View
3
Download
0
Category
Preview:
Citation preview
Intrahousehold Time Allocation in Rural Mexico: Evidence
from a Randomized Experiment
Marta Rubio-Codina�
June 1, 2009y
Abstract
This paper examines the e¤ects of Oportunidades, a conditional cash transfer program, on
the allocation of time of household members in rural Mexico. I exploit the random placement
of bene�ts across communities in the evaluation sample and the program�s eligibility crite-
ria and scheme of incentives to identify e¤ects. The majority of Oportunidades bene�ts are
linked to children�s school attendance, implying a reduction in the price of schooling. I argue
that changes in relative prices lead to substitution e¤ects whereas the (almost) unconditional
nutritional transfer translates into an income e¤ect. Findings show increases in schooling
and reductions in children�s participation in market and nonmarket work. Although the pro-
gram does not seem to substantially alter adults�time allocation, evidence suggests that adult
women substitute for children�s time in nonremunerated activities.
Keywords: Intrahousehold Time Allocation, Income and Substitution E¤ects, Condi-
tional Cash Transfer Programs, Mexico.
JEL Codes: D10, J22.
�The Institute for Fiscal Studies: marta_r@ifs.org.uk.yI am grateful to two anonymous referees, Orazio Attanasio, Tania Barham, Pierre Dubois, Paul Gertler and
Sylvie Lambert for useful comments and suggestions on earlier versions. The usual disclaimer applies.
1
1 Introduction
Time is both a fundamental and a scarce resource in the economy. The study of the role of
time in labor economics and the use of time budget data to model microeconomic behavior
have developed since Becker (1965) seminal piece. It is nowadays widely acknowledged that
time availability � or scarcity � and its allocation to various activities can ultimately deter-
mine the relative prices of goods and services, total production levels, and the distribution of
income (Juster and Sta¤ord 1991). However, the many methodological issues involved in the
measurement and treatment of time data have somehow limited the amount of applied research
in the area. Notable exceptions are the works by Gronau (1977), Evenson (1978), Mueller
(1984), Kooreman and Kapteyn (1987), Skou�as (1993) and Parker and Skou�as (2000). Ellis
(1994) provides a review of descriptive studies on time allocation in agricultural households in
developing countries.
Within a household, family members devote time to many di¤erent activities: participation in
the labor force, household chores, studying, or "pure" leisure activities, like hobbies or personal
needs and care (sleeping, eating, etc.).1 Moreover, the marginal rate of substitution of time
allocated to each of these activities between di¤erent household members is likely to be a¤ected
by certain government policies, such as work taxation, education stipends or child care subsidies.2
If available, detailed individual data on time use provides a unique research resource to examine
the impacts of public policies on individual behavior, both inside and outside the household.
This paper uses cross-sectional data on time allocation for all household members (older than
8 years old) to examine intrahousehold time responses to one such policy in rural Mexico. In
1997, the Mexican Federal Government launched the Education, Health and Nutrition Program,
Progresa, renamed Oportunidades under the Fox Administration. This antipoverty strategy pro-
vides monetary and in-kind bene�ts to targeted poor households conditional upon preventive
care and school attendance. The poor rural communities where the program operates are tra-
1Gronau (1977) �rst distinguished between leisure and work at home in theoretical models, while acknowledging
that the distinction is not straightforward. For example, eating and sleeping could be alternatively considered an
input in the household production function because they increase productivity.2The impact of a particular policy on intrahousehold time allocation depends on the household decision-making
model assumed. Behrman (1997) and Bergstrom (1997) provide excellent reviews of the existing theories.
2
ditionally characterized by very high labor force participation rates of men compared to those
of women. In addition, children �typically boys �tend to begin their labor force participation
at early ages in order to contribute to family income.
Oportunidades aims to increase basic education school enrollment and reduce child labor by
linking part of the bene�t to school attendance. This implies a reduction in the shadow wage
of children�s time allocated to activities other than schooling, and amongst other household
members �i.e. a change in relative prices. As a result, own-substitution and cross-substitution
e¤ects are likely to arise. For example, and assuming that school and work are substitutes,
one would expect increases in school participation and reductions in children�s time allocated
to market and nonmarket work. Moreover, a reallocation of adult household members�time to
market and nonmarket activities previously done by eligible children in the household are also
expected. The rest of the bene�t is an (almost) unconditional �xed amount that all treatment
households receive to spend on improved nutrition. This translates into an expansion of the
household�s budget set, and hence is likely to lead to a reduction in the labor supply of household
members through the income e¤ect.
The objective of this paper is to decompose any changes in the time allocation of house-
hold members as a result of the program into income, own-substitution, and cross-substitution
e¤ects. Identi�cation relies on the predictions of a very simple unitary household model and
exploits the random placement of bene�ts across communities in the evaluation sample, base-
line household composition, and the program�s scheme of incentives. Results show increases in
schooling and reductions in children�s labor participation in market and nonmarket activities,
primarily through the own-substitution e¤ect. Although the program does not seem to substan-
tially alter adults�time allocation, there is evidence that adult women in the house substitute
for children�s time in domestic and farm work. These �ndings indicate that Oportunidades is far
from providing disincentives to adult female labor supply or generating an increase in welfare
dependency amongst this population group. Rather, the program is motivating a reallocation of
time towards nonmarket work activities for adult female such that children can attend school.
This paper contributes to the existing literature on the e¤ects of Oportunidades on promoting
human capital accumulation. Overall, �ndings in the literature suggest positive impacts of the
3
program on school enrollment (Schultz 2004, Parker and Skou�as 2001) and on grade progression
(Behrman et al 2006, Dubois et al 2008). Concerning labor force participation, Parker and
Skou�as (2000, 2001) �nd reductions in child labor in paid activities for boys. For adults,
both Parker and Skou�as (2001) and Skou�as and DiMaro (2008) �nd very little impacts of the
program on adult labor supply and leisure. This piece builds on Parker and Skou�as (2000) in an
attempt to identify any changes in intrahousehold time allocation resulting from the intervention,
as well as the transmission channels of these e¤ects.
The remainder of the paper is organized as follows. Next section describes the Oportunidades
intervention in greater detail. In section 3, I discuss the expected e¤ects of the program on time
allocation within a unitary model of household behavior framework. Section 4 describes the
data. Section 5 and 6 develop the identi�cation strategy and present results on individual and
intrahousehold time allocation. Finally, section 7 concludes.
2 The Rural Oportunidades Program
Oportunidades is an antipoverty program that combines a traditional cash transfer approach
with �nancial incentives for positive behavior in health, education and nutrition. Over its �rst
three years, the program extended bene�ts to almost all eligible families living in rural areas.
Starting in 2001, it expanded to urban areas and currently covers around 5 million families all
over Mexico. It is one of the largest conditional cash transfers program in the world, distributing
approximately 3 billion US dollars to some 5 million bene�ciary households in 2008.3
2.1 Bene�t Structure
The cash transfers are disbursed bimonthly in two forms. The �rst is a �xed nutritional grant
intended for families to spend on more and better nutrition. It is complemented with nutritional
supplements and immunization directed to 0 to 2 year olds, and to pregnant and lactating women.
Children 2 to 5 also receive nutritional supplementation if signs of malnutrition are detected.
Families must complete a schedule of visits to health care centers and follow monthly talks on
disease prevention, nutrition and health �called "pláticas" �to receive the nutritional stipend.
3http://www.oportunidades.gob.mx/informacion_general/main.html
4
The second is an educational grant given for each child younger than 18 and enrolled in school
between the third grade of primary school and the third grade (last) of junior high.4 It rises
as children progress to higher grades in order to re�ect their monetary contribution to family
income if they were working. In particular, the transfer amount increases substantially after
graduation from primary school and is higher for girls than boys during junior high and high
school to try to reduce dropout during the transition year. Bene�ciary children also receive
money for school supplies once or twice a year. Students lose their bene�ciary status if they
miss school more than 15 percent of the time for unjusti�ed reasons or if they repeat a grade
twice.
There is an upper limit in the total transfer amount received per household. Whenever this
cap is hit, the educational monthly grants are proportionally reduced so that the total household
transfer amount equals the maximum. Table 1 contains the monetary values of the transfers
in October 1997 pesos (baseline). The cash transfer is directly given to the female head of the
household for two reasons: �rst, because Oportunidades seeks to promote gender equality; and
second, because it is believed that resources controlled by women are more likely to be translated
into greater improvements in children�s wellbeing than resources controlled by men. On average,
the total transfer received represents over 20 percent of total household income.
2.2 Bene�ciary Eligibility and Enrollment
When Oportunidades was �rst rolled out in rural areas in 1997, program eligibility was deter-
mined in two stages (Skou�as et al 2001). First, underserved communities were identi�ed based
on the proportion of households living in poverty as de�ned in the 1995 population census.
Second, low-income households in these communities were chosen using a proxy means test.
Pre-intervention (baseline) data to construct the index was collected in October 1997 on all
households in eligible communities through the Survey of Household Socioeconomic Characteris-
tics (Encuesta Socioeconómica de Hogares, or ENCASEH). This classi�cation scheme designated
52 percent of households in selected communities as eligible for bene�ts ("poor"). Subsequently,
the Government decided that a subset of the "non-poor" households had been unduly excluded.
They expanded the eligibility criteria to include a set of slightly wealthier households in a process
4Beginning in 2001, high school scholarships are granted to all bene�ciaries younger than 21 enrolled in school.
5
called "densi�cation". However, many of these households were not incorporated into the pro-
gram until later because of administrative delays and operational di¢ culties (Hoddinott and
Skou�as 2004).
All eligible households living in treatment localities were o¤ered Oportunidades and over 90
percent enrolled. Once enrolled, households received bene�ts for a three-year period with the
possibility of being recerti�ed if all household members obtained the prescribed preventive med-
ical care, children attended school and mothers participated in the "pláticas". About 1 percent of
the households were denied the cash transfer for non-compliance. New households were not able
to enroll until the next certi�cation period, which prevented migration into treatment communi-
ties for Oportunidades bene�ts. Households in rural areas were "recerti�ed" (reassessed with a
proxy means test) after three years on the program to determine future eligibility. If a household
was recerti�ed as eligible, it would continue receiving bene�ts. If not recerti�ed, the household
was guaranteed six more years of support before transiting o¤ the program. Thus, households
could expect a minimum of nine years of bene�ts upon enrollment (OPORTUNIDADES 2003).
2.3 Experimental Design and Data Sources
A remarkable aspect of Oportunidades was the commitment of the Government to conduct a
rigorous evaluation. Given budgetary and logistical constraints, the Government could not enroll
all eligible families in the country simultaneously and had to phase in enrollment over time. For
ease of implementation, the Government decided that whole communities would be enrolled at
a time and that they would be enrolled as fast as possible so that no eligible household would
be kept out of the program.
As part of this process, the Government randomly chose 506 experimental communities in
seven states in rural Mexico over which it randomly phased in treatment: 320 communities (60
percent) were then randomly allocated to the treatment group, and the remaining 186 commu-
nities (40 percent) to the control group. Eligible households in treatment communities began
receiving bene�ts in March/April of 1998; whereas eligible households in control communities
were not incorporated until November/December of 1999. In order to minimize anticipation
e¤ects, households in control communities were not informed that Oportunidades would provide
bene�ts to them until two months before incorporation. Behrman and Todd (1999) con�rm
6
that the randomization balanced control and treatment communities. Attanasio et al (2005)
explicitly test but �nd no evidence of anticipation e¤ects amongst control households.
Every six months, from October 1998 to November 2000, Oportunidades collected detailed
information on a host of topics in the Evaluation Surveys (Encuestas de Evaluación, or ENCEL).
The complete evaluation sample consists of a panel of the 24,077 eligible and ineligible households
in the 506 eligible experimental communities. In 2003, the evaluation added a new control group,
which was surveyed along with the original 506 communities. All these data can be merged at
the individual level to the ENCASEH census for a total of seven rounds of data between 1997
and 2003.
Skou�as (2005) provides a detailed description of the program and the evaluation design,
along with a summary of impacts over the �rst three years of implementation.
3 Oportunidades and Intrahousehold Time Allocation
This section develops a simple family labor supply model that characterizes the e¤ects of the
economic incentives o¤ered by Oportunidades in the allocation of time of all household members
under a unitary household framework. The model �based on Ashenfelter and Heckman (1974)
and Newman and Gertler (1994) �allows for choices between di¤erent work activities and breaks
down changes in individual time use due to the program into income, own- and cross-substitution
e¤ects.
Consider a household h consisting of i = 1; :::; I members (a = 1; :::; A � I adults, q =
1; :::; Q � I children and k = 1; :::;K � Q children eligible for bene�ts, such that A+Q = I and
K � Q) with a utility function of the form:5
U = U(L1; :::; LI ; C;X; ") (1)
where C is total (aggregate) household consumption and Li is the leisure time of individual i,
X represents observable heterogeneity (including individual, household and community char-
acteristics) and " represents unobservable household heterogeneity. Each household member is5A child is eligible for bene�ts if she: (i) lives in a poor family; (ii) has completed between second and eight
grade of basic education; and (iii) is younger than 18. Any eligible child e¤ectively receives bene�ts if she: (i)
lives in a treatment community; and (ii) is enrolled in grades 3 to 9 of basic education and attends school.
7
endowed with a total amount of time T that she can allocate between leisure and j = 1; :::; J
di¤erent activities: market work (m), farm work (f) and domestic work (d). Children can
additionally engage in schooling (s). The household faces the following budget constraint:
Pi
Pj 6=sW
ji H
ji + Y � PC +
Pi=qW
si H
si (2)
where Hji are the number of hours individual i spends on activity j and W
ji is i�s marginal
return to activity j 6= s.6 The price of schooling, W si , equals the real cost of schooling: fees,
books, transportation, etc. P is the price of the aggregate commodity good and Y is nonlabor
income. The I time constraints are:
Pj H
ji + Li = T 8i (3)
and the I � J non-negativity constraints are:
Hji � 0 8i; j (4)
After replacing the I time constraints in (1), the household optimization problem boils down to
choosing the hours each family member supplies to each of the J activities and total family con-
sumption C that maximize (1) under (2) and (4). Assuming that the second order conditions for
a maximum are satis�ed, the I � J �rst order conditions: Hji � 0 and H
ji
�@U(:)=@Li@U(:)=@C �
W jiP
�=
0;8i; j, plus the family budget constraint in (2) yield the resulting demands for aggregate con-
sumption and individual leisure:
C = C(W;P; Y ;X; ") (5)
Li = Li(W;P; Y ;X; ") 8i (6)
where W = (Wm1 ; :::;W
mI ;W
f1 ; :::;W
fI ;W
d1 ; :::;W
dI ;W
s1 ; :::;W
sQ). From the I time constraints in
(3), hours supplied by individual i to each activity j are:
Hji = H
ji (W;P; Y ;X; ") 8i; j (7)
6This amounts to assuming that household members have di¤erent productivities in the performance of a given
activity, and hence allows for specialization according to comparative advantages. This assumption is supported
by the data: indeed, (statistically signi�cant) di¤erent wages are reported at the community level for male, female
and children for di¤erent types of activities.
8
All standard consumer theory results hold under this framework. However, two substitution
e¤ects should be now considered: (i) the substitution e¤ect given the variation in a family
member�s (shadow) wage on its own labor supply, the own-substitution e¤ect ; and (ii) the
income-compensated e¤ect of such wage variation on any other family member�s labor supply,
the cross-substitution e¤ect.
In this stylized household, the Oportunidades intervention amounts to:
1. a reduction in the price of schooling given the educational grant, tk > 0; received by each
child k eligible for bene�ts and conditional on him attending school: dW sk = �tk < 0,
where d denotes variation.
2. an increase in nonlabor income given the (almost) unconditional nutritional grant n: dY =
n > 0.7
Therefore, for each household member i in a treatment household, the e¤ect of the Oportu-
nidades package (Pk dW
sk ; dY ) on her time supplied to activity j is:
dHji =
Pk@Hj
i@W s
kdW s
k +@Hj
i@Y dY 8i; j (8)
Using the Slutsky equation, the program�s total e¤ect on Hji can be re-written as:
dHji =
@ bHji
@W sidW s
i +Pk 6=i
@ bHji
@W skdW s
k +h�PkH
jkdW
sk + dY
i@Hj
i@Y 8i; j (9)
where bHji =
bHji (W;P;U ;X; ") is the Hicksian (utility compensated) labor supply function.
8
The �rst term in equation (9) is the own-substitution e¤ect on i = k�s time spent on activity
j from receiving the educational grant. By de�nition, this term will be nonzero for eligible7 It is not completely accurate to state that the nutritional supplement is unconditional. Mothers have to
perform a certain number of activities �namely, attend "pláticas" and take the children to health centers �which
are time consuming and therefore might change behavior through substitution e¤ects of a nature similar to those
generated by the educational grant. However, the time commitments for the mothers are much smaller than those
implied by the educational grant for the children and can arguably be ignored.8This expression assumes that Oportunidades has no e¤ect on local prices or wages (dP = dW j
i = 0;8j 6= s).
Hoddinot and Skou�as (2004) explicitly test for changes in food prices in treatment communities and �nd no
e¤ects. Similarly, Gertler et al (2008) �nd no signi�cant di¤erences in wages between treatment and control
communities. The theoretical model also ignores any community spillover e¤ects. This simplifying assumption
will be relaxed in the empirical part, which will allow for correlations across households in the same community.
9
children k living in treatment households and e¤ectively receiving the educational grant; and
zero otherwise. The second term is the sum of all cross-substitution e¤ects on Hji given the
educational grants received by the (other) k children in a treatment household �i.e. the (other)
children in a treatment household that receive the educational stipend. Hence, cross-substitution
e¤ects will arise for any individual i (or k) living in treatment households where there are (other)
bene�ciary children. The third term in equation (9) is the total income e¤ect coming from the
educational and the nutritional grants. All household members in treatment households are
subject to the income e¤ect given the (almost) unconditional nutritional grant, regardless of
whether there are eligible children enrolled in and attending school in the household.
It seems reasonable to assume that all j work activities are inferior goods�@Hj
i@Y < 0;8j 6= s
�,
and that schooling is a normal good�@Hs
i@Y > 0
�. This would imply a negative total income term
in equation (9) for the labor supply equations of any household member, and positive for the
schooling demand of school aged children.
For any bene�ciary child k, the own-substitution e¤ect will go in the same direction as the
income e¤ect under two further sensible assumptions: (i) schooling is an ordinary good, which
implies @ bHsk
@W sk< 0; and (ii) schooling and each of the j 6= s work activities are substitutes,
which implies�@ bHj
k@W s
k> 0
�.9 These e¤ects might however be o¤set by nonzero cross-substitution
e¤ects resulting from the reduction in the price of schooling of other bene�ciary children (likely
siblings) in the household. The sign of the cross-substitution e¤ect between any two individuals
will be positive or negative depending on whether schooling and work of the two individuals
are substitutes or complements. Thus, the total e¤ect on schooling demand and labor supply
depends on the magnitude and direction of the di¤erent e¤ects at play and remains an empirical
question.
For non-bene�ciary children in treatment households, an analogous reasoning predicts an
increase in schooling and a reduction in participation and hours spent in work related activities
9The existing empirical evidence suggests that market work and schooling are substitutes (see for example,
Ravallion and Wodon 2000). On the contrary, farm and domestic work are more likely to involve activities
compatible with schooling rather than perfect substitutes. Hence, as@ bHj
k@Ws
k! 0, smaller own-substitution e¤ects
on farm and domestic hours will be expected.
10
due to the income e¤ect. The own-substitution e¤ect for these children is zero as they are
not direct bene�ciaries of the educational grant. The total e¤ect of the program on their time
allocation will thus depend on the sign and size of the cross-substitution e¤ects, which will in
turn depend on whether schooling and work between non-bene�ciary and bene�ciary children
in the household are complementary or substitute activities.
Similarly, income and cross-substitution e¤ects are likely to a¤ect the labor participation of
adults in treatment households. For each activity j, a reduction in participation is expected as a
result of the income e¤ect, @Hji
@Y > 0. On the other hand, an increase in participation is expected
given the reduction in the price of schooling of k: @ bHjk
@W sk< 0 and dW s
k < 0. The underlying
assumption for positive cross-substitution e¤ects between adults and bene�ciary children is that
their work activities are substitutes. If so, the educational grant implies an increase in the
shadow value of adults�time to activities children would otherwise engage in if they were not in
school.10
Next, I turn to the description of the expected e¤ects of Oportunidades on the leisure time
of household members. Replacing equation (3) in equation (9), the total e¤ect of the program
on i�s leisure becomes:
dLji = ��@ bHs
i@W s
i+Pj 6=s
@ bHji
@W si
�dW s
i �Pk 6=i
�@ bHs
i@W s
k+Pj 6=s
@ bHji
@W sk
�dW s
k �
�hP
k
��T +Hs
k +Pj 6=sH
jk
�dW s
k + dYi�
@Hsi
@Y +Pj 6=s
@Hji
@Y
�8i; j (10)
As before, the �rst, second and third terms identify �in this order �the own-substitution, cross-
substitution and income e¤ects. For non-bene�ciary children and for adults, i 6= k, increases
in leisure time are expected if leisure is assumed a normal good. However, the expected total
program e¤ect on leisure is ambiguous as it depends on the sign of the cross-substitution e¤ects �
namely, whether the leisure times of household members i and k are complements or substitutes.
10Note that perfect substitutability between adults�and children�s work is not required for these implications to
hold. As a matter of fact, the size of the cross-substitution e¤ect �i.e., whether adults partially or fully substitute
for children�s time in activity j �will depend on the degree of substitutability. Because the time allocated to
nonmarket activities is likely to be more elastic � this is, more of a substitute � to changes in other household
member�s wages than the time allocated to market work, larger cross-substitution e¤ects regarding domestic and
farm work activities are expected.
11
For bene�ciary children, i = k, there is an added degree of uncertainty because of the nonzero
own-substitution e¤ect. This e¤ect will have a positive or a negative sign depending on whether
the reductions in the total number of hours working implied by the program are larger than the
increases in total schooling time, or vice versa.
4 The Data
4.1 Analysis Sample and Balance
The dependent variables in this paper are constructed using information on a time use module
associated to the May 1999 evaluation survey. This module contains information on the amount
of time each household member older than 8 allocated to the 19 di¤erent activities �listed in
Table 2 �during the day before the interview. For the purpose of analysis, these activities have
been classi�ed into four di¤erent work categories �market (paid) work, farm work, domestic
work and the sum of the three (all work) �schooling and leisure.11 Leisure is de�ned as a residual
(i.e. 24 hours minus total time devoted to other activities) and includes time spent sleeping,
eating and socializing, which is estimated around 9 daily hours for Mexico (INEGI 2000).12
The sample of analysis is restricted to individuals that were not interviewed on a Sunday
or a Monday. Weekday and weekend time use patters are likely to be di¤erent, specially for
children in school. Basic descriptive statistics evidence that children spend less time in school
and increase their participation in household chores and farm related activities over the weekend.
Similarly, adults tend to spend less time working for a wage and more time working in the house
or enjoying leisure during weekends. As the main interest here is the e¤ect of the Oportunidades
educational incentives on the time use of all household members, I focus on the study of weekday
time use exclusively.
11Other activities, such as community work and transportation, have also been considered. However, estimates
(available upon request) are highly unreliable given the overall low participation rates in these activities, which
is likely due to the short reference period �i.e. the day before the interview.12One of the main challenges in time use surveys is the adequate treatment of activities performed simultaneously
in order to avoid double accounting. The Oportunidades questionnaire addresses this issue with a complementary
question that asks individuals to rank activities executed at the same time by order of importance. However,
answers to these questions are di¢ cult to interpret as they often refer to consecutive �rather than simultaneous
�activities, such as transportation and schooling or paid work.
12
Moreover, the analysis is also limited to eligible ("poor") households as classi�ed under
the original classi�cation scheme. In rural Mexico, children enroll in school for the following
academic year as early as in May. Hence, time allocation decisions in May 1999 are a¤ected by
choices made in May 1998. In order to observe behavioral household responses (children school
enrollment) to the program incentives, households should have received or at least be aware of
their inclusion in the program by May 1998. Administrative data on transfer payment records
shows that no household re-classi�ed as eligible under the "densi�cation" process had received
payments by then (Hoddinott and Skou�as 2004). In order to not attribute treatment e¤ects to
untreated households, I exclude "densi�ed" households from the analysis.
The �nal sample consists of the 35,148 individuals in 7,967 eligible households with weekday
time use information in May 1999. These represent almost two thirds (64 percent) of all eligible
households originally classi�ed as "poor". The treatment group includes 2,911 (37 percent)
"poor" households in treatment communities, and the control group the remaining 5,056 (63
percent) "poor" households in control communities. Note that the proportion of households
with existing time use data in the treatment and control groups is very similar to the original
(randomized) distribution �60 percent treatment and 40 percent control. At the time of the
time use survey, households in the treatment group had enjoyed bene�ts for over a year (since
May/April 1998), whereas no household in control communities had yet received any program
bene�ts. Recall that randomized out communities were not phased in until November/December
1999.13 It is important to clarify at this point that, treatment is de�ned as "intention to treat"
rather than "actual treatment" in the current setting. This implies that all households that
were o¤ered treatment are included in the sample, regardless of their �nal decision to take up
the intervention.14
Table 3 presents pre-intervention characteristics for households in the treatment and control
communities in the estimation sample. The head of the household is often the man and the
principal source of family income. Both he and his spouse have, on average, less than primary
school completed, conditional on having some education. The average household size is six13Time use data were collected again in November 2000. By then, however, all eligible households in both
treatment and control communities were receiving bene�ts.14Previous studies report high compliance rates � at around 90 percent � and low non-compliance with the
program rules �of about 1 percent (see Skou�as 2005, Gertler et al 2008).
13
members and around 65 percent of the households have at least one child eligible to receive
primary school bene�ts. The closest secondary school, an indicator of the distance to the largest
locality in the area and a proxy of the availability of labor markets, is on average about 3 km
far. The last column tests the di¤erence in means for pre-intervention characteristics between
the treatment and control groups. Out of the 32 characteristics tested, only the number of hours
the head�s spouse works for a wage is statistically signi�cant � at the 10 percent level. This
well-balanced sample suggests that the randomization process was e¤ective in generating truly
exogenous variation in treatment, and that the two groups are comparable on observed (and
unobserved) characteristics.
4.2 Time Use Allocation Before the Intervention
This subsection discusses the pre-intervention time allocation patterns of children (Table 4 ) and
adults (Table 5 ) in the poor rural communities Oportunidades operates. Because no data on
time use was collected at baseline, Tables 4 and 5 report participation rates and daily hours
devoted to the di¤erent activities in May 1999 by household members in control communities.
Arguably, the time use of the control group in May 1999 is a good proxy of the time allocation of
the treatment group before the intervention for two reasons: (i) the treatment and control groups
share similar characteristics and are hence comparable, as seen in the previous subsection; and
(ii) the control group had not received bene�ts by May 1999.
A common trend in Table 4 is the negative correlation between children�s schooling and work.
While school enrollment falls with age, participation in any of the work activities considered �
market, farm or domestic work � increases. However, there are some di¤erences by gender.
Younger girls, ages 8 to 11, show higher school enrolment than boys in the same age group. But
this pattern is reversed during adolescence, suggesting that girls drop out of school earlier than
boys and especially during the transition from primary to junior high school (at ages 12 or 13,
on average).15 Moreover, while teenage girls �ages 14 and older �engage in household chores
more than in schooling, teenage boys�participation in any type of work is higher than schooling
15School attendance rates are underreported in this survey round compared to other ENCEL rounds. Parker
and Skou�as (2000) point out the fact that school may have already �nished in rural isolated communities or may
be close to its end at the time the time use module was carried out as a possible explanation.
14
only at later ages (16 and over). On average, boys�market and farm work labor supplies are
higher than girls�at all ages.
The amount of time boys and girls devote to work increases with age, although some types
of work are more time demanding �and hence interfere with schooling more �than others. For
instance, domestic tasks require no more than one and a half hours on average for boys and
between 2 and 4 hours for girls; market activities demand over 7 daily hours; and farm work
takes up between 3 to 4 hours per day for boys and less than 2 hours for girls. This suggests
that while domestic and farm work are compatible with schooling, participation in market work
is a full time occupation.
The time use patterns for adult men and female in Table 5 re�ect a traditional gender division
of labor in these communities. While less than 7 percent of women report working for a wage,
over 90 percent participate in domestic work almost full time �over 5 and a half hours per day,
on average. On the other hand, men show higher participation rates in market work (60 percent)
than in domestic work (42 percent). As for children, market work is a full-time activity for both
men and women, requiring 7 to 8 hours per day on average. One �fth of women and one third of
men participate in farm activities. Conditional on participation, however, men devote three and
a half hours more to working the farm land and taking care of animals than women. Finally,
men enjoy less leisure time than women, which is at odds with previous evidence from time use
studies in urban Mexico (INEGI 2002). One possibility is that female leisure is overestimated
in the data if, for example, not all of the relevant housework activities women engage in on a
typical day in rural areas were included in the questionnaire.
5 Program Impact on Individual Time Use
5.1 Empirical Strategy and Identi�cation
This section presents intent to treat estimates of Oportunidades on the time allocation of children
(ages 8 to 17) and adults (ages 18 and over) in May 1999. This accounts for 18 months of program
exposure for eligible households in treatment communities. Because randomized out (control)
communities had not yet been incorporated in the program at the time, the average treatment
15
e¤ect is unbiasedly identi�ed with a treatment dummy variable as follows:
Yih = �o + �1Th +Pr�rXihr + "ih 8i = 1; ::; N (11)
where Yih can either be leisure time of individual i in household h; participation of i in activity
j; or hours i spends on activity j. The following j activities are considered: market work, farm
work, domestic work and schooling (only for children). Th is a dummy that takes on the value
of one if i lives in a treatment community and 0 otherwise. Xihr represents a vector of r char-
acteristics including: current individual characteristics (age, age squared, education, ethnicity,
and rank amongst siblings and parental characteristics �for children only); baseline household
demographics (sex of the head, and household size and composition), baseline household assets
(bathroom, electricity, land ownership and draft animal ownership); community distance to the
closest municipality and state dummies. The error term "ih includes individual preferences for
schooling and work and other unobservables, which are uncorrelated with the treatment sta-
tus by construction. Indeed, randomization guarantees that Th ? "ih. Because Oportunidades
sampled (and randomized) entire communities, I cluster standard errors at the community level.
For each activity type, equation (11) is separately estimated for boys, girls, men and women.
The leisure equation is estimated using OLS. The impact on participation in each activity j is
estimated using a probit model (variation at the extensive margin). Conditional on participation
changes in the amount of daily hours spent on each activity (variation at the intensive margin)
are estimated using a left-censored tobit model with censor value at zero. The mean of the
observation-by-observation marginal e¤ects have been computed for the probit and tobit models
as discussed in section 8.1 in the Appendix. For the tobit models, marginal e¤ects are reported
for the mean of all positive outcomes �i.e. conditional on participation.
In a second set of estimates, the household treatment status is interacted with several dum-
mies, Dihg that equal one if individual i belongs to age group g. The estimation equation
becomes:
Yih = �o +Pg�gDihg +
Pg gThDihg +
Pr�rXihr + "ih 8i = 1; ::; N (12)
The coe¢ cient on the interaction term ThDihg is thus an estimate of the average program impact
for those individuals in age group g. I consider the following age groups for children: 8 to 11
and 12 to 17; and 18 to 24, 25 to 34, 35 to 44, 45 to 54 and more than 55 years, for adults.
16
Identi�cation relies on the exogeneity of the treatment status conditional on observables:
Th ? "ih p (Xihr; Dihg). This assumption, which is guaranteed by the randomized allocation
of treatment, implies that E ("ih j Xihr; Dihg) = 0. Randomization also implies that treatment
is independent of individual characteristics, (Th ? Dihg), and hence the treatment e¤ect for
di¤erent age groups is orthogonal to the unobservables: ThDihg ? "ih p (Xihr; Dihg). Therefore,
E ("ih j ThDihg; Xihr; Dihg) = 0 follows and the di¤erent b g�s are identi�ed.5.2 Results on Children�s Time Use
Table 6 presents estimates of the overall program impact on the time allocation to the di¤erent
activities considered (in columns) for boys (Panel I) and girls (Panel II). Estimates on the pooled
sample (Models A) and by age (Models B) are reported in rows.16
Results show a signi�cant reduction in leisure time for girls ages 12 to 17 (column 1, Model
IIB). Indeed, as a result of the program, teenage girls increase the amount of time they spend in
school (column 3, Model IIB) by more than they reduce the number of hours they work (column
5, Model IIB). For boys, the increase in school hours (column 3, Model IB) and reduction in
work time (column 5, Model IB) are quantitatively similar. As a result, there is no signi�cant
impact on their leisure time. Increases in school participation are also larger for teenage girls
(8.7 percentage points) than for teenage boys (4.3 percentage points). This �nding is consistent
with previous research on Oportunidades (Schultz 2004; Parker and Skou�as 2000; 2001) and is
likely to respond to the program structure of incentives �i.e. larger transfer amounts for girls
during junior high school. Note also that impacts are larger at the intensive margin, which is
not too surprising considering the relatively high primary school enrollment rates.
The reduction in work hours for girls (column 5, Model IIA) � and especially amongst
teenage girls (column 5, Model IIB) �is primarily driven by a reduction in the number of hours
they devote to household chores (column 11, Models IIA and IIB). This �nding is of particular
importance given that 60 percent of teenage girls spend, on average, more than 3 hours per day
performing domestic tasks (see Table 4 ). For boys, the program reduces their time in market,
farm and domestic activities, although these reductions are not statistically signi�cant. The sum
16All regressions include the covariates listed above � point estimates on the covariates are available upon
request. Results (upon request) are qualitatively robust to the exclusion of these covariates.
17
of e¤ects is, however, signi�cant (column 5, Models IA and IB). Further disaggregation of the
e¤ects by ages shows a dramatic reduction �of 3.3 percentage points or a 65 percent decrease!
�in market work participation for boys ages 12 to 13 (results available upon request).
Figures A1 to A3 plot the distribution of the estimated observation-by-observation mar-
ginal e¤ects for those activities where the program has a larger impact: boys�schooling, girls�
schooling, and girls�domestic work. For each �gure, the �rst column plots the unconditional
distribution of the marginal e¤ects; the second plots marginal e¤ects by age; and the third by
the number of children eligible for educational bene�ts in the household. E¤ects on participation
and hours are plotted on the �rst and second rows, respectively.
As the �rst graph in Figure A1 shows, around 20 percent of boys increase attendance by the
maximum value of 3.7 percentage points. On the other hand, the distribution of schooling hours
follows a normal, with a majority of boys experiencing increases similar to the average increase
of 0.17 hours. The distribution of e¤ects by age shows that the marginal impact on participation
is higher for teenage boys (ages 12 to 17). Nonetheless, e¤ects are rather constant by number of
eligible children in the household, suggesting little cross-substitution e¤ects in schooling. The
plots in Figure A2 show very similar distributions of marginal e¤ects on girls�schooling.
Concerning girls� domestic work, around 25 percent of girls experience the maximum re-
duction of 2.1 percentage points (�rst graph in Figure A3 ). By age, the distribution of e¤ects
suggests stronger and less spread e¤ects for girls 12 to 15. More children eligible for bene�ts in
the household is associated with larger reductions in domestic work participation, which suggests
that siblings substitute each other in the performance of household chores.
5.3 Results on Adults�Time Use
Table 7 presents estimates of impact on time allocation for men (Panel I) and women (Panel
II). The structure of this table is analogous to that of Table 6.
Overall, results suggest very little impact of the program on adult time use, which is con-
sistent with the �ndings in Parker and Skou�as (2000) and in Skou�as and DiMaro (2008).
For men, there is a reduction in the amount of total hours worked (column 3, Model IA) that
translates in an increase in leisure time, signi�cant at the 10 percent (column 1, Model IA).
18
E¤ects are stronger for men ages 25 to 34 (columns 1 and 3, Model IB). Concerning women�s
time allocation, there is a 0.2 hours increase in household chores by women 25 to 34. Results
also show an almost signi�cant increase �signi�cant at the 10 percent �of 4.1 percentage points
in market work participation for women 45 to 54.
Figure A4 plots the estimated marginal e¤ects on women�s participation and hours to do-
mestic work. In both cases, the distribution of marginal e¤ects is skewed to the right. This
implies that, while there is very little variation at the extensive margin, over 70 percent of the
women in the treatment group increase the number of hours to domestic work. By age, the
distribution of marginal e¤ects suggests small but positive program impacts for middle aged
women; and negative �and less precisely estimated (more disperse) �e¤ects for younger and
elder women.
6 Program Impact on Intrahousehold Time Allocation
The simple theoretical framework developed in section 3 predicts a reduction in children�s par-
ticipation in work activities and an increase in adults� participation in those same activities
given the reduction in the price of schooling. Indeed, the estimates presented in the previous
section suggest that Oportunidades has reduced teenage girls�time and participation in domestic
work, and boys�participation in market work. Moreover, there is evidence of an increase in the
number of hours prime-age women spend in household chores and working for a wage. Together
these �ndings are indicative of potential cross-substitution e¤ects in the household, channeled
through a reallocation of domestic (market) activities between mothers and daughters (sons).
This section aims to formalize this preliminary evidence. Based on equation (9), I use the
randomized treatment status and household demographic structure at baseline �this is to say,
the presence of eligible children in the household �to decompose the e¤ects of Oportunidades
on the labor supply and schooling demand of each household member into own-substitution,
cross-substitution and income e¤ects.
19
6.1 Identi�cation of Income and Substitution E¤ects
The reduced form proposed to disaggregate the total program impact for individual i in house-
hold h is:
Yi = �o + �1Th + �2Bi + �3ThBih + �4PBh + �5ThPBh + (13)
+4P
k 6=i;k=1 1ksexhk +
4Pk 6=i;k=1
2kgradehk + 3 ln(transfer)h +Pr�rXihr + "ih
where Th is a dummy equal to 1 if household h is in a treatment community and Bih takes on
the value of 1 if individual i in household h is eligible to receive the educational grant. Hence,
the coe¢ cient on the interaction term ThBih, b�3, identi�es the own-substitution e¤ect � i.e.the term
�@ bHj
i@W s
i
�in equation (9). Indeed, ThBih = 1 if i is a child eligible to receive bene�ts
(Bih = 1) and lives in a treatment household (Th = 1). As argued in section 3, for these children,
the own-substitution e¤ect arises as a result of the increase in the opportunity cost of time in
school that the educational grant they receive entails. Household members that are not directly
entitled to the educational stipend (Bih = 0), have no own-substitution e¤ect.
Similarly, PBh is a dummy variable that equals 1 if there are children eligible to receive the
educational grant in household h (other than i, if i herself is eligible for bene�ts). Arguably,
the coe¢ cient on the interaction term ThPBh, b�5, identi�es the sum of cross-substitution e¤ectsgiven the educational stipends that (other) eligible children living in the household receive �i.e.�P
k 6=i@ bHj
i@W s
k
�in equation (9). As seen, cross-substitution e¤ects may alter the time allocation
patterns of individuals living in treatment households (Th = 1) where there are children directly
bene�ting from the educational stipends (PBh = 1).17
In this setting, the coe¢ cient on the treatment variable, b�1, identi�es the (residual) incomee¤ect. In treatment households with no children eligible to receive the educational grant, both
Bih = 0 and PBh = 0, and the total program impact equals the income e¤ect coming from
the nutritional grant. However, in households with children eligible to receive the educational
17The speci�cation in (13) implicitely assumes that cross-substitution e¤ects are equal across households with a
di¤erent number of children receiving the educational stipend. I estimated a modi�ed version of (13) where PBh
is replaced with dummies accounting for the presence of di¤erent numbers of eligible children in the household
and tested the validity of this assumption.
20
grant, the income e¤ect is the remaining impact of the program, once the e¤ect of all educational
grants has been accounted for.
As noted, identi�cation strongly relies on both the randomization of the treatment status
and the use of pre-intervention data to construct the regressors of interest. The variables Bih
and PBh re�ect children eligible for bene�ts �i.e. potentially bene�ciary children �rather than
children e¤ectively receiving bene�ts � i.e. actual bene�ciaries � in order to correct for the
endogeneity linked to being a bene�ciary. Recall that eligible children receive the educational
grant conditional on attending school, which implies that being an actual bene�ciary re�ects in
part the decision of going to school. I determine potentially bene�ciary children in May 1999
taking the grade the child is enrolled in at baseline and moving it forward assuming no repetition
or dropout. The number of potential bene�ciaries is thus an overestimate of the number of actual
bene�ciaries. Moreover, because treatment status is independent of baseline characteristics and
school enrollment, Th ? "ih p (Xihr; Bih; PBh) and so ThBih ? "ih p (Xihr; Bih; PBh) and
ThPBh ? "ih p (Xihr; Bih; PBh) follow.
Besides the control variables in Xihr de�ned above, the speci�cation in (13) additionally
controls for the gender and the potential grade of the �rst four bene�ciary children (excluding i
if i is a bene�ciary child). These variables induce variation in the amount of transfers treatment
households receive.18 In order to obtain an estimate of the income e¤ect across households re-
ceiving the same monetary incentive, I also include the log of the potential total transfer amount
a household would receive in May 1999 as a regressor. As for potential children, this amount is
computed taking household composition at baseline and applying the program bene�t allocation
rules assuming no school dropout, no grade repetition and universal enrollment. Finally, note
that equation (9) implies an additional assumption to identify the income e¤ect: eligible children
in households receiving the same transfer amount (Pk dW
sk + dY ) must have similar time use
patterns �i.e. supply the same number of hours to activity j,�P
kHjk
�.
18Although a household can have up to eight potential bene�ciary children, 99 percent of the households in the
sample have no more than four. Moreover, the cap on the maximum amount e¤ectively implies that a household
does not receive transfers for more than three or four children.
21
6.2 Results on Intrahousehold Time Allocation
Tables 8 and 9 report coe¢ cient estimates of the income, own-substitution and cross-substitution
e¤ects resulting from the estimation of (13) for boys, girls, adult men and adult women.
For boys and girls, the estimation sample has been restricted to teenagers (children ages 12 to
17) since, as seen in section 5.2, it is within this age group that larger program impacts on time
allocation are observed. As expected, results show a positive and signi�cant own-substitution
e¤ect on participation and hours in school for boys (second and third columns in Table 8, Panel
I). Moreover, the presence of other bene�ciary siblings signi�cantly reduces the amount of time
they devote to school related activities (negative cross-substitution e¤ects). This suggests that
these households continue to face the so-called quantity/quality trade o¤ in educational choices,
even after extra resources are poured into the household. For girls, the own-substitution and
cross-substitution e¤ects go in the same direction as for boys but are not statistically signi�cant.
Consistent with the theory, there is strong evidence of a negative own-substitution e¤ect
(5.9 percentage points reduction) on market work participation for bene�ciary teenage boys.
For girls, there is weak evidence �signi�cant at the 10 percent �of a negative cross-substitution
e¤ect on market work hours. The income and own-substitution e¤ects are, however, very im-
precisely estimated and not signi�cantly di¤erent from zero. Estimates also suggest a 14.8
percentage point reduction in farm work participation for teenage girls through an income ef-
fect. Nonetheless, conditional on participation, the number of hours teenage girls devote to
farm activities increases with the number of (other) eligible children in treatment households.
In other words, (older) teenage girls in treatment households spend more time performing farm
related tasks to substitute for the time previously devoted by (other) bene�ciary children in the
household. Finally, own-substitution e¤ects result in a strong reduction in girls�leisure, which
is larger than the increase given the income e¤ect. This is in line with the overall reduction in
leisure time for teenage girls reported in Table 6. The e¤ects on leisure for teenage boys are
qualitatively similar but not statistically signi�cant.
Results in Table 9 show a common pattern regarding the e¤ects of Oportunidades on women�s
time allocated to the di¤erent work activities (with the exception of market work). On the one
hand, the negative income e¤ects imply lower adult female participation and less hours devoted
22
to farm and domestic activities, and to overall work. On the other hand, the positive cross-
substitution e¤ects suggest that adult female increase their participation and time in these
activities so as to substitute for the work that eligible children in the household were performing
before the intervention. These �ndings � in line with the evidence in sections 5.2 and 5.3 �
support the theoretically predicted positive intrahousehold interactions between adult women
and eligible teenagers to free up their time from domestic and farm tasks to go to school.
Not surprisingly, while the income e¤ect results in a positive and signi�cant increase of
leisure for adult women in the household, the cross-substitution e¤ect implies a reduction in
their leisure time. The total impact �reported in Table 7 �is not signi�cantly di¤erent from
zero. Indeed, further disaggregating the results by age (results available upon request) shows
that for prime age women 25 to 44, the income and substitution e¤ects cancel each other out.
For men, however, the positive income e¤ect is larger in size than the negative cross-substitution
e¤ect (column 1, Table 9 ), resulting in the almost signi�cant �signi�cant at the 10 percent �
overall increase in men�s leisure time reported in Table 7. These �ndings suggest that women
receive the burden of labor responsibility due to the program, while men do not substitute for
their children�s time to the same extent and increase leisure time instead.19
The qualitative evidence in Adato et al (2000) endorses these results. During the summer
of 1999, the authors conducted focus groups with 230 bene�ciary and non-bene�ciary women
to learn about their perceptions on the program. The authors report: "Another reason that
women�s time burden increases is because of the need to do work that was previously done by
children who are now attending school, particularly secundaria. However, their mothers see this
as worthwhile in order for their children to study. (...). Although some women said that the
father also does some of this work, more often it was the mother.", (Adato et al 2000, p. xiii).
19While the evidence of overall increase in men�s leisure is weak, it raises some concerns of whether the program
is promoting disincentives to adult men labor supply, in particular when combined with the increases in overall
work hours reported in Table 7. Parker and Skou�as (2000) and Skou�as and DiMaro (2008) use labor supply data
from the evaluation surveys dismiss this argument. The evidence in Gertler et al (2008) on investment of part of
the transfer in productive activities would also suggest that the program is not generating welfare dependency.
23
7 Conclusions
This paper examines the e¤ect of the economic incentives provided by the Oportunidades inter-
vention on individual and intrahousehold time allocations.
Overall, results on individual time use are consistent with previous evidence on the e¤ects
of the program on labor supply and school enrollment (see Skou�as 2005 for a review). For
teenage boys, impact estimates show reductions in market work and increases in schooling,
giving further evidence in support of the hypothesis that working for a wage and attending
school are substitutes. Market work also appears as a more important deterrent to school for
boys than for girls. The program has larger positive impacts on girls�school attendance than
on boys�, and reduces the amount of time teenage girls devote to household chores. On the
other hand, there is little evidence of time reallocations amongst adults, expect for increases in
domestic work hours and market participation for middle aged women.
Subsidizing school enrollment has repeatedly been acknowledged as an e¤ective mechanism
to reduce child paid labor (Ravallion and Wodon 2000, Parker and Skou�as 2001, for instance).
This paper shows that this remains true even when nonremunerated activities are included in the
de�nition of work. More generally, conditional cash transfer programs appear as a potentially
useful tool to reduce child labor in its various forms. Moreover, the �exibility in their design
make them particularly attractive to address speci�c (context�dependent) problems. As shown
above, the particular structure of incentives of the program �increasing transfers as the child
enrolls in higher grades and o¤ering larger transfers to girls �seems to operate in the desired
direction. Indeed, children older than 12 � in general � and girls � in particular � are the
demographic groups with larger impacts on time allocation.
The second part of the paper investigates the transmission channels of these e¤ects. The
randomized allocation of treatment and the program structure of incentives are exploited to
disaggregate the total program impact on individual time use into income, own-substitution
and cross-substitution e¤ects. Reduced form estimates suggested that there are positive cross-
substitution e¤ects between adult women and children concerning the two forms of unpaid work
considered: farm work and domestic work. This suggests that rather than promoting disin-
centives to adult female labor supply and create welfare dependency, conditional cash transfer
24
programs may indeed generate positive responses amongst adult household members �in par-
ticular women �to facilitate the investment in human capital of younger members.
References
[1] Ashenfelter, Orley C. and James J. Heckman (1974), �The Estimation of Income and Sub-
stitution E¤ects in a Model of Labor Supply�, Econometrica, 42(1):73-85.
[2] Adato, Michelle, Bénédicte de la Brière, Dubravka Mindek and Agnes Quisumbing (2000),
�The Impact of Progresa on Women�s Status and Intrahousehold Relations�, International
Food Policy Research Institute Report, Washington D.C.
[3] Attanasio, Orazio, Costas Meghir and Ana Santiago (2005), �Education Choices in Mexico:
Using a Structural Model and a Randomized Experiment to Evaluate Progresa�, Institute
for Fiscal Studies Working Papers EWP05/01, London.
[4] Becker, Gary S. (1965), �A Theory of the Allocation of Time�, The Economic Journal,
75(Sept):493-517.
[5] Behrman, Jere R. (1997), �Intrahousehold Distribution and the Family�, Handbook of Pop-
ulation and Family Economics, Mark R. Rosenzweig and Oded Stark eds. Elsevier Science.
[6] Behrman, Jere R. and Petra E. Todd (1999), �Randomness in the Experimental Samples
of Oportunidades�, International Food Policy Research Institute Report, Washington D.C.
[7] Behrman, Jere R., Piyali Sengupta and Petra E. Todd (2006), �Progressing through Pro-
gresa: An Impact Assessment of a School Subsidy Experiment�, Economic Development
and Cultural Change, 54(1):237-275.
[8] Bergstrom, Theodore C. (1997), �A Survey of Theories of the Family�, Handbook of Popu-
lation and Family Economics, Mark R. Rosenzweig and Oded Stark eds. Elsevier Science.
[9] Dubois, Pierre, Alain de Janvry and Elisabeth Sadoulet (2008), �E¤ects on School En-
rollment and Performance of a Conditional Cash Transfers Program in Mexico�, Toulouse
School of Economics, mimeo.
25
[10] Ellis, Frank (1994), Peasant economics: Farm households and agrarian development, Cam-
bridge University Press, Cambridge.
[11] Evenson, Robert E. (1978), �Time allocation in rural Philippine households�, American
Journal of Agricultural Economics, 60(1):322-330.
[12] Gertler, Paul J., Sebastian Martinez and Marta Rubio-Codina (2008) �Investing Cash
Transfers to Raise Long Term Living Standards�, Policy Research Working Paper No.
3994, The World Bank, Washington D.C.
[13] Gronau, Reuben (1977), �Leisure, Home Production, and Work -the Theory of the Alloca-
tion of Time Revisited�, Journal of Political Economy, 85(6):1099-1123.
[14] Hoddinott, John and Emmanuel Skou�as (2004), �The Impact of OPORTUNIDADES on
Consumption�, Economic Development and Cultural Change, 53(1):37-61.
[15] INEGI (2002), Uso del Tiempo y Aportaciones en los Hogares Mexicanos, Instituto Nacional
de Estadística (INEGI), Aguascalientes.
[16] Juster, F. Thomas and Frank P. Sta¤ord (1991), �The Allocation of Time: Empirical Find-
ings, Behavioral Models, and Problems of Measurement�, Journal of Economic Literature,
29(2):471-522.
[17] Kooreman, Peter and Arie Kapteyn (1987), �A Disaggregated Analysis of the Allocation
of Time within the Household�, Journal of Political Economy, 95(2):223-249.
[18] Mueller, Eva (1984), �The value and allocation of time in rural Botswana�, Journal of
Development Economics, 15(1-3): 329-360.
[19] Newman, John L. and Paul J. Gertler (1994), �Family Productivity, Labor Supply and
Welfare in a Low Income Country�, The Journal of Human Resources, 29(4):989-1026.
[20] OPORTUNIDADES. 2003. �Reglas de Operación 2003�. www.progresa.gob.mx
[21] Parker, Susan W. and Emmanuel Skou�as (2000), �The Impact of Progresa on Work,
Leisure, and Time Allocation�, International Food Policy Research Institute Report, Wash-
ington D.C.
26
[22] Parker, Susan W. and Emmanuel Skou�as (2001), �Conditional Cash Transfers and their
Impact on Child Work and School Enrollment: Evidence from the PROGRESA Program
in Mexico�, Economía, 2(1):45-96.
[23] Ravallion, Martin and Quentin Wodon (2000), �Does Child Labour Displace Schooling?
Evidence on Behavioural Responses to an Enrollment Subsidy�, The Economic Journal,
110, 158-175.
[24] Schultz, Paul (2004), �School Subsidies for the Poor: Evaluating the Mexican PROGRESA
Poverty Program�, Journal of Development Economics, 74(1): 199-250.
[25] Skou�as, Emmanuel (1993), �Labor market opportunities and intrafamily time allocation
in rural households in South Asia�, Journal of Development Economics, 40(2):277-310.
[26] Skou�as, Emmanuel (2005), �PROGRESA and Its Impacts on the Welfare of Rural House-
holds in Mexico�, International Food Policy Research Institute Research Report 139, Wash-
ington D.C.
[27] Skou�as, Emmanuel, Benjamin Davis and Sergio de la Vega (2001), �Targeting the Poor in
Mexico: An Evaluation of the Selection of Households into PROGRESA�, World Develop-
ment, 29(10):1768-1784.
[28] Skou�as, Emmanuel and Vincenzo Di Maro (2008), �Conditional Cash Transfers, Adult
Work Incentives, and Poverty�, The Journal of Development Studies, 44(7): 935-960.
27
8 Appendix
8.1 Computation of Marginal E¤ects
For probit models the estimated e¤ect of switching the j-th binary variable, xj , from 0 to 1 on
the probability that yi equals one is:
MPij =
4E[yijxij ]4xj = 4�(xi�̂)
4xj = �(x1i �̂)� �(x0i �̂)
where x1i is the vector of regressors with the j-th binary variable set to 1, and similarly for x0i .
�(:) and �(:) are the density and cumulative distribution functions of a standard normal. If xj
is a continuous variable, the e¤ect is:
MPij =
@E[yijxij ]@xj
= �̂j�(xi�̂)
For the tobit model, the estimated e¤ect of a binary variable on the mean of all positive yi�s,
i.e. conditional on participation is:
MTij =
4E[yijxij ;yi>0]4xj = 4[xi�̂+�̂�(zi)=�(zi)]
4xj =�x1i �̂ + �̂
�(z1i )
�(z1i )� x0i �̂ � �̂
�(z0i )
�(z0i )
�where z1i =
x1i �̂�̂ and z0i =
x0i �̂�̂ . If xj is a continuous variable, the e¤ect on the mean of all positive
yi�s becomes:
MTij =
@E[yijxij ;yi>0]@xj
= @[xi�̂+�̂�(zi)=�(zi)]@xj
=�1� zi �(zi)�(zi)
� �2(zi)�2(zi)
�For each observation i, the MP
ij and MTij depend on the values of all coe¢ cients and regressors
j. Mean observation-by-observation marginal e¤ects are computed by evaluating the marginal
e¤ect of variable j for each observation and taking the mean over the sample of these marginal
e¤ects:20
�MKj = 1
N
PIi=1M
Kij ;8K = P; T
The variance-covariance matrix of the observation-by-observation marginal e¤ects is calculated
using the Delta Method, in order to account for the existing covariance between the estimated
marginal e¤ects of any two observations. Given the gradient is a linear map, the gradient of
20One common alternative consists in evaluating the e¤ect of x at the mean value, �x, of the estimation subsample.
However, �x may not be meaningful, especially if x is binary.
28
the mean marginal e¤ect is the mean of the observation-by-observation gradients. Thus, the
variance-covariance matrix is calculated as:
V ( �MKj ) = V
�1N
PIi=1M
Kij
�= @
@�̂0
�1N
PIi=1M
Kij
�V (�̂) @
@�̂
�1N
PIi=1M
Kij
�0=
=
�1N
PIi=1
@MKij
@�̂0
�V (�̂)
�1N
PIi=1
@MKij
@�̂
�0;8K = P; T
where V (�̂) is the chosen covariance matrix for �̂.
8.2 Figures
Frac
tion
Boys' PpschObs-by-Obs Marginal Effect
.005463 .036876
0
.1
.2
Obs
-by-
Obs
Mar
gina
l Effe
ct
Boys' PpschAge
Obs-by-Obs Marginal Effect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
0
.01
.02
.03
.04
Obs
-by-
Obs
Mar
gina
l Effe
ct
Boys' Ppsch(hh) Number of potential benefic
Obs-by-Obs Marginal Effect Mean Marginal Effect
1 2 3 4 5 6 7 8
0
.01
.02
.03
.04
Frac
tion
Boys' schObs-by-Obs Marginal Effect
.046302 .247168
0
.02
.04
.06
Obs
-by-
Obs
Mar
gina
l Effe
ct
Boys' schAge
Obs-by-Obs Marginal Effect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
.05
.1
.15
.2
.25
Obs
-by-
Obs
Mar
gina
l Effe
ct
Boys' sch(hh) Number of potential benefic
Obs-by-Obs Marginal Effect Mean Marginal Effect
1 2 3 4 5 6 7 8
.05
.1
.15
.2
.25
Figure A1: Obs-by-Obs Marginal E¤ects of Oportunidades on Boys�Schooling: Participation
(top graphs) and Hours (bottom graphs). Unconditional Distribution, Conditional on Age and
Conditional on Number of Eligible Children in the Household.
29
Frac
tion
Girls' PpschObs-by-Obs Marginal Effect
.003827 .064191
0
.1
.2
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' PpschAge
Obs-by-Obs Marginal Effect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
0
.02
.04
.06
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' Ppsch(hh) Number of potential benefic
Obs-by-Obs Marginal Effect Mean Marginal Effect
1 2 3 4 5 6 7 8
0
.02
.04
.06
Frac
tion
Girls' schObs-by-Obs Marginal Effect
.075517 .442707
0
.02
.04
.06
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' schAge
Obs-by-Obs Marginal Effect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
.1
.2
.3
.4
.5
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' sch(hh) Number of potential benefic
Obs-by-Obs Marginal Effect Mean Marginal Effect
1 2 3 4 5 6 7 8
.1
.2
.3
.4
.5
Figure A2: Obs-by-Obs Marginal E¤ects of Oportunidades on Girls�Schooling: Participation
(top graphs) and Hours (bottom graphs). Unconditional Distribution, Conditional on Age and
Conditional on Number of Eligible Children in the Household.
30
Frac
tion
Girls' PpdomObs-by-Obs Marginal Effect
-.021169 -.009402
0
.1
.2
.3
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' PpdomAge
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
-.02
-.015
-.01
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' Ppdom(hh) Number of potential benefic
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
1 2 3 4 5 6 7 8
-.02
-.015
-.01
Frac
tion
Girls' domObs-by-Obs Marginal Effect
-.219454 -.045298
0
.02
.04
.06
.08
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' domAge
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
8 9 10 11 12 13 14 15 16 17
-.25
-.2
-.15
-.1
-.05
Obs
-by-
Obs
Mar
gina
l Effe
ct
Girls' dom(hh) Number of potential benefic
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
1 2 3 4 5 6 7 8
-.25
-.2
-.15
-.1
-.05
Figure A3: Obs-by-Obs Marginal E¤ects of Oportunidades on Girls�Domestic Work:
Participation (top graphs) and Hours (bottom graphs). Unconditional Distribution,
Conditional on Age and Conditional on Number of Eligible Children in the Household.
31
Frac
tion
Women's PpdomObs-by-Obs Marginal Effect
-.021143 -.000355
0
.02
.04
.06
.08
Obs
-by-
Obs
Mar
gina
l Effe
ct
Women's PpdomAge
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
18 25 35 45 55 65 75 85 95 98
-.02
-.015
-.01
-.005
0
Obs
-by-
Obs
Mar
gina
l Effe
ct
Women's Ppdom(hh) Number of potential benefic
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
1 2 3 4 5 6 7 8
-.02
-.015
-.01
-.005
0
Frac
tion
Women's domObs-by-Obs Marginal Effect
.009452 .057856
0
.05
.1
Obs
-by-
Obs
Mar
gina
l Effe
ct
Women's domAge
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
18 25 35 45 55 65 75 85 95 98
0
.02
.04
.06
Obs
-by-
Obs
Mar
gina
l Effe
ct
Women's dom(hh) Number of potential benefic
Obs-by-Obs Marginal Eff ect Mean Marginal Effect
1 2 3 4 5 6 7 8
0
.02
.04
.06
Figure A4: Obs-by-Obs Marginal E¤ects of Oportunidades on Women�s Domestic Work:
Participation (top graphs) and Hours (bottom graphs). Unconditional Distribution,
Conditional on Age and Conditional on Number of Eligible Children in the Household.
32
8.3 Tables
Table 1: Oportunidades Monthly Transfer Amounts at Baseline (October 1997)
Transfer Component Level Grade Boys GirlsEducational Stipend Primary School 3rd year 60 60
4th year 70 705th year 90 906th year 120 120
Junior High School 1st year 175 1852nd year 185 2053rd year 195 225
School Supplies Stipend Primary, 1rst payment 80 80Primary, 2nd payment 40 40Junior High School 150 150
Nutritional Stipend (per family)Transfer Cap (per family)
90550
Source: Oportunidades (www.oportunidades.gob.mx). Transfer amounts are adjusted for inflation every semester according to theConsumer Price Index published by the Bank of Mexico. Transfer Cap is the maximum transfer amount awarded for basic education(primary school and junior high) and nutrition.
Table 2: List of Activities in the May 1999 Time Use Module
Market Work 1. Working for a salary or wage2. Working in own business
Farm Work 3. Working the family land4. Taking care of animals
Domestic Work 5. Purchasing food or other products for the household6. Sewing and making clothes for household members7. Taking household members to school, clinic or work8. Cleaning the house9. Washing and ironing clothes for household members10. Preparing food11. Fetching water, firewood or throwing out trash12. Taking care of small children, elderly and sick13. Making household repairs
Schooling 14. Attending school15. Doing homework after school
Community Work 16. Community Work17. Voluntary work for neighbours or other relatives
Transportation 18. Transportation time to work, school, market, etc.Other 19. Other activities
33
Table 3: Test of Equality of Means in the Estimation Sample in October 1997 (Baseline)
N Mean SD N Mean SD t-statHead's CharacteristicsAge Household Head 5056 42.25 (14.331) 2911 42.80 (14.894) -1.078Female Head =1 5056 7.32 (0.260) 2911 8.55 (0.280) -1.597Indigenous Head =1 5047 39.45 (0.489) 2905 39.93 (0.490) -0.080Head's Education (Years)† 3516 4.07 (2.313) 1982 3.90 (2.260) 1.563Head has a Paid Occupation = 1 5054 85.42 (0.353) 2910 84.71 (0.360) 0.450Monthly Hours Head Works in Paid Occupation† 4294 171.83 (55.399) 2453 173.35 (51.794) -0.569
Spouse's CharacteristicsAge Spouse of Head 4501 36.76 (12.503) 2561 36.55 (12.528) 0.498Spouse's Education (Years)† 2912 4.14 (2.144) 1617 4.11 (2.191) 0.278Spouse has a Paid Occupation = 1 4963 11.93 (0.324) 2843 11.15 (0.315) 0.390Monthly Hours Spouse Works in Paid Occupation† 579 150.48 (65.274) 309 163.69 (67.189) -1.75+
Household CharacteristicsChildren 0 to 7 =1 5056 76.09 (0.427) 2911 77.02 (0.421) -0.606Children 8 to 17 =1 5056 70.63 (0.456) 2911 70.04 (0.458) 0.437Adult Men 18 to 54 =1 5056 87.28 (0.333) 2911 85.85 (0.349) 1.382Adult Female 18 to 54 =1 5056 91.53 (0.278) 2911 91.00 (0.286) 0.608Adults Older than 55 =1 5056 27.63 (0.447) 2911 27.96 (0.449) -0.246Household Size 5056 6.03 (2.433) 2911 6.00 (2.418) 0.367Children Eligible for Educational Grant =1 5056 65.17 (0.476) 2911 64.14 (0.480) 0.673Number of Children Eligible for Educational Grant† 3295 2.13 (1.100) 1867 2.11 (1.074) 0.606Home Ownership =1 5052 93.84 (0.240) 2911 92.44 (0.264) 1.406Dirt Floor =1 5043 73.19 (0.443) 2902 74.29 (0.437) -0.377Electricity =1 5052 62.55 (0.484) 2910 59.86 (0.490) 0.545Water in the House =1 5047 4.93 (0.217) 2899 3.41 (0.182) 1.605Bathroom =1 5046 54.36 (0.498) 2904 52.07 (0.500) 0.599Land Ownership =1 5043 59.94 (0.490) 2910 58.04 (0.494) 0.601Farm Animals Ownership =1 5050 34.89 (0.477) 2908 32.81 (0.470) 0.794
Community CharacteristicsPre-school =1 5056 91.42 (0.280) 2911 89.08 (0.312) 0.728Satellite TV Secondary School (Telesecundaria) =1 5056 23.60 (0.425) 2911 25.46 (0.436) -0.310Mobile Health Unit =1 5056 72.31 (0.448) 2911 71.90 (0.450) 0.070Distance to Secondary School (Km) 3979 2.98 (1.636) 2308 3.46 (2.604) -1.189Monthly Community Agricultural Male Wage (1997 Pesos) 4911 568.79 (168.711) 2764 561.29 (177.962) 0.363Monthly Community Agricultural Female Wage (1997 Pesos) 3483 488.81 (169.676) 1933 477.99 (136.345) 0.501Monthly Community Agricultural Child Wage (1997 Pesos) 3468 386.50 (161.363) 1704 379.38 (135.025) 0.351
Treatment Group Control Group
Notes: +significant at 10%. †Conditional on being positive. T-stat of differences in means computed clustering SE at the community level. Mean of dichotomous variablesexpressed in percentages.
34
Table 4: Time Use of Boys and Girls Before the Intervention - Control Households in May 1999
Participation (%) Daily Hours† Participation (%) Daily Hours†
Schooling 8 to 11 78.99 5.86 80.49 5.7512 to 13 70.59 5.98 68.62 6.0814 to 15 54.18 6.11 45.01 6.3516 to 17 26.91 6.22 21.34 6.1312 to 17 53.16 6.06 48.26 6.17All ages 64.52 5.95 62.74 5.93
Market Work 8 to 11 0.83 6.48 0.74 3.1112 to 13 5.02 6.22 1.01 6.8714 to 15 15.27 7.65 4.89 7.7316 to 17 29.38 7.53 10.28 7.5612 to 17 15.13 7.40 4.89 7.56All ages 8.84 7.37 2.95 7.06
Farm Work 8 to 11 11.95 2.24 6.89 1.6712 to 13 17.82 2.84 9.40 1.3314 to 15 22.36 4.10 11.74 1.3916 to 17 25.93 4.55 11.57 1.1412 to 17 21.59 3.85 10.76 1.30All ages 17.35 3.36 9.02 1.43
Domestic Work 8 to 11 30.95 1.26 34.26 1.7812 to 13 33.56 1.49 51.68 2.3914 to 15 39.45 1.40 64.97 3.2916 to 17 40.25 1.65 65.04 4.2112 to 17 37.44 1.50 59.69 3.24All ages 34.59 1.40 48.27 2.78
Leisure 8 to 11 - 18.09 - 18.0712 to 13 - 17.86 - 17.8414 to 15 - 17.46 - 18.0516 to 17 - 17.43 - 18.7312 to 17 - 17.60 - 18.14All ages - 17.81 - 18.11
BOYS GIRLS
† Number of daily hours conditional on positive participation.
35
Table 5: Time Use of Adult Men and Women Before the Intervention - Control Households in May 1999
Participation (%) Daily Hours† Participation (%) Daily Hours†Market Work 18 to 24 54.96 7.66 9.27 6.91
25 to 34 66.59 7.66 5.77 6.4535 to 44 67.25 7.48 5.16 6.4245 to 54 63.64 7.63 4.43 6.83More than 55 41.81 7.32 3.99 6.62All ages 59.87 7.57 5.97 6.65
Farm Work 18 to 24 26.37 5.03 15.34 1.5025 to 34 30.91 5.24 21.05 1.2835 to 44 33.10 4.99 22.63 1.3245 to 54 36.96 5.06 26.58 1.50More than 55 40.47 4.95 19.93 1.39All ages 33.07 5.06 20.67 1.38
Domestic Work 18 to 24 37.78 1.66 84.66 5.5225 to 34 45.45 1.74 95.03 6.2735 to 44 42.25 1.90 95.78 5.8745 to 54 43.87 1.80 92.83 5.21More than 55 38.13 1.76 79.55 4.14All ages 41.73 1.78 90.36 5.61
Leisure 18 to 24 - 16.81 - 18.2625 to 34 - 15.66 - 17.4435 to 44 - 15.74 - 17.6745 to 54 - 15.73 - 18.14More than 55 - 17.46 - 19.86All ages - 16.23 - 18.13
† Number of daily hours conditional on positive participation.
MEN WOMEN
36
Tab
le 6
: Opo
rtun
idad
es Im
pact
on
Boy
s' a
nd G
irls
' Tim
e A
lloca
tion
- Poo
led
Impa
ct a
nd b
y A
ge
Leisu
reH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
I. B
OY
S(N
= 7
691)
(OL
S)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
Mod
el IA
:Al
l Age
s0.
033
0.02
90.
166*
*-0
.014
-0.1
04+
-0.0
11-0
.150
-0.0
04-0
.035
-0.0
09-0
.020
(0.1
51)
(0.0
23)
(0.0
58)
(0.0
21)
(0.0
62)
(0.0
08)
(9.6
30)
(0.0
14)
(0.7
89)
(0.0
21)
(0.0
50)
Mod
el IB
:8
to 1
10.
048
0.01
00.
071
-0.0
12-0
.059
-0.0
03-0
.089
0.00
50.
025
-0.0
15-0
.030
(0.1
82)
(0.0
32)
(0.0
87)
(0.0
25)
(0.0
79)
(0.0
21)
(9.0
59)
(0.0
18)
(0.8
10)
(0.0
23)
(0.0
54)
12 to
17
0.02
40.
043+
0.26
3**
-0.0
18-0
.145
*-0
.012
-0.1
74-0
.010
-0.0
74-0
.005
-0.0
13(0
.167
)(0
.021
)(0
.081
)(0
.024
)(0
.069
)(0
.009
)(9
.147
)(0
.015
)(0
.774
)(0
.024
)(0
.054
)
II. G
IRLS
(N =
732
1)
Mod
el II
A:Al
l Age
s-0
.211
0.04
9*0.
290*
*-0
.015
-0.0
84+
-0.0
03-0
.097
0.00
40.
043
-0.0
19-0
.097
*(0
.147
)(0
.023
)(0
.058
)(0
.021
)(0
.045
)(0
.005
)(2
8.38
5)(0
.011
)(1
.422
)(0
.021
)(0
.044
)
Mod
el II
B:8
to 1
1-0
.001
-0.0
14-0
.001
0.00
10.
010
-0.0
000.
030
0.02
10.
085
-0.0
05-0
.028
(0.1
84)
(0.0
34)
(0.0
86)
(0.0
25)
(0.0
66)
(0.0
09)
(27.
771)
(0.0
15)
(1.4
19)
(0.0
25)
(0.0
61)
12 to
17
-0.3
87*
0.08
7**
0.58
2**
-0.0
28-0
.145
**-0
.004
-0.1
21-0
.007
0.01
4-0
.030
-0.1
39**
(0.1
63)
(0.0
20)
(0.0
86)
(0.0
22)
(0.0
54)
(0.0
05)
(27.
008)
(0.0
13)
(1.3
52)
(0.0
21)
(0.0
53)
Not
es:+
Sign
ifica
ntat
the
10pe
rcen
tlev
el.
*Sig
nific
anta
tthe
5pe
rcen
tlev
el.*
*Sig
nific
anta
tthe
1pe
rcen
tlev
el.S
Ecl
uste
red
atth
eco
mm
unity
leve
lin
para
nthe
ses.
Forp
robi
tand
tobi
tmod
els,
mea
nof
the
obse
rvat
ion-
by-o
bser
vatio
nm
argi
nal
effe
cts
repo
rted.
All
regr
essio
nsin
clud
eth
efo
llow
ing
cont
rols:
age,
age
squa
red,
educ
atio
ndu
mm
ies,
rank
amon
gsts
iblin
gs,m
othe
r'set
hnic
ity,s
exof
the
hous
ehol
d's
head
,bas
elin
eho
useh
old
dem
ogra
phic
san
dho
useh
old
size,
base
line
dwel
ling
char
acte
ristic
s (b
athr
oom
, ele
ctric
ity, l
and
owne
rshi
p an
d dr
aft a
nim
al o
wne
rshi
p), l
og m
ale's
agr
icul
tura
l w
age
(in 1
997
pric
es),
dist
ance
to th
e cl
oses
t mun
icip
ality
and
stat
e du
mm
ies.
Age
and
age
squa
red
are
not i
nclu
ded
in M
odel
s B.
Scho
olA
ll W
ork
Mar
ket W
ork
Farm
Wor
kD
omes
tic W
ork
37
Tab
le 7
: Opo
rtun
idad
es Im
pact
on
Men
s' a
nd W
omen
s' T
ime
Allo
catio
n - P
oole
d Im
pact
and
by
Age
Leisu
reH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
Part
icip
atio
nH
ours
I. M
EN (N
= 9
519)
(OL
S)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)M
odel
IA:
All A
ges
0.20
9+-0
.007
-0.1
41*
-0.0
12-0
.060
-0.0
05-0
.025
-0.0
04-0
.012
(0.1
27)
(0.0
11)
(0.0
72)
(0.0
18)
(0.0
66)
(0.0
19)
(0.1
65)
(0.0
20)
(0.0
29)
Mod
el IB
:18
to 2
4-0
.133
0.02
10.
058
-0.0
000.
003
-0.0
040.
000
0.01
60.
037
(0.2
42)
(0.0
15)
(0.1
64)
(0.0
27)
(0.1
51)
(0.0
29)
(0.2
01)
(0.0
34)
(0.0
51)
25 to
34
0.40
7*-0
.046
*-0
.247
+0.
005
0.06
9-0
.037
-0.2
41-0
.020
-0.0
17(0
.169
)(0
.022
)(0
.138
)(0
.025
)(0
.128
)(0
.024
)(0
.179
)(0
.028
)(0
.043
)35
to 4
40.
255
-0.0
11-0
.174
-0.0
36-0
.162
0.02
00.
116
-0.0
15-0
.062
(0.1
88)
(0.0
19)
(0.1
41)
(0.0
26)
(0.1
27)
(0.0
26)
(0.1
93)
(0.0
24)
(0.0
43)
45 to
54
0.23
70.
009
-0.1
70-0
.041
-0.3
14+
0.04
30.
238
0.00
1-0
.019
(0.2
28)
(0.0
20)
(0.1
82)
(0.0
31)
(0.1
61)
(0.0
32)
(0.2
10)
(0.0
33)
(0.0
54)
Mor
e th
an 5
50.
321
-0.0
13-0
.215
-0.0
03-0
.003
-0.0
37-0
.165
0.00
50.
009
(0.2
40)
(0.0
18)
(0.1
72)
(0.0
24)
(0.1
69)
(0.0
25)
(0.1
94)
(0.0
32)
(0.0
53)
II. W
OM
EN
(N =
103
70)
Mod
el II
A:Al
l Age
s-0
.159
-0.0
080.
052
0.00
90.
104
0.01
00.
012
-0.0
070.
046
(0.1
08)
(0.0
07)
(0.0
59)
(0.0
07)
(6.6
93)
(0.0
17)
(0.2
14)
(0.0
07)
(0.0
53)
Mod
el II
B:18
to 2
40.
035
-0.0
14-0
.192
-0.0
07-0
.098
0.00
10.
006
-0.0
14-0
.158
(0.1
91)
(0.0
12)
(0.1
26)
(0.0
09)
(6.2
78)
(0.0
27)
(0.2
13)
(0.0
13)
(0.1
14)
25 to
34
-0.2
38-0
.007
0.16
90.
005
0.05
00.
016
0.02
1-0
.001
0.20
1*(0
.168
)(0
.013
)(0
.109
)(0
.010
)(6
.370
)(0
.022
)(0
.213
)(0
.014
)(0
.099
)35
to 4
4-0
.230
-0.0
030.
134
0.01
90.
214
0.01
60.
028
-0.0
120.
087
(0.1
77)
(0.0
14)
(0.1
24)
(0.0
12)
(6.4
77)
(0.0
24)
(0.2
14)
(0.0
17)
(0.1
12)
45 to
54
-0.2
230.
003
0.12
20.
041+
0.48
5-0
.003
-0.0
220.
011
0.04
4(0
.228
)(0
.015
)(0
.167
)(0
.024
)(6
.501
)(0
.026
)(0
.213
)(0
.015
)(0
.150
)M
ore
than
55
-0.1
65-0
.009
0.02
40.
022
0.26
50.
013
0.00
9-0
.011
-0.0
05(0
.195
)(0
.011
)(0
.154
)(0
.019
)(6
.464
)(0
.025
)(0
.214
)(0
.013
)(0
.139
)N
otes
:+Si
gnifi
cant
atth
e10
perc
entl
evel
.*Si
gnifi
cant
atth
e5
perc
entl
evel
.SE
clus
tere
dat
the
com
mun
ityle
veli
npa
rant
hese
s.Fo
rpro
bita
ndto
bitm
odel
s,m
ean
ofth
eob
serv
atio
n-by
-obs
erva
tion
mar
gina
leffe
ctsr
epor
ted.
All
regr
essi
ons
incl
ude
the
follo
win
gco
ntro
ls:
age,
age
squa
red,
educ
atio
ndu
mm
ies,
mar
ital
stat
us,
head
ofth
eho
useh
old
stat
us,
ethn
icity
,ba
selin
eho
useh
old
dem
ogra
phic
san
dho
useh
old
size
,ba
selin
edw
ellin
gch
arac
teris
tics
(bat
hroo
m,e
lect
ricity
,lan
dow
ners
hip
and
draf
tani
mal
owne
rshi
p),l
ogm
ale
agric
ultu
ral
wag
e(in
1997
pric
es),
dist
ance
toth
ecl
oses
tmun
icip
ality
and
stat
edu
mm
ies.
Age
and
age
squa
red
are
noti
nclu
ded
inM
odel
s B.
All
Wor
kM
arke
t Wor
kFa
rm W
ork
Dom
estic
Wor
k
38
Tab
le 8
: Ow
n-Su
bstit
utio
n, C
ross
-sub
stitu
tion
and
Inco
me
Eff
ects
on
Tee
nage
rs' (
Age
s 12
to 1
7) T
ime
Allo
catio
n
Lei
sure
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsI.
BO
YS
(N =
428
7)(O
LS)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)Tr
eatm
ent (
T) =
10.
260
-0.0
67-0
.268
-0.0
120.
086
0.06
30.
676
0.02
4-0
.026
-0.0
63-0
.097
(Inc
ome
Effe
ct)
(0.5
98)
(0.0
77)
(0.3
43)
(0.0
66)
(0.2
71)
(0.0
40)
(5.6
50)
(0.0
50)
(0.7
91)
(0.0
65)
(0.1
11)
T *
Ben
efic
iary
= 1
-0.3
440.
119*
*0.
608*
*0.
033
-0.1
17-0
.059
*-0
.583
-0.0
16-0
.094
0.12
7**
0.15
7(O
wn-
Subs
t Eff
ect)
(0.3
54)
(0.0
45)
(0.2
35)
(0.0
41)
(0.1
94)
(0.0
30)
(3.0
96)
(0.0
34)
(0.7
40)
(0.0
40)
(0.1
07)
T *
Nb
Oth
er B
enef
icia
ries
0.25
1-0
.085
+-0
.558
*0.
030
0.17
4-0
.016
-0.1
850.
027
0.24
9-0
.047
-0.0
20(C
ross
-Sub
st E
ffec
t)(0
.401
)(0
.045
)(0
.251
)(0
.047
)(0
.209
)(0
.034
)(0
.322
)(0
.035
)(0
.181
)(0
.050
)(0
.077
)
II. G
IRL
S(N
= 3
992)
Trea
tmen
t (T)
=1
0.41
4-0
.015
0.05
7-0
.011
-0.2
000.
016
0.25
9-0
.148
**-0
.547
-0.0
10-0
.055
(Inc
ome
Effe
ct)
(0.4
99)
(0.0
74)
(0.3
51)
(0.0
68)
(0.2
49)
(0.0
24)
(23.
256)
(0.0
55)
(0.9
04)
(0.0
69)
(0.2
09)
T *
Ben
efic
iary
= 1
-0.7
30*
0.02
20.
013
0.04
60.
289+
0.00
90.
238
0.02
90.
099
0.02
60.
133
(Ow
n-Su
bst E
ffec
t)(0
.364
)(0
.043
)(0
.246
)(0
.046
)(0
.169
)(0
.015
)(2
2.75
1)(0
.027
)(1
.844
)(0
.046
)(0
.144
)
T *
Nb
Oth
er B
enef
icia
ries
0.63
8-0
.025
-0.2
03-0
.037
-0.1
67-0
.038
-0.7
23+
0.05
20.
223+
-0.0
18-0
.067
(Cro
ss-S
ubst
Eff
ect)
(0.4
05)
(0.0
49)
(0.2
66)
(0.0
53)
(0.1
95)
(0.0
23)
(0.4
23)
(0.0
30)
(0.1
22)
(0.0
55)
(0.1
65)
Not
es:+
Sign
ifica
ntat
the
10pe
rcen
tlev
el.
*Sig
nific
anta
tthe
5pe
rcen
tlev
el.*
*Sig
nific
anta
tthe
1pe
rcen
tlev
el.S
Ecl
uste
red
atth
eco
mm
unity
leve
lin
para
nthe
ses.
For
prob
itan
dto
bitm
odel
s,m
ean
obse
rvat
ion-
by-o
bser
vatio
nm
argi
nal
effe
cts r
epor
ted.
All
regr
essi
ons i
nclu
de th
e fo
llow
ing
cont
rols
: age
, age
squa
red,
edu
catio
n du
mm
ies,
rank
am
ongs
t sib
lings
, mot
her's
eth
nici
ty, s
ex o
f the
hou
seho
ld's
head
, bas
elin
e ho
useh
old
dem
ogra
phic
s and
hou
seho
ld si
ze, b
asel
ine
dwel
ling
char
acte
ristic
s(ba
thro
om,e
lect
ricity
,lan
dow
ners
hip
and
draf
tani
mal
owne
rshi
p),s
exan
dgr
ade
ofth
efir
stfo
urbe
nefic
iary
child
ren
inth
eho
useh
old,
log
ofth
eto
talp
oten
tialh
ouse
hold
trans
fera
mou
nt(in
1997
pric
es),
log
mal
e'sag
ricul
tura
lw
age
(in 1
997
pric
es),
dist
ance
to th
e cl
oses
t mun
icip
ality
and
stat
e du
mm
ies.Sc
hool
All
Wor
kM
arke
t Wor
kFa
rm W
ork
Dom
estic
Wor
k
39
Tab
le 9
: Ow
n-Su
bstit
utio
n, C
ross
-sub
stitu
tion
and
Inco
me
Eff
ects
on
Adu
lts' T
ime
Allo
catio
n
Lei
sure
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsPa
rtic
ipat
ion
Hou
rsI.
ME
N (N
= 9
519)
(OL
S)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)(P
robi
t)(T
obit)
(Pro
bit)
(Tob
it)
Trea
tmen
t (T)
=1
0.58
3*-0
.004
-0.2
620.
026
0.16
4-0
.047
-0.3
19-0
.053
-0.0
71(I
ncom
e Ef
fect
)(0
.280
)(0
.021
)(0
.190
)(0
.033
)(0
.173
)(0
.035
)(0
.186
)+(0
.036
)(0
.056
)
T *
Nb
Ben
efic
iarie
s-0
.119
-0.0
17-0
.035
-0.0
04-0
.051
0.03
70.
171
-0.0
14-0
.055
(Cro
ss-S
ubst
Eff
ect)
(0.2
04)
(0.0
17)
(0.1
58)
(0.0
23)
(0.1
46)
(0.0
27)
(0.1
13)
(0.0
25)
(0.0
43)
II. W
OM
EN
(N =
103
70)
Trea
tmen
t (T)
=1
0.44
3*-0
.027
+-0
.328
*0.
017
0.24
0-0
.051
+-0
.164
-0.0
26-0
.185
(Inc
ome
Effe
ct)
(0.2
23)
(0.0
14)
(0.1
54)
(0.0
14)
(7.2
06)
(0.0
30)
(0.1
62)
(0.0
16)
(0.1
39)
T *
Nb
Ben
efic
iarie
s-0
.245
0.01
00.
254+
-0.0
29*
-0.4
21**
0.04
3*0.
130*
*0.
011
0.26
2*(C
ross
-Sub
st E
ffec
t)(0
.194
)(0
.014
)(0
.131
)(0
.012
)(0
.165
)(0
.019
)(0
.040
)(0
.015
)(0
.118
)N
otes
:+Si
gnifi
cant
atth
e10
perc
entl
evel
.*S
igni
fican
tatt
he5
perc
entl
evel
.**S
igni
fican
tatt
he1
perc
entl
evel
.SE
clus
tere
dat
the
com
mun
ityle
veli
npa
rant
hese
s.Fo
rpro
bita
ndto
bitm
odel
s,m
ean
ofth
eob
serv
atio
n-by
-obs
erva
tion
mar
gina
leff
ects
repo
rted.
All
regr
essi
ons
incl
ude
the
follo
win
gco
ntro
ls:a
ge,a
gesq
uare
d,ed
ucat
ion
dum
mie
s,m
arita
lsta
tus,
hous
ehol
dhe
adst
atus
,eth
nici
ty,b
asel
ine
hous
ehol
dde
mog
raph
ics
and
hous
ehol
dsi
ze,b
asel
ine
dwel
ling
char
acte
ristic
s(b
athr
oom
,ele
ctric
ity,l
and
owne
rshi
pan
ddr
afta
nim
alow
ners
hip)
,sex
and
grad
eof
the
first
four
bene
ficia
rych
ildre
nin
the
hous
ehol
d,lo
gof
the
tota
lpot
entia
lhou
seho
ldtra
nsfe
ram
ount
(in19
97pr
ices
),lo
gm
ale
agric
ultu
ral
wag
e(in
1997
pric
es),
dist
ance
toth
ecl
oses
tmun
icip
ality
and
stat
edu
mm
ies.
Age
and
age
squa
red
are
noti
nclu
ded
inM
odel
sB
.
All
Wor
kM
arke
t Wor
kFa
rm W
ork
Dom
estic
Wor
k
40
Recommended