Economic Growth, Rural Educational Investment and The Level and Distribution of Rural Incomes

Abstract
Economic growth, rural educational investment and the level and

distribution of rural incomes
Reena Chandu Badiani
2010
This dissertation examines two related questions. First, it estimates the effect of
growth in the demand for skilled and unskilled labor on rural household incomes
and the rural wage structure. Second, it examines the effect of growth in household
incomes and in labor market returns to education on household educational investment. The contribution of this dissertation is to consider these questions jointly: I
first examine how variation in the demand for skilled and unskilled labor alters incomes and returns to education in rural areas, and then examine how this variation
in incomes and returns feeds through to household education choices. To do this,
I first develop a model of household education choice that illustrates the effects of
policy induced expansions in skilled and unskilled manufacturing labor demand
and increases in agricultural productivity on returns to education and income for
skilled and unskilled workers. Second, I exploit plausibly exogenous variation in
skilled and unskilled labor demand caused by increases in regional agricultural
productivity and policy driven variation in manufacturing employment for different skill levels. The results show that an increase in unskilled labor demand
decreases returns to education for all workers, but has income effects that vary by
skill and land ownership. An increase in skilled labor demand increases returns
to education for all workers but only increases income for skilled households. The
results on educational investment show that investment for boys responds positively to both income and returns to education. For girls, investment in education
only responds to increases in income, but not to returns.
Economic growth, rural educational investment and the

level and distribution of rural incomes
A Dissertation
Presented to the Faculty of the Graduate School
of
Yale University
in Candidacy for the Degree of
Doctor of Philosophy
by
Reena Chandu Badiani
Dissertation Director: Professor Mark Rosenzweig
May 2010
UMI Number: 3414959
All rights reserved

INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
UMI
Dissertation Publishing
UMI 3414959
Copyright 2010 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.
uest
A
ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, Ml 48106-1346
2010 by Reena Chandu Badiani

All rights reserved.
Contents
List of Tables
iii
List of Figures
Acknowledgements
vi
Chapter 1.
Introduction to Dissertation
Chapter 2.
How does growth in the demand for skilled and unskilled labor
affect household incomes and the rural wage structure?
2.1.
Introduction
2.2.
Related Literature: Rural Labor Markets, Determinants of Wages and
7
7
Incomes
10
2.3.
Setting: The Rural Indian Economy
11
2.4.
Theoretical Model
15
2.5.
Empirical Strategy
27
2.6.
Conclusion
59
Chapter 3.
What is the effect of growth in household incomes and returns to

education on household educational investment?
78
3.1.
Introduction
78
3.2.
Related Literature
81
3.3.
Setting: Education in India
86
3.4.
Theoretical Model
90
3.5.
Empirical Specification
95
3.6.
Conclusion
104
i
CONTENTS
Bibliography
ii
111
Appendix A.
Data and Description of the Environment
118
Appendix B.
Derivations
126
B.l.
Obtaining the Wage Equations
126
B.2.
Obtaining the Income equation
128
B.3.
Alternative specification for the wage equation
129
Appendix C.
Figures
134
List of Tables
Table la.
Descriptive Statistics
61
Table lb.
Descriptive Statistics - Literacy by Task and Sector
62
Table lc.
Descriptive Statistics - Wages by Task and Sector
62
Table 2a.
Descriptive Statistics - Education by Skill Bracket in Manufacturing 63
Table 2b.
Descriptive Statistics - Tasks by Skill Bracket in Manufacturing . . .
Table 3.
Descriptive Statistics - Decomposition of Variance in the Proportion
of Educated Workers
63
63
Table 4.
First Stage Static Results
64
Table 5.
First Stage Dynamic Results
65
Table 6.
Unskilled Wages
66
Table 7.
Skilled Wages
67
Table 8.
Wage Robustness Checks, Groundwater and Agricultural Prices . . .
68
Table 9.
Wage Robustness Checks, Instruments and Initial Characteristics . . 69
Table 10.
Wage Robustness Checks, Price of Agricultural Products and Proportion
of Linked Industries
70
Table 11.
Income
71
Table 12.
Income, Splitting Households by Land and Skill
72
Table 13.
Income, Checking Robustness of Land Thresholds
73
Table 14.
Poverty
74
Table 15.
Inequality
75
Table 16.
Income Robustness Checks, Alternative Income Sources

iii
76
LIST OF TABLES
iv
Table 17.
Days devoted to Work in Agriculture
77
Table 18.
Descriptive Statistics - Education
106
Table 20.
Structural Education Parameters
108
Table 21.
Robustness Check, Education Specification
109
Table 22.
Robustness Check, Education Infrastructure
110
Table A.1.
Density and Specific Gravity of Raw Materials
135
Table A.2.
Industries by Factor Intensity
135
List of Figures
Table 1.
Map of Limestone Deposits and Employment in the Cement Industry 136
Table 2.
Map of Forest Density and Employment in the Plywood Industry . . 137
Table 3.
Exerpt from the National Mineral Atlas
138
Table 4.
Map of sorghum regions from the GAEZ (FAO, 2001)
139
Acknowledgements
I owe a depth of gratitude to my intellectual parents. Thanks to Mark Rosenzweig
for guiding my search for interesting questions, for stretching my capacity and for
encouraging me to strive for intellectual rigor. You have led by example and have
changed the way that I think. Joe Altonji's phenomenal empirical rigor and ability
to make data speak has been a great motivation to me. However, I have not yet
met the standards you set for me early in the Ph.D.: my handwriting is still horrible and a 90 hour week still surpasses me. I was inspired by Chris Udry's work as
an undergraduate. I feel extremely fortunate to have learnt from you as a graduate
student. I thank Mike Boozer for investing his time and super brain power in guiding me. Your support has been exceptional, I would not have reached this point
without it. Finally, thanks to David Atkin, Abhijit Banerjee, Bob Evenson, Mushfiq
Mobarak, Nancy Qian, Gus Ranis, Steve Redding, Peter Schott, Paul Schultz, T.N.
Srinivasan and Melissa Tartari for comments and advice.
I am grateful to my parents, siblings and dogs for supporting my academic
pursuits with such patience. To my grandmother: you have provided a phenomenal litmus test, Toronto asked to hire you over me. To Andreas: thank you for a
balance without compromise. I dedicate the introduction to you, written from a
sandbank flecked with gold on the Congo river.
I would like to thank my friends and colleagues at the Economic Growth Center
and beyond for their support. Thanks in particular to Katrina Jessoe, Vis Taraz,
Rachel Heath, Gharad Bryan, and Alice Klygne, for intellectual and moral support,
gallons of tea, cupcakes and laughter, and to the Slutskys for making grad school
sing.
VI
CHAPTER 1
Introduction to Dissertation
Economic growth is associated empirically with rising investment in education.
This empirical regularity could be driven by an income effect as well as by a returns to education effect. Growth may directly raise the returns to schooling, for
example skill biased technological change that raises the demand for educational
skills (Nelson and Phelps, 1966; Schultz, 1975). Growth also alters the level and
distribution of income through changing the returns to unequally distributed productive assets (Kuznets, 1955). The literature on the demand for education in developing countries has found that education responds positively to both incomes
(Jacoby and Skoufias, 1997) and returns (Kochar, 2004; Foster and Rosenzweig,
1994a). The relative contribution of these channels is however unclear since the literature has typically focused on one determinant of education at a time. Focusing
on one channel may confound the two effects if growth raises wage and income
levels as well as alters returns to education.
The US literature on skill biased growth finds that growth in the demand for educated workers contributes to the explanation of changes in the US wage structure
(Card and Dinardo, 2002; Katz and Murphy, 1992). Translating this to a developing
country context, how does growth in the demand for skilled versus unskilled labor
in the non-agricultural sector alter incomes and inequality? The literature examining the impact of non-agricultural growth on rural outcomes has considered this
sector as a homogeneous entity (Eswaran et al., 2009; Lanjouw and Murgai, 2009;
Foster and Rosenzweig, 2004; Datt and Ravallion, 1999). However, industry level
data from India suggest that industries vary substantially in the tasks they require
their population to perform and in their demand for skill.
l
1. INTRODUCTION
This dissertation makes two principal contributions to understanding how economic growth alters household education choices and incomes. Chapter 3 jointly
investigates the effects of rising incomes and returns to education on educational
investment in rural India between 1983 and 1999. The two effects are separated
by examining two determinants of wage and income growth, which induce independent variation in household incomes and returns to education. The sources of
growth examined induce exogenous movements of labor between the agricultural
and non-agricultural sectors. Chapter 2 examines the response of the level and
distribution of incomes in rural areas to an expanding manufacturing sector.
The approach taken in this dissertation is to examine the response of wages,
household income and education to two determinants of growth: region-time varying agricultural productivity and policy driven region-time variation in skilled
and unskilled manufacturing employment. Agriculture and manufacturing differ in their demand for skill and land. Hired labor in agriculture conducts low
skill tasks such as harvesting. Labor in manufacturing conducts skilled and unskilled tasks ranging from managerial to manual. Exogenous variation in the size
of these sectors shifts skilled and unskilled labor demand differently, and changes
the return to land. Therefore, sector specific shocks differ in their effect on returns
to education, as captured by the ratio of skilled to unskilled wages, as well as
on household incomes, according to their skill and land assets. I investigate how
these two sources of growth alter returns to education and household incomes,
and how changes in returns and household incomes alter household educational
investment.
To frame my empirical analysis, I build a small-country general equilibrium
model of household education choices in rural labor markets. The model illustrates the effects of increases in agricultural productivity and policy induced expansions in skilled and unskilled manufacturing labor demand on returns to education and households income, according to their skill and land endowments. The
1. INTRODUCTION
model highlights a source of endogeneity bias: regional manufacturing employment reflects local wages and unobserved determinants of education. The model
motivates an instrumental variables strategy to overcome this bias: industries are
attracted to locations endowed with the raw materials they use. Changes to industry level import tariffs and regulations therefore alter labor demand in regions
endowed with the resources the industry uses.
The predictions of the model are tested using a district and household level
data-set covering over 250 districts in 15 major Indian states between 1983 and
1999. The data-set includes district-level mineral, metal and energy endowments
and is compiled from maps. India presents the perfect setting for studying how
shocks to the agricultural and manufacturing sectors alter equilibrium outcomes.
India in 1983 was split into 412 districts and migration between districts is low.
The empirical specifications examine wage, household income and education responses to within district over-time variation in agricultural productivity and instrumented skilled and unskilled manufacturing employment. Agricultural productivity is measured using a Lespeyres index of regional agricultural potential: I
apply the national time-varying yield frontier in seven crops to the fraction of each
region physically suited to each crop. Changes in the national crop frontier occur
due to the introduction of new varieties of seeds, for example.
In the first stage instrumental variables specification, I use the predictions of
the model to induce variation in manufacturing employment that is uncorrelated
with unobserved determinants of income and education. I exploit the observation
that in the early 1990s, Indian industries were located in districts endowed with
the materials they use intensively. For example, cement industry employment is
concentrated where limestone is found, while plywood industry employment is
higher in densely wooded districts (figure A.l). This may reflect Indian industrial
planning, which encouraged states to promote industries that use the resources
1. INTRODUCTION
found in the district (Government of India, 1969). It may alternatively reflect decisions by industries to locate near relatively immobile inputs (Krugman, 1991).
Changes to industry policy are predicted to directly affect employment in districts
endowed with the materials the industry uses. India relaxed industrial regulations
and reduced import tariffs during a series of crises starting in 1984 (Topalova, 2004;
Srinivasan, 2002). The timing and depth of policy changes varied across industries.
The paper exploits within-district over-time responses of manufacturing employment to policy changes, where the identifying variation comes from differences
across industries in the broad types of raw materials used and in industry policy
over time. Employment is broken apart by skill by exploiting variation in the timing and depth of policy changes across industries with different skill intensities.
The wage and income results in chapter 2 indicate that the two drivers of
growth differ in their impact on the returns to education and household incomes.
An increase in agricultural productivity or instrumented unskilled manufacturing
employment decreases returns to education, but has income effects that vary by
skill and land ownership. An increase in skilled employment increases returns to
education for all workers but only raises income in educated households. Households with different initial land and education endowments exhibit heterogeneous
income responses to different drivers of growth.
The education specifications in chapter 3 examine whether children aged between 5 and 9 start school. Unskilled manufacturing growth increases the probability a child starts school in unskilled landless households, while it reduces it in
landed households. The intuition lies in the income effect: unskilled manufacturing growth raises income in unskilled landless households, while it has no effect
in landed households.
The income and returns to education effects on educational investment are estimated using minimum distance. Since agricultural productivity and manufacturing employment alter educational investment through incomes and returns to
1. INTRODUCTION
education, the coefficients on these two terms in the education specification reflect weighted combinations of income and returns to education effects, where the
weights are wage and income responses to the two determinants of growth. I use
the estimated coefficients from the education, wage and income specifications to
jointly estimate the two effects for rural landless households.
I find that income growth accounted for a larger fraction of the rise in education seen between 1983 and 1999 than growth in returns to education. Male educational investment responds positively to rising household incomes and labor
market returns to education. In contrast, investment among girls only responds to
rising incomes. This finding fits with the observation of low female participation in
complex occupations in rural areas. I test whether raising the child wage reduces
education through the opportunity cost channel. I find no statistically significant
effect, consistent with the low incidence of child labor among young children.
The estimates shed insight into a fundamental question about the process of
development: how does an expanding non-agricultural sector alter the level and
distribution of incomes in predominantly agrarian areas? Between 1983 and 1999,
unskilled manufacturing growth reduced income inequality by a third, while agricultural productivity growth raised it by 150%. However, manufacturing growth
contributed little to raising poverty relative to agricultural productivity growth.
Manufacturing growth explains approximately 27% of the reduction in poverty
and 8% of the increase in unskilled wages, while agricultural productivity growth
explains 21% of the increase in unskilled wages and 53% of the decine in poverty.
The small relative impact of manufacturing is partly attributable to its skill composition: skilled manufacturing contributes little to reducing poverty or raising
unskilled wages. My estimates suggest that, had growth in the manufacturing
sector been purely unskilled, India would have experienced a 9% greater increase
in unskilled wages and a 13% greater reduction in poverty. The small impact of
manufacturing growth on rural poverty has repeatedly been asserted in the policy
1. INTRODUCTION
literature (Kocchar, 2004). The estimates in this paper are the first to empirically
validate this observation and to provide an explanation for w h y this is the case.
The results suggest that, in terms of male educational investment, a rising tide
lifts all boats. Regardless of the source of growth, education rises among the poorest households in India - rural, unskilled, landless households. Sources of growth
that raise incomes but reduce returns to education, such as agricultural technological change and unskilled manufacturing growth, increase educational investment
substantially. Skill biased growth raises returns but has little impact on incomes; it
raises education to a lesser degree through the returns to education channel. The
same is not true for girls, whose education was found to only respond to rising
income: skill biased growth has little impact on their education.
This dissertation contributes to an extensive literature on the demand for education in developing countries. 1 The two determinants of education investment
have not previously been captured in one paper; a contribution of this dissertation
is to estimate them jointly and capture their relative effects. The dissertation also
contributes to a growing literature on non-agricultural growth in rural areas.
deviate from the literature by breaking manufacturing apart by skill and finding
an exogenous source of variation in it.
"*To n a m e b u t a few, the d e m a n d for education has b e e n s h o w n to r e s p o n d to rising r e t u r n s to

e d u c a t i o n (Kochar, 2004; Rosenzweig a n d Foster, 1996), unanticipated a n d anticipated g r o w t h in
h o u s e h o l d incomes ( E d m o n d s , 2006; E d m o n d s et al, 2004; and Jacoby a n d Skoufias, 1997), the
o p p o r t u n i t y costs of schooling (Rosenzweig a n d Foster, 2004) a n d to changes in the direct cost of
schooling (Schultz, 2004; K r e m e r et al. 2003).
P r o m i n e n t p a p e r s i n c l u d e L a n j o u w and Murgai, 2009; E s w a r a n et al., 2009; Foster a n d Rosenzweig, 2004; Datt a n d Ravallion, 1999.
CHAPTER 2
How does growth in the demand for skilled and unskilled labor
affect household incomes and the rural wage structure?
2.1.
INTRODUCTION
Growth alters both the level and the distribution of income through changing the
returns to unequally distributed productive assets (Kuznets,
1955).
Households
endowed with multiple productive assets will therefore experience heterogeneous

income effects when faced with different sources of growth. The US literature
on skill biased growth finds that growth in the demand for educated workers contributes to the explanation of changes in the US wage structure (Card and Dinardo,
2002;
Katz and Murphy,
1992).
Translating this question to a developing country
context, how does growth in the demand for skilled versus unskilled labor alter
the level and distribution of household incomes and inequality in rural areas?
The agricultural sector is the primary sector of employment in rural areas in
developing countries. Tasks conducted by hired labor in this sector are not complementary with education (Rosenzweig,
1995;
Foster and Rosenzweig,
1993).
Growth
emanating from the agricultural sector therefore increases the demand for predominantly unskilled hired labor. 1 Growth in rural areas emerging from the nonagricultural sector may differ in its demand for skill. The literature examining the
impact of non-agricultural growth on rural outcomes has considered this sector to
be a homogeneous entity (Eswaran et al.,
and Rosenzweig,
2004;
2009;
Datt and Ravallion,
Lanjouw and Murgai,
1999).
2009;
Foster
However, industry level data
^Growth in the agricultural sector h a s been f o u n d to alter equilibrium w a g e s (Jayachandran, 2006),

child labor (Foster and Rosenzweig, 2004b) and inequality (Foster and Rosenzweig, 2004a).
7
2. ECONOMIC GROWTH, WAGES AND INCOMES
from India suggest that industries vary substantially in the tasks they require their
population to perform and in their demand for skill.
In this chapter, I examine how growth in the demand for skilled versus unskilled labor in the non-agricultural sector alters incomes and inequality in rural
agrarian areas. I focus on the manufacturing sector, the largest non-agricultural
sector of employment (Government of India, 2002). I contribute to the literature by
allowing the effect of manufacturing growth to vary by skill and by using a novel
instrumentation strategy to deal with the issues of causality that have plagued the
literature.
To frame the empirical analysis, I build a small country general equilibrium
model of rural labor markets. The model illustrates the effects of increases in
agricultural productivity and policy induced expansions in skilled and unskilled
manufacturing labor demand on wages and the incomes of skilled and unskilled
workers.
The model highlights a source of endogeneity bias: regional manufacturing
employment reflects local wages and unobserved determinants of education. The
model motivates an instrumental variables strategy to overcome this bias: industries are attracted to locations endowed with the raw materials they use. Changes
to industry level import tariffs and regulations therefore alter labor demand in regions endowed with the resources the industry uses.
I test the model's predictions using district and household level data from rural
India between 1983 and 1999. The data covers over 250 district level labor markets.
I compile a new dataset of district mineral and metal endowments which I combine
with data on industry raw material use and industry policies to instrument skilled
and unskilled manufacturing employment in my first stage regression. I then examine wage and household income responses to within district over-time variation
in agricultural productivity and instrumented skilled and unskilled manufacturing employment. Agricultural productivity is measured by a Lespeyres index of
regional agricultural potential: I apply the national time-varying maximum yield

in seven crops to the fraction of each region that is physically suited to each crop.
The national agricultural productivity frontier shifts out due to the introduction of
new seed varieties, for example.
The wage and income results indicate that the two drivers of growth differ in
their impact on the returns to education and household incomes. An increase in
agricultural productivity or instrumented unskilled manufacturing employment
decreases returns to education, but has income effects that vary by skill and land
ownership. An increase in skilled employment increases returns to education for
all workers but only raises income in skilled households. Households with different initial endowments exhibit heterogeneous income responses to the two drivers
of growth.
Between 1983 and 1999, agricultural productivity growth explained 53% of the
reduction in poverty and half of the rise in income among landless households,
nearly four times more than manufacturing growth explained. However, manufacturing growth dampened the rise in inequality over the period, while agricultural productivity growth raised it. The small relative impact of manufacturing on
poverty is partly due to its skill composition: skilled employment growth has contributed little to raising unskilled wages, a principal component of income among
poor, landless households. My estimates suggest that had growth in the manufacturing sector been purely unskilled, India would have experienced a 9% greater
increase in unskilled wages and a 13% greater reduction in poverty.
The chapter is structured as follows. The next section describes the related literature. Section 3 presents descriptive statistics which are used to motivate the
assumptions of my model. In section 4,1 put forward a model to frame the empirical analysis. I test the model's predictions in section 5 and section 6 concludes.
10
2.2. RELATED LITERATURE: R U R A L LABOR M A R K E T S , D E T E R M I N A N T S OF

WAGES AND INCOMES
A key characteristic of the classic models of economic growth is a m o v e m e n t out

of agriculture and into the non-farm workforce (Kuznets, 1955; Ranis-Fei, 1961;
Ahluwalia, 1975). A central question that has occupied the literature on labor markets in developing countries is the effect of moving labor out of agriculture on the
average and marginal product of the remaining labor in agriculture (Rosenzweig,
1988). In section 2.2.1, I examine the literature on rural w a g e determination and
h o w this paper fits into the literature. In section 2.2.2, I review the literature on
non-agricultural growth in rural labor markets.
2.2.1. Determinants of Wages in Rural Labor Markets

The early literature on rural labor markets characterized the w a g e determination
process as uncompetitive. Rural wages were presumed to be set above the equilibrium w a g e resulting in under and unemployment. In the surplus labor models
of Lewis (1954) and Ranis and Fei (1961), the agrarian w a g e is set according to
the average product of labor. A movement out of the agricultural workforce into
a non-farm sector has no to little impact on the marginal product of the remaining workforce in agriculture i.e. the shadow w a g e of an agricultural laborer is
zero. 2 A neoclassical framework, based on the assumption of labor market clearing and competitive labor markets, suggests that shifting labor from agriculture to
manufacturing w o u l d increase the marginal product of the remaining workforce
in agriculture, and manifest itself in increased agrarian wages.
2
T h e literature takes three approaches to model the h o u s e h o l d f o u n d a t i o n s of these macro develo p m e n t models (Rosenzweig, 1988). The first set of m o d e l s p r e s u m e s that n o labor m a r k e t exists
at all and m o d e l s family labor as disguised u n e m p l o y m e n t . The second a s s u m e s that labor markets operate perfectly. The third set of m o d e l s constitute the nutrition-efficiency w a g e m o d e l s of
involuntary u n e m p l o y m e n t (Leibenstein, 1957; Mirrlees, 1975; Stiglitz, 1976; Bliss and Stern, 1976).
In this theory, w a g e s are influenced by an explicit link b e t w e e n productivity and c o n s u m p t i o n at
low levels of nutrition. In these models, competition a m o n g w o r k e r s does not reduce the w a g e rate
below an efficiency w a g e since a lower w a g e w o u l d not allow w o r k e r s to c o n s u m e e n o u g h to w o r k
efficiently.
11
The empirical evidence of w a g e determination a n d labor supply in developing

countries is broadly inconsistent with the predictions implied by the micro m o d els u n d e r p i n n i n g the labor surplus hypothesis (Bardhan, 1979; Lau et al., 1978;
A d u l a v i d h y a et al., 1979; Barnum and Squire, 1978; Rosenzweig, 1980; Strauss,
1983). This paper takes the neoclassical f r a m e w o r k to derive testable predictions
on h o w growth in non-agricultural labor d e m a n d alters the rural w a g e and income
structure.
2.2.2. Non-Agricultural
Groivth in Rural Labor Markets
There is a growing literature exploring the impact of rural manufacturing growth

and agrarian productivity on agrarian wages, poverty and the welfare of rural
households in developing countries (Eswaran et al., 2009; Lanjouw and Murgai,
2009; Foster and Rosenzweig, 2004; Datt a n d Ravallion, 1999). The effects estimated in the existing literature have been exposed to concerns about endogeneity d u e to the simultaneous determination of wages and manufacturing employment. A notable exception is Foster and Rosenzweig (2004), w h o use the NCAER
ARIS/REDS data-set to examine the relative impact of growth in the tradable manufacturing sector and agrarian productivity growth on wages, incomes and inequality in villages in rural India. They instrument manufacturing employment
using the Besley-Burgess measure of labor regulations that varies across states and
over time. 3 The literature to date has considered the tradable manufacturing sector as a h o m o g e n e o u s entity. However, industry level data from India suggest that
industries vary substantially in the education profile of their workforce.
2.3. SETTING: T H E R U R A L I N D I A N E C O N O M Y
In this section, I provide descriptive evidence supporting the central arguments

of the chapter. Section 2.3.1 shows that occupations vary substantially in their
complementarity with education across the agricultural and manufacturing sector.
o
The predictive p o w e r of this i n s t r u m e n t is h o w e v e r low.
12
Hired workers in agriculture conduct unskilled tasks, while those in manufacturing conduct skilled and unskilled tasks. Section 2.3.2 shows that wages in the rural
labor market rise with education and reflect the skill levels of tasks conducted.
The descriptive statistics covering India's 15 major states are from four r o u n d s of
household data collected in 1983,1987,1993 and 1999 by the National Sample Survey Organization (NSSO). In addition, I m a k e use of the 1982 and 1999 r o u n d s of
data from the Rural Economic and Development Surveys (REDS) collected by the
National Council of Applied Economics Research (NCAER). The Data A p p e n d i x
describes the data in greater detail.
2.3.1. Complexity of Tasks and Education in Rural India, by Sector

In this section, I provide evidence that agriculture hires workers to conduct unskilled tasks while manufacturing hires workers to conduct both skilled and unskilled tasks. Workers conducting skilled tasks have higher education levels.
Table 1-a presents descriptive statistics that divide the rural Indian workforce
into agricultural and non-agricultural sectors in 1983 and 1999. Agriculture employed a large fraction of the rural workforce in 1999 - 65% of males a n d 34% of
females. The size of the agricultural workforce varies substantially across India: in
Madhya Pradesh 85% of the workforce was employed in agriculture compared to
41% in Kerala and 61% in Tamil N a d u . The manufacturing sector employs approximately 31% of non-agricultural workers, making it the largest non-agricultural
sector of employment (Government of India, 1999). In 1983, the m a n u f a c t u r i n g
sector employed 8% of the male workforce while 13% of rural working age males
were working in the service sector (retail, construction and transport). In 1992,
manufacturing accounted for 17% of GDP while agriculture accounted for 23%
(Panagariya, 2008).
Agricultural workers can broadly be divided into those from landed households, w h o cultivate their o w n land and may work on the w a g e labor market,
13
and those from landless households, w h o predominantly work in the wage labor market. Table 1-a shows that approximately half of the agricultural workforce
worked for wages, 28.5% of rural males. In a weekly time recall, hired workers
report spending 99% of their time on unskilled tasks such as weeding and sowing. These wage laborers are disproportionately illiterate - in 1983, 71% of males
were illiterate, compared to 53% in the working age population as a whole. Illiteracy among females wage laboreres is substantially higher - 94% of females were
illiterate, compared to 83% in the population as a whole.
Skilled work is more prevalent amongst individuals who cultivate their own
land - in the REDS data, 13% of household agricultural time among landed households is devoted to supervisory activities. Foster and Rosenzweig (1995) present
evidence that education raises the productivity of individuals in landed households w h o undertake managerial tasks, such as deciding what varietals of seeds
to sow. Tables 1-a and 1-b show that individuals w h o cultivate their own household's land have higher levels of education than hired labor. Household heads,
who are likely to be making input decisions, are slightly more educated than ownhousehold cultivators who are less likely to be making these choices.
Labor in the manufacturing sector conducts both skilled and unskilled activities. Table 2-a examines education levels by worker category: skilled (white-collar
and blue-collar skilled) and unskilled (blue-collar unskilled). Unskilled work is
that in which purely physical tasks were reported. Occupations were classified using descriptions in the National Classification of Occupation (Government of India, 1968). Occupations described as "work doing purely physical activity" were
classified as blue collar unskilled while more technical descriptions - working with
machines, mixing paint using the right proportions of different chemicals - were
defined as blue collar skilled. White collar occupations include managers, secretaries and production-floor supervisors. The codes and categorizations are available upon request.
14
Approximately 35% of male and 55% of female occupations in manufacturing

fall into this category. Education decreases in the skill category of the occupation
conducted. White-collar and blue-collar skilled workers are comparable in their
levels of education - 78% of male white collar workers are literate compared to
72% of blue-collar skilled workers. Blue-collar unskilled workers have lower levels of literacy relative to the more skilled manufacturing groups - 49% were literate
in 1987. This is shown in table 1-b, which conditions on village fixed effects to alleviate concerns that the descriptive statistics reflect decisions by the manufacturing
sector to establish itself in more educated regions.
Table 2-b describes the different types of tasks conducted in manufacturing and
provides an indication of why education varies by skill category. Task definitions
come from the US Dictionary of Occupational Titles, 1977. Since there are likely
to be technological differences between manufacturing industries in the US and
India, this data provides some intuition for why workers conducting skilled occupations have higher levels of education than those conducting unskilled occupations. A crosswalk was created between the 1968 Indian National Classification of
Occupations and the 1960 US Census of Occupation Titles. This is available upon
request. The data on occupational skills comes from Autor et al (2003). I thank
David Autor for sharing this data with me. Each task variable is assigned a value
ranging from zero to ten, with higher values representing greater intensity of a skill
used. The five broad measures of tasks are routine manual activity, non-routine
manual activity, routine cognitive tasks, non-routine interactive tasks (direction,
control and planning) and quantitative and analytical tasks (mathematical reasoning). White collar occupations use quantitative and non-routine cognitive skills
more than both sets of blue-collar occupations. The blue-collar occupations are
quite similar in all attributes, with the exception of routine cognitive tasks which
are conducted to a far higher degree in the skilled blue-collar bracket.
15
2.3.2. Wages by Sector, Occupation and Complexity of Task

The previous section showed that skilled tasks in both manufacturing and cultivation are associated with higher levels of education than unskilled tasks. This
section presents evidence showing that rural wages vary with education and the
complexity of task conducted.
Table 1-c displays descriptive statistics on w a g e earners by sex. Literate individuals working in the w a g e labor market earn a 20% p r e m i u m over illiterate
workers. Separating workers by sector as well as by literacy, illiterate manufacturing workers earn a 10% w a g e p r e m i u m over illiterate workers in other sectors
(column c). This p r e m i u m appears to be driven by the complexity of the task performed rather than by education level - an illiterate person conducting an unskilled
task in manufacturing earns the same w a g e as an illiterate person in the other nonagricultural sectors. N o n - m a n u a l workers in agriculture earn a p r e m i u m over agricultural laborers. This should however be taken with a note of caution since the
n u m b e r of workers falling into this industry-occupation cell is extremely small 0.7% of hired labor in agriculture reported conducting these occupations.
2.4. T H E O R E T I C A L M O D E L
This section presents a static model of rural labor markets and household income
that builds u p o n Foster and Rosenzweig (1995, 2004a) and Ellison and Glaeser
(1999). The assumptions of the model d r a w u p o n the descriptive statistics presented in the previous section. The model provides a setting for u n d e r s t a n d i n g
the determinants of rural wage and income growth in the context of a two-sector
rural labor market. The p u r p o s e of the model is two fold. First, it delivers 3 sets
of predictions that are tested in section 2.5. Second, it highlights a source of endogeneity bias and p u t s forward exclusion restrictions to overcome it.
2.4.1. Model
16
Environment
H o u s e h o l d s live in a region whose borders define a closed labor market. A country

is m a d e u p of D regional economies. There are two sectors in each economy: agriculture and manufacturing, which are modeled using a specific factors framework.
The t w o sectors overlap in two locally traded inputs - skilled and unskilled labor.
Equilibrium wages and household incomes are determined within the model, and
vary with the regional economic environment. Economy subscripts are omitted
until necessary.
2.4.2. Households: Endowments, Preferences and Maximization
Problem
Households have three e n d o w m e n t s in both periods: land (A), adult education

(s a ) and a unit of adult labor market time. Household subscripts are suppressed
in this section. A d u l t education takes the value of 0 if the adult is uneducated or
1 if educated. Households consume a marketed consumption good and supply
their one unit of labor market time inelastically. Utility is concave in consumption
- 0 < p < 1, V(ch) = cj.
Households are unable to borrow or save, implying that their b u d g e t constraints clear every period. For tractability, there is no leisure; in the empirical
framework, I discuss h o w incorporating leisure alters the testable predictions. 4
Adults split their labor market time between four activities: unskilled work in
agriculture, unskilled work in manufacturing, skilled work in manufacturing and
skilled work in agriculture. Unskilled work in agriculture and manufacturing are
activities which both educated and uneducated individuals can conduct; skilled
work is defined as activities that only educated individuals can do. Skilled work
in agriculture consists of supervisorial activities, which can only be conducted by
^Jayachandran (2006) presents evidence indicating that labor s u p p l y responses to a p r o d u c t i v i t y
shock are m o r e inelastic in districts in w h i c h credit constraints are greater, in w h i c h a greater proportion of w o r k e r s close to subsistence a n d in w h i c h migration costs are high. In the empirical
section, I e x a m i n e h o w allowing labor s u p p l y elasticity to vary according to labor m a r k e t alters the
interpretation of m y results.
17
members of landed households.

1
where
and
l + lM + l(sa = l)(lM +
l(Ah>0)lf)
is unskilled time in agriculture and manufacturing; if is skilled
time in agriculture and manufacturing.

Households earn income from supplying labor to the w a g e labor market a n d
from cultivation if landed. They earn income m from an exogenous source.
wu(l^
+ l i f ) + l(sa = l)wsl^
+ l(A>0)YlA{sn)
+ m
The household problem is given by:

max
cph s.t.
2.4.3. Economy Environment
pcch
and
<y(sa)
Endowments
Each region is e n d o w e d with four immobile factors of production - a w o r k i n g

age population (P), land (A), two non-renewable raw materials (T\ and T2). The
population consists of P households of four kinds: landed educated, landed uneducated, landless educated and landless uneducated.
Land is distributed un-
evenly over landed households. Total factor productivity in agriculture, dj, varies
across regions according to a region's agro-climatic conditions.
Each region is
a small, open economy - final agricultural and manufacturing goods are freely
traded across regions within the country.
The marginal cost of extracting the immobile raw material decreases in the resource endowment. 5 This implies a mechanical relationship between the price of
raw materials and a region's endowments. The marginal cost of extracting raw
T h e immobility of raw materials is for tractability. A n alternative a s s u m p t i o n , with similar testable

predictions, is that prices increase in the distance from extraction, d u e to transportation and t r a d e
costs across regions.
18
materials is denoted by cri = c(r;-, Ty) for j = 1,2, w h e r e rj denotes extracted resources and Tj denotes the total regional e n d o w m e n t . The raw material is priced
at marginal cost, pri = crK The cost of extracting the resource is decreasing in the
total regional e n d o w m e n t - p!/ < 0- F r simplicity, the marginal cost of extracting
the resource is the same across all units extracted:
= 0. 6 The rents from raw
materials flow outside the region.
2.4.4. Agricultural
Production
Agricultural production combines labor with land to produce a single agricultural

output according to a concave, CRS technology. In the absence of a land market,
only landless households cultivate land. 7 Two types of labor are d e m a n d e d in
agriculture: unskilled and skilled. Unskilled labor undertakes physical tasks such
as weeding and harvesting. This labor can be conducted by both educated and
uneducated workers. Hired and family labor are perfect substitutes in unskilled
work. 8 Skilled labor undertakes managerial tasks, such as deciding which varietals of seeds to sow. Skilled labor can only be conducted by landed household
members. 9 Labor inputs are chosen to maximize profits, given the household's
land and education e n d o w m e n t s and the agricultural productivity faced:
llA
F(dU/ds;9,A,sa)
pAF(du,
max
du,ds
9(ds(6l+S2sa))^d2
ds; 9, A, sa) wudu
(2.1)
M^
T h e simplest structure for the r a w materials m a r k e t is presented. The a s s u m p t i o n s are m o d i f i a b l e

with minimal repercussions to the m a i n testable predictions. A m o r e complex structure w o u l d
allow for a fixed total quantity of r a w materials to b e c o n s u m e d every period, a n d for convexity in
the cost of extracting the materials.
The REDS data s u p p o r t this: 5.5% of landless h o u s e h o l d s cultivated land in 1999. The distinction
b e t w e e n l a n d e d a n d landless h o u s e h o l d s changes little over time: 5% of landless h o u s e h o l d s f r o m
1971 o w n e d land b y 1982 a n d 8% of a d u l t s b o r n in landless h o u s e h o l d s o w n e d land in 1999.
8
T h i s a s s u m p t i o n is m a d e for tractability. B h a r a d w a j (2009) s h o w s that hired a n d family labor can
b e considered to b e perfect substitutes in some, b u t n o t all, agrarian tasks.
9
T h e data motivate this a s s u m p t i o n - less t h a n 0.5% of hired w o r k e r s in agriculture c o n d u c t m a n agerial w o r k .
19
where: EI'4 denotes profits from cultivating land, net of hired labor expenses; du
and ds denotes unskilled and skilled labor; pA denotes the (externally set) national
price of agricultural output; 9 is the agricultural productivity, which varies by region and over time; A is household land; sa is the household adult's education and
zvu is the labor market clearing w a g e of unskilled labor.
Education of the household adult raises the marginal product of skilled labor
- F^(sa
= 1) Fd.(sa = 0) > 0. The productivity of unskilled labor is not rising
in the education levels of individuals conducting these tasks. This implies that educated and uneducated individuals conducting unskilled work are paid the same
1n
wage, i.e. there is no return to education in the hired labor market in agriculture.
I simplify the problem faced by households by assuming an interior solution
for supervisorial time. 11 A household's d e m a n d for unskilled and skilled labor is
given by:
du
du (iv, ws; pA, 6, A, s)
1 sn7i s"7i
ws n
ds(wU/ivs-/pA/6,A,sn)
=
<y\1"y2)7l2pA9(51
+
?2(l-s") s(l-72)
S2sa)ri\^
where w s is the wage of a skilled worker in the manufacturing sector. In the absence of a manufacturing sector, this is the shadow w a g e of educated supervisors
in the agricultural sector. The agrarian profit function can therefore be written as:
nA(zvu,zvs;pA,e,A,s")
10
pAF(wll,ws-,pA,e,Ah,sa)
-wudu(wu,
iv5-,pA,e,
A,sn)
F o s t e r a n d Rosenzweig (1993) find this to b e the case using piece rates data f r o m the Philippines.
^ I f the m a r g i n a l p r o d u c t of supervisorial time is greater than the relevant w a g e , l a n d e d a d u l t s will

only c o n d u c t o w n - f a r m supervisorial activity. This is m o r e likely as land h o l d i n g s increase a n d as
the agrarian technological frontier shifts out.
20
Household profits increase in agricultural productivity, adult education, land, the

price of agricultural products and decrease in wages.
Total unskilled and skilled labor d e m a n d is given by s u m m i n g labor d e m a n d
across landed households:
D
duh(wUfw5-pA,e,Ah,sah)
hePA
dsh(ivu,ws;pAl9,Ah,sah)
E
hePA
U n d e r the assumption of a CRS technology, the distribution of land will not affect the d e m a n d for labor unless education is unevenly distributed across landed
households.
2.4.5. Manufacturing
Production
The manufacturing sector uses four inputs in production: unskilled labor, skilled
labor and two raw materials. Only educated workers can conduct skilled m a n ufacturing work. Manufacturing production is modelled using a Cobb-Douglas
technology, which is increasing and concave in all inputs. Labor and raw materials are complements in production. There are two manufacturing industries that
vary in the output shares of inputs. Therefore two industries faced with the same
vector of input prices will vary in their optimal input choices.
The manufacturing sector competes with the agricultural sector for labor. Unskilled and skilled labor is remunerated at the same rate across the two sectors.
Both industries choose labor and raw material inputs to maximize profits, given
the prices and wages faced:
n; =
max
p % j F j ( N
j , N
s j
N,tj,Nsj,rij,r2j
F j ( N
u j
, N
5 j
, r y , r
2 }
K H }
* l
rYj, r2j) -
u j
s j
r y
p''2r2j
21
where Ely denotes profits in industry j, Nuj and NSj are unskilled and skilled labor
in industry j, ry and r2j denotes raw material one a n d two's inputs in manufacturing; pj^ - is the output price facing industry j; wu and ws are the unskilled and
skilled wages; and pri and pr2 are the time invariant prices of raw materials 1 and
2.12
Total d e m a n d for labor and raw materials in the m a n u f a c t u r i n g sector is given

by:
NM
Ey=i
NSM
E-=i N5j(wUf ws, f \ prl', p)
r f
L U r y ( w
rf
Y.)=xr2](wu,ws,f\fhp%)
u j
( w , Ws,
, w
, f \ f
r l
) p )
; p M )
L j
= 1
a y w - ^ j
nj=1
p ?
= E/=i a2}w-lyj
nt=i Ps*
F g
^LU^pîjjYlUipl"
= 1
pas
Manufactured goods have a perfect substitute produced in the world market.

I place myself in a small open economy setting in which the world price of the
perfect substitute ties d o w n the domestic price faced by industry j, p^-. Xj denotes
industry varying tariff barriers. C i j represents an industry varying per-unit cost of
complying to labor and industrial regulations. Regulations and tariff barriers are
the same in all industries across all economies within the country.
= Pttj + T; " C!j
(2.2)
Zero profits conditions imply that industries choose inputs until the price of o u t p u t
is equal to the marginal cost of production.
2.4.6. Equilibrium
Conditions
Using the t w o labor market clearing conditions for skilled and unskilled labor, the
ten first order conditions from production and the utility maximization condition,
it is possible to solve for the equilibrium quantities and prices of interest, notably
19
Intuitively, this requires the decrease in the stock of raw materials d u e to c o n s u m p t i o n d u r i n g
the period to b e small.
22
the wages of skilled and unskilled workers, the allocation of labor across sectors,
raw material inputs a n d the schooling choices of households.
Equilibrium w a g e s are determined by the labor market clearing conditions:

P
DA(W*U,W*S,ZA) + GA(WIW*,ZA)
+ NM(W*U,W*S,ZM)
+n
Ps
Gf(w*,zv*,ZM)
+ NsM(w*,w*,ZM)
m
s
KX,Z
(2.3)
m
)
(2.4)
where P is the total working age population and Ps is the educated working age
population. G s is educated skilled (supervisorial) labor in agriculture. ZA denotes
determinants that only shift agricultural labor d e m a n d and only alter manufacturing labor d e m a n d through the labor market channel, such as agricultural productivity. ZM denotes determinants that only shift manufacturing labor d e m a n d , such
as industrial policy.
The wages of skilled and unskilled workers are identical if the supply of educated workers at the skilled w a g e is greater than or equal to the d e m a n d for skilled
workers at the unskilled wage. Intuitively, this is because educated workers can
conduct unskilled work. If the d e m a n d for skilled workers is greater than the supply of workers at the unskilled wage, the skilled w a g e rises above the unskilled
wage.
Under these conditions, educated workers no longer conduct unskilled
tasks. I focus on the more interesting case in which the wages are different and
educated workers only work in the skilled labor market.
There are two regions of interest for industries in the manufacturing sector: the
corner solution where production is zero and the interior solution. The corner solution occurs where the existing unskilled wage, the s h a d o w skilled wage and raw
material prices are sufficiently high that the marginal cost of the first unit of production is greater than the price of output, or where the supply of literate workers
23
is zero. In the interior solution, manufacturing o u t p u t and inputs are chosen such
that price equals to marginal cost. From an initial situation where a unit of good
is produced at a price greater than marginal cost, manufacturing o u t p u t expands
pulling labor out of agriculture and into manufacturing. This raises the marginal
product of the existing labor in agriculture. This process continues until the zero
profit conditions are fulfilled.
2.4.7. Testable Predictions from the Model

2.4.7.1. Comparative Statics 1 - Manufacturing
Employment.
Regional manufac-
turing labor d e m a n d is a function of locally determined prices, as seen in 2.4.5.

To empirically examine how exogenous shifts in skilled a n d unskilled manufacturing labor d e m a n d raise wages, incomes and education, w e will need to find a
source of variation in labor d e m a n d that is uncorrelated with local determinants
of the outcome variables. Since manufacturing labor d e m a n d is a function of locally determined prices, empirically the challenge reduces to finding variation in
manufacturing employment that is uncorrelated with local determinants of wages,
incomes and education.
The model suggests a source of within-district over-time variation that can be
used - since industries differ in their raw material intensities, two industries faced
with the same changes in policy over time will experience different changes in
employment response within a region. Since agriculture and manufacturing only
overlap in the labor market, changes to the wages comes through greater competition between agriculture and manufacturing in the local labor market. The model
suggests a w a y in which the effects of skilled and unskilled shifts can be separately
identified. Since industries differ in their skill intensity, the skill composition of
manufacturing d e m a n d varies across regions. Variation in the timing of policy
changes across industries imply that the skilled and unskilled manufacturing labor d e m a n d curves are shifted at different moments in time. These predictions
24
constitute the base of m y identification strategy. The intuition behind them is p u t

forward in comparative statitics below.
l.a Raw material prices: capture the static distribution and industrial composition of skilled and unskilled m a n u f a c t u r i n g employment. O u t p u t decreases as raw material prices increase, ceteris paribus. The d e m a n d for
unskilled and skilled labor in manufacturing changes as the price of a raw
material increases. 13 The response of manufacturing labor d e m a n d to a
change in raw material prices increases in the raw material's share in total
output. 14
dp? T1 U /
dp /
l.b Policy Changes: induce changes in skilled and unskilled manufacturing employment that vary by region. A n increase in industry level tariffs raises
the domestic price of a good and increases industry output. Regions vary
in their o u t p u t and labor d e m a n d responses to changes in industry prices
d u e to their initial raw material e n d o w m e n t s . Within a region, labor dem a n d responses to changes in policy varies across industries. This implies
that regions that vary in their raw material e n d o w m e n t s see different employment shifs at a given m o m e n t in time. O u t p u t and labor d e m a n d responses to industry policy vary across districts with different raw material
Two effects operate. First, an increase in r a w material prices implies that, at the labor m a r k e t
w a g e , the zero profit condition n o longer h o l d s p r o m p t i n g a decrease in m a n u f a c t u r i n g o u t p u t .
Second, a c h a n g e in the relative price of the r a w material a n d labor i n d u c e s substitution a m o n g s t
inputs. If the f o r m e r effect d o m i n a t e s , an increase in the price of r a w materials r e d u c e s the d e m a n d
for skilled a n d unskilled labor. This occurs if the increase in the price of r a w materials r e d u c e s
o u t p u t sufficiently to o v e r w h e l m the increase in labor d e m a n d d u e to an increase in o u t p u t . This
is the case for both industries if the second order effect of a c h a n g e in equilibrium w a g e s d u e to a
shift of labor out of the agricultural sector d o e s n ' t o v e r w h e l m the direct price effect.
14
In the extreme case w h e r e the share of a r a w material in total o u t p u t is zero the labor d e m a n d
response to a r a w material price c h a n g e will b e zero, ceteris paribus.
25
prices for skill level k = s,u and industry j = 1,2:
dTj ^
'
dT^dpy
'
\dTjdpn)dakj
2.4.7.2. Comparative Statics 2: Wage Responses to Changes in the Aggregate Economic
Environment.
2.a Shifts in Skilled and Unskilled Manufacturing
Labor Demand driven by tar-
iff/deregulation policy changes raise the skilled and unskilled wage, ceteris paribus. Since the agricultural and manufacturing sector only overlap
in the labor market, industrial policy only affects agriculture through this
market channel. A shift in unskilled manufacturing labor d e m a n d pulls
unskilled labor out of agriculture, raising the marginal product of the existing labor. A shift in unskilled labor d e m a n d reduces skilled wages, since
it reduces the marginal product of the remaining skilled labor in agriculture. Vice-versa for exogenous shifts in skilled labor d e m a n d .
dio dN
dNu dpM
dws
dN
^
^
n
u
dw
DN,
'
~ pi
9N dpM < u>
dN,
dpM
diOi dN<
dpM
> U
2.b Agricultural TFP: raises both skilled and unskilled wage, since it raises the
marginal product of labor in agriculture:
dwu
w >
^
~
dlL\
w
>
2.4.7.3. Comparative Statics 3: Incomes.
3.a Agricultural
TFP: raises both skilled and unskilled wages, as well as the
returns to land. It thus raises the incomes of both landed and landless
households. In the absence of land markets, cultivation profits reflect the
26
returns to land therefore agricultural TFP raises cultivation profits.

^ > 0
de
3.b Shifts in Skilled and Unskilled Manufacturing
Labor Demand: induced by changes
in industrial policy raise the incomes of landless households. The income

responses of landed households vary with land endowments. The m a n u facturing sector competes with the agrarian sector for labor. Shifts out in
manufacturing labor d e m a n d reduce labor in agriculture and decrease the
marginal product of land. The net effect is negative for net importers of
labor (A > Ax), while it is likely to be positive for small landowners.
d y
dwu
dyL,(A>Ax)
L L
dwu
d N
d N
u
dwu
dwu
^Vl,{A<Ax)
d N
<
dpM
dw
dwu
>
dp
d N
BNU
d N
u
dp
3.d Inequality: The difference in incomes between landless and landed households decreases in unskilled manufacturing labor d e m a n d a n d increases
in agricultural productivity.
d(yA-yLl)
de
d(]M - Vll) dNu
n
>
'
dNu
d p .
n
<
2.4.8. From Theory to Empirics

The 3 sets of comparative statics p u t forward in the theoretical model trace out
an intuitive order for approaching the empirical analysis. First, I d r a w u p o n the
first set of predictions to p u r s u e an instrumental variable strategy by examining
the heterogeneous response of regional manufacturing employment to variation
in industrial policies. In section 2.5.2, I use the estimates from m y first stage regression to evaluate whether growth in agricultural productivity and predicted
27
manufacturing employment increases unskilled and skilled wages (prediction 2).

In section 2.5.4,1 examine whether the two sources of growth alter household incomes differently according to their initial assets (prediction 3). Throughout the
analysis, I break predicted manufacturing apart by skill to test whether the relationship between manufacturing growth, wages and incomes varies with the skill
composition of the manufacturing sector.
The data used in this chapter can be divided into five categories: industryregion level data on employment, wages, regional raw material endowments, industry varying factor intensities and time and industry varying policies. The data
will be presented as it is used. A more detailed description of the data can be found
in the appendix 2.
2 . 5 . E M P I R I C A L STRATEGY
The model makes predictions on wage and income responses to agricultural productivity growth and shifts in manufacturing labor demand induced by changes in
industry policy. This section derives the empirical specification used to test these
predictions.
The model applies to a locally clearing labor market. The appropriate empirical analogue in the Indian context is a district. Migration across districts and state
boundaries within India is low both in absolute terms and relative to other comparable developing countries (Munshi and Rosenzweig, 2009). In the REDS data,
8% of adult males had migrated from their villages of birth, and of these over 80%
went iwthin the district. 5.9% of male respondents in the NSSO from 1999 had
moved between 1990 and 1999, 55% of these movements occurred within the district. NSSO figures on temporary migration indicate that these flows are also low in 1999, 3.06% of males had temporarily migrated for work purposes. 1 5
15
T h e survey asks w h e t h e r individuals h a d t e m p o r a r y migrated for 60 d a y s or m o r e for w o r k

p u r p o s e s . This question therefore elicits responses on longer-term migration episodes and m a y
miss shorter-term higher frequency trips.
28
2.5.0.1. Wage Specification. Equilibrium unskilled and skilled rural wages are
determined by setting district level skilled and unskilled labor d e m a n d of equal to
their labor supply. For convenience, I repeat the labor market clearing conditions
from the model:
Ps
Df(iv*,w*,ZM)
+ N s M (w*, W*, Z M )
The log-linearized supply and d e m a n d equations in the agricultural and manufacturing sectors are given by the following system of equations:
Labor D e m a n d ^
ao
+aiio l l d t
+a2u>sdt
+a3^dt
~ûdt
Labor Demand^ 4
bo
+biioudt
+b2wsdt
+b5Xdt
+b4Z+ubdt
Labor Demand^
c0
+ciivudl
+c2ivsdt
+c3Xdt
+ucdt
Labor D e m a n d ^
do
+d-\Wlldt
+d2wsdt
+d3Xc?f
Labor Supply !(
Pu
Labor Supply s
Ps
-f-^Z^
+uddt
wlldt and wsdt denote unskilled and skilled wages in district d at time t, Xdt denotes
all common determinants of agricultural and manufacturing labor d e m a n d , Z^J
are all variables that only shift the manufacturing d e m a n d equation, conditional
on equilibrium wages and Xdt, such as the price of raw materials and industrial
policy. Z d t refers to variables that only shift agricultural labor d e m a n d , such as
agricultural TFP or the price of agricultural products.
Setting d e m a n d for skilled and unskilled workers equal to supply, w e find equilibrium wages. I evaluate labor d e m a n d for manufacturing at the equilibrium
wages and solve for equilibrium wages as a function of manufacturing employment:
29
K i t
"0 +
(2.5)
+ X d f T + Z ^ + Sd + St + edt
<dt
Po
+ ^^miskilled,dt^dt)
skilled,dt^dt^
+XDT +
where w*dt and

d
at time
w*dt
t , E^lsklUed
dt
+ ^
(2-6)
+ <5f +
edt
denote equilibrium unskilled and skilled wages in district

and
E^
j j l e d
dt
denote equilibrium employment of unskilled
and skilled workers in the manufacturing sector, 9 dt denotes the agricultural productivity frontier. .
A d d i n g and subtracting
dt
and
dt
from these equations and
suppressing the notation indicating equilibrium outcomes, w e obtain the final

wage regressions:
Wudt =
0 + *ltotal,dt + (^2 - l)Eilled,dt + a36dt +

+Sd
v>n,dt =
Po + Pi!italidt
+ ( f a - Pi)&Uedld,
+Sd
+ St
AdlK
dt
+ M t + Adtf +
+ St
(2.7)
(2.8)
edt
Appendix B.l shows the derivation of equations (2.7) and (2.8) in greater detail
and defines the a and /3 terms as a combination of agricultural labor d e m a n d parameters. For example, a\ captures a non-linear combination of the o w n and crossprice elasticities of skilled and unskilled agricultural labor d e m a n d . The derivation
of the equations in the appendix is in a more general form, in which labor supply
is responsive to changes in equilibrium wages.
30
2.5.0.2. Income Specification. The model implies that household incomes are affected by changes in the aggregate economic environment through two components of income: (a) labor market earnings and (b) profits from cultivation. The
empirical specification used to test the income predictions is given by the following specification, derived in appendix B.2:
Vhdt
To + 7iEtoffl/,df
+ Unskilled,dt
73 Odt
74 Adt
+ y
H H
d t
+ }id
+ St +
<phdt
Household level factors, such as the household's land and skill e n d o w m e n t , are
captured by
HHdt.
2.5.0.3. Issues with the Empirical Wage and Income Specification. There are t w o
reasons for keeping manufacturing employment directly in the empirical specification, rather than directly estimating the reduced form specifications. First, it
sheds light on the distributional implications of an expanding non-agricultural sector in rural agrarian areas. A key characteristic of the classic models of economic
growth is a m o v e m e n t out of agriculture and into the non-farm sector (Kuznets,
1955; Ranis-Fei, 1961; Ahluwalia, 1975). This model predicts that this transition
will alter the distribution of incomes in rural areas.
Second, this approach provides insight into whether promoting the agricultural or non-agricultural sector has a greater impact on the level of wages and
incomes. The ideal conceptual experiment w o u l d be to consider whether moving
a portion of the workforce from agriculture to manufacturing, and vice-versa, has
a greater impact on rural wages. To investigate which m o v e m e n t raises wages to
a greater extent, w e need to estimate the o w n and cross-price w a g e elasticities of
labor d e m a n d in agriculture and manufacturing. The strategy I use approaches the
question from a different angle - given the tools through which policy makers promote the agricultural and manufacturing sector, do wages and incomes rise more
through channels that promote the agricultural or manufacturing sector? The promotion of agricultural technical change and manufacturing jobs are major channels
31
through which policy alters rural outcomes in India. For example, policy makers
directly target job creation through manufacturing employment subsidies in West
Bengal and Orissa to attract industries to rural areas. The Government of India invests extensive resources into the research and development of n e w seed varietals
and outreach p r o g r a m s which promote the adoption of n e w technologies. 1 6
17
Estimating these equations directly is likely to lead to biased estimates of the

parameters of interest. The simultaneous determination of manufacturing employment and wages implies that any locally unobserved c o m m o n determinants of
agricultural and manufacturing labor d e m a n d could be driving the relationship
between wages and manufacturing employment.
For example, changes in the
quality of electricity services may increase the d e m a n d for labor in both sectors.
Therefore the OLS estimates of equations (2.7) and (2.8) will yield biased estimates
of the parameters of interest.
2.5.1. First Stage

This section describes the two-stage least squares strategy used to identify the
causal effect of shifts in manufacturing labor d e m a n d on rural wages a n d incomes.
The strategy used is motivated by prediction 2 of the model, which states that the
manufacturing employment response to a change in industrial policy varies according to the region's e n d o w m e n t s of the resources the industry uses intensively,
i.e. if the industry is expected to initially have been located in the district. The
instruments used are the interactions between the region's resource endowments,
the industry's raw material intensity and the industry's policy.
alternative a p p r o a c h is to directly examine h o w w a g e s r e s p o n d to a c h a n g e in agricultural
relative to m a n u f a c t u r i n g e m p l o y m e n t . U n d e r labor m a r k e t clearing, the coefficients on agricultural a n d m a n u f a c t u r i n g e m p l o y m e n t will c a p t u r e a m e a s u r e of the labor s u p p l y elasticity w h i c h
is the s a m e for b o t h variables.
1 7
India e m p l o y s m o r e agricultural researchers in the public sector than the US, a l t h o u g h its R&D
e x p e n d i t u r e s on agricultural are only about one tenth the size of US e x p e n d i t u r e s (Evenson et al.,
1999). It is also placed a m o n g one of the biggest s p e n d e r s in d e v e l o p i n g countries in this area; India
w a s only s u r p a s s e d b y C h i n a and Brazil on the a m o u n t it s p e n d s on agricultural research.
32
Section 2.5.1.1 explains w h y this strategy has predictive p o w e r a n d section

2.5.1.2 motivates their validity. The results from the first stage estimations are presented in section 2.5.1.5.
2.5.1.1. Why are these instruments likely to be correlated with manufacturing
employ-
ment? I combine two different literatures to explain w h y the set of instrumental

variables are likely to be correlated with manufacturing employment. First, industry locations have been found to be related to resource and labor-market natural
advantages across US states (Ellison and Glaeser, 1999; Kim, 1999). Ellison and
Glaeser (1999) find that location responses to resources vary according to the intensity with which an industry uses a resource. 1 8 The relationship between the
natural resources and industrial employment m a y also reflect the 4th Industrial
Plan of India, which encouraged states to promote industries which intensively
used the raw materials found in those regions (NCAER, 1969a,b,c). Second, reductions in industrial regulations and trade barriers between 1985 and 1997 have been
f o u n d to have h a d large effects on manufacturing o u t p u t and employment in India
(Aghion et al 2009; Hasan et al, 2007; Chari, 2008; Topalova, 2004; Kambhampati,
1997). The first-stage empirical model is built on the overlap between these t w o
literatures. If the static distribution of industrial employment in Indian districts
can be explained by the interaction between an industry's resource usage and regional resources, the interaction of these variables with industrial policy captures
variation in employment responses across districts with different initial industrial
concentrations.
2.5.1.2. Validity of Instrumental
Variables. In this section, I discuss the validity
of the IV strategy. Prediction l.b states that within district employment responses
1s
For example, the highest concentration of a l u m i n i u m p r o d u c t i o n , an electricity intensive i n d u s try, is f o u n d in Washington, the state w i t h the lowest electricity prices.
33
across industries to changes in policy vary according to the industries' technologies. This implies the following empirical specification:
^ total/skilled,id
ft
+ M f + /32T,f * Sf + 0 3 T; t * S; * fld + 0 4 T ft * Hd
(2.9)
+fesi *nd + p6Si + p7nd + p8Xdt + Sd + St + uidt

where
^ total/skilled
idt
total or skilled manufacturing employment in industry z,
district d at time f, r,f are industry level policies at time t, s, is a measure of an

industry's use of a given input, and n d is a measure of the district level e n d o w m e n t of that input. 1 9 I treat the triple interaction term as excludable. 20 To consistently identify the coefficients on total and skilled manufacturing employment in
the w a g e and income specifications, the excluded variables should be uncorrelated
with unobserved determinants captured by the error terms in the outcome regressions. To separately identify the coefficients on total and skilled m a n u f a c t u r i n g
employment, the instruments m u s t also generate independent variation in these
t w o variables. I discuss this further in section 2.5.1.3.
The f r a m e w o r k implies that the only channel through which changes in industrial policy affects agrarian wages and incomes is through competition between
agriculture a n d manufacturing in the rural labor market. Empirically, this implies
19
A c o n c e p t u a l l y e q u i v a l e n t a p p r o a c h w o u l d b e to a g g r e g a t e i n d u s t r y - d i s t r i c t level e m p l o y m e n t
across i n d u s t r i e s to a r r i v e at a district level strategy:
^ total/skillrd,dt
where f t
t i m e t, TSf =
nd + J 4 f t * nd +
+ 75S * nd + j6s + 7 7 n d + 7 $ X d t + Sd+5t
70 + ? l T , + 7 2 t 5 , + 73TS, *
+ vd,
Tit c a p t u r e s t h e a v e r a g e level of the policy of interest in t h e e c o n o m y at

j
T,f * s, is a w e i g h t e d a v e r a g e of i n d u s t r y tariffs at t i m e t, a n d r s ( * nd
j
tn * s; * n d is t h e interaction of r e s o u r c e - u s e w e i g h t e d policy w i t h the r a w m a t e r i a l r e s o u r c e
e n d o w m e n t of t h e district. A g g r e g a t i n g tariffs across i n d u s t r i e s c o m p r e s s e s t h e v a r i a t i o n in timing a n d m a g n i t u d e , a l t h o u g h the c o n c e p t u a l s o u r c e of v a r i a t i o n is t h e s a m e . For this r e a s o n , t h e
d i s t r i c t - i n d u s t r y regressions are m y p r e f e r r e d specification. I r e p o r t results f r o m b o t h sets of regressions.
20
In t h e first s t a g e specification, I i n c l u d e o w n - i n d u s t r y policy c h a n g e s a n d o m i t c r o s s - i n d u s t r y policy channels. For t h e s e o m i s s i o n s to i n v a l i d a t e t h e identification strategy, t h e variables c a p t u r i n g
h o w i n d u s t r y /'s tariff c h a n g e affects j's e m p l o y m e n t n e e d to b e correlated w i t h t h e triple interaction t e r m as well as e n t e r directly into t h e w a g e regression. A p p e n d i x B.3 d i s c u s s e s t h e s e c o n c e r n s
in g r e a t e r detail a n d s u g g e s t s a l t e r n a t i v e a p p r o a c h e s w h i c h h e l p v a l i d a t e the strategy.
34
that all policy interactions can be treated as excludable from the w a g e and income
regressions. In the empirical strategy however all lower order terms are included
in the second stage regression a n d only the triple interaction terms are excluded.
Intuitively this allows industrial policy to alter agricultural productivity through
nationally traded agricultural inputs produced by the m a n u f a c t u r i n g sector, such
as fertilizers. This effect is allowed to vary according to the region's raw material
e n d o w m e n t s and the region's distance from trading h u b s (major p o r t s / s t a t e capitals). The identifying assumption needed is that labor is the only locally traded
overlapping input between the agricultural and manufacturing sectors, a n d that
output markets for neither agricultural nor manufacturing commodities clear at a
local level. I examine the robustness of m y results to these assumptions in section
2.5.3.
The next p a r a g r a p h s explain in greater detail w h y the lower order terms are
not excludable from the wage and income regressions b u t the triple interaction
terms are. First, district level time-invariant resource e n d o w m e n t s are likely to be
correlated with time invariant determinants of agrarian wages. For example, the
mineral e n d o w m e n t s of a region are correlated with its soil quality, an unobserved
input in the agricultural production function. 2 1 These time invariant district level
terms are thus captured by district fixed effects.
Second, average effects of policy changes are captured by time dummies. The
removal of import tariff barriers has been f o u n d to alter both the price and n u m b e r
of imported intermediate goods sold, as well as the n u m b e r of final goods produced by Indian firms (Goldberg et al, 2009). If the market for final manufacturing
products is at a national rather than regional level, the agricultural input response
to a change in goods markets can be captured using year fixed effects. To capture
21
Chemical and mechanical w e a t h e r i n g at the E a r t h ' s surface d e c o m p o s e s rocks a n d the m i n e r a l s

within those rocks to f o r m soil (Kessler, 1996). The soil's nutrient content, profile a n d composition
will d e p e n d on the texture, structure, chemical a n d mineralogical c o m p o s i t i o n of the p a r e n t m a terial. Therefore, the u n d e r l y i n g quality of the soil for agricultural p u r p o s e s is likely to b e highly
correlated with the m i n e r a l composition of ores in a district.
35
variations in a region's exposure to imported goods, I include the interaction of

average tariff changes and the distance from trading hubs.
Finally, the interactions between policy and endowments capture how employment response to policy changes vary with regional resource endowments. A concern is that aggregate policy changes may affect regions differently according to
their raw material endowments. 2 2 In addition, time trends in wages may vary
with initial conditions. For example, densely wooded regions may have fewer
urban settlements or the baseline size of the urban population may be associated
with differential time trends in rural wages. The policy and endowment interactions are therefore include in the second stage specification. I allow for regional
time trends in my specification to address the concern that districts may be experiencing different trends in wages and industrial growth.
Conditional on the lower order terms, the interaction between resource weighted
industrial policy and regional resources, r,-f * s, * nd, captures variation in regional
employment driven by technological variation across industries and variation in
the timing and depth of policy across industries. By definition, the variables inducing the identifying variation are location invariant. Therefore, these terms are
unlikely to capture any local unobserved variation in baseline conditions that may
be correlated with agrarian wages nor are they likely to be correlated with time
trends or aggregate policy responses that vary according to district level baseline conditions. I verify whether this is the case by examining whether change
in the excluded variables between 1983 and 1999 is partially correlated with baseline characteristics. The instrumental variables are not systematically and jointly
correlated with the level of wages in 1980, male or female literacy or the size of the
population. To address the second concern, I examine the robustness of my results
22
E x t e n d i n g the soil quality e x a m p l e above, if changes in aggregate policy r e d u c e the price of rice
seeds, w e w o u l d expect profits in areas to b e differently affected by policy changes according to
w h e t h e r rice is cultivable.
36
to including region level growth trends and time trends interacted with baseline
characteristics.
A n u m b e r of p a p e r s have examined the impact of changes in national level
industrial policy on total factor productivity, the capital to labor ratio, and input
mix in India (Aghion et al., 2009; Chari, 2008; Topalova, 2004; Krishna and Mitra,
1998). The estimation method proposed captures manufacturing labor d e m a n d
responses d u e to changes in industrial policy that alter the marginal cost to price
mark-up, but m a y also capture increases in labor d e m a n d d u e to changes in total
factor productivity.
A key identifying assumption is that the agricultural and manufacturing sectors only overlap in the local labor market. The coefficients on manufacturing employment in the w a g e regressions are identified if unobserved district-time varying
determinants of agricultural productivity (such as pesticide use) are uncorrelated
with the instrumental variables. The income regressions require the additional assumption that the instruments are uncorrelated with income from sources other
than cultivation and hired work in manufacturing. The instrumental variables
capture all industry-region level responses to a change in tariff changes, such as
changes in output, input d e m a n d for all inputs other than labor and prices. If agriculture and manufacturing are linked through other locally clearing markets and
the lower order interaction terms don't fully capture the price response in these
markets to changes in d e m a n d , the exclusion restrictions may be violated. I test
the robustness of m y results to other potential channels in section 2.5.3.
2.5.1.3. Identification of the Effects of Skill Composition. Shifts in labor d e m a n d
are broken apart by skill by exploiting variation in the timing and degree of tariff
The evidence on the relationship b e t w e e n t r a d e liberalization a n d total factor productivity is
mixed - Tybout et al. (1991) finds n o i m p a c t of increased p r o d u c t i v i t y after liberalization in the
Chilean context, b u t a n u m b e r of studies find evidence of productivity increases in other contexts
(Harrison (1994), Pavnick (2002) a n d Topalova (2004). The evidence e m e r g i n g f r o m the Indian
t r a d e r e f o r m s indicates that there is i n d e e d a case to b e m a d e for t r a d e liberalization in this context
h a v i n g led to an increase in TFP (Goldberg et al., 2009; Aghion et al., 2009; Topalova, 2004; Krishna
a n d Mitra, 1998).
37
movements across industries with different proportions of literate workers. For

example, assume there are two industries. Industry 1 employs 50% skilled workers and experiences a drop in tariff barriers 3 years before industry 2, which only
employs unskilled workers. The direct effect of changes in industry l ' s tariffs will
shift both the demand for skilled and unskilled labor in locations in which industry 1 is predicted to be present 3 years before industry 2's unskilled labor demand
OA
curves are shifted.
This strategy separately identifies variation in skilled versus unskilled employment if industries vary in their technological requirements for skilled labor. The
education and skill composition of the manufacturing workforce in India varies
substantially by industry. While part of the variation may be attributed to variations in tasks by industry, it is also likely to depend on the relative skilled and
unskilled wages in a given location. To examine how much of the variation in
the proportion of skilled workers in a region-industry cell is industry specific and
how much is due to regional variation in relative wages, I conduct an Analysis of
Variance of region-industry level skill and education proportions using region and
industry fixed effects. Industry level dummies capture variation in the proportion
of skilled workers that is common to the industry across all locations (e.g. technology, type of work conducted). District level factors capture variation in the skilled
proportion of the workforce that is common to all industries within that district.
For example, in areas where unskilled wages are low, both relative to the labor
market wage for literate individuals in that district and in absolute terms, all firms
within that district will be likely to employ a greater proportion of illiterates.
Table 3 reports the results from the variance decomposition for 1983 and 1999.
The industry level dummies capture 21.5% of the variance in the proportion of
literate and skilled workers employed and are jointly statistically significant at a
24
Both industries will of course b e affected b y b o t h sets of tariff r e f o r m s t h r o u g h general equilibr i u m effects, the m o s t simple of w h i c h w o r k s t h r o u g h equilibrium w a g e s at a district level. Focusing on the direct effect of o w n - i n d u s t r y r e f o r m s will n o t violate the exclusion restrictions as long as
labor is the only district level interaction b e t w e e n agriculture a n d m a n u f a c t u r i n g .
38
1% level with an F-Statistic of 17.32 for literate workers. Therefore industry level
factors explain a large and statistically significant fraction of the variance.
2.5.1.4. Data. I use data on industry-time level policies, industry-level raw material intensity, industry-region-time level skilled a n d unskilled employment and
regional resources. The data are discussed in greater d e p t h in appendix 2.
Resources groupings follow the main categories of industrial usage in the Mineral Atlas of India and various NCAER Economic Plans. Raw materials are g r o u p e d
into construction, metals, chemicals, ceramics, w o o d and energy. Mineral and
metal e n d o w m e n t s were captured by geo-coding the National Mineral Atlas of
India (Gov of India, 1995). Industry level policies from Aghion et al (2008) are captured using a continuous import tariff measure and a 0-1 industry deregulation
measure. Industrial employment is measured using 3-digit National Industrial
Classification codes (Government of India, 1987). Industry resource use is measured using the Input-Output matrix of India (1971 and 1993). I impute the share
of total costs for a resource group and convert it into an above-median d u m m y
and quantile measure.
2.5.1.5. First Stage Regressions: Results. In section 2.5.1.5,1 verify the framework
by examining whether regional resources a n d industry resource intensity predicts
the static distribution of industry-region level employment. In section 2.5.1.5 I
discuss the results from the first stage regressions.
Do resource endowments help to explain the static geographic distribution of employment?
The results from the static specification are presented in table 5. All specifications
include lower order interaction terms and region fixed effects. Standard errors are
clustered on a region level. The d e p e n d e n t variable is the log of total regional industry employment; factor intensity is measured using a d u m m y indicator. The
variables are presented in descending order of transportability. Appendix table
39
A . l presents the density and price per tonne of raw materials. The cost to the consumer of minerals with a low unit value increases with distance to the place of
use and the difficulty of transporting the material (the lower the density). Consequently, low unit value commodities are of little value unless processed close to
their source. I focus on these relatively immobile inputs.
The interaction between an industry's input intensity and the region's resource
e n d o w m e n t captures prediction 1-a: employment is more responsive to a raw m a terial in industries that use the material more intensively than those which use it
less intensively. The results indicate that the availability of bulky materials such as
wood, construction minerals and ceramics minerals are positively and significantly
correlated with employment. For example, in ceramics intensive industries, employment is 38% greater in areas in the top quartile of ceramics e n d o w m e n t s than
in the bottom quartile. In regions within the top quartile of ceramics materials,
employment in ceramics industries is 23.1% greater than in non-ceramics industries. Similarly, in regions within the top quartile of forest density, e m p l o y m e n t
in w o o d intensive industries is 48% higher than in non-wood intensive industries.
The coefficients are largest for the commodities which are bulkiest and least cost
effective to transport. By contrast, the availability of materials w h o s e price and
density combination make them more worthwhile to transport don't have a positive impact on the location of industries that use those materials most intensively.
The results support the hypothesis that raw material e n d o w m e n t s contribute to
the explanation of the static distribution of employment.
Does the interaction capture differential employment responses to policy changes? Tables 5 present Fixed-Effect specifications of changes in region-industry level employment over time. Columns (a) through (d) present estimates of employment responses to changes in industrial regulations while columns (e) through (h) present
40
results capturing the response to changes in import tariffs. All specifications include lower order interaction terms, region fixed effects and time variant district
characteristics including a region and time varying measure of agricultural productivity a n d the logarithm of the population, split by landless and landownership
quartiles.
The coefficients on the triple interactions confirm the hypothesis that employment responses to changes in industry time varying import tariffs and licensing
reforms vary by industry resource use a n d a region's resource endowments. The
lower order interaction terms are not s h o w n for reasons of parsimony, therefore
the heterogeneous employment response of industries across regions and industries is best illustrated by examples. In response to a 65% d r o p in tariffs (the average decrease between 1983 and 1999), densely w o o d e d areas saw on average a 4%
greater decline in employment than n o n - w o o d e d areas. The results are driven by
w o o d intensive industries: in densely w o o d e d areas, employment in w o o d intensive industries saw employment d r o p by 34% more than in non-wood intensive
industries. Similarly, a 65% drop in tariffs in the ceramics industry resulted in a
12% decrease in employment in areas in the top quartile of the ceramics e n d o w ment, compared to a 3% increase in areas in the bottom quartile of the ceramics
endowment.
2.5.2. Skilled and Unskilled Wage Regressions

In this section, I use the IV strategy from step 1 to estimate the response of wages
to changes in instrumented skilled and unskilled manufacturing employment. In
section 2.5.2.4,1 interpret the results and in section 2.5.3,1 examine the robustness
of the results to alternative explanations.
2.5.2.1. Empirical Specification. Prediction 2 suggests that both skilled and unskilled wages increase in agricultural productivity. Unskilled manufacturing employment is predicted to raise the unskilled wage, and to d o so by more than
41
skilled employment. The skilled w a g e is predicted to increase in skilled employment and to do so by more than in unskilled manufacturing employment.
The empirical strategy derived in section 2.5.1 is restated below for convenience:
0CQ + a !
Enl4t
(2 -
u>udt
0L\) billed,dt
Wmdt
= fa + faE^tahdt + (fa-fa)EîUeddt
30dt +
AdtK
+ Sd
+ fa9dt + Adtxp +
Sd
St + dt
+ St +
dt
where wudt and zvsdt denote equilibrium unskilled a n d skilled wages in district
d at time t, E^f
dt
and E^jl[ed
dt
denote instrumented total and skilled employment
in manufacturing; 9 denotes agricultural productivity, Adt captures all other determinants of wages. District d u m m i e s absorb all location invariant factors while
year d u m m i e s capture all c o m m o n year effects.
As s h o w n in appendix B.l, the structural parameters indicate that a.i > 0, oc\ >
a2 and ot3 > 0; and fa > 0, fa > fa and fa > 0. Therefore I anticipate that
the estimated coefficient on skilled manufacturing e m p l o y m e n t is negative in the
unskilled w a g e regression, while it is positive in the skilled w a g e regression.
2.5.2.2. Data. Prediction 2 is tested using measures of wages, agricultural productivity, characteristics of the rural population and instrumented manufacturing
employment. Agrarian daily wages are used to measure unskilled rural wages. 2 5
The m e d i a n district level daily w a g e earned by illiterate workers in the NSSO survey is used as an alternative measure. Skilled wages are measured as the median
daily w a g e of literate non-agricultural workers. Alternative measures include the
9c
W a g e w o r k e r s in the a g r i c u l t u r a l sector are d i s p r o p o r t i o n a t e l y illiterate relative to t h e p o p u l a t i o n
a v e r a g e a n d c o n d u c t m a n u a l p h y s i c a l tasks, as d i s c u s s e d in section 2.3.1. T h i s m e a s u r e t h e r e f o r e
c o n s t i t u t e s a g o o d p r o x y for the w a g e o b t a i n e d b y i n d i v i d u a l s w o r k i n g in u n s k i l l e d o c c u p a t i o n s .
42
wage of workers with primary and above primary education. 2 6 Skilled manufacturing employment is measured as literate workers. Alternative measures include
workers with at least primary education and those in skilled occupations.
I put forward a new approach for measuring agricultural technological change.
Agricultural TFP growth constitutes a moving out of the agricultural production
frontier. Changes in agricultural productivity can be captured using growth in
agricultural yields over time (Foster and Rosenzweig (2004), Jayachandran (2006),
Lanjouw and Murgai (2009)). Yields however reflect a combination of TFP growth
as well as endogenous responses of inputs to TFP growth (such as labor and irrigation). 27 Therefore using actual yields to measure agricultural productivity will
result in biased estimates of the parameters of interest.
Agricultural TFP is measured by combining information on the technological
frontier of crops across India with variation in the type of crops that can be grown
across districts. The technological frontier in India is captured by the maximum
of the India-wide yield in a particular crop. For example, Punjabi districts are
at the rice yield frontier. The national frontier is combined with a district-level
time-invariant measure of the physical suitability of crops to local agro-climatic
conditions. The Global Agro-Ecological Zones (GAEZ) project run by the Food
and Agriculture Organization (FAO) combines information on climatic conditions,
soil quality and terrain to make an index of the suitability of a region to growing
rice, maize, wheat, cotton, sugar, pulses and sorghum. For example, sorghum is
not physically suited to parts of Rajasthan but is well suited to most of Andhra
Pradesh. I combine the physical suitability of land to a crop with the crop level
yield frontier in India to capture the maximum potential productivity of a district
26
M e a s u r e s i m p u t e d f r o m the NSSO data exclude m a n u f a c t u r i n g sector w o r k e r s since the instrumental variables will b e directly correlated with the m a r g i n a l p r o d u c t of w o r k e r s in this sector
t h r o u g h various i n p u t channels.
27
Foster a n d R o s e n z w e i g (1996) m e a s u r e technical c h a n g e b y estimating the agricultural technology before a n d after the onset of the green revolution. This a p p r o a c h h o w e v e r requires detailed
cultivator level data on agricultural p r o d u c t i o n over time, w h i c h is unavailable for the years I examine.
43
at a point in time, where c represents crops, d districts a n d t time:
9dt
c
^ M a x i m u m National Yield in Crop c t *
(2.10)
*Proportion of District Most Suited to Crop c d * Crop Price c l 9 8 0

Weather shocks are computed using weekly rainfall data. The measures used are
total m o n s o o n rainfall, the square of total m o n s o o n rainfall and a weather "shock"
variable taking the value of -1 if rainfall is 50% below it's long term mean, and 1 if
it is 50% above.
2.5.2.3. Results. Wage estimates are presented in tables 6 and 7. The distinction
between skilled and unskilled m a n u f a c t u r i n g employment is p u t aside in Table 6,
panel A. The results for skilled wages are presented in table 7. In all tables, the
Fixed-Effect (FE) specification is presented in column (a) and column (b) presents
the Fixed-Effect Instrumental Variables (FE-IV) specification. C o l u m n s (c) through
(g) present specification and data checks using the FE-IV specification. The FE
coefficient captures two effects. Firstly as manufacturing e m p l o y m e n t grows, labor
is d r a w n out of agriculture raising the marginal product of the remaining agrarian
workforce. Secondly, manufacturing is attracted to areas with lower wages and
w a g e growth profiles. The FE-IV strategy isolates the first effect; therefore the FEIV coefficient is expected to be greater than the FE coefficient. 28
The FE coefficient on manufacturing employment reported in column (a) of
table 6-A shows a small statistically significant relationship between manufacturing employment and unskilled wages. The coefficient estimated using the FE-IV
strategy in column (b) is greater than the FE coefficient, confirming the hypothesis
that the estimated FE coefficient w a s d o w n w a r d biased. A 10% increase in manufacturing employment raises wages by between 1.1% and 1.7%, confirming the
78
For example, in the Indian context, cross subsidization b e t w e e n agricultural a n d industrial electricity prices implies that the t w o rates are negatively correlated. M a n u f a c t u r i n g is deterred f r o m
places with low agricultural electricity prices, w h i c h are in turn likely to raise p r o d u c t i v i t y in agriculture.
44
prediction of the model. An increase in agricultural productivity is predicted to

have a positive impact on agrarian unskilled wages. I find this to be the case: a
10% increase in the technological frontier increases wages by approximately 4%.
The slight increase in the coefficient on agricultural productivity between columns
(a) and (b) suggests that manufacturing employment orientates itself towards areas with low agricultural productivity, where the unskilled wage is relatively low.
Conditioning on region time trends in column (d) increases the estimated wage
impact slightly - this suggests that growth paths in manufacturing employment
and wages are negatively correlated. The estimated coefficients are of a similar
magnitude to the coefficient of 0.9% found by Rosenzweig and Foster (2004) using
village level data and a different IV strategy.
Table 6-B breaks manufacturing employment d o w n by skill. The model predicts that the coefficient on total manufacturing employment is positive, while
that on skilled employment is negative - (a.2 <x.\) < 0 . The FE-IV coefficient estimates in column (b) confirm these predictions: a 10% increase in purely unskilled
(illiterate) manufacturing employment raises agrarian wages by 3% while an increase in purely skilled (literate) employment raises wages by 0.6%. The larger
change in the coefficient estimate for unskilled manufacturing employment relative to the change in the coefficient on skilled manufacturing employment indicates that selection on the basis of the unobserved determinants of unskilled wages
occurs most predominantly in the unskilled manufacturing sector. Similar results
are found when using other measures of skilled employment.
Table 7 reports estimates of equation (2.8) in which the dependent variable is
log skilled wages. In the FE specification, skilled manufacturing employment has
a positive and statistically significant impact on skilled wages - a 10% increase in
skilled employment raises skilled wages by 1.2% - while an increase in unskilled
employment has no statistically significant effect. In the FE-IV specification, a 10%
45
increase in skilled employment rises to 2%, while that of unskilled employment remains unchanged. The estimated coefficients confirm the predictions of the model.
The direction of m o v e m e n t of coefficients between the FE and FE-IV specification
indicates that purely unskilled m a n u f a c t u r i n g employment does not select into locations on the basis of the unobserved determinants of skilled wages, mirroring the
observation for skilled employment and unskilled wages. Agricultural productivity is positively correlated with skilled wages, although the estimated coefficient is
not statistically significant.
2.5.2.4. Interpretation of Results from Wage Regressions. Between 1983 and 1999,
agricultural potential increased by 54% or at an annualized rate of 2.51%. The
estimates suggest that this raised real unskilled wages by 21.6% and accounted for
just u n d e r half of the total w a g e growth of 52%. The impact on skilled wages w a s
m u c h smaller - it raised them by 7.6%, or accounted for 18% of the total skilled
w a g e growth of 42%.
Total manufacturing employment increased by 43% over the same period, although the proportion of individuals in manufacturing has increased by only 14%.
The majority of employment growth consists of unskilled employment, which
grew by 67% compared to an increase of 24% in unskilled employment. Manufacturing e m p l o y m e n t growth raised the w a g e of rural unskilled workers by 8%
between 1983 and 1999 and raised the wages of skilled workers by 5%, accounting for 15% and 14% of unskilled and skilled wage growth respectively. H a d the
increase in manufacturing employment consisted of only unskilled growth, the
wages of rural unskilled workers w o u l d have increased by 13%.
The estimates suggest that drivers of growth vary in their impact on skilled and
unskilled wages. Low-skill biased growth, such as agricultural technical change,
has a greater effect on unskilled wages than on skilled wages. Skill-biased growth,
such as skilled manufacturing job growth, has a greater effect on skilled wages.
46
The estimates suggest that growth in agricultural potential and unskilled m a n u facturing employment decreased the ratio of skilled to unskilled wages by 12% and
9% respectively. Skilled manufacturing growth increased the ratio by 4%. Therefore, sources of low-skilled biased growth raise the levels of wages while reducing
the return to education.
The relative impact of agriculture and manufacturing are opposite to those
found in the existing literature. Using household panel data between 1982 a n d
1999, Foster and Rosenzweig (2004) find that a 74% increase in agricultural productivity resulted in a 47.7% increase in village level agricultural wages. A 900%
increase in factory workers resulted in a 93% increase in wages. The point estimates in the t w o papers are of similar magnitudes; the primary explanation for
the differences estimated is that the trends in manufacturing employment seen
in the REDS data do not match those seen in any other nationally representative
data-sets.
2.5.3. Robustness Checks

In this section, I verify that the coefficients estimated in the w a g e specifications are
indeed driven by movements in labor between the agricultural and manufacturing sector and don't capture alternative explanations. The coefficients may capture
overlaps between the manufacturing and agricultural sector in other locally clearing input and o u t p u t markets, such as local markets for agricultural or manufacturing products (Adhvaryu et al, 2009) or groundwater inputs (Keskin, 2009). In
this section, I discuss the robustness of m y results to these alternative explanations.
2.5.3.1. Agricultural
Outputs.
If supply and d e m a n d for agricultural products
clear at a district level, a reduction in industrial production and employment m a y

additionally impact the marginal revenue product of labor in agriculture through
47
a change in the demand for agricultural products and agricultural prices in a district. Atkin (2008) argues that movement restrictions for agricultural products imply that a district can be considered to be the relevant market for certain agricultural products. Demand may be affected through two main channels. Firstly, there
may be a direct demand effect from the agro-processing industry, since all input
demand functions are likely to be simultaneously affected by trade reform. Secondly, demand for products may be affected by a household level income effect.
Therefore, we would expect the coefficient on industrial employment to be u p w a r d
biased if the agricultural market clears at a local level.
To verify the robustness of the results to shifts in agro-processing demand, I in9Q
elude measures of the size of the agro-processing sector.
I examine how the coef-
ficient on predicted employment responds to conditioning on the ratio of predicted

agro-processing employment to total predicted industrial employment. Column d
of table 8 presents specifications testing whether the coefficient on industrial employment in the wage regressions is driven by local agro-processing demand. The
estimated wage-employment elasticity increases from 0.296 to 0.304, a difference
that is not statistically significantly different from zero at a 10% level. The coefficient on the ratio of agro-processing employment is not statistically different from
zero. These results suggest that local agricultural price variation driven by changes
in local agro-processing demand doesn't explain the relationship between wages
and predicted industrial employment.
The robustness of the results to the inclusion of the proportion of employment
in the agro-processing industry suggests that local agricultural demand effects
may not be a channel through which trade reform is exerting an impact on agrarian wages. However since the proportion of agricultural products consumed by
nn
The agro-processing sector c o r r e s p o n d s to industries w i t h N a t i o n a l Industrial Classification
C o d e s (1987) b e t w e e n 200 and 225.
48
the agro-processing sector during this period remained small relative to the market for agricultural products, a finding of no statistically significant effect may just
reflect the relatively small market weight of this purchaser.
A second channel through which local agrarian demand effects could exert an
u p w a r d bias on the coefficient of interest is through an income effect. Shifts in
industrial employment will induce changes in equilibrium wages and household
nominal income. Real income effects will of course depend on the total household
price vector, which is a function of agricultural prices as well as the prices of manufacturing goods. Assume first that the prices of all products other than agricultural
products are determined at a national level and are held constant. An increase in
agricultural wages at a local level will increase local income, demand and prices
of agricultural products, ceteris parabis. This will bias the coefficient on industrial
employment upwards.
Even with the assumption that all other markets clear at a national level, the
prices of all goods purchased by households are likely to be directly affected by
trade reform. Trade reform, particularly of consumption goods, will affect household consumption bundles directly through the prices of finished manufacturing
goods. Even in the absence of direct trade reform to consumption goods, this effect
would be expected in a more attenuated form since changes in the price of intermediate inputs into production will feed through to final goods prices of consumer
products. In the presence of non-homogeneous preferences for agricultural products across locations (Atkin, 2008) or non-homothetic Engel Curves, trade reform
is likely to induce agricultural prices over time to vary across locations if agricultural product markets clear regionally. The direction of the bias on the coefficient
on industrial employment is therefore unclear.
A straightforward approach to address this concern is to examine whether the
excluded set of instruments are conditionally correlated with farm-harvest prices
49
at a district level. Table 9 reports the results f r o m a regression of district level n o m inal wheat, rice, sorghum, cotton and sugar prices on the interaction of average
and use-weighted average tariffs with district level endowments. A F-Test over
the excluded set of instruments for sorghum, sugar and wheat confirms that they
are jointly unable to explain a significant fraction of the variation in farm-harvest
prices. However, for wheat and cotton the F-test indicates that the instruments are
jointly statistically significant.
I therefore examine whether conditioning on farm-harvest prices in the second
stage regression is able to explain away the relationship between instrumented
manufacturing employment and wages. While it is unclear h o w to interpret the
coefficient on a vector of equilibrium prices regressed on an equilibrium price, the
essence of this test is to examine how m u c h of the correlation between industrial
employment and wages is attributable to changes in equilibrium prices (Altonji,
Elder and Taber, 2005). The results of this specification are presented in table 8.
The coefficient on predicted employment goes d o w n slightly suggesting that local
prices may explain a small part of the effect, but the coefficient remains statistically
significant at a 5% level indicating that variation in local agricultural prices are not
enough to explain a w a y the effect.
2.5.3.2. Other Overlapping Input or Output Markets. If manufacturing and agriculture compete for other locally clearing inputs, the excluded instruments will
be correlated with unobserved local prices that directly enter into labor d e m a n d .
For example, if manufacturing uses g r o u n d w a t e r (Keskin, 2008), the instrumental
OA
variables will be correlated with the equilibrium price of water.
In the absence
of data on g r o u n d w a t e r prices, I examine whether m y instruments are correlated
ON
The bias generated b y other c o m p e t i n g i n p u t s r u n s is likely to b e negative - intuitively, a policy shock that shifts out industrial o u t p u t a n d i n p u t d e m a n d s in m a n u f a c t u r i n g is likely to (but
d o e s n o t necessarily) increase the equilibrium price of all locally clearing inputs. A decrease in
the a m o u n t of l a n d / g r o u n d w a t e r e m p l o y e d in agriculture will pull d o w n the m a r g i n a l p r o d u c t of
labor, while a decrease in labor e m p l o y e d will increase the m a r g i n a l p r o d u c t of labor.
50
with g r o u n d w a t e r usage. G r o u n d w a t e r use is measured as the proportion of villages within a district whose water table is below 5 different threshold depths.
An increase in the proportion of villages whose water table lies below 10 meters
captures whether g r o u n d w a t e r is extracted faster than it is being replenished. I
include region time trends to capture variations with districts in g r o u n d w a t e r usage over time in areas with c o m m o n rock types. 3 1 The results are presented in
columns (a) through (e) of table 9, panel B. A F-Test over the excluded set of instruments confirms that they are jointly unable to explain a significant fraction of the
variation in g r o u n d w a t e r depletion. 3 2 If industries vary in their intensity of water
usage, an alternative test is whether the ratio of employment in water intensive
industries is able to explain away the effect of predicted employment on wages. I
follow Keskin (2008) in separating industries into water and non-water intensive
industries. 3 3 The results are robust to this specification, presented in column (e) of
table 10.
2.5.3.3. Alternative Channels. The estimates may also be biased d u e to non-market
channels, for example through negative externalities such as air and water pollution. If the volume of the pollutant is positively correlated with manufacturing
output, and agricultural production decreases in the pollutant, this channel will
bias the coefficient d o w n w a r d s . It is harder to think of examples of positive externalities of industries which could be drive the results seen.
Trends in g r o u n d w a t e r depletion are likely to vary over time according to the rock t y p e in the
region (Jessoe, 2009), w h i c h is strongly correlated w i t h mineral composition. This w o u l d n o r m a l l y
be c a p t u r e d b y the lower order interaction terms.
32
T h r e e r o u n d s of data are required to estimate w h e t h e r the excluded variables are correlated w i t h

local g r o u n d w a t e r prices or m e a s u r e s of local g r o u n d w a t e r depletion. The rationale b e h i n d this
test is that in areas in w h i c h t r a d e r e f o r m is most likely to decrease d e m a n d for w a t e r f r o m the
m a n u f a c t u r i n g sector, g r o u n d w a t e r depletion should b e lower (Keskin, 2009). Since g r o u n d w a t e r
m e a s u r e s are h a r d to c o m e by, I use t w o w a v e s of the Minor Irrigation C e n s u s of India (1993 a n d
2000).
oo
Cotton, wool, silk, jute textile p r o d u c t i o n a n d p a p e r and p a p e r p r o d u c t i o n industries are defined

as water-intensive.
51
A series of specification checks are conducted to ensure the results aren't driven
by the level of aggregation used. NSSO regions are m a d e u p of 4 to 7 districts similar in their agro-climatic and geographic conditions, there is also a high degree of
correlation in their mineral and natural resource endowments. Therefore regional
growth in manufacturing employment and its skill composition are likely to be
highly correlated with their district level counterparts. In column (b) of table 6, the
log of industry-region level employment is instrumented using industry-region
level measures. Predicted industry-region level employment is then aggregated
to a NSSO region level. Column (d) uses the district level specification, in which
the log of district level manufacturing employment is instrumented using the interaction of local resource endowments with factor use weighted tariff changes as
instruments. The two methods give results of similar magnitudes. I do a Hausman
test to verify whether the results are within sampling error of each other, where the
null-hypothesis is that the district-industry exogeneity assumption is correct and
that this approach is more efficient. Under the null hypothesis the test statistics
is distributed as a chi-squared statistics with 20 degrees of freedom; the critical
value at a 5% level is therefore 31.4. The Hausman test statistic is 14.09, therefore
I do not reject the null hypothesis. I use district level industry-employment data
from the Economic Census. The magnitude and statistical significance of the estimated coefficient remains robust to the use of district rather than region level data 10% increase in manufacturing employment is predicted to increase male agrarian
wages by 1.4%.
Alternative wage measures are examined in column (g) of tables 6 and 7. The
unskilled wage is measured using the log of the median wage of illiterate workers
in the district. By including all agricultural and non-agricultural occupations, this
measure provides an indication that both non-agricultural and agricultural wages
for unskilled workers respond in similar ways to changes in the structure of production. An alternative approach is to look at the ratio of non-agricultural and
52
agricultural wages of unskilled workers, to examine whether they co-move. The

coefficient on the shocks is statistically insignificantly different from zero, confirming this hypothesis.
In table 9 I examine whether the instrumental variables are jointly correlated
with existing district level characteristics. I find that they are not, which provides
OA
empirical s u p p o r t for the identification strategy.
2.5.4. Income Regressions

The results from the previous section indicate the t w o sources of growth induce
positive responses in skilled and unskilled wages, and that they alter the return
to skilled and unskilled labor. This section examines the response of the level and
distribution of incomes to the two drivers of growth.
2.5.4.1. Empirical Specification. The model suggests that household incomes responses to aggregate economic change come through two components: (a) labor
market earnings and (b) cultivation profits. Agricultural productivity is predicted
to raise both components, while manufacturing growth will raise labor earnings
and reduce cultivation profits. Therefore the two drivers of growth are predicted
to induce heterogeneous income effects according to the household's initial education and land endowments. The predictions are tested using the following specification, derived in section B.2:
-frMnmt
-f^Mami
.
t t t t
c
10 + Jl^totaidt + Unskilled,dt + 730d/ + 7 4 ^ d ( + 7 5 HHdt + }ld + St + 6hdt
n
yhdt =
where HHdt captures household factors, such as the household's land e n d o w m e n t

and skill e n d o w m e n t . District d u m m i e s absorb location invariant factors and year
d u m m i e s capture all common year effects. If household's also derive income from
OA
This d o e s not h o w e v e r p r e c l u d e the possibility that m a n u f a c t u r i n g e m p l o y m e n t responses to

changes in industrial policy m a y b e h e t e r o g e n e o u s across regions, w h e r e the h e t e r o g e n e o u s res p o n s e s are correlated w i t h r a w material e n d o w m e n t s a n d the initial e d u c a t i o n a n d w a g e characteristics. This possibility should h o w e v e r largely b e s w e p t a w a y in the region time trends specification.
53
alternative activities, for example profits from self-employment in the service sector or manufacturing, the estimates will be biased if changes in agricultural productivity or manufacturing employment is correlated with income from these alternative activities. I test the robustness of m y results to these alternative channels
in appendix 2.5.7.
Households are separated into groups according to their land-holdings and the
education of the household head. As such, I divide households according to the
probability that they are working in the skilled and unskilled labor market, as well
as whether they are earning cultivation profits. The coefficients on agricultural
productivity and predicted manufacturing employment capture average income
responses across households within an e n d o w m e n t category.
2.5.4.2. Data. In the absence of income data, I use consumption as a proxy. This
is likely to be a better proxy for households at the bottom than at the top tail of the
income distribution. Consumption per capita, district level poverty and inequality
are measured using the NSSO consumption modules. 3 5 I follow Deaton (2003a,
2003b) in correcting the estimates for changes in the recall period used and use
an alternative set of poverty lines to those p u t forward by the Indian Planning
Commission. Poverty is measured as the head-count ratio and poverty g a p and
O/l
inequality is measured as the gini coefficient.

The model predicts that the income response of households to changes in m a n ufacturing labor d e m a n d will vary according to whether they are net exporters or
net importers of land. Since the NSSO survey has no information on inputs into
cultivation, I use the 1982 w a v e of the REDS data to identify the land cut-off at
which labor is imported. The REDS data indicate that 75% of households w h o
o w n less than 2 acres of land don't hired labor, while 90% of household w h o o w n
more than 2 acres of land do. Landed households are separated into net exporters
oc
36
I t h a n k Petia Topalova for s h a r i n g her district level p o v e r t y data with me.

District level identifiers w e r e not entered into the 1983 data therefore these years are n o t u s e d .
54
and importers of labor, where the cut-off is set at 2 acres. Landless households are
defined as those w h o own less than 0.1 acres of land.
2.5.5. Income Results

2.5.5.1. Consumption.
The results are presented in tables 11, 12 and 13. A n in-
crease in agricultural productivity is f o u n d to raise the consumption of all households, as predicted by the model. A 10% increase in agricultural productivity raises
the consumption of large landed households by approximately 1.5%, while it raises
the consumption of all other households by approximately 2%.
Manufacturing growth has a heterogeneous effect on household consumption
across land and skill groups. A 10% increase in unskilled manufacturing employment raises consumption in illiterate and literate households by 3.6% and 2.5%
respectively. The results from section 2.5.2.3 indicate that unskilled manufacturing
employment has n o effect on skilled wages. The consumption response of literate landless households to changes in unskilled manufacturing employment indicates that these households earn labor income in both the skilled and unskilled
labor market. In contrast, skilled manufacturing employment only has a positive
consumption effect on literate landless households - skilled growth raises their
consumption by 4.1%. These results indicate that illiterate households don't earn
income from skilled activities, while literate landless households earn income in
both the skilled and unskilled labor markets.
The model predicts that unskilled manufacturing growth unambiguously lowers the incomes of net importers of labor. The results corroborate this prediction
- in contrast to other household groups, they see no positive consumption effect
from manufacturing growth. By contract small landed households see a positive
consumption response to unskilled manufacturing growth. As predicted, the estimated effect is smaller than that of landless households w h o don't experience
reductions in cultivation income.
55
2.5.5.2. Poverty. Households declaring manual, unskilled labor as their prim a r y source of income constitute a substantial fraction of the rural poor (Eswaran
et al, 2009). Therefore determinants of unskilled w a g e growth are also expected to
decrease poverty. C o l u m n s (a) a n d (b) of table 14 present results from a FE specification, columns (c) t h r o u g h (f) present results f r o m the FE-IV specification. In
panel A the d e p e n d e n t variable is the headcount rate; in panel B it is the poverty
gap. Since the results are similar for the two measures, I only discuss those for the
headcount rate.
A 10% increase in total instrumented manufacturing employment decreases the
headcount rate by 0.08 points, or by 2.2% from the 1987 m e a n (column c). The skill
composition of manufacturing employment alters the effect: unskilled employment reduces the proportion of the population living in poverty by 0.13 points, or
by 3.3%, whereas an increase in skilled employment has no impact. Agricultural
productivity has a slightly larger impact than unskilled employment, mirroring
the unskilled wage results - poverty is reduced by 0.14 points or by 3.8%.
2.5.5.3. Inequality. In panel C of table 15,1 examine the impact of a change in
manufacturing employment on the gini coefficient of inequality. The model predicts that agricultural productivity growth will raise inequality, while a shift out in
unskilled manufacturing labor d e m a n d will decrease it. The estimated coefficients
support the predictions of the model: a 10% increase in agricultural productivity
raises the gini by 0.014 points, or by 5.4%. The effect of manufacturing growth
varies by skill: a 10% increase in unskilled employment reduces the gini by 0.013
points or 4.4%, while skilled employment reduces inequality by a m u c h smaller
margin - 0.003 points or 1%. The results therefore s u p p o r t the prediction that agricultural productivity increases inequality.
56
2.5.6. Interpretation of Income Results

I find that households in different asset groups display heterogeneous income responses to the different sources of growth and that the drivers of growth differ in
their impact on the income distribution. Data from the NSS indicate that average
real household consumption per capita increased by 22% between 1987 and 1999.
The head-count rate declined by 13 percentage points from 36% to 23% between
1987 and 1999 while the gini coefficient of inequality has increased by 0.03 over the
period.
Agricultural productivity growth accounts for a larger fraction of consumption
growth, poverty reduction and rising inequality than manufacturing job growth.
It has raised the gini coefficient of inequality by 0.045, explaining 150% of rural
inequality and has reduced poverty by 7 percentage points, accounting for over
53% of the total decline.
Unskilled manufacturing growth d a m p e n e d the rise in inequality by 0.01. Unskilled manufacturing growth reduced poverty by 3.6 percentage points, accounting for approximately 27% of the decline. On the other h a n d skilled manufacturing
growth has had no impact on poverty reduction. If all the growth in manufacturing employment over the period had been unskilled, poverty w o u l d instead have
been reduced by 5.5 percentage points. The skill-biased employment growth witnessed in India has implied that there has been relatively little effect of changes
in the size of this sector on the wages of unskilled individuals (Kochhar, 2006).
This is an observation that has been repeatedly asserted in the policy literature. To
my knowledge, the estimates in this p a p e r are the first to empirically validate it.
The estimates suggest that had growth in the manufacturing sector been purely
unskilled, India w o u l d have experienced a 9% greater increase in unskilled wages
and a 13% greater reduction in poverty.
57
2.5.7. Robustness Checks

If households earn profits from manufacturing, household incomes will be affected
by industrial policy through the labor market as well as self-employment channel,
violating the exclusion restrictions. A n additional concern which alters the interpretation of the results is that the drivers of growth m a y capture income responses
of households working in the service sector. To ascertain whether these alternative mechanisms are driving the results, I separate households according to the
occupation of the household head. Households are separated into two groups:
those w h o work on the wage labor market and those w h o conduct their o w n nonfarm business activities. To examine w h e t h e r the results are driven by households
in the non-farm sector, I remove all households working in the service sector or
w h o are self-employed in the manufacturing sector. The m a g n i t u d e s and statistical significance of the coefficient estimates remain broadly unchanged by this
exclusion, as shown in panel A of Table 16. In panel B of Table 16, I additionally
verify whether agricultural productivity or the instrumental variables are correlated with employment in the service sector in rural areas. I separate services into
construction, transport and retail (including shops, restaurants and hotels). I find
no statistically significant relationship between service sector employment and m y
drivers of growth, indicating that the results are unlikely to be driven by this channel.
2.5.8. Verifying the Mechanisms: Labor Market Time

. To show that the mechanisms through which shifts in manufacturing labor dem a n d affect wages correspond to m y model, I additionally examine the time devoted by households to agricultural production, by land ownership status.
58
The model predicts that landless households will be most responsive to shifts in
low-education employment, while large landed households will be most responsive to shifts in high-education employment. Panel A of Table 17 separates households by land holdings, a household is defined as landless if it owns less than
0.25 acres of land. Columns (a) and (b) report results using the FE specification,
columns (c) through to (h) report results for the FE-IV specification. The results
suggest that total days worked in agriculture decrease by 1% in response to a 10%
increase in total employment, decomposing the shift in high and low-education
employment we see that the response is much larger in response to a rise in loweducation employment - a 10% increase in low-education employment reduces the
number of days worked in agriculture by 2% compared to 0.5% for the corresponding increase in high-education employment. Separating agricultural time devoted
by landed and landless, we can see that the response comes from landless households. The time devoted to agrarian activities decreases by 7.3% in response to
a shift in low-education employment, and by 2.8% in response to a shift in high
education employment.
The model predicts that small landed households are more likely to respond
to shifts in low-education employment while large landed households are more
likely to respond to shifts in high-education employment.
The coefficients on
low-education manufacturing employment move from negative to positive as we

move u p the land distribution, although they are statistically insignificant for small
landed households. The data suggest an alternative explanation: the number of
individuals reporting conducting both manufacturing and agricultural activities
increases in these land groups. The largest landed group are responding to manufacturing growth by increasing the number of working days spent in agriculture.
59
As wages increase and the profits from cultivation fall, large landed households
appear to be substituting out of leisure and into work. 3 7
2.6.
CONCLUSION
This paper provides evidence that the level and distribution of incomes in rural
areas responds to changes in the demand for skilled and unskilled labor, driven
by changes in agricultural productivity and growth in manufacturing employment driven by changes in industrial policy. Households with different land and
skill endowments have been found to exhibit heterogeneous income effects to
the two drivers of growth in a manner which in consistent with the theoretical
model. Growth in unskilled labor demand, such as unskilled manufacturing employment has raised the incomes of unskilled households more than skilled households, and has also had a greater impact on the unskilled wage than on the skilled
wage. Growth in skilled labor demand, such as skilled manufacturing employment growth, has raised the incomes of skilled households more than those of
unskilled households, and has raised the skilled wage more than the unskilled
wages.
Growth in unskilled manufacturing employment reduces inequality in rural
India, while agricultural productivity growth has contributed to rising inequality.
Furthermore, agricultural productivity growth explains approximately half of the
reduction in poverty and rise in unskilled wages seen over the period. Manufacturing growth has played a small role in increasing the levels of incomes in rural
India and has only accounted for 20% of the reduction in poverty. This small impact is partly attributable to growth in skilled manufacturing employment over
the period. The estimates suggest that, had the manufacturing sector been purely
unskilled, India would have experienced a 13% greater decline in poverty. My
37
T h e m o d e l doesn't incorporate the leisure-work trade-off. Incorporating it w o u l d generate the

prediction that for h o u s e h o l d s w h o rely heavily on cultivation income, leisure w o u l d decrease as
m a n u f a c t u r i n g e m p l o y m e n t rises.
60
results suggest that, in the Indian context, skipping the second rung in the development ladder by promoting a skill biased rural non-farm sector is unlikely to
reduce poverty.
2. ECONOMIC GROWTH, WAGES AND INCOMES 72
Table l a : Primary O c c u p a t i o n s f o r W o r k i n g Individuals of W o r k i n g Age, 2 0 1

P a n e l A: M a l e s
1983
Occupation
Agricultural Sector
19
Percent in Proportion Proportion

Population Illiterate
Primary +
74.72
0.57
0.29
30.24
0.69
0.17
0.02
0.35
0.47
30.22
0.70
0.16
44.48
0.49
26.91
0.55
17.57
Land
Owned
facres)
4.99
Percent in Proportion
69.8
0.44
1.15
33.04
0.58
3.62
0.02
0.49
1.17
33.02
0.56
0.37
7.61
36.76
0.33
0.30
6.34
21.03
0.40
0.41
0.47
9.54
15.73
0.23
20.81
0.38
0.46
1.45
27.52
0.27
of which...
Hired Labor Market
Farm
manager
Manual
Laborer
Household Farm
Head
Other family
member
Non-Agricultural (excl gov)

of which...
Manufacturing
Worker
Other
TOTAL - RURAL (excl gov)
7.07
0.37
0.45
1.05
8.00
0.27
13.73
0.40
0.46
1.67
19.52
0.28
95.53
0.5
0.36
4.3
97.32
0.37
P a n e l B: F e m a l e s
1983
Occupation
Agricultural Sector
of which...
Hired Labor Market
Farm
Manual
manager
Laborer
Household I : arm
Head
Other family
member
Non-Agricultural (excl gov)
19
Percent in Proportion Proportion

Primary +
Land
Owned
facres)
4.27
86.08
0.90
0.06
41.21
0.93
0.04
0.04
0.81
0.19
99.96
0.93
0.04
1.20
44.87
0.87
0.08
8.92
0.88
0.07
91.08
0.81
6.26
Percent in Proportion
83.75
0.79
1.18
47.37
0.83
0.51
0.1
0.84
99.9
0.76
7.10
36.38
0.74
7.45
92.06
0.74
0.12
3.54
7.94
0.75
0.82
0.11
0.96
13.64
0.65
6.14
0.77
0.14
0.76
7.25
0.61
0.12
0.86
0.08
1.16
6.39
0.70
92.34
0.81
0.13
4.31
97.39
0.66
of which...
Manufacturing
Worker
Other
TOTAL - RURAL (excl gov)
TABLE l b : L i t e r a c y i n t h e R u r a l L a b o r F o r c e b y I n d u s t r y a n d O c c u p a t i o n
D e p e n d e n t V a r i a b l e : T a k e s a v a l u e o f 1 if a n i n d i v i d u a l is l i t e r a t e i n 1 9 8 7
Manufacturing
(a)
0.122*"
(b)
0.120***
(0.021)
(0.021)
(c)
0.185***
(d)
0.017
(e)
0.017
(0.021)
(0.026)
(0.026)
(0.027)
-0.188***
-0.188***
Unskilled Manufacturing
(0.011)
(0.011)
Own-Cultivation
0.101"*
(0.013)
Head*Own Cultivation
0.093***
ff)
0.061**
0.093***
-0.136***
-0.136***
-0.136***
(0.018)
(0.018)
(0.018)
(0.013)
(0.013)
0.024***
0.024***
0.011
0.011
(0.008)
(0.008)
(0.011)
(0.011)
-0.230***
-0.226***
-0.253***
-0.253***
(0.014)
(0.014)
(0.014)
(0.018)
(0.018)
(0.018)
Observations
91696
91696
91696
33076
33076
33076
R-squared
0.219
0.222
0.224
0.276
0.276
0.277
Agriculutral Laborer
-0.226*** -0.254***
AH columns include district identifiers and age polynomials.
TABLE l c : W a g e s b y I n d u s t r y a n d O c c u p a t i o n
D e p e n d e n t V a r i a b l e : L o g O n d i v i d u a l ' s W a g e in 1 9 8 7 )
Literate
(a)
0.189***
-0.007
Manufacturing
Agricultural Sector
(b)
0.171***
0.208***
fc)
(d)
0.167***
(e)
0.188***
(0.007)
(0
0.185***
(h)
(S)
0.245***
0.154***
(0.012)
-0.007
-0.008
-0.007
(0.007)
(0.016)
0.147**
0.112*
0.048
0.189***
0.245***
0.067
(0.062)
(0.065)
(0.062)
-0.049
(0.072)
-0.068
-0.042***
0.01
-0.056*"
0.031
-0.008
(0.010)
(0.008)
(0.010)
-0.267***
0.034
Literate*Manu
Literate*Ag
(0.026)
(0.041)
-0.139***
-0.189***
-0.025
(0.014)
Skilled*Manu
Unskilled*Ag
0.225***
0.106***
(0.023)
(0.039)
-0.042***
0.006
(0.008)
(0.010)
0.325**
Skilled*Ag
(0.134)
Observations
23598
23598
23598
23598
28273
28273
28273
33076
R-squared
0.513
0.516
0.518
0.518
0.259
0.259
0.26
0.276
All columns includc district identifiers and age polynomials.

Table 2a: Descriptive Statistics on Individuals Working in M a n u f a c t u r i n g in 1987, by Occupation Type
Occupation Bracket
Proportion of
Females
Males
Literate
Males
Females
At Least Primary
Males
Females
White Collar
0.17
0.13
0.78
0.29
0.69
0.19
Blue-Collar, Skilled
0.48
0.32
0.72
0.33
0.56
0.26
Blue-Collar, Manual
0.35
0.54
0.49
0.23
0.35
0.15
0.52
0.20
0.41
0.14
Average In NSS
Table 2b: Tasks Done by Occupational Bracket in the Manufacturing Sector in 1 9 8 7

Non-Routine
Analytic Tasks
Non-Routine
Routine Physical
Physical Tasks
Tasks
Non-Routine
Interactive
Tasks
Routine
Cognitive Tasks
Panel A: Males
W h i t e Collar
Blue-Collar, Manual
Average in DOT ( 1 9 7 7 )
4.19
0.56
3.35
3.65
3.32
(1.69)
(0.37)
(0.76)
(3.05)
(3.08)
2.68
1.64
4.58
0.54
8.49
(1.99)
(1.59)
(1.22)
(1.43)
(2.01)
2.10
1.63
4.23
0.39
6.81
(1.26)
(1.26)
(0.99)
(0.39)
(2.56)
3.84
1.45
4.02
2.01
4.94
Panel B: Females
W h i t e Collar
Blue-Collar, Manual
Average in DOT ( 1 9 7 7 )
4.50
0.16
3.31
3.75
1.56
(1.08)
(0.21)
(0.49)
(3.45)
(1.68)
1.48
0.82
4.77
0.04
8.49
(1.86)
(0.75)
(1.21)
(0.22)
(1.83)
0.88
0.89
3.92
0.07
4.85
(1.05)
(0.90)
(0.89)
(0.25)
(3.81)
3.63
1.31
4.03
1.86
5.04
A crosswalk was created between Indian Occupational Codes (NCO-68) and the US Census Occupation Codes (1960). TheDictionay ol Occupational Titles dataset
was assembled by Autor, Levy, and Murnane (2003): they collected data on job task requirements from the US Department of Labor's Dictionary of Occupational
Titles (DOT) and merged them with census occupation classifications. The classification of jobs into White Collar, Blue-Skilled and Blue Manual was conducted
using the NCO-68. The average characteristics in the DOT is an unweighted mean across all occupations. Definitions: Non-Routine Analytical Tasks -General
Educational Development, Maths: Non-Routine Physical Tasks - Bye-Hand Coordination; Routine Physical Tasks - Finger Dexterity; Non-Routine Interactive
Tasks - Direction, Control and Planning; Routine Cognitive Tasks - Set Limits, Standards
TABLE 3: Decomposition of Variance in the Proportion of Educated Workers in a

Dependent Variable: Proportion of Region-Industry Workforce by education
Literate
< Primary
< Middle
< Secondary
Panel A: 1 9 8 7
[bj
tj
Region
10.11
10.87
12.02
14.33
[d]
Industry
17.32
21.11
24.84
26.36
R-Squared
0.27
0.31
0.34
0.36
(d)
Panel B: 1 9 9 3
(a)
(b)
(c)
Region
5.09
5.8
7.41
9.2
Industry
17.52
20.26
23.17
24.99
R-Squared
0.26
0.29
0.32
0.34
8372
Obs
8372
8372
8372
N Regions
52
52
52
52
N Industries
161
161
161
161

TABLE 4 : E m p l o y m e n t in 1 9 8 3
D e p e n d e n t Varible; Log(Region-lndustry E m p l o y m e n t )
At Least
Total
M e a s u r e of Factor
Use
Literate
Primary
Total
Quantile
Dummy
(a)
(b)
(c)
(d)
Groundwater
0.050**"
0.049""*
0.047**"
0.015***
"Agro P r o p o r t i o n
(0.0150)
(0.0140)
(0.0110)
(0.0050)
Uncultivable Land
-0.335
-0.335
-0.248
-0.093
*Agro P r o p o r t i o n
(0.2690)
(0.2450)
(0.1970)
(0.0920)
Forest Cover
0.102"""
0.060*"
0.034
0.096***
Wood Proportion
(0.0280)
(0.0260)
(0.0270)
(0.0120)
Ceramics Minerals
0.016***
0.004
0.007
0.009***
"Ceramics Proportion
(0.0050)
(0.0060)
(0.0050)
(0.0020)
Construction Minerals
0.027***
0.021***
0.016***
0.002
"Construction Proportion
(0.0040)
(0.0030)
(0.0030)
(0.0020)
Chemicals Minerals
-0.030**"
-0.021***
-0.010*
-0.009***
"Chemicals Proportion
(0.0060)
(0.0060)
(0.0050)
(0.0020)
Metals
-0.051**"
-0.040""*
-0.032***
-0.010*"*
"Metal Proportion
(0.0050)
(0.0050)
(0.0040)
(0.0020)
-0.001
-0.021*"
-0.020*"*
0.003
(0.0090)
(0.0100)
(0.0070)
(0.0040)
Coal
"Energy Proportion
Electricity
"Energy Proportion
Constant
-0.315"
-0.115
-0.082
-0.146**
(0.1660)
(0.1430)
(0.1280)
(0.0660)
1.978***
1.559"*"
1.294**"
1.110***
(0.0710)
(0.0660)
(0.0590)
(0.1460)
5856
5856
5856
5856
Industries
96
96
96
96
Regions
61
61
61
61
Observations
" s i g n i f i c a n t a t 1 0 % ; ** s i g n i f i c a n t a t 5 % ; *** s i g n i f i c a n t a t 1 %
All specifications i n c l u d e NSS region fixed effects, s t a n d a r d e r r o r s are c l u s t e r e d at the r e g i o n level.
Columns 1-3 m e a s u r e the 0 / 1 use of a factor in an i n d u s t r y , w h e r e an i n d u s t r y is classified as u s i n g
the i n p u t if the m e a s u r e d factor i n t e n s i t y lies above t h e m e d i a n f o r all i n d u s t r i e s . In C o l u m n 4, the
i n t e n s i t y w i t h w h i c h an i n d u s t r y uses a g i v e n factor is m e a s u r e d u s i n g quantiles.

TABLE 5: First Stage Specification
D e p e n d e n t Varible: L o g ( R e g i o n - l n d u s t r y Total E m p l o y m e n t )
Panel A: No Policy ( c o l u m n a) and D e r e g u l a t i o n ( b - e )
Total
Literate
Primaryt
Dummy
Policy
Policy*Forest Cover
*Wood
Proportion
Policy'Ceramics
"Ceramics
Minerals
Proportion
Policy'Construction
"Construction
Proportion
Policy*Chemicals
"Chemicals
Proportion
Policy*Metals
"Metal
Proportion
Policy*Coal
"Energy
Proportion
Policy*Electricity
"Energy
Proportion
Constant
Observations
F-Statistic - Triple
Interactions
Total
Quantile
(d)
0.271
(a)
-0.384
(0.318)
0.487**"
(0.042)
0.026***
(0.009)
0.030***
(0.004)
-0.042***
(0.006)
-0.016***
(0.005)
-0.059***
(0.016)
-0.214**"
(0.024)
13.247
(15.636)
23450
34.49
(b)
-0.954***
(0.284)
0.384***
(0.039)
0.018*
(0.010)
0.031***
(0.004)
-0.037***
(0.005)
-0.009*
(0.005)
-0.047***
(0.015)
-0.204***
(0.024)
12.723
(13.916)
23450
32.35
(c)
-1.177***
(0.303)
0.338***
(0.037)
0.017*
(0.009)
0.030*"*
(0.004)
-0.033***
(0.005)
-0.009*
(0.005)
-0.033***
(0.012)
-0.175***
(0.021)
14.582
(14.287)
23450
21.98
Total
Dummy
Literate
Primary-)-
Total
Quantile
(g)
0.710**
(0.297)
0.288***
(0.044)
0.030***
(0.009)
(h)
1.417**
(0.636)
0.115*"*
(0.325)
0.103***
(0.013)
-0.001
(0.002)
0.008**"
(0.002)
-0.008**"
(0.002)
-0.003*
(0.002)
-0.016***
(0.005)
-0.054***
(0.009)
11.976
(15.795)
23450
22.93
Panel B: Policy - Import Tariffs
(e)
1.039**
(0.420)
0.336***
(0.044)
0.054***
(0.008)
0.016***
(0.004)
-0.007
(0.005)
0.017***
(0.006)
0.060***
(0.012)
-0.331***
(0.062)
-0.881
(16.383)
22539
23.76
(0
0.848**
(0.333)
Policy*Forest Cover
0.296***
"Wood Proportion
(0.043)
0.034*"*
Policy*Ceramics Minerals
(0.009)
Policy*Construction
0.013***
(0.005)
Policy*Chemicals
-0.005
"Chemicals Proportion
(0.005)
Policy'Metals
0.019***
"Metal Proportion
(0.005)
0.040***
Policy*Coal
"Energy Proportion
(0.012)
Policy"Electricity
-0.271***
"Energy Proportion
(0.049)
Constant
-1.246
(15.217)
22539
Observations
F-Statistic - Triple Interactions
13.16
* significant at 10%; ** significant at 5%; * ** significant at 1 %
Policy
0.011**
(0.004)
-0.002
(0.005)
0.017***
(0.005)
0.033***
(0.009)
-0.249***
(0.056)
-1.008
(14.819)
22539
11.39
(0.020)
0.014"**
(0.002)
0.016***
(0.002)
-0.005*
(0.003)
-0.002
(0.002)
0.021***
(0.004)
-0.075***
(0.026)
-0.863
(16.108)
22539
28.52
All specifications include NSS region fixed effects, standard errors are clustered at the region level. Columns
1-3 measure the 0 / 1 use of a factor in an industry, where an industry is classified as using the input if the
measured factor intensity lies above the median for all industries. In Column 4, the intensity w i t h which an
industry uses a given factor is measured using quantiles.

TABLE 6: Unskilled Wage R e s p o n s e to InstrumentedManufacturing Employment and Agricultural Productivity
Panel A: D e p e n d e n t Variable: Log Unskilled Wages
Log(Manufacturing
Log(Agricultural
Employment)
Productivity)
Constant
Adjusted
R-Squared
FE
FE-IV
FE-IV
FE-IV
FE-IV
FE-IV
(a)
(b)
(c)
(d)
(e)
(0
FE-IV
(g)
0.025"
0.144*
0.175**
0.159*
0.146**
0.118*
0.132
(0.012)
(0.085)
(0.076)
(0.086)
(0.074)
(0.062)
(0.145)
0.414***
0.428**
0.443*
0.214
0.439**
0.456
0.365**
(0.187)
(0.206)
(0.239)
(0.246)
(0.214)
(0.244)
(0.167)
-9.993***
-11.678***
-10.238***
-8.799***
-2.451***
-8.369***
-7.346***
(2.773)
(3.056)
(3.032)
(3.694)
(0.879)
(3.107)
(2.566)
0.800
0.806
0.756
0.842
0.888
0.674
0.747
District Fixed Effects
Region Time
34.49***
13.087***
23.57***
45.48***
34.49***
34.49***
930
930
678
930
930
930
1034
Trend
First Stage F-Stat

Observations
Panel B: D e p e n d e n t Variable:Log Unskilled Wages, Instrumented Manufacturing Broken Apart by Skill
Log(Manufacturing
Employment)
Log(Literate
Manufacturing)
Log(At Least
Primary)
FE
FE-IV
FE-IV
FE-IV
FE-IV
FE-IV
(a)
(b)
to
(d)
(e)
CO
(g)
0.029**
0.296**
0.309***
0.288*
0.299**
0.217
0.318*
(0.013)
(0.135)
(0.138)
(0.129)
(0.120)
(0.142)
(0.165)
-0.022
-0.239*
-0.251**
-0.155
-0.379*
(0.029)
(0.142)
(0.118)
(0.145)
(0.206)
-0.259**
(0.151)
Log(Skilled
FE-IV
_
-0.226
Manufacturing)
(0.222)
Log(Agricultural
Productivity)
0.386*
0.258
0.322
0.321*
(0.215)
(0.216)
(0.221)
(0.250)
(0.173)
-10.947***
-10.667***
-10.433***
-7.319**
-6.553**
-6.900***
(3.157)
(3.188)
(3.217)
(3.253)
(3.462)
(3.179)
(2.668)
0.801
0.801
0.801
0.801
0.842
0.683
0.743
(0.188)
Constant
Adjusted
R-Squared
0354**
(0.206)
_9 g g g * * *
0.415*
0.371*
Region T i m e
32.35***
21.98***
28.52**
11.82**
32.35***
32.35***
930
930
930
930
678
930
1034
Trend
First Stage F-Stat

Observations
The d e p e n d e n t v a r i a b l e is log male a g r a r i a n wages, f r o m the A g r i c u l t u r a l Wages of India. In c o l u m n (g) o f panel A, the d e p e n d e n t variable is the
m e d i a n wage of illiterates in alt sectors o t h e r t h a n m a n u f a c t u r i n g , i m p u t e d f r o m the E m p l o y m e n t - U n e m p l o y m e n t Rounds collected by the
NSSO. All specifications include d i s t r i c t fixed effects ( a p a r t f r o m (Q in b o t h panels w h i c h includes r e g i o n fixed effects), year d u m m i e s , r a i n f a l l
variables (total rainfall b e t w e e n j u n e and September, r a i n f a l l squared and a shock measure t a k i n g the value 1 if there is a positive r a i n f a l l shock,
0 if no shock and - 1 for negative shocks), the log of male and female p o p u l a t i o n for the landless and by land quantiles. Columns (b) o n w a r d s
a d d i t i o n a l l y c o n t r o l for the i n t e r a c t i o n of d i s t r i c t level resources w i t h average t a r i f f and delicensing reforms. The first stage i n s t r u m e n t s are the
average i n t e r a c t i o n b e t w e e n b e t w e e n d i s t r i c t level resources, i n d u s t r y resource usage and i n d u s t r y t a r i f f s and regulations. The n u m b e r of
observations in c o l u m n (c) is l o w e r t h a n the rest because d i s t r i c t i d e n t i f i e r s are not available for the 1 9 8 3 NSS. C o l u m n (e) in panel A uses
d i s t r i c t - i n d u s t r y representative data f r o m the Economic Census for 1 9 9 1 and 1998. Column (f) in b o t h panels includes region fixed effects.
2. E C O N O M I C G R O W T H , WAGES A N D I N C O M E S 78
TABLE 7: Skilled Wage Response to Instrumented Manufacturing Employment and
Agricultural Productivity
Dependent Variable: Log Male Skilled Wages
Log(Manufacturing
Log(Literate
Employment)
Manufacturing)
Log(At Least Primary
Log(Skilled
R-Squared
FE-IV
(c)
(d)
(e)
-0.032
0.020
0.015
0.017
0.082*
(0.041)
FE-IV
(0.021)
(0.040)
(0.042)
(0.038)
0.118***
0.203**
0.186*
(0.041)
(0.101)
(0.112)
Manufacturing)
0.231**
(0.106)
0.259*
(0.152)
0.153
0.140
0.162
0.114
0.059
(0.222)
(0.228)
(0.208)
(0.209)
(0.205)
-6.044*
-4.598***
-5.3061
-5.0383
-8.612
(3.434)
(5.241)
(5.224)
(5.199)
(3.732)
0.523
0.582
0.553
0.553
0.559
Trend
F-Statistic
32.35***
21.98***
28.52**
11.82**
1034
1034
1034
1034
1034
District Fixed
Region Time
No of
FE-IV
(b)
Productivity)
Constant
First Stage
FE-IV
Manufacturing)
Log(Agricultural
Adjusted
FE
(a)
Effects
Observations
The dependent variable is log median male wages of literate w o r k e r s in the non-agricuftural sector, excluding manufacturing. This is
i m p u t e d f r o m the E m p l o y m e n t - U n e m p l o y m e n t Rounds collected by the NSSO. All specifications include d i s t r i c t fixed effects, year
dummies, r a i n f a l l variables {total rainfall between june and S e p t e m b e r , rainfall squared and a shock measure t a k i n g the value 1 if
there is a positive rainfall shock, 0 if no shock and -1 for negative shocks), the log of male and female p o p u l a t i o n for the landless and
by land quantiles. Columns (b) o n w a r d s additionally c o n t r o l for the i n t e r a c t i o n of d i s t r i c t level resources w i t h average t a r i f f and
delicensing reforms. The first stage instruments are the average interaction between between d i s t r i c t level resources, i n d u s t r y
resource usage and i n d u s t r y tariffs and regulations. Manufacturing e m p l o y m e n t data in column (a) comes f r o m the 1 9 8 7 , 1 9 9 3 and
1999 NSSO, columns (b) o n w a r d s additionally includes 1983.

T a b l e 8: R o b u s t n e s s Checks, W a g e s
Including Price of Agricultural

Products
Log(Manufacturing Employment)
Log(Literate Manufacturing)
(b)
0.303"
(0.134)
-0.259*
(0.145)
0.299**
(0.153)
(d)
0.256*
(0.152)
-0.236
(0.171)
Log(At Least P r i m a r y )
Proportion Water Intensive

Log(Agricultural Productivity)
Constant
Adjusted R-Squared
Region T i m e Trend
First Stage F-Stat
Observations
Water
(0
0.317**
(0.133)
-0.267*
(0.140)
_
-0.760
(0.926)
Proportion Agro Processing
Agro
(e)
0.304**
(0.133)
-0.267*
(0.141)
_
-0.218
(0.208)
Log(Skilled M a n u f a c t u r i n g )
Proportion of Linked
"
0.201
(1.200)
0.417**
0.529***
0.455**
0.362*
0.369*
(0.201)
(0.207)
(0.200)
(0.205)
(0.213)
-10.856*** -12.929*** -10.433*** -10.335*** -10.353**"
(3.253)
(3.077)
(3.038)
(2.961)
(3.005)
0.803
0.807
0.801
0.804
0.804
-
N
32.35***
930
N
21.98***
930
N
28.52**
930
N
32.35***
930
N
32.35***
930
T h e d e p e n d e n t v a r i a b l e is l o g m a l e a g r a r i a n w a g e s , f r o m t h e A g r i c u l t u r a l W a g e s of I n d i a . All s p e c i f i c a t i o n s i n c l u d e
district fixed effects, y e a r d u m m i e s , rainfall variables (total rainfall b e t w e e n j u n e a n d S e p t e m b e r , rainfall s q u a r e d
a n d a s h o c k m e a s u r e t a k i n g t h e v a l u e 1 if t h e r e is a p o s i t i v e r a i n f a l l s h o c k , 0 if n o s h o c k a n d - 1 f o r n e g a t i v e s h o c k s ) ,
t h e log of m a l e a n d f e m a l e p o p u l a t i o n f o r t h e l a n d l e s s a n d b y l a n d q u a n t i l e s . M a n u f a c t u r i n g e m p l o y m e n t d a t a c o m e s
f r o m t h e NSSO s u r v e y s c o n d u c t e d in 1 9 8 3 , 1 9 8 7 , 1 9 9 3 a n d 1 9 9 9 .

Panel A: Characteristic of Districts in 1 9 8 3
Agricultural
Wage (1983)
Skilled Wage
Headcount
Poverty Line
Gini
(a)
0.0244***
(b)
0.007
(c)
0.004
(d)
-0.003
(d)
-0.0003
"Wood P r o p o r t i o n
(0.008)
(0.007)
(0.015)
(0.033)
Deregulated*Ceramics
-0.011
-0.004
0.009
(0.008)
(0.001)
0.000
(0.001)
0.003**
(0.001)
Deregulated*F'orest Cover
(0.007)
(0.004)
Deregulated*Construction
0.0044***
0.001
0.003
0.001
(0.001)
0.001
"Construction P r o p o r t i o n
(0.002)
(0.002)
(0.003)
(0.001)
Deregulated"Coal
0.0038*
-0.0075"*
-0.0027**
"Energy P r o p o r t i o n
(0.002)
(0.004)
-0.001
(0.001)
"Ceramics P r o p o r t i o n
0.000
(0.001)
(0.000)
0.0152*
Deregulated*Electricity
-0.074
0.120
-0.008
-0.048
(0.138)
(0.076)
(0.023)
(0.055)
(0.008)
Constant
-3.2203
-0.3495
0.0571
-0.1239
0.5100***
(0.112)
(1.997)
(1.105)
(0.339)
(0.795)
Observations
310
348
350
350
324
F-Statistic over Triple I n t e r a c t i o n s
1.84
2.43*"
1.04
2.06"
0.26
Proportion
Male Literate
Proportion
Female
Literate
Proportion
Male >=
Primary
Proportion
Female >=
Primary
Population
(d)
0.009
(0.007)
Panel B: Characteristic of Districts in 1 9 8 3
Deregulated*Forest Cover
"Wood P r o p o r t i o n
(a)
0.002
(b)
(c)
0.001
-0.001
fdj
-0.001
(0.002)
(0.002)
(0.002)
(0.001)
0.000
-0.002
-0.001
0.000
-0.007
(0.001)
(0.001)
(0.001)
(0.001)
(0.007)
0.001
0.000
-0.001
0.000
-0.0066**"
(0.001)
(0.001)
(0.001)
(0.000)
(0.002)
-0.0033*
-0.001
0.001
0.000
0.002
(0.002)
(0.001)
(0.001)
(0.001)
(0.005)
Deregulated"Coal
*Energy P r o p o r t i o n
Constant
0.113
0.093
-0.046
-0.029
-0.708
(0.129)
(0.074)
(0.107)
(0.049)
(1.199)
2.3095
2.2336"*
-0.4381
-0.3509
3.2129
(1.887)
(1.082)
(1.560)
(0.717)
(17.580)
Observations
350
350
350
350
350
1.19
1.08
0.45
0.29
2.62**
* significant at 10%; ** significant at 5 % ; *** significant at 1%

All specifications include d i s t r i c t fixed effects, year dummies, rainfall variables (total rainfall between june and September, rainfall squared and a
shock measure taking the value 1 if there is a positive rainfall shock, 0 if no shock and -1 for negative shocks), the log of male and female
population for the landless and by land quantiles. In addition, they include the interaction of d i s t r i c t level resources w i t h average t a r i f f and
delicensing reforms.

P a n e l A: C h a r a c t e r i s t i c of Districts in 1 9 8 3
Agricultural
Wage (1983)
Skilled Wage
Headcount
Poverty Line
Gini
(a)
0.0244***
(b)
0.007
(c)
0.004
(d)
-0.003
(d)
-0.0003
*Wood P r o p o r t i o n
(0.008)
(0.007)
(0.015)
(0.033)
(0.001)
-0.011
-0.004
0.009
Ceramics Proportion
(0.007)
(0.004)
(0.008)
0.001
(0.001)
0.001
(0.001)
D e r e g u l a t e d * F o r e s t Cover
0.0044***
0.001
0.003
(0.002)
(0.002)
(0.003)
Deregulated*Coal
0.0038*
-0.0075**
-0.0027**
'Energy Proportion
(0.002)
(0.004)
-0.001
(0.001)
-0.074
0.120
-0.008
-0.048
(0.001)
0.000
(0.001)
0.003**
(0.001)
0.000
(0.000)
0.0152*
(0.138)
(0.076)
(0.023)
(0.055)
(0.008)
Constant
-3.2203
-0.3495
0.0571
-0.1239
0.5100***
(0.112)
(1.997)
(1.105)
(0.339)
(0.795)
Observations
310
348
350
350
324
F-Statistic o v e r Triple I n t e r a c t i o n s
1.84
2.43**
1.04
2.06*
0.26
Proportion
Male Literate
Proportion
Female
Literate
Proportion
Male >=
Primary
Proportion
Female >=
Primary
Population
(b)
(d)
-0.001
(d)
0.009
(0.007)
P a n e l B: Characteristic of Districts in 1 9 8 3
(a)
0.002
0.001
M
-0.001
(0.002)
(0.002)
(0.002)
(0.001)
0.000
-0.002
0.000
-0.007
'Ceramics Proportion
(0.001)
(0.001)
(0.001)
(0.007)
0.001
0.000
-0.001
(0.001)
-0.001
0.000
-0.0066***
'Construction Proportion
(0.001)
(0.001)
(0.001)
(0.000)
(0.002)
-0.0033*
-0.001
(0.002)
(0.001)
0.001
(0.001)
0.000
(0.001)
(0.005)
Deregulated'Coal
'Energy Proportion
Constant
0.002
0.113
0.093
-0.046
-0.029
-0.708
(0.129)
(0.074)
(0.107)
(0.049)
(1.199)
2.3095
2.2336**
-0.4381
-0.3509
3.2129
(1.887)
(1.082)
(1.560)
(0.717)
(17.580)
Observations
350
350
350
350
350
1.19
1.08
0.45
0.29
2.62**
* significant at 10%; ** significant at 5%; *** significant at 1%

All specifications include district fixed effects, year dummies, rainfall variables (total rainfall between june and September, rainfall squared and a
shock measure t a k i n g the value 1 if there is a positive rainfall shock, 0 if no shock and -1 for negative shocks), the log of male and female
population for the landless and by land quantiles. In a d d i t i o n , they include the interaction of d i s t r i c t level resources w i t h average t a r i f f and

TABLE 11: Log(Consumption per cap), Instrumented Manufacturing
E m p l o y m e n t and Agricultural Productivity
D e p e n d e n t Variable: Log Consumption Per Capita
Panel A: FE
By Land Status
Landless
Log(Manufacturing Emp)
Log(Agricultural Potential)
Constant
Landed
By Education
Illiterate
Literate
(a)
(b)
(d)
0.020**
0.014*
0.018**
0.014
(0.095)
(0.008)
(0.009)
(0.009)
0.022
0.019
0.012
0.028
(0.022)
(0.017)
(0.018)
(0.019)
0.195
0.124*
0.215**
0.159**
(0.158)
(0.067)
(0.101)
(0.063)
1.167
2.761
2.147
1.159
(2.791)
(1.264)
(1.368)
(1.126)
Adjusted R-Squared
0.377
0.401
0.269
0.238
Observations
64855
127860
94650
98065
Panel B: FE-IV
By Land Status
Landed
Illiterate
Literate
(e)
(0
0.302***
(0.104)
0.035
(0.086)
0.184***
(0.066)
0.484
(3.003)
0.381
64855
0.029
(g)
0.137***
(0.047)
-0.117
(0.095)
0.204**
(0.099)
2.932
(1.818)
0.297
94650
0.059
(0.053)
0.036
(0.042)
0.206***
(0.071)
-1.816
(1.365)
0.232
98065
Landless
Log(Manufacturing Emp)
Log(Agricultural Potential)
Constant
Adjusted R-Squared
Observations
By Education
(0.056)
0.056
(0.045)
0.165**
(0.082)
1.787
(1.615)
0.360
127860
(h)
All specifications include d i s t r i c t and year fixed effects, rainfall variables (total rainfall
between june and S e p t e m b e r , r a i n f a l l squared and a shock measure t a k i n g the value 1 if there
is a positive rainfall shock, 0 if no shock and -1 for negative shocks), the log of male and
female population by land quantiles. Column (e) o n w a r d s a d d i t i o n a l l y controls for the
interaction of district level resources w i t h average t a r i f f and delicensing reforms. Household
controls are the sex of the household head, his age, age squared and social group. The first
stage instruments are the i n t e r a c t i o n between between district level resources, i n d u s t r y
resource usage and i n d u s t r y tariffs and regulations.

TABLE 12: C o n s u m p t i o n , I n s t r u m e n t e d M a n u f a c t u r i n g E m p l o y m e n t a n d
L a n d l e s s T h r e s h o l d is < 0 . 2 5 a c r e s . Small L a n d e d <2, Large L a n d e d >2
D e p e n d e n t Variable: Log C o n s u m p t i o n P e r Capita; FE-IV
Landless
Landed
Illiterate
Literate
Small
Large
(a)
0.370**
(b)
0.284**
(0
0.127***
(d)
0.022
(0.164)
(0.126)
(0.038)
(0.098)
-0.288*
0.185*
-0.154**
0.065
(0.171)
(0.099)
(0.078)
(0.064)
0.202***
0.156*
0.165*
0.201**
(0.076)
(0.086)
(0.099)
(0.088)
-1.634
4.302
0.946
2.927*
(1.265)
(2.036)
(1.712)
(1.722)
A d j u s t e d R-Squared
0.318
0.327
0.338
0.365
Observations
36042
28813
52787
75073
LogfAgricultural Productivity)
Constant
Small L a n d e d
Log(Manufacturing E m p l o y m e n t )
Log(Literate E m p l o y m e n t )
Constant
Large Landed
Illiterate
Literate
Illiterate
Literate
(e)
0.149***
(0
0.097**
(g)
0.136
(h)
0.014
(0.043)
(0.042)
(0.102)
(0.091)
-0.132
-0.139*
-0.056
0.173*
(0.097)
(0.083)
(0.070)
(0.071)
0.132
0.171*
0.215**
0.197*
(0.111)
(0.091)
(0.101)
(0.108)
3.325*
-1.472
3.604
1.677
(1.839)
(1.662)
(2.361)
(1.804)
Adjusted R-Squared
0.322
0.336
0.378
0.304
Observations
27026
25761
31582
43491
All specifications include district and year fixed effects, rainfall variables (total rainfall between june
and S e p t e m b e r , rainfall squared and a shock measure t a k i n g the value 1 if there is a positive rainfall
shock, 0 if no shock and -1 for negative shocks), the log of male and female population by land
quantiles. In addition, they control for the interaction of district level resources w i t h average t a r i f f
and delicensing reforms. Household controls are the sex of the household head, his age, age squared
and social group. The first stage i n s t r u m e n t s are the interaction between between d i s t r i c t level
resources, i n d u s t r y resource
are b e t w e n 21.98 and 32.35.
usage and i n d u s t r y tariffs and regulations. The F-Stats on the first stage

TABLE 13: Consumption, I n s t r u m e n t e d Manufacturing E m p l o y m e n t and
Examining Sensitivity to Land T h r e s h o l d s
D e p e n d e n t Variable: Log Consumption Per Capita; FE-IV
Landless
<2.5
<3.5
acres
acres
<3 acres
< 4 acres
Constant
Adjusted R-Squared
Observations
W
0.131***
(0.038)
-0.159**
(0.077)
0.171*
(0.099)
1.489
(1.812)
0.319
64605
>2.5
acres
Log(Literate Employment)
Constant
Adjusted R-Squared
Observations
(e)
0.018
(0.095)
0.064
(0.069)
0.169*
(0.091)
3.344*
(1.738)
0.362
63812
(b)
0.126***
(0.036)
-0.134**
(0.074)
0.192*
(0.100)
1.454
(1.829)
0.318
73213
M
0.128***
(0.037)
-0.124*
(0.074)
0.215**
(0.103)
1.165
(1.887)
0.313
78709
Landed
>3.5
>3 acres
acres
M
0.026
(0.103)
0.073
(0.076)
0.212**
(0.093)
2.740
(1.808)
0.360
55204
(8)
0.062
(0.100)
0.036
(0.073)
0.176*
(0.095)
3.037*
(1.821)
0.360
49708
(d)
0.133***
(0.037)
-0.124*
(0.075)
0.238**
(0.104)
0.996
(1.899)
0.309
84354
> 4 acres
(h)
0.080
(0.095)
-0.021
(0.064)
0.168*
(0.100)
4.510**
(1.929)
0.361
44063
All specifications include district and year fixed effects, rainfall variables (total rainfall between june
and S e p t e m b e r , rainfall squared and a shock measure t a k i n g the value 1 if there is a positive rainfall
shock, 0 if no shock and -1 for negative shocks), the log of male and female p o p u l a t i o n by land
quantiles. In addition, they control for the interaction of d i s t r i c t level resources w i t h average t a r i f f
and delicensing reforms. Household controls are the sex of the household head, his age, age squared
and social group. The first stage i n s t r u m e n t s are the interaction between between d i s t r i c t level
resources, industry resource usage and i n d u s t r y tariffs and regulations. The F-Stats on the first stage
are betwen 21.98 and 32.35.

TABLE 14: Poverty e s t i m a t e s
Panel A: D e p e n d e n t Variable: Headcount Rate
FE
FE-TSLS
Rural
LogfAgricultural Productivity)
-0.136*
(0.082)
(b)
-0.002
(0.003)
0.003
(0.003)
-0.136*
(0.081)
Constant
2.360**
(1.259)
2.374***
(1.259)
(a)
-0.002
(0.008)
-
Urban
(c)
-0.088*
(0.045)
CO
(e)
0.011
(0.012)
(d)
-0.127**
0.059
0.131*
(0.071)
-0.153*
(0.080)
-0.139*
(0.082)
-0.092
(0.061)
-0.012
(0.021)
0.027
(0.023)
-0.087
(0.061)
2.737**
(1.252)
2.337*
(1.301)
1.353
(1.058)
1.203
(1.041)
Panel B: D e p e n d e n t Variable: Poverty Gap

FE-TSLS
FE
Rural
Log(Agricu)tural Productivity)
-0.042*
(0.023)
(b)
-0.004
(0.003)
0.001
(0.005)
-0.041*
(0.023)
Constant
0.629
(0.396)
967
0.608
(0.408)
967
(a)
-0.003
(0.002)
-
Observations
(c)
-0.0274
(0.021)
Urban
-0.052**
(0.024)
(d)
-0.069***
(0.020)
0.077***
(0.026)
-0.059**
0.025
(e)
0.011
(0.012)
-0.092
(0.061)
(0
-0.012
(0.021)
0.027
(0.023)
-0.087
(0.061)
0.606**
(0.285)
967
0.715*
(0.395)
967
0.667
(0.720)
967
0.609
(0.718)
967
All specifications include d i s t r i c t fixed effects, year dummies, rainfall variables (total rainfall between june and S e p t e m b e r , rainfall
squared and a shock measure taking the value 1 if there is a positive rainfall shock, 0 if no shock and - 1 for negative shocks), and the log
of male and female p o p u l a t i o n for the landless and by land quantiles. Columns (b) t h r o u g h (f) a d d i t i o n a l l y c o n t r o l for the interaction of
district level resources w i t h average t a r i f f and delicensing reforms. The first stage instruments are the average interaction between
between d i s t r i c t level resources, i n d u s t r y resource usage and i n d u s t r y tariffs and regulations. The F-Stats on the first stage are b e t w e n
21.98 and 32.35.
* significant at 10%; ** significant at 5 % ; *** significant at 1%; H u b e r - W h i t e standard e r r o r s are r e p o r t e d in parentheses, standard
errors are clustered at a d i s t r i c t level in columns (a) and at a region-year level in columns (b) t h r o u g h (f).

TABLE 15: Inequality: Manufacturing E m p l o y m e n t by Literacy
D e p e n d e n t Variable: Gini Measure of Inequality
FE
FE-IV
M
-0.026*
(0.013)
(b)
-0.036**
(0.017)
0.010
(0.050)
0.046
(0.127)
0.036
(0.128)
0.056
(0.129)
0.062
(0.130)
0.055
(0.129)
0.047
(0.029)
0.066
(0.127)
Constant
0.588
(0.814)
Y
967
0.603
(0.820)
Y
967
0.554
(0.802)
Y
967
0.421
(0.808)
Y
967
0.532
(0.799)
Y
967
0.493
(0.796)
Y
967
Log(Primary and above Manufacturing)
Log(Skilled Manufacturing)

Observations
(a)
-0.038**
(0.017)
-
(d)
(e)
-0.071*** -0.049***
(0.026)
(0.026)
0.060***
(0.027)
0.033
(0.029)
-
(0
-0.049**
(0.022)
All s p e c i f i c a t i o n s i n c l u d e d i s t r i c t fixed effects, y e a r d u m m i e s , r a i n f a l l v a r i a b l e s ( t o t a l r a i n f a l l b e t w e e n j u n e a n d September, r a i n f a l l

s q u a r e d a n d a shock m e a s u r e t a k i n g the value 1 i f t h e r e is a p o s i t i v e r a i n f a l l shock, 0 if n o shock and - 1 f o r negative shocks), and the
l o g of male a n d female p o p u l a t i o n f o r the landless a n d by land quantiles. Columns (b) t h r o u g h (f) a d d i t i o n a l l y c o n t r o l f o r the
i n t e r a c t i o n of d i s t r i c t level resources w i t h average t a r i f f a n d d e l i c e n s i n g r e f o r m s . The f i r s t stage i n s t r u m e n t s are the average
i n t e r a c t i o n b e t w e e n b e t w e e n d i s t r i c t level resources, i n d u s t r y resource usage a n d i n d u s t r y t a r i f f s and r e g u l a t i o n s .
* s i g n i f i c a n t at 1 0 % ; * * s i g n i f i c a n t at 5 % ; * * * s i g n i f i c a n t at 1 % ; H u b e r - W h i t e s t a n d a r d e r r o r s are r e p o r t e d in parentheses, s t a n d a r d
e r r o r s are c l u s t e r e d at a d i s t r i c t level i n c o l u m n s (a) and at a r e g i o n - y e a r level i n c o l u m n s (b) t h r o u g h (f).

TABLE 1 6 : R o b u s t n e s s Checks, I n c o m e
Panel A: Examining Sensitivity to Removing Self-Employment in the Service or Manufacturing Sector
D e p e n d e n t Variable: Log C o n s u m p t i o n P e r Capita
Landless
<2 acres
>2 acres
Illiterate L i t e r a t e Illiterate L i t e r a t e I l l i t e r a t e L i t e r a t e
Log(Lit M a n u f a c t u r i n g E m p l o y m e n t )
Constant
A d j u s t e d R-Squared
Observations
(a)
0.362**
(0.162)
-0.289*
(0.173)
(b)
0.230*
(0.132)
0.194*
(0.106)
(c)
0.145***
(0.043)
-0.119
(0.098)
0.216***
(0.081)
1.907
(1.247)
0.146
(0.125)
-2.153
(2.106)
0.176*
(0.105)
2.456
(1.848)
0.324
36042
0.301
26600
0.319
24382
(d)
0.106**
(0.042)
(e)
0.134
(0.099)
(0
0.012
(0.112)
-0.152*
(0.082)
0.178**
(0.090)
-0.171
-0.039
(0.069)
0.201**
(0.102)
3.174
0.169*
(0.094)
0.192*
(0.109)
(1.683)
0.327
21173
(2.314)
0.383
30478
P a n e l B: E x a m i n i n g w h e t h e r S e r v i c e S e c t o r E m p l o y m e n t c h a n g e s w i t h i n s t r u m e n t s
D e p e n d e n t Variable: Log of District Level S e c t o r a l E m p l o y m e n t
Construction
Retail
Construction Proportion
Deregulated*Coal
Deregulated* Electricity
Constant
Observations
Region Time T r e n d s
(a)
-0.99
(0.723)
30.978
(25.202)
0.089
(0.209)
0.008**
(0.003)
-40.156
(62.847)
-0.848
(1.417)
15.200
(12.148)
967
N
1.78
(b)
-0.637*
(0.373)
4.868
(13.542)
0.030
(0.152)
0.001
(0.002)
24.035
(39.415)
0.414
(0.496)
3.297
(6.966)
967
Y
0.41
(c)
0.122
(0.442)
-19.070
(24.896)
0.208
(0.163)
0.000
(0.002)
45.085
(62.620)
-2.307*
(1.288)
(10.805)
(11.756)
967
N
0.78
(d)
-0.219
(0.542)
30.476
(25.986)
0.013
(0.205)
0.001
(0.003)
-8.531
(64.347)
-0.774
(1.352)
7.581
(13.328)
967
Y
0.94
1.778
(1.841)
0.344
40654
Transport
(e)
-0.412
(0.485)
8.390
(13.965)
0.028
(0.155)
0.003
(0.002)
42.874
(42.756)
0.812
(0.542)
0.518
(7.973)
967
N
1.12
(0
-0.269
(0.745)
-22.166
(25.820)
0.185
(0.172)
-0.001
(0.002)
28.195
(63.729)
-2.379*
(1.242)
(8.274)
(13.958)
967
Y
1.1
* s i g n i f i c a n t at 1 0 % ; * * s i g n i f i c a n t at 5 % ; * * * s i g n i f i c a n t at 1%
All specifications i n c l u d e d i s t r i c t fixed effects, year d u m m i e s , r a i n f a l l v a r i a b l e s ( t o t a l r a i n f a l l b e t w e e n j u n e and S e p t e m b e r , r a i n f a l l s q u a r e d and a
shock m e a s u r e t a k i n g the value 1 i f t h e r e is a positive r a i n f a l l shock, 0 if no shock a n d -1 f o r negative shocks), the log o f m a l e and female
p o p u l a t i o n f o r the landless a n d by land quantiles. In a d d i t i o n , t h e y i n c l u d e the i n t e r a c t i o n of d i s t r i c t level resources w i t h average t a r i f f and

TABLE 17; Time Devoted to Agriculutral Activities
Panel A: D e p e n d e n t Variable: Log Male Days Worked in Agriculture
FE
Total
Total
Constant
(e)
0.09
(0.058)
(c)
-0.088*
(0.045)
0.127
(0.144)
(b)
-0.021
(0.012)
-0.061*
(0.029)
0.096
(0.139)
6.654***
(1.031)
7.315
(1.030)
5.933*** 4.794*** 8.703*** 8.024***

(2.154)
(1.027)
(1.145)
(2.213)
-0.028**
(0.013)
-
(d)
-0.194**
(0.097)
0.140*
(0.077)
0.244*
(0.129)
FE-IV
Landed
0.037
(0.159)
-0.303
(0.183)
m
-0.086
(0.098)
0.062
(0.092)
-0.09
(0.427)
Panel B: D e p e n d e n t Variable: Log Male Days Worked in Agriculture, by 1 9 9 3 Land Quantile

FE-IV
3rd Quantile
Bottom Quantile
2nd Quantile
(e)
(d)
(c)
(b)
CO
-0.044
-0.114
0.006
-0.048
0.185
0.059
(0.104)
Constant
Observations
0.589*
(0.300)
2.175
(1.994)
Y
850
(0.130)
0.072
(0.129)
0.604**
(0.305)
0.137
(3.127)
Y
850
(0.107)
-
-0.241
(0.432)
2.625
(2.341)
Y
850
(0.113)
0.056
(0.116)
-0.23
(0.436)
2.697*
(1.625)
Y
850
Landless
(h)
(g)
-0.302** -0.737***
(0.125)
(0.151)
0.452***
(0.125)
0.097
0.187
(0.406)
(0.391)
8.900**
(3.546)
6.775**
(2.773)
Top Quantile
(g)
0.162**
(h)
0.431***
(0.078)
(0.114)
(0.125)
-0.279***
0.131
(0.098)
(0.112)
-1.237*** -1.211*** -1.327** -1.385**
(0.449)
(0.550)
(0.553)
(0.455)
7.463*** 8.801***
3.264*
3.022
(1.837)
(1.927)
(2.648)
(2.844)
Y
Y
Y
Y
850
850
850
850
(0.132)
-
All specifications include d i s t r i c t fixed effects, year dummies, rainfall variables (total rainfall between june and S e p t e m b e r , rainfall squared and a
shock measure t a k i n g the value 1 if there is a positive rainfall shock, 0 i f no shock and -1 for negative shocks), the square of the log agricultural
index, the log of male and female population for the landless and by land quantiles. Columns (c) - (f) additionally control for the interaction of
district level resources w i t h average t a r i f f and delicensing reforms. The first stage instruments are the average interaction between between
district level resources, i n d u s t r y resource usage and industry tariffs and regulations.
* significant at 10%; ** significant at 5 % ; *** significant at 1%; H u b e r - W h i t e standard e r r o r s are r e p o r t e d in parentheses, standard errors are
clustered at a d i s t r i c t level in columns (a), (b) and (e) and at a region-year level in columns (c) and (d)
CHAPTER 3
What is the effect of growth in household incomes and returns to

education on household educational investment?
3.1.
INTRODUCTION
Economic growth is empirically associated with rising investment in education.

This empirical regularity could be driven by an income effect as well as by a returns to education effect. Growth may directly raise the returns to schooling, for
example skill biased technological change that raises the demand for educational
skills (Nelson and Phelps, 1966; Schultz, 1975). Growth also alters the level and
distribution of income through changing the returns to unequally distributed productive assets (Kuznets, 1955), as shown in chapter 2.
The literature on the demand for education in developing countries has found
that education responds positively to both incomes (Edmonds, 2006; Edmonds et
al, 2004; Jacoby and Skoufias, 1997) and returns (Kochar, 2004; Rosenzweig and
Foster, 1996). The relative contribution of these channels is however unclear since
the literature has typically focused on one determinant of education at a time. 1
Focusing on one channel may confound the two effects if growth raises wage and
income levels as well as alters returns to education.
This paper deviates from the literature by separately jointly estimating the effect of incomes and returns on educational investment. Since both incomes and
returns are likely to be correlated with unobservable determinants of educational
investment, such as education quality. I use an instrumental variables strategy to
^The relative m a g n i t u d e of the t w o effects is unclear since the existing studies d o not cover overl a p p i n g time f r a m e s or populations. This reduces the comparability of the estimates since income
effects capture the education response of h o u s e h o l d s with a given marginal utility of income, t h u s
the estimated effects are likely to vary across h o u s e h o l d s and p o p u l a t i o n s according to income
levels.
78
3. DETERMINANTS OF EDUCATIONAL INVESTMENT
79
do this, where the instruments used are the two drivers of growth examined in
chapter 2 - district-time varying agricultural productivity and a district level measure of industrial policy changes. A primary empirical challenge in disentangling
the two effects is that we need variation independent and exogenous variation in
household incomes and returns. I exploit the observation that the agricultural and
manufacturing sector differ in their demand for skill and land. Therefore exogenous variation in the size of these two sectors alters the ratio of skilled and unskilled wages as well as their levels, and also differs in its impact on the incomes
of skilled and unskilled households, and landed and landless households.
I frame my empirical analysis using a model of household education choices.
The model discussed in this chapter is set within the general equilibrium model
of rural labor markets discussed in chapter 2. The model illustrates the effects
of increases in agricultural productivity and policy induced expansions in skilled
and unskilled manufacturing labor demand on returns to education and income
for skilled and unskilled workers. The model then predicts how changes in the
aggregate production function feeds through to education through the income and
returns to education channel.
I test the model's predictions using district and household level data from rural
India between 1983 and 1999. The data covers over 250 district level labor markets.
I compile a new dataset of district mineral and metal endowments which I combine
with data on industry raw material use and industry policies to instrument skilled
and unskilled wages, and skilled and unskilled manufacturing employment in my
first stage regressions. I then examine education responses to within district overtime variation in skilled and unskilled wages.
The education specifications examine whether children aged between 5 and
9 start school. Unskilled manufacturing growth increases the probability a child
starts school in unskilled landless households, while it reduces it in landed households. The intuition lies in the income effect: unskilled manufacturing growth
80
raises income in unskilled landless households, while it has no effect in landed

households.
The income and returns to education effects on educational investment are estimated using m i n i m u m distance. In the absence of data on incomes, I use two
specifications to capture the income and returns effects. In a linear log approximation to a household's decision rule, education is a function of the difference
between skilled and unskilled wages, which captures the return to education, and
household incomes. In the first specification, I focus on landless unskilled households who work for unskilled wages in the wage labor market. Substituting the
unskilled wage into their incomes, I find that the skilled wages captures only a returns to education effect while the unskilled wage captures both an income and a
returns effect. I instrument these wages using growth in agricultural productivity
and district level measures of industrial national policy.
In the second approach, I substitute the empirical approximations of wages
and incomes into the educational specification. Since agricultural productivity and
manufacturing employment alter educational investment through incomes and returns to education, the coefficients on these two terms in the education specification reflect weighted combinations of income and returns to education effects,
where the weights are wage and income responses to the two determinants of
growth. I use the estimated coefficients from the education, wage and income
specifications to jointly estimate the two effects for rural landless households.
I find that income growth accounts for a larger fraction of education growth
than changes in the returns to education over the period. A one standard deviation increase in household incomes raises male enrollment by 0.05, while a one
standard deviation increase in returns raises enrollment by 0.02. Male educational
investment responds positively to rising household incomes and labor market returns to education. In contrast, investment among girls only responds to rising
81
incomes. This finding fits with the observation of low female participation in complex occupations in rural areas. I test whether raising the child wage reduces education through the opportunity cost channel. I find no effect, consistent with the
low incidence of child labor among young children.
lifts all boats. Regardless of the source of growth, education rises among the poorest households in India - rural, unskilled, landless households. Sources of growth
that raise incomes but reduce returns to education, such as agricultural technological change and unskilled manufacturing growth, increase educational investment
substantially. Skill biased growth raises returns but has little impact on incomes; it
raises education to a lesser degree through the returns to education channel. The
same is not true for girls, whose education was found to only respond to rising
income: skill biased growth has little impact on their education.
The chapter is structured as follows. The next section describes the related
literature. Section 3.3 presents descriptive statistics which are used to motivate
the assumptions of my model. In section 3.4, I put forward a model to frame the
empirical analysis. I test the model's predictions in section 3.5 and section 3.6
concludes.
3.2. RELATED LITERATURE
This section puts forward a review of the literature estimating the determinants of
educational investment, with a focus on studies from developing countries. The
canonical theory of human capital starts with the hypothesis that education is an
investment, chosen by comparing the costs of education to the expected future income benefits (Becker (1964)). In this framework, the length of time spent in school
is influenced by individual characteristics, such as individual ability, the discount
rate and the returns to education, which is captured by the complementarity between h u m a n capital and productivity. This paper focuses on education decisions
82
driven by future monetary returns. While education m a y also be d e m a n d e d for

its positive effect on non-market outcomes, these non-market channels are outside
the scope of this paper.
The literature can broadly be divided into studies examining the supply of education, such as an expansion of the existing infrastructure, and those focusing on
the d e m a n d for education. Since the focus of this paper is on the d e m a n d side, the
o
literature review focuses on studies in this area. In section 3.3.1,1 review studies
which examine the d e m a n d response to a rise in household incomes. In section
3.3.2 I focus on education responses to rising returns.
3.2.1. Demand for Education: Income channel

In the Barro-Becker model of education choice, investment is independent of current family wealth and parental preferences (Becker, 1981). This f r a m e w o r k imposes strong assumptions, notably perfect altruism, perfect credit markets and
no symbolic consumption of education. Relaxing the assumptions of this model,
a positive relationship between educational investment and household income
arises d u e to credit constraints, symbolic consumption of education, or because
borrowing costs are correlated with parental income (Becker, 1975; Becker a n d
Tomes, 1986; Banerjee and N e w m a n , 1994; Baland and Robinson, 2000; Banerjee,
2004).
The empirical evidence finds a positive relationship between family income
and school attainment in both developed (Bowles, 1972; Jencks et al, 1972; Lazear,
1980) and developing country contexts (Behrman and Knowles (1999); Edmonds,
For example, studies h a v e f o u n d a strong relationship b e t w e e n parental education a n d child

health, a n d e d u c a t i o n a n d criminality (Currie a n d Moretti (2003), Lochner and Moretti (2004); see
D u n c a n a n d T h o m a s (1995) for a s u r v e y of the literature).
o
Recent studies on the s u p p l y side include, to n a m e b u t a few, Duflo (2000), Deaton a n d C a s e

(1997).
83
2009; E d m o n d s , 2006; E d m o n d s et al, 2004; Jacoby (1994, 2004); Jacoby and Skoufias, 2004). A strand of the literature examines the relationship between transitory variation in incomes and education choices. Jacoby a n d Skoufias (1997) use
data f r o m rural India to show that school attendance responds to seasonal income
fluctuations; they interpret this as evidence that households face incomplete financial markets. Using data from Tanzania, Beegle et al. (2006), present evidence that
transitory income shocks results in a reduction in child labor. They find that the
effect is reduced w h e n households have assets which can be used as collateral, and
interpret this as evidence of credit constraints. E d m o n d s (2006) uses the discontinuity of eligibility for pensions in South Africa to examine child labor a n d school
attendance responses to pension eligibility. Since pension incomes are entirely anticipated, the timing of anticipated changes in income should have no effect of
labor or schooling if households are able to borrow against p e r m a n e n t income.
E d m o n d s finds that families w h o become eligible for pension income exhibit large
child labor and schooling responses a m o n g 10-17 year olds. The evidence in the
papers discussed may however also be consistent with education being a normal
good.
A thorough review of income estimates can be found in Behrman and Knowles
(1999). Schooling is measured in the literature using a n u m b e r of outcomes, from
ultimate grade attainment and enrollment to grade repetition and subject specific
test scores. The median income elasticity estimates across these measures is 0.07,
estimates above 0.20 are found for samples with poorer households and in studies
conducted in low income regions.
3.2.1.1. Demand for education: Returns to Schooling Channel. In theoretical m o d els of educational investment, raising returns to education increases educational
investment u n d e r a range of scenarios. These results hold even if credit markets are
shut d o w n , as long as w e have either perfect altruism, no symbolic consumption
84
and no contraction, or no altruism and perfect contracting (Loury, 1981; Banerjee,

2004).
The empirical evidence examining the response of household educational investment to returns to education is somewhat more limited than the evidence on
the income channel. Foster and Rosenzweig (1994) present evidence that technological change is complementary with decision making activities in agriculture.
Since decision making tasks are not conducted on the hired labor market, and
land markets are thin, growth in agricultural productivity only alters the expected
returns to education of sons in landed households. In their rural labor market
framework in which only the agricultural sector is present, there are no returns to
education for landless households in rural labor markets. 4
Kochar (2004) argues that, despite residing in predominantly agrarian areas in
which expected returns to education are low, rural educational investment may
still occur in response to local urban returns to education. She examines the response of middle school enrollment in rural areas to variation in urban rates of
return. Since variation in urban rates of return may reflect rural preferences for
education, she instruments urban wages using the ratio of the number of urban
female workers with higher education in the region to those with less schooling. 5
^Foster a n d Rosenzweig s h o w that, d u r i n g a p e r i o d of technical c h a n g e in Indian agriculture bet w e e n 1971 a n d 1982, the educational enrollment of children in cultivator h o u s e h o l d s w a s responsive to changes in the returns to e d u c a t i o n facing these h o u s e h o l d s . They s h o w that the introduction of HYV seeds, raised the return to literacy a m o n g cultivator h o u s e h o l d s t h r o u g h . They u s e
changes in agricultural technology d u r i n g the green revolution in India to estimate the r e s p o n s e of
educational i n v e s t m e n t to c h a n g e s in the r e t u r n s to schooling. They find that technical c h a n g e in
agriculture increased the r e t u r n s to schooling a n d educational i n v e s t m e n t of l a n d - o w n i n g households, w h o c o n d u c t complex decision m a k i n g tasks on their o w n land-holdings. In contrast, e d u cational i n v e s t m e n t in landless h o u s e h o l d s , declines in the face of technical change, w h i c h is attribu t e d to increases in child labor m a r k e t w a g e s that raise the o p p o r t u n i t y cost of schooling (Foster
a n d Rosenzweig, 2004).
5
Since rural to u r b a n migration is low for females, u r b a n f e m a l e e d u c a t i o n is a r g u e d to b e uncorrelated w i t h schooling preferences of rural h o u s e h o l d s or other u n o b s e r v e d rural conditions affecting
educational investment.
85
Since households differ in their probability of migration, and hence in their

probability of working directly in the urban labor market, Kochar separately estimates the coefficients among households with a low and high probability of migration (landed and landless households respectively). She argues that, since landless
households face a higher probability of working in the urban sector, comparing
the estimates across landed and landless households provides an additional test
of whether the estimates are identifying rate of returns effects. The relationship
between urban wages and rural educational investment may also capture rural incomes, for example through rural or urban wage channels or through the demand
for rural products. Since landless households have low levels of higher schooling
and are more likely to earn low-education wages, if only the income effect were in
operation the relationship between urban wages and incomes would be negative.
However, finding of a positive effect for landless households does not rule out the
possibility that the income effect is captured in these estimates. 6
Jensen (2010) presents evidence that perceived returns to secondary schooling
differ from measured returns among 8th grade students in the Dominican Republic. The paper presents evidence that students display an education response to a
rise in expected educational investment. Schools in a random subset of schools are
told the measured returns to school. Treated students found to report increased
perceived returns 4 to 6 months after treatment, be 4% more likely to be enrolled
the next academic term and 4 years later are likely to have completed 0.2 more
years of schooling. The program was found to have the largest effect among higher
income households, with no effect on low income households. This paper therefore suggests an interaction between returns to education and income.
6
To see this, a s s u m e that the u r b a n w a g e s of high skilled w o r k e r s are held fixed a n d that e d u c a tional i n v e s t m e n t is only driven by variation in incomes, i.e. the rate of r e t u r n s effect is zero. First
take the case in w h i c h u r b a n r e t u r n s are low b e c a u s e the w a g e s of low-skilled w o r k e r s are high.
Then the incomes of landless w o r k e r s w o r k i n g in low-skilled occupations are high, a n d their e d u cational investment is also high. Second, take the case that u r b a n r e t u r n s are high b e c a u s e illiterate
w a g e s are low, then the incomes of landless h o u s e h o l d s are low a n d educational i n v e s t m e n t a m o n g
these h o u s e h o l d s is low. If the returns channel is also operational, it is possible that b o t h an i n c o m e
a n d returns channel is in effect b u t that the r e t u r n s channel o v e r w h e l m s the income effect.
86
3 . 3 . S E T T I N G : E D U C A T I O N IN I N D I A
In this section, I provide descriptive evidence supporting the central a r g u m e n t s of

the paper. Section 3.3.1 shows trends in incomes, labor market returns to education and education participation. The descriptive statistics point towards income
driving rising educational investment. Section 3.3.2 describes the supply side of
education in India, with a focus on the educational infrastructure. The descriptive
statistics covering India's 15 major states are from four r o u n d s of household data
collected in 1983,1987,1993 and 1999 by the National Sample Survey Organization
(NSSO). In addition, I m a k e use of the 1982 and 1999 r o u n d s of data from the Rural
Economic and Development Surveys (REDS) collected by the National Council of
Applied Economics Research (NCAER). The data are described in greater detail in
the Data Appendix.
3.3.1. Education, Incomes and Wages
The descriptive statistics presented in this section point towards incomes driving
the rise in educational investment: incomes and the proportion of children starting
school have risen over time, while the ratio of skilled to unskilled wages has fallen.
Descriptive statistics on educational and child labor are presented in table 18-a and
18.b.
Between 1982 and 1999, average real household income grew by approximately
70% in the REDS data while in the NSSO real per capita consumption grew by
22%. Agricultural wages grew by approximately 52% over this period.
The real
w a g e s received by literate and primary educated workers in the non-agricultural

sector grew by 43% and 42% respectively. The wage received by m a n u a l labor in
agriculture can be considered as the reservation wage for uneducated workers in
rural areas (Deaton and Dreze, 2002). I therefore use the agrarian w a g e to proxy
for the unskilled wage, and the wages of literate/primary educated workers as
n
The NSS data indicate that w a g e s grew by 59% b e t w e e n 1983 and 1999, while the REDS data point
to a m u c h sharper rise of agrarian w a g e s of 68%.
87
the skilled wage. The skilled wage ratio is defined as the ratio of the skilled wage
relative to the unskilled wage. The literate wage ratio has declined by 9%, from
1.72 in 1983 to 1.58, while the primary wage ratio has declined by 14% over the
period. Therefore, rural areas in India have experience substantial rises in the level
of incomes between 1983 and 1999, although the skilled wage ratio has decreased
on average over time.
The education margin examined in this paper is whether to send children to
school or not. This is an important margin of choice in India, despite technically
free education: 30% of boys and 52% of girls aged between 5 and 9 in 1983 never
started primary school. The Public Report on Basic Education in India (PROBE,
1999) reported that a third of all children aged between 6 and 14 were out of school
in 1995. India's education record is cast in a low light, both in absolute terms and
relative to other developing countries. For example, female literacy rates are lower
in India than in sub-Saharan Africa and the average adult has spent on average
two years at school, compared to five years in China, seven years in Sri Lanka
(PROBE, 1999).
School attendance has increased sharply between 1983 and 1999 - by 1999, 9%
Q
of boys and 14% of girls hadn't started school.
I examine the entry decisions in
retrospect, where children aged between 5 and 9 in 1983 are aged between 10 and
14 in the 1987-88. Education investment varies substantially across states - from
near universal male enrollment in Kerala as early as 1983 to 45% of boys out of
school in Bihar. Kerala is excluded from the analysis since the decision to start
schol doesn't appear to be a relavant margin of choice in this state.
Child labor is low among children aged 5 to 9. In 1987, 2.5% of boys and 8% of
girls reported conducting domestic or income generating activities. Among children not attending school, the majority report no activity. These patterns suggest
o
The startling trends seen in the NSS e m p l o y m e n t - u n e m p l o y m e n t r o u n d s are confirmed in the N a tional Family Health Survey w h i c h h a s a m o r e detailed education m o d u l e (Kingdon, 2007). T h e
PROBE report n o t e s that the n u m b e r of children n e v e r enrolled in school d r o p p e d from approximately half in 1983 to a p p r o x i m a t e l y 20 percent b y 1996.
88
that children w h o are not attending school are likely to be engaged in leisure activities. 9 The proportion of children working in either an income generating or
domestic capacity is strongly negatively correlated with w a g e levels, and in particular with the unskilled wage. This correlation is driven by households with
below m e d i a n land-holdings, w h o report the highest levels of child labor.
Table 18-b indicates that the proportion of children starting school in a district
is strongly positively correlated with the level of skilled and unskilled wages. Areas with higher levels of skilled and unskilled wages therefore see higher levels
of educational investment. The raw correlation between the ratio of skilled and
unskilled wages and education suggests that districts in which the returns to education are high also experience higher investment in education. The statistically
significant positive correlation between levels of wages and proportions of children attending school holds in each cross-section, b u t the statistically significant
relationship between returns and schooling holds only in 1993 and 1999.
3.3.2. Institutional Structure and Education
Infrastructure
This section presents evidence on India's education system, on the provision of

educational infrastructure and changes to the structure of the system over the time
period examined.
Elementary education is the joint responsibility of the state and central government. The vast majority Of the public expenditure on education is borne by the
state. Tuition fees in government primary schools are technically free across all
states (Mehta, 1996), b u t additional expenditures such as uniforms and books, account for approximately 300 Rs (1996 Rs) per a n n u m on average, or approximately
10 days of male agrarian w a g e labor (PROBE, 1999). The quality of educational
infrastructure, teaching and administrative resources leaves m u c h to be desired;
9
T h e s e p a t t e r n s d o not a p p e a r to b e an artifact of the b r o a d questions on an i n d i v i d u a l ' s principal

u s u a l activities since they are b o t h seen in the weekly time recall data a n d in the m o r e detailed
REDS data-set.
89
as such, the elementary education system has been under continual reform at both
a state and central level. A concern of this paper is that the provision of education infrastructure, and change in infrastructure at a district-year level may be
correlated with district level shocks. I therefore detail the major changes to infrastructure and management that have occurred since 1983. In section ??, I examine
whether districts experiencing greater increases in agricultural potential or greater
employment effects of industrial policies saw differential changes in their education infrastructure, costs or programs.
Several changes to the education environment have occurred during the period examined in this paper. In 1987, the government of India launched Operation
Blackboard, with the aim of ensuring that every primary school had a basic minim u m of facilities. 10 The PROBE report suggests that OB has achieved some results:
the number of single-teacher primary schools has declined and materials have improved somewhat. However, the achievements of OB are below target. Due to
a lack of data, I am unable to test the robustness of the results to changes in the
availability of classroom aids and materials. I am however able to verify their robustness to changes in the number of teachers per capita and the average distance
of a household from a primary school.
In 1994, the District Primary Education Program (DPEP) was launched in a bid
to increase both the quality and quantity of primary education infrastructure, with
the aim of achieving the objective of universal primary education (Government
of India, 2002). The program began towards the end of 1994, with 42 districts
spread over seven states (Government of India, 2004). However, the roll out of the
program was limited during the first few years of its existence, with the program
really .
10
T h e facilities and aids p r o m o t e d b y OB are: (i) At least t w o reasonably large all w e a t h e r r o o m s

a l o n g w i t h separate toilet facilities for b o y s a n d girls; (ii) at least t w o teachers, as far as possible o n e
of t h e m a w o m a n ; a n d (iii) essential teaching a n d learning material including blackboards, m a p s ,
charts, a small library, toys, g a m e s a n d s o m e e q u i p m e n t for w o r k experience.
90
In late 1995, the government announced a national school-meal p r o g r a m m e .

The argument for the program w a s that the provision of school meals improves
enrolment and attendance (PROBE, 1999). However, implementation of the program during the initial phase w a s quite slow - the PROBE report finds no evidence
that mid-day meals were in place by mid-1998 (PROBE, 1999). To examine the robustness of m y results to an increase in school meal provision, I examine w h e t h e r
districts experiencing greater increases in agricultural potential or greater employment effects of industrial policies also h a d a different roll out of the mid-day meals
program.
3.4. THEORETICAL M O D E L
This section presents a model of household education choices. The assumptions of

the model d r a w u p o n the descriptive statistics presented in the previous section.
The model of household education choices is nested in the General Equilibrium
framework in section 2.4. Therefore, changes in agricultural productivity and industrial policies feed through to education choices by altering the wages faced and
incomes earned by households.
3.4.1. Households: Endowments, Preferences and Maximization
Problem
Households are modeled over two periods and consist of t w o generations. In the
first period, each household includes an adult a and a child y. In the second period,
the adult dies and the child becomes a y o u n g adult. 1 1 Household subscripts are
suppressed in this section. Households have three e n d o w m e n t s in both periods:
land (A), adult education (s") and a unit of adult labor market time. In the first
period, households are additionally e n d o w e d with 1 unit of child time. This can
either be devoted to education or to producing a domestically consumed good.
Adult education takes the value of 0 if the adult is uneducated or 1 if educated.
Adult education in period 2 is the only endogenously determined e n d o w m e n t ,
abstract f r o m the child-bearing decisions that follow.
91
Adults choose the child's education in period 1. The choice is discrete - the child
either attends school or does not.
Households have time separable preferences over consumption in period 1 and
2 - C\ and c2. Utility is concave in consumption - 0 < p < 1.
V{clfc2)
1 9
= u(Cl) + fiu{c2) = c? + jSc
(3.1)
Consumption goods are of two perfectly substitutable varieties: h o m e produced

by the child using a unit of child time lXJ or purchased at price pc:
C
1 =
C2 =
ld + clp
+ Cip
C2p
Households are unable to borrow or save, implying that their b u d g e t constraints

clear every period. 1 3 In the first period, income is divided between purchased consumption and schooling. In the second period, income is spent only on purchased
consumption. For tractability, there is no leisure; in the empirical framework, I
discuss h o w incorporating leisure alters the testable predictions. 1 4
Adults split their labor market time between four activities: unskilled work in
agriculture, unskilled work in manufacturing, skilled work in manufacturing and
skilled w o r k in agriculture. Unskilled work in agriculture and manufacturing are
activities which both educated and uneducated individuals can conduct; skilled
work is defined as activities that only educated individuals can do. Skilled work
in agriculture consists of supervisorial activities, which can only be conducted by
members of landed households. In the first period, children either go to school
1 9
There is n o symbolic c o n s u m p t i o n of education - i.e. there is n o " w a r m g l o w " of giving to one's

child (Andreoni, 1989).
1^
Credit m a r k e t s in d e v e l o p i n g countries offer limited o p p o r t u n i t i e s to finance educational investm

e
n t (Banerjee, 2004), m o t i v a t i n g the a s s u m p t i o n that h o u s e h o l d s are credit constrained.
14
J a y a c h a n d r a n (2004) presents evidence indicating that labor s u p p l y responses to a productivity
shock are m o r e inelastic in districts in w h i c h credit constraints are greater, in w h i c h a greater proportion of w o r k e r s close to subsistence a n d in w h i c h migration costs are high. In the empirical
section, I e x a m i n e h o w allowing labor s u p p l y elasticity to v a r y according to labor m a r k e t alters the
interpretation of m y results.
92
or produce the domestic consumption good. The time constraints of adults and
children in period 1 and adults in period 2 are given by:
(1 - s\)lyl
+ s\lsl
li2 + l% + l&
= l)(lg
l(Ah>0)l)
where lAt and l^j is unskilled time in agriculture and manufacturing in period t =
1,2; lAt is skilled time in agriculture and manufacturing. Zyi denotes a unit of child
h o m e production time in period 1, while l s \ denotes a unit of time spent at school.
Households earn income from supplying labor to the w a g e labor market and
from cultivation if landed. They earn income m from an exogenous source.
yi
wu2(lA2
+ l^2) + l(sa2 = l)ws2l^
+ l(A>0)UA(sa2)
+ m2
The household problem is given by:

Cj + p>cp2
maxs.v
s.t.
pcclp + s\Cs
2
where s^ =
pcc2p
(3.2)
<yi(s?)
< y2{sa2)
- the schooling of the youth in the first period is the educational
e n d o w m e n t of the household adult in the second period. Since education choices

are discrete, households educate their children if the indirect utility from obtaining
schooling is greater than that from not doing so:
V(s? = l )
>
V(s? = 0)

0
93
{ y y ^ l ~ C s ) P + (i(w u2 l u 2 + W s2 l s2 + n A (s n 2 = l))P}
uA
(3.3)
( s 2 = o)) p ] > o
3.4.1.1. Comparative Statics: Education. This model leads to a series of testable

predictions for education:
a Income: A child is more likely to be educated as household income rises,
ceteris paribus. This prediction is driven by the assumption that the b u d get constraint clears every period.
a
dm l
- Q y
- (yhl + pay1)
> o
b Labor Market Returns to Education: A child is more likely to be educated as

the ratio of the skilled and unskilled w a g e rises, ceteris paribus.
a
( I \9
( i )
\Pc /
,w2
W
ll
c Returns to Education in Agriculture:
= i)r
>o
A landed household is more likely to
invest in education as agricultural productivity rises, in the land wealth of

the household and in their cross partial. Intuitively, agricultural productivity raises the return to education of landed households.
Since land-
less households don't cultivate land, their returns to education do not

change. 1 5
a
15
T h e s e predictions h a v e b e e n previously been tested b y R o s e n z w e i g a n d Foster (1996), w h o find

that the returns to schooling increased d u r i n g the green revolution in India. Areas with the greatest agrarian technical c h a n g e w i t n e s s e d the greatest average increases in school enrollment, conditional on the availability of a school.
94
l.d Opportunity Cost: Educational investment is decreasing in the opportunity

cost of schooling, lost domestic production.
= -p(yh
+ PeccY
<0
3.4.2. From Theory to Empirics

The education decision rule is a function of household income, as well as the returns to education. If w e were to take this model directly to the data, and to estimate the relationship between observed incomes, wages and household educational investment, one w o u l d obtain biased parameter estimates. For example,
both household incomes and returns are likely to be correlated with education
quality in a region.
Therefore, to capture how education choices respond to variation in incomes
and wages, w e need to capture exogenous and independent variation in incomes
and returns that is uncorrelated with unobserved determinants of educational investment. This paper uses an instrumental variables strategy to do this, where
the instruments used are changes in agricultural productivity and a regional measure of industrial policy. The instruments are motivated by the model and results
of chapter 2, where households are nested in a two-sector, small country general
equilibrium model of rural labor markets. Exogenous changes in agricultural productivity and variation in industrial policy alters wages and household incomes
through altering sectoral skilled a n d unskilled labor d e m a n d .
The data used in this paper can be divided into six categories: industry-region
level data on employment, wages, regional raw material e n d o w m e n t s , industry
varying factor intensities, time and industry varying policies and household education. The data will be presented as it is used. A more detailed description of the
data can be found in the appendix 2.
95
3.5. EMPIRICAL SPECIFICATION
In chapter 2, the drivers of w a g e and income growth vary in their effect on incomes as well as on the returns to education, as captured by the differences in
their estimated coefficients on unskilled and skilled wages. I use these observations to disentangle the effect of an increase in income from changes in the return
to education on educational investment.
The prediction from the model state that the probability that an individual is
educated increases in current household income, the anticipated return to schooling in the labor market (the relative wages of skilled and unskilled workers w h e n
the child enters the labor market), the anticipated return to education within the
agricultural sector, and decreases in the opportunity cost of schooling. I restate the
condition under which households educate their children for convenience (equation (3.3)):
sl = i
if
~ M ~
-(wl
+ nA(si2
+ nA{ s
- ">
=1))p
= o))] > o
Schooling decisions reflect expected returns to education. To empirically estimate

the determinants of education, w e need to m a k e assumptions about the formation
of expectations. I assume that parents have myopic expectations about the returns
to education faced by their children. They therefore believe that the future returns
faced by their children are the same as those faced today:
w
sd,t+1
sdt ,
w
r C,hdt
ud,t+l
udt
lv
9d,t+1 = 6dt + ahdt
where
96
= 0 and E[ah dt ] = 0. I use current w a g e s as a proxy for the wages
children receive u p o n entering the labor market. I test the robustness of the estimates to empirically relaxing this assumption. Since I examine whether a child
starts school of not, the return considered is that from becoming literate.
I specify the following linear approximation to equation (3.3):
SFO/F = 1
if
Ko + KIYWF + K2(wsdt
- wudt)
+ K3dhdt * Ah + JC 4 OQ,
(3.4)
+k5Csdt + }ih + Vd + Vt + ehdt > 0

where Sjldt is a d u m m y variable capturing whether the individual started primary
school and OCdt represents the opportunity cost of schooling. Educational investment is predicted to increase in household income (Ki > 0) and in the labor market
return to literacy (k2 > 0). For cultivator households, enrollment is predicted to
increase directly in agricultural productivity, d u e to its complementarity with education in cultivation activities (k3 > 0). Education is predicted to decrease with the
opportunity cost of schooling (K4 < 0) and the direct monetary cost of schooling
7C5 < 0. Education m a y also vary with time invariant district characteristics, such
as regional preferences for education.
As noted in section (2.5.5.1), landless households earn income in the w a g e labor
market and landed households additionally earn income from cultivation activities. Substituting household income into the education equation (3.4) and combining parameters, w e get an estimating equation that is a function of the wages of
skilled and unskilled workers, the cost of schooling and cultivation revenues:
S/u/f = 1 if
K0 + (KjA~K2
+ K4)wudt+
(k-1% + k2)ws + Kil(A
> 0 ) n A (3.5)
+K3edt * Ah + k5Csdt + nh + nd + }it + ehdt > 0

where A denotes the proportion of household time spent working in the unskilled
labor market and denotes the proportion of household time spent in the skilled
97
labor market, n" 4 denotes revenue from household cultivation, net of hired labor
and other input expenses.
The coefficients on wages and profits capture combinations of the reduced form
parameters and vary across households according to their skill a n d land holdings.
Firstly the parameters on wages vary across educated and uneducated households,
since they devote different a m o u n t s of time to skilled and unskilled activities.
Secondly landed households additionally earn income from cultivation activities.
Cultivation revenues are themselves decreasing functions of wages. In the absence
of data on cultivation revenues, the estimated coefficient on wages will additionally capture education responses to changes in revenues driven by w a g e variation.
Directly estimating equation (3.5) using OLS is unlikely to capture causal effects. For example, the ratio of skilled to unskilled w a g e s is likely to be correlated
with local preferences for education and household incomes may reflect school
quality. To overcome the endogeneity bias, I examine education responses to income and returns to education variation induced by changes to agricultural productivity and instrumented manufacturing employment. As discussed in section
(2.5.1.2), the excluded variables are unlikely to be partially correlated with local
education characteristics. I use two approaches to estimate the coefficients in this
specification: (1) an indirect least squares strategy and (2) an instrumental variables strategy.
In the indirect least squares approach, I substitute the empirical specifications
for cultivation revenues and wages into equation (3.5). The estimating equation
therefore examines education responses to changes to agricultural productivity
and predicted manufacturing employment. 1 6
Sfrrff =
+ lE total,dt + ^skilled,dt
+ a3&dt +
+ a5HHht
+a6Ad/t
16
T h e profit specifications are discussed in greater detail in a p p e n d i x B.2.
+ u"
(3.6)
98
The parameter combinations vary across g r o u p s of households according to their

skill and land assets. For landless households, w h o only earn income in the w a g e
labor market, the estimated coefficients are combinations of the a and
terms f r o m
the w a g e regressions and the k terms from the structural education specification
(equation 3.4). For example, for landless literate households the parameters on the
first three terms are given by:
(KiA + K 4 )ai + (Ki + K 2 ) ( 0 i - a i )
a2
(K1A~K2
a3
(KiA + K4)a3 + (Ki^ +
+ K4)(0i2-0Li)
[ k i
] ( P 2
pi)
K2){p3~a3)
For landed households, the parameters additionally capture terms from the cultivation profit equation. The parameters in equation (3.6) in chapter 2 therefore
capture a weighted combination of the k terms, where the weights capture the relationship between incomes, wages and the source of growth.
I use a probit and linear probability model to estimate the coefficients. The
structural parameters are estimated using the Optimal M i n i m u m Distance estimator, where the weight matrix is the inverse of the variance covariance matrix of
the reduced form coefficients. The weights are the estimated a and /3 terms from
sections 2.5.2.3 and 2.5.5.1. Since I am unable to empirically estimate the cultivation profit terms, I am only able to estimate the structural parameters for landless
households.
I first use the estimates of a\, a2 and a3 to estimate K\ and k2, setting k 4 equal
to zero. I estimate the parameters separately for literate and illiterate households,
as well as by sex. Secondy, I set k2 and k 4 to be equal for literate and illiterate
households, and test whether k 4 is statistically different from zero. The opportunity cost of schooling reflects productive uses of child time such as engaging in
domestic production or income generating activities. The descriptive statistics in
99
section 3.3.1 suggest that few children w o r k prior to age 10. Therefore the opportunity cost of y o u n g children attending school is likely to be small; I therefore test
whether k 4 = 0.
I include district fixed effects in all specifications to capture variation across
districts in unobserved time invariant determinants of education that m a y be correlated with the average level of manufacturing employment and agricultural productivity in a district, such as education quality. In the absence of data on the
costs of schooling facing households, I include state-year interactions which capture variations in the provision and cost of education at a state level. I additionally
include region time trends to capture trends in education provision within a g r o u p
of districts.
Using the IV strategy, I instrument wages using agricultural productivity and
the interaction of effective tariffs a n d district resource e n d o w m e n t . In the absence
of data on incomes or cultivation revenues, this approach can only be used for
landless households. I assume that illiterate households work only in the unskilled
labor market, an assumption consistent with the results presented in section 2.5.5.1.
Literate households work in both the skilled and unskilled labor markets. The
estimating equations are therefore:
Illiterate Landless:
S=1 hdt if
o + OI -
2 + k4)wudt
+ K2wsdt + k5Cdt +
(3.7)
+Vd + Ft + ehdt > 0

Literate Landless:
S=lhdt
if
/c0 + (zq A - k2 + K4)wudt + (ki + K2)ivsdt + k5Csdt -(3.8)

+Vd + Ft + ehdt > 0
The three structural parameters, K\, k2 and

imum.
are estimated using the optimal min-
100
3.5.0.1. Data. To capture education choices, I use 4 waves of NSSO data between 1987 and 2004 to examine ex-post education choices between 1983 and 1999.
I examine h o w the w a g e s and incomes faced w h e n children are aged 5 to 9 alter
the school entry decision. For example, children aged 5 to 9 in 1983 are aged 10 to
14 in 1987-88. The education measure is thus whether children aged between 10
and 14 in 1987 ever attended school. 17
3.5.0.2. Results. The estimates f r o m the education specifications are presented
in tables 19 and 20. Panels A and B of table 19 present the estimates f r o m equation
(3.6), while panel C presents the estimates from equations (??) and (3.8). The structural parameters estimated using the m i n i m u m distance estimator are presented
in table 20.
In table 19, the estimated parameters on agricultural productivity and instrumented manufacturing vary across household asset groups. In illiterate landless
households, a 10% increase in agricultural productivity raises the probability that a
boy attends school by 0.03. Since growth in agricultural productivity raises w a g e s
but reduces the returns to schooling, the coefficient sign indicates that the positive income effect outweighs the negative effect from a reduction in the returns to
schooling. Landed households also exhibit positive education responses to agricultural productivity - a the same increase in agricultural productivity raises the
probability that a boy in these households attends school by 0.03. This is likely to
reflect both income and returns to education effects since agricultural productivity raises the return to primary school through managerial activities in cultivation
(Foster and Rosenzweig, 1996).
In column (c), an increase in unskilled manufacturing employment raises the
probability that a child in a literate household acquires education, while it reduces
it in large landless households. Since both sets of households experience similar
17
H o u s e h o l d s are asked the level of educational a t t a i n m e n t that their child has achieved (Governm e n t of India, 2001). A child w h o has n e v e r a t t e n d e d school is reported as either illiterate or literate
w i t h n o f o r m a l schooling.
101
reductions in the return to literacy, the difference in education responses is likely

to reflect heterogeneous household incomes responses to unskilled growth. Both
literate and illiterate landless households display positive responses to increases in
predicted skilled manufacturing employment. Since illiterate landless households
are unlikely to be working in the literate labor market and skilled manufacturing
growth has been f o u n d to have a small positive effect on unskilled wages, the positive education response in these households is likely to reflect a positive returns
to education response.
Panels A and B of table 21 examine the sensitivity of the results to using a linear
probability model.
3.5.0.3. Robustness Checks. If the supply side of education responds to growth
in incomes or returns, the estimated coefficients may be capturing a this supply
side response. For example, Foster and Rosenzweig (2004) find that new schools
are allocated to areas in which agricultural technological change is expected to be
greatest.
To ascertain whether these alternative mechanisms are driving the results, table 22 displays the response of a n u m b e r of indicators of educational supply are
correlated with agricultural productivity and the excluded instruments. In the absence of information on school placement, I use data on the n u m b e r of teachers
in a district, the ratio of teachers to pupils and the average distance to the nearest
primary school as a measure of educational infrastructure (panel A, columns a-c).
To capture changes in costs, I examine primary school tuition fees within a district (panel A, columns d and e) as well as the proportion of primary school aged
children receiving tuition subsidies, free m i d d a y meals, books and travel subsidies
and scholarships.
The results indicate that there may be some case to be m a d e for a supply side
response to the changes in manufacturing employment driven by the excluded
instruments, but the response to changes in agricultural productivity appears to be
102
minimal. In order to ensure that m y results are not driven by changes in the supply
of education infrastructure or education costs, I include the n u m b e r of teachers,
average distance from schools, the proportion of children receiving a scholarship,
m i d d a y meals and books as explanatory variables in the education regressions.
The m a g n i t u d e and statistical significance of the coefficient estimates are robust to
this inclusion and are displayed in panel C, table 21.
3.5.0.4. Results - Income and Returns to Education Effects. The parameters estim a t e d capture combinations of the structural parameters, which are estimated using optimal m i n i m u m distance. The results are presented in table 20; columns
(a) through (d) present the results for boys, while columns (e) through (h) present
those for girls. In (a) and (b) the opportunity costs are constrained to zero, in (c)
and (d) I test whether they are indeed zero. The probability a child starts school
increases in household incomes. A 10% increase in income raises the probability
that children start school by 0.11 and by 0.09 in illiterate and literate households
respectively. The difference between the two effects is not statistically significant.
The positive income effect may represent household credit constraints or that education enters directly into household utility. I am unable to distinguish between
these two channels.
D e m a n d for education raises in the returns to education - a 10% rise in returns
to education increases the probability of attending school by 0.05 and 0.04 in illiterate and literate households. In other words, it pulls 5 in 100 children into school.
The relative m a g n i t u d e of the estimated coefficients indicates that education response to rising income are greater than those to returns to education. In comparison, the returns response for girls is small and insignificant. This m a y reflect
the absence of female labor d e m a n d in skilled occupations in rural areas (PROBE,
1999). Females are also more likely to marry outside of the district (Rosenzweig
and Stark, 1989), therefore females may also respond to a different geographic
103
return to education than boys. A large literature also suggests that female educational attainment is responsive to returns to education on the marriage market,
therefore it may be the case that m y model of educational attainment based on
labor market returns alone is incorrect for females.
In columns (c) and (d) of table 20, I test whether the opportunity cost effect
is different from zero. I find that the opportunity cost for both boys a n d girls is
not statistically different f r o m zero. This is likely to reflect the observation that
children w h o aren't at school also aren't working.
3.5.1. Interpretation of Results

The proportion of boys starting school in illiterate landless households has risen
by 34 percentage points from 0.45 in 1983 to 0.79 in 1999. Unskilled wages have increased by 54%, while the average ratio of skilled to unskilled wages in rural area
has decreased by 14%. The estimated coefficients therefore suggest that the rise
in income during this period raises the proportion of children in these households
by 0.52, while the reduction in the returns to education w o u l d reduce it by 0.07.
Overall, the estimates suggest a rise in the proportion of children starting school of
0.45 over the period. A large part of this effect can be attributed to rising agricultural productivity. The estimates suggest that educational investment w a s raised
by 0.21 through this channel, while rising unskilled manufacturing e m p l o y m e n t
has raised investment by 0.07 percentage points.
The education increase a m o n g girls in this household g r o u p is even more startling: the proportion of girls in illiterate landless households attending school has
increased from 0.26 in 1987 to 0.72 in 1999. The estimates suggest that growth in
incomes raised the proportion of girls starting school by 0.43. The majority of this
effect comes through rising agricultural productivity.
The implied income elasticities lie between 2 and 3 and are larger than those
found in the literature. In a review of the literature, Behrman and Knowles (1999)
104
find that the estimated income elasticity lies between 0.1 and 0.3. Estimates focusing on lower income samples are found at the top end of the income elasticity band.
However the income elasticities in the previous literature focus predominantly on
the decision to obtain a year of additional schooling or not. The education margin
considered in this paper is whether to send a child to school or not. The 1995 education module from the NSSO gives a sense of the number of years of primary
schooling attained by the 1983 cohort of 5 to 9 year olds, conditional u p o n having
enrolled in primary school. Children in this cohort attain on average 5.69 years
of schooling, conditional u p o n enrolling in primary school, with less than 13% of
children attaining 3 years or less of education. Therefore converting the elasticities
estimated in this paper into a figure based on years of schooling, the estimates imply income elasticity of 0.37 for boys and 0.54 for girls. These figures are still on the
high side for the literature, but since they focus on the poorest households are very
comparable in magnitude to those found elsewhere in the literature. Interestingly,
Edmonds et al
(2005)
find similarly high poverty-elasticity estimates of
0.7
for a
similar time frame in India.
3.6.
CONCLUSION
This paper estimates the effect of rising income and returnt to schooling on educational investment between 1983 and 1999. Male educational investment responds
positively to growth in incomes and returns to education. The estimates suggest
that rural income growth has raised the demand for education among primary
school aged boys by 23 percentage points over the period, while decreases in rural
returns to education have worked the other way. In contrast to male educational
investment, female investment only responds to rising incomes.
may lift all boats. Regardless of the source, growth increases educational investment at the bottom end of the income spectrum in rural India. Sources of growth
105
that raise the unskilled wage but have little impact on the skilled wage, such as
agricultural technological change and unskilled manufacturing growth, increase
educational investment substantially through the income channel. At the same
time, they increase household welfare by raising income and reducing poverty.
Skill biased manufacturing growth, while having a smaller impact on the levels of
current incomes, raises the educational investment of landless households through
rising returns to education.
3. DETERMINANTS OF EDUCATIONAL INVESTMENT 117

T a b l e 1 8 - a : E d u c a t i o n a l E n r o l l m e n t a n d Child Labor, C h i l d e n A g e d 5 - 9
Boys
Girls
Work
Domestic
S t a r t e d School
Work
Domestic
0.03
0.00
0.00
0.00
0.00
0.71
0.03
0.04
0.48
0.81
0.03
0.02
0.61
0.84
0.02
0.02
0.73
0.91
0.01
0.02
0.84
0.03
0.01
0.01
Table 18-b: Correlation b e t w e e n Starting Primary School, W a g e Levels a n d Returns in 1 9 8 7

P a n e l A: B o y s a g e d 5 - 9
Returns
Proportion
Unskilled
Skilled W a g e
(Skilled/
Wage
(Literate)
Unskilled)
P r o p o r t i o n S t a r t e d P r i m a r y School
Unskilled W a g e
0.589***
Skilled W a g e ( L i t e r a t e )
0.506***
Returns (Skilled/Unskilled Wage)
0.250***
0.498***
-0.318***
Proportion
Unskilled
Skilled W a g e
(Skilled/
Wage
(Literate)
Unskilled)
1
0.428*
P a n e l A: Girls a g e d 5 - 9
Returns
P r o p o r t i o n S t a r t e d P r i m a r y School
Unskilled W a g e
0.683***
Skilled W a g e ( L i t e r a t e )
0.546***
0.498***
Returns (Skilled/Unskilled Wage)
0.311***
-0.318***
0.428***
107

Table 19: Started School Between the Age of 5 and 9
Specification: Probit
Panel A: Dependent Variable: 1 if boy reports having started school
Landless
Landed
Illiterate
Literate
Net Exporters
(a)
(b)
(c-L)
(c-H)
0.251***
0.073**
-0.178***
-0.028
(0.095)
(0.031)
(0.051)
(0.047)
-0.146**
0.212***
0.050
0.076
0.071
(0.049)
(0.032)
(0.046)
0.418**
0.124**
0.234***
0.251***
(0.177)
(0.051)
(0.074)
(0.043)
State*Year
Log(Manufacturing
Employment)
Log(Manufacturing Literate)
Log{Agricultural
Productivity)
Dummies
Adjusted
R-Squared
Observations
Net Importers
0.144
0.128
0.136
0.202
14207
12524
24421
41315
Panel B: Dependent Variable: 1 if girl reports having started school

Landless
Illiterate
Log(Manufacturing
Employment)
Log(Manufacturing
Literate)
Landed
Literate
Net Exporters
Net Importers
(a)
(b)
(c-L)
(c-H)
0.084*
0.237***
0.289***
-0.093**
(0.041)
(0.051)
(0.048)
(0.082)
-0.029
-0.214***
-0.259***
0.085
(0.047)
(0.045)
(0.055)
(0.041)
0.252*
0.143
0.379***
0.276***
(0.152)
(0.091)
(0.127)
(0.078)
State*Year
Log(Agricultural
Productivity)
Dummies
Adjusted
R-Squared
Observations
0.223
0.223
0.257
0.267
13849
12865
20983
35845
Panel C: Specification - Instrumental Variables

Dependent Variable: 1 if individual reports having started school
Boys Aged 5 to 9
Girls Aged 5 to 9
Illiterate
Literate
Illiterate
(a)
(b)
(c)
(d)
0.534*
0.429*
0.254
0.269
(0.306)
(0.236)
(0.332)
(0.312)
0.263*
0.287**
0.126
0.536
(0.143)
(0.133)
(0.270)
(0.407)
State*Year
Log(Unskilled
Wages)
LogfSkilled W a g e s )
Dummies
Literate
All specifications include district fixed effects, year dummies, region time trends, rainfall variables (total rainfall between june
and September, rainfall squared and a shock measure taking the value 1 if there is a positive rainfall shock, 0 if no shock and - 1
for negative shocks), and the log of male and female population for the landless and by land quantiles. Household controls are
the sex of the household head, his age, age squared and social group. The first stage instruments are the average interaction
between between district level resources, industry resource usage and industry tariffs and regulations. The F-Stats on the first
stage are between 13.4 and 16.1. * significant at 10%; ** significant at 5%; " " significant at 1%; Huber-White standard errors
are reported in parentheses, standard errors are clustered at a region-year level in panels A and 13 and at a district level in panel
C.
108

Table 20; Structural P a r a m e t e r s of Education Decision Rule for children aged 5 to 9
Panel A: Setting Opportunity Cost Equal To Zero, Tau=0.4 (Proportion of Unskilled Time)
Boys
Girls
Indirect Least
Squares
Income
(Kappa 1)
Returns to Education
(Kappa 2)
Xi-Squared
Degrees of Freedom
Tau
Illiterate
(a)
1.079**
(0.416)
0.497**
(0.213)
1.18
1
-
Literate
fb)
0.904***
(0.202)
0.442**
(0.134)
2.059
1
0.4
Indirect Least
Squares
Illiterate
(c)
0.797
(0.539)
0.263
(0.182)
Literate
(d)
0.501**
(0.255)
0.279*
(0.143)
0
0.4
Panel B: T a u = 0 . 6 / 0 . 8 ; Testing if Opportunity Cost=Zero,

Boys
Girls
Literate
Literate
Literate
Literate
(c)
(al
fd)
fb)
0.919*** 0.920*** 1 079*** 1.079***
Income
(0.346)
(0.203)
(0.204)
(0.346)
(Kappa 1)
0.226
0.442
Returns to Education
0.266*** 0.634***
(0.321)
(0.100)
(0.168)
(0.265)
(Kappa 2)
Opportunity Cost
(Kappa 4)
0.583
0.583
Xi-Squared
2.061
2.075
1
1
1
1
Degrees of Freedom
0.8
0.6
0.8
Tau
0.6
Illiterate
Literate
Illiterate
(e)
0.791**
(0.341)
0.220
(0.175)
3.114
1
(g)
0.374
(0.572)
0.120
(0.408)
1.076***
(0.338)
-0.016
(0.199)
13.827
1
0.4
Boys
Literate
Literate
(e)
0.935***
(0.197)
(f)
0.879*
(0.495)
0 497***
(0.213)
-0.225
(0.394)
2.896
2
0.4
2
0.4
Literate
fh)
0.500
(0.446)
0.348
(0.216)
0
0.4
Girls
Literate
Literate
(hi
(g)
0.791**
0.534
(0.456)
(0.341)
0.2202
(0.178)
0.2572
(0.533)
7.272
2
2
0.4
0.4
Table c o n t a i n s m i n i m u m distance estimates of the s t r u c t u r a l p a r a m e t e r s , based o n a u x i l i a r y p a r a m e t e r s are f o u n d in table 10. In

c o l u m n s (e), (g), (f) a n d (h) of panel B, the o p p o r t u n i t y cost of e d u c a t i o n has been set to zero. T a u indicates the p r o p o r t i o n of t i m e spent
by s k i l l e d h o u s e h o l d s i n the u n s k i l l e d l a b o r m a r k e t .

Table 21: Robustness Checks, Education
Specification: Linear Probability Model
P a n e l A: D e p e n d e n t V a r i a b l e : 1 if b o y r e p o r t s h a v i n g s t a r t e d s c h o o l
Landless
Landed
Illiterate
Literate
Net Exporters
fa)
(b)
(c-L)
(c-H)
0.167***
0.042**
-0.091***
-0.126*
(0.057)
(0.021)
0.039
(0.075)
-0.067*
0.136***
0.095
0.028
(0.037)
(0.049)
(0.067)
(0.040)
Net Importers
0.343
0.207**
0.261
0.271
(0.553)
(0.100)
(0.231)
(0.486)
State*Year D u m m i e s
Adjusted R-Squared
0.178
0.132
0.179
0.2063
Observations
14207
12524
24421
41315
P a n e l B: D e p e n d e n t V a r i a b l e : 1 if g i r l r e p o r t s h a v i n g s t a r t e d s c h o o l
Landless
Landed
Illiterate
Literate
Net E x p o r t e r s
(a)
(b)
(c-L)
(c-H)
0.182**
0.189**
0.154*
-0.328***
(0.080)
(0.092)
(0.089)
(0.086)
-0.108*
-0.019
-0.186***
0.162***
(0.061)
(0.092)
(0.059)
(0.595)
0.236
0.201
0.589*
0.567*
(0.173)
(0.134)
(0.329)
(0.318)
Adjusted R-Squared
0.242
0.194
0.273
0.292
Observations
13849
12865
20983
35845
Net I m p o r t e r s
Specification: Probit; Include Education Infrastructure Controls a n d T i m e T r e n d s Interacted w i t h

Initial C h a r a c t e r i s t i c s
P a n e l C: D e p e n d e n t V a r i a b l e : 1 if b o y o r g i r l r e p o r t s h a v i n g s t a r t e d s c h o o l
Landless
Girls
Boys
Illiterate
Literate
Illiterate
(b)
(c)
(d)
(a)
0.279***
0.057**
0.093*
0.258***
(0.102)
(0.029)
(0.048)
(0.046)
-0.197***
0.172***
-0.036
-0.221***
(0.043)
Log(Agricultural
Productivity)
Literate
(0.076)
(0.043)
0.045
0.327*
0.125**
0.251*
0.127
(0.175)
(0.063)
(0.134)
(0.088)
Time Trends*lnitial
Literacy/Primary
Adjusted R-Squared
0.137
0.127
0.194
0.211
Observations
14207
12524
13849
12865
* significant at 10%; ** significant at 5%; *** significant at 1%

AH specifications includc district fixed effects, year dummies, rainfall variables (total rainfall between junc and September,
rainfall squared and a shock measure taking the value 1 if there is a positive rainfall shock, 0 if no shock and -1 for negative
shocks), the log of male and female population lor the landless and by land quantiles. In addition, they include the interaction
of district level resources with average tariff and delicensing reforms. Finally they include the sex of the household head, his
age, age squared and social group.

T a b l e 2 2 : R o b u s t n e s s Checks, E d u c a t i o n
P a n e l A: D e p e n d e n t Varibles: M e a s u r e s of E d u c a t i o n Cost a n d I n f r a s t r u c t u r e
Average
Average
N u m b e r of
Distance
Cost o f
Total Cost
from School
Teachers
Transport
Per Student
T a r i f P F o r e s t Cover
TarifPCeramics
TarifPConstruction
TarifPCoal
"Energy Proportion
TarifPElectricity
"Energy Proportion
Agricultural Productivity
Constant
Observations
F-Stat o v e r i n s t r u m e n t s a n d
a g r i c u l t u r a l index
Average
Tuition Per
Student
(a)
3.749
(b)
-0.016
(17.364)
0.008
(0.031)
-0.032*
(0.009)
0.004*
(0.018)
-0.013***
(0.004)
-0.010
(0.008)
0.001
(0.002)
-0.008
(0.010)
-0.658*"
(0.272)
0.710
0.004
(0.003)
0.473**"
(0.176)
0.951
(0.008)
-0.010
(0.021)
0.394
(6.813)
2.088
(0.437)
-13.158**
(6.365)
923
(0.949)
6.169
(14.942)
653
(0.902)
-22.966*
(12.377)
653
(2.289)
-60.489"
(34.886)
653
(1.276)
-26.931
(23.709)
653
2.67**
3.31***
1.63
0.55
0.94
(0.002)
0.017
(0.012)
-31.547*
(17.722)
0.499
(c)
0.004
(0.014)
(d)
-0.643
(1.012)
-0.019
(0.030)
-0.001
(e)
0.051
(0.037)
-0.026
(0.019)
0.002
(0.004)
0.000
(0.010)
0.689
(0.519)
0.941
P a n e l B: D e p e n d e n t Varibles: P r o p o r t i o n of P o p u l a t i o n R e c e i v i n g E d u c a t i o n S u b s i d i e s
F r e e T u i t i o n M i d d a y Meal
T a r i f P F o r e s t Cover
"Wood Proportion
TarifPCeramics
TarifPConstruction
TarifPCoal
"Energy Proportion
TarifPElectricity
Agricultural Index
Constant
Observations
F-Stat over i n s t r u m e n t s a n d
agriculutural index
(a)
-0.010
(0.008)
0.002
(0.008)
0.002"
(0.001)
-0.167
(0.115)
0.006
(0.004)
-0.117
(0.343)
5.671
(5.665)
653
1.77
Books
Transport
Scholarship
(b)
-0.041***
(0.011)
0.016
(0.014)
0.000
(0.003)
0.215
(0.176)
0.003
(0.010)
0.251
(0.490)
(c)
-0.008
(0.011)
-0.040***
(0.013)
-0.004**
(0.002)
0.371***
(0.136)
0.019***
(0.005)
-0.206
(0.565)
(d)
-0.029
(0.027)
0.035
(0.023)
-0.011"
(0.006)
-0.534*
(0.309)
-0.068**
(0.030)
-1.651
(1.060)
-8.13
(7.286)
653
-6.178
(7.454)
653
33.101"
(16.966)
653
(e)
0.013***
(0.004)
0.007
(0.005)
0.000
(0.001)
-0.175***
(0.056)
0.007***
(0.002)
-0.204
(0.244)
5.774"
(3.256)
653
3 42***
6.73***
1.76
6.98***
* s i g n i f i c a n t a t 1 0 % ; ** s i g n i f i c a n t a t 5 % ; *** s i g n i f i c a n t a t 1 %
All s p e c i f i c a t i o n s i n c l u d e d i s t r i c t fixed e f f e c t s , y e a r d u m m i e s , r a i n f a l l v a r i a b l e s ( t o t a l r a i n f a l l b e t w e e n j u n e a n d S e p t e m b e r ,
r a i n f a l l s q u a r e d a n d a s h o c k m e a s u r e t a k i n g t h e v a l u e 1 if t h e r e is a p o s i t i v e r a i n f a l l s h o c k , 0 if n o s h o c k a n d - 1 f o r n e g a t i v e
s h o c k s ) , t h e log of m a l e a n d f e m a l e p o p u l a t i o n f o r t h e l a n d l e s s a n d b y l a n d q u a n t i l e s . In a d d i t i o n , t h e y i n c l u d e t h e i n t e r a c t i o n of
d i s t r i c t level r e s o u r c e s w i t h a v e r a g e t a r i f f a n d d e l i c e n s i n g r e f o r m s .
Bibliography
[1] A d u l a v i d h y a , K., Kurida, Y., Lau, L., Lerttamrab, P. a n d "Yolopoulos, P. (1979) "A microecon o m i c analysis of the agriculture of T h a i l a n d " , Food Research Institute Studies, 17:79-86.
[2] Aghion, Philippe, Burgess, Robin, R e d d i n g , Stephen, a n d Zilibotti, Fabrizio, (2009) "The U n equal Effects of Liberalization: Evidence from D i s m a n t l i n g the License Raj in India", The American Economic Review, 98(4), p p . 1397-141
[3] A h l u w a l i a , M.S. (1975) "Income Distribution a n d D e v e l o p m e n t : S o m e Stylized Facts", The
American Economic
Review
[4] A h l u w a l i a , M.S. (2002) "State-Level P e r f o r m a n c e u n d e r Economic R e f o r m s in India", in A n n e

O. Krueger (ED.) Economic Policy R e f o r m s a n d the Indian Economy, N e w Delhi: Oxford University Press.
[5] Altonji, Joseph G., Todd E. Elder a n d C h r i s t o p h e r R. Taber, 2005. "Selection on O b s e r v e d a n d
U n o b s e r v e d Variables: Assessing the Effectiveness of Catholic Schools," Journal of Political
Economy, University of Chicago Press, vol. 113(1), pages 151-184, February.
[6] Atkin, D a v i d (2009) "Trade, Tastes a n d Nutrition in India" Working Paper
[7] Autor, David, Levy, a n d M u r n a n e (2003) "The Skill C o n t e n t of Recent Technological C h a n g e :
An Empirical Exploration", (2003) Quarterly Journal of Economics, 118(4), N o v e m b e r 2003,12791334.
[8] Baland, Jean-Marie a n d James A. Robinson, (2000) "Is Child Labor Inefficient?," Journal of
Political Economy, University of Chicago Press, vol. 108(4), p a g e s 663-679, A u g u s t .
[9] Banerjee, Abhijit (2004) " E d u c a t i o n Policy and the Economics of the Family" Journal of Development Economics, 74(1), N e w Research on Education in D e v e l o p i n g Economies, p p 3-32
[10] Banerjee, Abhijit a n d A n d r e w N e w m a n (1993) "Poverty a n d well-being in d e v e l o p i n g c o u n tries: poverty, incentives and d e v e l o p m e n t " Amiercan Economic Review, Papers and
Proceedings,
p p 211-215, May.
[11] Banerjee, Abhijit a n d Esther Duflo (2005) " G r o w t h T h r o u g h the Lens of D e v e l o p m e n t Econ o m i c s " Handbook of Development Economics, 1(a) pp. 473-552
111
3. BIBLIOGRAPHY
[12] B a r d h a n , P r a n a b K, 1979. "Labor S u p p l y Functions in aPoor A g r a r i a n E c o n o m y / '
112
American
Economic Review, A m e r i c a n Economic Association, vol. 69(1), p a g e s 73-83, March.

[13] B a r n u m , H . N . a n d Squire, L. (1978) " A n econometric application of the theory of the f a r m
h o u s e h o l d " , journal of Development
Economics, 6:79-102.
[14] Becker, G a r y (1964) Human Capital, N e w York: National Bureau of Economic Research
[15] Becker, G a r y S. a n d Nigel Tomes (1986) " H u m a n Capital a n d the Rise a n d Fall of Families"
Journal of Labor Economics, Vol. 4(3), p p . S1-S39.
[16] Beegle, Kathleen a n d Dehejia, Rajeev H. a n d Gatti, Roberta, 2006. "Child labor a n d agricultural
shocks," Journal of Development Economics, Elsevier, vol. 81(1), p a g e s 80-96, October.
[17] B e h r m a n , Jere B. a n d James C. K n o w l e s (1999) " H o u s e h o l d Income a n d Child Schooling in
V i e t n a m " The World Bank Economic Review, 13(2) p p . 211-56
[18] Besley, Timothy a n d Robin Burgess, (2004) "Can Labour Regulation H i n d e r Economic Perform a n c e ? Evidence f r o m India," The Quarterly Journal of Economics, MIT Press, vol. 119(1), p a g e s
91-134, February.
[19] B h a r a d w a j , P r a s h a n t (2009) "Fertility a n d Rural Labor M a r k e t Inefficiencies: Evidence f r o m
I n d i a " Working Paper
[20] Bliss, C. a n d N. Stern (1978) "Productivity, Wages a n d N u t r i t i o n , " Journal ofevelopent
Economics
Vol(5), 331-3.
[21] Bowles, Samuel (1972) "Schooling and Inequality from Generation to Generation," Journal of
Political Economy, ( M a y / J u n e 1972):S219-S251.
[22] Card David a n d John E. D i N a r d o (2002) "Skill-Biased Technological C h a n g e a n d Rising Wage
Inequality: Some P r o b l e m s a n d Puzzles," Journal of Labor Economics, University of Chicago
Press, vol. 20(4), p a g e s 733-783, October.
[23] Chari, A.V. (2009) "The aggregate p r o d u c t i v i t y effects of entry a n d o u t p u t restrictions: an
analysis of license r e f o r m in I n d i a " Working Paper
[24] Currie, Janet a n d Enrico Moretti " M o t h e r ' s E d u c a t i o n and the Intergenerational Transmission
of H u m a n Capital: Evidence f r o m College O p e n i n g s " (2003) The Quarterly Journal of Economics,
Vol. 118, N o . 4, p p . 1495-1532
[25] Datt, G a u r a v a n d Martin Ravallion (2002) "Is I n d i a ' s Economic G r o w t h Leaving the Poor Behind?," Journal of Economic Perspectives, 16(3), p p 89-108
[26] Deaton, A n g u s (2003) " A d j u s t e d Indian Poverty Estimates for 1999-2000", Economic and Political Weekly, 38(4) pp. 322-326
3. BIBLIOGRAPHY
113
[27] Deaton, A n g u s a n d Jean Dreze a n d "Poverty a n d Inequality in India: A R e - E x a m i n a t i o n "

Economic and Political Weekly, 37(36), p p . 3729-3748
[28] Deaton, A n g u s a n d A l e s s a n d r o Tarozzi (2005) "Prices a n d Poverty in India," C h a p t e r 16 in
Deaton a n d Kozel, Data and Dogma: The Great Indian Poverty Debate, N e w Delhi, M a c m i l l a n
India.
[29] Dreze, Jean a n d Geeta K i n g d o n (2001) "School Participation in Rural I n d i a " Review of Development Economics 5(1), p p 1-24
[30] Dreze Jean a n d A m a r t y a K. Sen, (1995) India: Economic Development
and Social
Opportunity.
O x f o r d University Press.
[31] Duflo, Esther "Schooling a n d Labor M a r k e t C o n s e q u e n c e s of School Construction in I n d o n e sia: Evidence f r o m an U n u s u a l Policy E x p e r i m e n t " American Economic Review 91(4) p p 7 9 5
[32] E d m o n d s , Eric (2006) "Child Labor a n d Schooling Responses to Anticipated Income in South
Africa", Journal of Development
Economics, D e c e m b e r 2006, 81(2), 386-414
[33] E d m o n d s Eric V. a n d N o r b e r t Schady (2009) "Poverty Alleviation a n d Child Labor,"
NBER
Working Papers 15345, N a t i o n a l Bureau of Economic Research

[34] E d m o n d s , Eric, N i n a Pavnick a n d Petia Topalova (2007) "Trade A d j u s t m e n t a n d H u m a n C a p ital Investments: Evidence f r o m Indian Tariff Reform," NBER Working P a p e r s 12884
[35] E d m o n d s , Eric a n d C. Turk "Child Labor in Transition in Vietnam" in P. G l e w w e , N. A g r a w a l
a n d D. Dollar (eds), Economic Growth, Poverty and Household Welfare in Vietnam.
Washington
DC: World Bank, 2004, p p . 505-550

[36] Ellison, Glen a n d E d w a r d Glaeser (1997) " G e o g r a p h i c Concentrateion in U.S. m a n u f a c t u r i n g
industries: A d a r t b o a r d a p p r o a c h " Journal of Political Economy 105(5), 889-927
[37] Ellison, G l e n n a n d E d w a r d L. Glaeser (1999) "The G e o g r a p h i c Concentration of I n d u s t r y : D o e s
N a t u r a l A d v a n t a g e Explain Agglomeration?," American Economic Review, A m e r i c a n Economic
Association, vol. 89(2), p a g e s 311-316, May.
[38] E s w a r a n et al., M. E s w a r a n , A. Kotwal, B. R a m a s w a m i and W. W a d h w a , (2009) "Sectoral
labour flows a n d agricultural w a g e s in India, 1983U2004: has g r o w t h trickled d o w n ? " , Economic and Political Weekly 44 (2) (2009), p p . 46U55.
[ev99] venson, Robert E., Carl E. Pray, a n d M a r k W. Rosegrant (1999) "Agricultural research a n d
productivity g r o w t h in India," International Food Policy Research Institute Series 109
[39] Foster, A n d r e w D. a n d M a r k R o s e n z w e i g (1993) "Information, Learning a n d Wage Rates in
Low-Income Rural A r e a s " Journal of Human Resources, Vol 28(4) pp759-790
3. BIBLIOGRAPHY
114
[40] Foster, A n d r e w D. a n d M a r k R o s e n z w e i g (1995) "Learning b y d o i n g a n d learning f r o m others:

H u m a n capital a n d technical c h a n g e in agriculture" Journal of Political Economy 103(6), 11761209
[41] Foster, A n d r e w D. a n d M a r k R o s e n z w e i g (1996) "Technical C h a n g e a n d H u m a n - C a p i t a l Ret u r n s a n d Investments: Evidence f r o m the Green Revolution," American
Economic
Review,
A m e r i c a n Economic Association, vol. 86(4), p a g e s 931-53, September.

[42] Foster, A n d r e w D. a n d M a r k R o s e n z w e i g (2004a) "Agricultural D e v e l o p m e n t , Industrialization a n d Rural Inequality" Working Paper
[43] Foster, A n d r e w D. a n d M a r k R o s e n z w e i g (2004b) "Technological c h a n g e a n d the distribution
of schooling: evidence f r o m the green revolution in I n d i a " Journal of Dvelopment
Economics,
74(1), N e w Research on Education in D e v e l o p i n g Economies, pp.87-111

[44] G l e w w e , Paul a n d Jacoby, H a n a n G. (2004) "Economic g r o w t h a n d the d e m a n d for education:
is there a w e a l t h effect?," Journal of Development
Economics, Elsevier, vol. 74(1), p a g e s 33-51,
June.
[45] Goldberg, Penny, A m i t K h a n d e w a l , N i n a Pavnick a n d Petia Topalova (2009) "Trade Liberalization a n d N e w I m p o r t e d I n p u t s / ' f o r t h c o m i n g , American Economic Review Papers and Proceedings
[46] G o v e r n m e n t of India (1968) National Classification of Occupations.
G o v e r n m e n t of India, Min-
istry of Labor.
[47] G o v e r n m e n t of India (1987) Natiottal Industrial Classification. G o v e r n m e n t of India, Central Statistical Organisation, D e p a r t m e n t of Statistics
[48] G o v e r n m e n t of India. 2000. " I n p u t - O u t p u t Transaction Table, 1993-94," Central Statistical Organization, D e p a r t m e n t of Statistics, Ministry of P l a n n i n g a n d P r o g r a m m e I m p l e m e n t a t i o n .
[49] G o v e r n m e n t of India (1995) Mineral Atlas of India G o v e r n m e n t of India, Geological Survey of
India.
[50] G o v e r n m e n t of India (1999) Employment
and Unemployment
situation in India, 1999-2000 Gov-
e r n m e n t of India, National Sample S u r v e y Organization, Report N o . 458

[51] G o v e r n m e n t of India (2001) C o n c e p t s a n d Definitions Used in NSS: G o l d e n Jubilee Publication. G o v e r n m e n t of India, Ministry of Statistics and P r o g r a m I m p l e m e n t a t i o n , National
Sample Survey Organisation.
[52] Griliches, Zvi (1980) Technology, Education and Productivity:
York: Basil Blackwell
Essays in Applied Econometrics, N e w
3. BIBLIOGRAPHY
115
[53] Gulati, A s h o k a n d S u d h a N a r a y a n a n (2003) The subsidy syndrome in Indian agriculture"
Oxford
University Press: N e w Delhi

[54] H a s a n , Rana, Devashish Mitra a n d K.V R a m a s w a m y (2007) "Trade Reforms, Labor Regulations, a n d L a b o r - D e m a n d Elasticities: Empirical Evidence f r o m India," The Review of Economics
and Statistics, MIT Press, vol. 89(3), p a g e s 466-481, 02
[55] Jacboy, H a n a n G (1994) " B o r r o w i n g Constraints a n d Progress T h r o u g h School: Evidence f r o m
P e r u , " Review of Economics and Statistics, 76(1) p p 151-60
[56] Jacoby, H a n a n G a n d E m m a n u e l Skoufias (1997) "Risk, Financial Markets, a n d H u m a n Capital
in a D e v e l o p i n g Country," Review of Economic Studies, 64(3), p p 311-35
[57] Jayachandra, Seema (2006) "Selling Labor Low: Wage Responses to Productivity Shocks in
D e v e l o p i n g C o u n t r i e s " Journal of Political Economy, 114(3), p p . 538-575.
[58] Jensen, Robert (2010) "The (Perceived) R e t u r n s to Education a n d the D e m a n d for Schooling,"
Quarterly Journal of Economics, 125(2).
[59] Jessoe, Katrina (2009) " I m p r o v e d Source, I m p r o v e d Quality? D e m a n d for D r i n k i n g Water
Quality in Rural I n d i a " Working Paper
[60] Katz, Lawrence F a n d M u r p h y , Kevin M (1992) " C h a n g e s in Relative Wages, 1963-1987: Supply a n d D e m a n d Factors," The Quarterly Journal of Economics, MIT Press, vol. 107(1), p a g e s
35-78, February.
[61] Keskin, Pinar (2009) "Thirsty Factories, H u n g r y Farmers: Intersectoral I m p a c t s of Industrial
Water D e m a n d "
[62] Kingdon, Geeta (2007) "The progress of school education in I n d i a " Oxford Review of Economic
Policy 23(2) p p 168-195
[63] Kim, S u k k o o (1995), "Expansion of M a r k e t s a n d the G e o g r a p h i c Distribution of Economic
Activity: The t r e n d s in U.S. Regional M a n u f a c t u r i n g Structure, 1860-1987" Quarterly Journal of
Economics 110, 881-908
[64] Kochar, Anjini (2002) " E m e r g i n g Challenges for Indian E d u c a t i o n Policy" in A n n e O. Krueger
(ED.) Economic Policy Reforms a n d the Indian Economy, N e w Delhi: O x f o r d University Press.
[65] Kochar, Anjini (2004) " U r b a n influences on rural schooling in India", Journal of
Development
Economics, 74(1), N e w Research on Education in D e v e l o p i n g Economies, p p 113-136

[66] Kochhar, Kalpana, Utsav Kumar, R a g h u r a m Rajan, A r v i n d S u b r a m a n i a n a n d Ioannis Tokatlidis (2006), "IndiaSs Pattern of D e v e l o p m e n t : W h a t H a p p e n e d , W h a t Follows?" IMF Working
Paper, W P / 0 6 / 2 2
3. BIBLIOGRAPHY
116
[67] K a m b h a m p a t i , U m a , Pravin K r i n s h n a a n d Devashish Mitra "The I m p a c t of Trade Policy Ref o r m s on Labor Markets: Evidence f r o m India," Journal of International
Trade and
Economic
Development
[68] Krishna, Pravin, a n d Devashish Mitra (1998) "Trade Liberalization, M a r k e t Discipline a n d
Productivity G r o w t h : N e w Evidence f r o m India,T Journal of Development
Economics
[69] Krueger, A n n e , a n d Sajjid C h i o n y (2002), "The Indian E c o n o m y in Global C o n t e x t " in A n n e

O. K r u e g e r (ED.) Economic Policy R e f o r m s a n d the Indian Economy, N e w Delhi: O x f o r d University Press.
[70] K r u g m a n , Paul, (1991), Geography and Trade, C a m b r i d g e , Mass: MIT Press.
[71] Kuznets, Simon (1955) "Economic G r o w t h a n d Income Inequality," American Economic
Review
65(1), p p . 1-28. See K u z n e t s Curve.

[72] Lanjouw, Peter a n d Rinku M u r g a i (2009), "Poverty Decline, Agricultural Wages a n d N o n F a r m e m p l o y m e n t " Agricultural
Economics Vol 40(2) pp243-263
[73] Lazear, E d w a r d (1980) "Family Background a n d O p t i m a l Schooling Decisions," The Review of

Economics and Statistics, Vol. 62(1), p p . 42-51
[74] Leibenstein, H., (1957) Economic Backwardness and Economic Growth N e w York: Wiley
[75] Lochner, Lance a n d Enrico Moretti (2004) "The Effect of Education on Crime: Evidence f r o m
Prison Inmates, Arrests, a n d Self-Reports," American
Economic Review, A m e r i c a n Economic
Association, vol. 94(1), p a g e s 155-189, March

[76] M e h t a B. (1996) "'Free' P r i m a r y E d u c a t i o n ' " Economic and Political Weekly 31(16), p p 1035-1036
[77] Mehta, R. (1999) "Tariff a n d Non-tariff Barriers of Indian Economy: A Profile," Research a n d
Information System for the non-aligned a n d O t h e r D e v e l o p i n g C o u n t r i e s (RIS), N e w Delhi.
[78] Mirrlees, J. (1975) "A P u r e Theory of U n d e r d e v e l o p e d Economies," in Agriculture
in Develoo-
ment Theory, ed. L.A. Reynolds, N e w H a v e n : Yale University Press

[79] M u n s h i , Kaivan a n d M a r k R o s e n z w e i g (2005), " W h y is Mobility in India so Low?
Social
Insurance, Inequality, a n d G r o w t h , " BREAD Working P a p e r No. 092.

[80] N C A E R Publication, (1969) "Techno-Economic Survey of A n d h r a P r a d e s h " a n d other states;
National Council of A p p l i e d Economic Research: N e w Delhi
[81] Nelson, R. and Phelps, E., 1966. " I n v e s t m e n t in h u m a n s , technological diffusion, a n d e c o n o m i c
g r o w t h " American Economic Review: Papers and Proceedings 51(2), p p . 69-75
[82] Panagariya, A r v i n d (2008) India: The Emerging Giant Oxford University Press: USA
[83] Pavcnik, N. (2002) "Trade Liberalization, Exit a n d Productivity I m p r o v e m e n t s : Evidence f r o m
Chilean Plans," Review of Economic Studies, Vol. 69(238).
3. BIBLIOGRAPHY
117
[84] PROBE Team (1999), Public Report on Basic Education in India ( N e w Delhi: O x f o r d University
Press).
[85] Ranis, G u s t a v a n d John Fei "A Theory of Economic D e v e l o p m e n t " The American
Economic
Review 51(4), p p . 533-565

[86] Rosenzweig, M a r k (1980) "Neoclassical Theory a n d the O p t i m i z i n g Peasant: A n Econometric
Analysis of M a r k e t Family Labor S u p p l y in a D e v e l o p i n g C o u n t r y " The Quarterly Journal of
Economics, Vol. 94(1), p p . 31-55.
[87] Rosenzweig, M.R. (1988). "Labor Markets in L o w Income C o u n t r i e s , " in H a n d b o o k of Develo p m e n t Economics, H. C h e n e r y a n d T.N. Srinivasan, Editors, N o r t h H o l l a n d , Vol. pg. 713-761.
[88] Rosenzweig, M a r k (1995), " W h y Are There R e t u r n s to Schooling?", American Economic
Review
(Papers and Proceedings), Vol 85(2), p p . 153-158

[89] Schultz, T.W. (1975) "The Value of the Ability to Deal w i t h Disequilbria" Journal of Economic
Literature 13(1) pp.827-46
[90] Schultz, Paul (2004) "School subsidies for the poor: e v a l u a t i n g the Mexican Progresa p o v e r t y
p r o g r a m , " Journal of Development Economics, 74(1), p p 199-250,
[91] Srinivasan, T.N. (2000), "Eight Lectures of India's Economic R e f o r m s " , O x f o r d India Paperbacks, N e w Delhi: O x f o r d University Press.
[92] Srinivasan, T.N. (2002), "The Indian E c o n o m y in Global C o n t e x t " in A n n e O. Krueger (ED.)
Economic Policy R e f o r m s a n d the Indian Economy, N e w Delhi: O x f o r d University Press.
[93] Stiglitz, J.E., "The Efficiency Wage H y p o t h e s i s , S u r p l u s Labor a n d the Distribution of Income
in L.D.C.s" Oxford Economic Papers, Vol 28(2), p p . 185-207.
[94] Topalova, Petia (2004) "Trade Liberalization and Firm Productivity: The Case of I n d i a " IMF
Working P p a e r W P / 0 4 / 2 8 , International M o n e t a r y F u n d
[95] Topalova, Petia (2005) "Trade Liberalization, Poverty, a n d Inequality: Evidence f r o m Indian
Districts," NBER Working P a p e r s 11614, N a t i o n a l Bureau of Economic Research, Inc.
[96] Tybout, ]., J. De Melo, a n d V. Corbo, (1991) "The Effects of Trade R e f o r m s on Scale a n d Technical Efficiency: N e w Evidence f r o m Chile," Journal of International
Economics, Vol. 31.
[97] Wooldridge, Jeffrey (2002) Econometric Analysis of Cross Section and Panel Data MIT Economic
Press.
APPENDIX A
Data and Description of the Environment

A.0.0.1. Employment Data: The Employment-Unemployment surveys conducted
by the National Sample Survey Organization (NSSO) are the main data source for
all employment and education data. I use the "thick"employment-rounds (round
10) conducted in 1983, 1987-88, 1993-94 and 1999-2000. The surveys collect information on approximately 75,000 rural and 45,000 u r b a n households and usually
cover all states in India. Employment by industry is constructed at a region level,
where a regions consist of 4 to 6 neighboring districts within the same agro-climatic
zone.
I use two different education thresholds to define a skilled group. These thresholds are determined by the data: in 1983, approximately 40% of manufacturing
employees were classified as illiterate, and 59% had less than primary education.
The employment surveys ask households and individuals within household to list
their primary and subsidiary usual occupations as well as their primary and subsidiary current occupations. I use primary occupation status to define an individual's sector and restrict the sample to individuals of working age, between 25 and
55. 1 A weekly time-use recall allows me to capture the days of w o r k devoted to
agricultural labor market, o w n farm and manufacturing activities over the course
ry
of the preceding week.

I use employment data from two waves of the Economic Census, a countryw i d e census of all economic activities other than those related to crop production
and plantation. It contains basic data on the principal activity conducted by the
-i
I test the sensitivity of m y results to including all w o r k i n g aged i n d i v i d u a l s over the age of 25.
The m a g n i t u d e a n d sign of the coefficient estimates in the w a g e e q u a t i o n s are largely u n c h a n g e d .
It is not necessary to account for seasonal variation w h e n u s i n g these m e a s u r i n g since s u r v e y i n g
is spread out u n i f o r m l y over the calendar year.
118
A. DATA AND ENVIRONMENT
119
firm and the n u m b e r of family and hired male and female labor employed. 3 Since
it is a census it captures district-industry level data on e m p l o y m e n t for both the
formal and informal sector.
A.0.0.2. Wages. Wage data on agricultural wages come from the Agricultural
Wages in India (AWI) collected by the Directorate of Economics and Statist at the
Ministry of Agriculture. Data are collected on a monthly basis for various agricultural operations, such as ploughing and sowing. Where possible, separate w a g e s
are collected by sex a n d for children. Data are not collected for all districts within
a state and data reporting is uneven across time and districts. The missing observations for male wages however do not appear to be related to any characteristic
of local labor markets b u t rather to failures in the data collection system. 4 Wages
are deflated using the State level Rural Agricultural Labor Price Index published
by the Indian Labor Bureau. Skilled wages are imputed as the m e d i a n of the nonfarm wages of literate non-manufacturing workers, using data on the weekly w a g e
labor income in the NSSO Employment surveys.
A.0.0.3. Education Data. The measure of education used in this p a p e r is whether
a child aged 10-14 is reporter to be illiterate. The NSSO classifies the education of
an individual according to the highest standard that they have obtained. Therefore in the case of an individual w h o has attended school but cannot read or write,
they should be classified as having some formal education (Government of India,
2001). To verify whether this is the case, the 55th round of the NSSO survey conducted in 1999 asks children to report whether they have ever attended school. Of
The Economic C e n s u s is conceived as a f r a m e of n o n - f a r m firms for c o n d u c t i n g detailed f o l l o w - u p

s u r v e y s on the u n o r g a n i z e d sector.
4
T h e actual data collection is left to i n d i v i d u a l states, w h o assign village level officials to collect
the locally c o m m o n c u r r e n t w a g e on a m o n t h l y basis ( H i m a n s h u , 2004). Since there a p p e a r s to
b e n o check or e n f o r c e m e n t of data collection at a state level b y the Directorate of Economics a n d
Statistics, data is regularly missing at a district or even state level.
120
the 33% of illiterate 9 to 14 year olds, 92% h a d never attended school while 3% attended school b u t d r o p p e d out. This provides some reassurance that this measure
is capturing individuals w h o decided not to enroll in school.
A.0.0.4. Income Data.
A.0.0.5. Natural Resources. Raw material e n d o w m e n t s are grouped according
to the main categories of industrial usage following the Mineral Atlas of India and
various NCAER Economic Plans. They are listed below.
Raw Material Groups using definitions from Mineral Atlas of I n d i a / N C A E R

Economic Plans (1960s-1990s):
Forestry : Proportion of District Covered in Forests
Metal: A l u m n i u m , C h r o m i u m , Copper, Iron Ore, Lead, Manganese, Zinc.
Ceramics: Kaolin, Feldspar, Glass and Foundry Sand.
Construction Sector: Calcite, China Clay, Limestone, Sandstone.
"Strategic" Chemicals - Asbestos, Baryte, Dolomite, Fluorite and Limonite.
Energy: Coking and Non-Coking Coal and Electricity Prices
To measure the mineral and metal endowments, I have geocoded the National
Mineral Atlas and Geological M a p of India published by the Geological Survey of
India (GSI). Figure A.4 displays the data in its raw form. I define a district's raw
material e n d o w m e n t using all resources f o u n d within the district boundaries.
Region level data on soil conditions and agricultural yields comes from the India Agriculture and Climate Dataset compiled by Robert E. Evenson and James W.
McKinsey, Jr., using data from the Directorate of Economics and Statistics within
the Indian Ministry of Agriculture. Wood e n d o w m e n t s are captured using the proportion of the district covered by forest, from the Forest Survey of India (1993).
A.0.0.6. Factor Intensity.
Data from the Input-Output Matrix is used to build
measures of the intensity of use of different inputs by industry. The measures

reflect the share of costs accounted for by a given category of inputs. The definition
121
of factor intensity used is given by:

^
.
Cost of Input k
T
Factor Intensity,, =
^
Cpst
I use discrete measures which are designed to capture the broader technological
requirements of an i n d u s t r y The quantile measure separates all non-zero factor
intensities into quantiles, while the d u m m y measure codes an industry as using
a factor if it lies above the median factor intensity for that raw material a m o n g
all non-zero factor intensities. Table A.3 lists the industries which lie in the top
quintile of the factor intensity measure for w o o d , ceramics, chemicals and metal
inputs. While there are a few surprises, the list largely conforms to expectations:
the iron producing and metal extracting industries are metal intensive; the glass,
cement and refractory industries are ceramics minerals intensive while the aluminium processing, chemicals and coke oven industries are energy intensive.
A.0.0.7. Industrial Policy and Regulations.
I use tariff measures and delicensing
reforms compiled by Aghion, Burgess, Redding and Zillibotti (2008). Post Independence Indian Industrial Policy consisted of state directed, centrally planned development strategies. The "Licence Raj" entailed a series of rigid controls over the
establishment, capacity, investment and production of the industrial sector (Srinivasan, 2000). Import-substitution policies, aimed at stimulating growth, gave rise
to a restrictive trade regime with high average and peak nominal tariffs and nontariff barriers. Reforms d u r i n g the 1980s and 1990s represented a major break from
the previous approach (Srinivasan, 2000). The first phase of reforms occurred in
1985, after a political crisis sparked by the assassination of the incumbent p r i m e
minister. A subset of manufacturing industries was removed from the jurisdiction
of the license regime whilst others were allowed more flexibility in their functioning. The government adopted an expansionary fiscal stance leading to large fiscal
deficits, high levels of external debt and falling foreign reserves by 1991. A serious
macroeconomic and balance of p a y m e n t s crisis ensued; in the w a k e of this crisis,
122
systemic structural reforms were introduced (Krueger a n d Chiony, 2002). Reforms

to industrial policy and excise tariffs altered the institutional f r a m e w o r k a n d profitability of private sector firms, irrespective of location. Fiscal reforms initiated by
the central government resulted in substantial reductions in customs duties. Both
average and peak tariffs have been drastically reduced since 1990-91, from an average nominal tariff of 125% to 25% in 1997-98. Excise duties were simplified and
harmonized across products.
A.0.0.8. Description of Indian Industrial Environment.
Post Independence Indian In-
dustrial Policy
The N e h r u v i a n view of economic development, d o m i n a n t in the period ensuing
independence, endorsed the need for economic development driven by state directed economic activity and central planning. India's economic strategy during
this period is articulated in a series of Five Year Plans which laid d o w n the philosophy and strategies for development in the subsequent years. The Second Five
Year plan (1957-1962) laid d o w n the foundations of India industrial policy until the
1990s (Krueger and Chiony, 2002). The "Licence Raj" entailed a series of rigid controls over the establishment, capacity, investment and production of the industrial
sector (Srinivasan, 2000). Certain industries were reserved for public ownership,
whilst others were circumscribed to public-private partnerships. Targets were established for permissable production levels and licences were required to expand
capacity beyond that permitted. Import-substitution policies, aimed at stimulating
growth, gave rise to a restrictive trade regime with high average and peak nominal
tariffs and non-tariff barriers.
Reforms initiated during the 1980s and 1990s represented a major break from
the inward oriented, state directed and public sector driven' approach p u r s u e d
since independence (Srinivasan, 2000). The first phase of reforms aimed at unraveling the regulatory framework occurred in 1985, after a political crisis sparked by
the assassination of the incumbent prime minister, Indira Gandhi, in 1984. During
123
this period, a subset of manufacturing industries w a s removed f r o m the jurisdiction of the license regime whilst others were allowed more flexibility in their functioning. The government a d o p t e d an expansionary fiscal stance leading to large
fiscal deficits, high levels of external debt and falling foreign reserves by 1991. A
serious macroeconomic and balance of p a y m e n t s crisis ensued; in the w a k e of this
crisis, systemic structural reforms were introduced alongside more "traditional"
stabilization measures (Krueger and Chiony, 2002). Reforms to industrial policy, trade and p a y m e n t s regime, excise tariffs and financial sector substantially
decreased the role of the public sector in economic activity a n d altered the institutional framework and profitability of private sector firms. Reforms affected all
firms in that industry or mining sector irrespective of location.
Fiscal reforms initiated by the central government resulted in substantial reductions in customs duties and some reduction and rationalization of union excise
duties (excise duties levied by the central government). Both average and peak
tariffs have been drastically reduced since 1990-91, f r o m an average nominal tariff
of 125% to 25% in 1997-98. Excise duties were simplified and harmonized across
products.
Indian Industrial Location and Expansion - Industrial Policies and Raw Materials:
In the model presented in section 3,1 assume that firms take into account natural resource e n d o w m e n t s and labor costs w h e n making their location decisions.
In this section I argue that this assumption is appropriate prior in the period prior
to 1987 during which Indian industrial policy prior to 1987 restricted the ability of
larger regulated industrial units to choose their locations. Location decisions appear to have been guided by political concerns (such as providing employment to
"backward" areas or targeting balanced regional development (Mookherjee, 1995))
as well as by consideration for costs. However it does appear that access to inputs
were an important condition precondition to the establishment of an industry.
124
The Industrial Programs component of the Fourth Five Year Plans suggests that
access to inputs was one of the primary motivations given for the dispersal of industries across geographic areas. "The present pattern of regional distribution of
industries has been formed by the accessibility of base raw materials. Thus agriculture based industries are mostly concentrated in North and South Bihar, while
industries based on minerals are to be found chiefly in Chotanagpur, with a dispersal of some small-scale industries in South Bihar." (NCAER Bihar, 1969). The
interlinkages between industries appears to also have been taken into consideration, as this excerpt on the production of sulphuric acid indicates: "The location of
these suggested capacities will be linked with the consuming industries to avoid
transportation" (NCAER Andhra Pradesh, 1969). Conditional on accessibility of
resources, the location of final demand was also a component to be considered, as
demonstrated by this excerpt on the establishment of fertilizer factories: "The demand for fertilizers is likely to be large in each State and it is considered justifiable
to contemplate setting u p of suitable capacities in each State to meet the demand
arising within the State, provided of course suitable raw materials and other facilities are there for the production of fertilizers." (NCAER Andhra Pradesh, 1969).
The implications of these decisions for the cement industry can still be readily
seen in the current geographic location of firms. Figure 1 in the appendix shows
the location of cement grade limestone resources and employment in the cement
industry in 1988 across districts in India. There is clearly a high degree of clustering
in the location of these industries near their most bulky input, limestone. 5 Figure
2 plots the location of firms set u p since deregulation of the cement industry in
1991; at this point firms were no longer subscribed by regulatory policy to obtain
a licence to produce. As can be seen, post deregulation cement firms continue to
locate in the proximity of limestone deposits, indicating that it is unlikely to be the
% i n c e the i n d u s t r y c o d e relating to cement p r o d u c t i o n includes several p r o d u c t s w h i c h use limestone to v a r y i n g d e g r e e s (for example, asbestos cement relies m o r e heavily o n asbestos resources
than on limestone), it is u n s u r p r i s i n g l y to see outliers.
125
case that industrial policy alone prioritized proximity to inputs. 6 Clearly access
to limestone resources continues to be a motivating force behind the placement of
plants. Similar scenarios can be depicted for other major regulated industries.
Other aspects of the Indian institutional environment and infrastructure provide further justification for w h y firms may consider locating in the vicinity of
important inputs. Evidence indicates that industries are clustered around important natural resources in the US. One explanation provided for locating near important resources is to diminish transportation costs. In the Indian context, where
transportation costs are substantially higher than in the US, that the fraction of agglomeration that may be attributed to natural advantages may be even higher. In
addition to the costs of moving goods across areas, costs of obtaining commodities
from a different states are markets for all commodities are not integrated across
India. A good example is the market for agricultural commodities, in which restrictions on the movement of some important commodities across states increase
the cost of the inter-state transportation of goods. In Maharashtra, farmers' sales
of cotton to neighboring states is restricted by the state operated cotton procurement monopoly, by imposing a higher price on interstate trade. Furthermore, a
Central Sales Tax applies to the inter-state sale of commodities, further penalizing movements of goods across state borders. Cascading taxes on raw materials,
intermediate goods and capital goods add on to the inter-state sales tax. As a consequence, the effective tax rate imposed on interstate sales was higher than the
nominal tax rate on interstate sales.
^There m a y of course b e other reasons w h i c h i n d u c e firms to locate in these areas, such as interfirm
spill-overs w h i c h m a k e it a d v a n t a g e o u s for firms to locate within proximity to firms w h i c h h a v e
already established. I a m u n a b l e to distinguish b e t w e e n these m o t i v a t e s for proximity.
APPENDIX
Derivations
B.L. O B T A I N I N G THE W A G E E Q U A T I O N S
Linearizing labor d e m a n d and supply for m a n u a l and skilled workers, w e get:

Labor Demand unsfc -^ erf ' ' m g
aQ + aiwudt
+ a2wsdt + a3Xdt + a4Zft
Labor D e m a n d ^ " ' ' " ' m g
b0 + hwudt
+ b2wsd/t + b3Xdt + b4Z%J + ubdt
Labor D e m a n d =
+ uadt
o + Ciwudt + c2wsdt + c3Xdt + c4Z% + ucdt
Labor D e m a n d ^ " 1 ' 1 " 6
d0 + dxwudt
Labor Supply H)1s/a//(?rf
/o + / i wudt + f2wsdt
Labor Supply sjW/erf
go + giwudt
+ d2wsdt + d3Xdt + d4Zdt + udt

+ f3Xdt
+ f4Zdt S +, d
u dt
+ g2wsdt + g3Xdt + g4Z% + uddt
Setting labor d e m a n d equal to labor supply for both m a n u a l and skilled labor,
but keeping manufacturing labor d e m a n d in it's raw form, w e get wages for skilled
and manual labor:
Wudt =
^ ) [(CO - /o) + (c 2 - fl)Wsd,t
+ (C3 - f3)Xdt + C4Z% ~ f*Zdt
FManufacturing
manual,dt
sdt =
Kdo ~ o) +
+ U
dt
__ f ,
U
dt J
- gl)w>m1,t + {d3 - g3)Xdt + d4Z}t f

vManufacturing
^ ^ skilled,dt
+ 11 dt
126
,
dt\
g4Z$t
B. DERIVATIONS
127
Solving the system of simultaneous equations, w e get:
Wudt =
0i0 +
^nskllled4t
+ oci^fuiled + *sAg Productivity^ + ocAZA + oc5Zsdt

+ix6Xdt + edt
sdt
Po + hûnsmied,dt + hilled
+ fcAg Productivity,, + frZ* + p5Zsdt

+fi6Xdt
* =
+ edt
- ) ) * = * 7 F 7 7 ' = "> * { H ) ft =
and ft = < > , . ( & ) .

Under the assumption that unskilled and skilled labor d e m a n d are complements in agricultural and manufacturing, c\ < 0, Ci < 0, d] < 0 and c/2 < 0.
If total labor supply is constant, it follows that fi + g\ + f i + 2 = 0 and that
f i > 0, gi
< 0, /2 < 0 and g2 > 0. > 0 if the own-price labor supply
and d e m a n d response is greater than the cross-price labor d e m a n d and supply

response for both unskilled and skilled labor. Therefore cc\ > 0 and
> 0. Un-
der the same conditions, a.2 < ol\ since f i + f 2 > 0 and c\ C2 < 0 therefore
fi C\ > C2 f2 > C2 _/~2- Similarly /3j < fed\ decreases in the w a g e elasticity of agricultural unskilled labor d e m a n d intuitively, as agricultural labor d e m a n d becomes less responsive to changes in
wages, a larger increase in agrarian wages is required to "release" a given quantity
of labor from agriculture. ol\ increases as the unskilled w a g e elasticity of unskilled
labor supply decreases - the intuition is that if m a n u a l labor supply is unresponsive
to m a n u a l labor wages, an increase in m a n u a l labor d e m a n d in the manufacturing
sector will have a greater impact on wages than if the labor supply of m a n u a l laborers is responsive to changes in the wage. Similarly, ct2 decreases in the w a g e
elasticity of skilled labor supply and in the elasticity of skilled agricultural labor
demand.
B. DERIVATIONS
128
B.2. O B T A I N I N G THE I N C O M E EQUATION
Household income is given by:

yh = wuAh + ws
+ HAg{wu,
ws, Landh, 9, sah, Xh, p, Rain) + mh
where y/, denotes household income, A denotes the proportion of household time
spent working in the m a n u a l labor market, denotes the proportion of household
time spent in the skilled labor market. IlA% is the household cultivation profit
function, which is a function of household land e n d o w m e n t s , Landh/ the agricultural technology 6, the education of the household head, farm assets {Xh) and p
denotes local prices of inputs (for example, the price of agricultural seeds), m;, denotes all other sources of household income, for example income earned t h r o u g h
the bullock rental market or through local non-agricultural enterprizes.
A linear approximation to the cultivation profit function is given by:
hdt
0 +
Wmdt + C2U>sdt + 3Odt + C4Landhdt
+C?9dt * shdt + l,gLandhdt * wmdt + 9LandMt

HnQdt
+ 5Shdt + (,69dt * Ahdt

* wsdt +
10Raindt
* Wmdt + 129dt * Wsdt + n + Fd + Vhdt
where the error term contains household level farm assets, local prices and potentially the interaction of these terms with agricultural productivity. The interaction
between agricultural productivity and education arises from the complementarity
between these two variables (Foster a n d Rosenzweig, 1996). In addition, if large
landowners hire more labor on the labor market, the relationship between wages
and agricultural profits will vary according to land holdings.
Substituting the empirical specifications for the profit function (B.L) and wages
(2.7 and 2.8) into incomes and assuming (at first) that there is no variation across
B. DERIVATIONS
129
households in unskilled and skilled labor market time, we get:

Vmdt =
go + giEj? f f l W f + 2 E ^ + giOdt + g*Adt + g5Landhdt

+g7EMd,dt
* Landhdt
+ gsEt * Landhdt
+gwShdt * landhat + gu^dt

+guAdt
* Landhdt
+ gi5Adt
+ g9ddt *
+ g6sahdt
Landhdt
* 0dt + g12Es^f * 9dt + gl3s * d

*9 + gi6Rain
+ g1702 + pih + }id + vhdt
where the g terms are linear functions of skilled and unskilled labor market time: 1
The error term includes other sources of household income, household level farm
assets, local prices and potentially the interaction of these term with agricultural
productivity.
I estimate a compressed version of this regression. I separate households into
groups according to their landholding status and the education of the household
head. The coefficients on agricultural productivity and predicted manufacturing
employment capture the average effect of a change in these terms on households
within that endowment category:
+ 76 sahdt + }ih + Vd + vhdt
Vmdt = 10 + 7 i Etotnhdt+ 72EJJ( + 73#df + 74 Adf + y5Landhdt

where:
= a^Ci
+ A) + fafa + 0 + Land,, * (Csi + CM
faiCi + C) + Landh *
handle
+ CM
72 = <*2(i + A) +
7S = a3(Ci + A) + faiCi + 0 + Cno + Cupo +
+ S/,^7
B . 3 . A L T E R N A T I V E S P E C I F I C A T I O N FOR THE WAGE E Q U A T I O N
In the first stage specification, employment in industry i is written as a function of

industry z's policy changes, industry i's resource usage and regional endowments.
The error term in the first stage specification includes the local wage, all other
1
go = Co + Cio + Z2P0 + Aa0 + /3o si = i(i + A) + J8j (2 +
c?3
= 3(Cl + ) +
g7 =
c?13
+
ll,
c ?14
g8 =
=
C84 +
2 = 2(Ci + A) + p2(b + ),
+ ) + ll0 + ZnPo, g4 =
+ A) + /34(^2 + ), g5 = U 86 =
+ t,9p2, g9 = U, #10 =
gll = Cll a l + ^12^1, #12 = ll2 + l2^2/
#15 = Cll4 + ^12^4/ gl6 = Cw,g\7 = C.lla3 + ^12^3
B. DERIVATIONS
130
industries' policy changes and potentially the interaction of industry i's policies
with industry j's policies.
To violate my identification strategy, the omitted variables in the first-stage error term must be correlated with the triple interaction terms as well as with unobserved determinants of the wage regression. If the instrumental variables are valid
then the partial correlation between the second-stage error term and each instrumental variable should be zero. If policy changes are correlated across industries
over time, the covariance between the interaction between own-industry tariffs,
own-industry resource use and regional raw material endowments and any interaction involving industry j's tariffs may be non-zero.
In the case that the manufacturing and agriculture only overlap in the labor
market, these additional terms are highly unlikely to be partially correlated with
the wage error term. If there is another plausible channel through which the additional terms directly enter into the second stage regressions, these effects should in
any case be largely absorbed by the interaction of average industry tariffs and local natural resource characteristics included in the second stage. Finally the nature
of the policy reforms conducted greatly reduces the possibility of concern about
restrictions on the first stage specification.
I estimate two alternative specification to show that my results are robust to
other approaches. In the first, I allow industry i to be affected by industry j's
tariff change through the labor market. In the second more general approach, I
allow industry i's employment to vary with policy changes across all industries.
With a long panel of industries, this alternative specification could in principal be
conducted at a 3-digit level. With only four data points in time, for reasons of
parsimony I use 2-digit industry categories.
Topalova (2004) s h o w s that the differential c h a n g e s in i m p o r t tariffs across industries d u r i n g the

t r a d e policy r e f o r m s c o n d u c t e d b e t w e e n 1991 a n d 1997 w e r e unrelated to the state of the industries
at the b e g i n n i n g of the reform.
B. DERIVATIONS
131
In the following, I allow shocks in industry j to enter into industry i's employment through the labor market, notably through the wage. Writing m a n u f a c t u r i n g
employment in industry i, district d, time t as:
k d t
0 +
it
u>dt
p2rit
frwdt *
Si
r i t
Ti
wdt +
i t
faTit
*nd
* wdt *
Si
* nd
PioTitSi
+Pn Xdt + Sd + 5t + uldt

Where r,f is the import tariff covering products in an industry at a given m o m e n t
in time, nd is the natural resource stock in district d, r( is a measure of natural
resource use, wdt are equilibrium district wages at time t, Sj is a measure of labor
usage, Xdt are other explanatory variables. Sd are district level fixed effects and St
are time dummies. The only difference between equation B.l and the first stage
equation introduced in the main b o d y of the text are terms w dt through r,/ * s,.
The presence of wdt in equation (5) highlights the simultaneity problem since in
a structural equation equilibrium employment is clearly a function of equilibrium
wages.
Aggregating equation this equation over / industries to obtain employment in
the manufacturing sector at a district level:
kt
JPo + J f r T t + JpiTft + I f o r f t * nd + //3 4 tt * nd + //3 5 r * nd

+ JhZVdt + Jp7SWdt + JpgTt * Wdt + JfoTSt * Wdt + /jSloTSf
+JPnXdt
where x = j
+ J6d + Judt
x,. Since district and time d u m m i e s are included in the specifica-
tion, only variables that vary at a district time level are included leaving:
Idt =
JPo + Jhrft
*nd + JfaTt *nd + Jp6wdt + Jfi7wdt * s + Jp8wdt * rf
+ Jp9Wdt * TSt + JftuXdt
+ JSd + }udt
B. DERIVATIONS
Inserting (33) into (31) w e get:
132
ht
bo + biTTt *nd + b2ft *nd + b3tldt + b4/tXdt + Ddt + dt
^ ( f a - fas - fart -
ht
farst)
+ biTft * nd + b2Tt * nd + b4tXdt + Ddt + dt]
{ht)
Inserting district level wages into the industry-district level w a g e regression, and
ignoring interactions of industry level variables with district and time dummies: 4
Udt =
o + fliT/t +
+a7Xdt
* ri
2Tit
* Xi
+ a3Tit
*nd
* nd
+ a4Tit
+ a5r:
* nd
* st + a8Xdt * Tit + a9Xdt * rit * s, + a10ldt + auldt
+ I 3 * D / * rit
* Si + aUTit
* Si + Dd
+ Dt
a6Xdt
* s, + auldt
* rit
idt
Substituting equation (34) into industry-level employment above: 5
Udt
c13tXdt
c20tTt
where
x r
* Si
* nd
( f a
i d t
( f a
= aifa,
Co
Cb
W/rff
( f a
fa
fas)),
fl]
Sj
ci\\
faT,
faTS,)),
C l ^ j ^ i f a i
2 Q t
} 2
C161
fa
fa
( f a \
"12 * ^ *
c
] 3
d i
fa
fa
fa,
faTit
+ i ( / 3
C t \ ( f a
C m
fa
fatx0,
=
a - i f a ,
( u , ,
+
a
+
farst)),
a
t x
S , )
fa
d J
fao,
faT,
fas
faT,
fa
* s,- + cl2tXdt
fas
faf,
a w
faYs,),
C u t
*^ *
fazs,))
4 /
faTS,))
fa
farst)),cl2t
fa
fa,
i2]~-(/3n
- ^ ( f a i
1 0
fa,
fc^
fa
c c
fa,
C l
* i ( f a
{ f a
t x ^ f a ,
= oc2fa, a13 = oc2fa,
f a ,
faTS,))
rit
+ d d ( f a + fi7s + far, +
fa
* tu
e,dt
( f a
fa,
fas
T ^ *
( i n *
+ cutXdt
( f a + fas + far, +
fl3
+ c6Xdt
Sj
fajs,)
= a2fa,
a ^ f a ,
3 / t
sf + c 1 6 f r r , * nd *
+ Dt
fa
f a ,
f t *nd*
fa,
*nd*
c15tTft
* s, * Tit + Dd
t o f l i O / C]
fl8
faT,t
a ^ y ,
Drfj = (3d + a0(/38t,
a i f a ,
fait
ail * ^ *
b ]
fas/))
* nd
faTs,))
fas,)),
cistTf *nd + c
faTs,)),
r ,
+ cutTft *nd +
faT,
( f a
T i
* Tit + c2lfTf
* o ( f a
Sj * Tit +
= a-ifa,
Vdt
( f a
/ 3 n
( f a
a$
* Tit
*nd*
+ c5r, *
c7Xdl * Sj + c8Xdt * rit + cgXdt * Tit * Si + cWtXdt
c3Tit
c4r/t *
(St + d , ( f a + fas +
Co + C] Tit
( f i n + i ( f a + fas + psf,
fa,
fas+
fa0
fas
fa
B. DERIVATIONS
CiotXdl,cUitXdt
* Sj,c12itXdt
* Tit and cmXdt
133
* s, * rit capture third-order effects of
e m p l o y m e n t responses to changes in aggregate employment driven by changes

in average tariffs vary with Xdt and interactions of Xdt with industry level labor
usage and tariffs. Since these third-order effects are likely to be small, I omit these
variables from the analysis. A similar case can be m a d e for excluding C\^ t rft *Kd*
Tn, c\7t^t
*nd * Sj * Tn, c20t^t *nd*
assumption that
hdt =
Tn, c2it?t * nd* Si * Tit. I n addition, u n d e r the
ftw = 0, c\ 3t through C\gt don't vary over time. 6 This leaves:
Co + ci rit + ciTit * rt + c3Tit * rj *nd + c4rit *nd + c5rl * nd + c6Xdt
c7Xdt * s, + c8Xdt
ci8rf * nd + C19ft * nd * Si + Dd + Dt + eidt
* Tit
+ CgXdt * Tit * sf +
Tf
*nd
+ c15Tft
*nd*
Si
I n c l u d i n g the i n t e r a c t i o n s of t h e s e v a r i a b l e s w i t h t i m e t r e n d s h a s little i m p a c t on the coefficient

e s t i m a t e s in t h e s e c o n d stage. T h e s e results are t h e r e f o r e n o t i n c l u d e d in this r o b u s t n e s s c h e c k s b u t
are a v a i l a b l e u p o n r e q u e s t .
APPENDIX
Figures
134
C. APPENDIX FIGURES
135
T a b l e A . l : Physical Costs of T r a n s p o r t a t i o n of R a w Materials

Specific
Gravity
Mean
0.32
0.4-0.86
0.56
0.58
0.96
1
Density
Mean
SD
320
400
560
580
960
1000
-
Tobacco
Timber
Cotton (w Lint)
Rice
Sugar (Raw)
Water
1200
2160
2620
2660
3684
4610
8850
19300
Coal (Bitumious)
Clay (Kaolin)
Ceramics
Construction
Chemicals
Steel
Metals
Gold
300
120
1100
2410
6400
-
1.2
2.16
2.62
2.66
3.68
4.61
8.85
19.3
Domestic Price/Ton
Mean
SD
36000
Density/
Price
0.01
12900
18400
32000
0.04
0.03
0.03
390
180
187
203
509
1050
45135
3.08
12.00
14.04
13.08
7.24
4.39
0.20
4.39E-06
160.10
40.41
297.59
1257.98
61042.80
4.40E+09
Volume
Specific Gravity, Source: Compiled f r o m Various Engineering Texts
P r i c e s : a ) M i n e r a l s - D o m e s t i c O r e P r i c e s in 1 9 9 0 Rs a t M i n e H e a d , b ) C r o p s - W h o l e s a l e p r i c e s , f r o m
A g r i c u l t u r a l S i t u a t i o n in I n d i a . S o u r c e M i n e r a l D a t a : M i n e r a l Y e a r b o o k , I n d i a n B u r e a u of M i n e s .
Metal
Wood
Ceramics
Bidi M a n u f a c t u r e
Structural Clay P r o d u c t s
Fertilizers and Pesticides
Sawing of Wood
Glass and Glass P r o d u c t s
R e f r a c t o r y and Structural Clay
Veneer M a n u f a c t u r e
E a r t h e n a n d Plaster P r o d u c t s
Semi Finished Iron
Structural W o o d e n P r o d u c t s
Non-Structural Ceramics
Ferro-Alloys
W o o d e n a n d Cane Boxes
C e m e n t and Plaster
Copper M a n u f a c t u r i n g
W o o d Industrial P r o d u c t s
Mica P r o d u c t s
Brass M a n u f a c t u r i n g
Cork P r o d u c t s
Structural Stone Goods
Aluminium M a n u f a c t u r i n g
Wooden Furniture
Asbestos C e m e n t
Zine M a n u f a c t u r i n g
Bamboo Furniture
Misc Non-Metallic Mineral P r o d u c t s
Processing of Metal Scraps
W o o d e n P r o d u c t s nec
Radiographic A p a r a t u s
Other N o n - F e r o u s Metal P r o d u c t s
Energy
Chemicals
Tea Processing
Pulp, P a p e r and P a p e r Board
Organic and Inorganic Chemicals
Containers and Boxes
Fertilizers and Pesticides
P a p e r n.e.c
. Organic and Inorganic Chemica Paints and Varnishes
Coke Oven P r o d u c t s
Drugs and Medecines
Coal and Coal T a r P r o d u c t s n.e.i P e r f u m e s and Cosmetics
Refining of Sugar
Cement, Lime and Plaster
Ferro-Alloys
Tyre a n d T u b e s
Footwear
Brass M a n u f a c t u r i n g
Aluminium M a n u f a c t u r i n g
Rubber Products
P r o p o r t i o n of Literate Labor
Plastics
Drugs a n d Medecines
Coke Oven P r o d u c t s
Batteries
Printing and Publishing of Books
Tyre and T u b e s
Refined P e t r o l e u m P r o d u c t s
Metal F u r n i t u r e and Fixtures
Insulated W i r e s and Cables
Electrical E q u i p m e n t n.e.c.
C. APPENDIX FIGURES 147
Figure Al-a: Map of Employment in the cement industry and

limestone deposits in 1990.
Map of Limestone Grade Cement Deposits
Employment in the Cement Industry, 1 9 9 1
C. APPENDIX FIGURES
Figure Al-b: Map of Employment in the plywood industry and

wood deposits in 1990.
Map of Forest Cover and Employment

in Wood Industries in 1990
(Wood Industries: NIC-S7 270-27S)
338 3 4 4 2 0 0 - 15GD.D0DDCD
1 5 1 4 C 0 3 3 0 3 - 780.223C 3 3
-, 7 8 1 .CODDCO - 32S4.475CDD
3 3 2 e 122D l IO - f - 1 S.Q-SSCOQ
5 5 i e . C 0 3 D 0 D - o 8 ? o 7 33C30C
137
C. APPENDIX FIGURES 149
Figure A2: Exerpt from the National Mineral Atlas - Map

showing Limestone, Barytes, and Asbestos Deposits in
Cuddapah, Andhra Prades
C. APPENDIX FIGURES
Figure A3: Map of Suitability for Rainfed Pulses in India
I I
[~]
I I
I I
n
' I
Undefined
SI > 85 : Very high
SI > 7 0 : High
SI > 5 5 : Good
SI > 40 : Medium
SI > 25 : Moderate
SI > 5 : Marginal
SI > 0 : Very marginal s.
Not suitable
Water
139

ROOWDET
for Offset Lithography
and Screen
Printing
Transilwrap
Company, Inc. has been manufacturing
and converting plastics for the printing industry for more than 70
Our goal is to provide you with products consistent in quality and value that meet your specific
requirements.
Achieving
this goal means our equipment

knowledgeable
and experienced
years.
is built and maintained to the most exacting standards, and Transilwrap's

dedicated,
people are the key to our ability in understanding
and meeting your needs.
The result is a full line of printable plastics recognized by our customers for their quality and value in meeting the needs of the
industry
Transilwrap Company is dedicated t o producing printable materials.
P H O - P r o n f i synthetic paper is a mineral-filled synthetic that is easy
O u r ISO 9002 certified manufacturing plants get their raw materials
t o print using conventional inks. PRO-Print can be die-cut, guillotine cut,
from the world's largest and most reputable companies. O u r converting
folded, sewn, riveted, stapled, embossed, hot-stamped, adhesive coated
sites provide precision converting services including slitting and sheeting.
and microwaved. It offers excellent heat, tear and impact resistance along
O u r quality control team continually monitors equipment and materials
with anti-curl properties. PRO-Print is available with a matte/matte o r
to ensure we meet the needs of the printing industry by consistently
gloss/matte finish, in white opaque or black, and in FDA-approved grades.
producing materials with:

MXM
good ink compatibility

corona t r e a t m e n t on one or t w o sides for
b e t t e r ink adhesion and drying
tight gauge control
excellent anti-curl properties
consistent opacity
&
synthetic paper is a mineral-filled polypropylene with the same
feel and texture as white offset paper, but with the tear and chemical
resistance of plastic. It can be die-cut, folded, hot stamped, punched,
perforated, riveted, stamped and sewn and is environmentally friendly.
MXM is ideal for maps, greeting cards, tags, charts, signs and other
point-of-purchase applications.This product is more economical than
PRO-Print, but has a shorter outdoor life.
Transilwrap is truly committed to serving the printing industry. As
H i g h Impcscf P o l y s t y r e n e is an inexpensive plastic with an
members of the GATF (Graphic Arts Technical Foundation) and SGIA
excellent printable surface. It is more durable than paper or card stock
(Screen Printing and Graphic Imaging Association International), we are
and will not warp from humidity. It can be die-cut and guillotine cut.
glad t o share our knowledge of inks, fountain solutions, drying processes
High Impact Polystyrene is available in white translucent or white
and materials handling to help you successfully complete your offset and
opaque, matte/matte o r gloss/matte finish, screen o r litho grade.
screen projects.
FDA-approved grades available by custom order.
P r i n t a b l e Plastics t o M e e t Y o u r N e e d s
Transalloy
For banners, book covers, booklets, calendars, cards, charts, displays,
formulated plastic that is easy to print and allows for sharper, more
P - 2 6 0 E X / P 3 0 0 / P - 6 0 0 is our specialty
labels, maps, menus, packaging, posters, signs, tags and more,Transilwrap
vibrant graphics than ordinary vinyls or styrenes.A multi-polymer alloy,
has the printable plastic for you. O u r sales representatives will help
P-260EX/P-300/P-600 formulations are more printable, durable and tear
you select the best plastic to meet the specific needs of any job. All
resistant than many other plastics.Transalloy P-300 is the premium
of the following plastics are available in precision-cut, press-ready
product in the line, offering the ultimate in printability and application
sheets or custom-slit rolls.These materials can be successfully printed
properties. Transalloy P-260EX's greater rigidity makes it an excellent
by offset lithography or screen printing t o help you meet your
choice for phone and loyalty cards. Transalloy P-600 is a gloss/gloss
customers' requirements.
surface productan alternative t o gloss vinyl.
CSingZ'* molecular enhanced film is our most versatile product for

temporary indoor signage and posters.This specialty polypropylene clings
to any indoor surfaceplaster, brick, wallpaper, glass, metal and stucco
without adhesive.
High
Density Polyethylene (HDPi)
is a durable extruded
polyolefin available in natural, white and colors.The product works well

in a variety of temperatures and has good hinge characteristics.
IM(SM;
is a matte surface
rigid PVC film that can be die-cut, guillotine cut, folded,

glued, welded, embossed, punched, laminated and
thermoformed. It offers good water and oil resistance,
puncture and tear strength, resistance t o acids and
alkalis, dimensional stability and flame resistance, and is
self-extinguishing. Matte Calendered Vinyl is available in clear
o r white; custom colors by special order.
Tfrgmsvy' gloss rigid vinyl is a PVC that is less expensive
than other vinyls.Transvy can be die-cut, guillotine cut,
welded, glued, folded, embossed, punched, laminated and
thermoformed. It offers high gloss, excellent puncture and
tear resistance, moisture resistance, flame resistance, and is selfextinguishing. Transvy is available in clear or white.
riersted F>|ysS@r film
C a r d a n d sign a p p l l c a t i o n S ^ ^ M , , ' jJjii'f^j^MW
is a high-clarity, rigid plastic that
can be die-cut and guillotine cut. It offers great durability and
^"ffisWfi^V-
^T^lW' ^
strength, heat resistance, dimensional stability and anti-curl properties.

Oriented Polyester is available treated for ink and adhesive receptivity.
OSiP" extruded
Tywetk spunbonded olefin is a plastic material that combines

the best properties of fabrics, films and papers.Tyvek offers great
polyester can be easily printed, sealed, folded and
formed. Although it is less expensive than Oriented Polyester, it still

offers great strength and durability.
Tr@ns-Flex-Ccs$t'" static
durability; puncture and tear resistance; moisture, chemical, r o t and

mildew resistance; clean room acceptance and dimensional stability.
It can be die-cut, guillotine cut, folded, grommeted, sewn and stapled.
Tyvek is available in a variety of grades.
cling vinyl is a soft, polished, flexible
vinyl film adhered to a cast-coated board allowing for easy printing,
T r a n s - B o n r a e r " polyolefin is a family of films designed specifically
die-cutting and guillotine cutting.Trans-Flex-Cast has no adhesive
for printing banners and pennants:
coating, yet will adhere to most smooth surfaces such as glass,

porcelain, glossy plastics and some metals. It is easily removable,
r?S1S-BijirO'ar" 8 polyolefin is a minimal shrink film designed for
repositionable and reusable.Trans-Flex-Cast static cling vinyl is
good registration during process printing.Trans-Banner I is available
available in clear or white, .006", .008" vinyl with 8 pt. or 10 pt. board.
in ultra-low slip through high-slip formulations.
It is available surface prepared for conventional offset printing and

non-surface prepared for UV offset and all screen printing.
T r e w s - B e n n e r " " i l polyolefin is a coextruded two-side corona

treated, white/gray/white film designed for banners where 0.5-1.0%
Trans-Cling ' 11
low-tack vinyl is a flexible vinyl film with a
low-tack, pressure-sensitive adhesive for use on smooth surfaces.
light transmission is acceptable.This material prints well with screen,

offset and flexo applications.
It is available in white or clear, polished o r matte surface.
Trsaris-llonrwsr"" ultra
is our premium poly banner film for
T r e i i n s - S f i c k " high-tack vinyl is a removable, standard-tack,
offset printing, process color work. It is dimensionally stable, more
pressure-sensitive, white flexible vinyl. It is available top coated
rigid for greater registration detail, and 98% opaque.
and non-top coated for indoor and outdoor applications.

T r a n s - l c a n M e r " i V is a family of reinforced vinyl banner
Sfeek-Msfe ' permanent
adhesive vinyl is a removable, permanent-
materials. Standard products are the scrim-reinforced 10-oz.
tack, pressure-sensitive, white flexible vinyl. It is available top coated
Dakota and 14-oz.Yukon for one side printing, indoor and outdoor
and non-top coated for indoor and outdoor applications.
applications; and 13-oz. Sierra, a mesh-reinforced vinyl with two-side
Clisip-Msst'" USM multi-polymer
weights and grades available on a custom basis.
printability for indoor and limited (3-month) outdoor use. O t h e r

alloy is a rigid indoor signage
material with a high-performance ultra removable adhesive (URA) for

surfaces such as metal, polypropylene, stainless steel, glass and many
W @ l k e r ? ! ( l i s e r : floor graphics is a two-component, advertising
plastics. Its white satin top surface is highly printable.
system.The first component, which can be printed by offset lithography

or screen printing, is a white vinyl film with a removable, pressure-
VmKS-l$S3irri@r"* block out film is a white opaque metallized
sensitive adhesive and a 92-lb. release liner.The second component,
print treated material, with a permanent acrylic adhesive and 50#
a high-quality, tough, scratch-resistant overlamination, protects the
liner. Used in place of opaque white flood coat to create 2-way
graphic surface from scuffing and meets UL standards for slip
read P.O.P materials.
resistance. WalkerTalker comes with a product warranty and a
YffifSte" synthetic print sheet is a white opaque, single-layer micro-
retailer indemnification policy.
porous printing material that behaves more like paper than plastic. Its
absorptive characteristics allow for superior ink and adhesive bonding
and can be printed using conventional inks.Teslin can be folded, sheeted,
die-cut, grommeted, glued, sewn and laminated with ourTrans-Kote
thermal laminating films. It is ideal for banners, membership and phone
cards, calendars, maps, tags and other point-of-purchase applications.
C l i n g Z ' is a registered t r a d e m a r k of Permacharge C o r p o r a t i o n . Teslin- is a registered t r a d e m a r k

of PPG Industries.Tyvek* is a registered t r a d e m a r k of D u P o n t Company. Indigo- is a registered
t r a d e m a r k of Indigo Electronic Printing Systems. A r i z o n a ' is the registered t r a d e m a r k of
Raster Graphics. All o t h e r t r a d e m a r k s are t h e exclusive p r o p e r t y of Transilwrap Company, Inc.
Suggested Applications for Printable
Plastics
as
"H
a.
12
B
Backlit Displays
Indoor Banners
d
o
> o
c:
O Q Q Q O
O u t d o o r Banners
Book Covers
Booklets
Business Cards
f :
C;j
Calendars
Charts
Counter Mats
0Si
W
2-Way Decals
@ CI
Digital Printing
Food Tags and Inserts
e
0k
Greeting Cards
#1
Maps
Membership Cards
Menus
Overlays
Packaging
Phone Cards
ifl
P.O.P. Displays
Posters
^l-S?
#14
|||
fe
w 1
f-m
O ^ v )
c> o
O -
Pressure-Sensitive Labels
Pump Toppers
Shelf Talkers
(I)
o
Signs
Tags
f^ t
~ -
Tickets
Toys and Games
W i n d o w Displays
Floor Graphics
C u s t o m e r has final responsibility t o determine suitability o r fitness of a product for the specific end use and application conditions.Transilwrap offers this chart merely as a quick reference tool.
TRANSil\VRAp C O M p A N y , ilMC.
9201 W. Belmont Ave.
Franklin Park, IL 60131
(847) 678-1800
www.transilwrap.com
For more information, call:

( 8 0 0 ) 3 2 1 - 8 5 4 4 or ( 4 4 0 ) 6 3 8 - 2 0 0 0 in the U.S.A.
( 8 0 0 ) 2 6 8 - 4 1 0 8 or (416) 2 9 2 - 6 0 0 0 in Canada
0 0 1 - 8 4 7 - 2 3 3 - 4 1 1 8 in Europe
Fax: 0 0 1 - 8 4 7 - 2 3 3 - 4 0 4 1
E-mail: exportsales@transilwrap.com
Or visit our w e b site at w w w . t r a n s i l w r a p . c o m

Economic Growth, Rural Educational Investment and The Level and Distribution of Rural Incomes

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Economic Growth, Rural Educational Investment and The Level and Distribution of Rural Incomes

Diunggah oleh

Hak Cipta:

Format Tersedia

Abstract

Economic growth, rural educational investment and the level and

Economic growth, rural educational investment and the

Reena Chandu Badiani

Dissertation Director: Professor Mark Rosenzweig

UMI Number: 3414959

All rights reserved

2010 by Reena Chandu Badiani

Related Literature: Rural Labor Markets, Determinants of Wages and

Setting: The Rural Indian Economy

What is the effect of growth in household incomes and returns to

Setting: Education in India

Data and Description of the Environment

Obtaining the Wage Equations

Obtaining the Income equation

Alternative specification for the wage equation

Descriptive Statistics - Literacy by Task and Sector

Descriptive Statistics - Wages by Task and Sector

Descriptive Statistics - Education by Skill Bracket in Manufacturing 63

Descriptive Statistics - Tasks by Skill Bracket in Manufacturing . . .

Descriptive Statistics - Decomposition of Variance in the Proportion

First Stage Static Results

First Stage Dynamic Results

Wage Robustness Checks, Groundwater and Agricultural Prices . . .

Wage Robustness Checks, Instruments and Initial Characteristics . . 69

Wage Robustness Checks, Price of Agricultural Products and Proportion

Income, Splitting Households by Land and Skill

Income, Checking Robustness of Land Thresholds

Income Robustness Checks, Alternative Income Sources

Days devoted to Work in Agriculture

Descriptive Statistics - Education

Structural Education Parameters

Robustness Check, Education Specification

Robustness Check, Education Infrastructure

Density and Specific Gravity of Raw Materials

Industries by Factor Intensity

Map of Limestone Deposits and Employment in the Cement Industry 136

Map of Forest Density and Employment in the Plywood Industry . . 137

Exerpt from the National Mineral Atlas

Map of sorghum regions from the GAEZ (FAO, 2001)

"*To n a m e b u t a few, the d e m a n d for education has b e e n s h o w n to r e s p o n d to rising r e t u r n s to

endowed with multiple productive assets will therefore experience heterogeneous

Katz and Murphy,

Translating this question to a developing country

Foster and Rosenzweig,

Datt and Ravallion,

Lanjouw and Murgai,

However, industry level data

^Growth in the agricultural sector h a s been f o u n d to alter equilibrium w a g e s (Jayachandran, 2006),

2. ECONOMIC GROWTH, WAGES AND INCOMES

2. ECONOMIC GROWTH, WAGES AND INCOMES

regional agricultural potential: I apply the national time-varying maximum yield

2. ECONOMIC GROWTH, WAGES AND INCOMES

2.2. RELATED LITERATURE: R U R A L LABOR M A R K E T S , D E T E R M I N A N T S OF

A key characteristic of the classic models of economic growth is a m o v e m e n t out

2.2.1. Determinants of Wages in Rural Labor Markets

2. ECONOMIC GROWTH, WAGES AND INCOMES

The empirical evidence of w a g e determination a n d labor supply in developing

Groivth in Rural Labor Markets

There is a growing literature exploring the impact of rural manufacturing growth

In this section, I provide descriptive evidence supporting the central arguments

The predictive p o w e r of this i n s t r u m e n t is h o w e v e r low.

2. ECONOMIC GROWTH, WAGES AND INCOMES