Distribution (Males/Females Aged 18-64) ............... 53
Figure A2. Plot of Quantiles of Simulated Residuals Against Quantiles ofNormal Distribution (Males/Females Aged 12-17) ..... .......... 54
-x -
1. INTRODUCTION
In developing countries the empirical investigation of standard issues such as the
returns to human capital, labor supply, and consumption decisions is complicated by
the structure of families. Small family enterprises, both agricultural and non-
agricultural, account for much of production, and many family members work only in
the enterprise. However, it is also common for some family members to work outside
for a wage, for some to do both, and for some not to work at all. For example, in a
1985 sample of 1390 households in rural Peru, 57 percent of prime age males worked
only in the family enterprise, 18 percent worked only for a wage, 11 percent did both,
and 14 percent did not work. Families in developing countries typically have a large
number of members and the young and the elderly often work. In our rural Peru
sample, the number of workers in a family ranged from one to ten with over 60 percent
of families having three or more workers.
Accounting for this reality within an empirical family model is difficult, even
assuming a single decision maker. Three problems present themselves: the jointness of
production and consumption decisions, the interdependence of family members
activities in the utility and enterprise production functions, and the difficulty in
measuring the marginaal returns to working in the enterprise.
Family consulmption decisions are linked to their decisions on enterprise
production, except under certain restrictions on nature of markets, on the enterprise
production technology, and on the family's preference structure.' Total production
depends on family labor supply to the enterprise. Family labor supply is based, in part,
1 See Singh, Squire, and Strauss (1986) and Benjamin (1992) for models of joint productionand consumption. They show that only when families are price takers in both consumption andproduction markets (including labor), when there are no nonparticipation corners, and hired labor andfamily labor are perfect substitutes can family consumption and production decisions be treatedseparately. These requirements are suspect in many developing countries. However, Pitt andRosenzweig (1986) and Benjamin (1992) cannot reject these restrictions for Indonesia.
on the family's relative marginal utilities of consumption and of leisure - i.e. time
spent in non-income producing activities, such as education, leisure, and home
production. Family labor supply to the enterprise also depends on the comparison of
individual members' marginal returns to working in the enterprise to what can be
earned in the wage sector.
The family cannot decide the time allocation of one family member
independently of another. Family members are linked through the utility and
enterprise production functions. The marginal return to one individual working in the
family enterprise depends on the labor supply of other family members. Moreover, the
family's marginal utility of one member's leisure depends on the amount of leisure
enjoyed by other family members as well as family consumption. We define leisure as
time spent in non-income producing activities.
Unlike wage work, the marginal return to working in a family enterprise - the
derivative of the enterprise net profit function - is unobservable. Only total
production and profits net of nonfamily labor expenses are, in principal, directly
observable. Even these data are extremely costly to collect and many times suffer from
substantial measurement error (Casley and Lury, 1990; Viverberg, 1991). While a
substantial number of farm production and net profit functions have been estimated,
most treat production and consumption decisions separately. Many treat labor as an
aggregate good and therefore can not examine individual returns to work. Many treat
factor inputs such as labor as exogenous.2 Because of these data problems little work
has been completed on the determinants of the marginal return to work in family
nonagricultural enterprises.
2 A recent paper that address this issues is Jacoby (1990) who estimates shadow wages inagricultural self-employment from the agricultural production function. However, Jacoby modelsproduction independently of consumption and off-farm labor supply.
-2 -
In this paper, we develop a unified framework to address many of problems that
make modelling family decisions in developing countries difficult. In the model, the
family jointly determines consumption and the labor supply of individual members.
The model allows for an arbitrary number of family members, each of whom may or
may not engage in multiple activities. We specify a structural model that identifies the
marginal returns to work in self-employment even though they are not directly
observed. Our approach does not require estimation of an enterprise production
function.
The model consists of two types of structural equations and an identity: marginal
return functions for each activity for each family member, the family's marginal rate of
substitution of family consumption for each family member's leisure, and the family
budget constraint. A family connection is maintained through several avenues. The
marginal return to self-employed work depends on own and other family members'
hours worked in that activity. In addition, the family's marginal rate of substitution of
an individual's leisure for family consumption depends on own leisure, the leisure of
other family members, and consumption. We treat this specification as a system of
simultaneous equations and use the first order Kuhn-Tucker conditions to model
possible nonparticipation of any family member in any and all activities.
An important advantage of working with the structural equations is the
opportunity to do welfare analysis. Estimating the structural parameters of the
marginal rate of substitution allows us to convert leisure to consumption units and
calculate the compensating and equivalent variations of the changes in exogenous
variables such as factors that influence the returns to work.
We apply this model to family consumption and labor supply decisions of rural
landholding households in Peru. We estimate coefficients of marginal returns to two
-3 -
activities, wage work and self-employed agriculture, and the marginal rate of
substitution of consumption for leisure. We then use the estimates to examine the
effects of poverty alleviation programs on labor supply, consumption and welfare of
families in rural Peru. The poverty alleviation policies include programs that increase
the returns to work in agriculture such as agricultural extension offices, increasing the
returns to wage work through for example training, targeting these programs to females
versus males, and direct transfer programs.
Many of the features that we incorporate into our unified framework have been
treated separately in previous literature. While structural models of individual labor
supply decisions date from Heckman (1974), structural models of family labor supply
are relatively rare.3 A larger literature deals with reduced form family models. One
strand of the literature focuses on the jointness of production and corLsumption
decisions4. A second looks at the interdependence of labor supply decisions of different
family members. However, the typical model of family labor supply usuallr consider
the case where the husband works full-time for a wage and the wife may or may not
work5 . A few papers look at labor supply to multiple activities6 . However, these
3 Kooreman and Kapteyn (1987) take an indirect utility function approach to estimatinghusband and wive's time allocation decisions among market work and multiple home productionactivities jointly with the demand for goods. However, they only consider the nonparticipation ofwives in market work. They are not able to estimate the marginal return to home productionactivities. Lopez (1986) estimates parameters of a household's consumption demand and productionfunction. However, because he uses aggregate data, he does not look at individual labor supply or atthe possibility of nonparticipation.
4 See Singh, Squire, and Strauss (1986) for a good review of the literature on joint productionand consumption decision making.
5 For example see Ashenfelter and Heckman, 1974; Hausman and Ruud, 1984. A notableexception is Pitt and Rosenzweig (1981), who look at the effect of children's illness on adults' laborsupply.
6 A few papers have looked at multiple activities. Rosenzwieg (1978 and 1980) considers off-farm labor supply of individuals living on farms, Huffman and Lange (1989) and Ransom (1987)consider family models where husbands work on the farm and wives may work on or off the farm.Graham and Green (1984) and Gronau (1977) examine individuals' time spent in wage work andhome production.
-4 -
models do not consider consumption decisions jointly with labor supply and do not yield
estimates of the welfare effects of changes in leisure7.
The paper is organized in the usual way: theory, empirical model, application,
and conclusions.
7 Only Ransom (1987) explicitly includes consumption as a determinant of labor supply, buttreats consumption as exogenous.
- 5-
2. A FAMILY MODEL
Consider a family with N members whose utility function8 is:
u U (C, 4el12, ... ) (1)
where: C is the value of total (aggregate) household consumption, Ii is theleisure of family member i.
The family faces a budget constraint and N time constraints, one for each
member. Each family member can engage in M possible activities. The family's budget
constraint is:
MV + E Qj(Hl,,...,HNj I Fj) = P C (2)
j=l
where: V is nonlabor income, Q, is the total value to the family of engagingin the jt activity, Hij is hours of work individual i spends in the jth activity,F. is a vector of semi-fixed enterprise inputs, and P is the price of theaggregate consumption good.
If activity j is wage work, then Qj is the wage income earned by the family (i.e.
Qj= wijH,j where w,j is individual i's wage). If activity j is a family enterprise, then
Qj(*) is the net profit function. The time constraints are:
ME Hj, + gi = T , i=1,...,N (3)j=1
where T is total time available for an individual and Hij > 0 for all i and j.
The optimization problem is to choose the hours that each member supplies to
8 This specification of the family utility function implicitly assumes that families use a two-stage decision making process. They first allocate resources among total family consumption and theleisure of each member and then allocate consumption among members.
-6 -
the various activities and the total value of family consumption so as to maximize
utility subject to budget, time and non-negativity constraints. The first order
conditions consist of the budget constraint in equation (2) and:
Him 2 0, Him ( U (C .. , (I,,...,HNI )0 V im (4)
1 Ou(C, e1'e2'---"N ) =_ S49 (l 12 .. 4N P (5)
where, via the time constraint, ii = T - E Hi,,,, . The term A is the Lagrange multiplierm=1
on the budget constraint and in equilibrium is equal to the family's marginal utility of
income.
We reformulate the first order conditions into expressions that will later prove
more empirically tractable. As specified above, the first order conditions depend upon
the marginal values of leisure and the marginal returns. The marginal value of leisure
functions depend upon the arguments of the utility function and the determinants of the
marginal utility of income (A). In equilibrium, the marginal utility of income depends
upon all endogenous and exogenous variables, which would render identification of a
specification based on the parameterization of the marginal value of leisure and
marginal returns impossible.
An alternative is to reformulate the first order conditions in terms of the
marginal rate of substitution of consumption for individual members' leisure. Using
equation (5) to substitute for (1/A) in equations (4) and dividing by P, the price of
-7 -
consumption, yields:
Hi > 0, Him (MRSj - H m V i, m (6)
a9 U (C, 9l,42, ...- N)
where: MRS = a e eU (C,914 2r--gN )
0C
The reformulated first order conditions consist of equations (2) and (6).
The conditions for participation and the determination of hours worked are
analyzed by comparing the real marginal returns with the marginal rate of substitution.
An individual i will not participate in activity j if the family's marginal rate of
substitution of consumption for i's leisure is greater than i's marginal return in activity
j, evaluated at zero hours of work in activity j. If, evaluated at zero hours, the
marginal return is greater, then the individual will work up to the point where the
marginal return is equated to the family's marginal rate of substitution.
An important issue for the empirical specification is that the margin functions
cannot have arbitrary properties. Specifically, if one of the marginal returns is constant
(i.e. wage work), then the marginal rate of substitution must increase with hours
worked (decrease with leisure) to assure the possibility of an interior solution for work
in the wage sector. In addition, for an individual to possibly be able to work in more
than one activity, at least one of the marginal return functions must depend negatively
on hours worked. Otherwise, if an individual worked at all, he or she would specialize
in the activity with the highest marginal return or wage. In general, to work in M
activities, M -1 marginal return functions must depend negatively on hours worked.
-8 -
Working with the marginal rate of substitution allows us to conduct welfare
analysis. Consider the effect of an increase in an individual i's market wages induced,
for example, by an exogenous increase in the demand for wage workers. Differentiating
the utility function with respect to the wage yields,
dU(C,el,..iN) a U DC +AU a +l dU 2 . U a8N
dwi =aCw awi +a, awi + s2 aWi *aN aWi
Dividing this by the marginal utility of consumption yields,
(du(CA-.9N) 8c+
dw; = 8C + MRS, al + MRS2 a .+ + MRSNagN (7)
In equilibrium a' the marginal utility of consumption, is cqual to AP, the
marginal utility of income. Thus, the left hand side of (7) is the change in utility
valued in real monetary units. The first term on the right hand side is the change in
consumption, where consumption is valued in real monetary units. The other terms on
the RHS are the changes in leisure for each individual in the family weighted by the
MRS1 . The MRSi converts individual i's leisure into consumption units. When the
MRSi is evaluated at the equilibrium values of consumption and leisure before the
change in wages, this expression measures the compensating variation. When it is
evaluated at the equilibrium values after the change, the expression is the equivalent
variation.
-9-
S. EMPIRICAL SPECIFICATION
In our empirical application we look at two activities - farm and off-farm work.
In this case, our family model consists of marginal return functions for both activities
for each family member, the family marginal rate of substitution of family consumption
for individual leisure for each family member, the Kuhn-Tucker equilibrium conditions,
and the family budget constraint. These equations are all that are needed to examine
the labor supply and consumption decisions, and to make welfare comparisons. It is not
necessary to recover either the underlying utility functions from the marginal rate of
substitution function or the underlying agricultural net profit function from the
marginal return functions.
Our approach is to parameterize and estimate the margin functions. The main
objectives in specifying the margin functions are to obtain parsimonious yet flexible
functional forms that are consistent with theory. In this section, we discuss
specification and identification of the marginal functions. In section 4, we present
estimation and comparative statics methods.
SPECIFICATION
Wage function
We define off-farm work as working for a wage, whether in agriculture or not.
The marginal return is therefore independent of hours worked. Following standard
practice in the literature, we specify family member i's real wage as a log-linear
function,
In Q p t In = Xii. + ei X i= 1, ... ,N (8)
- 10 -
where: Xi. is a vector of individual i's characteristics, such as age andeducation, that affect i's market wage and ei. is a zero mean randomdisturbance.
Specifying a common p across all individuals means that differences in the log
wages across individuals will arise only from differences in individual characteristics and
the random error terms. This restriction can be relaxed by allowing 6 to depend on
demographic characteristics of the individual. In our empirical specification we allow /
to depend on whether the individual is a young male, young female, prime-age male, or
prime-age female. This is equivalent to specifying separate log wage functions for each
of the four demographic groups. Within any given demographic group, the log wages
will vary with the characteristics and the random error terms.
Marginal return to farm work
In estimation, we exploit the Kuhn-Tucker conditions to model nonparticipation
corners and deal with the unobservability of the marginal return to farm work and the
marginal rate of substitution. This requires comparison of the marginal real return
functions with each other and with the marginal rate of substitution. These
comparisons are simpler if the marginal return to farm work and the marginal rate of
substitution are also specified as log-linear functions. Otherwise, the empirical model is
nonlinear in parameters and errors do not enter additively. Our specification the
natural log of the real marginal return to farm work of individual i is:
en p i 9Qf =n F= Xijf3if + yiHif + E i1 H,1 + (9)
where: H is i's hours of farm work; Hjf is other family member j's hours inagricultural production; X; is a set of personal, family, and communitycharacteristics, including other fixed factors of production such as land andfarm equipment; and eif is a random productivity shocks.
- 11 -
Note that we allow the coefficient of own hours to differ from that of other
family members' hours. Since the inputs into agricultural production enter the
marginal return separately, no second order restrictions are placed on the unlderlying
enterprise net profit function.9
The error term in the marginal return represents unobserved factors affecting
individual i's productivity in farm work. These unobservable productivity
characteristics in the marginal return to farm work may be the same as those
influencing the wage. Therefore, the error terms in (8) and (9) are likely to be
correlated.
As specified, individual i's characteristics such as age and sex only affect the
intercept, but do not affect the rate at which the marginal return declines with own
hours. In our empirical implementation, we allow not only the coefficient vector #if but
also the coefficient yj to differ for individuals across the four demographic groups. We
also allow zyj to depend on the specific demographic groups of individuals i and j.
Similar to the wage function, this is equivalent to specifying separate marginal return to
farm work equations for each of the four demographic groups. Within each group, the
marginal returns will differ according to the individual and family characteristics and
the hours worked by all family members.
The added flexibility introduced by allowing every other family member's hours
to have a separate impact on the marginal return would greatly increase the parameter
space. This is especially problematic because our sample contains some families with a
9 In general, other factors of production would affect the individual's marginal return to farmwork. If these factors reflect current period input choices of the household, then, assuming one hadinformation on the input use and relevant prices, one would want to estimate simultaneously otherinput demands, labor supply, family production, and consumption. With sufficient data, this wouldbe feasible. However, in this version, we have simplified the problem by imposing zero restrictions onthe cross partial effects of labor use and variable farm inputs (e.g. hired labor and fertilizer). Thisimplies that the variable farm inputs do not enter the marginal returns. We do allow for cross effectsof labor and fixed inputs - land and farm equipment.
- 12 -
large number of pot;ential workers. We make the model parsimonious by restricting the
effects of other family members' hours to be equal within a demographic group - that is,
zsj is restricted to be equal to Yik for individuals j and k in the same demographic group.
For example, the effect of the hours of work of a young male j on the marginal return of
a prime age male is restricted to be the same as the effect of the hours of work of
another young male on the marginal return of a prime-age male. If there is no other
family member in the demographic group, then the total number of hours worked in
that group is zero. The restriction implies that, in addition to own hours, the sums of
the hours worked by all members within each of the four demographic groups appears
on the right hand side of equation (9).
The effect of another individual j's hours of work on i's marginal return (yij),
will only be equal to the effect of i's hours on j's marginal return (7yj) if the two
individuals are in the same demographic group. However, the cross effects are not
restricted to be equal when they are in different demographic groups. For example, the
effect of a young male's hours of farm work on a prime-age male's marginal return need
not equal the effect; of a prime-age male's hours on a young male's marginal return.
The marginal rate of substitution
Recall that the marginal rate of substitution (MRS) is the slope of the
indifference curve. We enter all of the arguments of the utility function separately into
the log MRS, ensuring that there are no second-order restrictions on the shape of the
indifference curve and, therefore, on the underlying utility function.
NinMRS = Zia; + orjcC + bili + E jt + Uj , i= 1,...,N (10)
where: a's and 6's are parameters, and 1u are random taste shifters 'withzero mean.
- 13 -
The MRS1 changes with respect to family consumption, own leisure, and every
other family member's leisure. For a diminishing MRS,, the sign of 6i must be negative
and the sign of ai, be positive. That is, the MRS, decreases with own leisure and
increases with family consumption. The signs of the bj indicate how the MRS1 changes
with increases in j's leisure and defines whether the family views i and j's leisure to be
substitutes or complements.
The error term in the marginal rate of substitution represents unobserved
preferences and tastes. It is assumed to be uncorrelated with the unobserved
productivity shocks in the wage and marginal return functions, equations (8) and (9).
As specified, individual i's characteristics such as age and sex only affect the
intercept of MRS,. They do not affect the rate at which the MRSi varies with the
arguments of the utility function. To achieve greater flexibility, we allow all the
parameters of equation (9) to differ according to the demographic group of individual i.
As with the wage and marginal return to farm work functions, this is equivalent to
specifying a different function for each of the demographic groups.
Again, the flexibility introduced by allowing every other family members' leisure
to have a separate impact on the MRSi would greatly increase the parameter space.
We approach the problem in exactly the same way as we did before. Therefore, in
addition to own leisure and family consumption, the sums of leisure of all members
within each of the four demographic groups appears on the right hand side of equation
(9).
The effect of an individual j's leisure on the MRS; (6ij) will only be equal to the
effect of i's leisure on the MRSj (6j,) if the two individuals are in the same demographic
group. However, the cross effects are not restricted to be equal when the two
individuals are in different demographic groups.
- 14 -
Obseruability
A major stumbling block is that the marginal rate of substitution and the
marginal return to farm work cannot be observed directly. Utility functions and their
margins are never directly observable. An individual's marginal return - the
derivative of the net profit function - is only directly observable when the enterprise
uses a single input. However, because we observe the wage, we can use the Kuhn-
Tucker conditions to infer equilibrium values of the MRS and the marginal return to
farm work depending on participation in the two activities. In some regimes the
equilibrium values of the MRS, and the marginal return are equal to the wage rate. In
the other regimes in which we cannot observe the equilibrium values, the Kuhn-Tucker
conditions provide bounds on the equilibrium values. This will be exploited later in the
specification of the likelihood function. Below we summarize the information contained
Note that the marginal returns to farm work depend on the farm hours of both
individuals. Similarly, the marginal rates of substitution depend on the leisure of both
individuals. Following our discussion of the previous section, if the two individuals are
from the same demographic group the coefficients in their margin functions will be the
same. If the two individuals are from different groups, these coefficients will be
different.
We estimate the coefficients in equations (11-16). The wage equations are
identified as there are no right hand side endogenous variables. However, identification
is an issue in the marginal return to farm work and the MRS equations. Identification
is achieved through natural exclusionary restrictions.
The marginal return to farm work depends on own and the other individual's
hours of work. Excluded from this equation are individual, family, and community
variables that affect own wage, own MRS, and the other individual's wage, MRS, and
marginal return to farm work. Candidate variables include wage experience,
community-levei demand for wage workers, taste shifters, and the other individual's
education and farm experience.
The marginal rate of substitution depends on family consumption, own total
hours of work, and other's total hours work. Excluded from this equation are unearned
income and individual, family, and community variables that affect the wage and the
marginal returns to farm work. Candidate variables include own and other's work
experience, fixed factors affecting agricultural production (such as land), producer
prices, and community-level demand for wage workers. As long as there are personal
characteristics that affect the return to one activity that do not affect the return to
another, exclusionary restrictions also identify households with more than two members.
- 18 -
4. EMPIRICAL METHODS
ESTIMATION
We assume that the error terms in the (log) wage, marginal return to farm work,
and MRS equations are from a multivariate normal distribution and estimate the model
using maximum li]kelihood10 . With no restrictions on the covariances, the likelihood
function for a two-person household involves quadruple integration. The likelihood
function for an N-person household involves 2 N integration. Because our sample
includes householdls with up to ten members, allowing a completely unrestricted
covariance structure is computationally burdensome.
We restrict error terms to be uncorrelated across family members. Given the
assumption that the error terms across individuals are uncorrelated, the multivariate
normal distribution of the errors factors into independent components involving only
errors of a single individual. The likelihood function involves no dimension higher than
a bivariate density. In addition to simplifying the computation, these covariance
restrictions across individuals provide overidentifying information.
Even though the likelihood function factors into separate terms for each
individual, full information estimation of the model must still use the family as the unit
of observation. This is because family consumption appears in the MRS of each family
member. A full information approach requires us to substitute in the budget constraint
for consumption to account for endogeneity. This introduces nonlinearities in
parameters and error terms, which increases the complexity of estimation.
We follow an alternative, limited information approach, and do not impose the
10 In our empirical application we examine the validity of the normality assumption.Appendix 1 presents 1ilots of estimated residuals and normality tests based on the estimated residuals.
- 19 -
budget constraint in the estimation. First, we predict family consumption using
nonlabor income and the presence of other relatives living away from home as
identifying variables. As family consumption is a continuous variable, this
instrumenting can be done by OLS.'1 In the second step, we replace actual family
consumption with its predicted value from the first stage.
Together with the assumption of uncorrelated errors across individuals,
conditioning on consumption results in a family likelihood function that is separable
across the individual members. This allows us to estimate the structural coefficients of
the model in the second stage using individuals as the units of observation. Because we
allow the coefficients to differ for each one of four demographic groups, we estimate the
second stage separately for individuals in each group.
While the error terms are assumed uncorrelated across individuals, we allow the
error terms in the wage and marginal return to farm work equations for an individual to
be correlated. These error terms represent unobserved productivity factors. The error
terms in the marginal rate of substitution represent taste shifters, which we assume are
uncorrelated with the unobserved productivity factors.
11 This assumes that family consumption is a continuous function of the underlying latentvariables (the wage and marginal return to farm work). See Blundell and Smith, 1990.
- 20 -
The likelihood function in the second stage for individual 1 in our two-person
p, = Pwf Uw O, - PW;WvO' - Pfuafu + OIUp ol~~~~~w u * OJf-u
Estimating the structural coefficients in this likelihood function is a
computationally tractable problem. However, it still requires a considerable amount of
time for the estimation to converge. For example, estimating the model with 39
- 21 -
parameters for a sample of 2352 individuals takes overnight to converge using GAUSS
on a 25 Mhz 80486 machine.
It is possible to relax the covariance restrictions since the model is identified
through exclusionary restrictions. However, the added computational burden makes
these extensions impractical for the moment. The simplest way of relaxing the
covariance restrictions across individuals would be to include family specific error terms
in the marginal return to farm work and MRS equations. These terms could be treated
as random effects. Conditional on the common family effects, the error terms of the
individual family members would be independent and the common factors could be
integrated out. Even this step adds considerably to the computational complexity, for
it requires using the family, rather than the individual, as the unit of observation. It
also increases the order of integration in the family likelihood function by two.
Similarly, it is possible to impose the budget constraint at the cost of additional
computational complexity.12
COMPARATIVE STATICS
The estimated structural coefficients provide information on the effects of
explanatory variables on the wage, marginal return to farm work, and the MRS. This is
of interest in and of itself. For example, we can estimate directly the rate of return to
education in self-employed farm work. We can also learn how families value leisure of
family members in different demographic groups. However, knowledge of the structural
coefficients alone does not provide information on the effects of exogenous explanatory
variables on family choices and welfare.
To obtain the family choices and welfare, we use the estimated structural
12 We are investigating the feasibility of these extensions using recently developed simulatedintegration methods.
- 22 -
coefficients and the Kuhn-Tucker conditions to solve for the utility maximizing values
of labor supply and family consumption. The structural coefficients were obtained using
individuals as the units of observation. However, due to the interdependence of the
margin functions and the binding family budget constraint, solving for the equilibrium
choice values requires using the family as the unit of analysis.
To obtain comparative statics, we calculate the equilibrium values before and
after a change in an exogenous variable. Operationally, this involves computing the
solution for each of the possible regimes and checking which one satisfies the Kuhn-
Tucker conditions. :Since the logical consistency conditions are satisfied in estimation,
we are assured that only one solution satisfies the Kuhn-Tucker conditions.
For the two-person household, we use equations (11-16) replacing the coefficients
with their estimated values. These equations include unobserved errors. We could
replace them with their mean values. However, due to the nonparticipation corners,
there are multiple regimes. Conditional on being in a particular regime, the mean of
the error terms are means of truncated trivariate normals and are nonzero. The means
need to be calculated separately for each regime. With two household members there
are 16 possible regimes. With four family members, there are 256 possible reduced
forms. The means are different for each family because they depend on the
characteristics of the family and therefore must be calculated separately for each family.
Rather than using conventional numerical integration, we simulate the error
distributions. We work with each family separately. Given the estimated variance-
covariance matrix of the error terms in the wage, marginal return to farm work, and
marginal rate of substitution equations for the particular age and sex group, take a draw
of the error terms. Given the draw, all terms on the right hand side of the margin
equations are known. We evaluate each of the possible regimes to determine in which
- 23 -
regime the Kuhn-Tucker conditions are satisfied. We then repeated the procedure for
100 draws. The average values of the solutions provides a measure of the unconditional
expected values of the endogenous variables. The percentage of solutions which fall in a
particular regime provide an estimate of the probability of being in that regime.
- 24 -
5. FAMILY DECISIONS IN RURAL PERU
DATA AND SPECIFICATION
Data used in this study are drawn from the Peruvian Living Standards Survey
(PLSS), conducted between July 1985 and July 1986 as a collaborative effort between
the Instituto Nacional de Estadistica in Peru and The World Bank. A random sample
of 5,120 households was chosen to reflect the distribution of the population in urban and
rural areas and in four natural regions. The households were chosen with an equal
probability of being selected in any given month to minimize the impact of seasonal
variation. This multipurpose survey collected information on family background and
resources available to the households. Thus, it gathered data on health, education and
training, migration, housing, fertility, income, expenditures, assets, labor force, and
faxm and business activities. In rural areas, the household level information was
complemented by a community questionnaire which gathered information on public
services, transportation, communication, and prices.13 Our sample includes all
households outside of Lima with land holdings greater than 0.01 hectares.
We estimate structural coefficients for the wage, marginal return to farm work,
and marginal rate of substitution functions for all family members aged 12 to 64.
Following the discussion in section 3, we classify all family members into four age and
sex groups - males aged 12 to 17, females aged 12 to 17, males aged 18 to 64, and
females aged 18 to 64 - and estimate separate structural coefficients for each group.
We estimate two models. The unrestricted model allows the total hours of farm
work of other family members within each age and sex group to affect one's own
13 For more information on the Peruvian survey, see The World Bank (1986). The Peruviansurvey is part of a series of Living Standards surveys conducted in an increasing number of developingcountries by The World Bank and the central statistical agencies. Surveys have been conducted or arecurrently in operation in C6te d'Ivoire, Mauritania, Ghana, Jamaica, Bolivia, Morocco, and Pakistan.
- 25 -
marginal return to farm work. It also allows the marginal rate of substitution to depend
upon the total hours of leisure of other family members within each age and sex group.
The restricted model allows only the effect of one's own hours of farm work to affect
one's own marginal return to farm work. Similarly, the marginal rate of substitution is
restricted to depend only on one's own leisure.
As discussed earlier, we follow a two-step estimation procedure. In the first step
we predict family consumption using nonlabor income and the characteristics of
relatives living outside the household as the principal instrument.14 It is measured as
the sum of retirement and pension benefits, medical or life insurance, interest on savings
accounts or other forms of savings, dividends on bonds and profit shares, rentals for
buildings, machinery, and vehicles, and inheritances. The measure of total family
consumption includes imputed returns from consumer durables and a valuation for the
auto-consumption of agricultural crops.'" It is deflated to reflect June 1985 values by a
temporal price index specific to one of thirteen regions.
The variables in the marginal rate of substitution equation are family
consumption, own leisure of the prime age male and female and a set of taste shifters.
The total hours of leisure for an individual are calculated as the total hours in the week
minus the hours spent in farm work and the hours spent in off-farm work. Leisure
includes hours spent working at home, taking care of children, and attending school. In
subsequent analysis we plan to analyze home production and hours of schooling jointly
with market work. The number of potential workers in the family and the individual's
age and age squared make up the set of taste shifters.
The marginal return to work on the farm is assumed to depend on the hours the
14 The instrumenting equation for real monthly household consumption had an adjusted R2 of0.38. The coefficient on other income was positive with a t statistic of 3.7.
5 For further details on the construction of this variable, see Glewwe (1987).- 26 -
individual works on the farm, the family's farm assets, the log of land, seasonal
dummies, agricultural producer prices, and vector of individual characteristics. The
omitted season is that of June, July, and August. The family's farm assets are
measured as the total value of equipment and the value of livestock. The producer
price index is calculated from the median price of the 16 most important agricultural
crops in Peru, which account for 80 percent of the value of all production. With fixed
weights, variation in the producer prices arises from price variation over 13 regions of
the country. Ideally, it would be desirable to calculate the index only from the price of
cash crops so as to avoid correlation between producer and consumer price indices. The
personal characteristics are age, education, and the years of experience in agricultural
work.
The wage is assumed to depend on the provincial per capita GDP, the
population density in the province, the distance from the individual's residence to a
permanent market, a dummy variable equal to one if the family's community is served
by public transportation two or more times a day, and the individual's age, age squared,
education, and the years of experience working for others.
Following the discussion in sections 3A and 3B, we allow all coefficients, sigmas,
and correlation coefficients to differ for males and females. The means of the
explanatory variables for each of the age and sex groups are presented in Table 1.
ESTIMATES OF STRUCTURAL COEFFICIENTS
Tables 2 through 5 present the estimated coefficients for females aged 18 to 64,
males aged 18-64, females aged 12-17, and males aged 12-17. The first column presents
estimates for the restricted models and the third for the unrestricted models.
Likelihood ratio tests reject the restricted in favor of the unrestricted model. In both
- 27 -
Table 1. Means of Explanatory Variables
M 12-17 F 12-17 M 18-64 F 18-64
Variables in wage function
GDP per capita/1000 0.13 0.13 0.13 0.13Population density/1000 0.03 0.03 0.03 0.03Km. to permanent market/100 0.94 0.95 0.92 0.85Availability of Public Transport 0.52 0.53 0.54 0.55Wage experience/100 0.003 0.001 0.06Age/100 0.14 0.14 0.37 0.37Years of schooling/10 0.43 0.39 0.44 0.27
Additional variables in marginal return to farm work equation
We next considered how well the model predicts family behavior. We followed
the simulation procedure described in the estimation section to solve for the equilibrium
values of family consumption and the hours of work of all family members jointly. To
reduce the computational burden, we took a 10 percent random sample of households
and limited the calculations to those with no more than four potential workers. In
addition, we simulated the model only once for each household. The value of the
random error term assigned was calculated as the mean of 100 draws from the
distribution given by the estimated variance-covariance matrix."6
The mean actual consumption of the sample of 139 households was 1316.4 Intis
per month. The predicted mean value of consumption was 1337.5 Intis per month. The
model does a remarkably good job of predicting mean consumption given our limited
information approach. The structural coefficients were estimated to fit labor supply
behavior conditional on consumption. They were not estimated to fit family
consumption well.
POLICY SIMULA TIONS
Simulating the family model allows us to address important policy questions.
For example, should governments try to raise welfare by increasing returns to self-
employed agricultural activities or by expanding wage opportunities? What is the
extent to which they should target investments? Some investments, such as those in
infrastructure, are targeted towards communities and families and will influence the
returns of all family members. Other policies, such as education, can be targeted to
individuals within a household. The government may promote female education or
16 Even with these simplifications, solving for the four person household involves calculating256 solutions of a nonlinear simultaneous system of equations. This takes 3 hours on a 25Mhz 80486PC. The number of solutions that must be obtained for each family increases by a factor of four foreach additional potential worker. The calculation time doubles with each additional draw.
- 36 -
direct agricultural extension efforts towards the type of activities typically done by
women on the farm.
In this section, we use the model to evaluate the labor supply and welfare effects
of three poverty alleviation policies - increasing returns to wage work, increasing returns
to farm work, and direct monetary transfers. We calculate effects due to increases in
returns separately for males and females to investigate the merits of targeting. In this
exercise, we do not analyze the mechanisms and the costs of implementing these
policies, but focus on the benefits. However, the changes in returns that are feasible to
achieve. For example, we consider the effects of a 20 percent increase in female wages.
Our structural estimates suggest that such an increase could be achieved by raising
mean education levels by one year. Based on our structural coefficients, a 20 percent
increase in marginal return to farm work for prime age males could be achieved with
approximately a one standard deviation increase in the log of land. A 20 percent
increase in the- marginal return to farm work for prime age females and children could
be achieved by a one standard deviation increase in farm assets.
The base case is presented in Table 7. Table 8.a presents the results of the first
simulation - increasing the wage offers to prime age males (18-64) by 20 percent. The
top row reports the weekly unconditional changes to own hours worked in the wage and
farm sectors for prime age males 18-64 and the changes in leisure for young males 12-17,
young females 12-17, prime age males, and prime age females 18-64. On average over
the whole sample, prime age males work 3.17 hours more in the wage sector and 2.72
hours less in the farm sector in response to the 20 percent increase in wage offers. Ona
net, prime age males work 0.46 more hours. Young males receive 0.19 hours more of
- 37 -
Table 7. Base Case - 10 Percent Random Sample of HouseholdsNumber of potential workers < 4
Individual Data
Pc-t Both Pct E Pct. Farm Pct. None Ha Ifa His axm
Males 12-17
Base caseSimulation 0 3.0 81.8 3.0 0.2 10.5
Number in sample: 33
Females 12-17
Base CaseSimulation 0 0 62.1 37.9 0 6.8
Number in sample: 29
Males 18-64
Base CaseSimulation 15.9 18.6 65.5 0 10.9 27.5
Number in sample: 145
Females 18-64
Base CaseSimulation 4.5 6.2 83.9 5.8 2.0 21.6
Number in sample: 155
Family data
Distribution of family size
One person households 10.8Two person households 39.6Three person households 28.1Four person households 21.5Total number of households 139
Mean actual consumption 1316.4Std. dev. actual consumption (1214.9)Mean base case consumption 1337.5Std. dev. base case consumption (1031.8)
- 38 -
Ta-bl SPOUCY SIMULATIONS
.a 20S INCREASE IN MALE WAGE
Changin Cho a iWeek Cage in inahWeek HowN Hew of Leiswoe C _em,pei Weim
Wae Farm Mal" P w I M F_l Jwork Work 12-18 12-18 19.65 19-65
1d.dev. ( I ( (0.87) (0.10) (0.28) j (27.6) (11.0)
- 39 -
leisure and prime age females 0.10 additional hours of leisure. There is no effect on
young females.
In order for the increase in wage offers to affect the family at all, a prime age
male must either already be working in the wage sector or be induced to work in the
sector as a result of the change. Otherwise, the increase in wage offers is irrelevant.
The second row presents estimates of the changes in hours worked and leisure
conditional on a male already working in the wage sector. As expected, the conditional
effects are substantially larger.
The last two columns of Table 8.a report the effects of the increase in prime age
male wages on family consumption and welfare. The unconditional mean effect is an
increase of 44.9, approximately 3 percent of mean family consumption. Again, this
unconditional effect is averaged over some households for which the change is irrelevant.
If one conditions on there being an a male already employed in the wage sector, the
mean change in consumption is 3 times large, or 9.6 percent of meau family
consumption. The 20 percent increase in male wages does not lead to a 20 percent
increase in family consumption because other family members are contributing to
consumption.
The welfare effects takes into account the family's change in leisure as well as
the change in consumption. They were calculated using predicted marginal] rates of
substitution at base case values of consumption and leisure. While prime age males
work more in response to the wage increase, other family members enjoy more leisure.
However, the net effect of the changes in leisure on family welfare is negative. This is
evident from the fact that the change in welfare is less than the change in consumption.
In this case, the difference between the consumption and welfare effect is roughly 4
percent.
- 40 -
Table 8.b presents results for a 20 percent increase in female wage rates. Very
few females work in the wage sector in the base case or are induced to work by the
increase in wages. Thus, unconditional effects of the wage increase are small. However,
for those families where females already participate in the wage sector, prime age
females experience a larger reduction in leisure than was the case with the 20 percent
increase in male wages. Despite this, the conditional effect on family consumption was
smaller than that of the wage increase of males, reflecting, in part, lower wages earned
by females. There were essentially no effects on the leisure of other family members in
different age and sex groups.
Tables 8.c and 8.d present results for 20 percent increases in male and female
marginal returns to farm work. The same general pattern with respect to weekly
changes in hours of work and leisure is revealed as was the case for the change in wage
offers. However, the effect is larger for the change in female marginal farm returns.
The picture with respect to monthly changes in consumption and welfare is considerably
different than was the case for the change in wage offers. In the first place, more males
and females work in the farm sector and, thus, are more affected by changes in the
marginal returns to farm work than they were by the changes in wage offers. This is
revealed by the smaller difference in the unconditional and conditional consumption and
welfare effects than for the wage changes. The contrast with the results from changes in
wage offers is especially pronounced for females.
The most striking result from Tables 8.c and 8.d is the similarity in the
conditional consumption effects of increases to male and female returns. This is in
marked contrast to the relative conditional effects of changes in wages offers.
Conditional on having an effect, the 20 percent increase in male marginal returns to
farm work led to an increase of family consumption of 142.8 (roughly an 11 percent
- 41 -
increase), while that due to the same percentage increase in the marginal returns of
females led to an increase of 139.0.
The ranking of the relative effects on the family of increases in male versus
female marginal returns is reversed if changes in welfare rather than consumption are
considered. This reflects the family's differential valuation of leisure of the different
groups that reallocate labor in response to the two changes. With the increase in male
returns the increase in consumption is obtained at the expense of male leisure. In the
base case, prime-age males have the least amount of leisure of any group. With the
increase in female returns, prime age males and young males and females receive more
leisure, while prime-age females receive less. The end result is that the increase in male
returns leads to a smaller change in welfare than in consumption, while the increase in
female returns leads to a larger change in welfare than in consumption.
Although the relative benefits to family welfare are higher for increases in female
returns to farming, it is important to consider also the costs of achieving the
productivity gains. If the costs are not higher for increasing female returns, our results
imply that such policies would be a cost-effective way of raising rural family welfare.
Table 8.e presents the effect of an increase in nonlabor income of 280 intis, a
little over 20 percent of mean family consumption. Simulating the effects of this change
allows us to investigate how families value the leisure of different age and sex groups.
The first row indicates that prime age females receive the smallest increase in leisure.
Of course, those who are already not working cannot receive more leisure. With the
increase in nonlabor income, all families have a positive consumption effect. The mean
effect is lower than the 280 Intis that was given to each household because families have
chosen to enjoy more leisure. The change in welfare is 10 percent higher than the
change in consumption.
- 42 -
6. CONCLUSIONS
This paper has developed an analytical approach to estimating family labor
supply and consumption decisions appropriate for developing countries. The key feature
of the approach is to work with underlying structural marginal return and marginal rate
of substitution functions. This allows us to handle .the complexity of having multiple
corner solutions for families with a large and variable number of potential workers. To
reduce the computational burden, we estimated the model using a limited information
approach and restricted the error terms in the marginal functions to be uncorrelated
across family members. It is possible to relax the covariance restrictions since the
model is identified through exclusionary restrictions. It is also possible to impose the
budget constraint in estimation. Making these extensions greatly increases the
computational complexity. For this reason, we believe that the next step would be to
investigate the feasibility of these extensions using simulated integration methods.
The general model developed for farm and off-farm work in Peru is applicable to
other analyses of multiple activities and/or interdependent family time allocations.
Examples include analyses of the effect of the availability of child care on the mother's
work at home and in the market, of the effect of illness of one family member on
activities of another, and the effect of infrastructure investment of the family's decision
of allocating their children's time among market work, home work, and school
attendance.
- 43 -
REFERENCES
Amemiya,T. (1974) "Multivariate Regression and Simultaneous Equation Models whenthe Dependent Variables are Truncated Normal," Econometrica, Vol. 42, pp.999-1012.
Ashenfelter, 0. and J. Heckman (1974), "The Estimation of Income and SubstitutionEffects in a Model of Family Labor Supply", Econometrica, Vol. 42, pp. 73-85.
Barnum, H. and L. Squire (1978), "An Econometric Application of the Theory of theFirm Household", Journal of Development Economics, Vol. 6, pp. 79-102.
Benjamin, D. (1988), "Household Composition and Labor Demand: Testing for RuralLabor Market Efficiency", Industrial Relations Section Working Paper No. 244,Princeton University, Princeton, N. J.
Blundell, R. and R. Smith (1990), "Estimation in Simultaneous MicroeconometricModels with Censored or Qualitative Dependent Variables", mimeo, UniversityCollege London, January.
Casley, D. J. and D. A. Lury (1987) Data Collection in Developing Countries, 2nd. ed.,London: Oxford University Press.
Duncan, G. (1987), "A Simplified Approach to M-Estimation with Applications to Two-Stage Estimators", Journal of Econometrics, Vol. 34, pp. 373-389.
Glewwe, P. (1987), "The Distribution of Welfare in Peru 1985-86", LSMS WorkingPaper No. 42, The World Bank, Washington, D. C.
Gourieroux, C., A. Monfort, E. Renault, and A. Trognon (1987a), "GeneralizedResiduals", Journal of Econometrics, Vol. 34, pp. 5-32.
Gourieroux, C., A. Monfort, E. Renault, and A. Trognon (1987b), 'SimulatedResiduals", Journal of Econometrics, Vol. 34, pp. 201-252.
Graham, J. and C. Green (1984), "Estimating the Parameters of a HouseholdProduction Function with Joint Products", Review of Economics and Statistics,Vol. 66, pp. 277-83.
Gronau, R. (1977), "Leisure, Home Production and Work: The Theory of theAllocation of Time", Journal of Political Economy, Vol. 85, pp. 1099-1123.
Hausman, J. and P. Ruud (1984), "Family Labor Supply and Taxes", AmericanEconomic Review, Vol. 74, pp. 242-248.
Heckman, J. (1974), "Shadow Prices, Market Wages, and Labor Supply, Econometrica,Vol. 42, No. 4, pp. 679-694.
Huffman, W.E. and M. Lange (1989), "Off-farm Work Decisions of Husbands and- 45 -
Jacoby, H. G. (1990), "Shadow Wages and Peasant Family Labor Supply: AnEconometric Application to the Peruvian Sierra", LSMS Working Paper No. 73,The World Bank, Washington, D.C.
Kooreman,P., and A. Kapteyn (1987), "A Disaggregated Analysis of the Allocation ofTime within a Household," Journal of Political Economy, Vol. 95, pp. 223-249.
Maddala,G.S. (1983), Limited Dependent and Qualitative Variables in Econometrics.Cambridge University Press, Cambridge, England.
Newman, J.L. (1991), "A Test for Normality of Residuals in the Tobit Model Based onSimulated Residuals", mimeo, The World Bank.
Pagan, A. and F. Vella (1989), "Diagnostic Tests for Models Based on Individual Data:A Survey", Journal f Applied Econometrics, Vol. 4, S29-S59.
Pollak,R.A. and M. Wachter (1975), "The Relevance of the Household ProductionFunction and Its Implications for the Allocation of Time," Journal. j PolijicalEcon , Vol. 83, pp. 255-277.
Ransom, M. (1987), "An Empirical Model of Discrete and Continuous Choice in FamilyLabor Supply", Review of Economics and Statistics, Vol. 69, pp. 463-472.
Rosenzweig, M. (1978), "Rural Wages, Labor Supply and Land Reform: A Theoretical.a Empirical Analysis", American Economic Review, Vol. 68, pp. 847-861.
Singh, I., L. Squire, and J. Strauss (eds.) (1980), Agricultural Household Models:Extensions, Applications and Policy, The World Bank, Washington, D.C.
Waldman, D. (1981), "An Economic Interpretation of Parameter Constraints in aSimultaneous Equations Model with Limited Dependent Variables",Intemational Economic Review, Vol. 22, pp. 731-739.
- 46 -
Appendix L Residual Analysis
The importance of analyzing residuals in limited dependent variables models is
now well established (see Pagan and Vella, 1989; Blundell, 1987; Gourieroux et al 1987a,
1987b). Because the dependent variable is not continuously observed, it is not possible
to construct estimates of the residuals simply by subtracting the predicted value from
the actual value. Our approach to this problem constructs estimated residuals under
the maintained hypothesis that the error terms in the marginal return equations follow
a bivariate normal distribution and are independent of the error term in the marginal
rate of substitution equation. The latter error term is assumed to follow a univariate
normal distribution. The construction of the estimated residuals must be done
separately for each of the four possible outcomes in the estimation procedure. We
illustrate the procedure with the model with no family interactions. The procedure is
exactly the same in the model with family interactions.
Case I - Individual Works Both On and Off the Farm
In this case the marginal returns are equal to each other and to the marginal rate
of substitution. Because the real wage offer is observable, it is possible to construct
estimates of the three error terms. Thus,
u = In W- Z& + &aO + 6 (T- H.- Hf)
fw= In -F ~w#I
- ff = en - Xf,Sf - 4-Hf
-47 -
Case II - Individual Works Only in the Wage Sector
The equations relevant for this case are given by eqs. (19-21) on page 16 in the
text and are reproduced below.
e W =Z +: c (T- H) + uP
In p = X.P. +e.
In W > Xpf + ef
When the individual works only in the wage sector, the real wage is observable
making it possible to obtain estimates u1 and u, exactly as before. However, all that is
known about ef is that it is less than en W XfAf . One alternative as an estimate
of ef is E( ef I ef < gn W Xf1 f). This is the idea of generalized residuals that
form the basis for diagnostic tests proposed by Chesher and Irish (1987) and Gourieroux
et al (1987a). However, replacing ef by its conditional mean over the sample places too
much mass of the density away from the low end of the distribution. The number of
observations in the tail of the distribution will be understated. Our approach is as
follows.
* Take 500 draws from the distribution of ef given C' and the estimated variances.-
. Identify all draws e2 such that e2 < Qn -
. Randomly select one e2 satisfying the above condition as an estimate of 2f.
- 48 -
Because there is more mass of the conditional density function close to the
conditional mean, it is more likely that the single draw that is randomly selected will be
close to the conditional mean. However, this procedure does allow for the possibility of
selecting a draw from the far tail of the distribution. This approach is similar to that of
simulated residuals developed by Gourieroux et al (1987b), without their second-stage
estimation. A monte-carlo simulation comparing this approach to the results from
generalized residuals is given in Newman (1991).
Case III - Individual Works Only on the Farm
The equations relevant for this case are given by eqs. (22-23) on page 16 in the
text and are reproduced below.
Za + &cC + 6 (T - Hf) + u > XP,ffl, + c.
Xf + Hf + ef = Za + aCC + 6 (T - Hf) + U
The procedure for estimating residuals in this case is as follows:
* Take 500 draws of u, given its estimated variance.
* Solve for ef using the second equation above. This yields 500 values of cf.
* Given the conditional density of C. given ef and estimated variances, take one
random draw of ew for each of the 500 values of cf. This yields 500 values of
the triplet (c,,,, eU u).
* Select all triplets that satisfy the condition given in the first equation above.
* Randomly select one triplet from the set satisfying the condition.
- 49 -
Case IV - Individual Does Not Work
The relevant equations are given by eqs. (24-25) on page 16 in the text and are
reproduced below.
Za + a'C + 6 T + u > X.#. + ew
Za + a6 + a T + u > Xpf + cf
The procedure for estimating residuals in this case is as follows:
- Take 500 draws of u, given its estimated variance.
* Randomly draw 500 values of (c., ef) given its joint distribution and estimated
variance-covariance matrix. This yields 500 values of the triplet ('w, c1, u).
e Select all triplets that satisfy the conditions given in above two equations.
e Randomly select one triplet from the set satisfying the conditions.
Plots of Estimated Residuals and Normality Tests
After estimating the residuals separately for each case, they are combined to
yield estimated residuals for the entire sample. Figure Al plots the quantiles of the
estimated residuals against quantiles of the normal distribution for prime age males and
females. Figure A2 plots the same relationships for the younger males and females.
These figures indicate a divergence from normality in the tails of the distribution,
principally for the prime age male. This is confirmed by the Kolmogorov-Smirnov test
statistics presented in Table Al.
- 50 -
Table Al. Kolmogorov-Smirnov Statistics for Hypothesis thatResiduals are Normally Distributed
Fig. A2. Plot of Quantiles of Simulated Residuals AgainsLQuantiles of Normal Distribution
,0 /~~~~~~~~~~
-4.78 5.02 -8.82 9.32 -5.60 6.19W'Aage MRF MRS
Males 12-17
0 i
-9.45 9.46 -10.08 10.73 -5.32 5.97Wage MRF MRS
Females 12-17
- 54 -
Distributors of World Bank PublicationsARGENTINA EL SALVADOR JAPAN SOUTH AFRICA, BOTSWANACarlos Hirsch, SRL Fusades Eastern BookService ForsingLetitlat-GaJera Guemes ALam Dr. Maiwel Enrique Araujo #3530 Hongo 3-Chorme, Bunkyo-ku 113 Oxford University PrmFlorida 165,4th Floo.Ofc 453/465 Edificio SISA, ler. Pbo Tokyo Southern Africt1333 Bueno Air San Salvador O11 P.O. Box 1141
KENYA Cape Town s8O0AUSTRALIA, PAPUA NEW GUNEA, FINLANi Africa Book Sevice OA Ltd.FIJI, SOLOMON ISLANDS, Akatemtinen lariakauppa Quaran Housem Mfanpno Stst For s vo ie, onrdaVANUATU, AND WESTERN SAMOA P.O. Box 128 P.O. Box 4!245 Intemaabonal Subscipdon SevieD.A. Books & Jaumals SF-0101 Heisinid 10 Nairobi P.O. Box 41096648 Whitehorse Road CraiglallMitcham 3132 FRANCE KOREA,REPUBICOF Johannesburg2024Victoria World Bank Publicatons Pan Kom Book Corpodton
66, avenue d'I6na P.O. Box 101, Kwangwhauwn SPAINAUSTRIA 75116 Pa:s Seoul Mundi-Preu UlIv, S.A.GeroId and Co. Castello 37Graben31 GERMANY MALAYSIA 28001 MadridA-iOllWien UNO-Verlag Univeaty ofMalay Coopeative
Poppedorfer Aee 55 Bookshop, Limited Librerfa IntmacdonalAEDOSBANGLADESH -S-300 Bonn 1 P.O. Box 1127, Jalan Pantai Beru Consel de Cent 391Micro industries Development 59700 Kuala Lumpur 08009 Bansdona
AssistanceSociety(MIDAS) GUATEMALAHouse S, Road 16 Librrias Pledra Santa MEiDCO SRI LANKA AND THE MALDIVESDhanmondi R/Area 5a. Caile 7-55 INFOTEC Lake House BookshopDhaka 209 Zona I Apartado Postal 224-60 P.O. Box 24
Guatemala City 14060 Thpan, Mexco DF. 100,SitChitampalamA.Brech : Gairdiner Mawatia156,NurAhmedSarak HONG KONG, MACAO NETHERLANDS Colowabo2Chittaong;400W Asia 2000 Ltd. De Lindeboom/InOr-Publikaties
Centii Hong Kong NEW ZEALAND Regerinptan 12, Box 16356BELGIUM EBSCO NZ Ltd. S-103 27 StockholmJean De Lannoy INDIA PrivateMail Bag 99914Av. du Roi 202 Allied Publishers Prtivate Ltd. New Market Forsubwr btiO ordsa1060 Brubs 751 Mount Road Auckland Wennergren-WiiamsAB
Madras- 600 002 Box 30004CANADA NIGERIA S-104 25 StockhoimLa Diffuseut Brecht eJ0cw University Press LimitedC.P. 85, ISOIB rue Amply. 15 J.N. Heredia Maig Three Crowns BuiddingJericho SWITZERLANDBoucheriLe Qu4bec Ballard Estate PrivateMail Bag 595 For singte tbi:J4B Si6 Bomtay -400 038 ibadan Librairie Payot
1, rue de BourgCHiNA 13/14AsafAi Road NORWAY CH 1002 LausaneChina Financial & Economnk New Delhi -110 002 Narven information Center
Publishing Hous Book Department For suherfp orders-8. Da Fo Si Dong Jl 17 Chittaranjn Avenue P.O. Box 6125i Attsted Librairie PayotBeiing CAicutta -700 072 N-0602 Oslo 6 Service des Abonnehme
Case postale 3312COLOMBIA Jayedeva Hostel Building PAKISTAN CH 1002 Lausanneinfoenlace Lda. 5th Main Road Gandhinagar Mirza Book AgencyApartado Aereo 34270 Bangalore -560009 65, Shazraha-eQuaid-rAtam TANZANIABogota D.E. P.O. Box No. 729 Oxford Univerity Pres
3-5-1129 Kachiguda Cros Road Lahore 54000 P.O. Box 5299COTE D'iVOiRE Hyderabad -SD 007 Maktaba RoadCentre d'dition et de Diffusion PERU Dar esSalaam
Africaine (CEDA) Prarthana Pla, 2nd Floor Editorial Desarrodlo SA04 B.P. 541 Near Thakore Baug, Navrangpura Apartado3824 THAILANDAbidjan 04 Plateau Aitmedabad -380009 Lima 1 Central Department Store
306 Silomn RoadCYPRUS Patila House PHiLIPPINES BangkokCyprus Coleage Bookstor 16-A AshokMarg Intenational Book Center6. DiogeStreet, Engomi Lucknow-226001 FifthP oor, Filpin Life Building TRINIDAD & TOBAGO, ANTIGUAP.O. Box 2006 Ayala Avenue, Makati BARBUDA, BARBADOS,Nicoata Central BazaarRoad Metro Manila DOMINICA, GRENADA, CUYANA,
60 Bajaj Nagar JAMAICA, MONTSERRAT, ST.DENMARK Nagpur440010 POLAND KITTS & NEVIS, ST. LUCIA,SamfundsLitteratu ORPAN ST. VINCENT & GRENADINESRosenoens A116 11 INDONESIA Palac Kult"ry I Naukl SyataaticsStudis UnitDK-1970 FredediibSrgC Pt. Indira Limited 00-901 Warawa #9 Watt Street
JL Sam Ratulangi 37 CurepeDOMINICAN REPUBUC P.O. Box l18 PORTUGAL Trinidad, West IndiesEditors Taller C. por A. Jakarta Pusat Livraria PortugalRmtauracifn elIsb la Catldca 309 Rua Do Camo 70.74 UNIDiCNGDOMApartadode Coffeos 2190 Z1 ISRAEL 120D Lsbon Microinfo Ltd.Santo Domingo Yozovot Literatur Ltd. P.O. Box 3
P.O. Box 56055 SAUDI ARABIA, QATAR Alton. HampulIeGU342PGEGYPT, ARAB REPUBUC OF Tel Aviv 61360 Jarir Book Store EnglandAl Ahmram iael P.O. Box 3196Al Galaa Stret Riyadh 11471 VENEZUELACairo ITALY Librerba del Este
Lkosa Commionaia Swtsoni SPA SINGAPORE, TAIWAN, Aptdo. 60337The MiddlerEastObserver Via Duca Di Calabria, III MYANMARBRUNEI Caracas 1060-A41, SherifStreet Casela Postale 552 Inforanalion PubliationsCairo 50125 Fwenze Private, Ltd.
02-06 Ist Fl. Pae-Pu InduttiaBldg.
24 New Industrial RoadSingapore 1953
LSMS Working Papers (continued)
No. 54 The Willingness to Payfor Education in Developing Countries: Evidence from Rural Peru
No. 55 Rigidite des salaires: Donnees microeconomiques et macroeconomiques sur l'ajustement du marchedu travail dans le secteur moderne (in French only)
No. 56 The Poor in Latin America during Adjustment: A Case Study of Peru
No. 57 The Substitutability of Public and Private Health Care for the Treatment of Children in Pakistan
No. 58 Identifying the Poor: Is "Headship" a Useful Concept?
No. 59 Labor Market Performance as a Detenninant of Migration
No. 60 The Relative Effectiveness of Private and Public Schools: Evidence from Two Developing Countries
No. 61 Large Sample Distribution of Several Inequality Measures: With Application to Cote d'lvoire
No. 62 Testing for Significance of Poverty Differences: With Application to C8te d'Ivoire
No. 63 Poverty and Economic Growth: With Application to COte d'Ivoire
No. 64 Education and Earnings in Peru's Infonnal Nonfarm Family Enterprises
No. 65 Formal and Infornal Sector Wage Determination in Urban Low-Income Neighborhoods in Pakistan
No. 66 Testing for Labor Market Duality: The Private Wage Sector in Cote d'Ivoire
No. 67 Does Education Pay in the Labor Market? The Labor Force Participation, Occupation, and Earningsof Peruvian Women
No. 68 The Composition and Distribution of Income in C6te d'Ivoire
No. 69 Price Elasticities from Survey Data: Extensions and Indonesian Results
No. 70 Efficient Allocation of Transfers to the Poor: The Problem of Unobserved Household Income
No. 71 Investigating the Determinants of Household Welfare in Cote d'Ivoire
No. 72 The Selectivity of Fertility and the Determinants of Human Capital Investments: Parametricand Semiparametric Estimates
No. 73 Shadow Wages and Peasant Family Labor Supply: An Econometric Application to the Peruvian Sierra
No. 74 The Action of Human Resources and Poverty on One Another: What We Have Yet to Learn
No. 75 The Distribution of Welfare in Ghana, 1987-88
No. 76 Schooling, Skills, and the Returns to Government Investment in Education: An Exploration UsingData from Ghana
No. 77 Workers' Benefits from Bolivia's Emergency Social Fund
No. 78 Dual Selection Criteria with Multiple Alternatives: Migration, Work Status, and Wage;
No. 79 Gender Differences in Household Resource Allocations
No. 80 The Household Survey as a Tool for Policy Change: Less~ons from the Jamaican Survey of LivingConditions
No. 81 Patterns of Aging in Thailand and Cote d'Ivoire
No. 82 Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia
No. 83 Growth and Redistribution Components of Changes in Poverty Measure: A Decomposition withApplications to Brazil and India in the 1980s
No. 84 Measuring Income from Family Enterprises with Household Surveys
No. 85 Demand Analysis and Tax Refonn in Pakistan
No. 86 Poverty and Inequality during Unorthodox Adjustment: The Case of Peru, 1985-90
The World Bank
Headquarters European Office Tokyo Office1818 H Street, N.W. 66, avenue d'Iena Kokusai BuildingWashington, D.C. 20433, U.S.A. 75116 Paris, France 1-1, Mamnouchi 3-chome