University of Southern California Dissertations and Theses

Essays on education and institutions in developing countries

(USC Thesis Other)

Essays on education and institutions in developing countries

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON EDUCATION AND INSTITUTIONS IN
DEVELOPING COUNTRIES
by
Tushar Bharati
A dissertation presented to the
Faculty of the USC Graduate School
University of Southern California
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in the subject of Economics.
August, 2018
Copyright 2018 Tushar Bharati
Abstract
In this dissertation, I examine the impact of government policies on educational attainment in de-
veloping countries and how the lack of information and education might give rise to low quality
institutions in these regions. The first chapter examines the role of an education policy in helping
individuals recover from ‘early life developmental insults’ and documents near complete recovery
for those exposed to the policy. The second chapter evaluates the impact of two public policies in
Tanzania on completed schooling and school start age. Using the ratio of the impacts of the two
policies and their interaction on schooling and start age, it documents that those who benefited from
the first policy of iodine supplementation were better at converting extra years in school into com-
pleted years of schooling. We interpret this as evidence of dynamic complementarity between years
in school and iodine supplementation. The third chapter examines the role of co-ethnic voting in
the election of bad quality representatives. I propose an asymmetric information candidate choice
model and test it using data from elections in two states of India. I find that a party’s choice to field
a candidate of a particular ethnicity depends on the ethnic composition of the constituency’s voters
and the ethnicity of other candidates running for office. Representatives elected in constituencies
with open elections, where candidates from all ethnicities can contest election and diversification
along ethnic lines is possible and beneficial, are more likely to have a prior criminal record. How-
ever, conditional on the average criminal involvement of the pool of candidates to choose from, there
is no difference across constituencies in terms of choosing a criminal candidate for office. This sug-
gests that election of criminal candidates to offices is, in part, due to political parties disregarding
potential candidates’ involvement in crime in order to be able to diversify along ethnic lines. How-
ever, this strategic candidate selection seems to be absent when constituencies are smaller, voters
are more educated, or the cost of collecting information is lower.
i
Acknowledgments
I am deeply indebted to my advisors for their constant encouragement and guidance. John
Strauss has trained me to be clear in my thinking and methodical in my research. Jeff
Nugent has taught me the importance of context in my research and inspired me to work
tirelessly towards excellence. Daniel Bennett has been a constant voice of reason and has
helped me see every setback as a stepping stone.
I have learned immensely from Richard Easterlin, Guofu Tan, Anant Nyshadham, and
other faculty members at the Department of Economics, University of Southern California,
and for that I am thankful. At various stages during my research, I benefited tremendously
from detailed discussions with Joseph Cummins, Matthew Kahn, and Juan Saavedra. I am
grateful for the institutional support from Young Miller, Morgan Ponder, and the University
of Southern California.
I thank Dawoon Jung and Manpreet Sohanpal for being constant companions; without
them this would not have been even half as much fun. I also acknowledge the immense
learning opportunities I had from my co-authors, friends and colleagues. Finally, I am
grateful to my family for their love and support.
ii
Table of Contents
Abstract i
Acknowledgments ii
List of Tables vi
List of Figures ix
Introduction 1
1 Recovery from an Early Life Shock through Improved Access to Schools:
Evidence from Indonesia. 5
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Previous Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Rainfall Shocks as Proxies for Income Shocks . . . . . . . . . . . . . 9
1.3.2 Primary School Construction Program . . . . . . . . . . . . . . . . . 11
1.3.3 Interaction between Early Life Rainfall shock and Primary School
Construction Program . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Conceptual Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5.1 Indonesian Family Life Survey (IFLS) . . . . . . . . . . . . . . . . . 19
1.5.2 Rainfall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.3 Primary School Construction Program . . . . . . . . . . . . . . . . . 22
iii
1.5.4 School Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.5.5 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.6 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.8 Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.8.1 Increased Competition and School Infrastructure . . . . . . . . . . . . 33
1.8.2 Other Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.8.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
1.8.4 Alternative Explanations . . . . . . . . . . . . . . . . . . . . . . . . 41
1.8.5 Other Outcomes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.9 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.10 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2 Is 1+1 more than 2? Joint Evaluation of Two Public Programs in Tanzania. 61
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.1 Iodine Supplementation Program (ISP) . . . . . . . . . . . . . . . . . 63
2.2.2 Primary Education Development Program (PEDP) . . . . . . . . . . . 66
2.2.3 Interaction of ISP and PEDP and the Question of Dynamic
Complementarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
2.3 Data and Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.3.2 Iodine Exposure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.3.3 Empirical Specification . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.4.1 School Grade Attainment and Primary School Starting Age . . . . . . 75
iv
2.4.2 Delay in Starting Primary School . . . . . . . . . . . . . . . . . . . . 78
2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.6 Tables and Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3 Coethnic Voters and Candidate Choice by Political Parties: Evidence from
India. 95
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.3 Equilibrium and Empirical Analogue . . . . . . . . . . . . . . . . . . . . . 103
3.4 Data and Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.4.1 Elections in Uttar Pradesh and Bihar . . . . . . . . . . . . . . . . . . 106
3.4.2 Pradhan Mantri Gram Sadak Yojana . . . . . . . . . . . . . . . . . . 109
3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.6 Revisiting the assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
Conclusion 126
Bibliography 129
A Appendix to Chapter 1 149
A.1 Proof of Proposition 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
A.2 Teacher Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
B Appendix to Chapter 2 172
B.1 ISP treatment definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
B.2 Alternative Definitions of ISP Exposure . . . . . . . . . . . . . . . . . . . . 174
v
List of Tables
1.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
1.2 Summary Statistics: Teacher characteristics . . . . . . . . . . . . . . . . . 56
1.3 Impact of rainfall and primary school construction on completed years of
schooling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
1.4 Impact of rainfall and primary school construction (fully interacted model) 58
1.5 Impact of rainfall and primary school construction on cognition . . . . . . 59
1.6 Impact of rainfall on control cohorts (1957-1961) . . . . . . . . . . . . . . 59
1.7 Primary school construction and pupil-teacher ratio . . . . . . . . . . . . . 60
1.8 Impact on school completion . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.2 Impact of Iodine Supplementation Program on completed years of schooling 91
2.3 Impact of ISP and PEDP on completed years of schooling . . . . . . . . . . 91
2.4 Impact of ISP and PEDP on primary school starting age . . . . . . . . . . . 92
2.5 Conversion of an additional year into additional years of schooling . . . . . 92
2.6 Impact of ISP on height of the child (Height-for-age) . . . . . . . . . . . . 93
2.7 Within household impacts of ISP . . . . . . . . . . . . . . . . . . . . . . . 93
2.8 Impact of ISP and PEDP on hours worked . . . . . . . . . . . . . . . . . . 94
3.1 Co-ethnic preference and strategic choice of candidates. . . . . . . . . . . . 97
3.2 Parliamentary Elections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.3 Assembly Elections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
vi
3.4 PMGSY and Candidate Selection . . . . . . . . . . . . . . . . . . . . . . . 124
3.5 Criminal candidates in PE . . . . . . . . . . . . . . . . . . . . . . . . . . 124
3.6 Criminal candidates in AE . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.7 Incumbents in Parliamentary Elections . . . . . . . . . . . . . . . . . . . . 125
A1 Grade conversion rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
A2 Type of school adminstration . . . . . . . . . . . . . . . . . . . . . . . . . 162
A3 Primary school construction and teacher qualifications . . . . . . . . . . . 163
A4 Primary school construction and teacher characteristics . . . . . . . . . . . 163
A5 Primary school construction and other crude teacher characteristics . . . . 164
A6 Interaction impact on years of schooling by year of birth . . . . . . . . . . 164
A7 Alternative definitions of rainfall shock . . . . . . . . . . . . . . . . . . . . 165
A8 Robustness with trimmed extreme rainfall observations . . . . . . . . . . . 166
A9 Alternative radius for weighted rainfall . . . . . . . . . . . . . . . . . . . . 166
A10 Inclusion of rainfall shocks in later years . . . . . . . . . . . . . . . . . . . 167
A11 Robustness to different definitions of parental education . . . . . . . . . . . 168
A12 Robustness to specification changes . . . . . . . . . . . . . . . . . . . . . 169
A13 Worked while in school . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
A14 Impact on schooling by gender . . . . . . . . . . . . . . . . . . . . . . . . 170
A15 Impact of rainfall and primary school construction on health . . . . . . . . 170
A16 Impact on Employment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
A17 Impact on Consumption and Well-being . . . . . . . . . . . . . . . . . . . 171
B1 ISP Coverage Variation (from Field et al. (2009)) . . . . . . . . . . . . . . 172
B2 Probability of Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
B3 Robustness of ISP Exposure Definition . . . . . . . . . . . . . . . . . . . . 175
B4 Impact of ISP on Vaccinations . . . . . . . . . . . . . . . . . . . . . . . . 175
vii
B5 Across Household Impacts of ISP . . . . . . . . . . . . . . . . . . . . . . 176
B6 Primary School Starting Age as a Plausible Mechanism . . . . . . . . . . . 176
viii
List of Figures
1.1 Educational attainments across birth cohorts . . . . . . . . . . . . . . . . . . 50
1.2 Educational attainments across birth cohorts by INPRES intensity regions . . 51
1.3 Educational attainment and INPRES treatment intensity . . . . . . . . . . . 52
1.4 Pupil Teacher ratio in Primary School . . . . . . . . . . . . . . . . . . . . . 53
1.5 Pupil Teacher ratio in Secondary School . . . . . . . . . . . . . . . . . . . . 54
1.6 Coefficient of the interaction of dummy indicating high rain in the birth
month * INPRES treatment intensity * year of birth . . . . . . . . . . . . . . 54
2.1 Iodine Supplementation Program in Tanzania (from Field et al. (2009)) . . . 86
2.2 Trends in years of education and primary school strating age before treatments 87
2.3 Trend in completion of appropriate grade for age before treatments . . . . . 88
2.4 Trend in height before treatments . . . . . . . . . . . . . . . . . . . . . . . 89
A1 Time line - Recovery from early life shocks . . . . . . . . . . . . . . . . . . 157
A2 Rainfall shock variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
A3 Rainfall and agricultural in Indonesia . . . . . . . . . . . . . . . . . . . . . 159
A4 Rainfall and education in Indonesia . . . . . . . . . . . . . . . . . . . . . . 160
A5 Rainfall and education by INPRES intensity regions . . . . . . . . . . . . . 161
A6 New Primary School Teachers appointed (Jalal et al. (2009)) . . . . . . . . . 161
ix
Introduction
Since the seminal works of Schultz (1961) and Denison et al. (1962) that started the “human
investment revolution in economic thought” (Bowman (1966)), economists have become
increasingly interested in the process of human capital formation. Accumulation of human
capital is considered crucial for the diffusion of modern economic growth to the develop-
ing regions of the world (Counts (1931); Inkeles (1969); North (1973); Davis et al. (1971);
Rosenberg et al. (1986)). Higher levels of human capital are also associated with the evolu-
tion of better institutional infrastructure and are desirable in their own right (Pigou (1952);
Adelman (1975); Grant (1978), Streeten et al. (1981); Sen (1984); Easterlin (2009)). In
spite of large investments in education, educational attainments remain low in much of
the developing world. The dissertation examines the factors influencing the demand for
education in developing countries and the role of education and information in shaping
the political institutions that might feedback in the form of more efficient legislation and
execution of policies aimed at improving educational attainments.
The first chapter examines the role of public policy in helping individuals recover from
‘early life developmental insults’. Using information from the fifth wave of Indonesian
Family Life Survey, we show that the nation-wide INPRES primary school construction
program in Indonesia helped those negatively impacted by low rainfall in the first year of
their lives to recover completely from the education deficit that low rainfall would have
caused in the absence of the program. Next, individuals who did not experience low rain-
fall in their first year did not benefit from the school construction program. We present
supplementary evidence that suggests that this was as a result of deteriorating school in-
frastructure quality and increased competition to get into middle and high schools. Those
1
who did not experience the early life rainfall shock were more likely to go to school even
in the absence of the program and, therefore, suffered disproportionately from the quality
deterioration.
The second chapter examines the joint effect of two public policy programs in Tanza-
nia - the Iodine Supplementation Program (ISP) and the Primary Education Development
Program (PEDP). We evaluate the joint impact of two policies using information on a sam-
ple of 10 to 13 year-old school aged children from the Kagera region in Tanazania. We
find that individuals who received iodine supplementation in-utero but did not benefit form
the abolition of school fees under PEDP (ISP exposed group) started school later and had
completed less years of education by the time of the survey than those who did not benefit
from the supplementation or fee abolition (control group). Those who did not receive the
iodine supplementation but benefited from the fee abolition (PEDP exposed group) started
school earlier and had completed more years of than the control group individuals. Those
who received both the iodine supplementation and benefited from the fee abolition started
school the latest and had completed the least amount of schooling by the time of the survey.
Using the ratio of the impact of the two policies and their interaction on completed school-
ing and primary school starting age, we show that those who were exposed to ISP were
more productive in converting one extra year in school into completed years of schooling.
We interpret this as evidence in favor of dynamic complementarity between years in school
and iodine supplementation. We provide suggestive evidence that this delay in enrollment
for those exposed to ISP was due to their worse health and higher likelihood of working
before enrollment.
The first two chapters make a case for policy intervention to promote education. The
quality of political institutions can affect the efficient implementation of these policies.
Elected representatives can have a significant role to play in the implementation of such
policies and it is observed that the implementation is often fraught with corruption. An
2
interesting follow up question is why do voters in these regions elect corrupt and ineffi-
cient candidates to office. In the final chapter, I examine the role of co-ethnic voting in
the election of bad quality representatives who, allegedly, undermine the effectiveness of
such policies. Unlike previous studies that take a candidate’s ethnicity as exogenous, this
chapter introduces an asymmetric information candidate choice model ` a la Hotelling where
political parties internalize voters’ co-ethnic political preferences and choose candidates of
different ethnicity strategically to maximize their vote profits. Data from parliamentary and
assembly elections from two states in India, Bihar and Uttar Pradesh, supports the model
prediction, namely, that choice to field a candidate of a particular ethnicity depends both
on the ethnic composition of the constituency’s voters (demand volume) and the ethnic-
ity of other candidates running for office (competition). Next, representatives elected in
constituencies with open elections, where candidates from all ethnicities can contest elec-
tion and diversification along ethnic lines is possible and beneficial are more likely to have
a prior criminal record than elected representatives from constituencies where only can-
didates of a certain ethnicity can contest. However, conditional on the average criminal
involvement of the pool of candidates to choose from, there is no difference between the
two types of constituencies in terms of choosing a criminal candidate for office. This sug-
gests that the election of criminal candidates to offices is, in part, due to political parties
disregarding potential candidates’ involvement in crime in order to be able to diversify
along ethnic lines. However, this strategic candidate selection seems to be absent when
constituencies are smaller, voters are more educated, or benefit more from a nation-wide
road construction program. Since co-ethic voter preferences are often thought to be a result
of lack of information about the policy stances of contesting candidates, the last result is
hardly surprising. It also suggests the existence of a virtuous education-institutions cycle.
Together, the chapters verify that households respond to both changes in costs and ben-
efits and changes in the quality of schooling. Higher education might not always translate
3
into higher earnings due to low quality of education or low returns to education in the lo-
cal economy. As a result, a reduction in cost of schooling in itself might not always be
sufficient to secure an increase in living standards. However, quality education might have
non-pecuniary benefits in the long term in the form of better institutions. Better institutions,
in turn, might improve the efficiency of all government policies and contribute to overall
welfare. Therefore, the cost-benefit analysis of such education policies should take into
consideration the long-run rewards of these policies.
4
Chapter 1
Recovery from an Early Life Shock through Improved
Access to Schools: Evidence from Indonesia.
1
1.1 Introduction
Early life environment can have long lasting impact on health, ability, and well-being of
individuals (Almond et al. (2017)).
2
Two observations stand out. First, from diseases
to weather, from pollution to maternal stress, many dimensions of early life experiences
matter. And second, even mild and brief shocks can have sizable effects. A question that
then naturally follows is whether we can ameliorate the negative effect of such adverse
shocks with corrective investment.
3
Parental investments that follow an early life shock not only depend on parental pref-
erences and their access to information, investment technology, and resources, but they
also respond to the nature and severity of the shock. Therefore, comparing differences in
the endogenous levels of these investments by parents across children where some are af-
fected by early life shocks and other are not will generate biased estimates of the level of
remediation possible. Therefore, to identify the possibility and extent of remediation, we
require two sources of exogenous variation - exposure to an early life shock followed by
a corrective policy or investment affecting the same population. In other words, we need
1
Coauthored with Dawoon Jung and Seungwoo Chin.
2
Also, see Almond and Currie (2011) and Currie and Almond (2011). See Gillman (2005) and Heindel
and Vandenberg (2015) for recent reviews from epidemiology.
3
Since shielding or insuring all vulnerable individuals against shocks to these different dimensions of
early life environment is unfeasible, the question of mitigation is pertinent even from the standpoint of public
finance.
5
“... ‘lightning to strike’ twice: two identification strategies affecting the same cohort but at
adjacent developmental stages” (Almond and Mazumder (2013a)).
Against this background, we examine the extent to which the Indonesian primary school
construction (hereon, INPRES) program of the 1970s, that improved access to primary
schools, helped individuals overcome an early life resource shock as proxied by low rainfall
in the first year of life.
4
Combining individual level information on educational attainment,
time and place of birth and other demographics from the fifth wave of the Indonesian Fam-
ily Life Survey Wave (IFLS, 2014) with the district-month-year level rainfall and the dis-
trict level intensity of school construction data from Duflo (2001), we investigate the joint
effect of rainfall shocks in the first year after birth and of a large scale school construction
program, the INPRES program, on educational attainment of the exposed individuals in
Indonesia. Consistent with the findings in Maccini and Yang (2009) and Duflo (2001), we
find that both high rain in the first year of life and the primary school construction program
had positive effect on education separately. In terms of recovery from early life shock,
individual who experienced less than historical median level of rainfall in the first year of
life, but received the median number of schools in her district under the INPRES program,
recovered completely from the early life shock, catching up to those who experienced more
than the historical median level of rainfall in the first year of life (See figures 1.1 and 1.2).
Additionally, we find no impact of school construction for those who experienced good
rain in the first year of life. Evidence suggests that the school expansion may have been
accompanied by deteriorating school infrastructure and increased competition to get into
middle and high schools. We find that, even for cohorts not exposed to the INPRES school
construction program, above-median levels of rainfall in the first year after birth were asso-
ciated with higher probability of going to school and higher levels of education. Once the
4
In Indonesia, low rainfall is associated with low agricultural output and, therefore, lower household
incomes. See figure A3 and section 1.3 for details.
6
INPRES program was rolled out, the high rainfall individuals, who would have been more
likely to be in school even without the program, suffered more due to deteriorating quality
and increased competition than low rainfall individuals. Consistent with the deteriorating
quality argument, we find that as schools got more time to respond to the situation, the
difference in the impact of INPRES vanished.
Our paper contributes to the limited evidence on recovery from early life shocks. This
is an important result as it speaks optimistically to the problem of not being able to shield
all individuals from all kinds of mild early life shocks - nurturing can counteract differences
that result from nature playing dice. We underscore the need for joint evaluation of differ-
ent shocks affecting educational attainment - human capital attainments are determined not
only by all the investment made in the skill’s production process across the years but also
the interaction of these investments with each other. Our paper also underlines that exam-
ining or detecting dynamic complementarity in the production of human capital is difficult
without being fully able to specify and estimate the production function. The exogenous
natural shocks or policy intervention affecting investments in human capital might also af-
fect the levels of other complementary inputs or the production function itself.
5
On the
policy front, by identifying better the beneficiaries of large scale program like INPRES, we
contribute towards a more sophisticated cost-benefit analysis of such programs.
The remainder of the paper is organized as follows. Section 1.2 briefly summarizes the
previous literature on the topic of recovery. Section 1.3 provides a brief background of the
two ‘shocks’, previous evidence on their impacts on education and the plausible pathways
they propose. Section 1.4 presents a conceptual framework to understand how these shocks
might have interacted and to draw testable implications. Data and the empirical strategy
are discussed in section 1.6. We present the results in section 1.7 and discuss the possible
5
For example, in this study, the school construction not only improved access but also had general equi-
librium impacts on the quality of schooling infrastructure that, in turn, affected schooling attainments.
7
pathways in section 1.8. Section 1.9 concludes.
1.2 Previous Literature
Multiple studies exploit exogenous variations in public policy interventions to examine the
impact of specific early life investments on human capital outcomes later on.
6
Some stud-
ies find bigger impacts for the more disadvantaged sections of the population suggesting
that such policies might facilitate recovery from early life shocks.
7
However, recovery es-
timates obtained from such studies could be biased since unobservable factors could be
associated both with the endogenous gradients along these indicators of disadvantage and
the production of later life outcomes of interest.
8
Owing to the demanding identification requirements, the research on the topic using
two exogenous sources of variation to identify the extent of remediation is limited. Aguilar
and Vicarelli (2011) examine the extent to which Mexico’s conditional cash transfer Pro-
gresa program mitigated the negative impacts of rainfall shocks due to El Nino around birth.
They find that the negative effects of the rainfall shocks were not mitigated by Progresa in
the short term. Adhvaryu et al. (2015) use a similar design to look at educational outcomes
and at ages 12 to 21. In contrast to Aguilar and Vicarelli (2011), they find that Progresa
mitigated much of the harmful effects of rainfall shocks in the year of birth.
9
Rossin-Slater
6
See Chetty et al. (2011); Gould et al. (2011); Dahl and Lochner (2012); Bharadwaj et al. (2013); Black
et al. (2014), and Gertler et al. (2014), for example.
7
Almond et al. (2011); Løken et al. (2012); Hoynes et al. (2015); Hoynes et al. (2016); Aizer et al. (2016)
8
For example, assume that cognitive ability at birth is positively associated with the wealth of the family
an individual is born in. Next, consider a conditional cash transfer program that is observed to have higher
impact on the education of children from low income households. It is not clear whether this investment
allowed the low income household children to recover from the cognitive advantage or whether the cash
transfers shielded them further shocks that they would have otherwise experienced growing up due to the
lack of family resources to deal with such shocks. These cited studies, as a result, do not claim to have
estimated the remediation effect.
9
Our study is, perhaps, the closest to Adhvaryu et al. (2015). However, there are a couple of important
differences. First, while Progresa attempted to incentivize the demand, INPRES was a supply side interven-
tion. Also, households below a certain income level were eligible for the Progresa cash transfer program early
on. The impact of rainfall shock for these households are expected to be higher. Moreover, as pointed out
8
and W¨ ust (2015) examine a related question of how a nurse visit program implemented
between 1930s and 1950s in Denmark interacted with another public program that induced
changes in the quality of preschool attended. They find that positive impact of the improve-
ment in preschool quality only for children not exposed to the nurse visit program. They
hypothesize that this lack of effect of improvement in preschool quality for children ex-
posed to the nurse visit program was due to the fact that the two program provided similar
kind of services.
Overall, the evidence on recovery from early shocks, especially from developing coun-
tries where these shocks are more frequent and severe and the avoidance mechanism less
developed, is limited and somewhat conflicting. The few studies cited above are also data
constrained to be able to explain the mechanism behind the often negative interaction effect
observed. As we explain in the next section, the timing of the two exogenous shocks com-
bined with the timing and richness of the data we use provides an ideal setting to examine
the possibility of recovery.
1.3 Background
1.3.1 Rainfall Shocks as Proxies for Income Shocks
Agriculture is one of the most important sources of household income in Indonesia. Ac-
cording to World Bank indicators, during the 1970s, close to 65 percent of the country’s
labour force was engaged in agriculture, mostly in rice production.
10
Agriculture, and rice
production in particular, is highly dependent on the timing and amount of rain in Indonesia
(Kishore et al. (2000); Levine and Yang (2014)). The specific monsoon trajectory varies
across years and, as a result, the timing, intensity and duration of precipitation varies a lot
by Hoddinott and Skoufias (2004), vast majority of households eligible for the program actually did receive
benefits.
10
It is conceivable that this number was even higher during the late 1950s and 1960s.
9
across the different rice growing regions in the country in any year and across years within
a region. Planting of rice is done once a certain level of rainwater has accumulated in the
fields. Delayed planting in the main agricultural season leads to reduced crop yields not
only in that season but also reduces the secondary crop yields by delaying the harvesting
of primary season crops and planting of secondary crops. Levine and Yang (2014) find
that higher rainfall in Indonesia is associated with higher output of rice. Maccini and Yang
(2009) find that the benefits of rainfall do not diminish even at very high levels of rainfall.
Using information from world bank indicators, we plot the association between agricultural
production indicator and annual rainfall in figure A3. Consistent with Maccini and Yang
(2009), Levine and Yang (2014), and the studies they cite, the figure seems to suggest a
positive relationship between rainfall and agricultural produce. Due to its equatorial loca-
tion, other weather related factors such as temperature shows little variation within years
and across years.
According to statistics released by Indonesian Ministry of Agriculture, smallholder
farmers account for around 90 percent of Indonesia’s rice production. Given the high de-
pendence on rice for both household consumption and income, rainfall shocks are one of
the most important risk factors faced by households.
11
Within a household, availability of
food, nutrient composition of food, income availability for other consumption purposes,
time allocation for adults and children, general level of stress, etc., can all be influenced
greatly by rainfall shocks (Behrman and Deolalikar (1987); Bouis and Haddad (1992);
Blau et al. (1996); Rosenzweig and Udry (2014)). Pregnant women and young children
are particularly vulnerable to such shocks. As mentioned in the introduction, evidence in
favor of the ‘critical-period programming’ or ‘development origins of diseases’ hypothesis
is amassing from different disciplines. Rainfall in early years is one such shock.
12
Maccini
11
Rainfall shocks have been found to affect consumption in important ways in other parts of the world as
well (Fafchamps et al. (1998); Alderman et al. (2006); Gin´ e et al. (2008); Bobonis (2009)).
12
For example, see Rocha and Soares (2015) and Shah and Steinberg (2017) in addition to the literature
10
and Yang (2009) find that better rainfall was associated with better health, educational and
socio-economic outcomes for women born between 1953 and 1974. They argue that better
rain, through higher income, improves the ability of parents to provide better ‘nutrition,
medical inputs, and generally more nurturing environments for infant girls’.
13
As Mac-
cini and Yang (2009) acknowledge, other channels, like disease environment, availability
of portable water, that link rainfall to child health might exist. Cornwell and Inder (2015)
find that both the nutrition and disease-environment pathways are operational linking rain-
fall to health outcomes. Figure A4 presents a lowess plot between education attainment
of individuals and the quantity by which the rainfall in their first year of life differed from
the prior fifty year average rainfall in their district of birth and suggests that the positive
relationship holds for our sample even at very extreme levels of rainfall.
Rainfall shock in-utero serves as the early life shock to investment in our study that has
implications for educational attainments of the exposed individuals. We investigate whether
improved access to primary schools through the INPRES school construction program help
individuals recover from this early life shock and catch up to those who did not experience
this shock in early life.
1.3.2 Primary School Construction Program
In the 1970s, the Indonesian government sought to redistribute the country’s aggregate
gains from the oil booms. This was carried out according to detailed presidential instruc-
tions on the process to be followed, often refereed to as INPRES (Ravallion (1988)). Seko-
lah Dasar INPRES, the implementation of which started in 1973, was one of the flagship
programs to come out of these instructions. Between 1973-74 and 1978-79, more than
cited above
13
Better nutrition and medical inputs could lead to earlier start of schooling (Glewwe et al. (2001)) or less
days off school due to illness. It could also lead to better ability to fight infectious diseases. Better nutrition
is also known to improve the cognitive ability of children (Galler et al. (1983); Brown and Pollitt (1996);
Glewwe et al. (2001); Liu et al. (2003)).
11
61,800 primary schools were built in Indonesia, roughly about 2 schools per 1,000 children
enrolled in 1971. This was an attempt to increase primary enrollment rate for children in
the age group 7 to 12, which was around 69 percent in 1973. (Duflo (2001)). The program
was planned to target districts with low enrollment first but did not follow the instruction
perfectly. It was one of the fastest and largest primary school construction project in the
world (Bank (1990)). Along with the construction project, the government also supplied
textbooks and hired new teachers (Parinduri (2014)). The program was a success and pri-
mary school enrollment rate rose up to 91 percent in 1986 (Bank (1990)). Duflo (2001)
evaluates the program and finds an increase of around 0.2 extra years of schooling for
every school constructed per 1000 children.
The timing of INPRES school construction program relative to timing of the survey
makes it the perfect intervention to study to examine the possibility of recovery from early
life shock. Information on education attainments of the population of interest was collected
when these individuals were well over 40 years of age. Therefore, there is little possibility
of a change in their completed years of formal education now. It also gets around the puzzle
of finding initial effects that “fade out” by teenage but then reappear in adulthood observed
in much of the literature on early intervention programs (Currie and Almond (2011)).
1.3.3 Interaction between Early Life Rainfall shock and Primary
School Construction Program
Parental investment might have responded to the early life rainfall shock, especially if
the impact of these shocks was readily visible. However, whether parents reinforced or
compensated for these shocks depends not only on parental preference but also on their
access to resources, information and investment technology. Evidence from the developing
world finds parents reinforce initial advantages (Li et al. (2010); Yi et al. (2015); Adhvaryu
12
and Nyshadham (2016)). It is possible, therefore, that within households, high rainfall
individuals who received better nutrition due household’s improved access to resources in
their year of birth, and, presumably, had better ability, ability defined broadly to include any
cognitive or health advantage they had, were also given priority when it came to sending
children to school. However, it is not clear if this was because parents wished to maximize
total returns of their children or because they wanted to equalize outcomes but did not have
suitable investment technology with decent returns available to them. As Almond et al.
(2017) point out, even for parents who wish to compensate for the early life disadvantage,
if they cannot afford the cost of intervening to help a disadvantaged child but can afford
basic investment in a non-disabled sibling, the choice may be obvious.
As a result, a priori, it is not clear if improved access to primary schools would have
had higher impact on high rainfall individuals or low rainfall individuals. For example, if
due to parental preference or constraints, low rainfall individuals are less likely to be in
school, such a school construction program might disproportionately benefit low rainfall if
it drives them to enroll. However, if it does not increase enrollment for low rainfall chil-
dren, the aggregate benefits of reduced transportation costs will be higher for high rainfall
individuals. The effects are reversed if parents had been responding to the early life shock
by investing more in the education of low rainfall individuals. It is, therefore, an empirical
question whether the school construction program had higher or lower impact by rainfall
in early life.
1.4 Conceptual Framework
In this section, we model an individual’s decision about her optimal level of schooling.
Depending on the amount of rainfall in the first year of life, individuals belong to one
of two ‘ability’ types. If the rainfall in the first year of life of an individual exceeds the
13
historical median amount of rainfall in her district of birth, we assume that the individual
is of high ability. Those with less than historical median rainfall in the first year of life are
of low ability.
14
That is
f
L
;
H
g;
L
<
H
Once an individual’s type has been determined by nature, she lives for two periods.
In the first period, the child who consumes and goes to school, borrowing against future
income for expenses incurred. In the second period that starts after the individual finish
school, the adult works, earns, consumes, and pays off her borrowings from period 1. The
optimization problem can be formulated as
max
fc
ijk
;c
ij(k+1)
;S
ijk
g
U(c
ijk
;c
ij(k+1)
)
subject to y(S
ijk
) =c
ijk
+pc
ij(k+1)
+C(S
ijk
);
c
ijk
0;c
ij(k+1)
0;S
ijk
0:
(1.1)
wherec
ijk
andc
ij(k+1)
denote the consumption of individuali born in districtj in year
k in the first and the second period of life, respectively. Consumption in the first period
is the numeraire and in the second period costs p. We use S
ijk
to denote the amount of
schooling the individual decides to attain. y(S
ijk
) is the adult income of the individual in
year (k + 1) when she starts working.
Following Card (1994) and Duflo (2000), we assume that income is a linear function of
schooling and the cost of schooling is strictly convex in the level of schooling.
y(S
ijk
) =a +b
ijk
S
ijk
(1.2)
14
We use to denote an individual’s ability, where ability is defined broadly as an aggregation of all those
individual characteristics or abilities that have been found to be positively associated with the amount of
rainfall an individual experiences in the first year of her life (see section 1.3.1).
14
and
h(S
ijk
) =
1
2
(Z
jk
)S
2
ijk
(1.3)
The marginal returns to schooling,b
ijk
is allowed differ across districts and years and
across individuals of the same cohort in the same district. To capture the time cost element
of going to school, we assume that the cost depends onZ
jk
, the number of primary schools
in districtj when cohortk born in districtj goes to school. An increase in the number of
schools in districtj when individuals from cohortsk go to school reduces the time cost of
going to school. That is,
1
(Z
jk
;D
i
)< 0.
For returns to schooling, we model within ability group individual level heterogeneity
additively using an individual level random term with mean zero in the population of that
ability type:b
ijk
=b
jk
+e
i
. Next, we define regional returns to education as a linear func-
tion of supply and demand for workers with different levels of education. On the supply
side, marginal benefit of schooling for type depends on the average level of education of
working individuals from all cohorts of type in that district and on the quality of educa-
tion obtained.
15
The marginal returns to schooling should be inversely related to the aver-
age level of education of the workforce.An additional year of schooling will have a smaller
benefit in labor markets where everyone else in the labor force is highly educated than in a
labor market where every one has very low levels of schooling. We assume that there are
no big positive externalities of education, at least at the level of education that the aver-
age Indonesian worker had in 1970. At all levels of schooling, higher quality of education
should generate higher marginal returns from schooling. The demand side is summarized
inv
j
, a zero-mean random term representing regional economic circumstances that affect
individuals with ability. Therefore,
15
We make the simplifying assumption that the returns from education for an individual of type depends
on the average level of schooling of other individuals of type in the region and not on individuals with a
different level of ability to account for occupational sorting to some extent. Relaxing this assumption will not
change the results of the model.
15
b
jk
= 2
1
S
j
+
2
q
jk
+v
j
;
1
< 0;
2
> 0 (1.4)
whereS
j
denotes the average years of schooling across all individuals of ability in
districtj. q
jk
denotes the quality of learning in schools that might depend on factors like
pupil-teacher ratio, teacher quality, textbook availability in schools in districtj when cohort
k was in school. We allow for the effect of education quality to be different across ability
types. More specifically,
Assumption 1
2
H
>
2
L
> 0
That is, conditional on the level of schooling, quality of education matters more for the
high ability individuals than for the low ability individuals. This implies complementarity
between ability and education, an assumption that has often been made in models seeking
to explain schooling choices in developing countries (Rosenzweig and Zhang (2013); Pitt
et al. (2012)). The assumption derives its theoretical justification from the observation that
education levels drive occupational sorting and cognitive and non-cognitive abilities tend
to have a higher premium in white collar jobs.
Notice that the optimization problems facing the high and low ability types differ only in
their varied returns from education. Since this assumption will drive our result, it is crucial
that we test its validity. Assuming an interior solution to an individual’s optimization, the
optimal levels of schooling is
S

ijk
=
E
k
b
ijk
(Z
jk
)
(1.5)
The marginal returns to schooling,b
ijk
, has a random demand component,v
j
, that is
realized only in period 2 of an individual’s life. Therefore, when deciding on the level of
16
schooling in period 1, individuals have to form an expectation about the marginal returns
to education. Averaging (1.5) across all individuals of a particular type, we have
S

jk
=
E
k
b
jk
(Z
jk
)
=
E
k
(2
1
S
j
+
2
q
jk
+v
j
)
(Z
jk
)
(1.6)
and, conditional on the number of schools and quality of education,
sign(
S

jk

) = sign(

2

)> 0 (1.7)
That is, under assumption 1, individuals with high ability should be attaining more
schooling than those with low ability. Under the alternate assumption of
2
H

2
L
,
the educational attainment of high ability individuals should be less than or no different
from that of the low ability individuals. We test this empirically to verify the validity of
assumption 1.
Next, we investigate how the INPRES school construction program affected the optimal
schooling choices. Primary school construction under the INPRES program increased the
number of schools in a district,Z
jk
, and, therefore, decreased the transportation or opportu-
nity cost of schooling. For further simplification, following Duflo (2000), we assume that
there were only two cohorts in the population, those who were out of school in 1974 (not
exposed to INPRES) and those who began school in 1974 or later (exposed to INPRES).
Subscriptk takes value of 1 for the older cohort (not exposed) and 2 for the younger cohort
(exposed). The assumption implies
S
j
=
1
2
(S
j1
+S
j2
); "f
L
;
H
g
Combining this assumption with 1.5, we reach the following proposition.
Proposition 1 If the average quality of education improved or remained unaffected with
17
the increase in number of primary schools under INPRES primary school construction
program, under assumption 1, the following must have been true:
1. Exposed individuals of all ability types must have attained more schooling than those
not exposed.
dS
j2
dZ
j2
> 0
2. The increase in schooling attainment for the high ability exposed individuals must
have been higher than that for the exposed low ability individuals.
(
dS
j2
dZ
j2
)

> 0
For an intuitive understanding of the proof, note that the reduction in cost of school-
ing was unambiguously beneficial for all. If there was no deterioration in quality, the net
benefit was equal to the benefit from reduction in cost. If the quality improved with the
construction of schools, the benefit from the program would have been even higher. Since,
according to (1.7), individuals with high ability attended more school, or, alternative, were
in school more often even before the school construction program was conceived, the in-
crease in the net marginal benefit from schooling due to a reduction in the cost of schooling
and improvement in quality is higher for this group. Hence, the corresponding increase in
schooling must be higher for this group.
However, if quality deteriorates with school expansion, the net effect on change in
schooling attainment will depend on the relative magnitudes of the reduction in cost of ac-
cessing schools and the reduction in the marginal benefit of attending due to a reduction in
quality of education.
18
Corollary 1 If the average quality of education deteriorated with the increase in number
of primary schools under INPRES primary school construction program, under assumption
1:
1. Exposed individuals of all ability types might not have attained more schooling than
those not exposed.
dS
j2
dZ
j2
Q 0
2. The change in schooling attainment for the high ability exposed individuals might
not have been higher than that for the exposed low ability individuals.
(
dS
j2
dZ
j2
)

Q 0
In section 1.7, we use our empirical findings on the comparative statics laid out in
Proposition 1 and Corollary 1 to choose between the competing models.
1.5 Data
1.5.1 Indonesian Family Life Survey (IFLS)
For our main analysis, we use information on schooling attainments from the fifth wave of
Indonesian Family Life Survey (IFLS, 2014). IFLS is an on-going longitudinal household
survey of conducted by RAND corporation that began from 1993. The survey respondents
are representative of about 83 % of Indonesian population and contains over 50,000 indi-
viduals living in 13 of the 27 provinces in the country (Strauss et al. (2016)). It contains
information of a wide variety of topics at the individual, the household and the commu-
nity level. Individual level information includes information on health, education, employ-
19
ment, migration, etc. IFLS also collects some education related information retrospectively.
Whenever we make use of retrospective information, we use the information from the ear-
liest wave of IFLS in which a respondent responded to these questions to minimize the
errors in recall.
Duflo (2001) uses the the Intercensal Population Survey (SUPAS) of 1995 to examine
the impact of INPRES. There are at least three reasons why we do not use SUPAS 1995
for our main analysis. To understand better the various pathways via which INPRES and
rainfall shocks affected the exposed individuals, we wanted to explore the plausibility of
a wide variety of mechanisms. The relative richness of IFLS data permitted that. Second,
migration related issues can be studied much better using the panel nature of IFLS surveys.
IFLS notes the district of residence at birth, at age 12 and at the time of each of the IFLS
waves making tracking easier. Third, important covariates like parental education have a
very high number of missing observations in SUPAS. Month of birth is another important
variable not available in data from Duflo (2001). Also, since Duflo (2001) is interested
in the impact of the program on wages in adult life, she used the sample of males only.
It is, however, of interest to see how INPRES, an intervention that reduced transportation
cost amongst other things, affected the educational attainment of women. While the com-
prehensive nature of IFLS 5 allows us to test for the impact of the treatments on a variety
of outcomes, it comes at a price. The size of the sample we study is significantly smaller
than that used by Duflo (2001). Given our smaller sample, we will sometimes interpret
the findings based on the consistency in their magnitudes and signs and not the level of
significance alone.
20
1.5.2 Rainfall
We use University of Delaware Center for Climatic Research’s “Terrestrial Precipitation:
1900-2008 Gridded Monthly Time Series (1900-2008) (Version 2.01)” rainfall that uses
an algorithm based on the spherical version of Shepard’s distance-weighting method to
combine data from twenty nearby weather stations and come up with interpolated rainfall
figures for every 0.5 latitude by 0.5 longitude grid. We borrow the district level latitude-
longitude information file from Maccini and Yang (2009) to match the districts to monthly
rainfall information from all grids in a 200 kilometer radius around the district center. We
calculate a weighted average of rainfall across these grids weighing rainfall information
from each grid by the inverse of its distance from the district center. For every month of
every year, we calculate a district level median rainfall by using the month-district specific
rainfall in the previous 50 years. Next, we deduct the prior 50 year month-district specific
median rainfall from the rainfall in a particular month. We assume that if the rainfall in a
district is a particular month in a particular year is above the prior 50 year district month
median, it is a positive shock to income in that month. For each individual in our sample, we
aggregate this difference between the district-month-year specific rainfall and the district-
month specific median rainfall in the past 50 years for the 12 months after birth. Then, we
construct a dummy variable that takes value ‘1’ if this 12-month aggregate is greater than
0, ‘0’ otherwise. This serves as our main rainfall shock, rainfall shock in the year after
birth. We also construct rainfall shock variables for the nine months before pregnancy, the
nine months during pregnancy, the month of birth, and for each of the nine years form age
2 to 10 of the individual, similarly. We use these rainfall shocks in other periods of life as
controls and to check the robustness of our results.
21
1.5.3 Primary School Construction Program
The number of school constructed differed across kabupaten (Indonesian districts). We
borrow the information on intensity of school construction used in Duflo (2001). Intensity,
here, is the number of newly constructed schools per 1,000 children in the district. We
match the individual level information from IFLS 5 with the the school construction in-
tensity in the district in which an individual was born. We assume that an individual went
to school in the district where he was born. IFLS also collects information on place of
residence at age 12. Using this information, we find that around 93% of the individuals we
study were in their birth districts at age 12. In the appendix, we redo the analysis excluding
those who were not in their birth districts at age 12. Our results are not sensitive to this
omission.
While those born after 1968 got the full advantage of INPRES, given the late enrollment
in developing countries, it is possible that individuals who were born not long before 1968
might have also benefited partially from the primary school construction. For this reason,
our control group consists of individuals who were 13 to 17 years old in 1974 (born between
1957 and 1961) and our treatment group consists of those who were 1 to 5 in 1974 (born
between 1969 and 1973). We omit the cohorts in between as the extent of exposure to
INPRES for these cohorts is not known. The INPRES treatment variable is the product of
the district level intensity of school construction and the treatment status of the individual
based on the year of birth. The treatment variable takes value ‘0’ for all who were born in
or before 1961 and is equal to the intensity of the school construction in their district for
all those born in or after 1968.
22
1.5.4 School Infrastructure
To identify the pathways through which INPRES affected educational outcomes differently
for high and low rainfall individuals, we also make use of information from the Indonesian
Population Census (IPC) of 1971, 1980 and 1990, and the Intercensal Population Survey
(SUPAS) of 1976 and 1985.
16
One pathway that we will be particularly interested in is
the school infrastructure and quality. In order to utilize information about the number
of teachers, we employ a strategy similar to Behrman and Birdsall (1983). From each
survey, we extract the sample of individuals who report being a school teacher as their
primary occupation. Using the demographic information on these selected individuals,
we construct the school infrastructure quality measures. We calculate district-survey year
specific primary and secondary school pupil-teacher ratio from these surveys.
17
1.5.5 Summary Statistics
In table 1.1, we present the summary statistics by INPRES treatment status and by rainfall
in the first year after birth. For the INPRES control group cohorts, high rainfall individuals
complete higher years of school than low rainfall individuals. The difference, however,
is almost non-existent for those exposed to INPRES. The INPRES control group, as one
would expect, has a lower average educational attainment than the exposed group. Our
sample consists of almost an equal proportion of males and females. The individuals are
mostly from Javanese Muslims from rural area and their parents tend to have some edu-
cation. The exact number of completed years of schooling of an individual’s mother and
father is missing for much of our sample of interest. Therefore, throughout our analysis,
16
These survey data were drawn from IPUMS international. See Minnesota Population Center, Integrated
Public Use Microdata Series, International: Version 6.4 [Machine-readable database]. Minneapolis: Univer-
sity of Minnesota, 2015. The original producer of the survey data is Statistics Indonesia.
17
In addition, we check the robustness of our findings across datasets by repeating our analysis with the
data used by Duflo (2001), a sample of male individuals from SUPAS 1995.
23
we use separate dummy indicators for whether the mother and father of an individual had
any education, completed primary, completed middle school, completed high school, or
had some tertiary education. In the appendix, we show replacing these dummy indicators
with years of schooling even in the limited sample that we have this information for gives
us qualitatively similar results.
The historical monthly average rainfall across Indonesia is about 180 millimeters. How-
ever, there is plenty of variation across months and across geographical area. We categorize
individuals into high rain and low rain on the basis of rainfall amounts in the first year after
birth. As is clear from the table 1.1, this does not mean that high rainfall individuals also
experienced higher than average rain before and during pregnancy. It seems that there is a
fair degree of variation in each of these rainfall variables that is independent of the rainfall
variables in other periods. INPRES intensity is ‘0’ for the control group by construction
and is over two schools per 1,000 children on average for both high and low rain individ-
uals in the treatment group. The difference between the intensity in high and low intensity
region is around one extra school per 1,000 children.
1.6 Empirical Strategy
Our identification relies on the variation in INPRES intensity and rainfall in the first year
after birth at the district-year and district-year-month level. The independent impact of IN-
PRES is identified by comparing treatment and control cohorts from districts with varying
levels of intensity of school construction. The independent impact of rainfall in the first
year after birth is estimated by comparing the outcomes for high rainfall individuals and
low rainfall individuals across month-year of birth and district of birth. The interaction of
the two shocks is estimated by comparing individuals with varying levels of month-year-
district rainfall in the first year after birth from the control group with individuals with
24
varying levels month-year-district rainfall in the first year after birth from the treatment
group and subtracting from that the independent impact of the two treatments. The identi-
fication strategy is, basically, one of difference-in-difference-in-difference. More precisely,
we estimate the following:
S
ijmt
=+
1
R
jmt
+
2
I
j
T
t
+
3
R
jmt
I
j
T
t
+
X
ijmt
+
t
+
j
+
m
+
ijmt
(1.8)
whereR
jmt
is a dummy that takes value ‘1’ if the rainfall for monthm in yeart in a
kabupatenj is greater than the historical mean of rainfall of that season in that kabupaten,
‘0’ otherwise. I
j
represents the INPRES intensity in kabupatenj andT
t
represents if the
cohort born in yeart was exposed to the INPRES program. Those born between 1969 and
1973 form the treatment group for INPRES and those born between 1957 and 1961 form the
control group.
t
are year fixed effect,
j
are kabupaten fixed effect, and
m
is month fixed
effect. Kabupaten fixed effects absorb any unobservable time-invariant differences across
districts and account for endogenous program placement and intensity of INPRES across
kabupatens. X
ijmt
is a vector of controls that includes rainfall controls for nine months
before pregnancy, during pregnancy, type of residence (urban or rural), parental education,
gender dummy, a dummy each for whether the individual belongs to the majority ethnicity
and religion. In addition, following Duflo (2001), we include the interaction of the year of
birth with the population aged 5 to 14 in 1971, the enrollment rate in the district in 1971
and the allocation of the water and sanitation program, the second largest INPRES program
at the time, to account for omitted time-varying and region- specific effects correlated with
the school construction program. Standard errors are clustered at the province level.
Our main outcome variable of interest is completed years of schooling.
1
captures the
effect of high rainfall in the first year after birth for individuals not exposed to INPRES,
2
25
is the the effect of INPRES program for low rainfall individuals. To ascertain the extent of
recovery, we need to compare the positive effect of INPRES on those who were exposed
to low rain in the first year of their life (a negative shock) but benefited from INPRES to
the positive effect of high rain for those who were not exposed to INPRES later. That is,
at pointed out in A1, the extent of recovery is to be deduced from the relative magnitudes
of
1
and
2
. If at the median level of rain in the first year of life and the median level of
school construction intensity,
1
=
2
, it implies that for a median individuals, INPRES
brought about full recovery from the early life shock of low rainfall. The coefficient of
the interaction term,
3
, captures the heterogeneous effect of INPRES by rainfall in the
first year after birth. The impact of higher than average levels of rainfall in the first year
after birth for those exposed to INPRES is the sum of
1
and
3
. Similarly, the impact of
INPRES for high rainfall individuals is given by the sum of
2
and
3
. A positive
3
is
an indication of synergies between the two shocks at different stages in life. On the other
hand, a negative
3
will suggest that low rainfall individuals individuals made better use of
the opportunities provided by the INPRES program than high rainfall individuals.
As described in section 1.4, we test for the validity of assumption 1 by empirically
verifying equation (1.5). For the purpose, we employ a specification similar to equation
(1.8) but without the independent variables and controls related to INPRES school con-
struction program. Our outcome variable of interest here is completed years of schooling
and other educational attainment indicators. Our sample for this analysis will be limited
to pre-INPRES cohorts - those born between 1957 and 1961. A positive
1
would mean
that those with high rainfall in their first year of life attended more school even before the
INPRES program was conceived.
26
1.7 Results
We begin by examining the effect of above median rainfall in the first year of life and the
INPRES primary school construction program, separately. The results are presented in
table 1.3. Column (1) presents the impact of above median rainfall in the first yearof life.
Those who experience above median rainfall in the first year after birth attain 0.26 extra
years of schooling. Our estimates are very close to those reported by Maccini and Yang
(2009). The mean level of rainfall in the first year after birth is 2160.97 mm and 2682.46
mm for low rainfall and high rainfall individuals, respectively. Therefore, a movement from
the category of below median rainfall in year one to above median rainfall in the first year
of life roughly amounts to a little over 20% increase in rainfall, on average. Maccini and
Yang (2009) find a gain of 0.22 years of schooling for a 20% increase in rainfall in the first
year after birth. Column (2) presents the impact of one more school construction per 1,000
children under the INPRES program. As expected, school construction led to an increase
in the educational attainment of the exposed cohorts. One more school per 1,000 children
increased an individual’s years of schooling by 0.18 years on average. Even though the
estimate is not statistically significant, the magnitude is strikingly close to that reported
by Duflo (2001).
18
Duflo (2001) reports an increase of 0.19 years of schooling for each
school built per 1000 children. As is clear from column (3), including both the rainfall
shock and the INPRES school construction variable in the same specification leaves the
results virtually unchanged, suggesting that the variation in rainfall shock is orthogonal to
the variation in intensity of the school construction program.
In column (4), we examine the joint impact of above average rainfall in the first year af-
ter birth indicator, INPRES intensity and their interaction. The independent impacts of the
two treatments are significant and their magnitudes are around two to three times the mag-
18
It is worth reiterating that Duflo (2001) used only the male sample from different data set for her analysis.
27
nitudes in column (1) and (2). The interaction, too, is significant. Those who experienced
high rainfall in the first year of life did considerably better than those in the omitted cate-
gory with a negative rainfall shock and no INPRES program treatment; they attened 0.73
extra years of schooling, on average. Some of those who experienced low levels of rainfall
in the first year of life were later exposed to the INPRES school construction program.
Compared to the omitted category, they completed 0.37 more years of schooling for every
school constructed per thousand children. Taking into account that the median number of
schools per 1000 children in the district in our sample was 1.9, this meant 0.70 extra years
of schooling for the median child affected by the negative rainfall shock in early life but
exposed to the INPRES school construction program later on. Comparing this magnitude
with the impact of above median rainfall in the first year of life, improved access to primary
schools due to the INPRES primary school construction program facilitated full recovery
from the education deficit created by low rainfall in the first year of life.

3
captures the heterogeneity in the effect of INPRES school construction program by
rainfall in the first year of life. Table 1 also reports the the impact of INPRES school con-
struction program on those who had did not experience the early life shock (
2
+
3
= 0:01)
and the F-stat for the joint significance
2
and
3
. For ease of comprehension, in column
(1) of table 1.4, we present the results from column (4) of table 1.3 using a fully interacted
specification.
19
The impact of school construction on those who experienced above median
rainfall in the first year of life is very small in magnitude and insignificant. Columns (2)-(4)
report how this result changes under specification changes to the regression specification
or sample.
20
The coefficient for the impact of school construction on high rainfall individ-
19
For this, we modify our specification in (1.8) in the following way:S
ijmt
=+
1
R
jmt
+
2
(1
R
jmt
)I
j
T
t
+
3
R
jmt
I
j
T
t
+
X
ijmt
+
t
+
j
+
m
+
ijmt
. Here,
1
,
2
, and
3
are the impacts of
above median rainfall in the first year of life, school construction program for those who experienced below
median rainfall in the first year of life, and school construction program for those who experienced above
median rainfall in the first year of life, respectively.
20
The robustness of the results to most of these specification changes are explored in detail in section
1.8.3.
28
uals remains virtually unchanged across specifications. One can safely conclude that the
INPRES school construction had no impact on those who did not experience the early life
shock.
Figure 1.3 depicts the relationship between years of schooling and INPRES school
construction intensity after controlling for all the variables discussed in section 1.6. The
figure reiterates the observation of figure 1.1 and 1.2 and table 1.3. INPRES helped low
rainfall individuals to recover from the adverse experience but had no effect for high rainfall
individuals. As a result of the no impact of INPRES school construction program on those
who did not experience the early life shock, those who experienced the early life shock
not only recovered but also caught up to those who did not experience the shock. To
see this, note that the joint impact of above median rainfall in the first year of life and
the median number of schools constructed per 1000 children in the district is 0.77 extra
years of schooling (0.73+ (0.37-0.35) * 1.9 = 0.77). Since the impact of above median
rainfall in the first year of life for those who were not exposed to INPRES later was 0.73
extra years of school, the average difference between the educational attainment of those
who experienced higher than median level of rainfall in the first year of life and those
who had lower than median rainfall in the first year and were not exposed to INPRES, the
omitted category in our specification, is a weighted average of 0.77 and 0.73 extra years of
schooling. However, as explained above, the independent impact of INPRES program for
the median child is 0.70 extra years of schooling, suggesting that those who experienced
the negative early life shock but were exposed to the INPRES program caught up to those
who did not experience the early life shock.
Such a catchup effect has been observed earlier in Indonesia and elsewhere along di-
mensions of education and health. It is consistent with the findings of Adhvaryu et al.
(2015) about Mexico and of Rossin-Slater and W¨ ust (2015) about Denmark. Mani (2012)
used earlier rounds of IFLS survey to track cohorts who were 3 months to 6 years old
29
in 1993 over the next seven years and found that there is partial recovery from effects of
childhood malnutrition on adolescence height for children who benefited from health care
services in Indonesia. Alderman et al. (2006) and Hoddinott and Kinsey (2001) find some
evidence of catch up from the adverse effect of droughts in Zimbabwe.
21
Against the back-
drop of these findings from different countries and time periods, the results from table 1.3
suggest that recovery from early life shocks may be possible, at least along the dimension
of educational attainments, provided that the right investments are made at the right time.
At first glance, the negative interaction effect, or, alternatively, the zero impact of IN-
PRES school construction program on those who did not experience the early life rainfall
shock is puzzling. If better rainfall in the first year of life did improve ability, why did
the more able individuals fail to benefit from the school construction program? However,
given the developments that accompanied INPRES, some of which we discuss in the next
section, the lack of an effect of INPRES for those who experienced above median rainfall
in the first year of life is not surprising.
1.8 Mechanisms
According to the model of optimal schooling level choice from section 1.4, the impacts of
schooling on children with high rainfall in the first year of life is ambiguous if the quality
of schools were deteriorating with the school expansion. The results from the model rest
on two crucial assumptions - rain in the first year of life created differences in ability and
the returns from quality of schooling were higher for high ability individuals. We begin by
verifying the validity of these two assumptions.
21
Much of the previous literature, including Alderman et al. (2006), Hoddinott and Kinsey (2001), and
Mani (2012), uses the term ‘catch-up’ to describe a specific observation - a decrease in the difference by
which an individual or a group of individuals lag behind others along a dimension of interest across successive
waves of a longitudinal survey. Our study uses the term to describe the reduction in difference between the
average educational attainment of children with relatively high and with relatively low levels of rainfall,
across cohorts.
30
In table 1.5, we examine the impact of above median rain in the first year of life, the
school construction program, and their interaction on a series of indicators that are related
with the cognitive ability of individuals. Wave 5 of IFLS included a serial sevens subtrac-
tion test where the respondent was asked to serially subtract 7’s from 100. Serial sevens
subtraction is used as a quick and easy test of concentration and memory (Manning (1982)).
Both a failure to answer a subsequent subtraction question and an incorrect answer are indi-
cators of worse performance. Column (1) reports the results for the number of subsequent
subtraction questions a respondent answers. Column (2) reports the number of question
an individual gets right. For both these outcomes, relatively higher rain in the first year
of life is associated with better performance. Another test of the memory dimension of
cognition is the word recall test. In IFLS, 10 nouns are read slowly to respondents and
then the respondents is asked to repeat back the list, once immediately after the list is read
and a second time around 4 to 5 minutes later. Columns (3) and (4) present the results for
immediate and delayed word recall, respectively. The results are consistent with those for
serial seven tests. Across all these outcomes, high rainfall in the first year of life appears
to be associated with better cognitive functioning. We return to the impact of the school
construction program and its interaction with high rainfall on cognition later in this section.
The rate at which individuals convert each year in school into completed years of
schooling can be thought of another measure of ability that affects educational attainment.
IFLS collects retrospective information on the year in which an individuals started and
graduated or dropped out of primary school. Using this information, we calculate the years
an individuals spent in primary school. The ratio of completed grades to the number of
years spent in primary school serve as our measure of conversion, an indicator of ability.
The results are reported in table A1. In columns (1) and (2) we report that the age of entry
and exit does not respond significantly to the rainfall shock, the school construction pro-
gram, or their interaction. The conversion rate in primary school, however, is significantly
31
higher for those who experienced relatively high rain in the first year of their life, suggest-
ing their superior ability to covert years in school into completed years of schooling.
To verify the assumption of higher returns to quality of schooling for high ability indi-
viduals, we employ the strategy laid out in (1.7) of section 1.4. The source of asymmetry
in the model is the differential returns from the quality of schooling for the two ability
types. Under assumption 1, this implies higher educational attainments for those with high
rainfall in the first year of life even before the school construction program was conceived.
In table 1.6, we examine the impact of rain only on individuals from the control cohort -
those born between 1957 and 1961. We find that being exposed to above median levels of
rainfall in the first year after birth was associated with higher probability of enrolling for
primary, completing primary, middle and high school, and attaining higher years of educa-
tion. High rainfall individuals were more likely to be in school than low rainfall individuals
even before the INPRES program. This is consistent with (1.7) and serves as an indirect
test for assumption 1.
The two crucial assumptions of the model, therefore, appear to be valid. According to
corollary 1, the zero impact of INPRES school construction is plausible if the quality of
education deteriorated with the rapid rate of school construction. Evidence suggest that
the primary school expansion might have been accompanied by a deterioration of primary
school, middle and high school infrastructure. Individuals with low rainfall in the first year
of life were less likely to be in school before the program. Due the school construction
program, they transitioned from not being in school to being in a school, albeit of lower
quality. For them the school construction program was an unambiguously positive develop-
ment. The high rainfall children who would have otherwise been in a good quality school
now went to schools that were, perhaps, closer to their homes but, as we show below, of
worse quality
22
Therefore, whether the school construction program was, on net, a positive
22
Data on which school these individuals went to is difficult to find. However, using retrospective data
32
development for them is ambiguous.
Additionally, the number of middle and high schools, as we show below, did not grow at
the same pace as primary schools and primary school graduates. This created a bottleneck.
The increased competition hurt the group of high rainfall individuals, the group otherwise
more likely to enroll for middle and high school, disproportionately. While pre-INPRES the
competition to get into middle and high school was mainly between high rainfall individu-
als, post INPRES these individuals also faced competition from low rainfall individuals. As
a result, a smaller proportion of high rainfall individuals who had completed primary school
now got into middle and high school. Even for those who got in, the quality of middle and
high school was worse given the increased pressure on the existing infrastructure.
In the following sections, we substantiate these claims with information from 1971,
1980, and 1990 waves of the Indonesian Population Census (IPC) and the 1976 and 1985
waves of the Indonesian Intercensal Population Survey (SUPAS).
1.8.1 Increased Competition and School Infrastructure
Figure 1.4 presents the primary school pupil teacher ratio in the survey years from 1971
until 1990.
23
There was a marked increase in the primary school pupil teacher ratio between
1971 and 1976, after which it came down. The line presents the primary school pupil
teacher ratio for each year between 1971 and 1990 as per the data collected by UNESCO.
While the primary school pupil teacher ratios calculated from the IPC and the SUPAS
from IFLS on the type of administration of schools that ran the school individuals went to, we find some
evidence that individuals did go to the public school more often after INPRES. Results from table A2 show
that more individuals went to public non-religious schools post INPRES and this appears to not be specific
to either the high rain individuals or low rain individuals. Unlike Bold et al. (2014) who find that children
of more affluent families, in fear of deterioration of quality in government schools, responded to the Kenyan
policy of abolishing all fee in all government primary schools by moving to private schools, we find no
evidence of such a movement in response to INPRES.
23
Card and Krueger (1992) use pupil-teacher ratio as a measure of quality and find that a decrease in
pupil-teacher ratio from 30 to 25 in public schools in the United States is associated with a 0.4 percentage
point increase in the rate of return to education.
33
waves are slightly higher than that from UNESCO data, the trend over time is similar in
data from both these sources. Another observation that stands out is the rapid decline in the
pupil-teacher ratio between 1976 and 1980 to levels below that of 1971. This is consistent
with the rapid hiring of teachers around this period (figure A6). It is conceivable that the
teacher quality suffered due to this rapid hiring. We find weak evidence of deterioration in
the qualities of teachers, measured by there educational attainment and whether or not they
taught in the region of their birth. The results are discussed in appendix section A.2.
According to figure 1.5, the number of secondary school teachers could not keep pace
with the numbers of primary school graduates and secondary school enrollments.
24
The
pupil-teacher ratio in secondary school (depicted by the blue bars) increased from less
than 20 in 1971 to over 40 in 1980. And unlike the primary school pupil-teacher ratio,
it did not fall rapidly below its 1971 level soon after. Also, not all those who completed
primary school could enroll for secondary school. Assuming that a fixed proportion of
those who were enrolled in primary school would have otherwise gone on to enroll for
secondary school, we look at the ratio of primary school students to secondary school
teachers (depicted by the red bars). The results are more pronounced. The ratio of primary
school students to secondary school teachers more than doubled between 1971 and 1980.
While the both these pupil-teacher ratios came down over the following decade, in 1990
they were still higher than their level in 1971.
However, these pictures aggregate the information from districts with low and high
intensity INPRES school construction. What is of interest is whether the schooling infras-
tructure deteriorated more in districts where the intensity of school construction program
was higher. We compare the pupil teacher ratio across low and high intensity INPRES
districts before and after the INPRES program, controlling for initial differences between
24
Unfortunately, the data on teachers in IPC and SUPAS does not categorize them into middle and high
school teachers. Instead, they are aggregated as secondary school teacher.
34
the regions and differences in these outcomes across year of the survey. The results are
presented in table 1.7. The primary school pupil teacher ratio is not significantly different
across intensity regions. This is consistent with the rapid hiring of primary school teachers
for the newly constructed primary schools. However, the secondary school pupil teacher
ratio and the ratio of primary school pupil to secondary school teachers is higher in the high
intensity INPRES regions than in the low intensity regions post INPRES. This suggests that
the secondary school infrastructure could not keep pace with the expanding number of pri-
mary schools and primary school graduates leading to increased competition to get into
middle and high school and worse infrastructure for those who got in.
As an additional check, in table 1.8, we examine the effect of above median rainfall in
the first year of life, INPRES school construction program and their interaction on com-
pletion of primary and middle school. Above median levels of rainfall in the first year
was positively associated with the probability of completion of primary and middle school.
Improved access to primary schools due to the primary school construction program had a
strong and positive impact on the probability of primary school completion but not so much
on the probability of middle school completion. This is as expected since there was an in-
crease in primary schools but no proportionate increase in the number of middle schools.
The negative interaction effect, however, was significant even at the level of middle school.
This is consistent with there being a congestion and crowding at the middle school level.
Taken together, these results suggest that individuals who experienced high rainfall in the
first year of life and would have been in school more often even in the absence of INPRES
might have suffered due to the deterioration in school infrastructure along these dimen-
sions.
25
25
This is consistent with the finding from Bound et al. (2010) who find that that the increased pressure
on collegiate resources was the main factor behind a decline in college completion rates in the United States
between 1970-1990. This is also consistent with the findings of Currie and Thomas (2000) for the United
States that effect of the Head Start program, a pre-school education program for age between three and five,
faded out for children who went on to attend worse quality schools.
35
Also, as is clear from the findings, some of these infrastructure quality indicators im-
proved over time. Schools, with more time to hire and fire, might have been able to ensure a
higher number of teachers. More middle and high schools might have been established due
to the increased demand. Assuming that the deterioration explains the negative interaction
coefficient, the recovery of these quality indicators over time should imply that the magni-
tude of the interaction coefficient must be smaller for the later exposed cohorts than for the
earlier exposed cohorts. That is, the difference in the impact of INPRES between individu-
als born in above and below average levels of rainfall in the first year after birth should have
diminished with time. If the negative interaction effect was due to some other factors, there
is no apriori reason to believe that the negative interaction should fade away. To test for
this, we estimate the interaction effect separately for each exposed cohort. The estimated
interaction effected coefficients are plotted in figure 1.6. There is a clear trend towards a
zero interaction effect. The interaction effect did not persist suggesting that deterioration
in primary school teacher quality and secondary school infrastructure might indeed be the
reason behind lower gains from INPRES school construction program for those who did
not experience the early life low rainfall shock.
26
1.8.2 Other Mechanisms
There is a possibility that the school construction program had no benefit for those with
relatively higher rainfall in the first year of life because most of these individuals were
completing primary school even before the program. There was, therefore, not enough
scope for improvement due to improved access to primary schools. The mean rate of pri-
mary school completion before the program, reported in column (3) of table 1.6, was 62%.
For those with relatively high rainfall in the first year of life, the primary school completion
rate was 65%. Therefore, there appears to have been a reasonable margin for improvement
26
The regression results are presented in table A6.
36
due to the school construction program. The impact of high rainfall, the school construc-
tion program, and their interaction on gross rates of primary school enrollment further adds
to the evidence against the hypothesis . If the was indeed not a big margin for the school
construction program to affect the education decision, that would have been more visible
in a lack of impact of the school construction on the primary school enrollment decision.
In column (1) of table 1.8, the impact of school construction on the enrollment of both low
rainfall individuals(
2
) and high rainfall individuals (
2
+
3
) is insignificant. That there
was not enough scope of improvement here is a possibility. The interaction of high rain
and INPRES school construction, however, is negative and significant, and its magnitude
double that of the impact of the impact of school construction on the enrollment of low
rainfall individuals. This suggests that the impact of school construction on primary school
enrollment of high rainfall individuals was about as large as the impact on low rainfall in-
dividuals but in the opposite direction. There is no reason why the impact of the school
construction program would have been different for the two groups and, specifically, nega-
tive for the high rain individuals unless, as in the model, the quality deterioration affected
them differently. The no-scope-for-improvement hypothesis is also inconsistent with the
trend observed in figure 1.6. Of course, we cannot completely rule out the possibility that
the lack of impact of school construction on high rainfall individuals is due to a lack of
scope of improvement. But evidence suggests it is less likely to be an explanation, let alone
the only explanation.
Another mechanism that might have been operational is that through peer effect. Before
the school construction program, classes were smaller and, on average, the high rainfall
individuals, who were, most likely, of higher ability, formed the majority of the class.
After the school construction program, there was an increased proportion of low-ability
low rainfall individuals in the class room. The deterioration of peer quality, especially in
the absence of good educational infrastructure, might also have affected the educational
37
outcome of the high rainfall individuals. This would be consistent with figure 1.6. With
improvement in school infrastructure, the negative effect of deterioration in peer quality
might have gotten weaker. Moreover, if school infrastructure improved the peer quality
of low rainfall individuals over time, the negative impact of poor peer quality would have
vanished. Unfortunately, we do not have the relevant data to test this mechanism explicitly,
and, therefore, cannot rule out this possibility.
1.8.3 Robustness
The definition of the rainfall shock variable that we use for our main analysis assumes more
rain is better for the educational attainment of the individuals exposed to it. We verify this
in appendix table A7. We repeat the analysis using different definitions of rainfall shock.
In column (1), we reproduce our results from our main specification. In column (2), the
rainfall shock dummy takes vale ‘1’ if the rainfall in the month of birth in the district of
birth of an individuals was equal to or higher than the historical 75th percentile in that
district. That is, the rainfall treatment variable is a positive shock of extremely high levels
of rainfall. The estimated coefficient have signs consistent with our main specification.
But the impact of extreme rainfall, as expected, is bigger than that of rainfall levels above
median, suggesting very high rain is even better than moderately high rain. Defining the
rain shock variable in terms of above or below historical mean does not change the results.
When we define good rainfall dummy to take value ‘1’ if it is more than one standard
deviation above mean, the impact of the good rainfall, INPRES school construction, and
their interaction are all smaller in magnitude and significance. However, the signs and the
relative magnitude of the effects stay consistent.
In our main specification, we assume that the relationship between rainfall and our
outcomes is monotonic, at least the the median. However, extremely high rainfall can lead
38
to flooding that might affect educational outcomes. To ensure that the our findings are not
being driven by extreme rainfall observations, in table A8, we repeat our analysis after
trimming out extreme observations. The results remain remarkably similar even with large
reductions in sample size. We use the inverse distance weighted rainfall observations from
all 0.5 by 0.5 latitude-longitude grids that lie within a circle of radius 200 kilometers to
define our rainfall shock variable. In table A9, we check the robustness of our findings to
changes in the radius of the circle from within which we include the rainfall observations.
As is clear form the table, the results are robust to changes in the radius. Even extreme
values for the radius, like 50 and 500 kilometers, generate qualitatively similar results.
The impact of rainfall in the first year of life could, arguably, through its effect on
incomes and savings, to the time when a child enters school. To rule this out, in table A10,
in addition to rainfall in the first year of life, in the month of birth, during pregnancy, and
in the nine months preceding pregnancy, we control for rainfall in years two through five.
The results remain unaffected by the inclusion of rainfall in these years. Moreover, the
impact of rainfall in any of the other years is not significant or large enough relative to the
impact of rainfall in the first year of life (not presented here). This suggests that the impact
of rainfall in the first year of life is through nutrition and cognitive development.
In table 1.3, we control for mother’s education since it could be correlated with the de-
gree to which parents are able to shield their children from the negative effects of a rainfall
shock. For example, mother’s education has been observed to be systematically associ-
ated with the health outcomes of children (Horton (1986); Barrera (1990); Strauss (1990);
Thomas et al. (1990); Thomas et al. (1991); Thomas (1994)). In all our specifications
we include a dummy each for whether or not the mother of the child had any education,
completed primary school, completed middle school, completed high school, or had some
tertiary level education. We report these estimates separately in A11 and compare it with
different definitions of mother education that have been used in the literature before. Con-
39
sistent with earlier findings, education level of the mother of a child has a large significant
impact on the years of education completed. Whether we use dummies for some educa-
tion versus none or we use separate dummies for different levels of education as we do in
our main specifications does not make much difference. We do not have exact years of
schooling information for the mothers of much of our sample. However, column (3) we
show robustness to using completed years of schooling instead of these dummy indicators
for different levels of schooling. The results are qualitatively similar even with only forty
percent of the original sample. Also, the higher mean of dependent variable suggests that
the sample of individuals for whom we have their mother’s completed years of schooling
is a selected sample.
In table A12, we check the robustness of our findings to other specification changes. In
column (1), we reproduce our results from our main specification, where we assume that
children went to school in the district they were born. IFLS collects information on the
city that an individual resided in at the age of 12 for older cohorts. We find that 93 % of
our sample resided at age 12 in the same city in which they were born. It is safe to assume
that the percentage of individuals from our sample who went to primary school in their
city of birth was close to this number, if not higher. However, in column (2), we redo the
analysis with the sample of only those individuals who were living in their birth district at
the age of 12. The results suggest that our findings are not sensitive to this assumption.
In column (3), we replace the separate fixed effects in month and year that we employ in
our main specification with a month-year fixed effect. Our results are somewhat smaller
but qualitatively similar to those in column (1). In column (4), we cluster our results at the
province level instead of the district level. Our results remain unchanged in terms of their
significance. In column (5), we redo the analysis for the rural sample. The dependence on
agriculture is higher in rural Indonesia. As a result, the population from rural communities
are more vulnerable to rainfall shocks. Consistent with this, we find that the impact of
40
above median rainfall is much higher for the rural sample. The impact of INPRES school
construction and the interaction are also higher for the rural sample. The signs and relative
magnitudes of the three effects are consistent with those in column (4) of table 1.3.
There could be a concern that those who experienced high rainfall in the first year of
life might be more likely to survive until 2014 due to their better health. If that is the
case, our estimates might be prone to survival bias. If relatively low levels of rainfall
was indeed associated health and with survival until 2015, those who did survive must
have been the healthiest among the those who experienced low rainfall in the first year of
life. The survival bias would have been smaller for the sample of high rainfall individuals.
Since health and education are often positively related we must, therefore, be comparing
only those with the highest levels of education among the low rain individuals with a more
representative sample of high rainfall individuals. That would imply that our estimates on
the impact of rain are biased downwards, if at all. Moreover, as we discuss ahead, relatively
high rainfall in the first year of life did not give rise to a significant difference in the physical
health of those who exposed and those who were not.
1.8.4 Alternative Explanations
In this section, we examine some of the other possible explanations for the results we
observe in column (4) of table 1.3. For example, recall that INPRES intensity was higher
in poorer, educationally worse off regions. Could it be that more rain meant different things
for high and low INPRES intensity regions? For example, it could be that higher rain in
high INPRES region meant that more family members were required to work on fields in
high intensity regions. This can imply lower child care, and, therefore, worse outcomes
for with high rain in their first year of life and high inpres intensity in their districts. And
that, the negative interaction coefficient reflects this. The first piece of evidence against
41
this conjecture comes from figure 1.2. As is clear from the figure, across both low and high
intensity regions, high rain in the first year of life was associated with higher educational
attainments later in life and INPRES helped those with low rainfall in the first year of life
to catch up. Therefore, it seems unlikely that high rain in the first year of life had different
implications across these regions. Second, to test this explicitly, we examine the impact
on more rain in high intensity and low intensity districts for control cohorts of 1957-1961
in figure A5. As is clear from the figure, while high INPRES regions were indeed wore
off along educational outcomes before INPRES, more rain in the first year of life was
associated with more education across high and low intensity regions. The impact of very
high rain is, as expected, much larger in poorer, high intensity regions.
Another possibility could be that even though high rain in the first year of life has
similar impacts across regions in terms of education, it also meant more work on the field
or at home for high rainfall children due to their higher ability. Later when INPRES schools
were built, low rainfall individuals were sent to school more often because they were,
presumably, not as productive as high rainfall children. This decrease in labor supply
further increased the opportunity cost of going to school for the more productive, high
rainfall individuals and they chose to stay out of schools and work. Hence, the negative
interaction effect. While the primary school enrollment rates were not different across
different subgroups within the sample, one could argue that such a mechanism will not
show up in the enrollment rates if children are combining work with school as they do in
other parts of the developing world. To test this, we examine the impact of high rain in the
first year of life, INPRES school program and their interaction on whether children worked
while in school. While this analysis uses retrospective data about their working status
around 40 years ago, there is no reason to believe that the recall bias will be differential by
treatment status. The results are presented in appendix table A13. We find no evidence of
differences in the propensity of working between the different subgroups.
42
Yet another possibility is that suggested by Pitt et al. (2012). Men and women in devel-
oping countries sort into different occupations - men often take up jobs that reward brawn
and women find a comparative advantage in the sector that rewards education. The in-
vestments made even early on in the human capital of men and women often reflect this
structure of the economy. Investments in health often improved the earnings of men and
education of women while policy interventions attempting to improve education resulted in
better labor market returns for women (Miguel and Kremer (2004); Maluccio et al. (2009)).
The coefficients in table 1.3 are, perhaps, somehow identified off selected sections of the
population such that they reflect a mix of these gender specific impact. To test this, in
appendix table A14, we redo the analysis by gender. Consistent with Pitt et al. (2012),
we find that the impact of these shocks are bigger and more significant for the educational
attainment of women. However, the general direction of these impacts are same for both
males and females, suggesting that the gender level occupational sorting is not driving the
results.
1.8.5 Other Outcomes
A natural question that follows is whether the school construction program facilitated re-
covery along other dimensions of well-being. We examine three relevant dimensions -
cognition, physical health, and employment.
Consistent with previous literature on the topic and as discussed above, the findings
reported in table 1.5 suggest that relatively higher rain in the first year of life affected the
dimensions of cognitive ability captured by the outcomes in the table, positively. Improved
access to schools did not help low rainfall individuals recover from the cognition deficit rain
in the first year of life created. All the coefficients for the impact of school construction
program on cognitive outcomes of the low rainfall individuals are insignificant and have
43
relatively small magnitudes. This is consistent with the findings of Cunha et al. (2010) that
the elasticity of substitution for inputs in cognitive skill is substantially lower in later years
of a child’s life. Additionally, it would not be surprising that if the extra schooling failed to
improve the cognition of exposed individuals due to, as discussed above, its low quality.
The interaction effects of relatively high rainfall and the school construction program,
even though small in magnitudes, are sometimes significant. These negative coefficients,
that have signs consistent with the findings in table 1.3, suggest that school quality might
have been a relevant input into the production function of cognition even in later years. This
is not at odds with the findings in Cunha et al. (2010). There is no a priori reason to expect
that improved access to school, controlling for quality, might have an effect non-cognitive
outcomes. However, a deterioration in quality of school might have a negative effect on
non-cognitive skill formation, and through cross-productivity effect in the cognitive skill
production function, on the outcomes examined in 1.5.
The results for physical health, presented in table A15, are somewhat mixed. We look
at four indicators of later life health - length of upper arm, average lung capacity, domi-
nant hand grip, and investigator’s evaluation of respondent’s health. Adult height has often
been used as a proxy for early-life health and resource availability (See, for example, Fogel
(1994); Brunner et al. (1996); Strauss and Thomas (1998)). However, height shrinkage in
adults starts as early as forty years of age, especially under poor later life health conditions.
To distinguish the effect of early life conditions from later life conditions, we use the length
of the upper arm that does not shrink with age and is a better measure of maximum height
for older respondents (Huang et al. (2013)). Estimated coefficients for rainfall in the first
year of life, the school construction program, and their interaction are all in the right direc-
tion but not significant. IFLS also measures lung capacity using a peak flow meter. Lung
capacity is a correlate of later life health and functioning limitation. For our sample, above
median rain in the first year of life is associated with higher lung capacity. The school
44
construction program helps those who did not have high rain in the first year of life recover
from their disadvantage. Poor muscle strength is another predictor of later life functional
limitations and disability (Rantanen et al. (1999)). IFLS collects hand grip strength, an in-
dicator of muscle strenghth, using a dynamometer. The impacts of the high rain, the school
construction program, and their interaction on dominant hand grip strength are consistent
with the main results on education in their direction but not significant. The same is true for
investigator’s evaluation of the respondent’s health. It is important to note that the positive
impacts of school construction on most of these outcomes for the median individuals (scale
the coefficients by a factor of 1.9) are more significant and bigger in magnitude than the
independent impact of high rainfall in the first year of life. The results suggests that health,
as measured by these outcomes, was not significantly affected by rainfall in the first of year
life. If it was, improved access undo the disadvantage that rain generated. The impact of
education could have been through better access to information, better health practices, or
through income.
The results for employment related outcomes, presented in table A16, while mostly
consistent in their direction, are weak and somewhat uninformative.
27
From column (1),
high rain is associated with lower probability of working in 2014. However, from column
(2) and (3), it is associated with a higher probability of having worked in 2008 and 2007.
The impacts of the school construction program and the interaction are in the mostly in the
right direction but are not statistically significant. Outcomes related to occupational sorting,
intensive margin of labor supply, whether or not the individual has a pension plan, and
satisfaction with the job all have signs consistent with the possibility that the impacts were
mostly on the quality of employment. However, the coefficients are mostly insignificant.
The lack of a strong and consistent impact of rainfall in the first year on employment
makes question of recovery due to improved access to schools redundant. The results
27
Not reported here, the results are not different by gender.
45
for consumption expenditure and overall measures of well-being in table A17 too have
consistent signs but are not significant.
It is, of course, a possibility that we are unable to detect any effect on employment
related outcomes due to a small sample size and measurement errors in these outcomes.
The lack of strong and consistent impacts on employment, however, are consistent with
findings elsewhere. Duflo (2004), in a medium term evaluation of the school construction
program, finds that the school construction program might have depressed wages due to the
inability of physical capital to adjust to the increasing levels of human capital. Evaluating
the medium run impact of the PROGRESA conditional cash transfer program in Mexico,
Behrman et al. (2011) find a negative impact on the employment of young boys, no impact
on the employment of young girls and older boys, and an increase in employment of older
girls, suggesting that the impact of such educational policies could be very specific by age,
gender, and, perhaps, other characteristics. In Indonesia, propensity to work for a wage
is also greatly affected by the general economic conditions. For example, women have
been found to join the labor force to supplement household income in difficult times and to
exit wage work for household and childcare related work when times are good (Franken-
berg et al. (2003); Schaner and Das (2016)). The intra-household work allocation and the
brain-brawn based occupational sorting complicates the impact of education on employ-
ment further. Education could also impact employment decisions in a more fundamental
manner by affecting preferences that govern the labor-leisure trade-off and retirement de-
cisions (Becker and Mulligan (1997)). The lack of impact on employment could also be
a result of a ‘learning crisis’ - a situation where due to the low quality of education being
imparted in schools, individuals learn little even when they complete school more often
(Bold et al. (2017a)). With the limited evidence on the employment outcomes, it is difficult
to deduce anything conclusively. We leave this to future research on the topic.
46
1.9 Conclusion
Shocks early in life have long lasting impacts. This is an even bigger concern in devel-
oping countries where these shocks are often more severe and frequent and the ability of
parents to shield their children from such shocks or to make corrective or compensatory
investments limited. Most of the corrective investments are, therefore, carried out by the
state through public policies. It is, therefore, of immense importance from a policy stand-
point to understand if the negative effect of these early life shocks can be mitigated and to
what extent does this compensation work. Against this backdrop, we leverage two exoge-
nous treatments, an early life shock in the form of access to resources proxied by level of
rainfall in the first year of life and a policy aimed at improving access to primary schools,
to examine the possibility and extent of remediation. Our encouraging findings suggest
that almost complete recovery is possible, at least along the dimension of education attain-
ments, if adequate investment crucial for mitigating the negative effect of such shocks are
made. Using information from wave 5 of the IFLS, we document that the negative effect of
early life rainfall shock were mitigated completely if individuals received improved access
to primary school during childhood.
We show that while recovery was driven by the positive effect of improved access on
the disadvantaged individuals, the catch-up was partly a result of the inability of individu-
als who did not experience the early life shock to benefit from improved access to schools.
We provide suggestive evidence from IPC and SUPAS that this resulted from deteriorating
school infrastructure and teacher quality and increased competition that hurt individuals
who did not experience the early life shock disproportionately. Consistent with the obser-
vation of quality indicators improving over time, we find that the difference in the impact
of INPRES for high and low rain individuals vanishes over time.
Considering that the immediate goal of INPRES was to increase primary school enroll-
47
ment, it was, arguably, a big success. However, our analysis provides a good exposition
of some of the major trade-offs and challenges faced by almost all policies implemented at
such a large scale.
28
Often such immediate goals are set in pursuit of bigger goals of a better
educated populace and economic development. INPRES is redistributive in more than one
way in that it increased the educational attainment for one group (those who started school
because of INPRES) at, perhaps, a small cost to another group (those who were already in
schools). While such redistributions, given the overall gain, might still be well justified and
even desired, understanding the heterogeneity in its impact is beneficial for future policy
planning. It is important to realize that the benefits of the intervention could have been
bigger had it been accompanied by other improvements in infrastructure. Proper personnel
training and simultaneous expansion of middle and secondary school infrastructure might
have mitigated the negative effect of deteriorating quality and infrastructure on children
already in school and would have resulted in higher gains from INPRES.
Our paper, however, leaves a few important bigger issues unanswered. First, the reme-
diation we find is along the dimension of educational attainment. A question that future
research should focus on is to what extent did INPRES correct the possible negative effect
of early life rain shock along other dimensions of well being, such as health and cogni-
tion. Second, we cannot be sure that the corrective effect of improved access to schools
would hold if it benefited individuals at a different point in their life. For example, one
would expect the effect to be smaller for a high school construction program since dis-
advantaged individuals would have dropped out disproportionately before entering high
school. This is of immense importance from a policy perspective. If remediation efforts
early on in life are more effective than those later on, it might be a pareto improvement to
transfer resources from policies that increase investment in later years to those that focus
28
Concerns about the deteriorating quality was expressed even during the midwives training program in
Indonesia (Hatt et al. (2007); Makowiecka et al. (2008); Ronsmans et al. (2009)). It is plausible that quality
did suffer even during the INPRES school expansion program.
48
on early years. Third, the results cannot be generalized to other intervention policies and
contexts. While improved access to schools worked in Indonesia in the 1970s, it might not
have worked towards recovery from the harmful effects of being exposed to nuclear fallout
radiation on cognition in Sweden. In fact, the findings from these study should motivate
the examination of the remediation effect of various investment in a variety of setting. Fu-
ture research should aim towards developing an understanding of investments that work
in varied settings against a large number of early life negative shock as this will greatly
streamline policy aimed at tackling such disadvantages.
49
1.10 Tables and Figures
y
i,high
= α + g(yob
i,high
) + ε
i,high
y
i,low
= a + f(yob
i,low
) + e
i,low
6 7 8 9
Lowess - Years of schooling
1955
1960
INPRES
1965
1970
1975
Year of birth
High rain Low rain
FIGURE 1.1: EDUCATIONAL ATTAINMENTS ACROSS BIRTH COHORTS
50
6 7 8 9 10
Lowess - Years of schooling
1955
1960
INPRES
1965
1970
1975
Year of birth
High Rain High Inpres Low Rain High Inpres
High Rain Low Inpres Low Rain Low Inpres
FIGURE 1.2: EDUCATIONAL ATTAINMENTS ACROSS BIRTH COHORTS BY INPRES INTENSITY
REGIONS
51
y_res
i,high
= α + g(I_res
i,high
) + ε
i,high
y_res
i,low
= a + f(I_res
i,low
) + e
i,low
-4 -2 0 2
Lowess - Years of schooling (residual)
-2 -1 0 1
Number of schools per 1000 children (residual)
High rain Low rain
FIGURE 1.3: EDUCATIONAL ATTAINMENT AND INPRES TREATMENT INTENSITY
52
20 25 30 35 40 45
Primary School Pupil Teacher Ratio
1970 1975 1980 1985 1990
Survey year dummy
Primary School Pupil Teacher Ratio (UNESCO)
Primary School Pupil Teacher Ratio (IPUMS)
FIGURE 1.4: PUPIL TEACHER RATIO IN PRIMARY SCHOOL
53
0 50 100 150 200
Secondary School Pupil Teacher Ratio
1971 1976 1980 1985 1990
Secondary School Students: Secondary School Teachers
Primary School Students: Secondary School Teachers
FIGURE 1.5: PUPIL TEACHER RATIO IN SECONDARY SCHOOL
-1.5 -1 -.5 0 .5
Interaction coefficient (Dependent Var - Years of schooling)
1969* 1970* 1971* 1972* 1973*
Year of birth
FIGURE 1.6: COEFFICIENT OF THE INTERACTION OF DUMMY INDICATING HIGH RAIN IN THE
BIRTH MONTH * INPRES TREATMENT INTENSITY * YEAR OF BIRTH
54
TABLE 1.1: SUMMARY STATISTICS
Not exposed to INPRES Exposed to INPRES
Low Rain High Rain Low Rain High Rain
V ARIABLES Mean S.D. Mean S.D. Mean S.D. Mean S.D.
Demographics
N 551 540 796 1,205
Years of Education 6.626 4.506 7.150 4.500 9.397 3.919 9.399 3.935
% Enrolled in Primary School 0.960 0.978 0.992 0.993
% Completed Primary School 0.586 0.639 0.881 0.883
% Completed Middle School 0.323 0.350 0.613 0.611
% Completed High School 0.207 0.241 0.445 0.425
Gender (% Male) 0.474 0.481 0.511 0.503
Father has any education 0.711 0.769 0.813 0.813
Mother has any education 0.519 0.563 0.691 0.675
% Urban 0.281 0.293 0.260 0.300
Religion (% Muslim) 0.853 0.902 0.896 0.899
Ethnicity (% Javanese) 0.457 0.493 0.413 0.507
Rain (unit: mm)
Rain 9 months prior to pregnancy 1,769 540.6 1,971 463.9 1,983 463.3 1,745 576.0
Rain during pregnancy 1,834 482.0 1,990 507.5 1,983 425.3 1,749 566.2
Rain in the month of birth 156.6 124.7 202.0 127.7 189.0 130.8 193.3 130.7
Rain in the first year after birth 2,102 439.6 2,765 468.9 2,202 448.8 2,645 456.2
Rain in the second year after birth 2,263 489.8 2,483 517.5 2,626 450.7 2,472 486.2
Rain in the third year after birth 2,103 489.9 2,440 527.7 2,516 487.4 2,478 510.9
Rain in the fourth year after birth 2,143 542.3 2,299 460.5 2,560 432.5 2,434 561.5
Rain in the fifth year after birth 2,121 506.6 2,337 495.4 2,469 535.0 2,514 485.1
Rain in the sixth year after birth 2,139 476.5 2,275 517.1 2,525 511.3 2,520 536.5
Rain in the seventh year after birth 2,132 566.6 2,429 542.2 2,661 536.3 2,343 551.6
Rain in the eighth year after birth 2,444 671.8 2,313 483.2 2,507 604.0 2,520 499.4
Rain in the ninth year after birth 2,266 563.9 2,533 588.4 2,512 514.9 2,627 490.5
Rain in the tenth year after birth 2,371 498.3 2,602 580.2 2,680 485.8 2,370 507.5
INPRES Intensity (All regions)
N 47 53 67 48
Intensity 2.275 1.130 2.258 1.101
INPRES Intensity (Low intensity regions)
N 20 26 33 27
Intensity 0 0 1.626 0.559 1.850 1.054
INPRES Intensity (High Intensity regions)
N 27 27 34 21
Intensity 0 0 2.904 1.193 2.783 0.944
Notes: Intensity is the number of schools built per 1,000 students in the district.
55
TABLE 1.2: SUMMARY STATISTICS: TEACHER CHARACTERISTICS
Low Intensity Regions High Intensity Regions
V ARIABLES Mean S.D. Mean S.D.
Teacher Characteristics
N 22,159 19,096
Age 33.40 15.59 31.86 13.20
Gender (% Female) 0.473 0.395
Urban (%) 1.380 1.254
Completed High School (%) 0.845 0.860
Completed Diploma 4 (%) 0.083 0.051
Completed Diploma 3 (%) 0.047 0.028
Teaching in birth province (%) 0.890 0.916
Years spent in the current locality 29.79 28.96
Previous residence in a different province (%) 0.123 0.093
In a different province 5 years back (%) 0.025 0.035
Primary School characteristics
N 597 471
Primary School Pupil Teacher Ratio 29.28 21.65 31.57 24.74
Secondary School characteristics
N 523 385
Secondary School Pupil Teacher Ratio 28.99 23.96 35.60 30.00
56
TABLE 1.3: IMPACT OF RAINFALL AND PRIMARY SCHOOL CONSTRUCTION ON COMPLETED
YEARS OF SCHOOLING
(1) (2) (3) (4)
Years of Schooling
Above median rainfall in year 1 0.26* 0.25 0.73***
(0.15) (0.15) (0.22)
INPRES schools per 1000 children 0.18 0.18 0.37*
(0.20) (0.20) (0.21)
Above median rain * INPRES schools per 1000 children -0.35***
(0.11)
INPRES school per 1000 high-rain individuals 0.01
F-test p-value (0.97)
Median schools per districts 1.90 1.90 1.90
Mean of dependent variable 8.29 8.29 8.29 8.29
Observations 3,399 3,399 3,399 3,399
R-squared 0.40 0.40 0.40 0.41
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
57
TABLE 1.4: IMPACT OF RAINFALL AND PRIMARY SCHOOL CONSTRUCTION (FULLY
INTERACTED MODEL)
(1) (2) (3) (4)
Years of Schooling
Above median rainfall in year 1 0.73*** 0.59** 0.69*** 0.73***
(0.22) (0.24) (0.22) (0.24)
INPRES school per 1000 low-rain individuals 0.37* 0.32 0.38* 0.34
(0.21) (0.21) (0.21) (0.20)
INPRES school per 1000 high-rain individuals 0.01 0.03 0.02 0.01
(0.21) (0.21) (0.21) (0.21)
Median schools per disticts 1.90 1.90 1.90 1.90
Mean of DV 8.29 8.29 8.29 8.17
Specification change Main Month X Year FE Years 1-5 Non-movers
Observations 3,399 3,399 3,399 3,146
R-squared 0.41 0.43 0.41 0.41
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and
the water and sanitation program in the district during the time of school construction program. The specification used isS
ijmt
=
+
1
R
jmt
+
2
(1R
jmt
)I
j
Tt +
3
R
jmt
I
j
Tt +
X
ijmt
+t +
j
+m +
ijmt
. Here,
1
,
2
, and

3
, reported in the table, are the impacts of above median rainfall in the first year of life, school construction program for those who
experienced below median rainfall in the first year of life, and school construction program for those who experienced above median
rainfall in the first year of life, respectively.
58
TABLE 1.5: IMPACT OF RAINFALL AND PRIMARY SCHOOL CONSTRUCTION ON COGNITION
(1) (2) (3) (4)
Serial 7 questions Word recall
V ARIABLES answered correct Immediate Delayed
Above median rainfall in year 1 0.1507** 0.2777** 0.2294** 0.0583
(0.0706) (0.1190) (0.0896) (0.0963)
INPRES schools per 1000 children 0.0264 0.0858 0.0441 0.0780
(0.0506) (0.0804) (0.1153) (0.1078)
Above median rain * INPRES schools per 1000 children -0.0701** -0.0278 -0.1134** -0.0457
(0.0316) (0.0533) (0.0478) (0.0471)
INPRES school/1000 high-rain individual -0.04 0.06 -0.07 0.03
F-test p-value 0.40 0.52 0.51 0.77
Mean of DV 4.67 2.71 4.70 3.53
Observations 3,632 3,510 3,613 3,613
R-squared 0.1199 0.1537 0.2080 0.1825
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth, ethnicity, religion, urban community, rain during
pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had some
schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education. We also
control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the water and
sanitation program in the district during the time of school construction program.
TABLE 1.6: IMPACT OF RAINFALL ON CONTROL COHORTS (1957-1961)
(1) (2) (3) (4) (5)
Highest Enrolled Completed Completed Completed
V ARIABLES grade primary primary middle high
Above median rainfall in year 1 1.09*** 0.04*** 0.11*** 0.07* 0.06*
(0.34) (0.01) (0.03) (0.03) (0.04)
Mean of DV 6.61 0.96 0.62 0.39 0.26
Observations 1,287 1,439 1,439 1,439 1,439
R-squared 0.38 0.11 0.25 0.23 0.23
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother and the father of the
individuals had some schooling, completed primary school, completed middle school, completed high school, or had some tertiary
level education.
59
TABLE 1.7: PRIMARY SCHOOL CONSTRUCTION AND PUPIL-TEACHER RATIO
(1) (2) (3)
Pupil Teacher ratio in Primary pupil:
V ARIABLES Primary school Secondary school Secondary teachers
High intensity dummy * Survey 1976 -4.7522 6.6542 36.4285
(7.5060) (6.9226) (30.4713)
High intensity dummy * Survey 1980 -3.8968 10.3165** 98.7938***
(3.9178) (4.2582) (25.9944)
High intensity dummy * Survey 1985 -4.7114 9.1068** 34.6810**
(4.2985) (3.6759) (16.2590)
High intensity dummy * Survey 1990 -3.2665 6.1162 28.4008
(3.7086) (3.8529) (17.4169)
Mean of DV 30.29 31.79 111.46
Observations 1,068 908 908
R-squared 0.12 0.17 0.19
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses are clustered at the district level. Other controls
include survey year fixed effects, a dummy for gender of the teacher and a dummy for the type of residence, rural or urban.
TABLE 1.8: IMPACT ON SCHOOL COMPLETION
(1) (2) (3)
Enrolled in Completed
V ARIABLES Primary School Primary School Middle School
Above median rainfall in year 1 0.018** 0.062** 0.046*
(0.009) (0.024) (0.026)
INPRES schools per 1000 students 0.004 0.057*** 0.023
(0.005) (0.020) (0.027)
Above median rainfall in year 1 * INPRES schools/1000 -0.007* -0.031** -0.028**
(0.004) (0.012) (0.013)
INPRES school per 1000 high-rain individual -.002 0.03 -0.01
F-test p-value .47 0.25 0.85
Mean of DV .98 0.78 0.52
Observations 3,632 3,632 3,632
R-squared 0.080 0.230 0.254
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
60
Chapter 2
Is 1+1 more than 2? Joint Evaluation of Two Public
Programs in Tanzania.
1
2.1 Introduction
In-utero and early life investments in health and education have been found to have long run
impacts on educational attainments (Almond et al. (2017)). However, not much is known
about how early life investments in health and in education interact in determining the
education attainment of individuals. We further this discussion - we study the joint impact
of a health policy and an education policy in Tanzania on the educational attainments of
individual exposed.
Cunha and Heckman (2007) propose a theoretical framework that considers the forma-
tion of human capital as a dynamic process with the complementarity between investments
at different stages of life. They suggest that investments in human capital later in life
are likely to be more productive for individuals who receive higher investments in early
life than for those who received lower investments in early life. They term this ‘dynamic
complementarity’ in the production of human capital. A major empirical challenge in test-
ing the model of dynamic complementarity is the endogeneity of investments at different
points in life. As a result, the empirical evidence in favor of this model of complementarity
is limited. A small number of studies that evaluate the joint causal impacts of two public
programs or shocks affecting individuals at different stages in their lives fail to find any ev-
1
Coauthored with Dawoon Jung and Seungwoo Chin.
61
idence in favor of ‘dynamic complementarity’ (Adhvaryu et al. (2015); Rossin-Slater and
W¨ ust (2015); Gunnsteinsson et al. (2014); Malamud et al. (2016)). However, as we discuss
in detail ahead, it mostly due to their data or methodological limitations.
We use information from Kagera Health and Development Survey (KHDS) to examine
the joint effects of two public programs, the Iodine Supplementation Program (ISP) and
the Primary Education Development Program (PEDP), on the grade attainment of children
who were 10 to 13 year old at the time of the survey. We find that exposure to ISP, a health
policy that targeted pregnant women and, consequently, their in-utero babies, was nega-
tively associated with completed schooling of those exposed. PEDP abolished all primary
school fees and, as expected, had a positive impact on educational attainment. We also find
significant negative interaction effects of ISP and PEDP on educational attainments - indi-
viduals who were exposed to both ISP and PEDP had lower educational attainments by the
time of the survey than individuals not exposed to either or one of the two programs. We
show that the the two programs and their interaction have impacts on primary school start-
ing age that mirror their impacts on grade completed by 2004. ISP exposure is associated
with a delay in primary school starting age. PEDP had a negative effect on starting age.
The interaction of the two programs is positively correlated with starting age. The associ-
ation of the two programs and their interaction with educational attainments almost vanish
when we control for primary school starting age. Individuals who delay entering school
spend more time working on the family farm and doing household chores suggesting that
delayed entry is most likely due to positive returns from pre-school training (Bommier and
Lambert (2000)).
Next, the ratio of the impacts of the two programs and their interaction on educational
attainment by the time of the survey and on primary school starting age provides us with
a measure of the rate at which individuals convert years in school into completed years of
schooling. We find that in comparison to individuals exposed to PEDP but not ISP, those
62
exposed to ISP are better at converting years in school into completed years of schooling.
This is consistent with Field et al. (2009) claim that those exposed to ISP saw an improve-
ment in their cognition. This is also an indirect evidence of dynamic complementarity -
ISP exposure improves the productivity of each year spent in school.
This paper contributes to the literature in two ways. First, we re-evaluate the impact
of ISP on schooling attainments for individuals in Kagera region of Tanzania and present
results that help reconcile the divergent findings of two previous studies (Field et al. (2009)
and Bengtsson et al. (2013)) on the impact of ISP on schooling. We underline the impor-
tance of careful consideration of behavioral responses to ISP exposure to better understand
the reduced form impact of ISP. Second, we contribute to the methodological discussion
around identifying ‘dynamic complementarity’ by illustrating the need for more cautious
interpretation of the reduced magnitudes of the interaction of two exogenous shocks as
evidence for or against complementarity. Consequently, we use an alternative strategy to
present what we believe is the first piece of evidence in favor of ‘dynamic complementarity’
from developing countries.
The remainder of the paper is organized as follows. Section 2 provides a brief back-
ground of ISP and PEDP. Section 3 discusses the existing literature. Section 4 describes the
data and the empirical strategy. Section 5 presents the main results and robustness checks.
Section 6 concludes.
2.2 Background
2.2.1 Iodine Supplementation Program (ISP)
Lack of proper nutrient in utero or during early life is detrimental to the physical and cog-
nitive development of individuals (Barker (1990); Cao et al. (1994); Barker (1995); Barker
et al. (2002); Zimmermann et al. (2005)). One such important nutrient is iodine, essen-
63
tial for the synthesis of thyroid hormones. Adequate levels of these thyroid hormones in
pregnant mothers are important for physical and mental development of a fetus. Especially,
sufficient stock of iodine in a pregnant mother’s body in the first trimester of her pregnancy
is extremely crucial for the cognitive development of the baby in-utero. Brain development
during this period is sensitive to minor adjustments in thyroid hormone and mild maternal
iodine deficiency can impair the full cognitive development of an individual (Dugbartey
(1998); Pop et al. (1999)).
People from Tanzania, like those from many African countries, traditionally suffered
from high rates of Iodine Deficiency Disorder (IDD). According to a Tanzania nationwide
survey of iodine levels in the early 1970s, about 40% of the Tanzanian population lived in
iodine deficient areas and 25% of the population was estimated to have had IDD. The preva-
lence among pregnant and lactating women was as high as 52% (Van der Haar et al.,1988).
In response, beginning in 1986, the Tanzania Food and Nutrition Center (TFNC) started
distributing iodized oil capsules (IOC) to individuals in districts where more than 10% of
school children had some symptoms of goiter. The program, known as the Iodine Supple-
mentation Program (ISP), was expanded to 27 districts covering 7.3 million population by
1994 (Peterson et al. (1999)).
This program was one of the largest and most intensive iodine supplementation pro-
grams in the world (Peterson (2000)). The program was scheduled to begin in 1988 and
planned to distribute iodized oil capsules containing 400 mg of iodine amongst males and
females aged 2 to 45 years and a dose of 200 mg for children aged 12 to 23 months (Peter-
son et al. (1999)). However, the time line was not strictly followed. Four districts received
the supplementation in 1986 and only 10 had received it by 1988 while two districts did not
receive it until 1992. The coverage rate was never perfect in any district and the average
coverage rate was 64% (See Table A1). The delays in the program start date, in all cases,
were due to administrative issues arising from the logistical challenges of distributing IOC
64
throughout the district (Peterson (2000)). However, despite the delays, according to a con-
servative estimate from Peterson et al. (1999), the program protected 12 million individuals
from iodine deficiency (ID).
The program was considered a success, with several previous evaluations finding reduc-
tion in visible and total goiter rate (VGR, TGR) attributable to ISP (Peterson et al. (1999)).
In the early 1990s, a success of ISP led to the Universal Salt Iodization (USI) program
initiated by Tanzanian government. After the USI was introduced, ISP was used to com-
plement USI, focusing on districts not yet reached by the USI. Thus, during this period, the
absence of the ISP program in a district does not necessarily indicate that the individuals in
the district are unprotected from IDD.
Field et al. (2009) analyze the impact of ISP on grade attainment of the children of
treated pregnant mothers using the Tanzania Household Budget Survey 2000 (THBS 2000).
They find that ISP had significant positive impact on completed years of schooling of
treated children by the time of the survey. They find that treated children, who were still in
school at the time of the survey, had completed 0.345 more years of education. This, they
conjecture, must have been due to the improvement in cognition of those who received
the iodine supplementation. In contrast, using information from the 1999, 2004, and 2010
waves of Demographic and Health Survey, the 2008 wave of National Panel Survey, and the
2000 wave THBS, Bengtsson et al. (2013) find that the estimated impact of ISP on grade
attainment is not consistent across datasets and often negative in sign. The impact is pos-
itive and significant for only the THBS 2000 sample. They use a slightly different model
of iodine depletion over time than the one in Field et al. (2009) and find much smaller
magnitudes for the impact of ISP on grade attainments. They explore the robustness of
their findings across different definitions of ISP treatment and across different criteria for
selecting the sample and find no evidence of a significant consistently. They investigate
the attenuation bias that might result from selective migration and incomplete birth infor-
65
mation and conclude that these would not explain the differences between their results and
that in Field et al. (2009).
The main purpose of our study is to document the existence of dynamic complementar-
ity in the production of human capital and our evaluation of ISP is not directly comparable
to either to these studies. However, our findings help in reconciling the apparently contra-
dictory findings from Field et al. (2009) and Bengtsson et al. (2013).
2.2.2 Primary Education Development Program (PEDP)
Tanzania school system consists of seven years of primary school, four years of secondary
school and two years of upper secondary school. There are two national examinations
in primary school - one at the end of the 4th and another, primary school leaving exam
(PSLE), at the end of the 7th grade (Government Report, 2005). Students need to pass the
examination at the end of grade four to progress to grade 5 and the PSLE to advance to
secondary school. Children are expected to enroll in primary school at the age of seven
and complete primary school by the age of 13 (Ministry of Education and Culture, 1995).
However, delays in enrollment, dropouts and grade repetitions are common.
In January 2002, Tanzanian government launched the Primary Education Development
Program (PEDP) wherein tuition fees and other mandatory cash contributions to schools
were abolished (Tanzania Education Report, 2006). The primary purpose of PEDP was
to ensure the enrollment of all 7 to12 year olds by 2005. The net enrolment rate in pri-
mary school in the year preceding the launch of the Primary Education Development Plan
(PEDP) was less than 50%. The program began by targeting those who were seven to eight
years old in 2002, individuals born in 1993 or 1994. The coverage of the program was ex-
tended to 11 and 12 years old in 2004 (9 and 10 years old in 2002). However, the effort and
impact for these children was substantially lower and delayed. As a result, individuals born
66
in 1993 and 1994 were fully exposed to PEDP while those born before 1993 were partially
or never exposed to PEDP. Due to PEDP, net enrollment rates went up significantly from
66% in 2001 to 97.3% in 2007.
The program worked towards bringing down the cost of primary education by abolish-
ing all tuition fees. Moreover, a $10 capitation grant was also introduced and controlled
by school committees. This was intended to cover some of the additional school-based
costs. However, substantial indirect costs, such as an expense for instructional materials,
remained, the provision of which has not been sufficient to date.
2.2.3 Interaction of ISP and PEDP and the Question of Dynamic
Complementarity
If the ISP treated individuals indeed had higher cognitive ability than the untreated indi-
viduals, this could imply both a higher benefit and a higher opportunity cost of schooling
for the treated children. In the light of the possibility, the reduction of schooling cost due
to PEDP could potentially have differential impact on the ISP treated and untreated indi-
viduals. Since there was a fair degree of overlap between the two programs in terms of
the cohorts treated, a careful examination of interaction between these two programs is
warranted.
Of late, there has been a rising interest in studying interactions between two shocks
to human capital formation in developing countries. The primary motivation for studying
such interactions is to shed light on the production function of human capital in devel-
oping countries. In particular, many studies examine if inputs at different points in life
into the production of human capital exhibit any complementarity. However, as mentioned
before, endogeneity of the level of inputs at different points in life is a major empirical
challenge. A small number of studies have employed the ‘lightening strike twice’ (Almond
67
and Mazumder (2013b)) identification strategy - exogenous variation in the exposure to two
public programs or shocks that affect individuals at different stages in their lives (Adhvaryu
et al. (2015); Rossin-Slater and W¨ ust (2015); Gunnsteinsson et al. (2014); Malamud et al.
(2016)).
Adhvaryu et al. (2015) examine the interaction between early life adverse rainfall con-
ditions and Progresa conditional cash transfer in Mexico. They find that the conditional
cash transfer enabled individuals, who were otherwise lagging behind due to adverse rain-
fall shocks in early life, to catch up. Malamud et al. (2016) estimate the joint effect of
access to abortion facilities at the time of conception and access to better school later on.
Although they find no significant interaction effect, they acknowledge that their results do
not count as evidence against dynamic complementarity as behavioral responses by indi-
viduals or their parents between first intervention and second intervention could dampen
the joint effect.
In addition to assuming that the two sources of variations in investments are orthogo-
nal to each other, these studies also assume that an individual’s uptake or compliance or
avoidance or mitigation efforts in response to the second shock does not depend on her first
program’s treatment status. This second assumption might easily be violated. We use an
empirical specification similar to the ones in these studies to study the interactions between
the two programs, ISP and PEDP. However, we do not interpret the sign of the interaction
as evidence for or against dynamic complementarity. Instead, we use a comparison of the
ratio of the impact of ISP and PEDP on completed schooling by 2004 and primary school
starting age to shed light on the dynamic complementarity in the production of education.
68
2.3 Data and Empirical Strategy
2.3.1 Data
We use information from the Kagera Health and Development Survey (KHDS), a survey
representative of the population of the Kagera region from Tanzania. Located in the north-
western corner of Tanzania, Kagera is one of Tanzania’s 30 administrative regions. Kagera
is Tanzania’s fifteenth largest region and accounts for more than three percent of the coun-
try’s area (CIA (2010)). During 1980s, Kagera suffered from high rates of IDD. As a
result, four of its seven districts were targeted by ISP, the first one starting in 1989. Next,
the timeline of the KHDS survey waves overlap favorably with the high prevalence of IDD
and subsequent high intensity of ISP. KHDS households were originally interviewed in
four waves from 1991 to 1994. Follow-up surveys were then carried out in 2004 and 2010.
Therefore, it covers the implementation phase of both ISP and PEDP and still allows anal-
ysis of short and medium run outcomes like height and educational attainment. KHDS is
one of the longest-running African panel data set with an impressive tracking rate of around
90%. (Beegle et al. (2006a); De Weerdt et al. (2012)). Due to the overlap between the pro-
gram and the survey timelines, KHDS is a suitable survey to study the programs and their
impact. We use the 1991-1994 and 2004 waves for information on individual’s educational
attainments, primary school starting age, parental investments in the child and a variety of
covariates.
To calculate iodine exposure intensity, we use the district-year coverage rates from
Field et al. (2009). We match this coverage rate for each year for each district with the
corresponding observation from KHDS using the year and name of the district information
contained in KHDS. We follow Field et al. (2009) in their calculation of the probability that
an individual benefited from the supplementation. An individual’s probability exposure
69
depended on whether and when the iodine supplementation program was implemented in
the district of a district vis-` a-vis her mother’s first trimester of pregnancy. In that, we
make the assumption that mothers, through out their pregnancy, lived in the district where
they delivered their child. The details about the method followed are provided in the next
section. We restrict our sample to the cohorts born between 1991 and 1994. We do not
include cohorts born after 1994 because nation-wide iodine supplementation (USI) began
in late 1994. In addition, since PEDP (started in 2002) fully affected both 7 years-old and 8
years-old, cohorts born after 1992 were treated by PEDP and those born in 1991 and 1992
were not treated. We do not include the cohorts born before 1991 to avoid more serious
recall bias in the 1991-1994 waves and to balance then number of PEDP treatment and
control group across cohorts. Consequently, while the variation in ISP treatment is at the
level of cohorts and districts, the PEDP treatment varies only across cohorts. However,
since we make use of information from a small number of adjacent cohorts in a period
with no other major government program in the region, we believe that the bias due to
time-variant unobservable will be minimal.
The main reason we do not use information from other nationally representative surveys
like the Tanzanian Demography and Health Surveys (TDHS) or the Tanzania Household
and Budget Surveys (TBHS) is that the relevant waves from these surveys do not have
information on the district of birth of individuals. Internal migration across regions in Tan-
zania is common (Kudo (2015)). Kagera region is an ideal setting in this context because
migration outside Kagera is relatively low. Moreover, KHDS boasts of a high tracking rate
of individuals even when they move. The information allows us to restrict our attention to
individuals who report not having moved in the last ten years and report being a part of the
household in preceding years. This helps us minimize the attenuation bias from migration.
While children born in 1991 and 1992 were born twelve and eleven years before the 2004
wave, respectively, the probability that they moved in the first two years of their birth is
70
relatively small.
2.3.2 Iodine Exposure
As described before, sufficient levels of iodine are most crucial in the first trimester. There-
fore, the child of an iodine deficient mother who received an iodized oil capsule in the first
month of any year would not be protected unless the child was born in the eighth month
of that year or later. Following Peterson et al. (1999) and Field et al. (2009), we assume
that the timing of the distribution was uniform over the months of any year that the dis-
trict received the supplementation. We also maintain the assumption in Field et al. (2009)
that, conditional on the starting month, it took three months to complete the distribution of
these capsules. Therefore, for a district that received the supplementation program in the
first month of the year t, children born in the first seven months in that district were not
protected by the supplementation program. Research shows that the body stock of iodine
depletes at a certain rate after every such iodine supplementation. To account for this de-
pletion, we use the method used in Field et al. (2009). For those born in the eighth month
or later, protection, therefore, depended on whether the program started early enough to
have reached their mothers in time (first trimester or earlier) and whether their mothers had
retained adequate amounts of iodine throughout their first trimester after accounting for the
depletion of body iodine stocks with time.
The detailed table of probability of protection calculation is reproduced from Field et al.
(2009) in the appendix. For instance, we present here the calculation for those born in the
eighth, the ninth and the tenth month of year t. For those born in month 8, probability of
protection is equal to the probability that the program started in January that year (equal
to 1/12 using uniform timing assumption) and their mothers were reached in that very
month (equal to 1/3 using three-month diffusion time assumption.) Therefore, it is equal
71
to 1/36. For those born in the ninth month, the program reached their mother in the first
trimester if it started in January (1/12) and reached them by February (2/3) or if it started
in February (1/12) and reached them in February itself (1/3), therefore, the probability
of protection conditional on treatment in that year is 1/12. For those born in October,
the program reached their mothers in time if it started in January (1/12) and reached the
mother by March (1) or if it started in February (1/12) and reached them by March (2/3)
or if it started in March (1/12) and reached them in March itself (1/3). Therefore, the
probability of protection conditional on treatment in that year is 1/6. Given the assumption
on the rate of depletion made in Field et al. (2009), one that we maintain here, the stocks
of iodine retained in the body would be above the required levels for 24 months after the
administration of the pill. Therefore, one does not need to adjust for depletion for these
months. Finally, this calculated probability is multiplied by the coverage rate in a particular
district in a particular year to get the final treatment probability.
2.3.3 Empirical Specification
We begin by examining the direct impact of the iodine exposure on educational attainments
at the age of 10-13 in 2004. We follow Field et al. (2009) and Bengtsson et al. (2013), where
treatment is considered to vary exogenously at the district-birth year level after we control
for several observables and apply some fixed effects.
Y
1idb
=
1
+
11
ID
1idb
+
X
1idb
+
1b
+!
1d
+
1idb
(2.1)
whereY
1idb
is the years of schooling completed by an individuali born in districtd in
yearb by 2004. It depends onID
1idb
. the probability that individuali’s mother was treated
by the ISP program in the first trimester of her pregnancy. This treatment probability is
calculated as explained in the previous section. X
1idb
is a vector of covariates that include
72
a dummy each for whether the mother and father have some education or not, a dummy for
gender of the individual, a dummy each for whether the individual belongs to the majority
tribe or religion, and total land ownership of the household to which individuali belongs.
2
3
Standard errors are clustered at the district-year level in order to allow for arbitrary
correlation in the error terms within each cohort in a district.
4
Our specification is closer to that used in Bengtsson et al. (2013) and differs from Field
et al. (2009) in that it is more parsimonious with respect to the controls variables used.
Since the treatment occurred before birth, many of the potential controls run the risk of
being impacted by the treatment. For example, as pointed out by Becker and Tomes (1986)
and empirically verified by Rosenzweig and Wolpin (1980), fertility decisions are endoge-
nous to the quality of children, a dimension that ISP treatment might have affected. There-
fore, we exclude controls for birth order, number of children, distance to secondary school
and health clinic, food security measures, home ownership, housing quality, as used in
Field et al. (2009), and a dummy for household’s urban residence, as used in Bengtsson
et al. (2013).
Unlike Field et al. (2009), we do not use household fixed effect specification in our main
analysis. We believe that households where mothers gave birth twice within a span of four
years are different than those with one birth during the period and excluding household with
only one child born during this period will lead to substantial selection biases. While using
the month of birth information for the assignment of treatment probability would lead to
2
Land in rural areas were regulated by the Tanzanian government under the Village Land Act beginning
in 1999. To check for robustness, we replace the land ownership variable with the value of livestock owned
in an alternative specification. The results, not presented here, remain very similar.
3
Our sample has a very small percentage of HIV positive individuals. Out of 1784 individuals for whom
we clinical diagnosis, only four were diagnosed as HIV positive. Out of the 4240 individuals for whom we
have self diagnosis information, 10 diagnosed themselves to be HIV positive. Given the small percentage of
HIV positive individuals, we do not include HIV status as a control.
4
Since the number of districts is just seven, we prefer the district-year level clustering. Since the number
of clusters is relatively small, we repeat our analysis with cgm wildboot cluster method to correct for standard
errors (Cameron et al. (2008). The results (available on request) remain unchanged.
73
more accurate assignment, we do not have the month of birth information for a fairly large
number of individuals in our sample. Therefore, we follow Field et al. (2009) in our main
specification and assign treatment probabilities on the basis of month of birth. In section
B.2, we check the robustness of our results to the assignment of treatment probability on
the basis of month of birth for a smaller sample of individuals for whom we have the month
of birth.
Since the treatment status of the PEDP program is based on the year of birth, we cannot
use birth fixed effects in specifications that look to evaluate the impact of PEDP in addition
to ISP. Instead, we replace fixed effects in year of birth by a quadratic term in age. To
show that this quadratic term approximates the year of birth fixed effects closely, we re-
estimate (1) with the birth year fixed effect replaced by a quadratic in age. We, then,
look at the combined impact of ISP and PEDP on completed years of schooling using the
specification:
Y
2idb
=
2
+
21
ID
2idb
+
22
P
2idb
+
23
ID
2idb
P
idb
+
2
X
2idb
+
2
age+
2
age
2
+!
2d
+
2idb
(2.2)
For individual i, living in district d and born in year b, ID
2idb
represents the proba-
bility that individuali’s mother received iodine supplementation during the first trimester
of her pregnancy. P
2idb
indicates individual i’s exposure to PEDP and takes value ‘1’ if
the individual was born in 1993 or 1994, ‘0’ otherwise. Our specification includes district
fixed effects (!
d
), a quadratic in age, a dummy each for whether the mother and father have
some education or not, a dummy for gender of the individual, a dummy each for whether
the individual belongs to the majority tribe or religion, and total land ownership of the
household to which individuali belongs.
21
and
22
represent the independent impacts of
ISP and PEDP on the schooling attainments, respectively. Coefficient
23
is a measure of
74
heterogeneity in the impact of PEDP by ISP exposure status.
To provide suggestive evidence for the mechanism we propose, we begin by showing
that the ISP and PEDP treatment statuses and their interaction predict the primary school
start age and individual’s involvement in household work or work on family farm in a
manner that is consistent with the impact of these three variables on years of schooling.
For this we use a specification that is similar to (2), except now the outcome variable is the
age at which the individuals start primary school, probability that the individual worked
on-farm or at home in the last week, or the number of hours worked on-farm or at home in
the last one week.
We estimate the rate at which children treated by one or both of these programs convert
years at school into completed years of schooling by taking a ratio of estimated impact
of these two treatments and their interaction on completed years of schooling and primary
school starting age. We interpret these rates measures of the dynamic complementarity. A
higher value of this ratio for those exposed to ISP compared to those not exposed to ISP
would imply that the former group makes better use of each year spent in school.
2.4 Results
2.4.1 School Grade Attainment and Primary School Starting Age
Table 2.2 presents the impact of ISP exposure on completed schooling by 2004. The
columns differ in the controls used in the specification. For example, columns (1), (3),
(5), and (7) include age or birth year fixed effects to account for time varying unobservable
factors that might have impacted schooling levels in those years. Columnc (2), (4), (6), and
(8) replace the birth year fixed effect with a quadratic term in age. It is clear from the com-
parison of the coefficient across columns that a quadratic terms in age closely approximates
year of birth fixed effects in our analysis. Both the ISP coefficient magnitudes and the R-
75
squared fit of the model remains virtually unchanged. Controlling for tribe, religion and
total land makes almost no difference to the estimated impact of ISP on grade attainment.
When discussing the results, we will prefer the coefficient estimates from the specification
with the full set of controls and quadratic in age, similar to the one used in column (8).
According to the estimates, ISP exposure is associated with 0.70 fewer years of completed
schooling. Conditional on non-zero probability of exposure, the average probability of ex-
posure to ISP is 0.31. Therefore, those exposed to ISP, on an average, had completed 0.21
(0.70 * 0.31) fewer years of school.
At a first glance, the negative impact of ISP on grade attained is puzzling. There is no a
priori reason to expect a negative impact on grade attainment of a supplementation which is
expected to improve the cognition of those exposed. It also seems to contradict the results
from Field et al. (2009) that those exposed to ISP had completed more schooling. How-
ever, once we examine the behavioral responses to ISP exposure, the negative association
between ISP exposure and grade attainment in no longer a puzzle. But since the primary
objective of the paper is to examine how the two policies interacted, we discuss the joint
impact of these two policies before we document the behavioral response.
In table 2.3, we examine the joint impact of the two programs and their interaction on
years of schooling completed by the time of the survey in 2004. The coefficient estimates in
column (3) suggest that for those who were not exposed PEDP the impact of ISP remained
comparable to the estimates from table 2.2. However, there was a significant level of het-
erogeneity in the impact of PEDP by the ISP exposure status. Those not exposed to ISP but
exposed to PEDP had completed 0.18 extra years of education by the time of the survey
compared to those not exposed to either of the two programs. However, those exposed to
ISP and PEDP had, on average, completed 0.19 years of schooling (-0.71 * 0.31 + 0.18 -
0.47 * 0.31 = -0.19). They were comparable to those exposed to only ISP and not PEDP
who were lagging behind those who did not get exposed to either of the two programs by
76
0.22 years (-0.71 * 0.31).
The negative interaction effect is equally perplexing. Why will a reduction in the cost
of schooling hurt those who, most likely, have better cognition. One advantage of using
KHDS for our analysis is that it contains information on primary school starting age. By
examining the primary school starting age, a choice variables for individuals or their par-
ents, we can investigate if there were any behavioral responses to the policies. Table 2.4
reports the impact of the two policies and their interaction on primary school starting age.
The estimated coefficients seem to mirror the impact of the two policies and their interac-
tion on completed schooling. Those exposed to PEDP alone start school at a younger age
than those not exposed to either of the two programs. Those exposed to ISP only or both
the programs enter school later than those in the omitted category. On comparing 2.3 and
2.4, it is clear that the association of ISP exposure, PEDP exposure, and their interaction
with completed schooling is, at least, partly explained by changes in primary school start-
ing age in response to these treatments. In the next section, we provide suggestive evidence
to further explain why such a response might have arisen.
To be able to interpret the interaction coefficient as evidence in favor of or against
dynamic complementarity, one has to make the assumption that in the absence of dynamic
complementarity, the impact of the second policy on those who were exposed to the first
policy must be equal to the independent impact of the second policy on those who were
not exposed to the first policy. However, since ISP changes the primary start school age,
the cost and benefit from PEDP for those exposed to ISP might no longer be same as for
those who were not exposed to ISP. The difference in cost and benefits from PEDP might,
therefore, also invoke different behavioral response. However, since coefficient estimates
of the impacts of these two programs and their interaction on completed schooling and
primary school starting age from tables 2.3 and 2.4, respectively, are still unbiased, a ratio
of coefficient estimates for each of these policies and their interaction will be an indicator
77
of how productive a particular subgroup is in school. That is.
@(yearsof schooling)
@(schoolstartingage)
=
@(yearsof schooling)
@(exposuretoprogramX)

@(exposuretoprogramX)
@(primaryschoolstartingage)
(2.3)
whereX2fISP;PEDP;ISPPEDPg. We compute the conversion rate for the
subgroups in table 2.5. If those exposed to PEDP but not to ISP started school one year
earlier, they would have have attained 0.39 extra years of completed schooling. In com-
parison, those exposed to ISP alone would have attained 0.99 extra years of school. This
suggests that ISP exposure makes individuals more productive at school. If we consider
years in school as an input in the production function of human capital, the productivity of
this input is higher for those who benefit from the in-utero iodine supplementation. This
is an evidence of dynamic complementarity. We have to be careful when trying to make
a similar deduction for those who were exposed to both the programs. Their rate of con-
version should we a weighted average of the three conversion rates calculated in the table.
However, what should be the weights is not clear. However, for any combination of non-
trivial weights, their conversion rate will be better than those who were exposed to PEDP
but not ISP, lending further support to the dynamic complementarity argument made above.
2.4.2 Delay in Starting Primary School
But why might those exposed to ISP delay start of primary school more than those not
exposed? Late entry into primary school is very common in Tanzania (Burke (1998); Bom-
mier and Lambert (2000)) and elsewhere in Africa (Glewwe and Jacoby (1993); Glewwe
and Jacoby (1995); De Vreyer et al. (1998)). Several hypotheses have been proposed to
explain this delay in enrolment - existence of liquidity constraints (Jacoby (1994)), mal-
nutrition problems (Glewwe and Jacoby (1995)), considering children too young to be in
school (Burke (1998)), and pre-school training (De Vreyer et al. (1998)). There is no reason
78
to believe that ISP exposure of child was correlated with credit constraints that her family
faced, especially because ISP exposure depended on the timing of first trimester of preg-
nancy vis-a-vis supplementation and not on supplementation alone. Therefore, we focus on
the next two most important hypothesis - malnutrition problem and pre-school labor force.
Delay due to Worse Health
If those exposed to ISP delay starting school because they are malnourished, we might ex-
pect it to be reflected in their height-for-age. Low height-for-age is an indicator of stunted
growth reflecting a process of failure to reach linear growth potential is often associated
increased risk of early exposure to adverse conditions such as illness and/or inappropriate
feeding practices. Column (1) of table 2.6 presents the association between an individual’s
ISP exposure status and height for age. Those exposed to ISP, indeed, are shorter in 2004.
However, it is not clear why those exposed to ISP had worse growth. In-utero iodine sup-
plementation, especially in such low doses, has no adverse impacts on physical growth (Isa
et al. (2000)). Most individuals from the sample were interviewed at least once during the
first four waves of KHDS between 1991 and 1994. Height measurements were also taken.
Unfortunately, the number of individuals from each wave that we have height information
on is small. However, since the selection for being interviewed during any of these years
was unrelated to ISP exposure status, examining association between ISP exposure status
and height during these waves might still be informative. Columns (2)-(4) present the asso-
ciation between ISP and height for age during these waves. Even though the standard errors
are too large to interpret these coefficients without caution, the height for age during the
early years for individuals exposed to ISP seems to be higher than for those not exposed. If
the ISP exposure had an adverse impact on the physical health of those exposed, one might
expect to see an effect on height for age early on.
A second explanation for lower height attainment that seems to be consistent with the
79
trend in height differences across waves is that parents of those not exposed to ISP re-
sponded, either to their exposure status or to their lower cognition, by making compen-
satory investment in them. That might explain how the initial height advantage of those
exposed to ISP was reversed by 2004. However, previous studies from developing coun-
tries have mostly found that parental response in such scenarios is often to reinforce the
advantage that on of their children might have (Rosenzweig and Schultz (1982), Li et al.
(2010), Adhvaryu and Nyshadham (2014)). Most of these studies use a sibling fixed ef-
fects specification to check for reinforcement or compensation within families. Our main
sample consists of children born from 1991 to 1994. Households where mothers gave birth
twice or more within a span of four years are different than those with one birth during the
period. Therefore, excluding household with only one child born during this period will
lead to substantial selection biases. Therefore, the results of sibling fixed effects analysis
must be taken with a grain of salt.
The results are presented in table 2.7. Column (1) and (3) look at the association of
ISP exposure status with years of completed schooling by 2004 and height for age in 2004,
respectively. Since, here, we are interested in the impact of ISP alone and including those
born earlier than 1991 could, potentially, reduce some selection bias, in columns (2) and
(4), our sample consists of all those born between 1989 and 1994.
5
However, the results
do not indicate any compensatory response within the household. Most of estimated co-
efficients, even though not significant, are positive, consistent with reinforcement and not
compensation. This suggests the negative impact of ISP exposure on education and height
are identified from individuals with differential exposure to ISP born to households where
theirs was the single birth during this period (table B5). The results in 2.7 do not rule out
the possibility that households with one child born during this period and not exposed to
ISP tried to compensate for the lack of ISP exposure.
5
Kagera first received the program in one of its district in 1989.
80
Delay due to Child Labor
Next, we turn to the hypothesis with maximum empirical support in previous works - pre-
school training. According to the 2013 US Department of Labor report on worst forms of
child labor, as of 2011, over 25% of the Tanzanian children aged 5-14 were engaged in the
worst forms of child labor. A little over 20% of the children aged 7-14 were combining
work and school. Using information from the 2000 wave of the Tanzania Household and
Budget Survey, Kondylis and Manacorda (2012) find over 60% of children aged 7-14 en-
gaged in some form of productive activity and around 40% combining work and school.
Children worked on the family farm or did household chores. The average number of hours
worked every week was a little over 25 and around 20% of those who did not attend school
reported the reason as work or perceived uselessness of schools. Beegle et al. (2006b) use
information from the 2004 wave of KHDS and find that children aged 7-15 were found to
have worked a little over 18 hours in the week prior to their interview. Burke and Beegle
(2004) find the 10-15 year olds were working close to 9 hours on farming activities and
between 11-15 hours on household chores.
These findings are consistent with 2000-2001 integrated labor force and child survey by
the Tanzanian Ministry of Labor, Youth Development and Sports under International Labor
Organization’s International Program on Elimination of Child Labor. According to the re-
port, of the total number of children aged 5-17, 39.6 % were involved in economic activities
and 47.8% were engaged in housekeeping activities. Amongst those engaged in economic
activities, more than three quarter of them (78.8%) worked as unpaid family members in
their family farm or shamba and another 17.99% work as un-paid family members in non-
agricultural establishment. An estimated 34% of the total working children worked for
more than 4 hours per day or 30 hours per week. Beegle et al. (2006b) use crop and rainfall
shocks as instrumental variables for child labor and find that child labor has negative effects
81
on completed years of schooling. One of the ways in which child labor affects educational
attainment of children in Tanzania is through delayed enrolment. Even though children in
Tanzania are expected to enroll in primary school at the age of seven, enrollment is almost
always delayed by two, three, or even four years (Burke (1998); Bommier and Lambert
(2000)).
The returns to schooling in Tanzania are lower than other countries in the region (Knight
and Sabot (1990); Mason and Khandker (1997)). Since the country agricultural practices
mostly use traditional production methods, the returns to education in agriculture are low
(Mason and Khandker (1997)). Our findings are, therefore, not very surprising. Burke
and Beegle (2004) find that children in the Kagera region were not attending school due
to household demand of child labor and high opportunity cost of schooling. De Vreyer
et al. (1998), Bommier and Lambert (2000), and Burke and Beegle (2004) argue that the
main reason for delay in starting school in African countries, and in Tanzania in particular,
is the high opportunity cost of going to school. De Vreyer et al. (1998) present a model
where a household’s decision is similar to a portfolio choice among three assets - physi-
cal assets, ‘general’ human capital accumulation for the children through schooling, and
‘specific’ human capital accumulation for the children through participation in family eco-
nomic activities. Bommier and Lambert (2000) use information from the Human Resource
Development Survey conducted by the World Bank, the Dar-es-Salaam University, and the
Tanzanian government in 1992-1993 on 5000 households to test the model. They show
that the parents send their children to school later and for a smaller period of time since
Tanzania had high returns from accumulation of the ‘specific’ human capital.
From table 2.5, it does seem that ISP exposure made the exposed children smarter.
This, in turn, might have increased their opportunity cost of schooling more than those not
exposed. As a result, the ISP treated children might have chosen to start school later. More-
over, if they were aware that they were better at converting years in school into completed
82
years of schooling, this might have incentivized them further to delay schooling. This
would imply that those exposed to ISP were working more often than those not exposed
before school.
In 2004, at the time of the survey, all the individuals from the sample were in school.
It would have been ideal if we had information on the working status of these individuals
before they started school. Unfortunately, KHDS collected information on involvement
in market and non market labor activities only for the week preceding the interview date.
We assume that the number of hours worked in the week preceding the survey is correlated
with the number of hours worked every week in years preceding their enrollment in primary
school. Using information from THBS 2000, Kondylis and Manacorda (2012) find enrolled
children from all over Tanzania spent close to forty hours in school every week. They report
that hours of work among children in school was approximately half that of children out of
school. The alternative assumption that ISP treated children who enrolled later worked less
than those who were not exposed and enrolled at younger age is less plausible.
Most children generally work at home and on the family farm. These activities may
include but may but may not be limited to working in the fields or tending to livestock (cat-
egorized as farm activities in KHDS) or collecting water, fetching firewood, cleaning the
house, preparing meals, and time spent caring for other children or sick household mem-
bers (categorized as household chores in KHDS). Less than 0.5% of the children in our
sample were engaged in wage work outside the family farm. Therefore, we focus on work
on family farm and household chores only. Table 2.8 presents the association between pro-
gram exposures and number of hours worked in different activities preceding the survey.
ISP exposure seems to increase the number of hours worked on both family farms and un-
paid family chores. The coefficients, even though insignificant for the activities separately,
are in the right direction and large in comparison to the mean number of works worked in
these activities by the individuals in the sample. Moreover, when we combine the activities,
83
those exposed to ISP are working over five hours extra more than those not exposed. The
impact of PEDP exposure, even though positive, is small and insignificant. However, the
coefficient for the interaction large in columns (2) and (3), and significantly so for hours
spent on household chores. The results suggest that pre-school work is one of the reasons
behind delayed enrollment of those exposed to ISP.
While we find suggestive evidence in favor of both worse health and more pre-school
work for those exposed to ISP as possible reasons behind delayed enrollment, the evidence
is rather weak. Therefore, we do not want to claim one or both of these as the only or
even the strongest mechanism. What is clear, however, is that ISP invoked different be-
havioral responses from different subgroups and one needs to take those responses into
consideration when evaluating ISP and its interaction with PEDP.
2.5 Conclusion
There is now a broad consensus amongst demographers, sociologists and economists alike
that the diffusion of modern economic growth to the developing regions requires human
capital accumulation by the population in these regions (Counts (1931); Inkeles (1969);
North (1973); Davis et al. (1971); Rosenberg et al. (1986); Easterlin (1981); Easterlin
(2009)). A higher level of human capital is desirable in its own right (Pigou (1952); Adel-
man (1975); Grant (1978); Grant (1978); Streeten et al. (1981); Sen (1984)). However,
poor access to information and quality infrastructure, low levels of incomes, and imperfect
credit markets in these regions limit the possibilities of private investment in human capital
accumulation. State run policies, therefore, are of extreme importance in ensuring higher
levels of human capital (Easterlin (1981)). Given the limited state budget, the decision of
whether or not to roll out a particular program depends a lot on the cost benefit analysis
of the program. Traditionally, the cost benefit analyses of such development programs is
84
based on the evaluation of the single program as if it was implemented in isolation. How-
ever, if two independent programs interact in important ways, a partial equilibrium analysis
might greatly understate the net benefits of such programs. In such scenarios, it becomes
essential to jointly evaluate the impact of the two (or more) programs, allowing for possible
complementarity between the programs.
Keeping this in mind, we evaluate the Iodine Supplementation Program and Primary
Education Development Program in Tanzania. We find that ISP treatment was associated
with lower schooling achievements for the exposed kids in 2004. The effect operated en-
tirely through delays in enrollment. We provide suggestive evidence that this behavioral
response of delaying enrollment was because those exposed to ISP were in worse health
and spent more time working on the family farm or in the house. This, we conjecture,
might have been because their improved cognition made them better at these jobs. More
importantly, we find that those exposed to ISP were better at converting years in school into
completed years of schooling - a sign of dynamic complementarity between ISP exposure
led improved cognition and time spent in school.
The result that government policies interact in important ways might explain why the
short run impacts of many programs dissipate in the long run. The results also underscore
the need to raise the dimensionality of the policy space to be considered. However, it is im-
possible to evaluate the combined effect of all sort of different policy exposures and deter-
mine how they interacted. Perhaps a better combinations of theoretical, non-experimental,
quasi-experimental and experimental methods need to be developed to handle the situation.
85
2.6 Tables and Figures
FIGURE 2.1: IODINE SUPPLEMENTATION PROGRAM IN TANZANIA (FROM FIELD ET AL.
(2009))
86
0 5 10 15 20
Lowess
20 40 60 80 100
Age
Schooling (ISP treated districts)
Schooling (ISP untreated districts)
Starting age (ISP treated districts)
Starting age (ISP untreated districts)
FIGURE 2.2: TRENDS IN YEARS OF EDUCATION AND PRIMARY SCHOOL STRATING AGE
BEFORE TREATMENTS
87
-.05 0 .05 .1
Lowess - Completed apropriate grade for age
20 40 60 80 100
Age
ISP treated districts ISP untreated districts
FIGURE 2.3: TREND IN COMPLETION OF APPROPRIATE GRADE FOR AGE BEFORE
TREATMENTS
88
150 155 160 165
Lowess - Height (in cms)
20 40 60 80 100
Age
ISP treated districts ISP untreated districts
FIGURE 2.4: TREND IN HEIGHT BEFORE TREATMENTS
89
TABLE 2.1: SUMMARY STATISTICS
Outcomes Control Treatment
Ages 10-11
Mean SD Mean SD
Years of Schooling 1.87 1.00 1.44 0.96
Primary school start age 7.98 1.02 8.22 1.15
School Progression 0.82 0.30 0.79 0.40
HAZ in 2004 130.50 8.99 127.77 7.61
Proportion with
Vaccination card 0.92 0.85
Tb vaccination 1.00 0.97
Measles vaccination 1.00 0.92
Tetanus vaccination 0.46 0.39
Polio vaccination 0.52 0.55
Ages 12-13
Mean SD Mean SD
Years of Schooling 3.09 1.38 2.78 1.42
Primary school start age 8.57 1.45 9.01 1.46
School Progression 0.79 0.25 0.81 0.24
HAZ in 2004 141.21 8.47 139.64 8.44
Proportion with
Vaccination card 0.95 0.99
Tb vaccination 0.99 0.99
Measles vaccination 0.94 0.97
Tetanus vaccination 0.80 0.88
Polio vaccination 0.84 0.86
Independent variables Control Treatment
Ages 10-11
Mean SD Mean SD
Protection due to ISP 0 0 14.26 17.17
Age 10.34 0.47 10.13 0.34
Mother has any education 0.95 0.21 0.92 0.27
Father has any education 0.92 0.27 0.92 0.27
Household land per capita 0.48 0.53 0.55 0.47
Proportion
Sex = Male 0.54 0.48
Tribe = Mhaya 0.93 0.37
Religion = Catholic 0.65 0.53
N 133 185
Ages 12-13
Mean SD Mean SD
Protection due to ISP 0 0 70.05 27.72
Age 12.57 0.50 12.55 0.50
Mother has any education 0.97 0.17 0.89 0.32
Father has any education 0.95 0.21 0.88 0.32
Household land per capita 0.56 0.56 0.80 0.66
Proportion
Sex = Male 0.47 0.48
Tribe = Mhaya 0.94 0.01
Religion = Catholic 0.65 0.50
N 161 87
90
TABLE 2.2: IMPACT OF IODINE SUPPLEMENTATION PROGRAM ON COMPLETED YEARS OF
SCHOOLING
(1) (2) (3) (4) (5) (6) (7) (8)
V ARIABLES Years of education
Iodine Supplementation Program -0.70*** -0.70*** -0.69*** -0.70*** -0.69*** -0.70*** -0.69*** -0.70***
(0.12) (0.13) (0.12) (0.13) (0.12) (0.12) (0.12) (0.12)
Age fixed effect YES NO YES NO YES NO YES NO
Quadratic in age NO YES NO YES NO YES NO YES
Religion dummy NO NO YES YES NO NO YES YES
Tribe dummy NO NO NO NO YES YES YES YES
Land ownership control YES YES YES YES YES YES YES YES
Mean of dependent variable 2.20 2.20 2.20 2.20 2.20 2.20 2.20 2.20
Mean ISP treatment probability 0.31 0.31 0.31 0.31 0.31 0.31 0.31 0.31
Observations 518 518 518 518 518 518 518 518
R-squared 0.36 0.36 0.36 0.36 0.36 0.36 0.36 0.36
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at district-
year of birth groups. Other controls include a dummy each for whether the mother and father have some education or not, a
dummy for gender of the individual, and district fixed effects.
TABLE 2.3: IMPACT OF ISP AND PEDP ON COMPLETED YEARS OF SCHOOLING
(1) (2) (3)
V ARIABLES Years of education
Iodine Supplementation Program (ISP) -0.70*** -0.71***
(0.12) (0.11)
Primary Education Development Program (PEDP) 0.13 0.18**
(0.15) (0.07)
ISP * PEDP -0.47*
(0.26)
Mean of dependent variable 2.20 2.20 2.20
Mean ISP exposure probability 0.31 0.31
Observations 518 518 518
R-squared 0.36 0.35 0.36
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered
at district-year of birth groups. Controls include a quadratic in age, a dummy each for whether the mother and father
have some education or not, a dummy for gender of the individual, a dummy each for whether the individual belongs
to the majority tribe or religion, controls for total land holdings of the household, and district fixed effects.
91
TABLE 2.4: IMPACT OF ISP AND PEDP ON PRIMARY SCHOOL STARTING AGE
(1) (2) (3)
V ARIABLES Primary school starting age
Iodine Supplementation Program (ISP) 0.77*** 0.74***
(0.16) (0.15)
Primary Education Development Program (PEDP) -0.47** -0.47**
(0.19) (0.18)
ISP * PEDP 0.26
(0.21)
Mean of dependent variable 2.20 2.20 2.20
Mean ISP exposure probability 0.31 0.31
Observations 518 518 518
R-squared 0.17 0.16 0.17
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered
at district-year of birth groups. Controls include a quadratic in age, a dummy each for whether the mother and father
have some education or not, a dummy for gender of the individual, a dummy each for whether the individual belongs
to the majority tribe or religion, controls for total land holdings of the household, and district fixed effects.
TABLE 2.5: CONVERSION OF AN ADDITIONAL YEAR INTO ADDITIONAL YEARS OF
SCHOOLING
Treatment School starting age Years of schooling
@(yearsofschooling)
@(schoolstartingage)
PEDP only 0:47 0:18 0:38
ISP only 0:74 0:71 0:96
ISP * PEDP 0:26 0:47 1:8
Notes:
@(yearsof schooling)
@(schoolstartingage)
=
@(yearsof schooling)
@(exposuretoprogramX)

@(exposuretoprogramX)
@(primaryschoolstartingage)
, where X 2
fISP;PEDP;ISPPEDPg
92
TABLE 2.6: IMPACT OF ISP ON HEIGHT OF THE CHILD (HEIGHT-FOR-AGE)
(1) (2) (3) (4) (5)
Height for age Z-score in
V ARIABLES 2004 1994 1993 1992 1991
ISP -0.46** -2.34 3.33** 1.79 3.80
(0.21) (1.45) (1.40) (1.51) (2.48)
Observations 501 102 118 128 145
R-squared 0.04 0.09 0.08 0.17 0.13
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard
errors are clustered at district-year of birth groups. Controls a dummy each for whether the mother
and father have some education or not, a dummy each for whether the individual belongs to the
majority tribe or religion, controls for total land holdings of the household, and district fixed ef-
fects. We used the WHO Child Growth Charts and WHO Reference 2007 Charts for our height for
age analysis.
TABLE 2.7: WITHIN HOUSEHOLD IMPACTS OF ISP
(1) (2) (3) (4)
V ARIABLES Years of education Height for age in 2004
ISP 0.24 0.22 0.78 -0.39
(0.59) (0.46) (0.87) (0.47)
Observations 132 335 119 298
R-squared 0.86 0.86 0.66 0.67
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at
the level of the household. Controls include a dummy for gender of the individual, age and sibling fixed effects.
93
TABLE 2.8: IMPACT OF ISP AND PEDP ON HOURS WORKED
(1) (2) (3)
In the last week, how many hours did you work [...]?
V ARIABLES on the family farm in unpaid work in total
ISP 1.97 2.07 5.43***
(1.46) (1.75) (1.93)
PEDP 1.97 0.63 1.72
(1.18) (1.20) (1.08)
ISP * PEDP -1.85 10.33* 7.62
(3.07) (5.61) (8.69)
Mean of dependent variable 4.51 5.49 10.18
Mean ISP treatment probability 0.32 0.32 0.32
Observations 540 540 540
R-squared 0.04 0.05 0.04
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered
at district-year of birth groups. Controls include a quadratic in age, a dummy each for whether the mother and father
have some education or not, a dummy for gender of the individual, a dummy each for whether the individual belongs
to the majority tribe or religion, controls for total land holdings of the household, and district fixed effects.
94
Chapter 3
Coethnic Voters and Candidate Choice by Political
Parties: Evidence from India.
3.1 Introduction
Much has been written about the poor policy outcomes in situations where co-ethnic voting
has, allegedly, been salient.
1 2
It is a more serious problem in developing countries, where
the accountability mechanisms are often underdeveloped and discrimination on the basis
of ethnicity often more rampant (Vicente and Wantchekon (2009)). Banerjee and Pande
(2009), Vaishnav (2010), and Acharya et al. (2014) point out that ethnic voting explains
much of the inefficiency and criminalization plaguing the politics in India.
3
Most of these studies examine the impact of ethnic voting, taking as exogenous, the
ethnicity of the candidates running for elections in any constituency. However, a party’s
choice of their candidate to field for elections in a constituency may depend on ethnic com-
position of the constituency and ethnic preferences of the constituency’s population.
4
The
literature on how a party chooses its candidate to run for an office is scarce.
5
in comparison,
1
Co-ethnic voting is when voters of an ethnic group show an affinity for candidate belonging to the same
ethnic group in their voting behavior (Wolfinger (1965)).
2
Key (1949) argues that ethnic preference limited the competition and led to rampant corruption and
slowdown of economic development in the southern region of United States (Besley et al. (2005)). See also
Young (1979), Bates (1983), Horowitz (1985), and Dahl (2005).
3
These studies focus on the elections in the state of Uttar Pradesh in India. Uttar Pradesh is the biggest
state of India by population and by number of parliamentary and assembly constituencies.
4
In fact, the birth and rise of certain political parties in India is itself considered to be in response to the
ethnic fragmentation of the state (Jaffrelot (2003))
5
Field and Siavelis (2011) look at the selection process of candidate by a party in which each member of
the party maximizes her own profit and bargains for the position. In most situations, all members of the party
working towards maximizing the profit of the party appears to be a better assumption. An alternative middle
95
the literature on the related question of a political party’s choice of policy position is much
more voluminous. It builds on the classical works of Hotelling (1929) and Downs (1957)
on spatial competition.
6
This paper applies their model of spatial competition to study the
selection of candidates by political parties when they internalize the ethnic preferences of
the voters. In that, it replaces the policy space with an ethnicity space.
To understand better the need to study a political party’s candidate selection process,
consider the simple situation described in table 3.1. A constituency is composed of voters
from four ethnicities -E
1
;E
2
;E
3
;E
4
. All voters vote for co-ethnic candidates. However,
conditional on their co-ethnic preferences, they prefer a candidate with lower involvement
in crime. In this constituency, in terms of voters from each ethnicity, ethnicity E
4
is the
biggest ethnic group, followed by ethnicityE
3
, followed byE
2
. E
1
is the smallest ethnic
group. Two parties,X andY , are planning to field candidates for office in this constituency.
PartyX has two potential candidates,X
1
andX
2
, to choose from to field. CandidateX
1
hails from ethnicity E
1
and has no involvement in criminal activities. Candidate X
2
be-
longs to ethnicityE
2
and has a high degree of involvement in criminal activities. PartyY
plans to field eitherY
3
orY
4
.Y
3
is from ethnicityE
3
and has some involvement in criminal
activities. However, on a relative scale, she is less criminal thanX
2
and more criminal than
X
1
. Y
4
, hailing from ethnicityE
4
, had the highest involvement in criminal activities. Let
us assume that the higher a candidate’s involvement in crime, the worse she is as a political
representative.
7
If the political parties are interested in maximizing their votes shares but
do not internalize the ethnic preferences of the voters, they will choose to field candidates
based on their extent of involvement in criminal activities. PartyX will field candidateXa
1
and partyY will runY
3
for office. Since, by assumption, voters place a higher weight on
ethnicity of the candidate than on other characteristics,Y
3
, the worse of the two candidates,
path to take is to assume that individual party member’s profit are proportional to party’s profit
6
Stokes (1963) provides a good review of the early work on spatial competition models.
7
The assumption is supported by findings from India in Prakash et al. (2015) and Banerjee et al. (2014).
96
will be elected. Much of the focus in prior literature on the topic is on this inefficiency.
However, if parties internalize the ethnic preference of the voters, partyX will choose to
run candidateX
2
and partyY will run candidateY
4
. This result, then, will be worse - the
candidate with the highest involvement in crime will be elected.
TABLE 3.1: CO-ETHNIC PREFERENCE AND STRATEGIC CHOICE OF CANDIDATES.
PartyX PartyY
Candidate:X
1
Candidate:Y
3
Ethnicity:E
1
Ethnicity:E
3
Criminal Cases: Lowest Criminal Cases: Low
Candidate:X
2
Candidate:Y
4
Ethnicity:E
2
Ethnicity:E
4
Criminal Cases: High Criminal Cases: Highest
In this paper, I model the second source of inefficiency pointed out above. If political
parties internalize the co-ethnic preference of voters, they may, to avoid political com-
petition, differentiate themselves along ethnic lines in their choice of candidates to field.
If such strategic diversification indeed occurs, political parties might substitute favorable
ethnicity for quality resulting in the selection of worse quality candidates. I provide empir-
ical evidence from elections in Uttar Pradesh and Bihar in favor of the model predictions.
Political parties respond to both the ethnic composition of voters and ethnicity of candi-
dates from opposition parties in selection of candidates to field for elections. A comparison
of the candidates running for elections from constituencies where such diversification is
possible with candidates from constituencies where such diversification is limited due to
affirmative action laws suggests that candidate quality indeed suffers due to the strategic
diversification. However, this strategic diversification is much lower in smaller, well in-
formed, well educated constituencies. A policy implication is to ensure that voters have
97
detailed information about the candidates, parties and their political stances.
This paper points out that inefficiencies of co-ethnic voting might be underestimated if
we do not take into account that strategic ethnic diversification by political parties. It also
suggests that the election of a worse candidate to office is often due to the voters facing a
worse pool of candidates to choose from.
The paper is organized as follows: Section 3.2 describes the model and its assump-
tions. Section 3.3 characterizes the equilibrium and discusses the empirical analogue for
the model. Section IV describes the data and the setting in which the model is tested.
Section V presents the results. Section VI re-examines the underlying assumptions of the
model and comments on their plausibility. Section VII concludes.
3.2 The Model
Consider a representative democracy which consists of C constituencies, indexed by c.
Polling occurs at the constituency level. There are P political parties, indexed by r, all
or some of whom may be contesting the election in any constituency. Since the empirical
section of the paper uses information from Indian elections, we will keep our terminology
consistent with the ethnic classification in Indian politics. Affirmative action in Indian pol-
itics takes the form of candidate restrictions. In some constituencies, candidates from only
a subset of all ethnicities can run for elections. We call these ‘reserved constituencies’.
8
Other constituencies where these candidate restrictions do not apply and the elections are
open for candidates of all ethnicities will be called ‘general constituencies’. Accordingly,
8
The ethnic categories defined here are closely related to the caste classification in India. Each ethnic
category encompasses several caste group. However, there is no overlap - no caste can belong to more than
one ethnicity. The data from elections in India that we make use of contain information on ethnic identities
of the candidates running for office but not on their caste identity. For this reason and to maintain consis-
tency with previous literature, I call these aggregated caste group ‘ethnicities’. For reserved constituencies,
only candidates that belong to groups officially designated as Scheduled Castes or Scheduled Tribes by the
Constitution of India can run for elections.
98
we will assume that there are two ethnic groups in the population. Individuals - citizen vot-
ers or political party members - belong to one of the ethnicities, ‘general’ or ‘reserved’. In-
dividuals who belong to the ethnic groups that benefit from the candidate restriction clause
in any constituency, and not necessarily in the constituency of the voter concerned, belong
to the ‘reserved’ ethnicity. Individuals who belong to the ethnic groups that cannot contest
elections from reserved constituencies, regardless of the status of their home constituency,
belong to the ‘general’ ethnic group.
Since observable diversification ethnic lines is not possible in constituencies where
candidates from only one ethnic group can run for election, I consider only the general
constituencies in our model. The reserved constituencies, however, will serve as useful
counterfactuals when I compare the inefficiencies in outcomes due to the strategic diversi-
fication. In all general constituencies, each party can choose to field a candidate of either
general or reserved ethnicity. We assume that parties care only about maximizing their
vote shares. In any general constituency, each political party will choose to run a candidate
who will maximize the votes the political party earns in that constituency. We assume that
voters in each constituency are more likely to vote for candidates from their own ethnicity
than for a candidate from the other ethnicity.
The situation is similar to firms choosing a location for their store within a market.
9
As
in a firm location model, there are two main determinants of the ethnicity of the candidate
the party decides to run in the election. Similar to how a firm will want to choose a location
for the store where the demand for its product is high, if the parties internalize the co-ethnic
preference of the voters, they will want to run a candidate whose ethnicity is the same as
that of the majority of co-ethnic voters. However, all parties will want to capitalize on
the high voter population of a particular ethnicity. If all parties choose candidates of the
9
The model presented here is similar to the one Seim (2006) presents. Attempts have been made to keep
the notation similar to Seim (2006) to facilitate comparison.
99
same ethnicity, there will be increased competition for the votes from the that ethnic group.
The vote bank from this big ethnic group will, then, be split among the parties. Under
these circumstances, it might be beneficial for a party to deviate and run a candidate of a
minority ethnicity. Even though it might lose its share of votes from the majority ethnic
group as a consequence, it will have all the votes from the minority ethnicity. The gains
might outweigh the losses in some scenarios.
The percentage of votes that a potential candidate can fetch for its party depends, on
the population percentage of co-ethnic voter’s of her ethnicity, the number of candidates
of the same ethnicity as her own competing against her, and the number of candidates of
other ethnicities competing against her.
10
It might also depend on party and constituency
characteristics. For example, some parties may be more popular than others when all con-
stituencies in a state are considered together but others might enjoy more popularity in
particular constituencies. The percentage of votes won might also depend on a host of
other candidate characteristics. More often than not, parties do not even know all the po-
tential candidates being considered by any rival party, let alone having information on all
the characteristics about those potential candidates that matter for the selection procedure.
The characteristics of potential candidates that a party is choosing from is assumed to be
private information of the party. We use a constituency-party-candidate’s ethnicity-specific
error term, that we call candidate’s profitability type, to include all such omitted variables
related to candidate characteristics in the vote profit equation that we will specify below.
Assumption 1 (Independent Symmetric Candidate Profitability Type): The profitabil-
ity type of a candidate of ethnicity e from partyr in constituencyc, "
c
re
, is private infor-
mation to the party and "’s are independently and identically distributed draws from the
10
India follows the electoral system of first-past-the-post plurality voting where each voter is allowed to
vote for only one candidate, and the candidate who receives the highest percentage of votes casted in the
constituency is elected. Political parties, therefore, are interested in the percentage of votes polled for their
candidate in a constituency, and not the absolute number of votes.
100
distributionF (:). This distribution is common knowledge.
The percentage vote profit function for partyr, if it runs a candidate of ethnicitye from
constituencyc, can be written as

c
re
=X
c
e
+
c
+
r
+
c
r
g(n
c
;
c
:e
) +"
c
re
(3.1)
X
c
e
is the population share of voters of ethnicitye in constituencyc.
c
are constituency
level variables that might affect the vote profit of running a candidate.
r
are party-specific
variables, like the nationwide popularity of the political party, that might affect the votes a
candidate from that political party wins in a particular constituency.
c
r
are factors that affect
the vote profit that are different across parties within a constituency and different across
constituencies for the same party. For example, the popularity of the same party could be
different across constituencies and different parties will have different levels of popularity
within any given constituency. g(.) represents the competition dimension of the game.
c
:e
captures the magnitude of competition that a candidate of ethnicity e in constituencyc faces
from other candidates. n
c
is the vector of the number of other candidates of each ethnicity
contesting against a particular candidate.
Next, we assume that the competition effect is linear and additive. Each additional rival
candidate reduces the percentage of votes a candidate, let us call him candidate X, receives.
However, since all candidates of the same ethnicity split the votes from co-ethnic voters of
that ethnic group, the reduction in vote shares is higher when the new rival candidate is of
the same ethnicity as candidate X.
Assumption 2 (Linear Additive Competition Effect):
g(n
c
;
c
:e
) =
h

he
n
c
h
; e;hfGeneral, Reservedg
Assumption 3 :
(i) Proximity Competition Effect:
ee
>
ef
101
(ii) Symmetric Competition Effect:
fe
=
ef
e;f fGeneral, Reservedg
We expect
ee
to be positive but there are no clear priors on the sign of
ef
. Given the
assumptions, the percentage vote profit function can be rewritten as

c
re
=X
c
e
+
c
+
r
+
c
r

h

he
n
c
h
+"
c
re
; e;hfGeneral;Reservedg (3.2)
All parties move simultaneously when choosing their candidate. But parties do not have
perfect information about the ethnicity of the candidates that the rival’s parties will choose
and, therefore, do not known
c
h
for anyh. They form expectations about the number of rival
candidates of each ethnicity. The vote profit function becomes
E(
c
re
) =X
c
e
+
c
+
r
+
c
r

h

he
E(n
c
h
) +"
c
re
; e;hfGeneral;Reservedg (3.3)
Let the decision vector for each party be given by d
re
, whered
re
c
takes value 1 if the
party r chooses to run a general candidate for constituency c, and takes a value 0 if the
candidate is from the reserved ethnicity. Party s’s perception about party r’s choice of
candidate and, therefore, ethnicity in a particular constituency is given by
p
c
re
=Pr(d
c
re
= 1jX
c
e
;X
c
f
;
r
;
c
r
) =Pr(E(
c
re
)E(
c
rf
); e6=f;r6=s; 8rP (3.4)
where e, for example, might denote general ethnicity and f might denote the reserved
102
ethnicity. Therefore,
p
c
re
=Pr(X
c
e

h

he
E(n
c
h
) +"
c
re
X
c
f

h

hf
E(n
c
h
) +"
c
rf
);
8rP ; e;f;hfGeneral;Reservedg (3.5)
Since partys has similar information for all the other parties, from partys’s perspective,
p
c
re
=p
c
qe
=p
c
e
; 8r;qP ; r;q 6= s (3.6)
Therefore, the expectations of partyS about the number of candidates of any ethnicityh in
a particular constituency will be given by
E(n
c
h
) = (P 1)p
c
h
(3.7)
where (P1) is the number of other parties running a candidate against partys’s candidate.
3.3 Equilibrium and Empirical Analogue
Assumption 4: "’s are i.i.d. draws from standard normal distribution (or extreme value
type 1 distribution)
Given this assumption on the distribution of error terms
11
, equation (5) can be rewritten
as
11
In fact, the assumption required is weaker than the one made here. The required assumptions are dis-
cussed in greater detail in Section VI
103
p
c
re
=p
c
e
=Pr("
c
re
"
c
rf
(X
c
e

ee
(P 1)p
c
e

fe
(P 1)p
c
f
X
c
f

+
ef
(P 1)p
v
e
+
ff
(P 1)p
c
f
)
=F ((X
c
e
X
c
f
) (
ee

ef
)(P 1)(p
c
e
p
c
f
)) (3.8)
whereF () is the c.d.f. of normal (or logistic) distribution.
The situation can be described as a simultaneous move game where the probability
term is the fixed point of a set of simultaneous equations similar to (3.8) and the solution
isp
c
re
= p
c
e
. The outcome for each constituency, therefore, is one where each party does
its best given what its rivals are doing. SubstitutingX
c
f
= (1X
c
e
),the above equilibrium
generates the following comparative statics.
@p
re
@X
c
e
> 0;
@p
re
@n
c
e
< 0; (3.9)
@p
re
@n
c
f
T 0; if
ef
T 0 (3.10)
All things constant, if there are more voters of ethnicitye, all of whom have co-ethnic
preference, it is more profitable to run a candidate of ethnicitye. If a political party ran-
domly chose a candidate from the voter population to run for elections and was not acting
strategically, the probability of fielding a candidate of ethnicity e will still be positively
associated with the voter population of that ethnicity.
12
However, if this mechanical corre-
12
If 90% of the population in a constituency is from ethnicitye, there is a 90% chance that a candidate
selected at random my a party is also from ethnicitye.
104
lation was the only factor driving the association, a 1% increase in the voter population of
ethnicitye would translate into a 1% increase in the probability of a candidate of ethnicity
e being chosen by the parties. To test and reject the mechanical correlation, I test explicitly
the null hypothesis that
@pre
@X
c
e
= 1.
The higher the number of rival candidates of ethnicitye, the higher is the competition
for votes of the co-ethnic voters of ethnicity e, and, therefore, of fielding a candidate of
ethnicitye. The impact of an increase in the number of candidates of another ethnicity,f,
is not as clear. It depends on the percentage of voters in the constituency voting ethnically.
If all voters vote for co-ethnic candidate, then a rival party’s decision to field a candidate
of ethnicity f should increase (decrease) the probability of a party running a candidate
ethnicity e (f) to be able to diversify. However, there might voters in the constituency
who do not vote co-ethnically. For the votes of these voters, any competition for a party’s
candidate is bad news, regardless of the ethnicity of the candidate. The net impact will
depend on the interplay of these two forces. As long as the percentage of voters who
vote for co-ethnic candidates is reasonably large, the competition effect of an additional
candidate of some other ethnicity should be smaller than that of other candidates from the
same ethnicity (
@pre
@n
c
f
>
@pre
@n
c
e
).
The selection process can be rewritten as
G
c
r
=
8
>
<
>
:
1 if E(
c
r;general
)E(
c
r;reserved
):
0; otherwise
or
G
c
r
=
8
>
<
>
:
1 if X
c
gen
+
h

h
E(n
c
h
) +"
c
r;general
X
c
reserved
+
h

h
E(n
c
h
) +"
c
r;reserved
:
0; otherwise
(3.11)
105
whereG
c
r
is a dummy variable that takes value one if partyr chooses a general candidate
in constituencyc. We estimate this using a probit (or logit) specification.
3.4 Data and Setting
3.4.1 Elections in Uttar Pradesh and Bihar
India follows a dual polity. It has a union government at the center and state governments
at the periphery. There is a bicameral legislature at the center, consisting of an upper house
(Rajya Sabha or the Council of States) and a lower house (Lok Sabha or the House of
the People). A few states have a bicameral legislature, with an upper house (Legislative
Council or Vidhan Parishad) and a lower house (Legislative Assembly or Vidhan Sabha)
but the majority only have the lower house. The members of the lower house at both the
federal and state level are elected directly by adult universal suffrage. The elections for the
Lok Sabha are called parliamentary elections while those for the Vidhan Sabha are called
assembly elections.
I test the predictions of the model using information on the parliamentary election(PE)
and assembly elections(AE) from the states of Uttar Pradesh (UP) and Bihar in India. I
choose these states for two main reasons. First, ethnic rivalries and co-ethnic political pref-
erences are much more salient for these states (Acharya et al. (2014); Vaishnav (2010);
Banerjee and Pande (2009); Chandra (2007); Witsoe (2005)). Second, UP is the most
populous state of India and has the largest number of parliamentary and assembly con-
stituencies. Bihar ranks third in terms of population and fourth in terms of number of
parliamentary and assembly constituencies. Together, they account for around one fourth
of the country’s population. More than two-thirds of the population in both states is en-
gaged in agriculture and the literacy levels are below the country average. The population
in India is diverse but UP and Bihar, arguably, are two of the best representations of the
106
socio-economic and political climate in many parts of the country.
Elections in these two states are mostly dominated by the Hindus even though Muslim
candidates have a strong political presence in a few constituencies. In this paper, I do not
make a distinctions between candidates or voters from different religions.
13
The elections
in the state of Uttar Pradesh at both the parliamentary and the assembly level have been
dominated by four big parties - Bahujan Samaj Party (BSP), Samajwadi Party (SP), Indian
National Congress (INC) and Bhartiya Janta Party (BJP). All four parties, explicitly or
implicitly, appear to have affiliation or affinity to certain caste group or groups. But a clear
ethnicity (caste) affiliation is not observed in their actions. This is visible in the election
results of the past. The Scheduled Caste (SC’s) and Scheduled Tribes (ST’s) together form
around one-fourth of the state’s population. The Other Backward Castes (OBC’s) form
around 40% of the state’s population. BSP has explicitly identified itself to be representing
the SC’s and ST’s while the SP has been said to be targeting the OBC votes. However, In
the 2007 UP AE elections, BSP won 206 out of the 403 seats, a clear majority, while the SP
won 97 seats to finish second. In the 2012 UP AE, SP won 224 seats while the BSP bagged
80 seats in total. Strict affiliations to the caste groups in face of ethnic voting could not have
produced these results. In fact, the BSP’s 2007 victory was as a result of the fact that they
managed to bag a large number of votes from general ethnicity voters by running general
(and in particular, brahmin) category candidates even after their historically antagonistic
attitude towards the general ethnicity.
14
. It is this strategic behaviour that the present model
is trying to capture.
The situation in Bihar is similar. The state politics is dominated by Rashtriya Janta
Dal (RJD), Janta Dal (United), INC and BJP but has considerable presence of BSP, SP,
13
The fragmentation along the religion dimension and it’s complex relationship with the politics in the
country is interesting in its own right and a potential subject of future research.
14
Acharya et al. (2014) The BSP took this to an extreme in its 2002 slogan ”Thrash the Brahmin, the
Bania and the Rajput” (translated from Tilak, tarazu aur talwar, Inko maro joote chaar(Jain 1996, p. 215)
107
Lok Janshakti Party (LJP), National Congress Party(NCP) and Communist Party of India
(Marxist-Leninist) Liberation. While LJP enjoys considerable popularity among the SC
voters, RJD is popular among the OBC voters.
There are 80 parliamentary and 403 assembly constituencies in the state of UP and 40
parliamentary and 243 assembly constituencies in Bihar. As a aprt of the affirmative action
in politics, around a fifth of these constituencies are ‘reserved’ for the SC/ST candidates.
This means that only candidates from the SC and ST ethnic groups can run for office from
these constituencies. Since there is no scope for political parties to diversify along ethnic
lines, we drop these constituencies from our sample.
15
Next, I also drop the independent
candidates from our sample.
16
I do this because an independent candidate has no two
ethnicities to choose from and, therefore, such strategic concern.
17
I use data from the 2009 and 2014 PE, 2010 Bihar AE, and 2012 UP AE. I test the
predictions of the model separately for the parliamentary and assembly elections. For each
level of elections, I combine the data from the two states. I use information from elections
after the most recent delimitation of constituencies that started in 2008. The constituency
boundaries before the 2008 delimitation had remained unchanged since 1976 and were
based on 1971 census population figures. Since the smallest level of aggregation at which
the information on the population shares different ethnic groups is available does not fit the
assembly and parliamentary constituency boundaries perfectly (See Alam (2010), and the
literature it cites.), it is difficult to get reliable estimates of constituency-specific population
sizes of ethnic groups for elections before 2008. The 2008 delimitation was carried out on
the basis of the 2001 census figures. Official reports from the Delimitation of Parliamentary
15
Each of the two ethnic groups, general and reserved, are composed of several castes. A diversification
along caste lines is still possible in the reserved constituencies. However, since we only observe the ethnic
identity (general or reserved) and not the caste identity of voters and candidates in our data, we cannot observe
any diversification within reserved constituencies.
16
Independent candidates are those who are not affiliated to any party.
17
In this paper, I abstract away from the first stage the game presented here where people and parties
decide whether or not to run the elections from a constituency
108
and Assembly constituencies outlined the process and contain information on constituency-
specific population of ethnicity groups. The election results data were obtained from the
Election Commission of India’s
18
. The elections data contain, for each constituency, the
ethnicity of the candidates running for office from that constituency and an identifier for
the winner from each constituency. Data on candidate characteristics are obtained from a
series of affidavits released as a result of a Public Interest Litigation filed with the Delhi
High court. They contain information on age, assets, liabilities, education attainment and
criminal cases filed against the candidate.
3.4.2 Pradhan Mantri Gram Sadak Yojana
The most widely accepted explanation for the ethnic preferences of voters is that in con-
stituencies where voters are not fully aware of the policy stand of the candidates or can-
didates cannot make binding policy commitments, ethnicity serves as an imperfect signal
of the policies a candidate might enact once elected (Chandra (2007); Banerjee and Pande
(2009); Vaishnav (2010)). To test for the importance of information in the formation of
co-ethnic preferences, I utilize a policy intervention that created exogenous variation in the
information cost across constituencies.
The Pradhan Mantri Gram Sadak Yojna (PMGSY) is a federal mandated rural road con-
struction program that started in 2000 and is still underway. As of 2014, the program had
cost more than four billion dollars and constructed 75 thousand kilometers of all weather
roads in the states of Uttar Pradesh and Bihar combined. The program aims to connect
all habitations with a population of at least 500 to the nearest link road via an all-weather
road.
19
Aggarwal (2014) and Banerjee and Sachdeva (2015) describe the program in detail
and evaluate the impact of the program along different dimensions of development. Aggar-
18
(website: http://eci.nic.in/)
19
“A habitation is a sub-village level entity, and is defined as ‘a cluster of population, whose location does
not change over time’” (Aggarwal (2014)).
109
wal (2014) finds that the program reduced time, money and information cost of accessing
markets. Banerjee and Sachdeva (2015) find that the program increased social interaction
within and between villages. Both these studies suggest that the flow of information had
improved significantly due to the program. The program, since it makes use of the ex-
ogenously determined rule, is exogenous to the process of candidate selection by political
parties for each constituency. Since the program spelled out clearly that the construction
was to be prioritized using aggregate, and not ethnic group-specific, population-based rules,
the possibility that some favored regions received the program before others is ruled out.
Data on PMGSY road construction development are available at the habitation level
from the Online Management and Monitoring System (OMMS) of Government of India.
Along with the status of connectivity during the 2001 census and as of May 2016, data also
list the village, the block, the district, the assembly constituency, and the parliamentary
constituency to which the habitat belongs. Unfortunately, the boundaries of the assembly
and parliamentary constituencies have changed since then due to the 2008 delimitation.
We need to match the habitat-level data on PMGSY with their corresponding assembly and
parliamentary constituencies.
In India, the elections data available publicly are usually at the level of constituen-
cies. The information on socioeconomic indicators, on the other hand, are available at the
level of administrative units of districts and sub-districts. The administrative boundaries
of the country, that is the delineation of district, sub-district and block level boundaries,
do not bear a close relationship to the parliamentary and assembly constituencies’ bound-
aries. Therefore, it is a daunting task to match the habitations with the new assembly and
parliamentary constituencies.
20
20
This ‘lack of fit between the two vital maps of India’, as Alam (2010) puts it, has not only caused ‘admin-
istrative and political problems’ (see Sivaramakrishnan (1997), Sivaramakrishnan (2000), Sivaramakrishnan
(2001)), but has also created ‘some major obstacles .... for students of public policy, politics and political
economy’ of India.
110
I use an approach most similar to Brass (1975).
21
The basic building block for the
delimitation exercise in Bihar are ‘wards’ in the urban areas and community development
blocks or gram panchayats in rural areas. In Uttar Pradesh, these are wards in the urban
areas and tehsils, kanungo cicles, or patwari circles in rural areas. These smallest rural and
urban units can be made up of one or more habitations.
22
The delimitation orders contain
the names of the blocks and wards that make up an assembly constituency.
There are 330,000 habitations from the two states in the PMGSY data. I aggregate the
information to the ward or the block level. I then assign each ward or block, which can
contain multiple villages, to a single assembly constituency. Most of the blocks in UP and
Bihar are contained entirely in one assembly constituency.
23
Some blocks are split among
more than one assembly constituency. For these blocks, I assign them to the assembly
constituency which contains, within it, a larger number of villages and habitations from
the block. Some habitations were certainly assigned to a constituency they did not belong
to. But there is no reason to believe that the measurement error in this assignment method
is systematically related to the candidate choice by political parties across constituencies.
Mapping the blocks to the parliamentary constituencies is accomplished by matching their
assembly constituencies to each assembly constituency’s corresponding parliamentary con-
stituencies, the details of which are contained in the Delimitation of Parliamentary and
Assembly Constituencies Order, 2008.
Next, for each assembly and parliamentary constituency, I calculate the percentage of
habitations that were connected to the towns via an all-weather road in 2001 and the per-
21
There have been numerous attempts at generating socio-economic profiles of parliamentary and assem-
bly constituencies, each with its own weaknesses and pitfalls (See Alam (2010), Bhandari (2009) and the
literature cited in Alam (2010) for details.)
22
While tehsils are larger than commuity development blocks, for the sake of brevity, I will use ’blocks’
to refer to both tehsils and community development blocks. Similarly, I will use ’villages’ to refer to both
patwari circles and gram panchayats.
23
For example, in Bihar, over 80 percent of the blocks were fully contained in one assembly constituency.
Splitting of blocks was relatively more frequent in UP.
111
centage connected by May of 2016. I use the difference between these two figures, the
increase in connectivity between 2001 and 2016, as a proxy for the increase in availability
of information within the constituencies. For each assembly (parliamentary) constituency, a
dummy variable ‘High connect’ takes a value of ‘1’ if the particular constituency’s increase
in connectivity between 2001 and 2016 is more than the median increase in connectivity
across assembly (parliamentary) constituencies in the state, and ‘0’ otherwise. If the model
fits the data well, interaction of ‘High connect’ with the main variables of interest in (3.11)
should work towards muting the role of the main variables in the equation.
3.5 Results
Table 3.2 presents the results using the 2014 and 2009 PE elections data for the two states.
As per the model’s prediction, the higher the percentage of general ethnicity voters in the
constituency, the higher is the probability that a party will field a general candidate. A
one percent increase in the percentage population of general ethnicity voters increases the
probability of fielding a general candidate by 2.5 to 5 percentage points. To test that this
positive association is not as a result of the mechanical correlation that would arise even if
the parties chose a candidate randomly from the voter population, I test the null hypothesis
of whether this coefficient equal to 1. The main coefficient are much larger in magnitude
and test significantly different than 1 at the 1% level of confidence for all specifications.
Later in this section, while discussing the role of information in this model, we provide
additional evidence against the mechanical correlation hypothesis.
The number of general-ethnicity rival candidates significantly lowers the probability
that a party will choose to field a candidate of general ethnicity. The coefficient for number
of reserved-ethnicity rival candidates variable turns out to be insignificant. The direction,
however, suggests that candidates from the reserved ethnicity add to the competition that
112
general candidates face. Consistent with the prediction from the model, the magnitude of
this coefficient is smaller than the competition effect from general-ethnicity rival candi-
dates. The third and fourth column include only those parties who field candidates in at
least two constituencies. It is possible that a party contesting elections in only one con-
stituency is too small to have potential candidates from both ethnic groups to be able to
diversify strategically.
24
Overall, there is evidence in favor of the model at the parliamen-
tary level.
Table 3.3 presents the results for the assembly elections using data from 2010 Bihar AE
and 2012 UP AE. While the percentage of general-ethnicity voters is still positively and
significantly related to the probability that a party will chose a general-ethnicity candidate,
the magnitude is lower than for the parliamentary elections. The coefficient for percentage
general-ethnicity voter is still significantly different than 1 at the 1% level of confidence for
all specifications. The number of general-ethnicity rival candidates and reserved-ethnicity
rival candidates both have a negative impact on the probability of fielding a general candi-
date. However, general-ethnicity rival candidates seem to have a smaller, and statistically
insignificant, competition effect than reserved-ethnicity rival candidates.
It is not entirely clear why it might be so. One explanation could be that since assembly
constituencies are much smaller compared to the parliamentary constituencies, voters, due
to more frequent interactions with the candidates, are better informed about the policy
stands of the candidates and stop relying on ethnicity as a signal. Smaller size of the
constituency might also mean more accountability for the candidate elected.
25
Alternative
measures to influence voter, like election campaigning, might be much more effective at
the assembly constituency level due to the higher acquaintance of the candidate with the
24
Using larger cutoffs for this size variable does not affect the results.
25
Heath and Kumar (2012) find that the SC and ST populations of the state of Uttar Pradesh were unsat-
isfied with the performance of BSP government after the 2007 assembly elections suggesting that they had
some information about the workings of the state government.
113
issues facing that constituency. Also, since diversity in ethnic identities exists throughout
India, most of the policies that pertain to ethnic identities are debated on, formulated, and
legislated at the parliament level, and not in the state assemblies. As a result, having co-
ethnic representatives in the parliament might matter more than having co-ethnics in the
state assembly.
26
A commonality in these explanations is the crucial role of information. To test for
the importance of information, we interact the dummy variable ‘High connect’, denoting
high increase in road connectivity of the constituency between 2001 and 2016, described
in 3.4.2, with the main variables of our interest - percentage of general ethnicity voters,
number of general ethnicity rival candidates, and the number of reserved ethnicity rival
candidates. The results are presented in table 3.4. Column (1) presents the results for the
parliamentary elections. For regions with low increase in connectivity, the coefficient for
rival-general ethnicity candidates drops below that for rival reserved-ethnicity candidates
and is statistically insignificant. But the coefficients are in the expected direction. Interest-
ingly, the coefficient for interaction of high connectivity dummy with the main variables
suggest that the model is not a good predictor of the candidate selection process in regions
that saw a large increase in connectivity. The interaction of ‘High connect’ with percent-
age of general voters and with rival candidates of reserved ethnicity have coefficients with
signs opposite of their independent effects. If PMGSY indeed increased the extent to which
the voters in the constituencies had access to relevant information, the results suggest that
political parties do not utilize strategies that rely on co-ethnic voting to choose candidates
in regions where voters are more informed. The results also suggest that the positive co-
efficient that we observe for percentage general-ethnicity voter population variable is not
26
A similar argument has been made in Jaffrelot (2003) and Chandra (2007). They argue that the polit-
ical salience of caste identities become more pronounced in contexts where affirmative action policies are
involved. Historically, in post independence India, question of seat reservations and quotas for people to
redress caste discrimination are debated at the parliamentary level.
114
the result of the mechanical increase in the probability of selecting a general candidate at
random due to an increase in percentage general-ethnicity voter population.
For assembly elections, results presented in column (2), the independent coefficients
are in the same direction and of a similar magnitude as those in table 3.3. The interactions,
however, are close to zero in magnitude and insignificant. This is consistent with the re-
sults from table 3.3 - parties, to attract voters, do not rely as much on choosing co-ethnic
candidates in assembly elections as they do in parliamentary elections.
But does co-ethnic voting and this strategic candidate-choice result in undesirable out-
comes? To examine this, we compare the outcomes in general constituencies with open
elections with that from reserved constituencies where there is lower degree of ethnic vot-
ing and diversification along ethnic lines. As an indicator of quality of the outcome, we
use information on the number of criminal cases ever filed against the candidates running
for office from a constituency. Evidence suggests that criminality affects economic ac-
tivity and public good provision within the constituency (Banerjee et al. (2014);Prakash
et al. (2015)). Tables 3.5 and 3.6 present the results. Columns (1) from the tables show
that constituencies where only reserved ethnicity candidates can run for office tend to have
candidates with fewer criminal cases against them in both parliamentary and assembly
constituencies. Consistent with our hypothesis that the strategic candidate diversification
is less frequent in assembly constituency, the negative correlation between the reserved
constituency dummy and the number of criminal cases is higher for parliamentary con-
stituencies. Vaishnav (2010), too, finds that candidates are more likely to have criminal
records in general constituencies.
One explanation for higher criminality of fielded candidates in general constituencies
can be that voters in general constituencies have a preference for criminal representatives.
For example, such a situation may arise if criminal candidates are better at ensuring larger
benefits for their own ethnic groups (Vaishnav (2010)). This would be visible in the differ-
115
ence in distribution of public resources across ethnic groups in general and reserved con-
stituencies. However, studies on the impact of affirmative action in Indian politics find no
evidence of any distribution effects. Another explanation could be that since candidate di-
versification along ethnic lines is possible and profitable in general constituencies, political
parties, in these constituencies, place a much higher weight on the ethnicity of a potential
candidate. This reduces the importance of other desirable candidate characteristics in the
selection process and candidates with worse criminal history get fielded. Since such a di-
versification is not possible in reserved constituencies, the importance of other qualities,
like the absence of a criminal record, increase the probability of a potential candidate’s
chances of being fielded.
In column (2) of tables 3.5 and 3.6, we examine which of the two explanations ac-
counts better for the differences in the criminal record of candidates across these two types
of constituencies. The dependent variable is the number of criminal cases against the can-
didate finally elected to office from the constituency. The explanatory variable ‘proportion
of criminal candidates’ captures the inefficiency that arises from a worse pool of candidates
fielded by political parties in these constituencies in order to be able to diversify ethnically.
The ‘reserved constituency’ dummy now captures any differences across who is elected to
office conditional on the quality of the pool of candidates fielded from a constituency and
includes differences in preference for criminal candidates. It is evident that what explains
the election of a criminal candidate to office is the worse quality of the pool of candi-
dates voters choose from. In terms of their preference for criminal candidates, voters in
the two types of constituencies do not differ. Also, consistent with less frequent strate-
gic candidate-diversification in assembly constituency, association between the proportion
of criminal candidates and the number of criminal cases against the winner is smaller in
assembly constituencies than in parliamentary constituencies.
One characteristic of any candidate which is readily observable by all parties is whether
116
or not a candidate is the incumbent in an office. An incumbent is much more likely to be
fielded for elections since her victory in the previous elections is evidence of her vote-
garnering ability. The presence of an incumbent might affect selection process consider-
ably. For example, consider the case where party A fields the incumbent of general ethnicity
from a constituency. As per the model, this should decrease the probability that other par-
ties will field a candidate of general ethnicity. However, an incumbent of general ethnicity
might have a different competition effect than other candidates of general ethnicity. It is
possible that parties might perceive voters of the general ethnic group to be more loyal to
the incumbent than to any other candidate of the general ethnicity. Choosing a candidate
of the general ethnicity might become even less attractive in this scenario. Alternatively,
it is possible that the other parties in the constituency see the win of the incumbent in the
previous election as a signal of high degree of loyalty of the voters to the ethnicity of the
incumbent candidate, and not to the incumbent candidate herself. It might, therefore, be
more profitable to run a general-ethnicity candidate in such a constituency.
We examine the impact of an incumbent contesting elections in table 3.7. Since con-
stituency boundaries changed in 2008, we assume that there were no incumbents in 2009
PE elections.
27
For the parliamentary elections in 2014, we construct a dummy variable
which takes value ‘1’ for all candidates in a constituency if an incumbent of 2009 elections
was contesting again from that constituency in 2014.
28
All the incumbents in 2014 in the
two states, it turns out, are from the general ethnicity. As a result, we cannot distinguish
between the impact of an incumbent contesting elections and the impact of an incumbent
of general ethnicity contesting elections. In column (1) of table3.7, we include a dummy
indicator of whether the constituency has an incumbent contesting for office. The impact is
statistically insignificant and the statistical significance and magnitude of the main variables
27
This is a simplifying assumption that further research on the topic should seem to relax.
28
We cannot do a similar analysis for the impact of incumbency for the assembly elections since we have
information from only one round of assembly elections from each of the states.
117
in the model do not change much. In column (2), we interact the dummy capturing the pres-
ence of an incumbent with the main variables in the model. Interestingly, the presence of a
general-ethnicity incumbent candidate seems to increase the probability of parties fielding
a general candidate in constituencies with a lower percentage of general-ethnicity voters.
However, as the percentage of general-ethnicity voters increase, the presence of a general-
ethnicity incumbent candidate is associated with a lower probability of parties fielding a
general candidate from the constituency. The results suggest that parties might perceive the
win of a general-ethnicity candidate in the previous elections in constituencies with a lower
percentage of general-ethnicity voters as a signal of high loyalty of the general-ethnicity
voters to candidates of their ethnicity. Therefore, it might be more profitable to run a gen-
eral ethnicity in such constituencies. However, in constituencies with a higher percentage
population of general-ethnicity voters, the presence of a general-ethnicity incumbent might
mean a loyalty to the incumbent candidate. If so, fielding another general-ethnicity candi-
date might not be very profitable.
Taken together, the results suggest that strategic diversification of candidates along eth-
nic lines by political parties in response to their perceptions that voters vote co-ethnically,
especially in low information settings, leads to worse electoral outcomes.
3.6 Revisiting the assumptions
The model and the empirical verification of the model’s predictions makes some assump-
tions that merit further discussion. The assumption that forms the core of the model is that
political parties, in any constituency, care only about the percentage of votes. One might
argue that parties like BSP and SP have strong ethnic affiliations and might choose candi-
dates in accordance with that affiliation regardless of the vote profits from that candidates
in order to better represent the ethnic interests. There are at least three arguments that can
118
be made against it. First is the evidence from 2007 UP assembly elections where the BSP
chose to run many candidates from the general category even though it is known for its
strong affiliation to the SC’s and antagonistic outlook towards the general ethnic group.
Second, for these parties which have strong affiliations, the focus seems to be on having
representatives of the ethnicity in the government and not in every constituency. An ethni-
cally affiliated party might be open to running a candidate from any ethnicity as long as the
majority of the elected representatives it forms the ministry with are from the ethnicity it
is affiliated with. Third, ethnic affiliations can be and are incorporated in this model. Con-
sider this hypothetical case where three parties are contesting the election from a particular
constituency. Out of the three, one party has very strong ethnic affiliation. In that case, the
other two parties may be quite certain that the first party will run a candidate of the ethnic-
ity it is affiliated with. The remaining parties might strategically diversify their candidates
conditional on the first party running a candidate of a certain ethnicity. The game of strate-
gic diversification is played between all parties who do not have already committed to run
a candidate of a particular ethnicity. Strategic diversification may occur as long as as long
as at least one party is open to fielding candidates from any ethnicity. The impact of the
decrease in the number of parties strategically diversifying from that constituency will be
subsumed in the party fixed effects (party-specific characteristics) or party constituency
(party-constituency-specific characteristics) fixed effect.
One might argue that the assumption of asymmetric information is not as realistic and
that political parties might have all the relevant information about all the potential candi-
dates that every rival party is considering to run from a particular constituency. This is
highly unlikely. Usually, parties do not have very good information about the identity of all
the potential candidates that a rival party is considering. But let us consider that some party,
say party A, has this information about a rival party, party B. With complete information,
party A will be able to make the same calculation about party B’s candidate-ethnicity of
119
choice as party B. Party A, then, will treat party B’s choice as given and act strategically
against all other rival parties. Such a change will be subsumed in the constituency fixed ef-
fect (constituency-specific characteristics) in our model. Strategic candidate diversification
may occur even if there is one party without complete information about all the candidates
of at least one rival party.
A more plausible situation is that all parties might know exactly what candidate a par-
ticular party is going to field from a constituency. For example, in our context, it is well-
known that parliamentary elections from Amethi constituency for INC are almost always
contested by the most influential member of the Gandhi family. In such a scenario, other
parties will strategically diversify conditional on this knowledge. Such a situation is also
be accounted for in our model due to the inclusion of constituency fixed effects.
3.7 Conclusion
Corruption impedes economic growth (Shleifer and Vishny (1993); Mauro (1995)). Pre-
vious literature views the election of corrupt and criminal candidates as the failure of the
voters. Banerjee and Pande (2009) argue that ethnic biases result in selection of lower qual-
ity politicians. This paper suggest that strategic action of political parties might be a bigger
reason for selection if low quality candidates. In constituencies with open elections, parties
strategically diversify their candidates along the dimension of ethnicity and, in the process,
compromise on quality. In constituencies which are reserved for SC/ST candidates, diver-
sification along the ethnicity dimension is not possible. Therefore, the competition is more
intense. As a result, parties ensure that the candidates they run are not ‘tainted’. While
reserving all constituency for candidates of one or the other ethnicity might be a short-run
fix, it may work towards making these ethnic rivalries more pronounced. To address this
inferior candidate selection, we need to address the co-ethnic political preference. One way
120
in which that can be done is by providing the voters with more information about the can-
didate. If ethnicity works as signals for the policy stand of the candidates, this will weaken
the co-ethnic preferences.
The model presented in the paper is by no means a complete description of the complex
selection procedure for the candidates that a party uses. A more comprehensive model
should allow for other forces, like religion, gender, incumbency, etc., to play a role. This
paper abstracts away from some of these dimensions and attempts to model the candidate
selection process in the simple manner that still manages to explain certain stylized facts
about elections in India. Moreover, the selection process described by the model is more
of a second stage to a selection of parties into constituency, which is of great interest and
should be the topic of future research. Also, It is possible that the ethnic preferences of the
voters might evolve over time. A more sophisticated model should attempt to account for
these.
121
TABLE 3.2: PARLIAMENTARY ELECTIONS
(1) (2) (3) (4)
(PE (All): Probit) (PE (All): Logit) (PE ( 2):Probit) ((PE ( 2): Logit)
V ARIABLES General-ethnicity candidate selected
General V oters (%) 0.0256*** 0.0502*** 0.0280*** 0.0550***
(0.0086) (0.0168) (0.0094) (0.0182)
Number of general-ethnicity rival candidates -0.0407*** -0.0797*** -0.0318* -0.0635**
(0.0154) (0.0306) (0.0163) (0.0321)
Number of reserved-ethnicity rival candidates -0.0266 -0.0574 -0.0145 -0.0335
(0.0504) (0.0993) (0.0555) (0.1116)
Wald
2
forH
0
:
1
= 1 13193.50 3258.66 11233.15 2841.60
YEAR FE YES YES YES YES
STATE FE YES YES YES YES
Observations 2,071 2,071 1,908 1,908
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at parliamentary constituency level.
122
TABLE 3.3: ASSEMBLY ELECTIONS
(1) (2) (3) (4)
(AE (All): Probit) (AE (All): Logit) (AE ( 2): Probit) ((AE ( 2): Logit)
V ARIABLES General-ethnicity candidate selected
General V oters (%) 0.0207*** 0.0427*** 0.0207*** 0.0428***
(0.0039) (0.0078) (0.0039) (0.0079)
Number of general-ethnicity rival candidates -0.0088 -0.0174 -0.0086 -0.0171
(0.0075) (0.0147) (0.0075) (0.0148)
Number of reserved-ethnicity rival candidates -0.0839*** -0.1617*** -0.0862*** -0.1666***
(0.0259) (0.0502) (0.0262) (0.0508)
Wald
2
forH
0
:
1
= 1 62389.15 14730.55 61228.39 14456.51
STATE FE YES YES YES YES
Observations 6,117 6,117 6,053 6,053
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at assembly constituency level.
123
TABLE 3.4: PMGSY AND CANDIDATE SELECTION
V ARIABLES General-ethnicity candidate selected in
(PE) (AE)
General V oters (%) 0.0403*** 0.0281***
(0.0127) (0.0062)
Number of general-ethnicity rival candidates -0.0323 -0.0107
(0.0262) (0.0113)
Number of reserved-ethnicity rival candidates -0.0786* -0.0711*
(0.0409) (0.0380)
High increase in connectivity due to PMGSY 3.5687** -0.1074
(1.5184) (0.8011)
General V oters (%) * High Connect (%) -0.0413** 0.0008
(0.0187) (0.0102)
General-ethnicity rival candidates * High Connect -0.0579 0.0005
(0.0412) (0.0146)
Reserved-ethnicity rival candidates * High Connect 0.3758*** 0.0478
(0.1390) (0.0625)
Observations 2,002 5,326
YEAR FE YES NO
STATE FE YES YES
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard
errors are clustered at the constituency level.
TABLE 3.5: CRIMINAL CANDIDATES IN PE
(1) (2)
V ARIABLES Criminal cases against candidates Criminal cases against winner
Reserved constituency dummy -0.5046*** -0.2663
(0.1161) (0.6463)
Proportion of criminal candidates 5.2378**
(2.1388)
Observations 2,386 237
R-squared 0.1802 0.1160
STATE FE YES YES
YEAR FE YES YES
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The
standard errors are clustered at parliamentary constituency level.
124
TABLE 3.6: CRIMINAL CANDIDATES IN AE
(1) (2)
V ARIABLES Criminal cases against candidates Criminal cases against winner
Reserved constituency dummy -0.3946*** -0.3883
(0.0392) (0.2931)
Proportion of criminal candidates 2.1978**
(0.8639)
Observations 8,979 625
R-squared 0.0764 0.2044
STATE FE YES YES
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The
standard errors are clustered at assembly constituency level.
TABLE 3.7: INCUMBENTS IN PARLIAMENTARY ELECTIONS
(1) (2)
V ARIABLES General-ethnicity candidate selected
General V oters (%) 0.0486*** 0.1079***
(0.0185) (0.0230)
Number of general-ethnicity rival candidates -0.0812** -0.1175***
(0.0319) (0.0382)
Number of reserved-ethnicity rival candidates -0.0592 -0.0345
(0.0977) (0.0929)
Constituency incumbent of 2009 contesting elections in 2014 -0.1132 9.8967***
(0.2504) (3.3163)
General V oters (%) * Incumbent -0.1288***
(0.0372)
General-ethnicity rival candidates * Incumbent 0.0263
(0.0906)
Reserved-ethnicity rival candidates * Incumbent 0.1607
(0.1768)
Observations 2,071 2,071
YEAR FE YES YES
STATE FE YES YES
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The
standard errors are clustered at assembly constituency level.
125
Conclusion
Over the last fifty years, significant progress has been made in improving global primary
school enrollments. The gross primary enrollment rate in Sub-Saharan Africa has gone up
from a little over 50% in 1970 to 90% in 2014. South Asia has moved from around 70 %
in 1970 to 109 % in 2014. Rates of primary school completion have gone up from around
40% in 1970 to close to 70% in Sub-Saharan Africa and over 90% in South Asia. However,
universal primary education is still a distant goal.
The low educational attainment of those who are still left out can often be traced back
to ‘development insults’, their early exposure to adverse conditions such as illness and/or
inappropriate feeding practices, that have long lasting impact on their educational attain-
ment. Against this backdrop, the first chapter of this dissertation points out that, at least
along the dimension of education, the negative impact of these early life adverse circum-
stances can be mitigated with timely corrective policies. This provides a new impetus to
the public policy aimed at alleviating poverty of education and wealth.
However, it is worth noting that improvements in educational attainments in the last
fifty years have not translated proportionately into reductions in poverty. Over 50% of the
population of South Asia and 66.5% of the population in Sub-Saharan Africa were living
under the $3.20 per day PPP poverty line in 2013. Studies have found that even though
children are going to school more often, many of them lack basic reading, writing, and
arithmetic skills even after completing primary school (Bold et al. (2017a)). School infras-
tructure, like pupil-teacher ratio and teaching pedagogy, is pivotal to learning in classes
(Bold et al. (2017b)). Most of the increase in enrollment in primary enrollment across the
developing world in the recent years have been as a result of reduction or abolition of pri-
126
mary school fees or similar large-scale supply side measures. As the chapter documents,
while such measures might successfully increase enrollment, they might often strain the
existing school infrastructure leading to low levels of learning. Anticipating such increases
in demand in advance and planning steps, like recruitment and training teachers, to be able
to timely respond to these demands should be a part of the design of any such large scale
program.
The prevailing economic environment of a region also has important bearing on what
and how much the education policies can achieve. The second chapter documents that,
in Tanzania, the seemingly smarter individuals delayed enrollment and worked at home
or on the family farm - a behavior that is not completing surprising in an economy based
around traditional modes of production where returns to experience may be higher than
returns to education. When educational policies are not synchronized with the demands of
the local economic environment, improved educational attainments, even in places where
it is achieved, may fail to translate into growth and welfare in the short-run. And the lack
of returns to investment in education might depress the demand for education further. In
such scenario, education policies need to be complemented with policies like introduction
of subsides for modern agricultural equipment, so as to improve the returns from education
even in the short-run.
Such streamlining of government policies will require capable legislators at the helm.
However, developing countries often also impaired by high degree of incompetence and
corruption amongst its political leaders and policy makers. The third chapter shows that
political parties in these regions believe that they can win elections to an office by appeal-
ing to the ethnic preferences of voters. In an attempt to field a candidate of an ethnicity
that will maximize their vote profit, they disregard the inferior qualities of the candidates
and present the voters a worse pool of candidates to choose from. However, evidence
suggests that parties do not engage in this strategic behavior when the constituencies are
127
smaller, cost of amassing information is lower, or the voters are more educated. This sug-
gests that education might have non-pecuniary benefits in the short-run, which, in turn,
might improve economic welfare through more efficient planning and implementation of
government policies.
128
Bibliography
Acharya, A., Roemer, J. E., and Somanathan, R. (2014). Caste, corruption and political
competition in India.
Adelman, I. (1975). Development economics–a reassessment of goals. American Economic
Review, pages 302–309.
Adhvaryu, A., Molina, T., Nyshadham, A., and Tamayo, J. (2015). Helping children catch
up: Early life shocks and the Progresa experiment. Technical report, Mimeo.
Adhvaryu, A. and Nyshadham, A. (2014). Endowments at birth and parents’ investments
in children. Economic Journal.
Adhvaryu, A. and Nyshadham, A. (2016). Endowments at birth and parents’ investments
in children. Economic Journal, 126(593):781–820.
Aggarwal, S. (2014). Do rural roads create pathways out of poverty? evidence from india.
University of California, Santa Cruz, unpublished.
Aguilar, A. and Vicarelli, M. (2011). El Nino and Mexican children: Medium-term ef-
fects of early-life weather shocks on cognitive and health outcomes. Cambridge, United
States: Harvard University, Department of Economics. Manuscript.
Aizer, A., Eli, S., Ferrie, J., and Lleras-Muney, A. (2016). The long-run impact of cash
transfers to poor families. American Economic Review, 106(4):935–971.
Alam, M. S. (2010). On matching census tracts and electoral boundaries: The bottom-up
aggregation approach. Economic and Political Weekly, pages 64–72.
129
Alderman, H., Hoddinott, J., and Kinsey, B. (2006). Long term consequences of early
childhood malnutrition. Oxford Economic Papers, 58(3):450–474.
Almond, D. and Currie, J. (2011). Killing me softly: The fetal origins hypothesis. Journal
of Economic Perspectives, 25(3):153–172.
Almond, D., Currie, J., and Duque, V . (2017). Childhood circumstances and adult out-
comes: Act II. Journal of Economic Literature, 103.
Almond, D., Hoynes, H. W., and Schanzenbach, D. W. (2011). Inside the war on poverty:
The impact of food stamps on birth outcomes. Review of Economics and Statistics,
93(2):387–403.
Almond, D. and Mazumder, B. (2013a). Fetal origins and parental responses. Annual
Review of Economics, 5(1):37–56.
Almond, D. and Mazumder, B. (2013b). Fetal origins and parental responses. Annual
Review of Economics, 5(1):37–56.
Banerjee, A., Green, D. P., McManus, J., and Pande, R. (2014). Are poor voters indifferent
to whether elected leaders are criminal or corrupt? a vignette experiment in rural india.
Political Communication, 31(3):391–407.
Banerjee, A. and Pande, R. (2009). Parochial politics: Ethnic preferences and politician
corruption. mimeo.
Banerjee, R. and Sachdeva, A. (2015). Pathways to preventive health, evidence from india’s
rural road program. USC-INET Research Paper, (15-19).
Bank, W. (1990). Indonesia: Strategy for a sustained reduction in poverty. World Bank.
130
Barker, D. J. (1990). The fetal and infant origins of adult disease. Bmj, 301(6761):1111–
1111.
Barker, D. J. (1995). Fetal origins of coronary heart disease. BMJ: British Medical Journal,
311(6998):171.
Barker, D. J., Eriksson, J. G., Forsen, T., and Osmond, C. (2002). Fetal origins of adult
disease: strength of effects and biological basis. International journal of epidemiology,
31(6):1235–1239.
Barrera, A. (1990). The role of maternal schooling and its interaction with public health
programs in child health production. Journal of Development Economics, 32(1):69–91.
Bates, R. H. (1983). Modernization, ethnic competition, and the rationality of politics in
contemporary africa. State versus ethnic claims: African policy dilemmas, 152:171.
Becker, G. S. and Mulligan, C. B. (1997). The endogenous determination of time prefer-
ence. Quarterly Journal of Economics, 112(3):729–758.
Becker, G. S. and Tomes, N. (1986). Human capital and the rise and fall of families. Journal
of labor economics, pages S1–S39.
Beegle, K., De Weerdt, J., and Dercon, S. (2006a). Kagera health and develop-
ment survey 2004 basic information document. The World Bank. www. worldbank.
com/lsms/country/kagera2/docs/KHDS2004% 20BID% 20feb06. pdf[accessed March
13, 2007].
Beegle, K., Dehejia, R. H., and Gatti, R. (2006b). Child labor and agricultural shocks.
Journal of Development economics, 81(1):80–96.
Behrman, J. R. and Birdsall, N. (1983). The quality of schooling: Quantity alone is mis-
leading. American Economic Review, 73(5):928–946.
131
Behrman, J. R. and Deolalikar, A. B. (1987). Will developing country nutrition im-
prove with income? A case study for rural South India. Journal of Political Economy,
95(3):492–507.
Behrman, J. R., Parker, S. W., and Todd, P. E. (2011). Do conditional cash transfers for
schooling generate lasting benefits? A five-year followup of Progresa/Oportunidades.
Journal of Human Resources, 46(1):93–122.
Bengtsson, N., Peterson, S., and S¨ avje, F. (2013). Revisiting the educational effects of fetal
iodine deciency.
Besley, T., Persson, T., and Sturm, D. (2005). Political competition and economic perfor-
mance: Theory and evidence from the United states. Technical report, National Bureau
of Economic Research.
Bhandari, L. (2009). Socio-economic performance of constituencies: A response [with
reply]. Economic and Political Weekly, pages 61–63.
Bharadwaj, P., Løken, K. V ., and Neilson, C. (2013). Early life health interventions and
academic achievement. American Economic Review, 103(5):1862–1891.
Birdsall, N. (1985). Public inputs and child schooling in Brazil. Journal of Development
Economics, 18(1):67–86.
Black, S. E., Devereux, P. J., Løken, K. V ., and Salvanes, K. G. (2014). Care or cash?
The effect of child care subsidies on student performance. Review of Economics and
Statistics, 96(5):824–837.
Blau, D. M., Guilkey, D. K., and Popkin, B. M. (1996). Infant health and the labor supply
of mothers. Journal of Human Resources, pages 90–139.
132
Bobonis, G. J. (2009). Is the allocation of resources within the household efficient? New
evidence from a randomized experiment. Journal of Political Economy, 117(3):453–503.
Bold, T., Filmer, D., Martin, G., Molina, E., Stacy, B., Rockmore, C., Svensson, J., and
Wane, W. (2017a). Enrollment without learning: Teacher effort, knowledge, and skill in
primary schools in africa. Journal of Economic Perspectives, 31(4):185–204.
Bold, T., Filmer, D., Martin, G., Molinad, E., Rockmore, C., Stacy, B., Svensson, J., and
Wane, W. (2017b). What do teachers know and do? Does it matter? Evidence from
primary schools in Africa. World Bank Policy Research Working Paper, 7956.
Bold, T., Kimenyi, M., Mwabu, G., and Sandefur, J. (2014). Can free provision reduce
demand for public services? Evidence from Kenyan education. World Bank Economic
Review, 29(2):293–326.
Bommier, A. and Lambert, S. (2000). Education demand and age at school enrollment in
tanzania. Journal of Human Resources, pages 177–203.
Bouis, H. E. and Haddad, L. J. (1992). Are estimates of calorie-income fxelasticities
too high?: A recalibration of the plausible range. Journal of Development Economics,
39(2):333–364.
Bound, J., Lovenheim, M. F., and Turner, S. (2010). Why have college completion rates de-
clined? An analysis of changing student preparation and collegiate resources. American
Economic Journal: Applied Economics, 2(3):129–157.
Bowman, M. J. (1966). The human investment revolution in economic thought. Sociology
of Education, 39(2):111–137.
Boyd, D., Lankford, H., Loeb, S., and Wyckoff, J. (2005). The draw of home: How teach-
133
ers’ preferences for proximity disadvantage urban schools. Journal of Policy Analysis
and Management, 24(1):113–132.
Brass, P. R. (1975). Ethnic cleavages and the Punjab party system, 1952-1972. Studies in
Electoral Politics in the Indian States: The Impact of Modernisation, 4:3–62.
Brown, J. L. and Pollitt, E. (1996). Malnutrition, poverty and intellectual development.
Scientific American, 274(2):38–43.
Brunner, E., Marmot, M., Canner, R., Beksinska, M., Smith, G. D., and O’brien, J. (1996).
Childhood social circumstances and psychosocial and behavioural factors as determi-
nants of plasma fibrinogen. Lancet, 347(9007):1008–1013.
Burke, K. and Beegle, K. (2004). Why children aren’t attending school: the case of north-
western tanzania. Journal of African Economies, 13(2):333–355.
Burke, K. A. (1998). Investing in children’s human capital in Tanzania: does household
illness play a role? PhD thesis, State University of New York at Stony Brook.
Cameron, A. C., Gelbach, J. B., and Miller, D. L. (2008). Bootstrap-based improvements
for inference with clustered errors. The Review of Economics and Statistics, 90(3):414–
427.
Cao, X.-Y ., Jiang, X.-M., Dou, Z.-H., Rakeman, M. A., Zhang, M.-L., O’Donnell, K., Ma,
T., Amette, K., DeLong, N., and DeLong, G. R. (1994). Timing of vulnerability of
the brain to iodine deficiency in endemic cretinism. New England journal of medicine,
331(26):1739–1744.
Card, D. (1994). Earnings, schooling, and ability revisited. Technical report, National
Bureau of Economic Research.
134
Card, D. and Krueger, A. B. (1992). Does school quality matter? Returns to education and
the characteristics of public schools in the United States. Journal of Political Economy,
100(1):1–40.
Chandra, K. (2007). Why ethnic parties succeed: Patronage and ethnic head counts in
India. Cambridge University Press.
Chetty, R., Friedman, J. N., Hilger, N., Saez, E., Schanzenbach, D. W., and Yagan, D.
(2011). How does your kindergarten classroom affect your earnings? evidence from
project STAR. Quarterly Journal of Economics, 126(4):1593–1660.
Cornwell, K. and Inder, B. (2015). Child health and rainfall in early life. Journal of
Development Studies, 51(7):865–880.
Counts, G. S. (1931). Education: History. Encyclopaedia of the Social Sciences, 5:403–
414.
Cunha, F. and Heckman, J. (2007). The technology of skill formation. American Economic
Review, 97(2):31–47.
Cunha, F., Heckman, J. J., and Schennach, S. M. (2010). Estimating the technology of
cognitive and noncognitive skill formation. Econometrica, 78(3):883–931.
Currie, J. and Almond, D. (2011). Human capital development before age five. Handbook
of labor economics, 4:1315–1486.
Currie, J. and Thomas, D. (2000). School quality and the longer-term effects of head start.
Journal of Human Resources, 35(4):755–774.
Dahl, G. B. and Lochner, L. (2012). The impact of family income on child achievement:
Evidence from the earned income tax credit. American Economic Review, 102(5):1927–
1956.
135
Dahl, R. A. (2005). Who governs?: Democracy and power in an American city. Yale
University Press.
Davis, L. E., North, D. C., and Smorodin, C. (1971). Institutional change and American
economic growth. Number 123. CUP Archive.
De Vreyer, P., Sylvie, L., and Thierry, M. (1998). Educating children: a look at family
behaviour in cote d’ivoire. Unpublished manuscript.
De Weerdt, J., Beegle, K., Lilleør, H. B., Dercon, S., Hirvonen, K., Kirchberger, M., and
Krutikova, S. (2012). Kagera health and development survey 2010: Basic information
document. Technical report, Rockwool Foundation Working Paper Series.
Denison, E. F. et al. (1962). Sources of economic growth in the united states and the
alternatives before us.
Downs, A. (1957). An economic theory of political action in a democracy. Journal of
Political Economy, pages 135–150.
Duflo, E. (2000). Schooling and labor market consequences of school construction in
Indonesia: Evidence from an unusual policy experiment. Technical report, National
Bureau of Economic Research.
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.
Duflo, E. (2004). The medium run effects of educational expansion: Evidence from a
large school construction program in Indonesia. Journal of Development Economics,
74(1):163–197.
136
Dugbartey, A. T. (1998). Neurocognitive aspects of hypothyroidism. Archives of internal
medicine, 158(13):1413–1418.
Easterlin, R. A. (1981). Why isn’t the whole world developed? Journal of Economic
History, 41(01):1–17.
Easterlin, R. A. (2009). Growth triumphant: the twenty-first century in historical perspec-
tive. University of Michigan Press.
Eltom, M., Karlsson, F., Kamal, A., Bostr¨ om, H., and Dahlberg, P. (1985). The effective-
ness of oral iodized oil in the treatment and prophylaxis of endemic goiter*. The Journal
of Clinical Endocrinology & Metabolism, 61(6):1112–1117.
Fafchamps, M., Udry, C., and Czukas, K. (1998). Drought and saving in West Africa: Are
livestock a buffer stock? Journal of Development Economics, 55(2):273–305.
Field, B. N. and Siavelis, P. M. (2011). Endogenizing legislative candidate selection pro-
cedures in nascent democracies: evidence from Spain and Chile. Democratization,
18(3):797–822.
Field, E., Robles, O., and Torero, M. (2009). Iodine deficiency and schooling attainment
in tanzania. American Economic Journal: Applied Economics, pages 140–169.
Fogel, R. W. (1994). Economic growth, population theory, and physiology: the bearing
of long-term processes on the making of economic policy. American Economic Review,
84(3):369–395.
Frankenberg, E., Smith, J. P., and Thomas, D. (2003). Economic shocks, wealth, and
welfare. Journal of Human Resources, 38(2):280–321.
137
Galler, J. R., Ramsey, F., Solimano, G., Lowell, W. E., and Mason, E. (1983). The influ-
ence of early malnutrition on subsequent behavioral development: I. Degree of impair-
ment in intellectual performance. Journal of the American Academy of Child Psychiatry,
22(1):8–15.
Gertler, P., Heckman, J., Pinto, R., Zanolini, A., Vermeersch, C., Walker, S., Chang, S. M.,
and Grantham-McGregor, S. (2014). Labor market returns to an early childhood stimu-
lation intervention in Jamaica. Science, 344(6187):998–1001.
Gillman, M. W. (2005). Developmental origins of health and disease. New England Journal
of medicine, 353(17):1848.
Gin´ e, X., Townsend, R., and Vickery, J. (2008). Patterns of rainfall insurance participation
in rural India. World Bank Economic Review, 22(3):539–566.
Glewwe, P. and Jacoby, H. (1993). Delayed primary school enrollment and childhood
malnutrition in Ghana: An economic analysis, volume 23. World Bank Publications.
Glewwe, P. and Jacoby, H. G. (1995). An economic analysis of delayed primary school
enrollment in a low income country: The role of early childhood nutrition. Review of
Economics and Statistics, pages 156–169.
Glewwe, P., Jacoby, H. G., and King, E. M. (2001). Early childhood nutrition and academic
achievement: A longitudinal analysis. Journal of Public Economics, 81(3):345–368.
Glewwe, P., Muralidharan, K., et al. (2016). Improving education outcomes in developing
countries. Handbook of the Economics of Education, 5:653–743.
Gould, E. D., Lavy, V ., and Paserman, M. D. (2011). Sixty years after the magic carpet
ride: The long-run effect of the early childhood environment on social and economic
outcomes. Review of Economic Studies, 78(3):938–973.
138
Grant, J. P. (1978). Disparity reduction rates in social indicators. Overseas Development
Council.
Gunnsteinsson, S., Adhvaryu, A., Christian, P., Labrique, A., Sugimoto, J., Shamim, A. A.,
and West Jr, K. P. (2014). Resilience to early life shocks. Technical report, Technical
report.
Hanushek, E. A. (2003). The failure of input-based schooling policies. Economic Journal,
113(485).
Hanushek, E. A. and Rivkin, S. G. (2006). Teacher quality. Handbook of the Economics of
Education, 2:1051–1078.
Hapsariputri, W. M. (2010). Evaluating teachers’ quality improvement policy in Indonesia:
To meet the UNESCO-EFA criteria.
Hatt, L., Stanton, C., Makowiecka, K., Adisasmita, A., Achadi, E., and Ronsmans, C.
(2007). Did the strategy of skilled attendance at birth reach the poor in Indonesia? Bul-
letin of the World Health Organization, 85(10):774–782.
Heath, O. and Kumar, S. (2012). Why did dalits desert the Bahujan Samaj Party in Uttar
Pradesh? Economic and Political Weekly, 47(28).
Heindel, J. J. and Vandenberg, L. N. (2015). Developmental origins of health and dis-
ease: A paradigm for understanding disease etiology and prevention. Current opinion in
pediatrics, 27(2):248.
Hoddinott, J. and Kinsey, B. (2001). Child growth in the time of drought. Oxford Bulletin
of Economics and statistics, 63(4):409–436.
Hoddinott, J. and Skoufias, E. (2004). The impact of progresa on food consumption. Eco-
nomic development and cultural change, 53(1):37–61.
139
Horowitz, D. L. (1985). Ethnic groups in conflict. Univ of California Press.
Horton, S. (1986). Child nutrition and family size in the Philippines. Journal of Develop-
ment Economics, 23(1):161–176.
Hotelling, H. (1929). An empirical model of firm entry with endogenous product-type
choices. Economic Journal, 39:41–57.
Hoynes, H., Miller, D., and Simon, D. (2015). Income, the earned income tax credit, and
infant health. American Economic Journal: Economic Policy, 7(1):172–211.
Hoynes, H., Schanzenbach, D. W., and Almond, D. (2016). Long-run impacts of childhood
access to the safety net. American Economic Review, 106(4):903–934.
Huang, W., Lei, X., Ridder, G., Strauss, J., and Zhao, Y . (2013). Health, height, height
shrinkage, and ses at older ages: Evidence from China. American Economic Journal:
Applied Economics, 5(2):86–121.
Inkeles, A. (1969). Making men modern: On the causes and consequences of individual
change in six developing countries. American Journal of Sociology, pages 208–225.
Isa, Z., Alias, I. Z., Kadir, K. A., and Ali, O. (2000). Effect of iodized oil supplementation
on thyroid hormone levels and mental performance among orang asli schoolchildren and
pregnant mothers in an endemic goitre area in peninsular malaysia. Asia Pacific journal
of clinical nutrition, 9(4):274–281.
Jacoby, H. G. (1994). Borrowing constraints and progress through school: evidence from
peru. The Review of Economics and Statistics, pages 151–160.
Jaffrelot, C. (2003). India’s silent revolution: The rise of the lower castes in North India.
Orient Blackswan.
140
Jalal, F., Samani, M., Chang, M. C., Stevenson, R., Ragatz, A. B., and Negara, S. D. (2009).
Teacher certification in Indonesia: A strategy for teacher quality improvement. Jakarta:
Ministry of National Education Indonesia/ The World Bank.
Jong, H. N. (2015). New breed of teachers to improve education.
Jun, W. and Jianqun, L. (1985). Metabolism of iodized oil after oral administration in
guinea pigs. Nutrition Reports International, 31(5):1085–1092.
Key, V . (1949). Southern politics in state and nation. University of Tennessee Press.
Kishore, K., Subbiah, A., Sribimawati, T., Diharto, I. S., Alimoeso, S., Rogers, P., and
Setiana, A. (2000). Indonesia country study. Asian Disaster Preparedness Center.
Knight, J. B. and Sabot, R. H. (1990). Education, productivity, and inequality: The East
African natural experiment. Oxford University Press.
Kondylis, F. and Manacorda, M. (2012). School proximity and child labor evidence from
rural tanzania. Journal of Human Resources, 47(1):32–63.
Kremer, M., Brannen, C., and Glennerster, R. (2013). The challenge of education and
learning in the developing world. Science, 340(6130):297–300.
Kudo, Y . (2015). Female migration for marriage: Implications from the land reform in
rural tanzania. World Development, 65:41–61.
Levine, D. I. and Yang, D. (2014). The impact of rainfall on rice output in Indonesia.
Technical report, National Bureau of Economic Research.
Li, H., Rosenzweig, M., and Zhang, J. (2010). Altruism, favoritism, and guilt in the alloca-
tion of family resources: Sophie’s choice in mao’s mass send-down movement. Journal
of Political Economy, 118(1):1–38.
141
Liu, J., Raine, A., Venables, P. H., Dalais, C., and Mednick, S. A. (2003). Malnutrition at
age 3 years and lower cognitive ability at age 11 years: Independence from psychosocial
adversity. Archives of Pediatrics & Adolescent Medicine, 157(6):593–600.
Løken, K. V ., Mogstad, M., and Wiswall, M. (2012). What linear estimators miss: The
effects of family income on child outcomes. American Economic Journal: Applied
Economics, 4(2):1–35.
Maccini, S. and Yang, D. (2009). Under the weather: Health, schooling, and economic
consequences of early-life rainfall. American Economic Review, 99(3):1006–26.
Makowiecka, K., Achadi, E., Izati, Y ., and Ronsmans, C. (2008). Midwifery provision in
two districts in Indonesia: How well are rural areas served? Health Policy and Planning,
23(1):67–75.
Malamud, O., Pop-Eleches, C., and Urquiola, M. (2016). Interactions between family
and school environments: Evidence on dynamic complementarities? Technical report,
National Bureau of Economic Research.
Maluccio, J. A., Hoddinott, J., Behrman, J. R., Martorell, R., Quisumbing, A. R., and Stein,
A. D. (2009). The impact of improving nutrition during early childhood on education
among Guatemalan adults. Economic Journal, 119(537):734–763.
Mani, S. (2012). Is there complete, partial, or no recovery from childhood malnutri-
tion? Empirical evidence from Indonesia. Oxford Bulletin of Economics and Statistics,
74(5):691–715.
Manning, R. T. (1982). The serial sevens test. Archives of Internal Medicine, 142(6):1192–
1192.
142
Mason, A. and Khandker, S. (1997). Household schooling decisions in tanzania. World
Bank (mimeo).
Mauro, P. (1995). Corruption and growth. The quarterly journal of economics, 110(3):681–
712.
Miguel, E. and Kremer, M. (2004). Worms: Identifying impacts on education and health
in the presence of treatment externalities. Econometrica, 72(1):159–217.
North, D. C. (1973). The rise of the western world: A new economic history. Cambridge
University Press.
Parinduri, R. A. (2014). Do children spend too much time in schools? Evidence from a
longer school year in Indonesia. Economics of Education Review, 41:89–104.
Peterson, S. (2000). Controlling iodine deficiency disorders: Studies for program manage-
ment in sub-saharan africa.
Peterson, S., Assey, V ., Forsberg, B. C., Greiner, T., Kavishe, F. P., Mduma, B., Rosling, H.,
Sanga, A. B., and Gebre-Medhin, M. (1999). Coverage and cost of iodized oil capsule
distribution in tanzania. Health Policy and Planning, 14(4):390–399.
Pigou, A. C. (1952). Essays in economics. London: Macmillan.
Pitt, M. M., Rosenzweig, M. R., and Hassan, M. N. (2012). Human capital investment and
the gender division of labor in a brawn-based economy. American Economic Review,
102(7):3531–3560.
Pop, V . J., Kuijpens, J. L., van Baar, A. L., Verkerk, G., van Son, M. M., de Vijlder, J. J.,
Vulsma, T., Wiersinga, W. M., Drexhage, H. A., and Vader, H. L. (1999). Low mater-
nal free thyroxine concentrations during early pregnancy are associated with impaired
psychomotor development in infancy. Clinical endocrinology, 50(2):149–155.
143
Prakash, N., Rockmore, M., Uppal, Y ., et al. (2015). Do criminally accused politicians af-
fect economic outcomes? evidence from india. Households in Conflict Network (HiCN),
The Institute of Development Studies, University of Sussex.
Rantanen, T., Guralnik, J. M., Foley, D., Masaki, K., Leveille, S., Curb, J. D., and White, L.
(1999). Midlife hand grip strength as a predictor of old age disability. Jama, 281(6):558–
560.
Ravallion, M. (1988). Inpres and inequality: A distributional perspective on the centre’s
regional disbursements 1. Bulletin of Indonesian economic studies, 24(3):53–71.
Rice, J. K. (2003). Teacher quality: Understanding the effectiveness of teacher attributes.
ERIC.
Rocha, R. and Soares, R. R. (2015). Water scarcity and birth outcomes in the Brazilian
semiarid. Journal of Development Economics, 112:72–91.
Ronsmans, C., Scott, S., Qomariyah, S., Achadi, E., Braunholtz, D., Marshall, T., Pambudi,
E., Witten, K., and Graham, W. (2009). Professional assistance during birth and mater-
nal mortality in two Indonesian districts. Bulletin of the World Health Organization,
87(6):416–423.
Rosenberg, N., Birdzell, L. E., and Mitchell, G. W. (1986). How the West grew rich.
Popular Prakashan.
Rosenzweig, M. R. and Schultz, T. P. (1982). Market opportunities, genetic endowments,
and intrafamily resource distribution: Child survival in rural india. American Economic
Review, pages 803–815.
Rosenzweig, M. R. and Udry, C. (2014). Rainfall forecasts, weather, and wages over the
agricultural production cycle. American Economic Review, 104(5):278–283.
144
Rosenzweig, M. R. and Wolpin, K. I. (1980). Testing the quantity-quality fertility model:
The use of twins as a natural experiment. Econometrica: journal of the Econometric
Society, pages 227–240.
Rosenzweig, M. R. and Zhang, J. (2013). Economic growth, comparative advantage, and
gender differences in schooling outcomes: Evidence from the birthweight differences of
Chinese twins. Journal of Development Economics, 104:245–260.
Rossin-Slater, M. and W¨ ust, M. (2015). Are different early investments complements or
substitutes? Long-run and intergenerational evidence from denmark.
Schaner, S. and Das, S. (2016). Female labor force participation in Asia: Indonesia country
study. Technical report, Asian Development Bank.
Schultz, T. W. (1961). Investment in human capital. The American economic review,
51(1):1–17.
Seim, K. (2006). Stability in competition. RAND Journal of Economics, 37(3):619–640.
Sen, A. (1984). The living standard. Oxford Economic Papers, pages 74–90.
Shah, M. and Steinberg, B. M. (2017). Drought of opportunities: Contemporaneous and
long-term impacts of rainfall shocks on human capital. Journal of Political Economy,
125(2):527–561.
Shleifer, A. and Vishny, R. W. (1993). Corruption. The quarterly journal of economics,
108(3):599–617.
Sivaramakrishnan, K. (1997). Under-franchise in urban areas: Freeze on delimitation of
constituencies and resultant disparities. Economic and Political Weekly, pages 3275–
3281.
145
Sivaramakrishnan, K. (2000). North-south divide and delimitation blues. Economic and
Political Weekly, pages 3093–3097.
Sivaramakrishnan, K. (2001). Constituencies delimitation: Deep freeze again? Economic
and Political Weekly, pages 4694–4696.
Stokes, D. E. (1963). Spatial models of party competition. American Political Science
Review, pages 368–377.
Strauss, J. (1990). Households, communities, and preschool children’s nutrition out-
comes: Evidence from rural cˆ ote d’ivoire. Economic Development and Cultural Change,
38(2):231–261.
Strauss, J. and Thomas, D. (1998). Health, nutrition, and economic development. Journal
of Economic Literature, 36(2):766–817.
Strauss, J., Witoelar, F., and Sikoki, B. (2016). The fifth wave of the Indonesia family life
survey: Overview and field report.
Streeten, P., Burki, S. J., Haq, U., Hicks, N., and Stewart, F. (1981). First things first:
Meeting basic human needs in the developing countries.
Summers, A. A. and Wolfe, B. L. (1977). Do schools make a difference? American
Economic Review, 67(4):639–652.
Supriadi, D. and Hoogenboom, I. (2004). Introduction: Teachers in Indonesia through
times. Teachers in Indonesia: Their education, training, and struggle since colonial era
until reformation era, pages 1–20.
Thomas, D. (1994). Like father, like son; like mother, like daughter: Parental resources
and child height. Journal of Human Resources, pages 950–988.
146
Thomas, D., Strauss, J., and Henriques, M.-H. (1990). Child survival, height for age and
household characteristics in Brazil. Journal of Development Economics, 33(2):197–234.
Thomas, D., Strauss, J., and Henriques, M.-H. (1991). How does mother’s education affect
child height? Journal of Human Resources, pages 183–211.
Untoro, J., Schultink, W., Gross, R., West, C. E., and Hautvast, J. G. (1998). Efficacy of
different types of iodised oil. The Lancet, 351(9104):752–753.
Vaishnav, M. (2010). The market for criminality: Money, muscle and elections in India.
Unpublished paper.
Vicente, P. C. and Wantchekon, L. (2009). Clientelism and vote buying: lessons from field
experiments in african elections. Oxford Review of Economic Policy, 25(2):292–305.
Witsoe, J. (2005). Democracy against the state: The politics of caste empowerment in
Bihar. PhD thesis, University of Cambridge.
Wolff, J. (2001). Physiology and pharmacology of iodized oil in goiter prophylaxis.
Medicine, 80(1):20–36.
Wolfinger, R. E. (1965). The development and persistence of ethnic voting. American
Political Science Review, 59(4):896–908.
Yi, J., Heckman, J. J., Zhang, J., and Conti, G. (2015). Early health shocks, intra-household
resource allocation and child outcomes. Economic Journal, 125(588).
Young, C. (1979). The politics of cultural pluralism. University of Wisconsin Press.
Zimmermann, M. B., Aeberli, I., Torresani, T., and B¨ urgi, H. (2005). Increasing the io-
dine concentration in the swiss iodized salt program markedly improved iodine status in
147
pregnant women and children: a 5-y prospective national study. American Journal of
Clinical Nutrition, 82(2):388–392.
148
Appendix A
Appendix to Chapter 1
A.1 Proof of Proposition 1
The cohorts that went to school before the INPRES program was conceived and were out
of school by 1974 could not have anticipated the large scale school construction program.
Therefore, the older cohorts must have expected the number and quality of schools to be
the same for themselves and and younger cohorts. This implies
E
1
(Z
j1
) =E
1
(Z
j2
)
Combining with the equation 1.5, we obtain that for cohort 1 (older cohort)
S
j1
=E
1
S
j1
=E
1
S
j2
=E
2
S
j1
That is, individuals from the older cohort are observe their own schooling attainments
and, when making their schooling decisions, expect the regional average years of schooling
of a particular ability type to be identical for the older and newer cohorts. The newer cohorts
can also observe the educational attainment of the older cohorts. Using a similar logic,
E
1
q
j1
=E
1
q
j2
=E
2
q
j1
=q
j1
149
,
S
j2
=E
2
S
j2D
and
E
2
q
j2
=q
j2
However, for the younger cohort, the cost of schooling parameter(Z
j2
), and the quality
of education,q
j2
, changed with the increase in number of schools,Z
j2
, due to the INPRES
primary school construction program. Combining these assumptions about the differences
between expected and actual number and quality of schools with equations (1.5) and (1.6),
we get
((Z
j2
)
1
)S
j2
((Z
j1
)
1
)S
j1
=
2
(q
j2
q
j1
) (A.1)

dS
j2
dZ
j2
=

2
dq
j2
dZ
j2

1
(Z
j2
)S
j2
(Z
j2
)
1
(A.2)
and

=

2

dq
j2
dZ
j2

1
(Z
j2
)
S
j2

((Z
j1
)
1
)
(A.3)
If we assume that the average quality of education improved or remained unaffected
with the increase in number of primary schools under INPRES primary school construction
program, i.e.
dq
j2
dZ
j2
0, it generates the following comparative statistics.
dS
j2
dZ
j2
> 0;

> 0 (A.4)
Under the assumption of quality deterioration,
dq
j2
dZ
j2
< 0, the model predictions change
to
150
dS
j2
dZ
j2
Q 0;

Q 0 (A.5)
A.2 Teacher Quality
There is no clear consensus on what makes a good teacher (Hanushek and Rivkin (2006)).
Therefore, we consider a few different characteristics that could, arguably, be correlated
with the quality of a teacher. We use two broad measures of teacher quality: educational
qualifications of the teachers and residence of teachers. In their review of the role of teach-
ers in developing countries, Bold et al. (2017a) state that while it is now widely acknowl-
edged that teacher quality is a key determinant of student learning, little is known about
what specific dimensions of teacher quality matter. They provide evidence from a service
delivery indicator program that collected information through direct observations, unan-
nounced visits, and tests from primary schools in seven countries in Sub-Saharan Africa
suggesting that teacher absenteeism and their lack of pedagogical and subject specific
knowledge are surprisingly high. They conclude that pedagogical interventions and im-
proved accountability and incentives for teachers are the most urgent policy requirements
to ensure improved learning outcomes (See also Kremer et al. (2013) and Glewwe et al.
(2016)). In related work drawing information form the same survey, Bold et al. (2017b)
find a teacher’s knowledge of the subject and pedagogy and her pedagogical skills are most
strongly associated with her educational qualifications. In his review of the evidence on
the effectiveness of education policies in the United States and outside, Hanushek (2003)
reports that in contrast to the US experience, educational qualifications of teachers have
been found to have a much stronger effect on student performance in developing countries.
1
1
See also Summers and Wolfe (1977), Behrman and Birdsall (1983), Birdsall (1985), and Rice (2003).
151
Teacher quality, especially the qualifications of the primary school teachers, has been a
concern in Indonesia for quite sometime. According to a 2009 report released by Ministry
of National Education, Government of Indonesia, in collaboration with World Bank, more
than 60 percent of the total 2.78 million teachers in 2006 did not have the academic quali-
fication to be a school teacher - a four-year bachelor’s degree (Jalal et al. (2009)). Majority
of them had either a two-year diploma or a senior secondary certificate qualification. Most
teachers from this group (about 70 percent) teach in the primary schools. In 1970s, it is
conceivable that the teacher qualifications, in terms of their education attainements, was
worse. According to the report:
”.. [T]he quality of the teachers began to decline with the expansion of the primary
school (SD INPRES) program. In order to meet the surge in demand for teachers created
by the rapid increase in the number of primary schools, quality was sacrificed for quantity.
In general, recruitment into these programs became less selective and the average ability
of teachers fell. Consequently, the prestige of teachers also fell. Teacher salaries have
declined in real terms when compared to national average salaries in inverse proportion
to the number of teachers inducted into the profession and there has been less incentive for
the brighter students to enter the teaching service. ”
In an interview with the Jakarta Post in June 2015, the evaluation directorate program
planning head of the Culture and Elementary and Secondary Education Ministry of Indone-
sia, Tagor Alamsyah Harahap, acknowledged that the quality of the teachers recruited dur-
ing the SD INPRES policy was not up to par as they were hastily recruited (Jong (2015)).
Supriadi and Hoogenboom (2004) and Hapsariputri (2010) also find this to be true. Figure
A6 depicts the massive surge in the number new primary teacher appointment around this
time. It is, therefore, highly likely that apart from improving access to primary schools, the
152
INPRES program brought about some other significant changes. An important one among
them was the decrease in the quality of the newly recruited teacher.
However, the metric of teacher quality used in these studies is not clear. We, there-
fore, test the decline in teacher quality explicitly. Obtaining information on the quality of
teachers in Indonesia from the 1970s is not easy. To get around this problem, we employ a
methodology similar to that suggested in Behrman and Birdsall (1983). We pool together
the 1971, 1980, and 1990 waves of the IPC and the 1976 and 1985 waves of the SUPAS
and, from the pooled data, separate out individuals who report their main occupation as
primary school teachers. Out of these, the 1971 wave of IPC was conducted before the
INPRES school construction program started, the 1976 wave of SUPAS is from when the
school construction was underway and the 1980 and 1990 waves of IPC and the 1985 wave
of SUPAS are from years after the school construction program. We employ a difference-
in-difference strategy to check how the composition of teachers have been changed over
the years in regions with high and low intensity of INPRES school construction.
In table A3, we examine the impact of school construction intensity on the qualifica-
tions of the primary school teachers. The recommended qualification to be a primary school
teacher in Indonesia was a four-year bachelor’s degree (DIPLOMA IV or SARJANA 1),
three year college (DIPLOMA III) or a two-year diploma (DIPLOMA II) degree after com-
pleting high school (senior secondary school). We regress a dummy variable indicating if a
teacher has a high school degree, a DIPLOMA II, a DIPLOMA III or a DIPLOMA IV on a
dummy variable indicating if the district witnessed high or low intensity of INPRES school
construction interacted with the survey year. We include, in this analysis, survey year fixed
effects and intensity-region fixed effect and controls for the gender of the teacher and the
urban or rural status of the community. Comparing to the 1971 survey year benchmarks, the
average education of the primary school teachers improved considerably over the years -
the survey year dummies identify an increasingly improving education levels of the teach-
153
ers over the periods (not presented in the table). However, most of them still lacked the
recommended qualification levels for becoming a primary school teacher. High INPRES
intensity regions, regions where school construction was more intense, had teachers who
were less qualified to begin with. This is consistent with the observation that INPRES
targeted regions that were lagging behind much more intensively.
Our main interest is to examine if the INPRES school construction affected teacher
quality differently across high intensity and low intensity regions. We find a negative and
significant coefficient on the interaction of the survey year and INPRES intensity region in
1976 - teachers in regions that witnessed high intensity of school construction post INPRES
program were much less likely to have completed high school or the teaching diploma
degrees in 1976. In high intensity INPRES regions, the teachers were 6.8, 6.3, 6.3, and 5.2
percentage points less likely to have completed high school, DIPLOMA II, DIPLOMA III
and DIPLOMA IV , respectively, than in low intensity regions even after controlling for the
initial differences between regions. Moreover, if we compare the differences across regions
in 1971 and 1976, the differences seem to have increased even with the greater focus of the
policy on high intensity regions. However, the negative interaction decreases in magnitude
and in significance as time goes by. This may be due to the fact that with more time to hire
and fire teachers, schools were able to improve the quality of teachers over time. However,
most of the coefficients seem to be negative, even if not always significant, suggesting the
deterioration of quality persisted for some time.
Further, as suggested in Boyd et al. (2005), teachers prefer to work in regions that are, in
terms of characteristics and environments, close to where they grew up. Bold et al. (2017b)
also hypothesize that whether or not the teacher teaches in regions close to his home has
implications for teacher effort. There might also be higher levels of local accountability
that teachers from the region are subjected to.
2
Also, with over 17,000 islands, over thirty
2
Bold et al. (2017b) argue that the effect on student performance mediated through teacher effort could
154
provinces, multitude of faiths and ethnicity, and around 700 hundred living local languages,
Indonesia is very diverse it its culture. It is plausible that if a teacher understands the culture
of the region in which she teaches, she is better able to communicate with and understand
the students. If geographical distance is any proxy of cultural distance in Indonesia, teach-
ers who grew up in regions close to the one they teach in might be better in this regard. In
table A4, we look at teacher characteristics in terms of their residence. In column (1), we
check if a primary school teacher was born in the same district that she teaches in. Teachers
in regions with high intensity of school construction post the INPRES program were much
less likely to be born in the same district (18 percentage points less). In column (2), the
dependent variable takes a value ‘0’ if the teacher’s previous district of residence was the
same district as the one she is teaching in, ‘1’ if her previous district of residence was not
the one she is teaching in but was in the same province, ‘2’ her residence was in a different
province in the country and ‘3’ if the teacher is from a different country. As the coefficient
suggests, teachers after the INPRES program in high intensity school construction regions
were much more likely to be from outside the district. In column (3), the dependent vari-
able takes value ‘0’ if the teacher is teaching in the same district she was in five years ago,
‘1’ if she migrated from a different district in the same province, ‘2’ if she migrated from
a different province in same country and ‘3’ if the teacher came from a different country.
Again, teacher in high intensity regions post program were much more likely to be living
in a far off place five years ago. Column (4) uses the duration of stay in the present district
as the dependent variable. Teachers in high intensity regions post program had, on average,
spent around 3.7 years less in the district they are teaching in than teachers in low inten-
sity regions even after controlling for any initial differences. In sum, higher intensity of
INPRES regions had more teachers from outside the regions.
go in either direction. Teachers from the local region might be absent less often because they care more about
the students from the region or are more accountable to the community. Alternatively, they might be absent
more often if they have access more outside opportunities in the local economy.
155
In appendix table A5, we look at some other very crude indicators that might capture
a teacher’s ability to relate with the students and local accountability. We find that that the
teachers in high intensity region after the school construction program were less likely to be
from a religion followed by the majority of the students, less likely to speak the language
spoken by the majority in the region, more likely to be living in rented accommodation
without their families.
156
TIMELINE
YEARS OF
EDUCATION
IN 2014
INPRES ENDLINE
HIGH RAIN
EXPSD CHRTS IN
SCHOOL
16
HIGH RAIN
CNTRL CHRTS IN
SCHOOL HIGH RAIN EXPOSED COHORTS
14
CATCH-UP
HIGH RAIN EXPOSED COHORTS
12
HIGH RAIN
CNTRL CHRTS IN
YEAR 1
HIGH RAIN
EXPSD CHRTS IN
YEAR 1 HIGH RAIN EXPOSED COHORTS
10
CONTROL
COHORTS
BORN
EXPOSED
COHORTS
BORN
LOW RAIN
EXPSD CHRTS IN
SCHOOL
8
ACTUAL
RECOVERY
6
LOW RAIN
CNTRL CHRTS IN
YEAR 1
LOW RAIN
EXPSD CHRTS IN
YEAR 1
4
HIGH RAIN EXPOSED COHORTS
2
LOW RAIN
CNTRL CHRTS IN
SCHOOL
RECOVERY SPECTRUM
1950 1960 1970 1980 1990 2014
YEAR
FIGURE A1: TIME LINE - RECOVERY FROM EARLY LIFE SHOCKS
157
-2000 -1000 0 1000 2000
Rainfall a month - Historical average rainfall in the month
650 700 750 800 850 900
Month and year(century month code): January 1955 - December 1975
FIGURE A2: RAINFALL SHOCK VARIATION
158
-2 -1 0 1 2
Agricultural indicators (standardized)
1500 2000 2500 3000
Annual rainfall in millimeters
Cereal yield (corr: 0.390)
Land under cereal production (corr: 0.541)
Notes: Data from Food and Agricultural Organization of the United Nations. Years included: 1960-1990.
FIGURE A3: RAINFALL AND AGRICULTURAL IN INDONESIA
159
7.5 8 8.5 9
Years of education
-500 0 500
Rainfall in the year of birth - Prior 50 year average rainfall
FIGURE A4: RAINFALL AND EDUCATION IN INDONESIA
160
5.5 6 6.5 7 7.5 8
Lowess - Years of schooling (Control cohorts)
-500 0 500
Rainfall
All regions
Low INPRES intensity regions
High INPRES intensity regions
FIGURE A5: RAINFALL AND EDUCATION BY INPRES INTENSITY REGIONS
0 50,000 100000 150000
New primary school teacher appointments 1974 - 1975
1975 - 1976
1976 - 1977
1977 - 1978
1978 - 1979
1979 - 1980
1980 - 1981
1981 - 1982
1982 - 1983
1983 - 1984
1984 - 1985
1985 - 1986
1986 - 1987
1987 - 1988
1988 - 1989
1989 - 1990
1990 - 1991
1991 - 1992
1992 - 1993
1993 - 1994
1994 - 1995
1995 - 1996
1996 - 1997
1997 - 1998
1998 - 1999
FIGURE A6: NEW PRIMARY SCHOOL TEACHERS APPOINTED (JALAL ET AL. (2009))
161
TABLE A1: GRADE CONVERSION RATE
(1) (2) (3)
Start year of End year of Conversion
V ARIABLES Primary school rate
Above median rainfall in year 1 -0.1406 0.0535 0.0403**
(0.1384) (0.1925) (0.0181)
INPRES schools per 1000 children 0.1045 0.0803 0.0218
(0.1315) (0.1404) (0.0169)
Above median rainfall in year 1 * INPRES schools/1000 0.1068 -0.0277 -0.0030
(0.0960) (0.0851) (0.0051)
Mean of DV 1975 1981 0.92
Observations 2,933 3,004 2,880
R-squared 0.8587 0.8741 0.1143
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
TABLE A2: TYPE OF SCHOOL ADMINSTRATION
(1) (2) (3) (4)
Went to [...] school
V ARIABLES public primary private primary
Above median rainfall in year 1 -0.0097 -0.0023
(0.0189) (0.0094)
INPRES schools per 1000 children 0.0186** 0.0173 -0.0074 -0.0079
(0.0074) (0.0110) (0.0079) (0.0086)
Above median rainfall in year 1 * INPRES schools/1000 0.0018 0.0016
(0.0109) (0.0051)
Mean of DV 0.88 0.88 0.02 0.02
Observations 3,059 3,059 3,059 3,059
R-squared 0.1731 0.1746 0.1299 0.1306
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
162
TABLE A3: PRIMARY SCHOOL CONSTRUCTION AND TEACHER QUALIFICATIONS
(1) (2) (3) (4)
Graduated
V ARIABLES High School Diploma II Diploma III Diploma IV
High INPRES intensity * Survey year 1976 -0.0677* -0.0633** -0.0632** -0.0523***
(0.0402) (0.0300) (0.0300) (0.0179)
High INPRES intensity * Survey year 1980 0.0023 -0.0150 -0.0148 -0.0248***
(0.0133) (0.0130) (0.0130) (0.0083)
High INPRES intensity * Survey year 1985 0.0100 -0.0042 -0.0006 -0.0118
(0.0159) (0.0153) (0.0144) (0.0102)
High INPRES intensity * Survey year 1990 0.0133 0.0006 0.0082 -0.0067
(0.0129) (0.0155) (0.0158) (0.0115)
Mean of DV 0.85 0.07 0.07 0.05
Observations 41,243 41,243 41,243 41,243
R-squared 0.2818 0.0286 0.0289 0.0205
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses are clustered at the district level. Other controls
include survey year fixed effects, a dummy for gender of the teacher and a dummy for the type of residence, rural or urban.
TABLE A4: PRIMARY SCHOOL CONSTRUCTION AND TEACHER CHARACTERISTICS
(1) (2) (3) (4) (5)
Born in Distance to Residence Years at
V ARIABLES province previous residence? five years ago? current residence Age
High INPRES intensity * Survey year 1976 -0.1826*** -0.8402
(0.0647) (0.7004)
High INPRES intensity * Survey year 1980 -0.1761*** 0.1554*** 0.0323 -3.6800** -1.3202***
(0.0520) (0.0508) (0.0239) (1.5313) (0.4828)
High INPRES intensity * Survey year 1985 -0.2046*** 0.1717*** 0.0561* -4.1066** -1.6303**
(0.0586) (0.0654) (0.0300) (1.7495) (0.6358)
High INPRES intensity * Survey year 1990 -0.2001*** 0.1807*** 0.0211 -4.1542** -0.7103
(0.0598) (0.0584) (0.0231) (1.8549) (0.5907)
Mean of DV 0.90 1.06 0.93 29.40 32.58
Observations 41,255 40,102 39,180 33,782 41,241
R-squared 0.0418 0.2372 0.6598 0.1486 0.0166
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses are clustered at the district level. Other controls
include survey year fixed effects, a dummy for gender of the teacher and a dummy for the type of residence, rural or urban.
163
TABLE A5: PRIMARY SCHOOL CONSTRUCTION AND OTHER CRUDE TEACHER
CHARACTERISTICS
(1) (2) (3) (4)
Religion same as Language same as Lives in a rented Does not live
V ARIABLES that of majority students that of majority students accommodation with family
High INPRES * Survey year 1976 -0.0360 -0.0007 0.0171
(0.0441) (0.0024) (0.0134)
High INPRES * Survey year 1980 -0.0301 -0.0008 0.0870** 0.0333***
(0.0257) (0.0024) (0.0340) (0.0117)
High INPRES * Survey year 1985 -0.0397 -0.0007 0.1236*** 0.0458***
(0.0295) (0.0024) (0.0411) (0.0151)
High INPRES * Survey year 1990 -0.0362 -0.0007 0.1193*** 0.0146
(0.0292) (0.0024) (0.0399) (0.0124)
Observations 41,255 41,255 40,291 40,789
R-squared 0.0091 0.0004 0.0201 0.0049
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses are clustered at the district level. Other controls
include survey year fixed effects, a dummy for gender of the teacher and a dummy for the type of residence, rural or urban.
TABLE A6: INTERACTION IMPACT ON YEARS OF SCHOOLING BY YEAR OF BIRTH
(1) (2) (3) (4)
Highest school Completed
V ARIABLES grade completed Primary School Middle School High School
Above median rain * INPRES * Survey year = 1969 -0.8529*** -0.0738*** -0.1184** -0.0947**
(0.2751) (0.0212) (0.0476) (0.0402)
Above median rain * INPRES * Survey year = 1970 -0.4648 0.0130 -0.0234 -0.0182
(0.2731) (0.0318) (0.0665) (0.0635)
Above median rain * INPRES * Survey year = 1971 -0.3878 -0.0430 0.0433 -0.0673
(0.3931) (0.0474) (0.0424) (0.0545)
Above median rain * INPRES * Survey year = 1972 -0.1838 0.0017 -0.0499 -0.0321
(0.2704) (0.0342) (0.0637) (0.0488)
Above median rain * INPRES * Survey year = 1973 0.1331 0.0061 -0.0648** 0.0216
(0.1801) (0.0294) (0.0255) (0.0376)
Mean of DV 8.48 0.79 0.53 0.39
Observations 3,087 3,281 3,281 3,281
R-squared 0.4478 0.2539 0.2822 0.2668
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother and the father of the
individuals had some schooling, completed primary school, completed middle school, completed high school, or had some tertiary
level education. We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate
in 1971 and the water and sanitation program in the district during the time of school construction program.
164
TABLE A7: ALTERNATIVE DEFINITIONS OF RAINFALL SHOCK
Years of Schooling
V ARIABLES (1) (2) (3) (4)
Rain fall shock in year 1 0.7308*** 0.98*** 0.76*** 0.31
(0.2243) (0.25) (0.22) (0.28)
INPRES schools per 1000 children 0.3653* 0.2724 0.3896* 0.1987
(0.2136) (0.2059) (0.2164) (0.2053)
Rainfall shock in year 1 * INPRES schools/1000 -0.3539*** -0.32** -0.38*** -0.13
(0.1125) (0.13) (0.11) (0.12)
Mean of DV 8.29 8.29 8.29 8.29
Rainfall dummy = 1 Above median Above 75 %ile Above mean Above mean+SD
Observations 3,399 3,399 3,399 3,399
R-squared 0.4055 0.4068 0.4057 0.4035
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
165
TABLE A8: ROBUSTNESS WITH TRIMMED EXTREME RAINFALL OBSERVATIONS
(1) (2) (3)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.47 1.00*** 0.73***
(0.85) (0.37) (0.22)
INPRES schools per 1000 children 0.39 0.79*** 0.37*
(0.53) (0.29) (0.21)
Above median rain * INPRES schools/1000 -0.59 -0.45** -0.35***
(0.40) (0.17) (0.11)
Mean of DV 7.55 8.16 8.29
Trimmed sample 10%ile - 90%tile 5%ile - 95%tile Full
Observations 509 1,418 3,399
R-squared 0.55 0.45 0.41
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
TABLE A9: ALTERNATIVE RADIUS FOR WEIGHTED RAINFALL
(1) (2) (3) (4) (5) (6)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.16 0.36* 0.50** 0.73*** 0.65*** 0.49*
(0.21) (0.20) (0.23) (0.22) (0.24) (0.25)
INPRES schools per 1000 children 0.25 0.31 0.32 0.37* 0.35 0.34
(0.21) (0.21) (0.22) (0.21) (0.22) (0.22)
Above median rain * INPRES schools/1000 -0.12 -0.20** -0.21** -0.35*** -0.25** -0.25**
(0.10) (0.10) (0.11) (0.11) (0.10) (0.11)
Mean of DV 8.29 8.29 8.29 8.29 8.29 8.29
Radius (in kms) 25 50 100 200 300 500
Observations 3,399 3,399 3,399 3,399 3,399 3,399
R-squared 0.40 0.40 0.40 0.41 0.40 0.40
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
166
TABLE A10: INCLUSION OF RAINFALL SHOCKS IN LATER YEARS
(1) (2) (3)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.73*** 0.69*** 0.70***
(0.22) (0.22) (0.23)
INPRES schools per 1000 children 0.37* 0.38* 0.43
(0.21) (0.21) (0.27)
Above median rain * INPRES schools/1000 -0.35*** -0.36*** -0.37***
(0.11) (0.11) (0.12)
Mean of DV 8.29 8.29 8.29
Years of life included -2 to 1 Years -2 to 5 Years -2 to 5
Interactions included -2 to 1 Years -2 to 1 Years -2 to 5
Observations 3,399 3,399 3,399
R-squared 0.41 0.41 0.41
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
167
TABLE A11: ROBUSTNESS TO DIFFERENT DEFINITIONS OF PARENTAL EDUCATION
(1) (2) (3)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.7210*** 0.7308*** 0.7560**
(0.2315) (0.2243) (0.3644)
INPRES schools per 1000 children 0.3198 0.3653* 0.6041
(0.1986) (0.2136) (0.3792)
Above median rainfall in year 1 * INPRES schools/1000 -0.3160*** -0.3539*** -0.3565**
(0.1129) (0.1125) (0.1622)
Mother enrolled in school 0.7500**
(0.3203)
Mother completed primary school 1.6628***
(0.1487)
Mother completed middle school 3.3535***
(0.3225)
Mother completed high school 4.6145***
(0.3297)
Mother enrolled at tertiary level 6.5745***
(0.5869)
Mother’s education dummy 1.9739***
(0.1458)
Mother’s years of education 0.4140***
(0.0371)
Mean of DV 8.28 8.28 9.66
Parental education variable Dummy Categorical Continuous
Observations 3,399 3,399 1,661
R-squared 0.3759 0.4055 0.4085
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
168
TABLE A12: ROBUSTNESS TO SPECIFICATION CHANGES
(1) (2) (3) (4) (5)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.73*** 0.73*** 0.59** 0.73*** 0.97***
(0.22) (0.24) (0.24) (0.18) (0.27)
INPRES schools per 1000 children 0.37* 0.34 0.32 0.37 0.45
(0.21) (0.20) (0.21) (0.23) (0.27)
Above median rain * INPRES schools/1000 -0.35*** -0.33*** -0.29*** -0.35*** -0.48***
(0.11) (0.11) (0.11) (0.11) (0.13)
Mean of DV 8.29 8.29 8.29 8.29 7.70
Specification change Main Non-movers Month X Year FE Province cluster Rural sample
Observations 3,399 3,146 3,399 3,399 2,475
R-squared 0.41 0.41 0.43 0.41 0.38
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
TABLE A13: WORKED WHILE IN SCHOOL
(1) (2) (3) (4) (5) (6)
V ARIABLES Did you work while you attended [....] school
Elementary Junior high Senior high Elementary Junior high Senior high
Above median rainfall in year 1 -0.0710 -0.2247 -0.3078 -0.0417 -0.0790 -0.0664
(0.0941) (0.1408) (0.3104) (0.0660) (0.0941) (0.1537)
INPRES schools per 1000 children -0.0527 -0.0290 -0.0179
(0.0466) (0.0673) (0.0914)
Above median rainfall * INPRES schools/1000 0.0014 0.0459 0.0320
(0.0386) (0.0531) (0.0787)
Observations 1,113 435 275 3,093 1,780 1,227
R-squared 0.1905 0.4291 0.5717 0.1257 0.1822 0.2511
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth , gender, ethnicity, religion, urban community, rain
during pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had
some schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education.
We also control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the
water and sanitation program in the district during the time of school construction program.
169
TABLE A14: IMPACT ON SCHOOLING BY GENDER
(1) (2) (3)
V ARIABLES Highest school grade completed
Above median rainfall in year 1 0.7308*** 0.5619 0.8321***
(0.2243) (0.3505) (0.3076)
INPRES schools per 1000 children 0.3653* 0.3052 0.5126*
(0.2136) (0.3474) (0.2815)
Above median rainfall * INPRES schools per children -0.3001** -0.3999**
(0.1476) (0.1594)
Gender Pooled Male Female
Mean of DV 8.29 8.76 7.83
Observations 3,399 1,668 1,731
R-squared 0.4055 0.4048 0.4971
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth, ethnicity, religion, urban community, rain during
pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had some
schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education. We also
control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the water and
sanitation program in the district during the time of school construction program.
TABLE A15: IMPACT OF RAINFALL AND PRIMARY SCHOOL CONSTRUCTION ON HEALTH
(1) (2) (3) (4)
Length of Average Dominant Investigator’s
V ARIABLES upper arm Lung capacity hand grip evaluation of health
Above median rainfall in year 1 0.21 9.54* 0.42 0.12
(0.16) (5.08) (0.38) (0.07)
INPRES schools per 1000 children 0.04 7.35** 0.24 0.16***
(0.16) (3.43) (0.32) (0.05)
Above median rainfall * INPRES schools per children -0.06 -1.62 -0.09 -0.04
(0.10) (2.61) (0.20) (0.03)
Mean of DV 33.64 353.37 27.52 6.98
Observations 3,808 3,796 3,756 3,814
R-squared 0.23 0.50 0.55 0.22
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth, ethnicity, religion, urban community, rain during
pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had some
schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education. We also
control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the water and
sanitation program in the district during the time of school construction program.
170
TABLE A16: IMPACT ON EMPLOYMENT
(1) (2) (3) (4) (5) (6) (7)
Work for pay White Weeks worked Pension Job satisfaction
V ARIABLES 2014 2008 2007 collar job per year plan scale 1 to 4
Above median rainfall -0.05** 0.04* 0.03* 0.001 -1.70* 0.04* 0.04
(0.02) (0.02) (0.02) (0.03) (0.95) (0.02) (0.04)
INPRES schools per 1000 children 0.002 0.02 0.01 0.03 -0.64 0.05** 0.01
(0.02) (0.02) (0.01) (0.02) (0.84) (0.02) (0.03)
Above median rainfall * INPRES schools per children 0.01 -0.002 -0.003 -0.03** 0.72 -0.01 -0.06***
(0.01) (0.01) (0.01) (0.01) (0.51) (0.01) (0.02)
Mean of DV 0.80 0.86 0.87 0.26 42.72 0.12 2.96
Observations 3,632 3,295 3,295 3,009 3,076 2,468 3,076
R-squared 0.21 0.17 0.17 0.13 0.14 0.19 0.10
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth, ethnicity, religion, urban community, rain during
pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had some
schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education. We also
control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the water and
sanitation program in the district during the time of school construction program.
TABLE A17: IMPACT ON CONSUMPTION AND WELL-BEING
(1) (2) (3) (4) (5) (6)
Log expenditure on
food non-food all Overall satifaction Wealth ladder Happiness
V ARIABLES (weekly) (monthly) (yearly) scale 1 to 5 scale 1 to 5 scale 1 to 4
Above median rainfall 0.06 0.03 0.02 0.07 0.05 0.02
(0.04) (0.07) (0.05) (0.04) (0.05) (0.03)
INPRES schools per 1000 children 0.02 0.06 0.03 0.03 0.06 0.01
(0.04) (0.06) (0.05) (0.04) (0.04) (0.02)
Above median rainfall * INPRES schools per children 1000 -0.04 -0.01 -0.01 -0.04 -0.02 -0.01
(0.03) (0.04) (0.03) (0.02) (0.03) (0.02)
Mean of DV 12.88 13.50 17.58 3.25 2.98 2.99
Observations 3,630 3,629 3,630 3,632 3,615 3,632
R-squared 0.18 0.23 0.21 0.11 0.12 0.10
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at the district level.
All columns include controls for month of birth, year of birth, kabupaten of birth, ethnicity, religion, urban community, rain during
pregnancy, rain in nine months prior to pregnancy, and dummy indicators each for whether the mother of the individuals had some
schooling, completed primary school, completed middle school, completed high school, or had some tertiary level education. We also
control for year of birth interacted with the number of children of school going age in 1971, enrollment rate in 1971 and the water and
sanitation program in the district during the time of school construction program.
171
Appendix B
Appendix to Chapter 2
B.1 ISP treatment definition
This section draws heavily from Field et al. (2009). Information has been reproduced for
clarity in understanding of how the iodine treatment variables were defined.
TABLE B1: ISP COVERAGE VARIATION (FROM FIELD ET AL. (2009))
Region District Year 1 Coverage 1 Year 2 Coverage 2 Year 3 Coverage 3 Year 4 Coverage 4 Year 5 Coverage 5
Dodoma Mpwapwa 1990 0.65 1992 0.58
Arusha Monduli 1992 0.71
Arusha Arumera 1991 0.89
Kilimanjaro Rombo 1990 0.68
Mororgoro Ulanga 1988 0.73 1991 0.61 1992 0.34
Ruvuma Songea Rural 1987 0.91 1991 0.74 1995 0.85
Ruvuma Mbinga 1995 0.92
Iringa Mufundi 1986 0.41 1991 0.63 1995 0.54
Iringa Makete 1986 0.2 1991 0.62 1993 0.62 1996 0.49
Iringa Njombe 1989 0.76 1992 0.68 1995 0.64
Iringa Ludewa 1989 0.59 1992 0.62 1995 0.47
Mbeya Chunya 1990 0.49
Mbeya Mebya Rural 1986 0.44 1989 0.84 1990 0.9 1993 0.53 1997 0.53
Mbeya Kyela 1989 0.91 1993 0.57
Mbeya Rungwe 1986 0.35 1990 0.73 1993 0.49
Mbeya Ileja 1989 0.94 1992 0.71
Mbeya Mbozi 1989 0.67 1991 0.63
Rukwa Mpanda 1987 0.79 1991 0.6 1993 0.72
Rukwa Sumbawanga 1987 0.76 1990 0.89 1993 0.72 1996 0.51
Rukwa Nkansi 1987 0.89 1991 0.49
Kigoma Kibondo 1989 0.73 1992 0.75 1996
Kigoma Kasulu 1987 0.5 1990 0.66 1996 0.49
Kigoma Kigoma Rural 1991 0.91
Kagera Karagwe 1990 0.96 1994 0.85
Kagera Bukoba Rural 1994 0.78
Kagera Biharamulo 1990 0.96 1994 0.38
Kagera Ngara 1989 0.29 1994 0.51
Calculation of probability of protection: The treated mothers received and iodine dosage
of 380 mg via the IOC [Peterson (2000); Peterson et al. (1999)]. However, as described
172
in Field et al. (2009), Wolff (2001), Jun and Jianqun (1985) and Untoro et al. (1998) pro-
vide a review of literature that finds that majority of iodine stored in the fatty tissue is
depleted rapidly within the first week and an hyperbolic rate thereafter. Following Field
et al. (2009), we assume that 85 percent (323) of the 380 mg dose was extracted away
immediately within the first month and the depletion followed the simple hyperbolic dis-
counting formulaV =A=(1 +kt) after that, wherek
1
is the half life of iodine in months.
Using the observation from Cao et al. (1994) and Eltom et al. (1985), which use similar
dosages of IOC provides full protection for 24 months and that 6.5 mg is the minimum io-
dine requirement for one full month of protection, Field et al. (2009) calculate the half life
to be 3 months. This implied half life is consistent with other studies of the approximate
half lives of urine iodine excretion after oral iodine administration to human populations
with iodine deficiency (See Wolff (2001)). The probability of protection in a month of
the first trimester, therefore, is the probability that the program had started and reached
the mother of the child by that month and the stocks of iodine had not depleted to levels
insufficient for protection (< 4:2 mg as per Field et al. (2009)) in that month.
1
A child is
protected in the first trimester if she is protected throughout weeks 1 to 12 (roughly three
months)
Based on the information and assumption above, probability of protection from in utero
IDD if the child district of birth received the ISP in year t (by month of birth):
TABLE B2: PROBABILITY OF PROTECTION
Year Jan Feb March April May June July Aug Sep Oct Nov Dec Birth year average
t 0 0 0 0 0 0 0 0.028 0.083 0.167 0.250 0.333 0.072
t+1 0.417 0.5 0.583 0.667 0.75 0.833 0.917 1 1 1 1 1 0.806
t+2 1 1 1 1 1 1 1 1 1 0.998 0.991 0.977 0.997
t+3 0.955 0.927 0.891 0.849 0.802 0.749 0.69 0.627 0.559 0.488 0.419 0.353 0.668
t+4 0.292 0.237 0.189 0.148 0.112 0.082 0.057 0.037 0.022 0.011 0.004 0.001 0.099
1
The 6.5 mg and 4.2 mg figures are calculated based in the recommended daily allowance (RDA) for
pregnant women
173
B.2 Alternative Definitions of ISP Exposure
Recall that our definition of ISP exposure probability used the variation in the coverage
rate across districts and years and the probability of protection based on the availability of
adequate amount of iodine in the mother’s body which depends on the relative timings of
supplementation and conception. Since we did not have the exact date of supplementation
and birth, we assumed that the probability of supplementation and birth are uniform across
the year. This assumption can, however, bias our results. We check the robustness of
our results to different definitions of ISP exposure in table B3. The results from the most
preferred specification (specification (8) in table 2.2) are reproduced in column (1) for
comparison. In column (2), we use the district-year specific coverage rate as the probability
of ISP exposure. This avoids biases due to differences in exact date of supplementation.
In column (3), we use only the depletion formula for the exposure probability calculation
assuming that if a district was treated in a particular year, all individuals from the district
received the supplementation. This avoids biases due to measurement and reporting errors
in coverage rate. This is the main specification used in Field et. al (2009). In column (4),
we use a dummy indicator of whether or not the probability of exposure of an individual
was non-zero. The results suggest our findings are robust to the alternative ways of defining
ISP treatment.
174
TABLE B3: ROBUSTNESS OF ISP EXPOSURE DEFINITION
(1) (2) (3) (4)
V ARIABLES Years of Schooling
ISP -0.71***
(0.15)
ISP2: coverage only (no depletion) -0.52***
(0.14)
ISP3: depletion only (no coverage) -0.55***
(0.13)
ISP4: depletion dummy (= 1, if exposed at all) -0.40***
(0.10)
Religion dummy YES YES YES YES
Tribe dummy YES YES YES YES
Livestock Value YES YES YES YES
Mean of dependent variable 2.20 2.20 2.20 2.20
Mean ISP treatment probability 0.31 0.80 0.39
Observations 518 518 518 518
R-squared 0.3641 0.3624 0.3633 0.3627
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at geoage level where geoage groups are district-
year of birth groups. Other controls include a dummy each indicating whether the mother and the father of the child had some education, age, gender and primary
enumeration area fixed effects.
TABLE B4: IMPACT OF ISP ON VACCINATIONS
(1) (2) (3) (4) (5) (6) (7)
V ARIABLES Has a vaccination card Tetanus vaccine Polio vaccine Measles vaccine TB vaccine Score 1 Score 2
ISP 0.0330 -0.0408 -0.1127 -0.0235 -0.0097 -0.1537 -0.1866
(0.0840) (0.0996) (0.0954) (0.0768) (0.0729) (0.2550) (0.2119)
Religion dummy YES YES YES YES YES YES YES
Tribe dummy YES YES YES YES YES YES YES
Livestock Value YES YES YES YES YES YES YES
Mean of dependent variable 0.5597 0.3180 0.3392 0.4416 0.5724 2.1557 1.5960
Mean ISP treatment probability 0.3220 0.3228 0.3228 0.3228 0.3228 0.3220 0.3220
Observations 552 552 552 552 552 552 552
R-squared 0.6637 0.3947 0.4161 0.5865 0.6954 0.7121 0.6869
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at district-year of birth groups. Controls include a
quadratic in age, a dummy each for whether the mother and father have some education or not, a dummy for gender of the individual, a dummy each for whether the
individual belongs to the majority tribe or religion, controls for total land holdings of the household, and district fixed effects.
175
TABLE B5: ACROSS HOUSEHOLD IMPACTS OF ISP
(1) (2) (3) (4)
V ARIABLES Years of education Height for age in 2004
ISP -0.95*** -0.69** -0.76 -0.06
(0.28) (0.34) (0.50) (0.44)
Observations 386 398 355 356
R-squared 0.40 0.49 0.08 0.08
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at
district-year of birth groups. Controls include a dummy each for whether the mother and father have some education
or not, a dummy for gender of the individual, a dummy each for whether the individual belongs to the majority tribe or
religion, controls for total land holdings of the household, age and district fixed effects.
TABLE B6: PRIMARY SCHOOL STARTING AGE AS A PLAUSIBLE MECHANISM
(1) (2) (3) (4) (5) (6)
V ARIABLES Years of Schooling
ISP -0.7330*** -0.3029*** -0.7267*** -0.2937*** -0.7663*** -0.3074**
(0.1171) (0.1022) (0.1138) (0.1000) (0.1341) (0.1243)
PEDP 0.1840*** -0.0931 0.1840*** -0.0941 0.1801*** -0.0951
(0.0499) (0.0891) (0.0509) (0.0916) (0.0481) (0.0923)
ISP * PEDP -0.5187** -0.3362 -0.5063* -0.3141 -0.4910* -0.3093
(0.2428) (0.2253) (0.2510) (0.2265) (0.2641) (0.2328)
Primary starting age -0.5941*** -0.5950*** -0.5944***
(0.0380) (0.0389) (0.0395)
Mean of dependent variable 2.2023 2.2023 2.2023 2.2023 2.2023 2.2023
Mean ISP treatment probability 0.3199 0.3199 0.3199 0.3199 0.3199 0.3199
Religion dummy NO NO YES YES YES YES
Tribe dummy NO NO YES YES YES YES
Livestock Value NO NO NO NO YES YES
Observations 519 519 519 519 519 519
R-squared 0.3635 0.6290 0.3637 0.6298 0.3651 0.6299
Notes: *** p<0.01, ** p<0.05, * p<0.1. Robust standard errors in parentheses. The standard errors are clustered at geoage level where geoage are district X year
of birth groups. Other controls include a quadratic in age, total land ownership of the household in which the child was born, a dummy each indicating whether the
mother and the father of the child had some education, gender and primary enumeration area fixed effects.
176

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on development and health economics

PDF

Essays on economics of education and private tutoring

PDF

Essays in development economics

PDF

Three essays in education and urban economics

PDF

Essays on health insurance programs and policies

PDF

Three essays on health & aging

PDF

Essays on human capital accumulation -- health and education

PDF

Essays on health economics

PDF

Essays in political economy and mechanism design

PDF

Essays on family planning policies

PDF

Essays on health and aging with focus on the spillover of human capital

PDF

Selection and impacts of early life events on later life outcomes

PDF

Three essays on cooperation, social interactions, and religion

PDF

Three essays on economics of early life health in developing countries

PDF

Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes

PDF

Politics is something we do together: identity and institutions in U.S. elections

PDF

Essays on family and labor economics

PDF

Essays on development economics

PDF

Essays on the empirics of risk and time preferences in Indonesia

PDF

Essays in development and experimental economics

Asset Metadata

Creator Bharati, Tushar (author)

Core Title Essays on education and institutions in developing countries

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 06/27/2018

Defense Date 04/10/2018

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Crime,Education,Elections,health,OAI-PMH Harvest,rain,School

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Strauss, John (committee chair), Bennett, Daniel (committee member), Meijer, Erik (committee member), Nugent, Jeffrey (committee member)

Creator Email bharatitushar24@gmail.com,tbharati@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-506771

Unique identifier UC11266323

Identifier etd-BharatiTus-6349.pdf (filename),usctheses-c40-506771 (legacy record id)

Legacy Identifier etd-BharatiTus-6349.pdf

Dmrecord 506771

Document Type Dissertation

Format application/pdf (imt)

Rights Bharati, Tushar

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA