Does Female Schooling Reduce Fertility?
The Case of Universal Primary Education in Nigeria
Una Okonkwo Osili*
Department of Economics
Indiana University-Purdue University at Indianapolis
[email protected]
Bridget Terry Long
Harvard Graduate School of Education and NBER
[email protected]
March 2004
ABSTRACT
Empirical studies based generally point to a robust, negative association between female
education and fertility (see Schultz, 1994; Schultz, 1997; Ainsworth et al, 1996). Citing this evidence,
policymakers have advocated educating girls and young women as a means to achieve lower rates of
population growth, and thus for sustained economic and social welfare in developing countries. This
paper uses an unusual policy experiment from Nigeria to investigate the impact of female education on
early fertility decisions. Our results suggest that changes in schooling costs and other policy variables
associated with universal education policies can have a substantial impact on female education and early
fertility. We provide evidence that female education has a strong, negative association with early fertility
even after accounting for the possible endogeneity of the schooling decision. We also consider alternative
explanations for the rapid advances in female schooling and demographic outcomes including changes in
social norms and attitudes towards female education and economic opportunities for women.
*
Corresponding author: Una Okonkwo Osili, Department of Economics, IUPUI, 425 University Boulevard,
Indianapolis, IN 46202, U.S.A. Phone: (317) 274-4755, Fax: (317) 274-0097, Email: [email protected].
The authors would like to thank Rina and Paul Okonkwo for assistance with data collection. We have also benefited
greatly from discussions with Ifeanyi Osili, Izevbuwa Osayimwese, Omozuwa Osayimwese, Anna Paulson, Peter
Rangazas, and Anne Royalty. Yuling Han and Xiaojun Feng provided valuable research assistance.
1
I.
INTRODUCTION
Reducing population growth remains a high priority for many developing countries. Among the
policy alternatives, female education is considered a highly effective means of lowering fertility and
population growth (United Nations, 1995). Many empirical studies point to a robust, negative association
between female education and fertility (Schultz, 1997; Ainsworth et al, 1996), and this negative
relationship has been described as “one of the most clear-cut correlations” in the social science literature
(McGreevy and Birdsall, 1974; Cochrane, 1979). Moreover, there is growing consensus that investments
in the education of young girls and women yield other private and social returns including improved child
health and nutrition outcomes (Schultz, 2002; Thomas, 1991).
There are several reasons rooted in economic theory on why education may affect fertility. First,
female schooling may increase the opportunity cost of childbearing and rearing among educated women
(Becker, 1981; Schultz, 1981). Second, education may lower fertility through improvements in child
health and reduced rates of child mortality (Schultz, 1994; Lam and Duryea, 1999). Finally, female
schooling may affect fertility through the more effective use of contraception (Rosenzweig and Schultz,
1985; 1989) or by increasing female autonomy and bargaining power in fertility decisions (Mason, 1986).
However, a survey of the existing literature suggests a need for caution in interpreting the observed
relationship between female education and fertility as causal (Bledsoe et al, 1999). In particular, a
negative association may arise due to omitted variables such as individual ability or household and
community resources, which affect both schooling and fertility decisions. In addition, schooling
opportunities are often not randomly placed in communities (Duflo, 2001; Pitt, Rosenzweig and Gibbons,
1993). Furthermore, if fertility choices lead to interruptions in schooling, then fertility may be an
endogenous variable within the context of schooling decision (Angrist and Evans, 1996).
The ideal method to address the issue of causality is to use an exogenous source of variation in
schooling that is not related to fertility outcomes. This paper uses this technique by focusing on the
Universal Primary Education (UPE) program in Nigeria. Introduced in September 1976, the UPE was a
large-scale, nationwide policy designed to increase educational attainment.
Funded by the federal
government, it provided tuition-free primary education throughout the country and marked a significant
change in the educational opportunities available to men and women. Before 1976, regional governments
controlled education, and only two regions in the country had provided tuition-free primary schooling.
However, following the introduction of UPE, primary school enrollments increased from 4.4 million
students in 1974 to 14.5 million primary school students in Nigeria by 1982 (Federal Office of Statistics,
various years).
Using this policy as an exogenous source of variation in primary schooling rates, this paper
examines whether female schooling causes reductions in fertility. First, the paper examines how the UPE
affected female education levels by exploiting differences in the extent of the treatment. Second, the
paper links these changes in education to fertility rates. The UPE program was discontinued in 1981.
2
Using the 1999 Nigerian Demographic Health Survey, the paper employs several empirical strategies
including differences-in-differences and instrumental variables to estimate the relationship between
schooling and fertility. This paper joins a growing number of studies that identify the impact of education
policies and programs in developing countries.1 Moreover, the paper addresses the question of whether
investments specifically in primary education impact fertility, as there is little consensus about what level
of education should be expanded to affect the birth rate.
The results suggest that changes in schooling costs and other policy variables associated with a
universal education policy can have a substantial impact on female education and early fertility. We
provide evidence that female education has a strong, negative association with early fertility even after
accounting for the possible endogeneity of the schooling decision.
We also consider alternative
explanations for the rapid advances in female schooling and demographic outcomes including changes in
social norms and attitudes towards female education and economic opportunities for women.2
This paper is organized as follows: Section II provides background on the UPE program in
Nigeria and a descriptive overview of its impact on educational outcomes. Section III presents data
sources and the empirical framework. Section IV examines the impact of UPE on educational outcomes,
the treatment in this study. Section V analyzes the impact on fertility, and Section VI concludes.
II.
BACKGROUND: EDUCATIONAL POLICY IN NIGERIA AND THE UPE
Nigeria is an intriguing environment to study the impact of female education on fertility. Some
of the earliest research on the economic determinants of fertility is based on survey evidence and field
research from Nigeria and other parts of Sub Saharan Africa (Van De Walle, 1965; Caldwell, 1968; and
Caldwell, 1977). In addition, population growth in sub-Saharan Africa remains among the highest in the
developing world, at 3.2 percent a year.
One important feature of the Nigerian environment is the
absence of a sustained national family planning program, or government efforts to promote sterilization or
birth control pills (Caldwell et al, 1992).
Since the 1960s, UNESCO and other international agencies have called for universal education.
Nigeria’s UPE initiative has been characterized as “one of the most ambitious education projects in
African history” due to the magnitude of the program in terms of government resources and the number
of children who benefited (Bray, 1981:1). While the female primary enrollment rate was only 40 percent
in 1974, by the end of the program in 1981, it had risen to 104.7 (World Bank, 2002 African Development
1
For example, Case and Deaton (1999) use data from South Africa to study the impact of increased resources on
schooling outcomes. Duflo (2001) relies on an individual’s date of birth and region of schooling to identify the
impact of a large-scale school construction program in Indonesia on educational attainment and wages. Angrist et
al (2002) study the impact of school vouchers in Columbia.
2
Unlike other related work, however, changes in contraceptive use and legal statues, such as the legalization of
abortion and anti-discrimination laws, are less likely to be relevant in our environment.
3
Indicators).3 These achievements are particularly impressive given that Sub-Saharan and northern Africa
have some of the lowest levels of female educational attainment, and many young girls and women have
little or no exposure to formal schooling.
Education in Nigeria before the UPE program
Nigeria became independent from the United Kingdom in 1960. Prior to independence, it was
divided into distinct administrative units governed as semi-autonomous regions under colonial rule. 4
The first universal education scheme in Nigeria was launched in the Western region in January
1955.5 In that year, tuition fees were abolished for all levels of primary schooling. According to
Awokoya, then Education Minister for Western region, education was seen as “imperative” and “urgent.”6
However, these early programs were generally limited in its overall impact. Appendix Table 1
summarizes the important dates.
Following the model of the Western Region, the Eastern Region
launched a universal primary school program in January 1957 that provided education without fees. Due
to financial difficulties, the program was scaled back to provide free schooling only in the first two years
of primary schooling. Lagos, the capital of Nigeria after independence, also introduced a universal
primary education program in January 1957 and had achieved near-universal enrollments by the late
1960s.
In comparison to the rest of the country, the Northern region of Nigeria lagged in educational
access. 7Although there had been some limited efforts to increase adult literacy during the 1950s, access
to formal schooling remained low throughout much of the North (Bray, 1981). By 1970, less than 10
percent of the primary school age population was enrolled in school in several Northern states (UNESCO,
1971). Therefore, an important goal of the nationwide UPE program was to reduce education imbalances
between the Western region and the rest of Nigeria.
An Overview of the Impact of UPE
3
An enrollment rate is calculated as the percentage of school-age population who are enrolled in primary school in a
given year. It can be over 100.0 due to the fact that older or younger students may be in the enrollment count.
4
Currently, there are over 30 states in Nigeria. The number of states has changed over time to improve equity in the
revenue-sharing system at the federal level. Federal government revenue allocations to states and local governments
are governed by a formula based on population, need, and, to a lesser extent, derivation
5
The Western region was a forerunner in education. In 1843, the Methodist mission established the first primary
school in the Western region partially due its proximity to the Atlantic coast, which enabled the early settlement of
the Christian missionaries in this region (Fafunwa, 1974).
6
The priority accorded to education was reflected in the Western Region’s budget, with primary education
consuming nearly 40 percent of the government’s recurrent budget (Nwachukwu, 1985). Although schooling was
not compulsory, the student population increased significantly in the Western Region. While there was no explicit
targeting of girls, female enrollments rose during this period. Girls made up less than 20 percent of primary school
students in 1954, but their proportion increased to nearly 40 percent by 1966 ( Nwachukwu, 1985). Awokoya
(1952) “Proposals for an Educational Policy.” Western Regional Gazette, vol. 13, no. 10.
7
The distance from the Atlantic coast made it more difficult for Christian missionaries to establish schools in the
Northern states. In addition, British indirect rule and colonial policy of non-interference in Islamic religious
practices meant that primary school enrollment rates, particularly for women were extremely low in the North.
4
Fueled by revenues from the oil boom of the time, the UPE Program was formally announced on
October 1, 1974. It provided tuition-free primary education funded at the federal level and universally
available throughout the country starting in September 1976. To detail the scope and impact of the UPE
program, Figures 1A to 1D summarize the major trends in the number of students, schools, and teacher
training institutions before and after 1976 with information from various years of the Annual Abstract of
Statistics and Social Statistics in Nigeria. As shown in all the figures, the introduction of the nationwide
UPE had a major impact on enrollment and the number of schools and teacher training institutions.
Figure 1A displays how the number of primary school students increased dramatically after the
introduction of the UPE. During the first year of UPE, the number of students enrolled in grade one
increased by 82 percent to 3 million students. This exceeded predictions that only 2.3 million would
enroll (Federal Republic of Nigeria, 1978-79). These changes were not distributed evenly across the
country. Since primary education was already more prevalent in the Western region, the introduction of
UPE had a larger impact in the Northern and Southeastern regions. As shown in Figure 1A, growth in
student enrollment after 1976 tended to be higher outside the Western region.
To enable the expansion of primary education, the government needed to construct many new
classrooms. According to the Nigerian Third National Development Plan (1975-1980), the country
planned for the provision for 150,995 new classrooms at the primary school level. Figure 1B shows how
this substantial grown translated into the number of schools in Nigeria. To meet the increased demand for
teaching resources, plans were also announced to build new teacher training institutions and expand
existing teacher training institutions. According to Nwachukwu (1985), the UPE program required about
80,000 new teachers by 1976, and the government earmarked $320 million for the construction of teacher
training facilities. In addition to the new primary school classrooms, the country planned for 6,699 new
classrooms in teacher training institutions (Third National Development Plan, 1975-1980).8
The Nigerian UPE ended in September 1981 when the federal government handed over the
financing of primary schools to states and regional governments. With reduced funding for primary
schools, the reintroduction of school fees, and declining oil revenues in the 1980s, primary enrollments
fell or stagnated in some states after the program ended (Francis, 1998). As shown in Figure 1A, primary
school enrollments began to fall in the non-Western regions.
8
The construction projections for primary and teacher training institute classrooms were based on the 1963 census
of primary school age children (Third National Development Plan, 1975-1980). Therefore, much of the variation in
the planned number of classrooms is explained by state population and enrollment rates. Analysis of the projected
number of primary school classrooms found a positive estimate associated with the 1970 state population
(coefficient 0.004, standard error 0.0002) and negative estimate associated with the primary school enrollment rate
(coefficient -15.23, standard error 1.943). Likewise, for the number of planned teacher training institutions, the
analysis found a positive relationship with the 1970 state population (coefficient 0.012, standard error 0.0027) and a
negative relationship with the enrollment rate (coefficient -0.73, standard error 0.226). The adjusted R-squared was
0.96 for planned classrooms in primary schools and 0.68 for planned classrooms in teacher training institutions.
5
III.
DATA AND EMPIRICAL FRAMEWORK
The Data: The Nigerian Demographic Health Survey
To study the impact of the UPE program, we use the 1999 Nigerian Demographic Health Survey
(NDHS). This dataset is a nationally-representative sample of women in Nigeria aged 15-49 with rich
information on socioeconomic and demographic variables for 8,199 women. The NDHS survey is ideal
for this study because by design it gives reliable information on levels and trends in education, fertility,
and family planning practices for a large number of women. Table 1 provides summary statistics for the
data. To study the impact of the UPE on education, and early fertility, we use information on from the
NDHS on a respondent’s region of birth to link individuals to educational polices and institutions.
While the NDHS surveys have many important advantages, the data do have several
shortcomings. First, there is no good information on wages or income at the household or individual
level.
Secondly, there is only have limited information on the migration status of an individual.
Specifically, we know how long an individual has lived in area and whether they have always lived in a
given location, but we do not know where they lived prior to migration. Because we cannot identify the
state where an individual was educated if she moved from her place of residence after age six, we restrict
our analysis to the sample of nonmovers. The right section of Table 1 details this sample of movers,
which are not included in the analysis. Finally, because the cohort born after the UPE (1976-1981) was
just reaching their middle twenties by 1999, we do not use them as a control group in estimating the
impact on early fertility.
Empirical Strategy #1: Differences-in-Differences (DD)
Estimating the Impact of UPE
We use the difference-in-differences technique to evaluate the impact of the UPE program on
years of completed schooling. By noting that the UPE occurred between 1976 and 1981 and affected
certain regions more than others, we examine whether the introduction of universal primary education
caused discontinuities in the variables of interest. This approach has been used in recent studies including
Duflo (2001).
The first difference we use in the analysis is age at the time of the policy. Given the timing of
UPE, it mainly affected individuals born between 1970 and 1975. We assume that most individuals start
school at age 6, and to capture the length of exposure to the UPE program, we construct variables based
on an individual’s year of birth. However, one concern with this analysis is the possible prevalence of
underage and overage enrollments. According to a recent Multiple Indicator Cluster Survey (MICS)
conducted by UNICEF in 1999, underage enrollments, overage enrollments, and grade repetitions are
likely to be of importance in Nigeria. For example, of the sample of 977 female students enrolled in first
grade (primary one), 19 percent were aged five, 31 percent were six years of age, and 21 percent were
seven years of age. About 4 percent of the sample was 4 years old and about 15 percent were above 7
6
years of age. The high prevalence of underage and overage enrollments in first grade suggests that the
program may have affected individuals other than the main target group. Therefore, the cohort we use for
the control group were age 13 to 18 when the UPE was initiated (born between 1958 and 1963) and were
therefore less likely to be susceptible to overage enrollment.
The second difference we use is by region. The introduction of UPE to Nigeria had varying
effects on the states and regions within the country. As mentioned earlier, the Western region had the
advantage of being the first region where Christian missionaries established schools in Nigeria.
Furthermore, the Western region (including Lagos) had adopted and maintained a policy of Universal
primary education since the 1950s. Thus, the national UPE policy that eliminated school fees at the
primary school level in 1976 should have had a relatively limited impact on primary school enrollments in
Western states, while having large effects outside the Western region. While schooling levels increased
in both high-intensity and low-intensity areas, schooling increased faster in the high intensity areas. In
order to account for any general trends that affected all states, individuals in the Western region that did
not experience large changes in enrollments and schooling inputs are used as a control group.
Our difference-in-differences estimation can be described as follows:
(1)
Sijk = αi + BXijk + δ(High intensityk * Youngjk) + α1High Intensityk + α2 Youngjk + εijk
where i indexes individuals, j indexes cohorts, and k indexes states. Sijk is the years of schooling
completed by individual i in state k. Xijk includes controls for linear, quadratic and cubic age terms, an
urban dummy, age of the household head, religion dummies, and living arrangements (living with
parent=1). The parameter δ is the reduced-form estimate of universal primary education – it measures
whether individuals in states outside the Western region that experienced large changes (or high intensity
regions) due to the UPE program also experienced rapid growth in schooling differently from individuals
in states in the Western region that did not experience much change due to the UPE program. The
variable “High intensity” is a dummy variable that is equal to one if the person was educated in a state
that experienced a lot of change due to the UPE program (i.e. the non-Western states). “Young” is a
dummy variable captures whether an individual was born between 1970 and 1975; or between the ages of
one and six in 1976 when the UPE program was implemented, otherwise these variables are equal to zero.
Empirical Strategy #2: Differences-in-Differences in-Differences (DDD)
Likewise, the DD estimation strategy can be used to exploit three-way differences. For example,
examining changes in school resources:
(2)
Sijk = α0 + BXijk+ θ (Schooling Inputs* Young*High Intensity) + δ (Schooling Inputs* Young)
7
+ α1 Schooling inputs + α2 Young + α3 High Intensityjk +εijk
We expect that the change in schooling outcomes for women should be larger in the non-Western states
for the “young” cohort than in the Western region because schooling inputs expanded significantly in
non-Western states. The parameter θ is the reduced-form effect of an expansion in schooling inputs
associated with the program. More specifically, it measures whether individuals in states outside the
Western region that experienced large changes in schooling inputs due to the UPE program experienced
rapid growth in schooling differently from individuals in states in the Western region that did not
experience much change in schooling inputs.
Empirical Strategy# 3: Instrumental Variable Approach
To estimate the impact of education on fertility, we use a reduced-form model of the fertility
decision. As noted earlier, an important issue within existing literature concerns the causal interpretation
of the effect of education on fertility. If education has a causal effect on fertility, then an expansion in
schooling should induce lower fertility rates, holding other variables constant. However, unmeasured
individual, household and community-level constraints and resources may affect both education and
fertility decisions. For example, an increase in community wealth or level of economic development may
lead to higher educational attainment and lower fertility. Furthermore, education may serve as a proxy for
less measurable factors such as ability, cognitive skills, motivation, and parental background, and these
factors may be important determinants of a woman's fertility choices (Thomas, 1991). Ignoring these
factored will lead to biased estimates of the impact of education on fertility within the context of Ordinary
Least Squares (OLS) estimation.
One response to this identification problem is the instrumental variable approach.
Valid
instruments are variables that affect the level of educational attainment, but have no direct impact on
fertility. If we assume that the UPE program had no direct effect on fertility, other than through its effect
on educational attainment, then exposure to the UPE program can be used to construct instrumental
variable estimates of the impact of education on fertility outcomes. An individual’s exposure to the UPE
program is determined by age and region of schooling. The main dependent variable is the number of
children ever born.
The instrumental variable approach is described below:
(3)
Nijk = αi + BXijk + Sijk +εijk
In the equation (3) above, Nijk represents the number of children ever born to an individual i of cohort j
living in state k. More formally, OLS estimates may lead to biased estimates if ε ijk is correlated with
8
schooling due to unmeasured ability, and other factors such as family background or community wealth.
Two assumptions are required for exposure to the UPE program (which is based on age and region of
birth) to be a valid instrument. The first assumption is that exposure to the program is correlated with
schooling outcomes (once other explanatory variables have been controlled for). Equations (1) and (2)
above form the basis for the first stage regression. The second assumption that is required is that UPE
program had no direct impact on fertility but only through its impact on schooling.
We test this
assumption in our analysis below.
IV.
THE IMPACT OF THE UPE
DD Analysis: The Impact on Education
For this study, we first demonstrate the impact of the UPE program on female schooling. Beyond
the larger question of the paper about fertility, these results reflect on the impact of an education policy on
female schooling decisions. The results also form the basis for our first stage regression when we
investigate the effect of female schooling on fertility outcomes. To identify the impact of the UPE
program, we limit our sample to women of two cohorts: those born between 1958 and 1963 (age 13 to 18
when the program started) and those born between 1970 and 1975. Individuals born between 1970 and
1975 were aged 1-6 when the program started and are likely to have been the primary beneficiaries from
the program (we also refer to this cohort as the “UPE cohort”). In contrast, individuals who were born
between 1958 and 1963 are likely to have been too old to benefit from the UPE program and were
unlikely to be contaminated with overage enrollment like those born from 1969 to 1963. Our sample is
also restricted to individuals who have not moved since they were six years of age.
We compare the years of schooling for individuals in high-intensity (non-Western Region) and
low-intensity states (Western Region). The Westerns states had implemented tuition-free schooling prior
to the UPE program. In addition, the construction of classrooms in primary and teacher training schools
was generally more widespread outside the Western region. While schooling levels increased in both
high-intensity and low-intensity areas, schooling increased faster in the high intensity areas. If we
assume that the increase in schooling would not have been systematically different across high-program
and low-program regions in the absence of the program, then the difference-in-differences results can be
interpreted as a causal outcome of the program.
Ordinary least squares (OLS) estimation may lead to biased estimates of the impact of the UPE
program in the presence of unobserved state-level heterogeneity. This is an important concern because
states that experienced large changes in enrollments and schooling inputs during the UPE program (states
in non-Western regions) also had significantly lower educational and social indicators than the Western
region prior to the introduction of the UPE. For example, teacher quality tended to be lower outside the
Western region, particularly in the Northern States. Therefore, all of the models include state fixed
effects to deal with time-invariant unobserved heterogeneity at the state-level. The fixed-effects
9
regression allows us to “sweep out” time invariant unobserved state characteristics such as pre-UPE
teacher and school quality and availability, and initial differences in the level of economic development in
the state, which may affect educational attainment. The state fixed-effects that we include refer to an
individual’s state of birth and education.
Table 2 reports the effect of UPE on the completion of more than six years of schooling and the
number of years of schooling completed. The top panel displays the experiment of interest. As shown in
the baseline model, the UPE appears to have increased years of schooling by 1.09 years in the lowintensity states relative to the Western region. We also find that UPE program is associated with an 11
percentage point increase in the probability that an individual has completed more than six years of
schooling. All specifications includes controls for linear, quadratic, and cubic age terms, an urban
dummy, age of the household head, state-fixed effects and religion dummies, and living arrangement
controls.
One possible explanation for our results is mean reversion that may have taken place in the
absence of the program since the program had a higher impact in states/regions with relatively low female
enrollments. Therefore, in columns 2 and 5 we introduce the interaction of the female share of primary
school enrollment interacted with the young dummy. The inclusion of this variable appears to increase
the overall impact of the UPE program on schooling. In fact, we find that UPE program is associated
with a 19 percentage point increase in the probability that an individual has completed more than six
years of schooling. We also find that the UPE appears to have increased years of schooling by 1.3 years
in the low-intensity states relative to the Western region.
Another concern that arises in interpreting the results in the baseline model is that our estimates
do not account for the impact of time-varying state-level heterogeneity. In particular, schooling may have
increased faster in high-impact regions (outside the West) not solely because of the UPE program, but
rather due to omitted variables including time-varying changes in government programs and policies that
took place during this period. The UPE program was implemented during the “Oil Boom” era in Nigeria.
During this period, there was a significant expansion in public sector employment, as well as wages of
civil servants. This expansion in the federal civil service labor force may be correlated with the timing of
the UPE program. To account for this factor, we control for state-level growth in the civil service by
interacting the share of female employment in the civil service in 1971 with the dummy variable for the
1970-75 cohort. From columns 3and 6, our results are robust to the inclusion of this additional control
variable.
The bottom panel of Table 2 provides a robustness check of the results. One concern with the
analysis is that schooling outcomes may be higher for the treated cohort not because of the UPE program,
but rather due to a pre-existing trend in schooling. To address this concern, we compare 1958-1963
cohort to the 1952-57 cohort. Neither group should have been affected by UPE. If schooling levels were
10
increasing faster in the high intensity regions (mainly outside the Western region) prior to the UPE
program, then we should find a spurious positive coefficient for the unaffected cohorts (individuals 13-24
when the program started) when we compare the cohorts. When we compare successive older cohorts
(born before 1963) that were unaffected by the program, we generally find that education had generally
increased faster in the low-intensity areas in the Western region prior to the UPE program.
DD Analysis: The Impact on Early Fertility
Table 3 repeats the above analysis focusing on the impact of UPE on fertility. Since the UPE
cohort is a relatively young group, we limit our analysis to early fertility indicators. Our key dependent
variables here are whether a woman gave birth before age 15, and also the number of children born before
age 25. All estimates here include controls for age, age squared and age cubed, and a cohort dummy for
the “young” sample, and all other control variables (urban residence, religion, age of household head,
living arrangements, and state fixed effects).
From our results, the UPE cohort was about 9-11 percentage points less likely to have a first birth
before age 15 (column 1). Our baseline estimates for the impact of UPE on early fertility show a negative
but insignificant impact on the number of births before age 25 (column 4). However, in the final two
columns (columns 5-6), we present results that include controls for 1970 enrollment rates and female civil
service employment rates.
Here, we find that the UPE program appeared to have a negative and
statistically significant on the number of births before age 25, reducing the number of births before age
25. Our estimates of the impact of the program range from -0.83 to -0.91.
We interpret the above results as evidence that the UPE program had an impact on the number of
early births primarily through its effect on female schooling. However, we should caution that it is also
possible that the number of children born before age 25 may also be lower among the UPE cohort due to
other omitted variables that are correlated with the program. One possibility is that cohort difference in
contraceptive usage (an omitted variable in our analysis) may be correlated with the timing of the
program. We note that changes in the diffusion of modern contraceptive methods are less likely to be
important here because the prevalence of modern contraceptive methods remains relatively low in Nigeria
(Caldwell et al , 1992).
The bottom panel of Table 3 provides a robustness check of the results on early fertility. If early
fertility was decreasing faster in the high intensity regions (mainly outside the Western region) prior to
the UPE program, then we should find a spurious negative coefficient for the young ”unaffected” cohorts
when we compare the cohorts. When we compare successive older cohorts (born before 1963) that were
unaffected by the program, we generally find no significant trends in fertility by cohort and region prior to
the UPE program.
11
DD Analysis: The Impact of UPE Exposure and For Different Groups
Because the program had a limited duration (1976-1981), we can also construct a variable (UPE
exposure) that captures the number of years of the UPE program that an individual was exposed to from
the time of enrollment. The UPE exposure variable ranges from 0 to 6 years. Individuals who were born
in 1970, and who enrolled in primary school in 1976 (and attended school continuously) were most likely
exposed to a maximum of six years of the UPE program. It is possible that women who were over the age
of six (but less than 13) when the program was launched may also have also benefited from the program.
In addition, individuals born after the program ended (after 1976) may have also enjoyed some benefits, if
the program had a long-lasting impact on schooling. However, we focus our analysis on those born
between 1970 and 1975 because they were most likely to have benefited from the program.
Table 4 presents results on UPE exposure. We find that the number of years that an individual is
exposed to UPE had a positive and statistically significant impact on years of schooling completed. In
particular, an additional year of UPE exposure increased the years of schooling completed by 0.11-0.14
years. We also find that the length of exposure to the UPE program interacted with the high intensity
region dummy, has a positive but statistically insignificant impact on the years of schooling completed.
We also examine the effect of UPE exposure on the number of children ever born before age 25. Here,
we find that an additional year of UPE exposure reduces the number of early births by 0.02-0.06. We
should note that there is a negative association between the number of children ever born and the length
of exposure to the UPE program interacted with the high intensity region dummy, although this effect is
not statistically significant. These regressions include state fixed effects to account for time invariant
heterogeneity at the state level.
Table 5 provides results by location and religion. We note that the program had a larger impact
in urban areas and among non-Muslim households. The finding that schooling grew faster in urban areas
is not surprising given the difficulty of recruiting qualified teachers in rural areas. We should note that
our sample is composed of individuals who have never moved. Due to the self-selected nature of the
migration process, our rural sample may consist of individuals with lower levels of completed schooling
(if better educated workers migrate to urban areas).
The Impact of Educational Inputs on Schooling and Fertility
The results presented in the tables above consider all high-intensity states to be homogeneous.
However, our administrative data suggest that there was a great deal of variation in the program’s impact
in high-intensity areas (outside the Western region). To test the specific aspects of the program that had
an effect on schooling and demographic outcomes (other the elimination of tuition costs), we use data on
the planned number of new classrooms to be constructed in an individual’s state of origin at primary
schools and teacher training institutions. All the regressions discussed below include age controls,
12
controls variables for religion, age and gender of household head, urban status, living arrangements,
region dummies and state fixed effects.
If changes in schooling inputs during the program induced an
expansion in educational attainment, we expect that these effects will be larger outside the Western states
(because the program was more intensive in these states measured by the projected new number of
classrooms in primary schools and teacher training institutions).
From Table 6, we find very strong evidence that changes in schooling inputs induced significant
increases in years of schooling completed for the UPE cohorts educated in high-intensity areas. To
measure the effect of exposure to the UPE program on schooling and fertility, the coefficient of interest is
an interaction term that measures a woman’s age in 1976 interacted with the number of projected
classrooms in the state of origin.
We also measure program exposure by interacting whether an
individual was of primary school age at the onset of program (i.e. born between 1970 and 1975), the
number of planned classrooms, and whether an individual was educated in a high-intensity region
(outside the West). They key assumption here is that individuals who were 13 and above were less likely
to benefit from the program.
The results from Table 6 are consistent with large and significant changes in years of schooling
completed for states with large changes in schooling inputs. We find striking results associated with
classroom construction in teacher training institutions. We find that an additional classroom in a teacher
training institution (per 1000 pupils) has a larger impact on the years of completed schooling that the
construction of primary school classrooms. Finally, the construction of additional primary schools and
teacher training institutions had a larger impact on years of completed schooling in the high-intensity or
non-Western region.
Finally, Table 6 presents results with the number of children ever born as the dependent variable.
Here, we investigate the impact of the expansion in schooling inputs (classrooms in primary schools and
teacher training institutions) on fertility outcomes. These results provide convincing evidence that the
expansion in schooling inputs that occurred during the UPE program are negatively associated with the
number of early births measured by the number of children born before age 25. For classroom
construction, we find that an additional classroom constructed in teacher training institutions tended to
have a more negative impact on the number of children ever born than additional classrooms built in
primary schools. Finally, we should note that the reduction in number of children ever born associated
with classroom construction appears to have been greater for the UPE cohort in the high-intensity or nonWestern states.
The Instrumental Variable Strategy
In Table 7, we present results from our first stage regression with years of schooling
completed as the dependent variable. We use the ratio of actual to projected enrollments ( measured at
the 1976 state level) as an instrument for schooling in our analysis. The observed actual/projection ratio in
13
the state can be interpreted as the “actual” program impact of the UPE scheme in a given state. In states
with low enrollments relative to projected enrollments, years of schooling completed were less likely to
have increased for individuals in our sample. The results from Table 7 are consistent with negative and
significant association between years of schooling completed and actual to projected enrollments at the
1976 state level.
Consistent with earlier results, we find a strong positive association between years of schooling
completed and the interaction term (Age in 1976*projected number of primary school (teacher training)
classrooms)) and UPE exposure (measured in years).
As shown earlier, the construction of additional
primary schools and teacher training institutions had a larger impact on years of completed schooling in
the high-intensity regions outside the Western region.
Table 8 (Panel A) reports Ordinary Least Squares (OLS) estimates of the effect of education on
fertility measured as the number of children ever born before Age 25.
These results ignore the
endogeneity of the schooling decision. In Panel B, we present indirect least squares results, while in
panels C-F, we present two-stage least squares estimates of the effect of schooling on fertility. Our
results here include state dummies, and we also include additional control variables in some specifications
to deal with the concern that time-varying heterogeneity at the state-level and household-level variables
may impact our results. All of our estimates include controls for age, urban status, religion, living
arrangements (with parent=1).
First, we discuss results from the OLS estimation (Panel A) which ignore the endogeneity of
schooling. The OLS results suggest that schooling has a negative and statistically significant impact on
the number of children ever born. In particular, an additional year of schooling reduces the number of
children ever born by about 0.098-0.116. We compare random-effects estimates to our fixed-effects
model and do not find any statistically significant differences in these estimates. Since we have multiple
observations for women within a given household, thus we can also estimate a household-fixed effects
model, to account for unobserved heterogeneity at the household level due to differences in household
income or economic status (Column 5). The coefficient from the household-fixed effects model is
slightly lower (-0.111), but not statistically different from earlier results (columns 1-4). We also include
husband’s education and household assets and find that the effect of schooling on early fertility is only
slightly lower (-0.098 to -0.106).
Our indirect least squares results are presented in Panel B. Here, we use a single instrument—the
size of the program enrollment shock in an individual’s state of birth. From Panel B, (Columns 1-5) we
find that an additional year of schooling reduces the number of children ever born by 0.06-0.081. Panel
C uses two instruments –the size of the enrollment shock and Age in 1976 interacted with projected
number of new teacher training institutions in the state. Panel D uses the state-level enrollment shock as
well as the length of UPE exposure (0-6 years). In Panel E, we use all three instruments.
14
When we instrument for schooling, we find that the impact of schooling on fertility generally
become more negative. Our estimates from Panel C through Panel E range from -0.093 to -0.241. This
suggests that the OLS estimates may underestimate the magnitude of the effect of schooling on fertility.
However, not all differences between OLS and 2SLS estimates are statistically significant.
We can compare our estimates on the impact of female schooling on fertility to a number of
recent studies that have examined the relationship between fertility and female schooling in developing
countries. Ainsworth et al (1996) use data from the Demographic and Health Surveys for 14 Sub
Saharan African countries, and find schooling has a negative and significant association with the number
of children ever born in 13 out of the 14 countries included in their study. Their point estimates on the
impact of additional year on the number of children ever born range from -0.06 to -0.134, although they
do not attempt to instrument for schooling.
In the results presented in Table 8, education is assumed to have a linear impact on fertility. A
more flexible specification may be required to provide further insights into the impact of education on
fertility. In Table 9 (Panel B), we include 8 dummy variables to capture the impact of schooling using a
more flexible approach (the omitted category is no formal schooling). From column 1 (Panel B), our
results support our earlier findings from the linear specification in that education has a negative impact on
the number of children ever born for all levels of schooling. However, the negative effect of schooling is
only statistically significant for individuals who have completed more than four years of schooling.
Jejeebhoy (1995) and Schultz (1998) note that in some countries, the completion of one or two years of
primary schooling may not have a significant effect on the number of children ever born. In fact, women
with incomplete primary schooling may actually have slightly higher fertility rates than women with no
schooling due to the lower prevalence of traditional birth spacing among this group (Cleland and
Rodriguez, 1996). From our results, (column 1, Panel C) show that women with 4-6 years of completed
primary schooling are associated with 0.49 fewer children than women who have less than four years of
schooling.
This supports the near-universal finding that completed primary schooling and secondary
schooling are strongly negatively linked with the number of children ever born.
Results from column 1(Panels B and C) point to significant non-linearities in the impact of
schooling on fertility. There are two potential interpretations of the non-linearities observed in our
results. Thomas (1999) discusses these explanations using data from South Africa. One explanation is a
human-capital based in which higher levels of primary schooling (four or more years) have a more
productive effect in reducing fertility than earlier years. While the completion of more years of schooling
may lead to a larger negative effect of education on fertility due to productivity differences, it is also
likely that selection in the educational attainment process may be useful in explaining these observed nonlinearities. In particular, Thomas (1999) describes this pattern as one of “steps and flats” in which selfselection in educational attainment may drive the non-linear relationship between education and fertility.
Families may choose different levels of education for their female children based on tastes, abilities,
15
and/or resources. For example, women who complete only one year of secondary school could be drawn
from a different segment of the ability distribution than women who complete primary school and then
drop out of school. Furthermore, women who complete secondary school (particularly in regions with
low levels of schooling) are likely to be highly selective group. Lam and Duryea (1999) and Malacca and
Thomas (1999) also find significant non-linearities on the effect of schooling on fertility using data from
Brazil and Zimbabwe, respectively.
The effect of education on fertility may also depend on socio-economic environment in which
fertility decisions are made. The NDHS provides a rich setting in which explore the role of contextual
factors may affect the impact of education on fertility. There is a great deal of variation in community
variables in our data including access to health care, sanitation, and transportation infrastructure. Our
results in column 1 include community fixed effects. The inclusion of community fixed-effects allow us
to account for the role of time-invariant observed and unobserved community determinants of fertility.
For example, Jejeebhoy (1995) presents evidence that the highest differential in fertility for secondary
school educated women compared to women with no schooling is found in Latin America, while the
lowest differentials are found in Sub Saharan Africa. Cleland and Rodriguez (1996) highlight the
importance of contextual factors in their study of the effect of education on fertility using data from the
World Fertility Surveys for 31 countries.
In particular, the level of economic development in a
community, the presence of family planning programs, employment opportunities for women, as well the
extent of mass schooling (Caldwell, 1980) can affect the impact of education on fertility.
In fact,
Caldwell (1980) argues that it is community-level education rather than an individual’s educational
attainment that affects fertility. In addition, the inclusion of community fixed-effects also allows us to
deal with the role of social and cultural norms towards education and fertility decisions.
Our result on the negative impact of female schooling on fertility appears robust to the inclusion
of community fixed effects. When we include state fixed effects (column 2), however, the impact of low
levels of primary schooling on fertility is positive and statistically insignificant. However, we still find
that completing more than four years of schooling is consistently negatively associated with the number
of children ever born. From column 2 (Panel C), women with 4-6 years of completed primary schooling
are associated with 0.42 fewer children than women who have less than four years of schooling.
The results presented thus far do not account fully for the impact of spousal characteristics,
including spousal education. Within the NDHS sample, nearly 70 percent of women have been, or are
currently married. When we include an indicator for whether the spouse is present in the household, our
sample size falls (N=3293). The inclusion of this dummy variable does not appear to have effect the
negative impact of female education on fertility within the linear specification (both the size of the
coefficient and the level of significance remain the same). However, when we examine results with
schooling dummies, we a find a positive, but insignificant effect of education on fertility for women who
have completed less that five years of schooling. These results are consistent with results from other
16
studies from sub Saharan Africa, where education may have an insignificant impact on fertility at low
levels of schooling (Ainsworth et. al, 1996). Results from our spline specification (Panel C, column 3),
suggest that 4-6 years of completed female schooling are associated with 0.43 fewer children.
When we include spousal years of schooling, our results on the effect of female education on
fertility remain comparable to earlier results. In addition, we find that spousal education does not have a
statistically significant impact on fertility after we have controlled for female education. This finding is
consistent with evidence from other countries in Sub Saharan Africa. Ainsworth et al (1996) find that for
a sample for ever-married women, husband's schooling has no significant relation with fertility in about
one-third of the sub Saharan African countries in their study. Spousal education is highly correlated with
female schooling in our sample (correlation coefficient=0.71), which may help explain its insignificant
impact on fertility once we have controlled for female education. Basu (1999) provides evidence that
women with more years of completed schooling tend to marry highly educated men.
Household resources may also have an important effect on fertility decisions. One shortcoming
of the NDHS data is that there is limited information on household income. Ideally, a measure of the
household permanent income would allow us to control for two effects associated with household
resources. First, an increase in household income may induce households to have more children due to a
positive income effects. Second, with higher income the opportunity costs of time spent in child-related
activities including child-bearing and child-rearing are likely to increase. The inclusion of household
fixed-effects in column 5 (which should account for household characteristics such as household income)
does not appear to affect our estimates. Results from our spline specification (Panel C, column 5),
suggest that 4-6 years of completed female schooling are associated with 0.36 fewer children, compared
to individuals with 0-3 years of completed schooling.
V. CONCLUSIONS
According to a recent World Bank and United Nations publications, the “single most effective
policy action” that can be taken to lower fertility outcomes is to increase female education levels. In this
paper, we investigate the causal link between education and fertility using a large-scale policy experiment
from Nigeria. Results from the Nigeria suggest that the expansion in female education had a significant
impact on both education and fertility decisions. We find that exposure to the UPE program is associated
with 0.65-1.01 years of additional schooling. Under the assumption that exposure to the UPE program is
a valid instrument for schooling, we can construct 2SLS estimates of the impact of female education on
schooling. Our 2SLS estimates are slightly higher than OLS estimates, and suggest that an additional
year of schooling reduces the number of children ever born by 0.114-0.13.
17
REFERENCES
Ainsworth , M., Beegle K. and A. Nyamete, (1996). “The Impact of Women's Schooling on Fertility and
Contraceptive Use: A Study of Fourteen Sub-Saharan African Countries,” World Bank Economic
Review, 10(1):85-122
Angrist, J.D., E. Bettinger, , E. Bloom, E. King and , M. Kremer (2002). “Vouchers for Private Schooling
in Colombia: Evidence from a Randomized Natural Experiment,” American Economic Review,
92(5):1535-58.
Angrist, J. D. and W. N. Evans (1999). “Schooling and Labor Market Consequences of the 1970 State
Abortion Reforms”, Research in Labor Economics. 18: 75-113
Becker, G. S. (1981). Treatise on the Family Cambridge: Harvard University Press.
Bray, M. (1981). Universal Primary Education in Nigeria: A Study of Kano State London: Routledge and
Kegan Paul Ltd
Case, A. and A. Deaton (1999). “Schooling Inputs and Educational Outcomes in South Africa” Quarterly
Journal of Economics, 114(3) :1047-84
Caldwell, J.C. (1980) Mass Education as a Determinant of the Timing of Fertility Decline,” Population
and Development Review 6(2): 225-255.
Cleland, J. and G. Rodriguez (1988). “The Effect of Parental Education on Marital Fertility,” Population
Studies, 42(3):419-442.
Cochrane, S.H. (1979) Fertility and Education: What Do We Really Know? Baltimore: Johns Hopkins
Press
Duflo, E. (2001). “Schooling and Labor Market Consequences of School Construction in Indonesia:
Evidence From an Unusual Policy Experiment” American Economic Review 91:4:795-813
Glewwe, P. (2002) “Schools and Skills in Developing Countries: Education Policies and Socioeconomic
Outcomes,” Journal of Economic Literature 11 (2):436-482
Hirschman, C. (1985). “Premarital Socioeconomic Roles and the Timing of Family Formation: A
Comparative Study of Five Asian Societies.” Demography 22:35-59
Francis, Paul A (1998). “Hard Lessons: Primary Schools, Community and Social Capital in Nigeria”
World Bank Technical Paper No. 420 Africa Region Series Washington D.C. World Bank
Federal Government of Nigeria, Implementation Committee for the National Policy on Education
Blueprint 1978-1979
Federal Office of Statistics, Annual Abstract of Statistics various years
Federal Office of Statistics, Social Statistics in Nigeria various years
Jejeebhoy, S. J. (1995) Women’s Education, Autonomy and Reproductive Behavior: Experience from
Developing Countries New York: Oxford University Press.
18
Lam, D. and S. Duryea, (1999). “Effects of Schooling on Fertility, Labor Supply and Investments in
Children, with Evidence from Brazil,” Journal of Human Resources, 34(1):160-192
Mason, K.O. (1986) The Status of Women: Conceptual and Methodological Debates in Demographic
Studies: Sociological Forum 1:284-300
Nwachukwu, A.E. (1985). An Historical Analysis of the Roots of Universal Public Primary Education in
Nigeria (1900-1980) University of Kansas Ph. D. dissertation
Odaga, A. and Heneveld, W. (1995) Girls and Schools in Sub-Saharan Africa: From Analysis to Action
Washington D.C.: World Bank
Pitt, M. M., M.R.. Rosenzweig and D. M. Gibbons, (1993). “The Determinants and Consequences of the
Placement of Government Programs in Indonesia,” World Bank Economic Review, (7)3:319348.
Rosenzweig, M. R. and T.P. Schultz (1989). “Schooling, Information and Nonmarket Productivity:
Contraceptive Use and Its Effectiveness,” International Economic Review, 30(2):457-77
Rosenzweig, M. R. and T.P. Schultz (1985). “The Demand and Supply of Births and Its Life-Cycle
Consequences,” American Economic Review, 75(5): 992-1015
Schultz, T. P., ed (2002) “Why Governments Should Invest More to Educate Girls.” World Development
30:207-225
Schultz, T. P., ed (1998) “Demand for Children in Low Income Countries” Handbook of Population and
Family Economics volume 1A eds. Mark Rosenzweig and Oded Stark Amsterdam, The
Netherlands Elsevier Press
Schultz, T. P., ed (1995) Investment in Women's Human Capital. 1995, Chicago and London: University
of Chicago Press
Thomas D. (1999) “Fertility, Education and Resources in South Africa” in Critical Perspectives on
Schooling and Fertility in the Developing World Bledsoe C.H, J.B. Casterline, JA. JohnsonKuhn, and J.G. Haaga (eds), pp 138-176. Washington D.C.: National Academy Press.
Thomas, D. and J. Maluccio, (1996) “Fertility, Contraceptive Choice and Public Policy in Zimbabwe,”
World Bank Economic Review 10(1):189-222
Thomas, D., J. Strauss, J. and M. Henriques, (1991) “How Does Mother's Education Affect Child
Height?” Journal of Human Resources, 26(:2): 183-211
UNESCO, 1971. Education In Nigeria, Educational Projects for External Aid Vol. 1 Report 560
Paris:UNESCO
World Bank, 2002. African Development Indicators database
19
Figure 1A: Growth Rate of Primary School Enrollments by Region, 1970-1981
Annual Growth Rate
(in %)
Growth Rate: Primary School Enrollments
50
40
30
20
10
0
71
72
73
74
75
76
77
78
79
80
81
Year
Nonwest
West
Figure 1B: Growth Rate of Number of Primary Schools by Region, 1970-1981
Growth Rate: Number of Primary Schools
Annual Growth Rate
(in %)
50
40
30
20
10
0
-10
71
72
73
74
75
76
77
78
79
80
81
-20
Year
Nonwest
W est
Source: Annual Abstract of Statistics various years, Federal Office of Statistics, Nigeria
Social Statistics in Nigeria various years, Federal Office of Statistics, Nigeria
20
Growth Rate: Number of Students in Teacher Training Inst.
Annual Growth Rate
(in %)
100
80
60
40
20
0
-20
71
72
73
74
75
76
77
78
79
80
81
Year
Nonwest
West
Figure 1D: Growth Rate of Number of Teacher Training Institutions by Region, 1970-1981
Growth Rate: Number of Teacher Training Inst.
Annual Growth Rate
(in %)
50
40
30
20
10
0
-10
71
72
73
74
75
76
77
78
79
80
81
-20
Year
Nonwest
West
Source: Annual Abstract of Statistics various years, Federal Office of Statistics, Nigeria
Social Statistics in Nigeria various years, Federal Office of Statistics, Nigeria
21
Figure 3: Education and Fertility: Regional Differences by Birth Cohort
Education: Regional Differences by Birth Cohort
Years of Schooling
10
8
6
4
2
0
1940-45
1946-51
1952-57
1958-63
1964-69
1970-75
1976-81
Cohort
Nonwest Schooling
W est Schooling
Fertility: Regional Differences by Birth Cohort
No of Births before 25
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1940-45
1946-51
1952-57
1958-63
1964-69
1970-75
1976-81
Cohort
Nonwest Births before 25
W est Births before 25
Source: 1990 & 1999 Nigerian Demographic Health Survey (NDHS)
22
Table 1: Summary Statistics of the 1999 NDHS Data
Never Moved
(Sample for analysis))
Mean
Std Dev
Demographic Variables
No. of Children Ever Born
No. of Births before Age 25
Yrs of Education: Female
Yrs of Education: Husband
Completed Primary School
Age at First Marriage
Age of Female
Age of Head
Living with Parents
Head of Household
Year of Birth
Born 1976-1981
Born 1970-1975
Born 1964-1969
Born 1958-1963
Born 1952-1957
Born 1946-1951
Religion and Ethnicity
Muslim
Christian
Other Religion
Hausa
Yoruba
Ibo
Residency Characteristics
Urban
UPE Variables
High-Intensity: Non-west
Actual/Projected Enrollment
Years of Exposure to UPE
Female Enrollment in 1970
Projected No. of Classrooms –
Teacher Training Institutes (in
1000)
Projected No. of Classrooms –
Primary Schools (in 1000)
Female Civil Service in 1971
Observations
Moved
(Excluded from sample)
Mean
Std Dev
2.16
1.27
5.06
5.50
0.50
17.14
24.28
46.33
0.31
0.07
(2.78)
(1.68)
(4.79)
(5.66)
(0.50)
(4.39)
(10.31)
(13.83)
(0.46)
(0.25)
2.74
1.60
4.89
5.32
0.48
17.09
27.17
45.99
0.31
0.06
(2.92)
(1.78)
(4.80)
(5.58)
(0.50)
(4.49)
(9.17)
(13.63)
(0.46)
(0.24)
0.49
0.16
0.15
0.10
0.06
0.02
(0.50)
(0.36)
(0.36)
(0.31)
(0.24)
(0.13)
0.49
0.16
0.15
0.15
0.07
0.03
(0.50)
(0.37)
(0.35)
(0.36)
(0.26)
(0.16)
0.44
0.54
0.02
0.23
0.23
0.14
(0.50)
(0.50)
(0.14)
(0.42)
(0.42)
(0.34)
0.44
0.55
0.02
0.25
0.15
0.19
(0.50)
(0.50)
(0.14)
(0.43)
(0.36)
(0.39)
0.34
(0.48)
0.31
(0.46)
0.79
1.37
0.55
34.03%
(0.41)
(0.50)
(1.39)
(6.36)
-
-
2.41
(1.75)
60.66
(39.83)
-
-
1029.26
(2493.02)
6,208
-
3602
23
Table 2: The impact of the UPE program on Schooling
A. Experiment of Interest
Treatment Group: Born 1970-1975 (Age 1-6 in 1976)
Control Group: Born 1958-1963 (Age 13-18 in 1976)
Dependent Variable: Completed
Seven Years of Education
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(1)
(2)
(3)
Dependent Variable: Years of Schooling
(4)
Adding
Female
Enrollment
(5)
Adding
Female
Civil Service
(6)
Baseline
Model
Born 1970-75 *
High-Intensity State
0.11
(0.05)
0.19
(0.06)
0.19
(0.06)
1.09
(0.49)
1.27
(0.65)
1.11
(0.66)
Observations
1655
1655
1655
1655
1655
1655
B. Unaffected Cohorts (Robustness Check)
Treatment Group: Born 1958-1963 (Age 13-18 in 1976)
Control Group: Born 1952-1957 (Age 19-24 in 1976)
Dependent Variable: Completed
Seven Years of Education
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(1)
(2)
(3)
Born 1958-63 *
High-Intensity State
-0.08
(0.05)
-0.06
(0.06)
-0.07
(0.07)
Dependent Variable: Years of Schooling
(4)
Adding
Female
Enrollment
(5)
Adding
Female
Civil Service
(6)
-0.64
(0.60)
-0.23
(0.77)
-0.39
(0.77)
Baseline
Model
Observations
1018
1018
1018
1018
1018
1018
Notes: All results include 1970 state fixed effects. The baseline models include: age as linear, quadric, and cubic
terms, urban dummy variable, age of head of household, dummy variable for whether the individual lives with a
parent, and religion dummies.
24
Table 3: The impact of the UPE program on Fertility
A. Experiment of Interest
Treatment Group: Born 1970-1975 (Age 1-6 in 1976)
Control Group: Born 1958-1963 (Age 13-18 in 1976)
Dependent Variable: First Birth
Before Age 15
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(1)
(2)
(3)
Dependent Variable: No of Births Before
Age 25
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(4)
(5)
(6)
Born 1970-75 *
High-Intensity State
-0.09
(0.04)
-0.11
(0.06)
-0.10
(0.06)
-0.07
(0.20)
-0.91
(0.27)
-0.83
(0.27)
Observations
1658
1658
1658
1658
1658
1658
B. Unaffected Cohorts (Robustness Check)
Treatment Group: Born 1958-1963 (Age 13-18 in 1976)
Control Group: Born 1952-1957 (Age 19-24 in 1976)
Dependent Variable: First Birth
Before Age 15
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(1)
(2)
(3)
Born 1958-63 *
High-Intensity State
0.07
(0.05)
0.07
(0.07)
0.07
(0.07)
Dependent Variable: No of Births Before
Age 25
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(4)
(5)
(6)
0.16
(0.26)
-0.13
(0.34)
-0.15
(0.35)
Observations
1021
1021
1021
1021
1021
1021
Notes: All results include 1970 state fixed effects. The baseline models include: age as linear, quadric, and cubic
terms, urban dummy variable, age of head of household, dummy variable for whether the individual lives with a
parent, and religion dummies.
25
Table 4: The Impact of the Intensity of UPE Exposure
Treatment Group: Born 1970-1975 (Age 1-6 in 1976)
Control Group: Born 1958-1963 (Age 13-18 in 1976)
Dependent Variable: Years of
Schooling
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(1)
(2)
(3)
Years of UPE
Exposure
0.14
(0.07)
0.13
(0.07)
Dependent Variable: No of Births Before
Age 25
Adding
Adding
Baseline
Female
Female
Model
Enrollment Civil Service
(4)
(5)
(6)
0.11
(0.07)
-0.02
(0.03)
-0.06
(0.03)
-0.05
(0.03)
Observations
1655
1655
1655
1658
1658
1658
Notes: All results include 1970 state fixed effects. The baseline models include: age as linear, quadric, and cubic
terms, urban dummy variable, age of head of household, dummy variable for whether the individual lives with a
parent, and religion dummies.
Table 5: The Impact of UPE by Location and Religion
Treatment Group: Born 1970-1975 (Age 1-6 in 1976)
Control Group: Born 1958-1963 (Age 13-18 in 1976)
Born 1970-75 *
High-Intensity State
Dependent Variable: Years of
Schooling
NonUrban
Rural
Muslim
Muslim
(1)
(2)
(3)
(4)
Dependent Variable: No of Births Before
Age 25
NonUrban
Rural
Muslim
Muslim
(5)
(6)
(7)
(8)
4.38
(1.30)
-0.72
(0.44)
-1.09
(0.77)
2.02
(0.84)
-1.29
(1.29)
-1.02
(0.36)
-0.79
(0.30)
-0.69
(0.62)
Observations
575
1080
891
764
577
1081
893
765
The models include the following controls: age as linear, quadric, and cubic terms, urban dummy variable, age of
head of household, dummy variable for whether the individual lives with a parent, religion dummies, an interaction
term between young cohort and state female enrollment in 1970, an interaction term between young cohort and
female civil service in 1971 and 1970 state fixed effects.
26
Table 6: The Impact of Expansion in Schooling Inputs during UPE Program
Dependent Variable: Years of
Schooling
(1)
(2)
(3)
(4)
Age in 1976 * Projected Primary 0.004
School Classrooms (* 1000) (0.002)
Age in 1976 * Projected Teacher
Institute Classrooms (* 1000)
Born 1970-75 * Projected
Primary School Classrooms
* High Intensity Region
Born 1970-75 * Projected
Teacher Institute Classrooms
* High Intensity Region
Dependent Variable: No of Births Before
Age 25
(5)
(6)
(7)
(8)
-0.004
(0.001)
0.0527
(0.043)
-0.071
(0.018)
0.019
(0.010)
-0.005
(0.004)
1.968
(0.629)
-0.546
(0.260)
Observations
1655
1655
1655
1655
1658
1658
1658
1658
All models include 1970 state fixed effects and the following controls: age as linear, quadric, and cubic terms, urban
dummy variable, age of head of household, dummy variable for whether the individual lives with a parent, religion
dummies.
27
Table 7: First-Stage Estimates of the Instrumental Variables
Dependent Variable: Years of Schooling
Method Instrument(s) Used
A OLS None
B ILS
Actual/Projected Enrollment
Actual/Projected Enrollment
C 2SLS
Age in 1976 * Projected No. of
Teacher Training Classrooms
Actual/Projected Enrlmt (1976)
D 2SLS
UPE Exposure
Actual/Projected Enrlmt (1976)
E 2SLS UPE Exposure
(1)
Adding
Female
Enrollment
(2)
Adding
Civil
Service
(3)
-
-
-
-0.015
(0.003)
-0.015
(0.003)
-0.016
(0.003)
-0.016
(0.003)
0.05
(0.04)
-0.017
(0.003)
0.12
(0.05)
-0.017
(0.003)
0.12
(0.05)
-0.014
(0.003)
0.10
(0.07)
-0.014
(0.004)
0.09
(0.07)
-0.015
(0.004)
0.08
(0.07)
-0.014
(0.003)
0.17
(0.08)
-0.014
(0.003)
0.17
(0.08)
-0.015
(0.004)
0.16
(0.08)
Baseline
Age in 1976 * Projected No. of
0.10
0.10
0.16
Teacher Training Classrooms (0.05)
(0.05)
(0.05)
The baseline model includes the following controls: age as linear, quadric, and
cubic terms, urban dummy variable, age of head of household, dummy variable
for whether the individual lives with a parent, and religion dummies.
28
Table 8: Second-Stage Estimates of the Effect of Education on Fertility
Dependent Variable: No of Births before Age 25
Results reported: Coefficients of Years of Schooling
Method Instrument(s) Used
Baseline
Random
Female
Effects Enrollment
Civil
Service
(1)
(2)
(3)
(4)
A OLS None
-0.116
(0.01)
-0.116
(0.01)
-0.116
(0.01)
-0.115
(0.01)
B ILS
-0.081
(0.09)
-0.081
(0.09)
-0.061
(0.09)
-0.080
(0.09)
Actual/Projected Enrollment
C 2SLS Age in 1976 * Projected No. of
Teacher Training Classrooms
-0.182
(0.09)
[14.23]
-0.182
(0.09)
[14.23]
-0.142
(0.08)
[4.14]
-0.150
(0.08)
[3.64]
Actual/Projected Enrlmt (1976)
-0.093
(0.09)
[0.17]
-0.093
(0.09)
[0.17]
-0.101
(0.09)
[2.81]
-0.105
(0.08)
[1.99]
-0.241
(0.08)
[15.56]
-0.24
(0.08)
[15.56]
-0.203
(0.07)
[7.78]
-0.196
(0.07)
[6.62]
D 2SLS
Actual/Projected Enrollment
UPE Exposure
Household Adding
Adding
Fixed
Husband's Household
Effects Education
Assets
(5)
(6)
(7)
-0.110
(0.01)
-0.106
(0.01)
-0.098
(0.01)
Actual/Projected Enrlmt (1976)
E 2SLS
UPE Exposure
Age in 1976 * Projected No. of
Teacher Training Classrooms
State Dummies
Yes
No
Yes
Yes
Yes
Yes
State Random Effects
No
Yes
No
No
No
No
Born 1970-75*State Female Enrlmt 1970
No
No
Yes
Yes
Yes
No
Born 1970-75*Female Civil Service 1971
No
No
No
Yes
Yes
No
Household Fixed Effects
No
No
No
No
Yes
No
The baseline model includes the following controls: age as linear, quadric, and cubic terms, urban dummy variable,
age of head of household, dummy variable for whether the individual lives with a parent, and religion dummies.
Over-identification test statistic is shown in square brackets. For Method C and D, the 5% critical value of Chisquare with 1 Df is 3.84; for Method E, the 5% critical value of Chi-square with 2 Df is 5.99.
29
Yes
No
No
No
No
Table 9: Impact of Education on Fertility: Specification Checks
Dependent variable: No of Births before Age 25
(1)
(2)
(3)
(4)
(5)
-0.12
(0.01)
-0.12
(0.01)
-0.10
(0.01)
-0.10
(0.01)
-0.09
(0.02)
1 year
-0.17
(0.54)
0.09
(0.48)
0.28
(0.67)
0.28
(0.67)
0.61
(0.79)
2 years
0.71
(0.59)
0.51
(0.47)
0.78
(0.55)
0.83
(0.55)
1.49
(0.62)
3 years
0.35
(0.42)
0.22
(0.39)
0.53
(0.45)
0.14
(0.48)
0.01
(0.53)
4 years
-0.31
(0.34)
-0.29
(0.31)
-0.26
(0.37)
-0.44
(0.40)
-0.87
(0.45)
5 years
-0.61
(0.41)
-0.26
(0.37)
-0.48
(0.41)
-0.39
(0.41)
-0.63
(0.47)
6 years
-0.61
(0.15)
-0.42
(0.13)
-0.39
(0.14)
-0.31
(0.15)
-0.25
(0.17)
7 years
-0.51
(0.60)
-0.16
(0.52)
-0.14
(0.55)
0.06
(0.58)
0.35
(0.67)
More than 7 years
-1.35
(0.15)
-1.27
(0.13)
-1.12
(0.14)
-1.00
(0.17)
-0.93
(0.19)
4-6 years
-0.60
(0.14)
-0.42
(0.12)
-0.43
(0.14)
-0.35
(0.14)
-0.36
(0.17)
7 or more years
-1.36
(0.15)
-1.27
(0.12)
-1.14
(0.14)
-1.00
(0.16)
-0.95
(0.19)
No
Yes
Yes
Yes
Yes
Panel A
Linear term
Years of Schooling
Panel B
With Schooling dummies
(No Schooling is the omitted category)
Panel C
Spline estimates (0-4 years is the omitted category)
State Fixed Effects (1970)
Community Fixed Effects
Yes
No
No
No
No
Spouse in Household (Dummy variable)
No
No
Yes
Yes
Yes
Husband's Education
No
No
No
Yes
Yes
Household Fixed Effects
No
No
No
No
Yes
The baseline model includes the following controls: age as linear, quadric, and cubic terms, urban dummy variable,
age of head of household, dummy variable for whether the individual lives with a parent, religion dummies.
30
Appendix 1: The Full Coefficients Set for Column (6) in Table 2 & 3
A. Experiment of Interest
Treatment Group: Born 1970-1975 (Age 1-6 in 1976)
Control Group: Born 1958-1963 (Age 13-18 in 1976)
Born 1970-75 * HighIntensity State
Age
Age Squared
Age Cubed
Born 1970-75
Born 1970-75 * State
Female Enrollment 1970
Born 1970-75 * Female
Civil Service 1971
Living with Parent
Age of Head
Urban
Muslim
Christian
Constant
Dependent Variable: Years of
Schooling
Dependent Variable: No of
Births Before Age 25
1.11
(0.66)
-0.83
(0.27)
-5.09
(6.21)
0.17
(0.20)
-0.002
(0.002)
-0.61
(2.44)
0.06
(0.05)
-0.28
(0.16)
2.11
(0.40)
-0.02
(0.01)
2.56
(0.23)
1.10
(0.78)
4.61
(0.76)
52.20
(60.37)
3.60
(2.56)
-0.11
(0.08)
0.001
(0.001)
2.44
(1.00)
-0.11
(0.02)
0.15
(0.07)
-1.14
(0.17)
-0.0001
(0.004)
-0.28
(0.09)
-0.53
(0.32)
-1.07
(0.31)
-34.55
(24.89)
F Statistic
52.19
15.91
Observations
1655
1658
The models include the following controls: age as linear, quadric, and cubic terms, urban dummy variable, age of
head of household, dummy variable for whether the individual lives with a parent, religion dummies, an interaction
term between young cohort and state female enrollment in 1970, an interaction term between young cohort and
female civil service in 1971 and 1970 state fixed effects.
31
Appendix 2: The Full Coefficients Set for First and Second Stage Regression
First-Stage Regression: Method E, Column 3 in Table 7
Second-Stage Regression: Method E, Column 4 in Table 8
First Stage:
Dependent Variable: Years of
Schooling
IVs
Actual/Projected Enrlmt (1976)
-0.015
(0.004)
UPE Exposure
0.16
(0.08)
Age in 1976 * Projected No. of
Teacher Training Classrooms
0.16
(0.05)
Age
Age Squared
Age Cubed
Born 1970-75 * State Female
Enrollment 1970
Born 1970-75 * Female Civil
Service 1971
Living with Parent
Age of Head
Urban
Muslim
Christian
Constant
State Dummies
0.04
(5.06)
-0.01
(0.16)
0.0001
(0.002)
-0.008
(0.04)
-0.31
(0.16)
2.12
(0.40)
-0.02
(0.01)
2.48
(0.23)
0.92
(0.77)
4.59
(0.76)
11.48
(50.12)
Yes
Second Stage:
Dependent Variable: No of
Births Before Age 25
-0.196 1
(0.07)
4.83
(1.61)
-0.15
(0.05)
0.001
(0.0005)
-0.06
(0.02)
0.11
(0.07)
-0.73
(0.23)
-0.005
(0.004)
0.23
(0.21)
-0.31
(0.33)
-0.15
(0.47)
-45.28
(15.92)
Yes
Adj. R2
0.44
0.18
Observations
1655
1655
1
This result represents the coefficient on Years of Schooling in the No of Births Before Age 25 equation after we
have instrumented for years of schooling completed.
The models include the following controls: age as linear, quadric, and cubic terms, urban dummy variable, age of
head of household, dummy variable for whether the individual lives with a parent, religion dummies, an interaction
term between young cohort and state female enrollment in 1970, an interaction term between young cohort and
female civil service in 1971 and 1970 state fixed effects.
32