ESSAYS ON IDENTITY, WORKFARE PROGRAMS, AND LABOR MARKETS

By

Sudhir K. Singh

i

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

Agricultural, Food and Resource Economics—Doctor of Philosophy

2023

ABSTRACT

This dissertation explores three distinct aspects of labor markets in India: the impacts of the

public works program on agricultural productivity, the impact of gender reservation in local

leadership on labor market dynamics, and the role of identity-based political connections

in public employment.

The first essay of the dissertation studies the impacts of India’s National Rural Employ-

ment Guarantee Scheme (NREGS) on agricultural productivity. While previous research

has primarily focused on wage effects, this study investigates the effects on agricultural pro-

ductivity, considering wages, income, and infrastructure development. By employing the

difference-in-differences method and utilizing detailed agricultural data, the analysis reveals

that NREGS has led to a significant increase in agricultural productivity for marginal farm-

ers and no effects for large farmers. The increase in agricultural productivity for marginal

farmers can be attributed to the more intensive use of family labor by marginal farmers,

coupled with mechanization and increased irrigation.

The second essay of the dissertation focuses on the decline in female labor force partici-

pation in rural India. The study explores the effects of gender reservation in village council

leadership on women’s access to job opportunities, labor force participation, income, and

intra-household bargaining in the short-and medium term. The econometric results suggest

that gender reservation in local leadership primarily affects female participation in public

works and regular labor markets and their income and influence on household decisions.

The effects of reservation are observed with a lag, indicating the influence of social norms

and stereotypes.

The third essay investigates the role of identity-based political connections in public

employment. A measure of political connection is constructed by matching surnames

representing the caste of individuals and local leaders. The caste-based reservations are used

to instrument the identities of council leaders to address the potential endogeneity problem.

i

i

The findings indicate that political connections significantly increased participation in public

employment, both on intensive and extensive margins. The results are robust to the inclusion

of controls and various fixed effects. Moreover, the study finds that connections with council

leaders have a stronger effect compared to connections with village council members. This

research emphasizes the dynamics of identity-based political networks and their implications

for the distribution of public sector opportunities in India.

i

i

i

Copyright by
SUDHIR K. SINGH
2023

v

i

v

This dissertation is dedicated to
Elaine, Ben, and Lucy!

v

i

v

ACKNOWLEDGEMENTS

I am deeply grateful to my advisors and committee members, Songqing Jin, Chris Ahlin,

Nicole Mason-Wardell, Kyoo Il Kim, and Eduardo Nakasone, for their invaluable and

generous guidance and encouragement throughout my doctoral studies. A specific thank

you goes to Songqing Jin and Chris Ahlin for their rigorous examination of each draft of

the essays.

I extend my gratitude to Jeffrey Wooldridge, Robert Myers, Scott Swinton, Titus

Awokuse, Richard Horan, Soren Anderson, Mywish Maredia, and Thomas Reardon for

their unwavering support throughout my graduate journey. I also benefited from discus-

sions with Hari K. Nagarajan, Nese Yildiz, Klaus Deininger, Yanyan Liu, and Caitlin Brown.

Additionally, I wish to convey my sincere appreciation to my professors, colleagues, and

friends from both the Economics and Agricultural, Food, and Resource Economics depart-

ments for their invaluable support.

At the University of Rochester, I am thankful to my advisor, Travis Baseler, my col-

leagues and friends, and the Department of Economics for providing an intellectually stim-

ulating environment that has greatly benefited me during my graduate studies. Gratitude

is owed to countless friends—far too numerous to acknowledge individually—spanning

across Michigan State University and the University of Rochester.

Above all, I want to express my heartfelt gratitude to my family. Without them, it

would not have been possible for me to be where I am today. I am particularly grateful

to my partner, Elaine, for taking care of our son, Ben, and for her understanding and

support—especially throughout the pandemic when I needed to stay apart. I truly love you

both. I’m sincerely grateful to my mother-in-law, Sandra Hill, for her constant support,

particularly throughout the pandemic. Lastly, I am immensely thankful to my parents and

extended family members, who have always been there for me. Thanks to my pup, Lucy,

for staying up with me during late working hours.

vi

i

i

v

TABLE OF CONTENTS

LIST OF TABLES . .

LIST OF FIGURES . .

.

.

.

.

.

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . .

ix

. . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

CHAPTER 1

.

.

.

. .

. .

Introduction .

Potential Impact Pathways

HOW PRODUCTIVE ARE WORKFARE PROGRAMS?
EVIDENCE FROM INDIAN AGRICULTURE . . . . . . . . . .
1.1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Program Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2.1 National Rural Employment Guarantee Scheme . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . .
1.2.2

1
1
5
5
8
1.3 Data .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.5 Results .
1.5.1 Effects on Agricultural Productivity . . . . . . . . . . . . . . . . . 15
1.5.2 Effects on Agricultural Labor Use . . . . . . . . . . . . . . . . . . 18
1.5.3 Mechanization and Input intensification . . . . . . . . . . . . . . . 19
1.5.4 High Value Crops . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
Impacts of Public Goods Provision . . . . . . . . . . . . . . . . . . . . 23
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
. .
.
1.6.1 Results
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
. .
1.7 Conclusions .
.
.
REFERENCES .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
. .
. .
APPENDIX .

1.6

. .

. .

CHAPTER 2

EMPOWERING WOMEN: THE IMPACT OF FEMALE
LEADERSHIP ON LABOR MARKET DYNAMICS . . . . . . . 44
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
. . . . . . . . . . . . . . . . . . 48

.
2.1
2.2 Background and Institutional Context

Introduction .

.

2.2.1

2.2.2

India’s Declining Female Labor Force Participation: Evidence
and Policy Implications
. . . . . . . . . . . . . . . . . . . . . . . 49
Political Reservations and Females’ Labor Market Outcomes . . . . 50
2.3 Data and Empirical Approach . . . . . . . . . . . . . . . . . . . . . . 53
2.3.1 Data and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . 53
2.3.2 Econometric Approach . . . . . . . . . . . . . . . . . . . . . . . . 58
2.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 61
Impacts of Reservation on Female Labor Market Participation . . . 61
. . . . . . . . . . . . . . . . . . . . . . . 64
. . . . . . . . . . . . . . . . . . . . . 66
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

2.4.1
2.4.2 Heterogeneity of Effects
2.4.3 Exploring Impact Pathways

2.5 Conclusion .
.
REFERENCES .
. .
APPENDIX . .

.
.
.
.
. .

.
.
.

vii

CHAPTER 3

.

.

.

.

.

.

3.3.1

3.3 Data .

Introduction .

.
3.1
3.2 Background . .

IDENTITY-BASED POLITICAL CONNECTIONS AND
PUBLIC EMPLOYMENT . . . . . . . . . . . . . . . . . . . . . 83
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.2.1 The Caste System . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.2.2 The Role of Caste Identity in the Indian Economy . . . . . . . . . . 88
Public Works Program . . . . . . . . . . . . . . . . . . . . . . . . 90
3.2.3
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Instrumental Variable Framework . . . . . . . . . . . . . . . . . . 98
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
. . . . . . . . . . . . . . . . 102
Same Caste Group . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.6 Robustness Checks and Mechanism . . . . . . . . . . . . . . . . . . . 106
Sample Restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . 106
. . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Falsification Test
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
.

.
3.5.1 Role of the Village Council Members
3.5.2

3.6.1
3.6.2
3.6.3 Mechanism .
.
.
.
.
. .

3.7 Conclusion .
.
REFERENCES .
. .
APPENDIX . .

3.5 Results . .

3.4.1

. .

. .

i

i

i

v

viii

x

i

LIST OF TABLES

Table 1.1 Effects of NREGS on Agricultural Productivty . . . . . . . . . . . . . . 16

Table 1.2 Effects of NREGS on Agricultural Productivity by Farm Size . . . . . . . 17

Table 1.3 NREGS and Agricultural Labor Use . . . . . . . . . . . . . . . . . . . . 19

Table 1.4 Agricultural Inputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

Table 1.5 NREGS Increased Share of High-Value Crops . . . . . . . . . . . . . . . 22

Table 1.6 NREGS Related Investment and Productivity by Farm Size . . . . . . . . 25

Table A.1 Summary Statistics of Key Variables by Treatment Status and Year . . . . 35

Table A.2 Summary Statistics of Key Variables by Farm Size . . . . . . . . . . . . 36

Table A.3 Test of Parallel Pre-Trends for Key Outcome Variables Using REDS

1982-1999 Data .

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Table A.4 Test of Parallel Pre-Trends by Farm Size Using REDS . . . . . . . . . . . 38

Table A.5 Test of Parallel Pre-Trends Using Cost of Cultivation Survey Data . . . . 39

Table A.6 Agricultural Labor Use by Male and Female . . . . . . . . . . . . . . . . 40

Table A.7 No Effects on Total Cropped Area . . . . . . . . . . . . . . . . . . . . . 41

Table A.8 Labor Supply Decision . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

Table A.9 Estimated Productivity Effects Using Two-Stage DID Approach . . . . . 43

Table 2.1 Labor Force Participation by Reservation Status and Gender

. . . . . . . 57

Table 2.2 Estimated Effects of Political Reservation on Labor Supply at Extensive

and Intensive Margin . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

Table 2.3 Estimated Effects of Reservation Status on Labor Force Participation

by Marital Status .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

Table 2.4 Estimated Effects of Reservation on NREGS Governance . . . . . . . . . 67

Table 2.5 Reservation and Women’s Participation in Households’ Day to Day

Decision Making .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

ix

Table B.1 Household and Individual Characteristics in Ever and Never Reserved

Villages .

.

.

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

Table B.2 Village and Pradhan Characteristics in Ever and Never Reserved Vil-

lages, 1990/1991, and 2014/2015 . . . . . . . . . . . . . . . . . . . . . . 78

Table B.3 Estimated Effects of Reservation on NREGS Governance

. . . . . . . . 79

Table B.4 Reservation and Labor Force Participation by Marital Status and Types

of Employment

.

.

.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

Table B.5 Estimated Effects of Reservation on NREGS Governance

. . . . . . . . 81

Table B.6 Estimated Effects of Reservation on Decision Making . . . . . . . . . . . 82

Table 3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

x

Table 3.2 Political Connections and Its Effects on Employment

. . . . . . . . . . . 101

Table 3.3 Village Council Members of 𝑆𝑎𝑚𝑒 Caste (Jati) and Employment . . . . . 103

Table 3.4 Village Council Head of 𝑆𝑎𝑚𝑒 Caste Group and Employment

. . . . . . 105

Table 3.5 Effects of Political Connections by Sample Restrictions . . . . . . . . . . 107

Table 3.6 Placebo Regression of Political Connection’s Effects

. . . . . . . . . . . 108

Table 3.7 Local Panchayat Meetings and Leaders Accessability . . . . . . . . . . . 110

Table 3.8 Likelihood of Winning and Caste Support in Elections . . . . . . . . . . 111

Table C.1 Household Characteristics by Job Card Holding

. . . . . . . . . . . . . 121

Table C.2 Summary Statistics by Caste Groups for Job Card Holders

. . . . . . . . 122

Table C.3 Summary Statistics by Caste Groups for All Households

. . . . . . . . . 123

Table C.4 Political Connections and Its Effects on Employment

. . . . . . . . . . . 124

Table C.5 IV Estimates of Political Connections Impacts . . . . . . . . . . . . . . . 125

Table C.6 Political Connections and Jobcard . . . . . . . . . . . . . . . . . . . . . 126

Table C.7 Caste Heterogeneity in Effects of Political Connections . . . . . . . . . . 127

x

Table C.8 Do Village Council Members and Chairperson Have Differential Ef-

fects? .

.

.

.

.

. .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

Table C.9 Different Caste (Jati) but Same Caste Group . . . . . . . . . . . . . . . . 129

i

x

xi

LIST OF FIGURES

Figure 1.1 Distribution of Workers in NREGS . . . . . . . . . . . . . . . . . . . .

Figure 1.2 Distribution of Projects Carried Out Under NREGS . . . . . . . . . . .

6

7

Figure 1.3 Location of Sample REDS Villages by Phase of NREGS Program

Implementation .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

Figure 1.4 Gender Differences in NREGS Workers

. . . . . . . . . . . . . . . . . 33

Figure 1.5 Residualized Value of Output Per Acre . . . . . . . . . . . . . . . . . . 34

Figure 2.1 Location of Sample Villages

. . . . . . . . . . . . . . . . . . . . . . . 76

Figure 3.1 Distribution of Workers in NREGS . . . . . . . . . . . . . . . . . . . . 91

i

i

x

Figure 3.2 Share of Same Caste Population as Village Head . . . . . . . . . . . . . 94

Figure 3.3 Same Caste Population by Village . . . . . . . . . . . . . . . . . . . . . 119

Figure 3.4 Same Caste Population by Age . . . . . . . . . . . . . . . . . . . . . . 120

xii

1

CHAPTER 1

HOW PRODUCTIVE ARE WORKFARE PROGRAMS?
EVIDENCE FROM INDIAN AGRICULTURE

1.1

Introduction

Workfare programs are increasingly popular in developing countries as a means to

provide income support to the poor conditional on work while also promoting infrastructure

development. One of the most notable and largest workfare programs in the world is India’s

National Rural Employment Guarantee Scheme (NREGS)1. Given its size and ambition,

not surprisingly, NREGS has attracted considerable scholarly interest. NREGS has led to

an increase in wages (Azam, 2011; Berg et al., 2014; Merfeld, 2019), income (Cook and

Shah, 2022; Klonner and Oldiges, 2014), food security (Ravi and Engler, 2015), female

empowerment (Afridi, Mukhopadhyay, and Sahoo, 2016), and reduced migration (Imbert

and Papp, 2020), school enrollment, and scores (Shah and Steinberg, 2021), as well as

conflict (Fetzer, 2020). There is also evidence that the program provides a safety net in

response to shocks (Imbert and Papp, 2015), with positive impacts on reductions in infant

mortality (Banerjee and Maharaj, 2020), stunting (Dasgupta, 2017), and higher human

capital accumulation (Garg, Jagnani, and Taraz, 2020).

Yet, although most of India’s rural poor are in the farming sector, and some of the above

effects may be due to NREGS impacts on agricultural yields, program effects on agriculture

productivity received little attention in the literature. Nationally, it was shown that program-

induced wage increase triggered the adoption of small-scale labor-saving technologies,

mainly by marginal farmers (Bhargava, 2014). In Andhra Pradesh, the safety net provided

by NREGS helped smooth consumption in response to shocks and allowed farmers to adopt

more risky crops with higher returns (Gehrke, 2019). While this suggests that NREGS

1The National Rural Employment Guarantee Scheme was introduced through the National Rural Em-
ployment Guarantee Act of 2005. Subsequently, on 2 October 2009, it was renamed the Mahatma Gandhi
National Rural Employment Guarantee Act.

1

2

design features are well suited to provide targeted support to poor and smallholder farmers

and act as a productive enhancing safety net, little is known about its impact.

In this paper, we aim to fill this gap by assessing NREGS impacts on agricultural

productivity at the farmer level2. We draw on rich data from 1998/99 and 2007/08 rounds

of the nationwide Rural Economic and Demographic Survey (REDS) panel household

survey that covers all aspects of the production process (cultivated area, outputs produced

in each season, input use, technology choice, etc.). To estimate the effects of NREGS on

farmers and agricultural productivity, we exploit the variation from the phase-wise rollout

of the program and compare farmers in early phase districts (treatment) to farmers in later

phase districts (control)3. This allows us to assess the short-term productivity effects of

NREGS, explore pathways for these to materialize, and draw out and test implications in

terms of heterogeneity of effects across the farm size distribution. Additionally, we also

use the data collected in 2014-16 on the number of employment days generated and capital

use in NREGS projects for a subset of states to explore if NREGS affected agricultural

productivity via project-related works generating public goods, an issue that to the best of

our knowledge has thus far not been addressed in the literature.

Conceptually, NREGS’ impact on agricultural production can materialize via several

channels. By putting a floor under wages, the program is likely to raise the cost of

agricultural production and reduce hired labor use (Santangelo, 2016). This may give rise

to a substitution of the family for hired labor or efforts to replace labor with capital via

mechanization. In the short term, higher wages, together with NREGS-induced expansion

of labor demand, will result in an income effect that will transfer resources to benefit net

sellers (i.e., small and marginal farmers) at the cost of net buyers (i.e., medium and large

farmers) of labor.

2We use real output per acre as a measure of agricultural productivity throughout the paper.
3NREGS was implemented in a phased manner. In 2006, the program was initially introduced in the
first 200 poorest districts. Subsequently, in 2007, the program extended to an additional 136 poorest districts.
Finally, by April 2008, the scheme was expanded to cover the remaining districts.

2

3

As NREGS generates predictable employment in situations when, for example, exoge-

nous shocks such as droughts or floods and other opportunities have dried up, it provides an

implicit safety net (Godfrey-Wood and Flower, 2018). Possibly reinforced by the income

effects noted above, this may encourage the adoption of income-generating strategies with

higher risk-return profiles than those pursued traditionally by producers.

Finally, the program supports works to enhance agricultural productivity directly via

land improvement and minor irrigation or indirectly via rural roads to improve market access.

The size and incidence of productivity benefits from such works will depend on program

management, including selecting improvements with high potential impact, implementing

them transparently, and ensuring proper maintenance. NREGS works are thus expected to

have the most impact in states with high implementation quality. Size and incidence of

effects will depend on the type of work: land improvement or irrigation are often targeted

towards small producers (Ranaware et al., 2015) whereas general-purpose investments such

as roads are likely to benefit producers in proportion to their size of production.

Taken together, the program’s self-targeting, the positive income effect, and the fact

that an implicit safety net mainly benefits the poor lead us to expect NREGS effects to be

particularly pronounced for small and marginal farmers, an effect possibly reinforced by

the portfolio of work. Our DID regression suggests that NREGS indeed led to a significant

increase of 18% in agricultural productivity for marginal farmers with an operating area of

less than 2.5 acres (∼ 1 ha).

We find that the program affected the input used in agricultural production for marginal

and large farmers differently. It appears to have no effect on hired labor use for marginal

farms, but it decreased hired labor use for large farms by six days per acre. This decrease

seems to be offset by increased investment in irrigation and intensive use of fertilizer.

Simultaneously, the program increased family labor use by 13 days per acre for marginal

farms, coupled with mechanization and intensive irrigation use. However, it had no effects

3

4

on family labor use for large or medium farms.

We also find that NREGS-related investment in infrastructure increased agricultural

productivity. The estimated effects of total investment and employment generation on

agricultural output per acre are 4% and 3%, respectively. For marginal farmers, these

effects of investment and employment are 6% and 3%, respectively.

This paper contributes to two strands of literature. We add to the small literature explor-

ing NREGS impacts on agricultural productivity by documenting different channels for such

impact, including program-created works. While our findings regarding mechanization and

risk-coping are in line with what was found by studies such as Bhargava (2014) and Gehrke

(2019) and recent contributions that emphasized the role of NREGS as a safety net to cope

with shocks, our figures illustrate that most benefits accrue to marginal and small farmers.

We also find that NREGS-backed projects aimed at supporting agricultural production have

a positive productivity impact and that the size of such productivity benefits is potentially

large enough to allow recouping some of the program costs.

A second strand of the literature we contribute relates to the debate on the merits

and demerits of workfare compared to cash transfers (Dreze and Sen, 1990; Ravallion,

2022). While this issue is of interest in many countries (Gehrke and Hartwig, 2018),

it became salient in India following recent proposals to replace a plethora of subsidies

with a basic income (Coady and Prady, 2018; Khosla et al., 2018). We show that, if

effectively implemented, a workfare program can generate productivity benefits by building

public infrastructure and providing a safety net while work requirements act as a screening

device (Besley and Coate, 1995) that can sharpen targeting and enhance the program’s

cost-effectiveness. Implementation capacity thus seems a key factor: although workfare is

preferable in principle, capacity gaps that would lead to rationing (Dutta et al., 2012) or

works with high social payoffs difficult to implement (Alik-Lagrange and Ravallion, 2018)

could tilt the balance in favor of cash transfers. Quantifying the associated trade-offs as well

4

5

as ways to address capacity constraints via IT solutions, would thus be of great interest.

The rest of the paper is structured as follows: In section 1.2, we describe the program

characteristics and potential impact pathways. Section 3.3 discusses the data used in the

paper. In section 1.4, we discuss the empirical approach, the method used to evaluate it,

the identification strategy, and the extent to which it is supported by pre-program parallel

trends between the treatment and control groups. In section 1.5, we present the estimated

impacts of NREGS on agricultural productivity and mechanism: labor use, mechanization,

and input use. In section 1.6, we discuss the impact of different NREGS-supported works

on agricultural productivity. And conclude in section 3.7.

1.2 Program Characteristics

To motivate our analysis, we describe National Rural Employment Guarantee Scheme

(NREGS) characteristics and implementation arrangements and briefly summarize the lit-

erature assessing the program’s impact on wages, households’ ability to cope with shocks,

and the provision of local public goods. We use this to set out a conceptual framework and

hypotheses to be tested in our empirical investigation.

1.2.1 National Rural Employment Guarantee Scheme

India’s National Rural Employment Guarantee Scheme (NREGS) is the largest workfare

program of this nature globally. NREGS was implemented in a phase-wise manner. In

2006, the program was initially introduced in the first 200 poorest districts. Subsequently,

in 2007, the program extended to an additional 136 poorest districts. Finally, by April

2008, the scheme was expanded to cover the remaining districts4. Building on the country’s

long tradition of food-for-work schemes (Dutta et al., 2012; Subbarao, 1997)5, NREGS

4The selection of districts was based on a poverty index developed by the Planning Commission using
data from the 1993/94 National Sample Survey, a routine survey used to produce among others estimates of
poverty.

5The work requirement differentiates NREGS from conditional cash transfer programs such as PRO-
GRESA (Attanasio et al. 2012) or Bolsa Escola (Glewwe and Kassouf 2012) that focus on education or from
unconditional cash transfers.

5

6

Figure 1.1 Distribution of Workers in NREGS

guarantees employment of up to 100 days per year to households who registered locally

to obtain a job card and have applied for jobs 6. In line with the program’s self-targeting,

anybody can apply for a job card and then a job. Once a job application is lodged, authorities

are obliged to provide work opportunities within 15 days of application. However, if

applicants do not receive employment within this timeframe, they become eligible to receive

an unemployment allowance retroactively from the date of their application.

Physical labor is sourced to build productive assets such as access to roads and water

harvesting structures. 60% of the total NREGS workers are landless and are likely to be net

sellers of labor. The share of workers who are marginal farmers is 28 percent, while it is

only 12 percent for small and medium farmers combined.7 (See Figure 3.4)

6Applicants are eligible to receive a job card containing photos of all adult household members free within
15 days of application. Job-card holders’ indicative work demand leads to the elaboration of an annual plan
which, once ratified by the village assembly, is transmitted for consolidation at the district level, although in
practice, a more top-down process is often followed, based on central budget allocations.

7Marginal, small, medium, and large farmers are defined as those owning less than 2.5 acres, 2.5-5 acres,

6

7

Figure 1.2 Distribution of Projects Carried Out Under NREGS

NREGS also explicitly encourages female participation by paying equal wages to men

and women by requiring work to be provided in ways attentive to women’s needs and a

gender quota of one-third. Among landless workers, 68% are women while only 32% are

men. Women are more likely to work among the marginal farmers, too, but the difference

between men and women starts to disappear for small, large, and medium landholders

(Figure 1.4). The largest share of work is carried out in the road and land-related works, and

water conservation is the second largest work in the NREGS. Rural drinking water-related

work, sanitation work, and construction of aganwadis and play fields together generate about

19 percent of total work in NREGS. Some of these projects, such as water and irrigation-

related work, can directly affect agricultural productivity, while others, i.e., roads, can affect

agricultural productivity indirectly by reducing the market access frictions that are likely to

be faced by small and marginal farmers (See Figure 1.2).

5-10 acres, and more than 10 acres, respectively.

7

8

The overall responsibility for implementation is with states, and many functions are

delegated to villages or panchayats8. States put in place controls, including the use of

electronic payments and social audits, to minimize abuse, adjust to variations in capacity,

and ensure the program is implemented transparently.

1.2.2 Potential Impact Pathways

NREGS is likely to impact agricultural production through three channels, namely, (i)

a wage and income effect, the size and incidence of which will differ between net sellers

and buyers of labor; (ii) a risk reduction effect that, in line with the program’s self-targeting

character, will mainly accrue to small producers; and (iii) a public good effect through

works constructed by the program.

Wage and income effects arise as, by offering a minimum daily wage equally to men and

women, NREGS puts a floor under wages. In the short run, the exogenous introduction of

alternative employment opportunities will reduce labor supply to agriculture and tradable

sectors and increase production cost (Santangelo, 2016). The direction of associated income

effects depends on individuals’ position in the labor market: NREGS will increase income

for net sellers of labor but have an ambiguous or negative impact on net buyers. Producers

can respond to wage increases by (i) substituting the family for hired labor; (ii) investing

in labor-saving mechanical technology; or (iii) changing the output mix including the

possibility of reducing output

Many studies indeed find NREGS increased wages. Monthly wage data for 2000-2011

suggest that NREGS increased the growth of real agricultural wages by 4.3% during the peak

season, with gender wage gaps and skilled wages remaining unaffected (Berg et al., 2014)9.

8Village governments (gram panchayats) and their staff are responsible to match jobseekers to jobs
throughout the year. These include the preparing a list of projects at a reasonable distance from the village,
measures to allow female participation (e.g. child care), supervision of ongoing projects, worker identification
and assignment to specific sites, and financial management including wage payment

9A 2005-10 district panel that combines NSS data with information on rainfall and NREGS-induced asset
generation suggests that, partly due to the assets generated by NREGS, the sensitivity of wages to rainfall
shocks decreased by some 10 points Shah (2012).

8

9

National Sample Survey (NSS) data point towards program-induced wage increases of about

5 points overall and 9 points in ‘star’ states characterized by good program implementation

(Azam, 2011; Imbert and Papp, 2015). Such increases have been shown to be confined to

the dry season, reinforcing the notion of NREGS acting as a safety net (Imbert and Papp,

2020). This is consistent with reports of the program having reduced short-term ‘distress’

migration by unskilled workers in the dry season, in addition to, directly and indirectly,

enhancing the returns to small and medium farmers’ assets through water-conservation

structures and improving liquidity without affecting long-term migration by the higher-

skilled (Reddy, Reddy, and Bantilan, 2014). Econometric evidence from data specifically

collected for this purpose supports the notion of a program-induced reduction in short-term

dry-season migration but no effect on long-term migration (Coffey, 2013; Imbert and Papp,

2020).

In line with the notion that program-induced wage increase will provide the biggest ben-

efit to net sellers of labor, NSS data point towards large season-specific effects for scheduled

castes and tribes (SCs and STs) but not the general population10, as NREGS enabled these

groups to smooth consumption across seasons (Klonner and Oldiges, 2014)11. In Andhra

Pradesh, one of the better performing states, panel data suggest that NREGS was well tar-

geted, improving food consumption and asset accumulation by the poor (Deininger and Liu,

2019). This is likely to be the mechanism driving positive impacts on food security, savings,

and health outcomes by the poor (Ravi and Engler, 2015), female empowerment (Afridi,

Mukhopadhyay, and Sahoo, 2016), and less gender-based violence (Amaral, Bandyopad-

hyay, and Sensarma, 2015), especially for the poor. Positive investment effects are suggested

by increased school participation by primary-aged children (Islam and Sivasankaran, 2015)

10Data from the India Human Development Survey (IHDS) have been used to argue that NREGS increased
income and wages for the unskilled, reduced poverty by up to 32% and prevented 14 million from falling into
poverty (Desai et al. 2015)

11A back-of-the-envelope cost-benefit analysis suggests that consumption smoothing benefits for the SC/ST
population would have been sufficient to justify the wage outlays incurred by NREGS (Klonner and Oldiges,
2014).

9

0

1

and better primary achievement scores and grade progression (Mani et al., 2014). Effects on

enrollment and test scores for those in secondary school are, however, negative, most likely

because at this age children can substitute for adult labor (Shah and Steinberg, 2021)12.

Potential risk reduction benefits from NREGS are well recognized (Zimmermann 2015b)

and a series of recent studies link the NREGS roll-out with climatic data to document

the program’s impact on mitigating the impacts of drought on children’s development

(Dasgupta, 2017), infant mortality (Banerjee and Maharaj, 2020), cognitive development

(Garg, Jagnani, and Taraz, 2020), and agricultural yields (Taraz, 2019). Access to implicit

insurance via the program, particularly if combined with positive income effects, can allow

investment in high-risk high return activities that, although available earlier, would have

entailed too much of downside risk, especially in agriculture where NREGS access allowed

mechanization (Bhargava, 2014) and a shift towards riskier but more remunerative crops

(Gehrke, 2019) by small farmers.

If the program is implemented as directed, the public infrastructure created by NREGS

could further enhance agricultural productivity. Variation in NREGS implementation is

likely to affect the quality and longevity of any structures built under the program. Moreover,

the size and incidence of benefits from such structures will depend on their type with

investments facilitating market access likely to benefit all producers in proportion to their

level of market participation. NREGS guidelines allow land improvement and minor

irrigation activities to be conducted on marginal farmers’ private land (Babu et al., 2014),

an option that was widely utilized and rather well-targeted even in states not known for

high-quality implementation.

Few studies tried to assess the impact of infrastructure constructed via NREGS. Desai

et al. (2015) report that about 51% of program-related works are ‘completed’ but fail to

provide information on quality. While effects of such infrastructure were indeed limited in

12The program is also shown to have affected election outcomes (Zimmermann 2015a) and local violence

(Khanna and Zimmermann 2014).

10

1

1

Andhra Pradesh (Deininger and Liu, 2019), high quality of NREGS works was found to

increase the share of area devoted to cash crops (Shah 2013) although it is not clear whether

this is due to the works or the risk-reduction afforded by having employment through the

program available in times of need.

1.3 Data

For our main empirical analysis, we use the data from three rounds of the Rural Economic

and Demographic Survey (REDS) conducted by India’s National Council of Applied Eco-

nomic Research in 1981-82, 1998-99, and 2007-0813. The REDS builds on the Additional

Rural Income Survey (ARIS), first conducted in 1969 and 1971 that provides comprehen-

sive information on labor supply and wage receipts by each household member and inputs

and outputs from agricultural production for 4,257 households in 259 villages located in

17 major Indian states that were representative of India’s rural population, though with an

emphasis on higher-potential areas, at the time the sample was drawn14.

In 1981-82 surviving households and descendants from the original sample who resided

in the same village and a random sample of new households were surveyed (Vashishta 1989).

A similar procedure was applied to subsequent survey rounds resulting in a sample of 7,474

and 8,659 households across 242 villages in 17 major states in 1998/99 and 2007/08,

respectively15. The survey is particularly suited for the analysis as it contains detailed

information on labor supply and wage receipts by each household member and inputs and

outputs from agricultural production. To assess NREGS impacts, we draw primarily on the

survey’s labor and agricultural modules to assess the impacts of the program being available

13Until 1981-82, the survey was conducted as Additional Rural Income Survey. The REDS 2006 survey
started in 2006 with a listing of all households in sample villages. As agricultural data refer to the 2007/08
season, we refer to this survey as 2007/08 throughout.

14Unfortunately, data from the first round, which was based on a frame including one district per state in

the Intensive Agricultural District Program (IADP) and a random sample of other districts

15Political unrest precluded data collection in Jammu & Kashmir and Assam in 2006 and is thus dropped
from our analysis. Also, delayed survey implementation in Kerala implied that, when the 2006 REDS
questionnaire was administered, all the state’s districts had already been covered by NREGS, forcing us to
drop observations from this state.

11

2

1

on labor markets and agricultural productivity16.

Socio-Economic Profile of India (SEPRI) We also use data from the Socio-Economic

Profile of India (SEPRI), conducted between 2014 and 2016, to enrich our analysis. The

SEPRI is a follow-up survey of REDS and is administered to all households residing

in the REDS villages across the 13 states. The SEPRI survey employed the complete

set of questions of village modules from REDS, augmented by additional information

on village-level activities related to the National Rural Employment Guarantee Scheme

(NREGS). However, only a selected subset of questions from the household module of

REDS was used. The SEPRI contains detailed information on inputs and outputs from

agricultural production, consumption, assets, as well as NREGS activities such as project

types, employment generated, and investment in each project types17

Summary Statistics Key household and agricultural production statistics for 1998/99 and

2007/08 and in treatment and control districts are in Table A.1. Household size significantly

declines from 6.7 to 5.8 members on average. Partly due to subdivision with succession,

the mean land endowment declined from 5.9 to 5.1, with a modest increase in the incidence

of irrigation from 55% of land to 60% of land. Control areas are characterized by higher

mean land endowment and a higher share of irrigated land. Although NREGS-related work

is not identified separately in our data for the main analysis, we have detailed information

on the number of days devoted to crop cultivation by male and female family members and

hired workers. As a main outcome in the paper, we use harvest value per acre. Inputs such

as seed, machinery, and fertilizer are measured as the value per acre18.

Information on agricultural production outcomes in Panel B, Table A.1 points toward

16The sample includes all of a household’s descendants who still reside in the village in case of a split.
17The SEPRI was conducted in 13 Indian states, namely Andhra Pradesh, Bihar, Chhattisgarh, Gujarat,
Haryana, Jharkhand, Madhya Pradesh, Maharashtra, Orissa, Rajasthan, Tamil Nadu, Uttar Pradesh, and West
Bengal. Unlike the extensive questionnaire used in the REDS survey, SEPRI uses a subset of questions from
REDS with some additional questions on NREGS while sampling the entire population of the REDS survey
villages.

18All values of output and inputs are in real term.

12

3

1

modest increases in the real value of output (from Rs. 7784 per acre to Rs. 8717 per acre).

While the total number of hired laborers used increased from 16 to 20 days as well as family

labor, there was no increase in female family labor use per acre between the two rounds of

the survey. The summary table by farm size group is provided in Table A.2.

1.4 Empirical Strategy

The National Employment Guarantee program was implemented in three phases. Phase

I was implemented in February 2006, Phase II was implemented in April-May 2007, and

Phase III was implemented in April 2008. We combine the phase-wise implementation of

the NREGS and REDS 1999 and 2007-08 to assess the program’s impact on agricultural

productivity. When the 2007/8 round of REDS was administered, NREGS had been

implemented in phase I and II districts (treatment group) but not yet in those to be covered

under phase III of the program, allowing us to use the latter as a control group in the

difference-in-differences (DID) regression19. Letting 𝑖 denote households, 𝑣 village, 𝑑

district, and 𝑡 time. We estimate the following equation.

𝑌𝑖𝑣𝑑𝑡 = 𝛽0 + 𝛽1𝑁 𝑅𝐸𝐺𝑆𝑑𝑡 + 𝜆𝑋𝑣 × 𝑅𝑜𝑢𝑛𝑑𝑡 + 𝑍𝑣𝑑𝑡 + 𝜇𝑖 + 𝜙𝑡 + 𝜖𝑖𝑣𝑑𝑡

(1.1)

where 𝑌𝑖𝑑𝑡 is the outcome variable of interest such as output and inputs per acre for

farmer 𝑖, in village 𝑣 and district 𝑑 in year 𝑡, 𝑁 𝑅𝐸𝐺𝑆𝑑𝑡 is an indicator variable that equals

one if district 𝑑 was the part of phases I or II of the program roll-out in year 𝑡 and zero

otherwise; 𝑋 is a vector pre-treatment time constant village level control interacted with

survey round year. 𝑅𝑜𝑢𝑛𝑑 is a dummy variable for the survey round and takes on the value

one if t=2007/08 and zero otherwise. The time-invariant controls are constructed using data

19As discussed above, for Andhra Pradesh, Bihar, Chattisgarh, Haryana, Madhya Pradesh, Maharashtra,
Rajasthan, and Tamil Nadu, we were also able to obtain administrative data on employment generated by
NREGS overall and for irrigation-related activities, mainly cleaning and rehabilitating canals. If implemented
effectively, the latter would quickly improve agricultural productivity even for non-participants while program-
induced wage payments would affect risk coping mainly by direct beneficiaries. Due to delays in implementing
the survey in Kerala, all of the state’s districts had been covered by NREGS at the time the survey was
administered, forcing us to drop observations from this state. Our final analysis sample thus includes 16 of
the original states.

13

4

1

from REDS 1999 data and Census 2001. 𝑍 is a vector of time-varying controls such as

village-level rainfall. 𝜇𝑖 is either district or household fixed effects and 𝜖𝑖𝑑𝑡 is an error term.

𝜙𝑡 includes year-fixed effects, season-fixed effects, as well as stat-by-year fixed effects. We

run the variation of the equation 1.1 to account for other sources of differential changes

when necessary. Our main DID coefficient of interest is 𝛽1, which measures the impact of

NREGS exposure. Standard errors are clustered at the district level throughout.

Identification The identification of DID requires that households in both treated and control

districts follow a similar agricultural productivity growth path in the absence of treatment.

Even though a test of parallel trend is not possible, we can approximate the test using

pre-treatment data. We use the data from 1982 and 1999 of REDS to test for the parallel

pre-trend (see Table A.3). Test of parallel pre-trend is conditioned on household fixed

effects, allowing us to control time-invariant unobservable that might affect selection into

treatment. A notable caveat of using 1982 and 1999 REDS data is that there is a large gap

between the survey periods. Although several challenges, including household splits and

attrition, must be considered, results from tests for parallel trends in pre-program trajectories

can, in principle, be conducted using the 1982 to 1999 rounds of the REDS panel. Results

in Appendix Table A.3 suggest that the hypothesis of parallel pre-trends cannot be rejected.

We also test for parallel pre-trend of the key variables by farm size group. Results are

presented in Table A.4, suggestive of the fact that parallel pre-trends hold for all farm size

groups.

As trends over the 1982-99 period may not be relevant for what occurred closer to

NREGS introduction, we also use production data collected through the cost of cultivation

survey (CCS) by the Ministry of Agriculture in 2000-2005, i.e., directly before NREGS

became available. Contrary to Merfeld (2019) who uses NSS data to do so, we rely on

the cost of cultivation survey because it provides information on key agricultural outcomes,

including total and per acre agricultural output, days of family and hired labor, casual wage

14

rates and value of main inputs (fertilizer, manure, machinery, and irrigation) in Rs/ha 20 We

restrict our sample to the REDS districts and construct a pseudo-panel at the block level.

Results in Appendix Table A.5 suggest that there is no evidence of pre-trends between

treated and non-treated districts just before the program became effective.

1.5 Results

1.5.1 Effects on Agricultural Productivity

The NREGS can affect agriculture directly via land and soil improvement and indirectly

through wages and reallocation of inputs. However, before discussing specific channels

through which NREGS might affect productivity, we start with the evidence of aggregate

5

1

productivity. Using the difference-in-differences (DID) framework discussed in equation

1.1, we estimate the program effects on agricultural productivity measured as real output

per acre.

Results of the aggregate agricultural productivity impact are reported in Table 1.1. In

the first column, we provide the baseline specification and the effect of NREGS estimated

using the difference-in-differences (DID) method after accounting for household, year,

season, and state-by-year fixed effects. By including household fixed effects, we are able to

account for unobservable factors at the household level that could affect productivity. The

inclusion of year-fixed effects helps capture any policy changes at the national level that may

affect productivity. Additionally, we include season and state-by-year dummy variables to

account for seasonal variations and state-specific trends in prices and agricultural policies.

In the subsequent columns, we further enhance our baseline specification from column

1 of Table 1.1 by progressively incorporating additional fixed effects. This allows us to

20These data have been collected on an annual basis by the Directorate of Economics and Statistics in
India’s Ministry of Agriculture since 1971/72 mainly to determine the minimum support price at which the
Government will procure main crops. For sampling purposes, states are divided into homogenous zones
based on cropping patterns, soil quality, rainfall, and irrigation. A multi-stage stratified random sampling
design was adopted, selecting districts and subdistricts (tehsil) in the first stage, villages in the second, and
operational holding in the third stage. Operational holdings are divided into five categories, such as up to 1
ha, 1-2 ha, 2-4 ha, 4-6 ha, and above 6 ha, and at least two holdings are selected from each category.

15

6

1

control for differential changes at the year-season level, changes specific to different crops,

and crop-by-season level variations such as the selection of crops influenced by seasonal

factors.

Table 1.1 Effects of NREGS on Agricultural Productivty

(1)

(2)

(3)

(4)

(5)

(6)

NREGS

0.115
(0.095)

0.108
(0.093)

0.121
(0.084)

0.101
(0.078)

0.127
(0.085)

0.107
(0.079)

Observations
R-squared

14,620
0.462

14,620
0.463

14,620
0.578

14,620
0.581

14,620
0.585

14,620
0.587

Household, year FEs
Season FEs
State-by-year FEs
Season-by-year FEs
Crop FEs
Crop-by-season FEs
Controls

Y
Y
Y

Y
Y
Y
Y

Y
Y
Y
Y
Y

Y
Y
Y
Y
Y

Y

Y
Y
Y
Y

Y

Y
Y
Y
Y

Y
Y

Note: Dependent variable is the log of output value per acre throughout. NREGS is
a dummy and takes on the value one if districts are from phases 1 and 2; otherwise,
zero. Each column reports the coefficients of interest from separate regressions. The
vector of controls includes rainfall for each village and survey year. It also includes
time-constant pre-program village-level controls such as distance to the nearest town,
access to the local market, distance to district HQ, log of the population schools, etc.,
constructed from census 2001 and REDS 1999, interacted with year. Standard errors
are clustered at the district level and reported in parentheses; *** p<0.01, ** p<0.05,
* p<0.1

We also include sets of controls. These controls consist of time-varying rainfall at

the village level and time-constant pre-program variables constructed from REDS99 and

Census 2001 interacted with the year variable to account for divergent trends in village-

level developments. We did not find a statistically significant effect of NREGS on aggregate

productivity. Although coefficients are positive and economically meaningful, ranging from

0.107 to 0.127 percentage points, they are not precisely estimated.

16

7

1

1.5.1.1 By Farm-Size

To explore the heterogeneity in effects of NREGS by farm size, we divided farms into

four categories closely following the government of India’s definition 21. We conducted

similar regression analyses as presented in Table 1.2 for each farm size category. The

results revealed that NREGS had a positive impact on agricultural productivity for small

farmers, specifically those with land holdings of less than 2.5 acres. However, we did not

find any significant effects of NREGS on the productivity of other farmer categories. For

marginal farmers, the implementation of NREGS led to a remarkable 18 percent increase

in agricultural productivity. This increase is equivalent to a gain of 1636 Rs per acre.

Table 1.2 Effects of NREGS on Agricultural Productivity by Farm Size

VARIABLES Marginal (< 2.5 ac.)

Small (2.5-5 ac.) Medium (5-10 ac.)

(1)

(2)

(3)

(4)

(5)

(6)

(7)
(8)
Large (>10 ac.)

NREGS

0.182**
(0.088)

0.177**
(0.081)

0.015
(0.132)

-0.044
(0.125)

0.128
(0.106)

0.071
(0.107)

0.131
(0.137)

0.083
(0.131)

Observations
R-squared

4,267
0.595

Controls

4,267
0.601

Y

3,719
0.576

3,556
0.604

3,719
0.585

Y

3,556
0.609

Y

3,071
0.584

3,071
0.594

Y

Note: Dependent variable is the log of output value per acre throughout. NREGS is a dummy and
takes on the value one if districts are from phases 1 and 2; otherwise, zero. Each column reports the
coefficients of interest from separate regressions. Household and year-fixed effects as well are included
throughout. Regressions are also adjusted for season, season-by-year, state-by-year, and crop-by-season
fixed effects. The vector of controls includes rainfall for each village and survey year. It also includes
time-constant pre-treatment village-level controls such as distance to the nearest town, access to the
local market, distance to district HQ, log of the population schools, etc., constructed from census 2001
and REDS 1999, interacted with year. Standard errors are clustered at the district level and reported in
parentheses; *** p<0.01, ** p<0.05, * p<0.1

The coefficients are stable and effects size are similar for marginal farmers with and

without the controls. We explore in more detail whether the incidence of benefits on

marginal farmers can be attributed to the program being self-targeting, the fact that small

21Our definition of farm size categories closely mimics the definition used by the government of India.
GOI defines farmers as follows marginal (<1 ha) farmer, small (1-2 ha), small-medium (2-4 ha), medium
(4-10 ha).

17

8

1

farmers are likely to be net sellers of labor and thus benefit most from NREGS-induced

wage increases, and the provision of implicit safety nets through program-supported public

goods are disproportionately valuable for marginal farmers.

1.5.2 Effects on Agricultural Labor Use

Regression results of labor use separated by family and hired as well as farmer types

are reported in Table 1.3. In Panel A of Table 1.3, we examine the effect of the NREGS

program on hired labor days in agriculture. We did not observe any significant aggregate

effect of NREGS on hired labor days. However, it is important to consider that the labor

market in India is incomplete, and farmers with different land size holdings may face varying

constraints (Deininger et al., 2018). After separating the sample by farm size groups, we

find that large farms faces labor shortage in hiring labor due to NREGS with an estimated

coefficient of 6.21 per acre. Which is equivalent to a 40 percent decrease in the hired labor

use per acre with respect to control districts. The large farmers offset the decrease in the

hired labor use with intensive use of fertilizer per acre (see Table 1.4).

In panel B of Table 1.3, we report the estimates for family labor use in agricultural

production. The NREGS program is likely to affect the allocation of labor in both the

agricultural and non-agricultural sectors through an increase in wages (Azam, 2011; Berg

et al., 2012; Imbert and Papp, 2015; Merfeld, 2019). Our estimated coefficient for family

labor is positive and statistically significant for marginal farmers, indicating that they are

more likely to work on their own farms. This finding can be rationalized by the fact

that NREGS reduced out-migration by 22 percent (Imbert and Papp, 2020). Moreover, as

NREGS provides 100 days of job opportunities and supposes to complement agricultural

activities by providing more work in the lean season, it may reduce the need for farmers to

search for additional employment, allowing them to dedicate more time to their own farms.

The observed increase in family labor usage is primarily driven by more male members

of the family working on their own farms. This aligns with the notion of reduced migration,

18

9

1

Table 1.3 NREGS and Agricultural Labor Use

VARIABLES

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

(5)
Large

NREGS

Observations
R-squared

NREGS

Observations
R-squared

Panel A: Hired Labor Days

-4.15
(3.54)
4,267
0.62

-6.30
(4.16)
3,719
0.63

Panel B: Family Labor Days

12.34**
(5.58)
4,267
0.62

2.43
(7.16)
3,719
0.54

-1.13
(3.57)
3,556
0.63

0.91
(5.85)
3,556
0.53

-3.69
(2.74)
14,620
0.60

4.43
(5.00)
14,620
0.59

-6.21**
(2.90)
3,071
0.58

-4.57
(3.04)
3,071
0.58

Note: Dependent variable is labor days per acre throughout. NREGS is a dummy and
takes on the value one if districts are from phases 1 and 2; otherwise, zero. Each column
reports the coefficients of interest from separate regressions. Marginal, small, medium,
and large farmers are defined as owning less than 2.5 acres, 2.5-5 acres, 5-10 acres, and
more than 10 acres, respectively. Household and year-fixed effects as well are included
throughout. Regressions are also adjusted for season, season-by-year, state-by-year, and
crop-by-season fixed effects. The vector of controls includes rainfall for each village and
survey year. It also includes time-constant pre-treatment village-level controls such as
distance to the nearest town, access to the local market, distance to district HQ, log of the
population schools, etc., constructed from census 2001 and REDS 1999, interacted with
year. Standard errors are clustered at the district level and reported in parentheses; ***
p<0.01, ** p<0.05, * p<0.1

as males are typically more prone to out-migrate during lean agricultural seasons in search

of employment (Table A.6). Furthermore, we also test whether it is true that marginal

farmers are working on their farms more than the control groups in the appendix (Table

A.8). We utilize the member-level labor data for all households and find that in districts

with the program, marginal farmers are 7% more likely to work on their own farms.

1.5.3 Mechanization and Input intensification

While substituting family for hired labor is one way to respond to higher wages, it has

limited scope and may be associated with undesirable side effects. As rental markets for

ploughing services to support field preparation are active in India (Binswanger and Singh,

2018), mechanization is an option that can be pursued in parallel or as an alternative. In

19

0

2

Table 1.4, we present the estimated coefficient of the program on the intensification of input

use. In Panel A of Table 1.4, we do not find any significant effects of NREGS on machinery

use per acre overall. However, there was an observed increase in machinery usage per acre,

specifically among marginal farmers. The estimated increase in machinery rental changes

amounted to 70 Rs per acre, which is equivalent to a 23% increase compared to the mean

in control districts.

Although the coefficients are positive for large farms as well but are not precisely

estimated, this suggests that expansion of rental equipment use was more pronounced for

marginal farmers than for large farmers, consistent with the notion that the marginal labor-

saving effects of using small-scale equipment are highest for marginal producers (Bhargava,

2014). Such a finding is also consistent with the self-targeting nature of the program and

the fact that farmers who primarily sell their labor will benefit the most from mechanization

as the program could increase available liquidity for those who were previously constrained

In panel B of Table 1.4, we report the estimated coefficient of NREGS on fertilizer use.

The estimated coefficient for fertilizer with respect to the program is positive across all farm

size groups but only statistically significant for large farmers. The increase for large farms is

about 129 Rs/acre, which is equivalent to an increase of 26 percent from the control group.

The increase in fertilizer intensity for large farms net out the effects of a decrease in hired

labor per acre for large farmers, which was about 33 percent (see Table 1.3). In panel C, we

report the effects on irrigation intensity of the program and find that NREGS increases the

intensive use of irrigation coupled with an increase in family labor use for marginal farmers.

The increase in the family labor use for marginal farmers is almost entirely driven by more

males working more days in agriculture in the family. This is also consistent with the

notion that irrigation is almost entirely carried out by men since it requires complementary

physical laborers.

20

1

2

Table 1.4 Agricultural Inputs

VARIABLES

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

(5)
Large

NREGS

Observations
R-squared

NREGS

Observations
R-squared

NREGS

Observations
R-squared

11.05
(49.65)
14,620
0.62

61.83
(57.27)
14,620
0.55

33.78
(44.18)
14,620
0.60

Panel A: Machinery
1.93
70.05*
(66.11)
(37.97)
3,719
4,267
0.64
0.68

Panel B: Fertilizer
32.11
99.74
(73.60)
(64.13)
3,719
4,267
0.55
0.59

Panel C: Irrigation
50.78
(44.23)
3,719
0.56

120.43*
(64.99)
4,267
0.67

-37.18
(81.34)
3,556
0.58

8.47
(76.89)
3,556
0.53

-38.24
(40.41)
3,556
0.54

45.87
(74.87)
3,071
0.58

129.20*
(70.38)
3,071
0.55

-7.35
(22.40)
3,071
0.53

Note: Dependent variables are per acre throughout. NREGS is a dummy and takes on
the value one if districts are from phases 1 and 2; otherwise, zero. Each column reports
the coefficients of interest from separate regressions. Marginal, small, medium, and large
farmers are defined as owning less than 2.5 acres, 2.5-5 acres, 5-10 acres, and more than
10 acres, respectively. Household and year-fixed effects as well are included throughout.
Regressions are also adjusted for season, season-by-year, state-by-year, and crop-by-season
fixed effects. The vector of controls includes rainfall for each village and survey year. It also
includes time-constant pre-treatment village-level controls such as distance to the nearest
town, access to the local market, distance to district HQ, log of the population schools, etc.,
constructed from census 2001 and REDS 1999, interacted with year. Standard errors are
clustered at the district level and reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

1.5.4 High Value Crops

Public works programs can effectively mitigate income uncertainty by enhancing con-

stant income support and increasing liquidity for farmers by providing additional employ-

ment opportunities. This improved financial stability, in turn, enables them to allocate

a greater portion of their resources toward cultivating high-value crops Gehrke (2019).

Diversification of output beyond the traditional staples of rice and wheat and increases in

cropping intensity have long been identified as essential for Indian agriculture to achieve

higher levels of productivity and competitiveness. Crops other than rice and wheat are,

however, riskier as they are not covered by government-imposed floor price schemes and,

21

2

2

especially if grown in the off-season, more vulnerable to climatic risk. Intensification via

multi-cropping requires liquidity for the timely provision of inputs (including field prepa-

ration) and, where water is scarce, involves higher risk and the need to coordinate water

supply. Changes in the number of crops or the share of area devoted to non-traditional crops

could thus be interpreted as proxy for farmers’ higher ability to assume risk or to manage a

more complex production process better.

Table 1.5 NREGS Increased Share of High-Value Crops

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

(5)
Large

NREGS

Observations
R-squared

0.031*
(0.018)

14,620
0.821

0.027*
(0.015)

4,267
0.802

High Value Crops
0.052**
(0.025)

3,719
0.834

0.034
(0.028)

3,556
0.815

0.030
(0.027)

3,071
0.835

Note: Dependent variable is dummy if the crops are of high value. These crops include
pulses, oilseeds, fiber crops, and fruits and nuts. NREGS is a dummy and takes on the
value one if districts are from phases 1 and 2; otherwise, zero. Each column reports the
coefficients of interest from separate regressions. Marginal, small, medium, and large
farmers are defined as owning less than 2.5 acres, 2.5-5 acres, 5-10 acres, and more than
10 acres, respectively. Household and year-fixed effects as well are included throughout.
Regressions are also adjusted for season, season-by-year, state-by-year, and crop-by-season
fixed effects. The vector of controls includes rainfall for each village and survey year. It also
includes time-constant pre-treatment village-level controls such as distance to the nearest
town, access to the local market, distance to district HQ, log of the population schools, etc.,
constructed from census 2001 and REDS 1999, interacted with year. Standard errors are
clustered at the district level and reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

Results in Table 1.5 suggest that NREGS helped increase the share of high-value crops

grown by 3.1 percentage points. This is consistent with the notion of NREGS having

enhanced farmers’ liquidity and risk-bearing capacity Gehrke (2019). The size of estimated

effects, especially for increases in high-value crops, is most pronounced among marginal

and small farmers whose limited land endowment, together with the weakness of rental

markets, may preclude output increases via expanding cultivated area.

22

3

2

1.6

Impacts of Public Goods Provision

A key potential advantage of NREGS as a public work rather than a cash transfer

program is the scope for creating productivity-enhancing public good infrastructure such

as watersheds, rural roads, land improvement, and minor irrigation works (see Figure 1.2).

While they could generate high returns, the lack of data on the location and quality of

such works and the rather unreliable nature of data on labor input provided by the NREGS

management information system (Bose and Das, 2018) has traditionally made it difficult to

study this issue.

We use primary data collected through the Socio-Economics Profile of India (SEPRI)

survey- digitized from village records -that often are different from official MIS figures-

for a subset of states in our sample. The data include detailed information on NREGS and

contains information about the number of days spent under NREGS overall and by type of

activity. We have data on employment generated and total investment (both capital and labor

expenditure) in NREGS programs for 192 villages in 13 states. One way of assessing if

public goods created under NREGS affect agricultural productivity is to include information

on village-level employment generation and investment in the NREGS program as a shifter.

To do so we estimated reduced form equation of the following form.

𝑌𝑖𝑣𝑡 = 𝛽0 + 𝛽1𝐼𝑛 𝑓 𝑟𝑎𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒𝑣𝑡 + 𝜆𝑋𝑣 × 𝑅𝑜𝑢𝑛𝑑𝑡 + 𝑍𝑣𝑡 + 𝜇𝑖 + 𝜙𝑡 + 𝜖𝑖𝑣𝑡

(1.2)

where 𝑌𝑖𝑣𝑡 is the log of output value per acre produced by farm 𝑖 in village 𝑣 at time 𝑡,

𝐼𝑛 𝑓 𝑟𝑎𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒𝑣𝑡 is the log transformation of either employment generated under NREGS

or per capita investment on NREGS-related projects at time 𝑡 in village 𝑣. 𝑋 is a vector of

time-invariant controls that interacted with the survey round year. 𝑅𝑜𝑢𝑛𝑑 is an indicator

variable and takes on the value one if the survey round is 2015/16 and zero otherwise. 𝑍

is a vector of time-varying controls such as rainfall. 𝜇𝑖 is household fixed effects, 𝜙𝑡 is

year fixed effects, season fixed effects, state-by-year fixed effects and 𝜖𝑖𝑣𝑡 is a white noise

23

4

2

error term. The parameter of interest is 𝛽1, which can be interpreted as the elasticity of

agricultural output with respect to NREGS labor input (or investment per capita). It differs

from the estimates provided earlier in that it reflects implementation arrangements and

investment opportunities in a village and is thus more akin to an average treatment effect (a

dose-response) rather than an intention to treat.

1.6.1 Results

We report the results estimated using equation 1.2 in Table 1.6. Panel A of Table 1.6

reports results for all farmers. In columns 1 and 2, We report the output elasticity with

respect to investment per capita and employment generated. Since NREGS can induce

demographic shifts, we use population from 2001 to convert expenditure into per capita.

Results without controls are reported in columns 1 and 3 and with controls in columns

2 and 4. Results reported in Table 1.6 point towards a positive productivity effect of

NREGS-related investment and employment days with an estimated elasticity of 0.037

(col. 2) to .018 (col. 4), contrary to what was found elsewhere (Deininger and Liu,

2019).

In other words, We find that a 1 percent increase in NREGS related per capita

investment increases agricultural productivity by 3.5-3.7 percent. While a one percent

increase in total employment generated through NREGS, i.e., an increase of 42 days,

increases agricultural productivity by 2 percent overall, which is equivalent to 853 Rs/ac.

The results are robust to the inclusion of controls. If the program is implemented effectively,

productivity improvements due to NREGS-generated public goods would thus recover some

of the annual program costs.

1.6.1.1 By Farm Sizes

We further explore the heterogeneity in the impact of NREGS-related public goods

investment. Panel B, C, D, and E report estimated results by farm size groups. The

coefficients in Table 1.6 regarding the program effects from NREGS-provided public goods

24

5

2

Table 1.6 NREGS Related Investment and Productivity by Farm Size

(1)

(2)

Expenditure

(3)

(4)

Employment

Infrastructure

Observations
R-squared

Infrastructure

Observations
R-squared

Infrastructure

Observations
R-squared

Infrastructure

Observations
R-squared

Infrastructure

Observations
R-squared
Controls

0.035**
(0.017)
12,949
0.623

0.052**
(0.024)
5,245
0.674

0.043***
(0.015)
3,198
0.695

0.016
(0.025)
2,541
0.682

-0.001
(0.027)
1,348
0.615

Panel A: Overall

0.037**
(0.017)
12,949
0.624

0.016*
(0.009)
12,949
0.622

Panel B: Marginal (<2.5 acre)

0.061**
(0.029)
5,245
0.680

0.021
(0.015)
5,245
0.673

Panel C: Small (2.5-5 acre)
0.019*
(0.011)
3,198
0.694

0.048***
(0.014)
3,198
0.697

Panel D: Medium (5-10 acre)

0.036**
(0.014)
2,541
0.686

0.013
(0.013)
2,541
0.682

Panel E: Large (>10 acre)
0.008
(0.011)
1,348
0.615

-0.021
(0.014)
1,348
0.621
Y

0.018**
(0.009)
12,949
0.624

0.028*
(0.016)
5,245
0.679

0.027**
(0.010)
3,198
0.697

0.021**
(0.010)
2,541
0.686

0.001
(0.010)
1,348
0.621
Y

𝐼𝑛 𝑓 𝑟𝑎𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒
Note: Dependent variable is the log of output per acre throughout.
is the log of per capita investment in NREGS projects such as watersheds, roads, etc.,
per person for columns 1 and 2, and NREGS generated total employment for columns
3 and 4. Each column reports the coefficients of interest from separate regressions.
Household and year-fixed effects are included throughout. Regressions are also adjusted
for season, season-by-year, state-by-year, and crop-by-season fixed effects. The vector of
controls includes rainfall for each village and survey year. It also includes time-constant
pre-treatment village-level controls such as distance to the nearest town, access to the
local market, distance to district HQ, log of the population schools, etc., constructed from
census 2001 and REDS 1999, interacted with year. Standard errors are clustered at the
district level and reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

25

6

2

do not significantly differ across farm size groups when using employment generated as

the treatment variable. With an F-statistic of 1.48, we fail to reject the null hypothesis that

the coefficients across all farm size groups are the same (col. 4). However, when we use

NREGS-related per capita expenditure as our explanatory variable, we do find a significant

difference in the incidence of productivity effects across farm types (with an F-statistic of

2.99). The estimated output elasticities with respect to total investment per capita are 6%

for marginal farmers, 5% for small farmers, and 4% for medium farmers. The estimated

coefficient is not statistically significant for large farmers. Moreover, the estimated output

elasticities with respect to employment are 3% for marginal and small and 2% for medium

size farmers.

1.7 Conclusions

We explore the short-term productivity effects of NREGS using a DID strategy on

national farm-level data. We find weak support for aggregate productivity,i.e., estimated

coefficients are imprecisely estimated, but strong evidence that NREGS increased agri-

cultural productivity for marginal farms by 18 percent. The increase in productivity for

marginal farmers appears to be due to the intensive use of family labor, a fact that is con-

sistent with reduced migration and job search friction, mechanization, and more intensive

use of irrigation. Given the nature and size of the program, such an indirect effect on

agricultural productivity is large enough to recover a significant amount of program cost.

Furthermore, the positive effect of NREGS on agricultural productivity has the potential

to stimulate additional economic activity. By increasing the demand for local non-tradable

goods and services, the program can contribute to the growth of non-agricultural employ-

ment (Deininger et al., 2018; Emerick, 2018)), and promote ongoing development in rural

areas (Cook and Shah, 2022). However, to what extent they will be sustained depend on

various factors, such as the availability of investment opportunities for mechanization to

replace labor and the functioning of the land market to facilitate the expansion of cultivated

26

land. Moreover, local governance and supportive policies are essential to maintain the

momentum of development and ensure that the positive impacts are sustained in the long

term.

7

2

27

REFERENCES

Afridi, Farzana, Abhiroop Mukhopadhyay, and Soham Sahoo. 2016. “Female labor force
participation and child education in India: evidence from the National Rural Employment
Guarantee Scheme.” IZA Journal of Labor & Development 5 (1):1–27.

Alik-Lagrange, Arthur and Martin Ravallion. 2018. “Workfare versus transfers in rural

India.” World Development 112:244–258.

Amaral, Sofia, Siddhartha Bandyopadhyay, and Rudra Sensarma. 2015. “Employment
programmes for the poor and female empowerment: The effect of NREGS on gender-
based violence in India.” Journal of interdisciplinary economics 27 (2):199–218.

Azam, Mehtabul. 2011. “The impact of Indian job guarantee scheme on labor market

outcomes: Evidence from a natural experiment.” Available at SSRN 1941959 .

8

2

Babu, V Suresh, C Dheeraja, G Rajani Kanth, and SV Rangacharyulu. 2014. “Frequently
Asked Questions on MGNREGA Operational Guidelines.” Ministry of Rural Develop-
ment, Government of India :02.

Banerjee, Rakesh and Riddhi Maharaj. 2020. “Heat, infant mortality, and adaptation:

Evidence from India.” Journal of Development Economics 143:102378.

Berg, Erlend, Sambit Bhattacharyya, Rajasekhar Durgam, and Manjula Ramachandra. 2012.

“Can rural public works affect agricultural wages? Evidence from India.” .

Berg, Erlend, Sambit Bhattacharyya, Durgam Rajasekhar, and R Manjula. 2014. “Can public
employment schemes increase equilibrium wages? evidence from a natural experiment
in india.” .

Besley, Timothy and Stephen Coate. 1995. “The design of income maintenance pro-

grammes.” The Review of Economic Studies 62 (2):187–221.

Bhargava, Anil K. 2014. “The impact of India’s rural employment guarantee on demand

for agricultural technology.” .

Binswanger, Hans P and Sudhir K Singh. 2018. “Wages, prices and agriculture: how
Journal of Agricultural Economics

can Indian agriculture cope with rising wages?”
69 (2):281–305.

Bose, Nayana and Shreyasee Das. 2018. “Political reservation for women and delivery of

public works program.” Review of Development Economics 22 (1):203–219.

28

9

2

Coady, Mr David and Delphine Prady. 2018. Universal basic income in developing coun-

tries: Issues, options, and illustration for India. International Monetary Fund.

Coffey, Diane. 2013. “Children’s welfare and short-term migration from rural India.” The

journal of development studies 49 (8):1101–1117.

Cook, C Justin and Manisha Shah. 2022. “Aggregate effects from public works: Evidence

from india.” The Review of Economics and Statistics 104 (4):797–806.

Dasgupta, Aparajita. 2017. “Can the major public works policy buffer negative shocks in
early childhood? Evidence from Andhra Pradesh, India.” Economic Development and
Cultural Change 65 (4):767–804.

Deininger, Klaus, Songqing Jin, Yanyan Liu, and Sudhir K Singh. 2018. “Can labor-
market imperfections explain changes in the inverse farm size–productivity relationship?
Longitudinal evidence from rural India.” Land Economics 94 (2):239–258.

Deininger, Klaus and Yanyan Liu. 2019. “Heterogeneous welfare impacts of national
rural employment guarantee scheme: Evidence from Andhra Pradesh, India.” World
Development 117:98–111.

Desai, Sonalde, Prem Vashishtha, Omkar Joshi et al. 2015. “Mahatma Gandhi national

rural employment guarantee act: A catalyst for rural transformation.” Tech. rep.

Dreze, Jean and Amartya Sen. 1990. Hunger and public action. Clarendon Press.

Dutta, Puja, Rinku Murgai, Martin Ravallion, and Dominique Van de Walle. 2012. “Does
India’s employment guarantee scheme guarantee employment?” Economic and Political
Weekly :55–64.

Emerick, Kyle. 2018. “Trading frictions in Indian village economies.” Journal of Develop-

ment Economics 132:32–56.

Fetzer, Thiemo. 2020. “Can workfare programs moderate conflict? Evidence from India.”

Journal of the European Economic Association 18 (6):3337–3375.

Garg, Teevrat, Maulik Jagnani, and Vis Taraz. 2020. “Temperature and human capital in
India.” Journal of the Association of Environmental and Resource Economists 7 (6):1113–
1150.

Gehrke, Esther. 2019. “An employment guarantee as risk insurance? Assessing the effects
of the NREGS on agricultural production decisions.” The World Bank Economic Review
33 (2):413–435.

29

0

3

Gehrke, Esther and Renate Hartwig. 2018. “Productive effects of public works programs:

What do we know? What should we know?” World development 107:111–124.

Godfrey-Wood, Rachel and Benjamin CR Flower. 2018. “Does guaranteed employment
promote resilience to climate change? The case of India’s Mahatma Gandhi National
Rural Employment Guarantee Act (MGNREGA).” Development Policy Review 36:O586–
O604.

Imbert, Clement and John Papp. 2015. “Labor market effects of social programs: Evidence
from india’s employment guarantee.” American Economic Journal: Applied Economics
7 (2):233–63.

Imbert, Clément and John Papp. 2020. “Short-term migration, rural public works, and urban
labor markets: Evidence from India.” Journal of the European Economic Association
18 (2):927–963.

Islam, Mahnaz and Anitha Sivasankaran. 2015. “How does child labor respond to changes

in adult work opportunities? evidence from nrega.” Harvard University .

Khosla, Saksham et al. 2018. India’s Universal Basic Income: Bedeviled by the Details.

Carnegie Endowment for International Peace Washington.

Klonner, Stefan and Christian Oldiges. 2014. “Safety net for India’s poor or waste of public
funds? Poverty and welfare in the wake of the world’s largest job guarantee program.”
Tech. rep., Discussion Paper Series.

Mani, Subha, Jere R Behrman, Shaikh Galab, and Prudhvikar Reddy. 2014. “Impact of the

NREGS on schooling and intellectual human capital.” .

Merfeld, Joshua D. 2019. “Spatially heterogeneous effects of a public works program.”

Journal of Development Economics 136:151–167.

Ranaware, Krushna, Upasak Das, Ashwini Kulkarni, and Sudha Narayanan. 2015. “MGN-
REGA works and their impacts: A study of Maharashtra.” Economic and Political Weekly
:53–61.

Ravallion, Martin. 2022. “On the Gains from Tradable Benefits-in-kind: Evidence for

Workfare in India.” Economica .

Ravi, Shamika and Monika Engler. 2015. “Workfare as an effective way to fight poverty:

The case of India’s NREGS.” World Development 67:57–71.

Reddy, D Narasimha, A Amarender Reddy, and MCS Bantilan. 2014. “The impact of

30

Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA) on rural
labor markets and agriculture.” India Review 13 (3):251–273.

Santangelo, Gabriella. 2016. “Firms and farms: The impact of agricultural productivity on

the local indian economy.” Unpublished Manuscript .

Shah, Manisha and Bryce Millett Steinberg. 2021. “Workfare and Human Capital Investment

Evidence from India.” Journal of Human Resources 56 (2):380–405.

Shah, Vivek. 2012. “Managing productivity risk through employment guarantees: Evidence

from India.” Unpublished, India School of Business, Hyderabad .

Subbarao, Kalanidhi. 1997. “Public works as an anti-poverty program: An overview of
cross-country experience.” American journal of agricultural economics 79 (2):678–683.

Taraz, Vis. 2019. “Weather shocks, social protection, and crop yields: Evidence from
India.” Social Protection, and Crop Yields: Evidence from India (October 11, 2019) .

1

3

31

APPENDIX

2

3

Figure 1.3 Location of Sample REDS Villages by Phase of NREGS Program Implementation

32

3

3

Figure 1.4 Gender Differences in NREGS Workers

33

4

3

Figure 1.5 Residualized Value of Output Per Acre

34

Table A.1 Summary Statistics of Key Variables by Treatment Status and Year

Control

Treatment

Mean

SD

Mean

SD

Year=1998-99
SD

Mean

Year=2007-08
SD

Mean

Panel A: Household Characteristics
0.95
Male head
50.82
head’s age
0.90
Married head
4.38
Head’education
6.26
Household size
2.17
Working male
2.01
Working female
8.10
Total cropped area
5.65
Land owned
3.86
Irrigated land

0.22
13.21
0.31
4.14
3.09
1.28
1.13
9.39
6.42
5.65

0.94
51.29
0.89
5.14
6.32
2.14
2.05
6.69
5.47
2.69

0.24
13.06
0.31
4.55
3.43
1.33
1.23
10.79
8.83
4.76

0.96
49.76
0.90
4.33
6.72
2.29
2.09
7.21
5.95
3.30

0.19
13.35
0.30
3.71
3.58
1.41
1.27
9.34
8.57
5.47

0.93
52.54
0.88
5.32
5.82
2.01
1.96
7.44
5.10
3.12

0.26
12.71
0.32
4.98
2.84
1.17
1.09
11.09
6.94
4.90

9087.31
724.49
738.72
314.74
611.42
183.28
49.34
256.04

7805.25
1449.22
1926.10
368.17
505.11
457.51
152.96
470.68

Panel B: Agricultural Inputs and Output Per Acre
Value (Rs.) per acre...
Harvest value
Seed
Manure
Machinery
Fertilizer
Irrigation
Bullock
Other inputs
Labor days per acre...
Hired labor
Family labor
Total labor
Male hired labor
Female hired labor
Male family labor
Female family labor
Net seller of labor
Net buyer of labor
Share of high value crops

21.43
40.83
52.32
12.14
11.88
32.12
17.41
0.48
0.50
0.41

16.07
37.08
55.65
8.14
7.17
29.07
12.50
0.36
0.56
0.22

7400.42
589.15
543.63
226.09
578.50
153.70
49.03
134.34

6400.45
1117.20
1517.27
327.79
507.97
652.36
147.56
307.40

7784.68
425.58
206.37
206.45
630.09
113.71
55.39
257.11

7049.41
785.04
459.79
290.74
531.64
369.92
157.02
470.99

8717.14
928.03
1153.88
343.07
551.29
232.85
41.73
115.29

7253.44
1666.32
2413.59
398.35
471.87
732.69
141.17
269.73

20.51
35.68
58.40
10.21
9.17
29.27
11.03
0.40
0.56
0.29

24.07
38.18
50.15
14.04
12.98
31.26
15.41
0.49
0.50
0.24

16.21
34.65
53.11
7.46
7.73
25.60
11.91
0.40
0.51
0.27

20.90
38.02
47.45
11.74
12.00
28.43
15.63
0.49
0.50
0.44

20.98
38.39
61.84
11.33
8.80
33.47
11.53
0.36
0.62
0.23

24.94
41.07
55.02
14.49
13.06
34.69
17.31
0.48
0.48
0.42

Note: Our own calculation from Rural Economics and Demographic Survey

35

Table A.2 Summary Statistics of Key Variables by Farm Size

Marginal

Small

Medium

Large

Mean

SD

Mean

SD

Mean

SD

Mean

SD

Panel A: Household Characteristics
Male head
head’s age
Married head
Head’education
Household size
Working male
Working female
Total cropped area
Land owned
Irrigated land

0.93
49.37
0.88
4.35
5.85
1.96
1.84
2.66
1.74
1.05

0.25
12.71
0.33
4.19
2.91
1.17
1.08
3.31
1.93
1.65

0.94
50.17
0.90
4.48
6.04
2.06
1.95
5.04
3.51
2.07

0.24
12.69
0.30
4.29
2.96
1.23
1.11
4.20
2.09
1.93

0.96
51.35
0.90
5.03
6.33
2.21
2.06
8.63
6.31
3.77

0.21
13.25
0.30
4.61
3.09
1.30
1.15
5.99
3.85
4.02

0.97
55.62
0.90
5.89
7.54
2.61
2.48
18.57
15.45
8.66

0.18
13.40
0.30
4.40
4.25
1.56
1.44
18.50
14.06
9.30

1.24

7569.69
1554.20
2116.29
377.62
526.78
673.50
180.82
394.63

8921.50
797.00
864.07
321.58
680.89
280.45
71.87
156.99

Panel B: Agricultural Inputs and Output Per Acre
Net cropped area per season
1.46
Value (Rs.) per acre...
Harvest value
Seed
Manure
Machinery
Fertilizer
Irrigation
Bullock
Other inputs
Labor days per acre...
Hired labor
Family labor
Total labor
Male hired labor
Female hired labor
Male family labor
Female family labor
Net seller of labor
Net buyer of labor
Share of high value crops

18.54
52.27
73.74
9.80
7.66
40.37
16.46
0.58
0.36
0.17

24.66
46.08
56.94
14.45
12.97
36.04
19.55
0.49
0.48
0.37

2.02

1.94

3.21

2.80

5.84

6.36

8244.03
656.73
713.48
282.06
585.65
165.52
53.96
185.08

7275.13
1314.09
1865.42
353.58
503.51
780.88
158.65
382.97

7721.08
591.36
488.93
263.01
559.92
108.79
32.25
205.58

6320.51
1126.36
1377.30
341.71
487.30
276.37
113.99
386.39

7740.05
524.59
400.93
184.98
524.22
82.81
31.48
236.34

7257.26
968.87
1190.88
298.11
488.21
217.49
122.42
431.42

18.68
38.31
58.88
9.00
8.59
30.15
12.55
0.44
0.49
0.25

23.78
39.50
50.61
13.49
12.80
31.14
16.23
0.50
0.50
0.42

18.92
29.92
50.16
9.20
8.79
24.18
9.71
0.27
0.68
0.29

22.35
34.61
46.48
12.75
12.69
27.45
14.10
0.44
0.47
0.46

17.17
19.31
39.78
8.71
7.86
18.24
6.54
0.17
0.79
0.33

19.91
22.28
40.40
11.40
11.15
24.49
11.76
0.37
0.41
0.47

Note: Our own calculation from Rural Economics and Demographic Survey

36

Table A.3 Test of Parallel Pre-Trends for Key Outcome Variables Using REDS 1982-1999
Data

VARIABLES

NREGS x Year(=1999)

Observations
R-squared

(1)

(2)
Cropped area Crop Output Crop Output/ac

(3)

1.185
(1.032)
[0.253]
4,934
0.778

0.196
(0.308)
[0.526]
4,934
0.716

-0.134
(0.262)
[0.610]
4,934
0.732

(4)
Labor/ac

0.164
(0.398)
[0.682]
4,934
0.686

7

3

Machine/ac

Fertilizer/ac

Irrigation/ac

Seed/ac

NREGS x Year(=1999)

Observations
R-squared

-0.210
(0.469)
[0.655]
4,934
0.628

0.311
(0.324)
[0.338]
4,934
0.773

-0.172
(0.365)
[0.639]
4,934
0.644

-0.122
(0.207)
[0.558]
4,934
0.806

Pesticide/ac

Manure/ac

Animal/ac

Other Inputs/ac

NREGS x Year(=1999)

Observations
R-squared

-0.117
(0.410)
[0.776]
4,934
0.725

-0.268
(0.291)
[0.359]
4,934
0.748

0.204
(0.354)
[0.567]
4,934
0.596

0.080
(0.380)
[0.833]
4,934
0.693

Note: NREGS is a dummy and takes on the value one if districts are from phases 1 and 2;
otherwise, zero. Each column reports the coefficients of interest from separate regressions.
Household and year-fixed effects as well are included throughout. Standard errors are clustered
at the district level and reported in parentheses as well as associated p-value in the brackets; ***
p<0.01, ** p<0.05, * p<0.1

37

Table A.4 Test of Parallel Pre-Trends by Farm Size Using REDS

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)
Per Acre

(9)

(10)

(11)

(12)

VARIABLES

Area

Output

Output/ac

Labor

Seed

Manure

Fertilizer

Pesticide

Irrigation Animal Machine Other inputs

NREGS x Year(=1999)

Observations
R-squared

NREGS x Year(=1999)

Observations
R-squared

NREGS x Year(=1999)

Observations
R-squared

NREGS x Year(=1999)

Observations
R-squared

0.138
(0.152)
[0.365]
1,564
0.801

0.077
(0.249)
[0.757]
1,130
0.879

0.118
(0.618)
[0.849]
1,177
0.859

2.313
(3.771)
[0.541]
1,063
0.829

-0.398
(0.963)
[0.681]
1,564
0.806

0.643
(0.679)
[0.346]
1,130
0.878

0.222
(0.559)
[0.692]
1,177
0.913

-0.355
(0.395)
[0.370]
1,063
0.942

-1.183
(0.995)
[0.238]
1,564
0.817

0.521
(0.596)
[0.384]
1,130
0.896

0.104
(0.482)
[0.830]
1,177
0.927

-0.407
(0.345)
[0.242]
1,063
0.948

-0.084
(1.065)
[0.937]
1,564
0.800

-0.141
(0.967)
[0.885]
1,130
0.912

0.196
(0.794)
[0.806]
1,177
0.893

0.296
(0.713)
[0.679]
1,063
0.871

-0.638
(0.655)
[0.332]
1,564
0.848

0.463
(0.649)
[0.477]
1,130
0.937

-0.152
(0.533)
[0.776]
1,177
0.947

-0.306
(0.422)
[0.471]
1,063
0.936

-0.615
(0.654)
[0.349]
1,564
0.828

0.535
(0.816)
[0.514]
1,130
0.911

-0.686
(0.684)
[0.319]
1,177
0.919

-0.123
(0.717)
[0.864]
1,063
0.866

Marginal
-0.719
(0.683)
[0.295]
1,564
0.842

Small

0.995
(0.922)
[0.283]
1,130
0.907

Medium
0.377
(0.985)
[0.703]
1,177
0.912

Large

0.282
(0.529)
[0.596]
1,063
0.936

0.197
(0.664)
[0.767]
1,564
0.808

0.000
(0.959)
[1.000]
1,130
0.909

-0.599
(0.806)
[0.459]
1,177
0.923

-0.153
(0.848)
[0.857]
1,063
0.878

-0.922
(0.945)
[0.332]
1,564
0.783

0.058
(0.961)
[0.952]
1,130
0.878

-0.318
(0.906)
[0.726]
1,177
0.860

-0.635
(0.729)
[0.386]
1,063
0.813

0.440
(0.743)
[0.555]
1,564
0.776

0.178
(1.109)
[0.873]
1,130
0.861

-0.172
(0.731)
[0.815]
1,177
0.849

0.290
(0.475)
[0.543]
1,063
0.731

-0.876
(1.128)
[0.439]
1,564
0.776

0.219
(1.059)
[0.837]
1,130
0.884

0.404
(1.023)
[0.694]
1,177
0.863

-0.031
(1.304)
[0.981]
1,063
0.738

-0.649
(0.739)
[0.382]
1,564
0.790

0.459
(0.810)
[0.572]
1,130
0.906

0.413
(0.924)
[0.656]
1,177
0.886

0.067
(0.615)
[0.914]
1,063
0.879

Note: NREGS is a dummy and takes on the value one if districts are from phases 1 and 2; otherwise, zero. Each column reports the coefficients of interest
from separate regressions. Household and year-fixed effects as well are included throughout. Standard errors are clustered at the district level and reported
in parentheses as well as associated p-value in the brackets; *** p<0.01, ** p<0.05, * p<0.1

38

Table A.5 Test of Parallel Pre-Trends Using Cost of Cultivation Survey Data

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Per Acre

VARIABLES

Wage

Crop Output Crop Output

F. Labor H. Labor Machine

Fertilizer

Irrigation

NREGS x 2001-02

NREGS x 2002-03

NREGS x 2003-04

NREGS x 2004-05

Observations
R-squared

0.015
(0.016)
[0.329]
-0.008
(0.022)
[0.722]
-0.015
(0.021)
[0.477]
-0.007
(0.021)
[0.737]
123,614
0.627

-0.021
(0.054)
[0.699]
0.043
(0.078)
[0.582]
-0.001
(0.078)
[0.988]
-0.010
(0.072)
[0.889]
139,211
0.353

-0.023
(0.037)
[0.543]
0.044
(0.054)
[0.410]
0.033
(0.051)
[0.526]
0.021
(0.050)
[0.674]
139,211
0.393

0.021
(0.052)
[0.692]
-0.025
(0.083)
[0.767]
0.052
(0.079)
[0.511]
0.041
(0.080)
[0.607]
139,211
0.276

0.008
(0.073)
[0.911]
0.083
(0.127)
[0.515]
0.127
(0.117)
[0.279]
0.163
(0.120)
[0.174]
139,211
0.447

0.045
(0.049)
[0.359]
-0.013
(0.081)
[0.868]
-0.018
(0.076)
[0.817]
0.025
(0.078)
[0.752]
139,211
0.446

0.059
(0.078)
[0.446]
0.128
(0.145)
[0.378]
-0.003
(0.136)
[0.981]
0.135
(0.128)
[0.292]
139,211
0.383

-0.045
(0.103)
[0.661]
0.054
(0.138)
[0.693]
0.119
(0.134)
[0.373]
0.152
(0.140)
[0.277]
139,211
0.440

Notes: The cost of cultivation survey is administered by the Ministry of Agriculture, Government of India. The survey
collects plot-level summaries for sub-districts. Each column presents separate regression using population weights; NREGS
is a dummy for phase 1 or 2 districts. Dependent variables are measured in logs throughout, and units for casual wage,
agricultural output, fertilizer, and manure are in Rs. whereas labor, machinery, and irrigation use are in hours. Each
regression also includes crop fixed effects and state-by-year fixed effects, and standard errors are clustered at the district level.
*** p<0.01, ** p<0.05, * p<0.

39

Table A.6 Agricultural Labor Use by Male and Female

VARIABLES

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

NREGS

Observations
R-squared

NREGS

Observations
R-squared

NREGS

Observations
R-squared

NREGS

Observations
R-squared

-1.41
(1.76)
14,620
0.59

-3.15***
(1.12)
14,620
0.61

3.65
(4.13)
14,620
0.57

-1.16
(2.27)
14,620
0.58

Male: Hired Labor Days
-3.33
(2.55)
3,719
0.61

-0.42
(2.22)
4,267
0.59

-0.97
(1.98)
3,556
0.62

Female: Hired Labor Days
-4.72**
(1.83)
3,719
0.63

-4.09***
(1.45)
4,267
0.62

-0.81
(1.81)
3,556
0.63

Male: Family Labor Days
2.35
(5.76)
3,719
0.54

12.72***
(4.39)
4,267
0.59

0.53
(4.27)
3,556
0.54

Female: Family Labor Days
1.68
(3.30)
3,719
0.54

1.94
(2.27)
4,267
0.64

-2.05
(2.15)
3,556
0.53

(5)
Large

-2.40
(1.86)
3,071
0.58

-4.09**
(1.62)
3,071
0.58

-7.03
(4.24)
3,071
0.54

-6.01***
(2.15)
3,071
0.53

0

4

Note: Dependent variable is labor days per acre throughout. NREGS is a dummy and takes
on the value one if districts are from phases 1 and 2; otherwise, zero. Each column reports the
coefficients of interest from separate regressions. Marginal, small, medium, and large farmers
are defined as owning less than 2.5 acres, 2.5-5 acres, 5-10 acres, and more than 10 acres,
respectively. Household and year-fixed effects as well are included throughout. Regressions
are also adjusted for season, season-by-year, state-by-year, and crop-by-season fixed effects.
The vector of controls includes rainfall for each village and survey year.
It also includes
time-constant pre-treatment village-level controls such as distance to the nearest town, access
to the local market, distance to district HQ, log of the population schools, etc., constructed
from census 2001 and REDS 1999, interacted with year. Standard errors are clustered at the
district level and reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

40

Table A.7 No Effects on Total Cropped Area

VARIABLES

NREGS

Observations
R-squared

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

-0.034
(0.107)
14,620
0.688

Total Cropped Area
-0.021
-0.147
(0.144)
(0.150)
3,719
4,267
0.604
0.650

0.042
(0.126)
3,556
0.585

Large

0.086
(0.124)
3,071
0.576

Note: NREGS is a dummy and takes on the value one if districts are from phases 1 and 2;
otherwise, zero. Each column reports the coefficients of interest from separate regressions.
Marginal, small, medium, and large farmers are defined as owning less than 2.5 acres, 2.5-5
acres, 5-10 acres, and more than 10 acres, respectively. Household and year-fixed effects
as well are included throughout. Regressions are also adjusted for season, season-by-year,
state-by-year, and crop-by-season fixed effects. The vector of controls includes rainfall
for each village and survey year. It also includes time-constant pre-treatment village-level
controls such as distance to the nearest town, access to the local market, distance to district
HQ, log of the population schools, etc., constructed from census 2001 and REDS 1999,
interacted with year. Standard errors are clustered at the district level and reported in
parentheses; *** p<0.01, ** p<0.05, * p<0.1

1

4

41

Table A.8 Labor Supply Decision

(1)
(2)
Self Employed

(3)

(4)

(5)

Worked as Labor in Activities. . .

VARIABLES

Agr.

Nonagr.

Daily wage

Salaried

HH

2

4

NREGS x landless

NREGS x marginal

NREGS x small

NREGS x medium

NREGS x large

Observations
R-squared

-0.013
(0.040)
0.073**
(0.036)
0.005
(0.029)
-0.012
(0.037)
0.054
(0.049)
53,764
0.452

-0.002
(0.013)
-0.008
(0.013)
-0.006
(0.012)
-0.018
(0.013)
0.008
(0.012)
53,764
0.322

0.061*
(0.035)
0.002
(0.037)
0.017
(0.030)
0.023
(0.035)
0.026
(0.021)
53,764
0.461

0.002
(0.010)
-0.003
(0.009)
-0.016*
(0.009)
0.006
(0.011)
0.014
(0.012)
53,764
0.253

-0.064
(0.049)
-0.028
(0.046)
-0.042
(0.046)
-0.044
(0.057)
-0.043
(0.053)
53,764
0.555

Note: NREGS is a dummy and takes on the value one if districts are from phases 1 and 2; otherwise,
zero. Each column reports the coefficients of interest from separate regressions. Household and
year-fixed effects as well are included throughout. The vector of controls includes rainfall for each
village and survey year. It also includes time-constant pre-treatment village-level controls such as
distance to the nearest town, access to the local market, distance to district HQ, log of the population
schools, etc., constructed from census 2001 and REDS 1999, interacted with year. Standard errors are
clustered at the district level and reported in parentheses; *** p<0.01, ** p<0.05, * p<0.1

42

3

4

Table A.9 Estimated Productivity Effects Using Two-Stage DID Approach

VARIABLES

(1)
All

(2)
Marginal

(3)
Small

(4)
Medium

(5)
Large

Panel A: No Inputs

NREGS

Observations
R-squared

0.123**
(0.061)

14,620
0.007

0.184**
(0.073)

4,267
0.011

0.054
(0.074)

3,722
0.003

Panel B: With Inputs

NREGS

0.115**
(0.055)

0.179***
(0.066)

0.050
(0.067)

Observations
R-squared

14,620
0.008

4,267
0.013

3,722
0.003

0.125*
(0.068)

3,558
0.009

0.115*
(0.065)

3,558
0.009

0.134
(0.111)

3,073
0.010

0.113
(0.091)

3,073
0.012

Note: Dependent variable is the adjusted log of output value per acre using Gardener’s (2021)
approach throughout. NREGS is a dummy and takes on the value one if districts are from phases 1
and 2; otherwise, zero. Each column reports the coefficients of interest from separate regressions.
Clustered bootstrapped standard errors with 500 replications are reported in parentheses; *** p<0.01,
** p<0.05, * p<0.1

43

4

4

CHAPTER 2

EMPOWERING WOMEN: THE IMPACT OF FEMALE
LEADERSHIP ON LABOR MARKET DYNAMICS

2.1

Introduction

Since the 1990s, India has experienced robust economic growth, declines in fertility,

expansion of education, and improved access to infrastructure. These are all factors that

are generally believed to be associated with sustained increases in female labor force

participation (Klasen, 2019). Yet, India’s female labor force participation rate was 48

percent in 1984, low by global standards to start with, and, in rural areas, declined further

to 33 percent in 2012 (Andres et al., 2017). Reductions in rural women’s labor force

participation e.g. due to life cycle events or exogenous shocks that were intended to be

temporary, often were difficult to reverse and led to permanent dropping out of the labor

force (Sarkar, Sahoo, and Klasen, 2019). The effects of such shifts go well beyond foregone

income as women’s labor market participation status will affect their welfare as well as that

of future generations through impacts on autonomy, outside options, and ability to invest in

children’s education and health (Afridi, Mukhopadhyay, and Sahoo, 2016). Women’s labor

market participation will thus reduce human and physical capital accumulation and India’s

ability to take advantage of its ‘demographic dividend. Identifying ways to reverse or at

least arrest this decline is thus a priority for policy (Fletcher, Pande, and Moore, 2017),

especially in light of the devastation wrought by the COVID pandemic.

The economic literature identifies supply and demand factors as important determinants

of labor market participation. Agricultural mechanization and manufacturing’s rising capital

intensity reduced female labor demand as many women lack the education and skills that

A version of this chapter was previously published in The Journal of Development Studies and is reproduced
with the permission of the co-authors. DOI:10.1080/00220388.2022.2043278 © 2022 International Bank for
Reconstruction and Development/The World Bank

44

5

4

would allow them to shift to higher-paying sectors. This interpretation and the importance

of demand-side rationing are supported by women’s strong response to workfare programs

(Desai, 2018; Sarkar, Sahoo, and Klasen, 2019). On the other hand, some studies argue

that social norms may have interacted with higher real wages in rural areas affected supply

through an income effect (Mehrotra and Parida, 2017), a reaction that may be reinforced by

changes in educated women’s returns to home production vs. market participation (Afridi,

Dinkelman, and Mahajan, 2018).

Beyond economic factors, laws and social norms affect female labor force participation.

Regulations restricting factory work by women during certain hours, such as the 1948

Factory Act may reduce demand for female labor. In rural areas where the role of married

women is widely perceived to be limited to taking care of domestic duties (Bernal, 2008),

having married women work outside the home may reflect badly on their family (Eswaran,

Ramaswami, and Wadhwa, 2013). This, together with the challenges it may pose to their

role, may lead men to oppose their spouses participating in labor markets as reported

from rural Madhya Pradesh (Bernhardt et al., 2018). Yet, social norms change only slowly

(Kandpal and Baylis, 2019) and are closely aligned with broader shifts in culture, perception,

and learning (Fernández, 2013) that often persist over time across generations (Dhar, Jain,

and Jayachandran, 2019).

Recent studies suggest that social and economic change is most likely and sustainable

if they affect demand and supply-side factors. Gender norms are closely linked to social

attitudes (Dhar, Jain, and Jayachandran, 2022) that are not amenable to be changed through

short-term interventions (Jensen, 2012) but can be modified through longer-term exposure

(Dean and Jayachandran, 2019), as through initiatives that capitalize on other Government

programs (Field et al., 2021). Yet, changes in such norms are likely to have far-reaching

effects, including on the likelihood of labor force participation which, via the economic

resources generated, could trigger a virtuous cycle of economic empowerment to reinforce

45

6

4

changes in social norms.

A large body of literature has assessed the impact of reserving village leadership po-

sitions for women on the of village level resources across different types of public goods

(Chattopadhyay and Duflo, 2004) and, possibly after incurring some learning costs (Afridi,

Iversen, and Sharan, 2017), long-term outcomes. Studies suggest that by affecting gendered

stereotypes and attitudes regarding women’s ability (Beaman et al., 2012) the latter can

give women a voice (Iyer et al., 2012), altering long-held beliefs on the status of girls

vs. boys (Kalsi, 2017) and the importance of having adolescent girls enrolled in school

(O’Connell, 2018) that can shift labor market outcomes including self-employment in the

informal manufacturing sector (Ghani, Kerr, and O’Connell, 2014).

By providing predictable labor demand, the National Rural Employment Guarantee

Scheme (NREGS) not only increased wages, especially for women (Azam, 2011), in the dry

season (Imbert and Papp, 2015), and for the unskilled (Berg et al., 2014) but also reduced

short-term migration (Imbert and Papp, 2020), encouraged diversification of cropping

patterns (Gehrke, 2019), and reduced rural violence (Fetzer, 2020).

While India’s combination of reservation of village leadership positions for women that

aims to affect social norms with a direct increase in labor demand via workfare offers an

opportunity to observe the interaction of supply and demand side factors, few studies have

linked the two.

In Uttar Pradesh, Bose and Das (2018) uses data from a large number

of panchayats to show that having a female leader in a reserved position increased female

interest in public works as measured by the number of job cards issued as well as actual

demand for work under the National Rural Employment Guarantee Scheme (NREGS) but

did not affect actual employment outcomes, possibly due to measurement error in the

administrative data they use. In Andhra Pradesh, Afridi, Mukhopadhyay, and Sahoo (2016)

use child-level panel data to show that shifts in female labor force participation brought

about by NREGS improve children’s educational outcomes but do not explicitly discuss

46

7

4

reservation as a potential channel for such effects to materialize.

In this paper, we bring these two strands together more explicitly by using detailed indi-

vidual data to assess the short- and medium-term impact of female reservation together with

NREGS on female labor supply, identify mechanisms that might underpin such changes,

and, for a sub-sample for which such data were collected, explore impacts on female em-

powerment. Identification relies on the fact that, in each period, villages to be reserved

were randomly chosen. We analyze outcomes at the individual level by linking household

data for India’s 12 main states from the Rural Economics and Demographic Survey (REDS)

to villages’ reservation history. As our data were collected when NREGS was active, we

can assess if exogenous exposure to female leadership improved women’s ability to take

advantage of the income-earning opportunities associated with this program, potentially

catalyzing broader changes in their access to and use of resources. Moreover, by providing

estimates of the impact of female reservation in the current and the previous local election

period, we can assess longer-term effects on labor force participation, agency, demand for

work, and involvement in household decision-making.

We find that female contemporaneous reservation of leadership for women affected local

governance as measured by the quality of NREGS implementation but had no measurable

impact on female labor supply. At the same time, female leadership reservation affected

female labor supply to public workfare and private sector labor markets in the following

period. Effects were quantitatively large (half a standard deviation) and most pronounced

for married women. The past reservations also increased women’s income, their demand for

work, and -presumably due to increased access to resources- their participation in household

decisions relating to spending on food items, health, and education.

This paper contributes to the literature in several respects: It points towards a role

model effect as the avenue through which political reservation affects behavioral norms,

women’s economic participation, their control over resources, and their bargaining power.

47

8

4

Although a large part of this effect seems to be brought about by workfare participation

(Deininger, Nagarajan, and Singh, 2020), there is also evidence of reservation-induced

effects on participation in regular labor markets. We also add to the evidence regarding

the impact of gender preferences, including quotas, by showing that, even if women leaders

may lack experience or have to contend with male backlash (Gangadharan et al., 2016) so

that pre-existing gaps cannot be fully closed (Iyer and Mani, 2019), politically empowering

women can have positive effects in the medium term, consistent with the notion that agency

problems may hinder female political participation (Casas-Arce and Saiz, 2015).

The rest of the paper is organized as follows. Section 2 describes institutional background

and context by documenting the paradox of India’s secular decline of female labor force

participation and discusses the origin, nature, and evidence of the impact of the country’s

reservation policy as well as its Employment Guarantee Scheme. Section 3 describes data

and estimation strategy, including balance tests to ascertain that random assignment of

reservation status as mandated by legislation was indeed implemented. Section 4 presents

results regarding the impacts of reservation on (i) female labor force participation (separately

for NREGS-related and other employment) and the heterogeneity of these impacts by marital

status and age; (ii) individual income, desire to work, and participation in household

decision-making; and (iii) voice in terms of affecting the way NREGS is implemented

and tests for their robustness. Section 5 concludes by discussing policy implications and

suggestions for further research.

2.2 Background and Institutional Context

We show that despite favorable external conditions such as increased levels of income and

education and declining fertility over the last decades, India’s female labor force participation

level has declined from an already low level. This is likely due to a combination of supply-

and demand-side factors, including strong social norms. We discuss the background of

political reservation and how reservation of local political leadership for women could

48

9

4

possibly reverse this trend by altering social norms and, in interaction with other Government

policies such as NREGS, generate mutually reinforcing feedback loops between economic

and political empowerment.

2.2.1

India’s Declining Female Labor Force Participation: Evidence and Policy Im-
plications

A large literature explores the determinants and impact of female labor market partici-

pation within and across countries. Early studies often assumed that, due to growth-related

changes in countries’ economic structure, education, and fertility, labor force participation

would increase and then possibly decline with income. While support for this hypothesis

is weak (Gaddis and Klasen, 2014), female empowerment seems a key driver of labor force

participation; from the 1970s, gender-friendly legal reforms consistently led to higher fe-

male labor force participation (Hyland, Djankov, and Goldberg, 2020) and greater female

participation in legislative bodies is also found to be associated with higher female labor

force participation (Lv and Yang, 2018).

India is characterized by some of the most glaring levels of gender inequality globally

and very low female involvement in wage work. Although sustained growth of GDP,

education, and access to key infrastructure (electricity, cooking gas, and piped water) vastly

improved Indians’ lives since the early 1990s, women’s labor force participation stagnated

in urban areas (Klasen and Pieters, 2015) and actually declined in rural ones. Drops in

labor market participation have been pronounced after 2005 for those aged 15-24 years

and married (Andres et al., 2017). While educated women may have decreased their labor

market participation by choice, lack of opportunity and social stigma are key factors for the

less educated.

Agricultural mechanization and increased capital intensity in manufacturing are argued

to have limited opportunities for casual work by low-skilled females (Das et al., 2019). Lower

labor force exit (Sarkar, Sahoo, and Klasen, 2019) and significantly increased workforce

49

0

5

participation by women (Desai and Joshi, 2019) in response to workfare support this notion

and point towards job creation, reinforced by overall growth Fletcher, Pande, and Moore

(2017), as a key to higher female labor force participation (Chatterjee, Murgai, and Rama,

2015).

Social norms also affect female autonomy (Debnath, 2015). In central India, husbands

have been shown to be opposed to their wives’ taking up employment as this would reduce

their social standing (Bernhardt et al., 2018). Changing such attitudes is difficult in the

short term (Dean and Jayachandran, 2019) and requires a long-term approach (Jensen,

2012). Similarly, economically empowering women, e.g. via financial literacy training

and transfer of wage payments to their own bank accounts, did help to increase their labor

supply (Field et al., 2021).

2.2.2 Political Reservations and Females’ Labor Market Outcomes

Together with a constitutional amendment to give more power to local Governments,

reservation of village council leadership positions for women and scheduled castes (SCs)

or tribes (STs) was introduced in India in 1993 to, among others, overcome long-standing

inequalities and discrimination. The share of seats reserved for women was fixed at the state

level. Unlike reservations for SCs, seats to be reserved for women are selected randomly

in every election. Studies have shown that reservation-induced female leadership can

directly alter the nature and quality of public goods supplied locally, e.g., by women leaders

providing goods such as water and roads preferred by women Chattopadhyay and Duflo

(2004). At the same time, their longer-term impact on creating role models (Beaman et al.,

2009) may be equally or even more important as such role model effects can explain the

effects of female leadership reservation on rates of breastfeeding and immunization, as well

as higher child survival (Bhalotra and Clots-Figueras, 2014). Children’s exposure to the

reservation in utero or early in life has also been shown to be associated with better learning

outcomes in primary school (Pathak and Macours, 2017) through role model effects.

50

1

5

Concerns have been raised that reserving leadership positions for females who lack

relevant experience and connections may, in the short term, reduce the quality of program

implementation (Afridi, Iversen, and Sharan, 2017), trigger male backlash (Gangadharan

et al., 2016), or lead to females only standing in for their husbands. Yet, studies show that in

the medium term, such issues may be sorted out, and the policy has considerable potential

to affect a wide range of outcomes. Female leadership has been shown to increase the

level and quality of women’s political participation, their willingness to contribute to public

goods, and their ability to hold leaders accountable (Deininger et al., 2015). Exposure to

female leaders acting as role models triggered higher school enrollment by adolescent girls,

especially those from poorer and less-educated households (O’Connell, 2018). It narrowed

gender gaps (Beaman et al., 2012), improved female labor force participation (Duflo, 2005;

Iyer et al., 2012), and raised girls’ educational attainment and aspiration. Changes in beliefs

regarding gender roles and greater voice by women are argued to be central reasons for the

increased survival of higher-birth-order girls where local seats were reserved for women

(Kalsi, 2017). Enhanced female participation in program oversight, civic engagement,

and electoral participation in ‘reserved’ villages point towards potential complementarities

between political and economic empowerment (Deininger, Nagarajan, and Singh, 2020).

The National Rural Employment Guarantee Scheme (NREGS) has been designed to

expand the demand for unskilled work, especially among women. Building on the country’s

long tradition of food-for-work schemes (Dutta et al., 2012; Subbarao, 1997), this program

guarantees up to 100 days per year to households with registered locally and established

eligibility by obtaining a job card. Unskilled labor supplied by locals is expected to build

productive assets (access roads, water harvesting structures, etc.) to increase agricultural

productivity. NREGS explicitly encourages female participation by paying equal wages to

men and women and requiring that a minimum share of work be performed by women.

While there is considerable heterogeneity in program implementation across states, e.g.,

51

2

5

the use of electronic payment of wages directly into beneficiaries’ accounts (Muralidharan,

Niehaus, and Sukhtankar, 2016), major program-induced effects have been confirmed in

three areas. First, NREGS increased wages, especially for women (Azam, 2011), in the

dry season Imbert and Papp (2015), and for the unskilled (Berg et al., 2014). Second, by

providing a predictable source of income, it helped reduce seasonal short-term migration

(Imbert and Papp, 2020), encouraged diversification of cropping patterns (Gehrke, 2019),

and improved agricultural productivity Deininger et al. (2016). Finally, as the program

is self-targeting, distributional effects have been largely positive: NREGS enhanced con-

sumption (Bose 2017) and asset accumulation by the poor (Deininger et al., 2015). Through

the provision of an effective safety net, it can reduce rural violence (Fetzer, 2020) and has

been shown to have a positive impact on health Ravi and Engler (2015), primary school

participation (Islam and Sivasankaran, 2015), learning outcomes in primary (Mani et al.,

2014) although not secondary (Shah and Steinberg, 2021) school, gender-based violence

(Amaral, Bandyopadhyay, and Sensarma, 2015).

Yet, despite far-reaching positive impacts (Duflo 2005; Iyer et al. 2012), the literature

finds, at best tenuous links between political reservation and labor force participation. Using

state-level data, Ghani et al. (2014) found that female reservation did not increase female

manufacturing employment. A study closely related to ours is Bose and Das (2018), who use

administrative data from 6,000 panchayats in the state of Uttar Pradesh to show that having

a female leader increased the issuance of job cards and women’s demand for NREGS work

without affecting actual female employment. Detailed individual-level data on participation

in labor markets and NREGS governance allow us to (i) assess gender-specific empowerment

effects in greater detail, discerning in particular if reservation affected women’s labor market

participation beyond the increased labor demand by the NREGS program and expand

the sample to cover more than a single state; (ii) reduce measurement error invariably

associated with administrative data and control for covariates at the individual level; and (iii)

52

3

5

explore potential synergies between a role model effect brought about by female leadership

reservation and an empowerment effect arising from independent income earned through

labor market participation.

2.3 Data and Empirical Approach

We use descriptive data to check for the balance in pre-program characteristics between

ever and never reserved villages and differences in program-affected variables that, if

allocation was random, can be interpreted as causal interpretation. Data are consistent

with random allocation of reservation, suggest it brought to power leaders with less formal

education, and point towards gender differences in the impact of reservation on labor market

participation at the extensive and intensive margin

2.3.1 Data and Descriptive Statistics

To explore possible links between political and economic empowerment, we use indi-

vidual data from a complete enumeration of all adult residents in 190 villages in 13 states

implemented in 2014/15 as part of the long-running ARIS-REDS panel. Information was

collected on 275,677 individuals in 91,984 households, of which 23,350, generally the

most disadvantaged ones, had a job card allowing household members to apply for work

under NREGS. As earlier studies have shown, access to job cards was not affected by a

village’s current or past reservation status (Deininger, Nagarajan, and Singh, 2020), and to

obtain a conservative estimate of reservation-induced effects, our analysis focuses on these

households.

In addition to standard demographic and socio-economic characteristics at individual

and household levels, the survey obtained detailed information on actual and desired labor

market participation at the individual level. Individuals who participated in NREGS were

also asked about key program implementation features, including whether dated work

receipts were issued, payment was deposited in beneficiaries’ accounts, and wages were in

53

4

5

line with regulations (or if not, whether a complaint was lodged). A village questionnaire

was administered to, among others, elicit characteristics of all village leaders elected from

2005 together with election details, including if the election was ‘reserved.’ In a sub-sample

of states with traditionally high levels of discrimination against women, an extra module

was administered asking about individuals’ involvement in key household-level decisions.

Following the constitutional mandate for female reservation, most states held local

government elections in 5-year intervals, i.e. in 1995/96, 2000/01, and 2005/06. NREGS

was rolled out in a phased manner starting in 2006 and administered by local governments

that had been recently elected when NREGS was launched in 2006-2008. Another round

of elections was held in 2010 or 2011, and the village council leaders elected then had

just completed their terms when our data were collected. Under the assumption of pre-

program balance, random assignment of female leadership reservation to villages provides

an opportunity to assess if exposure to female leadership in the current or immediately

preceding election period improved women’s ability to take advantage of labor market

opportunities in NREGS or the private sector although we are unable to analyze impacts of

reservation and NREGS separately.

To check the balance in observables between treated and untreated villages before

the reservation was mandated in 1993, supporting the notion of reservation having been

assigned randomly so that results can be given a causal interpretation, we use 1991 Census

data accessed via the Socioeconomic High-resolution Rural-Urban Geographic Platform

(SHRUG) for India (Asher et al., 2021) together with 1990 Economic Census data. The latter

is important as it differentiates the non-agricultural labor force by gender. We combine this

with detailed information on individuals and village leaders from 2014/15. Household and

village characteristics are reported in Tables B.1 and B.2 (see online appendix) separately

for the entire sample (col. 1) and for villages that had or had not been reserved in the two

previous election periods (cols. 2 and 3) with col. 4 reporting p-values for equality of

54

5

5

means in the 2014/15 periods. None of the individual or household characteristics differs

significantly between ever and never reserved villages, suggesting that the policy of random

assignment of villages to female leadership was adhered to (please see the appendix for a

detailed description of the data).

Table 2.1 presents information on individuals’ actual and desired labor market partic-

ipation, involvement in household decision-making, and if they participated in NREGS,

program implementation, and governance with data for males in cols. 1-3 and for females

in cols. 4-6. In line with the literature, data show that labor force participation rates and the

number of days worked by men (87% participation with 189 days worked annually) exceed

those for women (64% and 65 days). Significant gender differences are visible in the way

labor days are allocated across sectors. Men spend close to 50% of their working time

in non-agricultural casual employment, followed by agricultural self-employment (40%),

casual labor in agriculture (34%), and salaried work (7%), and rather limited use of NREGS

(23%) which accounts for less than 5% of their time. Women, by contrast, rely much more

on employment in agriculture and workfare as they spend more than 60% of their time in

agriculture (33% self-employed and 30% in casual labor), followed by NREGS (27%) and

non-agricultural casual labor (10%). Such disproportional reliance on unskilled agricul-

tural work makes women more susceptible to being displaced by agricultural mechanization

(Mehrotra and Parida, 2017) with access to workfare possibly providing a safety net uptake

of which could be affected by women’s voice.

As these variables may be affected by female leadership reservation, tests for the sig-

nificance of differences in cols. 4 and 8 are of interest. We find that reservation-induced

effects are more pronounced for females than for males: while there is no difference in labor

force participation for males between ever (88%) and never (87%) reserved villages, and

males even work and earn significantly more in never (192 days and Rs. 66,000) vs. ever

(186 days and Rs. 63,724) reserved villages, the opposite is true for women for whom labor

55

6

5

force participation (68% vs. 59%), number of days worked per year (68 vs. 62), and total

earnings (Rs. 22,490 vs. Rs. 19,804) are all significantly higher in ever vs. never reserved

villages. At the same time, the willingness to work more is markedly higher for males and

females in ever vs. never reserved villages. The difference is more prominent for women

than for men (9.1 vs. 4.5 percentage points), possibly pointing toward greater rationing for

female labor market participation (Desai, 2018).

The reservation also appears to affect adherence to program rules and, for indicators

in which women were particularly disadvantaged, allowed them to achieve gender parity.

In ever-reserved villages, the share of women who got a dated work receipt and were

paid directly into their bank account increased from 62% to 68% and from 80% to 91%,

respectively. Reservation does not seem to have affected the share of females who were

underpaid (about 45% for ever and never reserved villages) and increased it for males (35%

in never vs. 41% in ever reserved villages), though close to two-thirds of those who did not

get paid the set amount did launch a complaint, much higher than those who did so in never

reserved villages (39% of men, and 46% of women). For the smaller sample where such

data was collected, evidence on involvement in decisions on food, non-food, health, and

education suggests reservation led to significant, though quantitatively modest, increases

in involvement in all these decisions by males as well as females; with 76% in ever vs

70% in never reserved villages, potential reservation-induced effects are largest for females’

participation in education decisions.

Appendix Table B.1 panel A presents data on the 23,350 households with job cards and

their 66,362 working-age members in sample villages. The average household includes 4.5

individuals, has a head who is aged 49 years, spent 3.8 years in school, is married in 85%,

widowed in 13.6%, and female in 11.6% of cases. The data further show that 89% of sample

households are Hindus, 42% belong to scheduled castes or tribes, 58% own agricultural

land, and 48% have a proper (pucca) house. Panel B presents means at the individual level,

56

highlighting that 21% had education at primary, 29% between primary and high school,

and 11% above the high school level. None of the individual or household characteristics

nor the timing of NREGS roll-out differs significantly between ever and never reserved

villages, suggesting that the policy of random assignment of villages to female leadership

was adhered to.

Table 2.1 Labor Force Participation by Reservation Status and Gender

Total

0.877
0.400
0.057
0.340
0.502
0.233
0.071
189.313
20.197
13.992
33.807
93.228
6.672
21.417
65,000
0.274

0.734
0.889
0.388
0.557
33887

0.655
0.828
0.798
0.854
11,628

Men
Reservation status
Never
Ever

Total

Test

Women
Reservation status
Never
Ever

0.882
0.418
0.051
0.353
0.501
0.254
0.068
186.509
22.095
12.684
32.719
91.063
7.525
20.423
63,724
0.296

0.744
0.905
0.412
0.649
17694

Panel A: Labor Supply
0.872
0.380
0.064
0.326
0.504
0.210
0.074
192.376
18.122
15.421
34.996
95.595
5.740
22.503
66,000
0.251

0.008
0.000
0.000
0.000
0.387
0.000
0.030
0.000
0.000
0.000
0.000
0.000
0.000
0.013
0.328
0.000

0.638
0.331
0.013
0.303
0.099
0.272
0.014
65.764
12.490
2.399
23.896
12.455
10.943
3.581
20,986
0.316

0.721
0.865
0.35
0.39
16193

0.315
0.000
0.001
0.000

0.66
0.866
0.454
0.59
32475

0.682
0.357
0.012
0.315
0.103
0.314
0.014
68.846
14.537
2.301
22.528
12.402
13.461
3.617
22,490
0.359

0.684
0.905
0.453
0.658
17017

Panel B: Intra-Household Decision Making

0.669
0.835
0.807
0.862
5,119

0.638
0.819
0.786
0.844
6,509

0.002
0.138
0.007
0.495

0.839
0.761
0.866
0.737
10,839

0.851
0.769
0.875
0.763
4,739

0.590
0.302
0.014
0.291
0.094
0.227
0.013
62.370
10.238
2.507
25.401
12.513
8.171
3.540
19,804
0.268

0.615
0.795
0.463
0.464
15458

0.824
0.751
0.855
0.704
6,100

7

5

Test

0.000
0.000
0.053
0.000
0.010
0.000
0.666
0.000
0.000
0.341
0.000
0.601
0.000
0.974
0.594
0.000

0.002
0.000
0.459
0.013

0.037
0.423
0.532
0.001

in NREGA

Participated in labor market
. . . self-empl. in agric.
. . . self-empl. in non-agric.
. . . casual labor in agric.
. . . casual labor in non-agri.
. . .
. . . regular salaried work
No of days worked
. . . self-empl. in agric.
. . . self-empl. in non-agric.
. . . casual labor in agric.
. . . casual labor in non-agri.
. . .
. . . regular salaried work
Individual income (Rs.)
Would like to work more
If participated in NREGS work
Got dated receipt
Paid directly to bank account
Was paid less than was due
If less, did complain
No. of obs.

in NREGA

Participates in decisions on . . . .
. . . food
. . . nonfood
. . . health
. . . education
No. of obs.

Note: Author’s own calculation from 2014/15 REDS follow-up survey. As discussed in the text, due to
funding constraints data on intra-household decision-making was limited to the states of Gujarat, Uttar
Pradesh, Maharashtra, Orissa, and West Bengal. To test differences in means, p values from regressions with
district fixed effects and standard errors clustered by village panchayat are reported in the last column.

In Appendix Table B.2, village-level data from the 1991 Census (panel A) and the 1990

Economic Census (panel B) suggest that sample villages are typical of rural India with a

population of 412 households (2,231 individuals of which 18% belonged to scheduled castes

57

8

5

and 5% to scheduled tribes. Some 57% of villages can access a good road. An average

village has about 1.4 elementary schools and 0.93 middle/high schools. About half of the

residential buildings have access to electricity, and 40% of the agricultural area is electrified.

Not surprisingly, there were limited nonfarm activities in 1990, which is supported by the

fact that only 105.8 people (equivalent to 4.7% of the total population of an average village)

worked in non-farm employment (panel B). Of the 105.8 employees working in the non-farm

sectors, 89.9 were male, and 15.9 were female. None of these variables differ significantly

between ever and never reserved villages. Pradhan characteristics in panel C suggest that,

in ever-reserved villages, the share of pradhans who either held or contested the position

of village leader before is slightly but not significantly lower in villages that had been

reserved compared to those that had not. At the same time, we find significant differences

in leaders’ attributes between the two types of villages, consistent with the notion that

female reservation opened the way for less educated non-Hindu leaders: while only 26%

and 14% of leaders in ever reserved villages had secondary or, higher education and 48%

were Hindus, corresponding figures for never reserved villages are 42%, 19%, and 64%,

respectively.

2.3.2 Econometric Approach

To assess the impacts of political preferences on women’s economic empowerment, we

use the fact that, in each period, a predetermined share of villages is randomly chosen to

have the leadership position reserved for a woman. Data on current and previous reservation

status allows us to test for the persistence of such effects, i.e., if -in line with the notion

that gender attitudes change slowly with individuals altering their attitude only after having

been exposed to female leadership for some time (Beaman et al., 2012)- past reservation of

a village for female leadership affects current outcomes. Synergies between political and

economic empowerment (Deininger, Nagarajan, and Singh, 2020) would yield the same

result. Letting 𝑣 denote villages, 𝑖 individuals, and 𝑡 time, we assess the impacts of female

58

reservation on outcome variables relating to individual i’s labor force participation as well

as other outcome variables by estimating the following equation.

𝑌𝑖𝑣 = 𝛽0 + 𝛽1𝑅1

𝑣 + 𝛽2𝑅2

𝑣 + 𝛽3𝑋𝑖𝑣 + 𝛽4𝑉𝑣 + 𝜇𝑑 + 𝜖𝑖𝑣

(2.1)

where 𝑌𝑖𝑣 is the outcome variable of interest for individual 𝑖 in village 𝑣, 𝑅1

𝑣 is an indicator

variable that equals one if council leadership in village 𝑣 was reserved for women in the

most recent election (i.e., the pradhan at the time of the survey was a woman who assumed

her position as a result of reservation) and zero otherwise; 𝑅2

𝑣 is an indicator variable that

equals one if council leadership in village 𝑣 had been reserved for a woman in the previous

election and zero otherwise; 𝑋 is a vector of household and individual controls; 𝑉 is a

vector of village and pradhan characteristics; 𝜇𝑑 a district fixed effect; and 𝜖𝑖𝑣 an error term.

Our main interest is in 𝛽1 and 𝛽2, the parameter estimates of current or past reservation on

individual outcomes relative to the base category of a village having never been reserved.

Similarly, 𝛽1 + 𝛽2 denotes the impact of political leadership reservation in a village both

in the present and the past, and it is straightforward to test for this joint impact either being

significantly different from zero or from the estimated coefficients for the current and past

reservation.

To explore the gender dimension of reservation, we let 𝑓𝑖𝑣 be an indicator variable taking

a value of one if the respondent is female and zero otherwise. With interactions between

the respondent’s gender and current or past reservation, our estimating equation becomes:

9

5

𝑌𝑖𝑣 = 𝛽0 + 𝛽1𝑅1

𝑣 + 𝛽3(𝑅1

𝑣 × 𝑓𝑖𝑣𝑡) + 𝛽2𝑅2
𝑣+

(2.2)

𝛽4(𝑅2

𝑣 × 𝑓𝑖𝑣) + 𝛽5𝑋𝑖𝑣 + 𝛽6𝑉𝑣 + 𝜇𝑑 + 𝜖𝑖𝑣

where parameters are as above, and the main difference here is that the parameters

estimated are gender specific. In other words, 𝛽1 and 𝛽2 are the estimated impact of current

or past reservations on men, and 𝛽1 + 𝛽3, as well as 𝛽2 + 𝛽4 are estimated impacts of current

59

and past reservations on women. The significance of any linear combinations of estimated

parameters can be tested via F-tests, which are reported in the results tables throughout.

An equivalent econometrics strategy that provides a more intuitive distinction between

the effects of one-time vs. cumulative female reservation on individual 𝑖′𝑠 labor force

participation and other outcome variables as above is:

𝑌𝑖𝑣 = 𝛼0 + 𝛼1𝑅𝑐

𝑣 + 𝛼2𝑅 𝑝

𝑣 + 𝛼3𝑅𝑐 𝑝
𝑣 +

𝛼4𝑋𝑖𝑣 + 𝛼5𝑉𝑣 + 𝜇𝑑 + 𝜖𝑖𝑣

(2.3)

where 𝑅𝑐

𝑣, 𝑅 𝑝

𝑣 , and 𝑅𝑐 𝑝

𝑣 are indicator variables of whether or not council leadership in

village 𝑣 was reserved for women in the most recent election only; the election previous

to it; or both periods, respectively, and parameters of interest are 𝛼1, 𝛼2, and 𝛼3, the

estimated effect of current only, past only, both current and past reservation on individual

outcomes relative to the base category of a village having never been reserved.

In this

specification, results of which are reported in the appendix tables, the gender dimension of

the reservation are obtained by adding interaction terms between the respondent’s gender

and the reservation variables to (3)

0

6

𝑌𝑖𝑣 = 𝛼0 + 𝛼1𝑅𝑐
𝛼3𝑅𝑐 𝑝

𝑣 + 𝛾3(𝑅𝑐 𝑝

𝑣 + 𝛾1(𝑅𝑐

𝑣 × 𝑓𝑖𝑣𝑡) + 𝛼2𝑅 𝑝

𝑣 + 𝛾2(𝑅 𝑝

𝑣 𝑓𝑖𝑣𝑡)+

𝑣 × 𝑓𝑖𝑣𝑡) + 𝛼4𝑋𝑖𝑣 + 𝛼5𝑉𝑣 + 𝜇𝑑 + 𝜖𝑖𝑣

(2.4)

In this specification, 𝛼1, 𝛼2, and 𝛼3 are the estimated impact of current only, or past

only, or both current and past reservation on men, and 𝛼1 + 𝛾1, 𝛼2 + 𝛾2, as well as 𝛼3 + 𝛾3,

are estimated impacts of current only, past only, and both current and past reservation on

women. The significance of linear combinations of estimated parameters can be tested

via F-tests, which are reported in the results tables throughout. The significance of linear

combinations of estimated parameters can be tested via F-tests, which are reported in the

results tables throughout.

60

1

6

2.4 Results and Discussion

Regressions at the household- and individual level suggest that reservation had no

concurrent impact on female labor force participation but affected modalities of NREGS

implementation, e.g., if work receipts were issued and those receiving less than the stipulated

wage complained. Past reservation is estimated to have led to gains in female labor force

participation at the extensive and intensive margins via a role model effect and associated

increased female labor demand. Combined with an NREGS-induced shift in labor supply,

it, in turn, triggered higher female labor market participation, leading to higher levels of

income and intra-household bargaining power for women.

2.4.1

Impacts of Reservation on Female Labor Market Participation

Table 2.2 reports results from regressions of labor force participation without and with

gender-differentiated effects that correspond to equations (1) and (2) in panels A and B,

respectively. Beyond results for overall participation along the intensive (col. 1) and

extensive (col. 4) margin, estimated coefficients are reported separately for NREGS-related

activities (cols. 3 and 6) and all activities except NREGS (cols. 2 and 5). The equivalent

specification with three indicators for the incidence of reservation as discussed above

(equations (3) and (4)) is in Appendix Table B.3.

Concurrent reservation is estimated to have had no impact on participation at the exten-

sive margin. At the intensive margin, there is some evidence (imprecisely estimated) that

introduction of NREGS crowded out non-NREGS activities with a marginally significant

increase in NREGS days (coefficient of 0.150 in col. 6 of panel A), substituting for a reduc-

tion in non-NREGS-related-labor supply (coefficient of -0.063 in col. 5). Differentiating

by gender in panel B suggests that this is driven by a small contraction of male labor supply.

Thus, we cannot reject the hypothesis that, during the reserved period, reservation has no

impact on the extent or the intensity of overall female labor market participation.

By contrast, we find highly significant gender effects of reservation in the previous

61

2

6

Table 2.2 Estimated Effects of Political Reservation on Labor Supply at Extensive and
Intensive Margin

Participation

No. of days worked

Total

Other

NREGS

Total

Other

NREGS

0.001
(0.008)
0.027***
(0.010)
66,362
0.277

-0.009
(0.010)
0.025**
(0.011)
66,362
0.314

Panel A
0.035
(0.022)
0.063***
(0.019)
66,362
0.241

-0.037
(0.038)
0.168***
(0.048)
66,362
0.375

-0.063
(0.045)
0.138***
(0.046)
66,362
0.404

0.150**
(0.073)
0.202***
(0.061)
66,362
0.243

0.003

0.159

0.000

0.009

0.154

0.000

-0.016
(0.026)
-0.029
(0.023)
0.035
(0.049)
0.118**
(0.046)
66,362
0.282
0.75
0.86
0.62

0.116
0.007
0.480
0.001
0.001

-0.024
(0.024)
-0.005
(0.021)
0.030
(0.045)
0.065
(0.043)
66,362
0.316
0.69
0.84
0.51

0.283
0.066
0.791
0.027
0.023

Panel B
0.028
(0.041)
-0.019
(0.040)
0.015
(0.062)
0.169**
(0.072)
66,362
0.248
0.25
0.23
0.27

0.855
0.043
0.200
0.000
0.000

-0.116
(0.117)
-0.108
(0.116)
0.164
(0.228)
0.574**
(0.228)
66,362
0.379
3.65
4.51
2.69

0.146
0.013
0.700
0.001
0.001

-0.139
(0.102)
-0.013
(0.096)
0.159
(0.190)
0.322
(0.196)
66,362
0.405
3.324
4.34
2.11

0.226
0.036
0.847
0.011
0.009

0.115
(0.148)
-0.110
(0.148)
0.076
(0.225)
0.648**
(0.276)
66,362
0.254
1.466
0.71
0.92

0.983
0.042
0.106
0.001
0.000

Reserved now 𝛽1)

Reserved before (𝛽2)

Observations
R-squared
Test: (p values)
F test (𝛽1 + 𝛽2 = 0)

Reserved now (𝛽1)

Reserved before (𝛽2)

Res now × fem (𝛽3)

Res. before × fem (𝛽4)

Observations
R-squared
Dep. Var Mean
. . . males
. . . females
Test: (p values)
F test (𝛽1 + 𝛽2 = 0)
F test (𝛽3 + 𝛽4 = 0)
F test (𝛽1 + 𝛽3 = 0)
F test (𝛽2 + 𝛽4 = 0)
F test (𝛽1 + 𝛽2 + 𝛽3 + 𝛽4 = 0)

Note: ‘Reserved now’ and ‘Reserved before’ are indicator variables of whether village panchayats are
reserved in the current or the previous panchayat periods and the sample is limited to NREGS job card
holders. Control variables included throughout but not reported include household size, composition,
land ownership, and the head’s marital status, gender, age, and education; village-level access to road,
distance to town and district HQ, population, share of SCs, STs, and key religions; years since the
last village election; pradhan characteristics (education, caste, religion, previous tenure and candidacy
for office) , district fixed effects and for individual-level regressions individuals’ gender, marital status,
age, education and their squared terms. Standard errors are clustered at village panchayat level Robust
standard errors reported in parentheses and multiple hypotheses tests are adjusted using the Bonferroni
method. p values of the F-test are reported in the table. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

62

3

6

period: the likelihood of labor market participation overall is estimated to have increased

by 2.7 percentage points or 3.6 percent at the mean, an effect comprised of estimated

increases by 6.3 percentage points (25.2%) and 2.5 percentage points (3.6%) for NREGS

and non-NREGS work, respectively (Table 2.2). A similarly significant effect -with an

estimated size of 0.202 and 0.138 percentage points for NREGS- and non-NREGS-related

work, equivalent to a 13.7% and 4.1% increase at the mean, respectively (cols. 6 and 5),

emerges at the intensive margin.

Disaggregating these effects by gender (Table 2.2 panel B) highlights that virtually all

long-term impacts can be attributed to changes in women’s rather than men’s labor market

participation. F-tests in the bottom rows indicate that estimated impacts of past reservations

on women’s labor supply (𝛽2+𝛽4) are significant at the 1% level throughout. Past reservation

is estimated to have led to an 8.9 percentage point increase in the likelihood of female labor

force participation, equivalent to an 11.8% increase at the mean, comprised of estimated

increases of 15 and 6 percentage points in women’s likelihood of participating in NREGS

and non-NREGS work, respectively. With 46 percentage points (12.7%) overall (col. 4)

-31 percentage points (9.3%) for non-NREGS (col. 5), and 54 percentage points (37%) for

NREGS work (col. 6)- estimated effects at the intensive margin are even larger.

While the evidence presented here is thus consistent with the lack of a reservation-

induced impact on female participation in the wage labor force in the short term (Bose and

Das, 2018; Ghani, Kerr, and O’Connell, 2014), in the medium term, it increased women’s

response to labor demand from NREGS and beyond. Estimated coefficients for non-

NREGS work are significant throughout, consistent with the notion that, beyond potential

effects on how workfare was provided and participation in NREGS work, reservation

increased women’s demand for wage work more broadly. While a cumulative income-

induced empowerment effect of participation in NREGS could be consistent with these

facts, we will below discuss evidence on changes in levels and quality of women’s political

63

4

6

participation in line with what was suggested by Deininger et al. (2015). It is suggestive

that the reservation of panchayat leadership positions for women has provided a role model

and thus performed an additional catalytic role.

2.4.2 Heterogeneity of Effects

If, as the literature suggests, the scope for labor market participation is particularly lim-

ited for married women (Eswaran, Ramaswami, and Wadhwa, 2013), reservation-induced

effects may be more pronounced for this group, either by providing them with economic

resources and social connections that they would not otherwise have access to or by helping

to change their husbands’ attitude to general gender roles and particularly female labor force

participation Bernhardt et al. (2018). To test this, we run the above regressions separately

for the sub-samples of married and unmarried individuals.

Results from doing so in Table 2.3 indeed support this notion, suggesting that estimated

effects are consistently more significant and larger for married than for unmarried individ-

uals: First, in contrast to insignificant aggregate effects of concurrent reservation on labor

supply in the total sample, current reservation is estimated to increase married women’s

likelihood of labor force participation by 2.6 percentage points with marginal significance.

The main channel for concurrent effects to materialize is via NREGS-related work, which

is estimated to increase by 4.5 percentage points due to reservation (panel A, col. 3), largely

by substituting for self-employment by males and, with a slightly smaller point estimate,

females. Appendix Table B.4 includes the equivalent specification with three reservation

indicators. Aggregate effects of current reservation on unmarried individuals’ participation

are insignificant (panel B col. 1): while the negative effect on self-employment (col. 2 and

3) is consistent with findings for married individuals, reservation has no significant effect on

NREGS participation by unmarried ones, consistent with the notion that they have access

to different opportunities in the labor market or different returns to work at home (Afridi,

Dinkelman, and Mahajan, 2018).

64

Table 2.3 Estimated Effects of Reservation Status on Labor Force Participation by
Marital Status

Participation

No. of Days Worked

Total

Other NREGS

Total

Other NREGS

Reserved now (𝛽1)

Reserved before (𝛽2)

Res now × fem (𝛽3)

Reserved before × fem (𝛽4)

Observations
R-squared
DepMean
Test:
F test (𝛽1 + 𝛽2 = 0; p val)
F test (𝛽3 + 𝛽4 = 0; p val)
F test (𝛽1 + 𝛽3 = 0; p val)
F test (𝛽2 + 𝛽4 = 0; p val)
F test (𝛽1 + 𝛽2 + 𝛽3 + 𝛽4 = 0; p val)

Reserved now (𝛽1)

Reserved before (𝛽2)

Res now × fem (𝛽3)

Reserved before × fem (𝛽4)

Observations
R-squared
DepMean
Test:
F test (𝛽1 + 𝛽2 = 0; p val)
F test (𝛽3 + 𝛽4 = 0; p val)
F test (𝛽1 + 𝛽3 = 0; p val)
F test (𝛽2 + 𝛽4 = 0=0; p val)
F test (𝛽1 + 𝛽2 + 𝛽3 + 𝛽4 = 0; p val)

-0.013
(0.032)
-0.035
(0.027)
0.039
(0.059)
0.126**
(0.054)
50,323
0.285
0.808

0.144
0.008
0.386
0.003
0.000

-0.026
(0.018)
-0.005
(0.021)
0.020
(0.031)
0.079**
(0.038)
16,039
0.256
0.612

0.251
0.051
0.789
0.004
0.035

Panel A: Married
0.037
(0.051)
-0.033
(0.050)
0.008
(0.075)
0.203**
(0.085)
50,323
0.248
0.286

-0.099
(0.143)
-0.131
(0.136)
0.157
(0.271)
0.613**
(0.267)
50,323
0.419
3.885

-0.024
(0.031)
-0.008
(0.027)
0.038
(0.055)
0.071
(0.053)
50,323
0.338
0.742

0.344
0.082
0.638
0.049
0.021

0.945
0.048
0.241
0.000
0.000

0.182
0.017
0.673
0.001
0.001

Panel A: Unmarried
0.008
(0.021)
0.014
(0.021)
0.030
(0.030)
0.057
(0.038)
16,039
0.273
0.145

-0.030
(0.019)
0.007
(0.018)
0.012
(0.026)
0.043
(0.029)
16,039
0.265
0.556

-0.168
(0.103)
-0.007
(0.115)
0.177
(0.149)
0.367*
(0.187)
16,039
0.282
2.913

0.336
0.116
0.392
0.024
0.248

0.410
0.077
0.092
0.010
0.001

0.231
0.035
0.938
0.008
0.028

5

6

-0.122
(0.130)
-0.018
(0.124)
0.159
(0.237)
0.332
(0.246)
50,323
0.464
3.567

0.361
0.078
0.767
0.030
0.015

-0.182*
(0.107)
0.041
(0.102)
0.133
(0.115)
0.236
(0.143)
16,039
0.286
2.650

0.260
0.033
0.639
0.016
0.103

0.150
(0.181)
-0.166
(0.181)
0.049
(0.271)
0.776**
(0.325)
50,323
0.249
0.917

0.945
0.045
0.133
0.001
0.000

0.030
(0.073)
0.024
(0.078)
0.132
(0.118)
0.233
(0.155)
16,039
0.300
0.491

0.596
0.073
0.046
0.014
0.002

Note: ‘Reserved now’ and ‘Reserved before’ are indicator variables of whether village panchayats are
reserved in the current or the previous panchayat periods and the sample is limited to NREGS job card
holders. Control variables included throughout but not reported include household size, composition,
land ownership, and the head’s marital status, gender, age, and education; village-level access to road,
distance to town and district HQ, population, share of SCs, STs, and key religions; years since the
last village election; pradhan characteristics (education, caste, religion, previous tenure and candidacy
for office) , district fixed effects and for individual-level regressions individuals’ gender, marital status,
age, education and their squared terms. Standard errors are clustered at village panchayat level Robust
standard errors reported in parentheses and multiple hypotheses tests are adjusted using the Bonferroni
method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

65

6

6

Second, past reservation is estimated to have had a gender-differentiated impact whereby

a reduction in the likelihood of married males’ participation -by 3.5 percentage points- is

more than compensated for by an increase in married females’ propensity to participate

to yield a net increase of 9.1 percentage points overall due to female reservation. Disag-

gregating by type of labor suggests that most of this effect can be attributed to increased

participation in NREGS activities, estimated to increase by 17 percentage points, versus a

6.3 percentage point gain in non-NREGS activities. The comparison of estimated elastici-

ties at the intensive margin between NREGS and non-NREGS demonstrates an even large

difference between the two types of job activities (0.66 for NREGS vs. 0.31 for non-NREGS

activities).

By comparison, for unmarried individuals, past reservation is estimated to have had

smaller effects that do not differ by gender and are less dominated by NREGS. For example,

the past reservation would increase females’ probability to participate in non-NREGS and

NRGES by 5 percentage points and 7.1 percentage points, respectively. The gains in

intensive margins are even more similar as the estimated effects for NREGS and non-

NREGS activities are 0.26 and 0.27 percentage points, respectively.

2.4.3 Exploring Impact Pathways

To explore if reservation affected supply- or demand-side factors, we report effects on

modalities of NREGS implementation that are likely to have affected the supply of jobs and

women’s bargaining power within the household separately.

Results from regressions (1) and (2) with the key indicators of program implementation

in Table 2.4 (or the equivalent regressions in Appendix Table B.5) suggest that current

and past reservations helped improve the quality of program implementation in several

dimensions: The share of those who received a dated receipt for work performed under

NREGS (col. 1) increased significantly during the reserved period and beyond (with

elasticities of 27 percentage points (38%) and 50 percentage points (70%), respectively).

66

7

6

Table 2.4 Estimated Effects of Reservation on NREGS Governance

dated

Get
receipt

Payment
account

to

Payment
less
assessed

than

If less, did
complain

Complaint
addressed

Reserved now (𝛽1)

Reserved before (𝛽2)

Observations
R-squared
Test:
F test (𝛽1 + 𝛽2 = 0; p val)
F test (𝛽1 + 𝛽2 = 0; p val)

Reserved now (𝛽1)

Reserved before (𝛽2)

Res now × fem (𝛽3)

Reserved before × fem (𝛽4)

Observations
R-squared
Dep. var mean
Test:
F test (𝛽1 + 𝛽2 = 0; p val)
F test (𝛽3 + 𝛽4 = 0; p val)
F test (𝛽1 + 𝛽3 = 0; p val)
F test 𝛽2 + 𝛽4 = 0; p val)
F test (𝛽1 + 𝛽2 + 𝛽3 + 𝛽4 = 0)

0.298***
(0.039)
0.740***
(0.041)
6,712
0.605

0.453
0.000

0.292***
(0.037)
0.736***
(0.043)
0.019
(0.019)
0.039*
(0.022)
6,712
0.606
0.712

0.000
0.061
0.000
0.000
0.000

0.032
(0.030)
0.227***
(0.034)
6,712
0.698

0.322
0.000

0.037
(0.029)
0.227***
(0.035)
-0.018
(0.021)
-0.011
(0.008)
6,712
0.699
0.882

0.000
0.266
0.587
0.000
0.002

Panel A
-0.057
(0.043)
-0.530***
(0.033)
6,712
0.386

0.492
0.000

Panel B
-0.049
(0.039)
-0.535***
(0.033)
-0.036
(0.036)
0.023
(0.040)
6,712
0.387
0.409

0.000
0.837
0.200
0.000
0.000

0.152***
(0.034)
0.313***
(0.035)
6,712
0.254

0.430
0.000

0.169***
(0.031)
0.316***
(0.035)
-0.061**
(0.025)
-0.051***
(0.018)
6,712
0.256
0.245

0.000
0.002
0.027
0.000
0.000

0.071**
(0.027)
0.188***
(0.031)
6,712
0.199

0.394
0.000

0.089***
(0.024)
0.188***
(0.029)
-0.070**
(0.028)
-0.033
(0.021)
6,712
0.201
0.192

0.000
0.012
0.673
0.002
0.043

Note: ‘Reserved now’ and ‘Reserved before’ are indicator variables of whether village panchayats are
reserved in the current or the previous panchayat periods and the sample is limited to NREGS job card
holders. Control variables included throughout but not reported include household size, composition, land
ownership, and the head’s marital status, gender, age, and education; village-level access to road, distance to
town and district HQ, population, share of SCs, STs, and key religions; years since the last village election;
pradhan characteristics (education, caste, religion, previous tenure and candidacy for office) , district fixed
effects and for individual-level regressions individuals’ gender, marital status, age, education and their
squared terms. Standard errors are clustered at village panchayat level Standard errors are clustered at
village panchayat level. Robust standard errors reported in parentheses and multiple hypotheses tests are
adjusted using the Bonferroni method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

67

8

6

The likelihood of lodging complaints in case of under-payment also increased in the reserved

period (with an elasticity of about 27 percentage points, or 47% at the mean level), though

no longer thereafter (col. 4). Significant lagged effects are observed for an increased

likelihood of wages being paid directly into beneficiaries’ account (col. 2) with an estimated

elasticity of 23 percentage points (25.8%); the likelihood of complaints for underpayment

being addressed (col. 6 with an elasticity of 24 percentage points (57.3%)) and possibly

as a result, a reduction in the likelihood of under-payment (col. 4). While reservation

undeniably improved program governance and enhanced females’ ability to access jobs

under NREGS, none of these effects are gender-specific; to the contrary, for some, mainly

lodging and response to complaints, regression results suggest that women still lag behind

men, consistent with the notion that changes in social norms are not instantaneous.

Results from regressions in Table 2.5 (equivalent to Appendix Table B.6) allow us to

explore if reservation increased women’s demand for work as well as their income and

bargaining power. As one would expect, if reservation relaxed constraints to female labor

supply, it triggers significant lagged increases in women’s individual income, estimated to

have increased by some 76 percentage points (8.4%), and females’ (but not males’) demand

for work by some 16 percentage points. Although not available for the entire sample, data on

intra-household bargaining power support the notion of a role-model effect of reservation

having, with a lag, led to higher levels of female autonomy: The share of women who

participate in decision-making on food, health, and education is estimated to have increased

by 18, 15, and 8 percentage points, respectively.

We conclude that, beyond improving the supply of suitable and attractive jobs for

females, reservation enhanced female decision-making autonomy and their potential and

actual participation in the labor force. A plausible interpretation of this evidence is that the

role model effect provided by female leaders had an enduring effect that enhanced women’s

ability to take advantage of changing labor demand, irrespective of whether such demand

68

9

6

Table 2.5 Reservation and Women’s Participation in Households’ Day to Day Decision
Making

VARIABLES

Reserved now (𝛽1)

Reserved before (𝛽2)

Res now × fem (𝛽3)

Reserved before × fem (𝛽4)

Observations
R-squared
Dep. Var Mean
Test:
F test (𝛽1 + 𝛽2 = 0)
F test (𝛽3 + 𝛽4 = 0)
F test (𝛽1 + 𝛽3 = 0)
F test (𝛽2 + 𝛽4 = 0)
F test (𝛽1 + 𝛽2 + 𝛽3 + 𝛽4 = 0)

Income Work More

Participation in Household Decisions on

Food

Nonfood

Health

Education

0.009
(0.226)
-0.221
(0.211)
0.072
(0.425)
0.983**
(0.388)
66,362
0.286
9.118

0.399
0.009
0.747
0.003
0.002

0.014
(0.044)
0.005
(0.047)
0.025
(0.059)
0.163**
(0.071)
66,362
0.251
0.296

0.724
0.046
0.302
0.000
0.000

0.013
(0.039)
-0.029
(0.045)
-0.094
(0.063)
0.204***
(0.048)
22,467
0.277
0.733

0.713
0.098
0.068
0.000
0.036

-0.005
(0.038)
0.143**
(0.054)
-0.004
(0.045)
-0.009
(0.042)
22,467
0.220
0.781

0.000
0.800
0.833
0.016
0.002

0.000
(0.031)
0.085*
(0.045)
-0.014
(0.025)
0.057**
(0.022)
22,467
0.229
0.816

0.001
0.135
0.693
0.002
0.000

0.018
(0.025)
0.044
(0.032)
-0.001
(0.028)
0.040*
(0.023)
22,467
0.290
0.336

0.009
0.212
0.561
0.014
0.000

Note: ‘Reserved now’ and ‘Reserved before’ are indicator variables of whether village panchayats are
reserved in the current or the previous panchayat periods. Regressions for the desire to work in column 2
and individual income in column 1 include the entire sample whereas those for intra-household bargaining are
limited to the states of Gujarat, Uttar Pradesh, Maharashtra, Orissa, and West Bengal where a supplemental
questionnaire on intra-household bargaining was administered. Control variables included throughout but
not reported include household size, composition, land ownership, and the head’s marital status, gender,
age, and education; individuals’ gender, marital status, age, education, and their squared terms; village-level
access to road, distance to town and district HQ, population, the share of SCs, STs, and key religions;
years since the last village election; pradhan characteristics (education, caste, religion, previous tenure, and
candidacy for office) and district fixed effects. Standard errors are clustered at the village panchayat level
Robust standard errors are reported in parentheses and multiple hypotheses tests are adjusted using the
Bonferroni method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

69

came from Government programs with preferential treatment for females or not.

2.5 Conclusion

Motivated by the recent decline of female labor force participation in India, this pa-

per explores if the random reservation of political leadership positions for women affects

women’s labor force participation as well as supply- and demand-related factors. While

there is no contemporaneous effect, past leadership reservations for women significantly in-

creased females’ labor supply by allowing individuals to join the labor force and increasing

the amount of time spent working by those already in work.

While a large part of the observed effects is attributable to females’ improved ability

to take advantage of public workfare under NREGS, female participation in non-NREGS

labor markets (especially non-agricultural casual and self-employment) expands as well.

Estimated effects are stronger for married than for unmarried women. Labor force partici-

pation allows women to obtain higher levels of individual income, increases their demand

for work, and affects bargaining power by enhancing their participation in intra-household

decision-making on spending for consumption, health, and education. Avenues to enhance

these effects by combining them with the targeted provision of information and training to

change norms regarding women’s labor force participation and equip them with the skills

to adapt to changing labor market conditions are a priority area for further research.

0

7

70

1

7

REFERENCES

Afridi, Farzana, Taryn Dinkelman, and Kanika Mahajan. 2018. “Why are fewer married
women joining the work force in rural India? A decomposition analysis over two decades.”
Journal of Population Economics 31 (3):783–818.

Afridi, Farzana, Vegard Iversen, and MR Sharan. 2017. “Women political leaders, cor-
ruption, and learning: Evidence from a large public program in India.” Economic
Development and Cultural Change 66 (1):1–30.

Afridi, Farzana, Abhiroop Mukhopadhyay, and Soham Sahoo. 2016. “Female labor force
participation and child education in India: evidence from the National Rural Employment
Guarantee Scheme.” IZA Journal of Labor & Development 5 (1):1–27.

Amaral, Sofia, Siddhartha Bandyopadhyay, and Rudra Sensarma. 2015. “Employment
programmes for the poor and female empowerment: The effect of NREGS on gender-
based violence in India.” Journal of interdisciplinary economics 27 (2):199–218.

Andres, Luis Alberto, Basab Dasgupta, George Joseph, Vinoj Abraham, and Maria Correia.
2017. “Precarious drop: Reassessing patterns of female labor force participation in India.”
World Bank Policy Research Working Paper (8024).

Asher, Sam, Tobias Lunt, Ryu Matsuura, and Paul Novosad. 2021. “Development research
at high geographic resolution: an analysis of night-lights, firms, and poverty in India
using the shrug open data platform.” The World Bank Economic Review 35 (4):845–871.

Azam, Mehtabul. 2011. “The impact of Indian job guarantee scheme on labor market

outcomes: Evidence from a natural experiment.” Available at SSRN 1941959 .

Beaman, Lori, Raghabendra Chattopadhyay, Esther Duflo, Rohini Pande, and Petia
Topalova. 2009. “Powerful women: does exposure reduce bias?” The Quarterly journal
of economics 124 (4):1497–1540.

Beaman, Lori, Esther Duflo, Rohini Pande, and Petia Topalova. 2012. “Female leadership
raises aspirations and educational attainment for girls: A policy experiment in India.”
science 335 (6068):582–586.

Berg, Erlend, Sambit Bhattacharyya, Durgam Rajasekhar, and R Manjula. 2014. “Can public
employment schemes increase equilibrium wages? evidence from a natural experiment
in india.” .

Bernal, Raquel. 2008. “The effect of maternal employment and child care on children’s

cognitive development.” International Economic Review 49 (4):1173–1209.

71

2

7

Bernhardt, Arielle, Erica Field, Rohini Pande, Natalia Rigol, Simone Schaner, and Charity
Troyer-Moore. 2018. “Male social status and women’s work.” In AEA Papers and
Proceedings, vol. 108. 363–67.

Bhalotra, Sonia and Irma Clots-Figueras. 2014. “Health and the political agency of women.”

American Economic Journal: Economic Policy 6 (2):164–97.

Bose, Nayana and Shreyasee Das. 2018. “Political reservation for women and delivery of

public works program.” Review of Development Economics 22 (1):203–219.

Casas-Arce, Pablo and Albert Saiz. 2015. “Women and power: unpopular, unwilling, or

held back?” Journal of political Economy 123 (3):641–669.

Chatterjee, Urmila, Rinku Murgai, and Martin Rama. 2015. “Job opportunities along the
rural-urban gradation and female labor force participation in India.” World Bank Policy
Research Working Paper (7412).

Chattopadhyay, Raghabendra and Esther Duflo. 2004. “Women as policy makers: Evidence

from a randomized policy experiment in India.” Econometrica 72 (5):1409–1443.

Das, Maitreyi Bordia, Soumya Kapoor Mehta, Ieva Zumbyte, Sanjeev Sasmal, and Sangeeta
Goyal. 2019. “Does Culture Matter or Firm? Demand for Female Labor in Three Indian
Cities.” Demand for Female Labor in Three Indian Cities (February 12, 2019). World
Bank Policy Research Working Paper (8736).

Dean, Joshua T and Seema Jayachandran. 2019. “Changing family attitudes to promote

female employment.” In AEA Papers and Proceedings, vol. 109. 138–42.

Debnath, Sisir. 2015. “The impact of household structure on female autonomy in developing

countries.” The Journal of Development Studies 51 (5):485–502.

Deininger, Klaus, Songqing Jin, Hari K Nagarajan, and Fang Xia. 2015. “Does female
reservation affect long-term political outcomes? Evidence from rural India.” The Journal
of Development Studies 51 (1):32–49.

Deininger, Klaus, Hari Nagarajan, Sudhir Singh, and Hari Krishnan Nagarajan. 2016.
“Short-term effects of India’s employment guarantee program on labor markets and
agricultural productivity.” World Bank Policy Research Working Paper (7665).

Deininger, Klaus, Hari K Nagarajan, and Sudhir K Singh. 2020. “Women’s political
leadership and economic empowerment: Evidence from public works in India.” Journal
of comparative economics 48 (2):277–291.

72

Desai, Sonalde. 2018. “Do public works programs increase women’s economic empower-

ment?: evidence from rural India.” .

Desai, Sonalde and Omkar Joshi. 2019. “The paradox of declining female work participation
in an era of economic growth.” The Indian Journal of Labour Economics 62 (1):55–71.

Dhar, Diva, Tarun Jain, and Seema Jayachandran. 2019.

“Intergenerational transmis-
sion of gender attitudes: Evidence from India.” The Journal of Development Studies
55 (12):2572–2592.

———. 2022. “Reshaping adolescents’ gender attitudes: Evidence from a school-based

experiment in India.” American Economic Review 112 (3):899–927.

Duflo, Esther. 2005. “Why political reservations?”

Journal of the European Economic

Association 3 (2-3):668–678.

3

7

Dutta, Puja, Rinku Murgai, Martin Ravallion, and Dominique Van de Walle. 2012. “Does
India’s employment guarantee scheme guarantee employment?” Economic and Political
Weekly :55–64.

Eswaran, Mukesh, Bharat Ramaswami, and Wilima Wadhwa. 2013. “Status, caste, and the
time allocation of women in rural India.” Economic Development and Cultural Change
61 (2):311–333.

Fernández, Raquel. 2013. “Cultural change as learning: The evolution of female labor force

participation over a century.” American Economic Review 103 (1):472–500.

Fetzer, Thiemo. 2020. “Can workfare programs moderate conflict? Evidence from India.”

Journal of the European Economic Association 18 (6):3337–3375.

Field, Erica, Rohini Pande, Natalia Rigol, Simone Schaner, and Charity Troyer Moore.
2021. “On her own account: How strengthening women’s financial control impacts labor
supply and gender norms.” American Economic Review 111 (7):2342–75.

Fletcher, Erin, Rohini Pande, and Charity Maria Troyer Moore. 2017. “Women and work

in India: Descriptive evidence and a review of potential policies.” .

Gaddis, Isis and Stephan Klasen. 2014. “Economic development, structural change, and
women’s labor force participation.” Journal of population economics 27 (3):639–681.

Gangadharan, Lata, Tarun Jain, Pushkar Maitra, and Joseph Vecci. 2016. “Social identity
and governance: The behavioral response to female leaders.” European Economic Review
90:302–325.

73

4

7

Gehrke, Esther. 2019. “An employment guarantee as risk insurance? Assessing the effects
of the NREGS on agricultural production decisions.” The World Bank Economic Review
33 (2):413–435.

Ghani, Ejaz, William R Kerr, and Stephen D O’Connell. 2014. “Political reservations and
women’s entrepreneurship in India.” Journal of Development Economics 108:138–153.

Hyland, Marie Caitriona, Simeon Djankov, and Pinelopi Koujianou Goldberg. 2020. “Gen-

dered laws.” World Bank Policy Research Working Paper (9080).

Imbert, Clement and John Papp. 2015. “Labor market effects of social programs: Evidence
from india’s employment guarantee.” American Economic Journal: Applied Economics
7 (2):233–63.

Imbert, Clément and John Papp. 2020. “Short-term migration, rural public works, and urban
labor markets: Evidence from India.” Journal of the European Economic Association
18 (2):927–963.

Islam, Mahnaz and Anitha Sivasankaran. 2015. “How does child labor respond to changes

in adult work opportunities? evidence from nrega.” Harvard University .

Iyer, Lakshmi and Anandi Mani. 2019. “The road not taken: Gender gaps along paths to

political power.” World Development 119:68–80.

Iyer, Lakshmi, Anandi Mani, Prachi Mishra, and Petia Topalova. 2012. “The power of
political voice: women’s political representation and crime in India.” American Economic
Journal: Applied Economics 4 (4):165–93.

Jensen, Robert. 2012. “Do labor market opportunities affect young women’s work and family
decisions? Experimental evidence from India.” The Quarterly Journal of Economics
127 (2):753–792.

Kalsi, Priti. 2017. “Seeing is believing-can increasing the number of female leaders reduce

sex selection in rural India?” Journal of Development Economics 126:1–18.

Kandpal, Eeshani and Kathy Baylis. 2019. “The social lives of married women: Peer effects
in female autonomy and investments in children.” Journal of Development Economics
140:26–43.

Klasen, Stephan. 2019. “What explains uneven female labor force participation levels and
trends in developing countries?” The World Bank Research Observer 34 (2):161–197.

Klasen, Stephan and Janneke Pieters. 2015. “What explains the stagnation of female labor

74

5

7

force participation in urban India?” The World Bank Economic Review 29 (3):449–478.

Lv, Zhike and Rudai Yang. 2018. “Does women’s participation in politics increase female
labor participation? Evidence from panel data analysis.” Economics Letters 170:35–38.

Mani, Subha, Jere R Behrman, Shaikh Galab, and Prudhvikar Reddy. 2014. “Impact of the

NREGS on schooling and intellectual human capital.” .

Mehrotra, Santosh and Jajati K Parida. 2017. “Why is the labour force participation of

women declining in India?” World Development 98:360–380.

Muralidharan, Karthik, Paul Niehaus, and Sandip Sukhtankar. 2016. “Building state ca-
pacity: Evidence from biometric smartcards in India.” American Economic Review
106 (10):2895–2929.

O’Connell, Stephen D. 2018. “Political inclusion and educational investment: Estimates
from a national policy experiment in India.” Journal of Development Economics 135:478–
487.

Pathak, Yuvraj and Karen Macours. 2017. “Women’s political reservation, early childhood
development, and learning in India.” Economic Development and Cultural Change
65 (4):741–766.

Ravi, Shamika and Monika Engler. 2015. “Workfare as an effective way to fight poverty:

The case of India’s NREGS.” World Development 67:57–71.

Sarkar, Sudipa, Soham Sahoo, and Stephan Klasen. 2019. “Employment transitions of

women in India: A panel analysis.” World Development 115:291–309.

Shah, Manisha and Bryce Millett Steinberg. 2021. “Workfare and Human Capital Investment

Evidence from India.” Journal of Human Resources 56 (2):380–405.

Subbarao, Kalanidhi. 1997. “Public works as an anti-poverty program: An overview of
cross-country experience.” American journal of agricultural economics 79 (2):678–683.

75

APPENDIX

6

7

Figure 2.1 Location of Sample Villages

76

7

7

Table B.1 Household and Individual Characteristics in Ever and Never Reserved Villages

Female head
Head’s age
Head’s education
Head married
Head widowed /separated
Household size
Members 15-65 years
Members 15-65 years
Children <15 years
Female children <15 years
Max. educ. in hh (years)
Hindu
SC/ST
Owns agricultural land
Has pucca house
No. of obs.

Female
Age
Educ. primary.
Educ. up to high school
. . . up to graduate
Others
Married
Unmarried
No. of obs.

Total

Reservation status
Never
Ever

Test
(p-values)

Panel A: Household characteristics

0.116
49.200
3.800
0.848
0.136
4.480
1.650
1.560
1.080
0.530
14.260
0.888
0.419
0.579
0.476
23,350

0.490
39.800
0.213
0.291
0.110
0.015
0.748
0.174
66,362

0.115
49.300
3.830
0.842
0.143
4.480
1.660
1.570
1.070
0.520
14.420
0.882
0.404
0.599
0.483
12,678

0.117
48.900
3.750
0.855
0.128
4.470
1.630
1.550
1.110
0.540
14.070
0.894
0.438
0.555
0.467
10,672

0.138
0.748
0.593
0.588
0.252
0.428
0.489
0.363
0.496
0.440
0.144
0.907
0.882
0.034
0.043

Panel B: Individual characteristics

0.491
39.900
0.208
0.291
0.116
0.015
0.742
0.176
34,707

0.489
39.700
0.218
0.292
0.102
0.014
0.754
0.172
31,655

0.184
0.108
0.723
0.019
0.111
0.513
0.115
0.268

Note: Author’s own calculation from 2014/15 REDS follow-up survey. To test differ-
ence in means, p values from regressions with district fixed effects and standard errors
clustered by village panchayat are reported in the last column.

77

8

7

Table B.2 Village and Pradhan Characteristics in Ever and Never Reserved
Villages, 1990/1991, and 2014/2015

Total

Reservation status
Never
Ever

Test
(p-values)

Panel A: 1991 Census Information

No. of households
Population
Scheduled caste pop.
Scheduled tribe pop.
Literate population
No. of primary schools
No. of middle schools
No. of high schools
No. of secondary schools
Good road
Bad road
Cultivated area (ha.)
Habitations w. access to electricity
Agriculture w. access to electricity

Non-farm employment
of which male
of which female
of which in manufacturing
of which in services
Observations

Earlier contested
Held position before
Up to high school
High sec. & above
Higher education
SC
ST
OBC
OC
Hindu
Muslim
No. of obs.

379
2058
380
125
762
1.267
0.507
0.200
0.034
0.581
0.593
533.9
0.487
0.486

443
2400
439
105
917
1.522
0.597
0.435
0.088
0.567
0.633
590.4
0.521
0.373

0.676
412
0.804
2231
0.702
410
0.699
115
0.871
840
0.214
1.398
0.813
0.551
0.158
0.320
0.282
0.061
0.591
0.574
0.412
0.614
0.614
565.2
0.921
0.503
0.433
0.735
Panel B: 1990 Economic Census Information
105.80
89.90
15.90
67.40
38.40
176

115.10
97.10
17.90
79.10
36.00
90

96.80
82.80
14.00
56.10
40.70
86

0.60
0.64
0.44
0.30
0.81

Panel C: 2014/15 REDS Information

0.158
0.474
0.263
0.342
0.163
0.537
0.116
0.126
0.216
0.563
0.089
190

0.137
0.442
0.263
0.263
0.137
0.579
0.116
0.105
0.2
0.484
0.084
95

0.179
0.505
0.263
0.421
0.189
0.495
0.116
0.147
0.232
0.642
0.095
95

0.088
0.087
0.486
0.007
0.29
0.583
0.458
0.838
0.735
0.629
0.147

Note: Author’s own calculation from 2014/15 REDS follow-up survey. Numbers in panel A and
B are calculated from The Socioeconomic High-resolution Rural-Urban Geographic Platform
for India (SHRUG). To test difference in means, p values from regressions with district fixed
effects and standard errors clustered by village panchayat are reported in the last column.

78

9

7

Table B.3 Estimated Effects of Reservation on NREGS Governance

Res. now only (𝛼1)
Res. before only (𝛼2)
Res. now and before (𝛼3)

Observations
R-squared
Test:
(𝛼1=(𝛼2)=(𝛼3)=0)
(𝛼1=(𝛼2)==0)
(𝛼2=(𝛼3)==0)

Res. now only (𝛼1)
Res. before only (𝛼2)
Res. now and before (𝛼3)
Res. now only x female (𝛾1)
Res. before only x female (𝛾2)
Res. now and before x female (𝛾3)

Observations
R-squared
Test:
𝛾1 = 𝛾2 = 𝛾3 = 0
𝛾1 = 𝛾2 = 0
𝛾1 = 𝛾3 = 0
𝛾2 = 𝛾3 = 0

Participation

No. of Days Worked

Total

Other

NREGS

Total

Other

NREGS

-0.002
(0.013)
0.024***
(0.009)
0.030**
(0.013)
66,362
0.277

0.198
0.147
0.694

-0.015
(0.033)
-0.024
(0.032)
-0.045
(0.030)
0.026
(0.063)
0.099
(0.065)
0.157***
(0.055)
66,362
0.282

0.124
0.319
0.042
0.377

-0.018
(0.016)
0.015
(0.012)
0.024
(0.015)
66,362
0.314

0.118
0.0892
0.638

-0.044
(0.032)
-0.043
(0.030)
-0.012
(0.027)
0.054
(0.056)
0.121**
(0.061)
0.077
(0.050)
50,592
0.316

0.615
0.329
0.700
0.491

Panel A
0.062**
(0.028)
0.095***
(0.023)
0.070**
(0.028)
66,362
0.241

0.550
0.822
0.455

Panel B
0.096**
(0.047)
0.106***
(0.030)
-0.041
(0.057)
-0.068
(0.068)
-0.023
(0.042)
0.231**
(0.096)
66,362
0.255

0.013
0.445
0.005
0.005

-0.031
(0.060)
0.176***
(0.047)
0.124*
(0.072)
66,362
0.375

0.0103
0.158
0.538

-0.078
(0.144)
-0.027
(0.159)
-0.249
(0.164)
0.096
(0.279)
0.418
(0.329)
0.777***
(0.295)
66,362
0.379

0.099
0.355
0.032
0.319

-0.075
(0.071)
0.123**
(0.056)
0.086
(0.064)
50,592
0.404

0.0245
0.118
0.653

-0.192
(0.133)
-0.121
(0.144)
-0.109
(0.125)
0.238
(0.230)
0.507*
(0.299)
0.421*
(0.225)
50,592
0.406

0.646
0.402
0.474
0.786

0.246**
(0.095)
0.315***
(0.073)
0.256***
(0.090)
66,362
0.244

0.743
0.930
0.569

0.377**
(0.165)
0.388***
(0.100)
-0.188
(0.211)
-0.269
(0.237)
-0.149
(0.149)
0.921**
(0.375)
66,362
0.263

0.009
0.544
0.003
0.003

Note: ‘Reserved now only’ and ‘Reserved before only’ are indicator variables of whether village
panchayats are reserved in the current or the previous panchayat periods. “Reserved now and
before” is an indicator variable if panchayats were reserved in both elections. The sample is limited
to those who are eligible to work under NREGS. Control variables are included throughout but
coefficients that are not reported include household size, composition, land ownership, and the
head’s marital status, gender, age, and education; village-level access to road, distance to town and
district HQ, population, share of SCs, STs, and key religions; years since the last village election;
pradhan characteristics (education, caste, religion, previous tenure, and candidacy for office) and
for individual-level regressions individuals’ gender, marital status, age, education, and their squared
terms. Robust standard errors reported in parentheses and multiple hypotheses tests are adjusted
using the Bonferroni method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

79

Table B.4 Reservation and Labor Force Participation by Marital Status and Types of
Employment

Total

Participation
Other

NREGS

No. of Days Worked
Other

NREGS

Total

Res. now only (𝛼1)

Res. before only (𝛼2)

Res. now and before (𝛼3)

Res. now only x female (𝛾1)

Res. before only x female (𝛾2)

Res. now and before x female (𝛾3)

Observations
R-squared
Test:
𝛾1 = 𝛾2 = 𝛾3 = 0
𝛾1 = 𝛾2 = 0
𝛾1 = 𝛾3 = 0
𝛾2 = 𝛾3 = 0

Res. now only (𝛼1)

Res. before only (𝛼2)

Res. now and before (𝛼3)

Res. now only x female (𝛾1)

Res. before only x female (𝛾2)

Res. now and before x female (𝛾3)

Observations
R-squared
Test:
𝛾1 = 𝛾2 = 𝛾3 = 0
𝛾1 = 𝛾2 = 0
𝛾1 = 𝛾3 = 0
𝛾2 = 𝛾3 = 0

-0.019
(0.042)
-0.044
(0.038)
-0.044
(0.032)
0.045
(0.075)
0.139*
(0.075)
0.162***
(0.059)
50,323
0.285

0.282
0.281
0.114
0.756

-0.005
(0.022)
0.041*
(0.024)
-0.042
(0.029)
-0.031
(0.033)
-0.033
(0.042)
0.126**
(0.050)
16,039
0.257

0.005
0.950
0.002
0.004

-0.056
(0.040)
-0.069*
(0.037)
-0.005
(0.033)
0.075
(0.068)
0.160**
(0.073)
0.077
(0.061)
36,802
0.339

Panel A: Married
0.117**
(0.058)
0.115***
(0.037)
-0.057
(0.069)
-0.086
(0.085)
-0.015
(0.051)
0.266**
(0.113)
50,323
0.255

-0.100
(0.180)
-0.133
(0.190)
-0.228
(0.178)
0.158
(0.336)
0.614
(0.383)
0.770**
(0.320)
50,323
0.419

0.496
0.307
0.982
0.282

0.0174
0.340
0.00564
0.00876

0.237
0.280
0.0968
0.704

Panel A: Unmarried
0.037
(0.025)
0.071***
(0.019)
0.003
(0.029)
-0.021
(0.030)
-0.056*
(0.031)
0.114**
(0.049)
16,039
0.276

-0.022
(0.120)
0.290**
(0.142)
-0.270*
(0.159)
-0.096
(0.157)
-0.243
(0.209)
0.687***
(0.251)
16,039
0.283

-0.018
(0.025)
0.033
(0.026)
-0.031
(0.025)
-0.015
(0.029)
-0.016
(0.041)
0.073**
(0.036)
13,790
0.265

0.053
0.970
0.023
0.062

0.003
0.292
0.007
0.001

0.002
0.451
0.001
0.001

0

8

-0.239
(0.165)
-0.245
(0.175)
-0.044
(0.153)
0.306
(0.287)
0.680*
(0.354)
0.369
(0.275)
36,802
0.465

0.615
0.341
0.848
0.418

-0.069
(0.133)
0.250*
(0.147)
-0.230*
(0.129)
-0.020
(0.125)
-0.110
(0.203)
0.469***
(0.177)
13,790
0.287

0.017
0.665
0.008
0.018

0.461**
(0.204)
0.427***
(0.123)
-0.251
(0.252)
-0.341
(0.291)
-0.133
(0.180)
1.054**
(0.437)
50,323
0.261

0.0111
0.406
0.00353
0.00449

0.141*
(0.085)
0.247***
(0.062)
-0.020
(0.111)
-0.069
(0.121)
-0.215**
(0.108)
0.471**
(0.208)
16,039
0.304

0.003
0.214
0.010
0.001

Note: ‘Reserved now’ and ‘Reserved before’ are indicator variables of whether village panchayats are
reserved in the current or the previous panchayat periods and the sample is limited to those who worked
under NREGS. Control variables are included throughout but coefficients that are not reported include
household size, composition, land ownership, and the head’s marital status, gender, age, and education;
village-level access to road, distance to town and district HQ, population, the share of SCs, STs, and key
religions; years since the last village election; pradhan characteristics (education, caste, religion, previous
tenure, and candidacy for office) and for individual-level regressions individuals’ gender, marital status,
age, education, and their squared terms. Robust standard errors reported in parentheses and multiple
hypotheses tests are adjusted using the Bonferroni method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

80

1

8

Table B.5 Estimated Effects of Reservation on NREGS Governance

Get dated
receipt

Payment
to account

0.106***
(0.034)
0.300***
(0.029)
0.961***
(0.077)
6,712
0.605

0.000
0.000
0.000

0.101***
(0.033)
0.308***
(0.032)
0.950***
(0.078)
0.007
(0.021)
-0.013
(0.025)
0.064**
(0.030)
6,712
0.606

0.007
0.210
0.036
0.002

-0.012
(0.027)
0.126***
(0.027)
0.242***
(0.062)
6,712
0.698

0.000
0.000
0.003

-0.006
(0.027)
0.132***
(0.027)
0.247***
(0.062)
-0.022
(0.025)
-0.027
(0.024)
-0.027
(0.024)
6,712
0.699

0.707
0.471
0.444
0.999

Payment
less
assessed

than

Panel A
0.048
(0.052)
-0.287***
(0.033)
-0.544***
(0.065)
6,712
0.386

0.000
0.000
0.000

Panel B
0.059
(0.048)
-0.274***
(0.027)
-0.542***
(0.061)
-0.052
(0.040)
-0.046
(0.035)
-0.005
(0.063)
6,712
0.387

0.666
0.795
0.378
0.389

If less, did
complain

Complaint
addressed

0.145***
(0.033)
0.297***
(0.029)
0.462***
(0.065)
6,712
0.254

0.000
0.000
0.000

0.166***
(0.030)
0.316***
(0.026)
0.483***
(0.063)
-0.069**
(0.031)
-0.087***
(0.024)
-0.107***
(0.029)
6,712
0.256

0.159
0.367
0.097
0.093

0.061**
(0.029)
0.166***
(0.028)
0.255***
(0.055)
6,712
0.199

0.000
0.000
0.003

0.084***
(0.025)
0.187***
(0.024)
0.274***
(0.051)
-0.084**
(0.034)
-0.089***
(0.029)
-0.096***
(0.033)
6,712
0.202

0.849
0.766
0.596
0.623

Res. now only (𝛼1)

Res. before only (𝛼2)

Res. now and before (𝛼3)

Observations
R-squared
Test:
𝛼1=(𝛼2)=(𝛼3)=0
𝛼1=(𝛼2)==0
𝛼2=(𝛼3)==0

Res. now only (𝛼1)

Res. before only (𝛼2)

Res. now and before (𝛼3)

Res. now only x female (𝛾1)

Res. before only x female (𝛾2)

Res. now and before x female (𝛾3)

Observations
R-squared
Test:
𝛾1 = 𝛾2 = 𝛾3 = 0
𝛾1 = 𝛾2 = 0
𝛾1 = 𝛾3 = 0
𝛾2 = 𝛾3 = 0

Note: ‘Res. now only’ and ‘Res. before only’ are indicators of whether village panchayats are reserved in the
current or the previous panchayat periods. “Reserved now and before” is an indicator variable if panchayats
were reserved in both elections. The sample is limited to those who are eligible to work under NREGS.
Control is included throughout but coefficients that are not reported include household size, composition, land
ownership, and the head’s marital status, gender, age, and education; village-level access to road, distance
to town and district HQ, population, the share of SCs, STs, and key religions; years since the last village
election; pradhan characteristics (education, caste, religion, previous tenure, and candidacy for office) and
for individual-level regressions individuals’ gender, marital status, age, education, and their squared terms.
Robust standard errors reported in parentheses and multiple hypotheses tests are adjusted using the Bonferroni
method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

81

Table B.6 Estimated Effects of Reservation on Decision Making

VARIABLES

Income Work more

Participation in Household Decisions on

Food

Nonfood Health Education

Res. now only (𝛼1)

Res. before only (𝛼2)

Res. now and before (𝛼3)

Res. now only x female (𝛾1)

-0.084
(0.287)
-0.376
(0.257)
-0.135
(0.262)
0.154
(0.547)
1.173**
(0.479)
Res. now and before x female (𝛾3) 1.009**
(0.388)
66,362
0.286

Res. before only x female (𝛾2)

0.065
(0.048)
0.112***
(0.039)
-0.014
(0.062)
-0.061
(0.065)
-0.034
(0.042)
0.236**
(0.099)
66,362
0.256

-0.011
(0.046)
-0.052
(0.048)
0.034
(0.046)
-0.122
(0.081)
0.165***
(0.057)
0.124**
(0.060)
22,571
0.260

-0.077*
-0.054
0.021
(0.039)
(0.034)
(0.033)
0.036
0.013
-0.006
(0.033)
(0.050)
(0.052)
0.240*** 0.178*** 0.102**
(0.040)
(0.034)
(0.040)
-0.074**
-0.032
0.006
(0.032)
(0.030)
(0.048)
0.068*
0.054
0.050
(0.038)
(0.036)
(0.040)
-0.022
0.040*
-0.030
(0.044)
(0.023)
(0.042)
22,571
22,571
22,571
0.187
0.216
0.207

2

8

Observations
R-squared
Test:
𝛾1 = 𝛾2 = 𝛾3 = 0
𝛾1 = 𝛾2 = 0
𝛾1 = 𝛾3 = 0
𝛾2 = 𝛾3 = 0

0.225
0.106
0.128
0.740

0.0114
0.601
0.00450
0.00343

0.000316
6.33e-05
0.000730
0.323

0.128
0.332
0.454
0.0450

0.00733
0.0194
0.00235
0.635

0.00319
0.000731
0.267
0.0773

Note: ‘Res. now only’ and ‘Res. before only’ are indicators of whether village panchayats are reserved
in the current or the previous panchayat periods. “Reserved now and before” is an indicator variable
if panchayats were reserved in both elections. The sample is limited to those who are eligible to work
under NREGS. Regressions for the desire to work in column 2 and individual income in column 1
include the entire sample whereas those for intra-household bargaining are limited to the states of
Gujarat, Uttar Pradesh, Maharashtra, Orissa, and West Bengal where a supplemental questionnaire
on intra-household bargaining was administered. Control is included throughout but coefficients
that are not reported include household size, composition, land ownership, and the head’s marital
status, gender, age, and education; village-level access to road, distance to town and district HQ,
population, the share of SCs, STs, and key religions; years since the last village election; pradhan
characteristics (education, caste, religion, previous tenure, and candidacy for office) and for individual-
level regressions individuals’ gender, marital status, age, education, and their squared terms. Robust
standard errors reported in parentheses and multiple hypotheses tests are adjusted using the Bonferroni
method. ∗ ∗ ∗𝑝 < 0.01, ∗ ∗ 𝑝 < 0.05, ∗𝑝 < 0.10.

82

3

8

CHAPTER 3

IDENTITY-BASED POLITICAL CONNECTIONS AND
PUBLIC EMPLOYMENT

3.1

Introduction

A growing literature suggests that identity-based political and social connections play

an important role in the distribution of public goods and services. The political connections

between politicians and beneficiaries who share similar identities can have both positive

and negative impacts on the local labor market and economy. While identity-based political

connections can help ensure that historically marginalized groups have a voice in decision-

making processes (Jaspal, 2011; Sankaran, Sekerdej, and Von Hecker, 2017), which can lead

to a more equitable distribution of wealth and social mobility (Cassan, 2019; Deshpande,

2018, 2019; Hoff, Kshetramade, and Fehr, 2011), affect hiring practices and can facilitate

the exchange of information in the absence of efficient market institutions (Schmutte, 2016;

Munshi and Rosenzweig, 2015).

It may also run the risk of perpetuating group favoritism. Politicians may use their

network and identity for preferential targeting (Besley et al., 2004; Chattopadhyay and

Duflo, 2004; Banerjee and Somanathan, 2007), and reward those who actively support their

party and candidacy (Besley et al., 2004; Chau, Liu, and Soundararajan, 2018; Moser, 2008;

Sheahan et al., 2016). Such favoritism may distort the allocation of resources and jobs and

disproportionately favor specific groups of local citizens over others (Colonnelli, Prem, and

Teso, 2020; Gille, 2018; Markussen and Tarp, 2014), which can lead to increasing group

inequality, misallocation of resources (Hsieh et al., 2019; Colonnelli, Prem, and Teso, 2020),

and corruption (Bardhan and Mookherjee, 2005).

Identity and connections are particularly important in rural India, where caste, reli-

gion, and ethnicity play a crucial role in shaping political dynamics and resource distribu-

tion (Chattopadhyay and Duflo, 2004; Banerjee and Somanathan, 2007). In recent years,

83

4

8

identity-based political connections have become an important aspect of politics, reflecting

the interests and concerns of various social and demographic groups in the allocation of

resources in India (Gaikwad, 2022). However, the role of political connections in public

employment remains relatively understudied. We study the effects of identity-based political

connections in India’s largest national workfare program, the National Rural Employment

Guarantee Scheme (NREGS). The program offers an interesting case study to examine the

effects of identity-based political connections, as local leaders possess significant discretion

in the allocation of benefits. However, estimating the effects of political connections is chal-

lenging due to the non-random election of local leaders, which we overcome by leveraging

the unique caste (jati) system in India1.

We constructed a measure of political connections between local leaders, including

village council chairs and members, and individuals based on a shared identity determined

by their surnames. In India, surnames typically reflect people’s caste, although they can

vary across different states and regions. To ensure accurate matching, we carried out the

matching procedure conditional at the village, district, and state levels. We then proceed to

estimate the labor-supply effects of political connections in NREGS.

Matching surnames enables us to compare individuals of different castes within the same

caste group and village, providing valuable insights into social dynamics and relationships2.

We also adjust for village-level unobservables and caste (jati) compositional differences

across villages by including village dummies and controls at the level of caste-by-village.

While this approach helps to mitigate potential bias, it does not completely eliminate the

concerns about endogeneity. Some unobservable factors at the caste (jati) level, such

as political influence within villages and caste-level hierarchy, could still be correlated

1Jati and caste are used interchangeably
2Caste groups such as Scheduled Caste (SC), Scheduled Tribe (ST), and Other Backward Castes (OBC)
are designations used by the Indian government for affirmative action; caste-group membership determines
reservations for jobs and elections. Caste is a much finer delineation than caste group, which usually subsumes
many sub-castes.

84

5

8

with both NREGS participation and political leadership. Therefore, we also employ an

instrumental variables (IV) approach in the main regression analysis. In this context, we

utilize political reservations for lower-caste groups, which are mandated by constitutional

amendments 73 and 74. These amendments require a share of the panchayat chairperson

seat to be reserved for lower caste groups (SC/ST) that is proportional to the state population

share of the lower caste group. The panchayats are then selected on a rotational basis for

the reservation3.

The main results suggest that political connection based on shared caste identity with

local leaders increased individual labor supply on both the intensive and extensive margins.

The estimates suggest an 11.5 percent increase in labor supply on the extensive margin.

The effect increases to 17 percent when reservation policy is used as an instrument. We

also find an increase in the number of days worked by individuals when local leaders are of

the same caste. The increase in labor supply on the intensive margin mirrors the increase

on the extensive margin. We find evidence that results are partially driven by favoritism by

local leaders.

Related literature NREGS is a decentralized program where the implementation is carried

out at the village level, with the village-council head playing a crucial role in the distribution

of public works and ensuring effective oversight of projects in consultation with the elected

council members. Since the power to distribute jobs rests with the local leaders, their

identities—gender, caste, religion—are likely to affect programs, and the political connec-

3The details are given at https://pib.gov.in/PressReleaseIframePage.aspx?PRID=1658145. The following
is the exact description of the clause. "As per clause (4) of Article 243D of the Constitution, the positions
of Chairpersons within Panchayats, whether at the village level or any other tier, are mandated to be reserved
for Scheduled Castes, Scheduled Tribes, and women. This reservation is to be carried out in the manner
prescribed by the State Legislature, and it is specified that the number of Chairperson positions reserved for
Scheduled Castes and Scheduled Tribes in Panchayats at each level in any State should, to the best extent
possible, be proportional to the total number of such positions in the Panchayats at each level. This proportion
is to be determined based on the population of Scheduled Castes in the State or Scheduled Tribes in the State
in relation to the overall population of the State. Moreover, it is further stipulated that a minimum of one-third
of the total number of Chairperson positions at each level in the Panchayats must be reserved for women.
Additionally, the allocation of positions reserved under this clause is to be rotated among different Panchayats
at each level".

85

6

8

tions of these local leaders can have an impact on the distribution of jobs in NREGS, which

may also allow local politicians to reap political gains (Bardhan and Mookherjee, 2000)

through allocation. Local leaders might prefer to allocate some resources to their own kind,

who have similar preferences (Chattopadhyay and Duflo, 2004) and shared identity 4.

This paper makes a significant contribution to the extensive literature on political econ-

omy by examining the role of political connections, specifically focusing on the effects of

caste identity in public works programs in India that are not yet studied in the literature.

This paper contributes to the studies on reelection incentives and political targeting that have

shed light on the motivations and strategies employed by politicians to secure their positions

in office (Besley and Case, 1995; Besley et al., 2004; Chau, Liu, and Soundararajan, 2018;

Persson, Roland, and Tabellini, 2000). Research on identity and representation that has ex-

plored the influence of identity factors, such as gender and ethnicity, on political processes

and labor market outcomes (Beaman et al., 2012; Chattopadhyay and Duflo, 2004; Cassan,

2015; Oh, 2019). And, political connections and allocation, which have demonstrated

how political networks and affiliations impact resource distribution(Choi, Penciakova, and

Saffie, 2021; Fafchamps and Labonne, 2017; Gille, 2018; Gagliarducci and Manacorda,

2020; Caeyers and Dercon, 2012; Markussen and Tarp, 2014).

The paper also builds on previous findings by Besley et al. (2004) and Kumar, So-

manathan et al. (2017), who have demonstrated the importance of political connections and

the role of caste identity in the intra-village allocation of public goods and the distribution

of benefits from social programs, respectively. However, this paper goes a step further by

providing novel evidence on how political connections formed around caste identity impact

4If the local leaders are elected from historically minority groups, they can also generate demand effects
in public work via role model channel (Chattopadhyay and Duflo, 2004; Beaman et al., 2009; Jaspal, 2011;
Sankaran, Sekerdej, and Von Hecker, 2017). A large literature has studied the effects of having female
leaders in many different settings and documents that they are more likely to invest more in water and health
(Chattopadhyay and Duflo, 2004), they raise aspiration and education among women (Beaman et al., 2012;
Clots-Figueras, 2012), and they increase participation in the labor market (Bose and Das, 2018; Deininger,
Nagarajan, and Singh, 2020). Hence it’s important to disentangle if the effects are through the demand channel
(role model effects) or the supply channel (preferential allocation).

86

7

8

the labor supply of workers in public works programs in India. By exploring this previ-

ously unaddressed aspect, the study contributes to a more comprehensive understanding of

the intricate dynamics between political connections, identity, and labor outcomes in the

context of public works programs.

The rest of the paper is organized as follows. In section 3.2, we present a survey on

the history of caste, the role of caste identity in the Indian economy, and briefly document

the background of the public program and its effects. We describe my data and descriptive

statistics in 3.3. Empirical strategy, the estimation and identification challenges are dis-

cussed in section 3.4. Section 3.5 presents my main results and discusses heterogeneity in

the results. In section 3.5.1, we test whether the strength of the connections matters. We

check for robustness of the results and lay out the mechanism that is likely to be driving the

results in section 3.6. In section 3.7, we conclude and discuss the policy implications.

3.2 Background

3.2.1 The Caste System

The roots of caste in India can be traced back to 1500-500 BCE. Historically, caste was

a system of social stratification that was hereditary and based on a person’s occupation.

During this time, society was divided into four main varnas or social classes: the Brahmins,

the Kshatriyas, the Vaishyas, and the Shudras. Dalits were not included in the original varnas

(Munshi, 2019; Macdonell, 1914). Within these varnas and dalits, there exist hundreds of

subgroups known as castes or jatis5 6. Each varna had its distinct roles and responsibilities,

and social mobility between them was limited. Over time, this rigid social structure became

more complex and the caste (jati) system became deeply ingrained in Indian society. Jati,

meaning "birth group" or "sub-caste," became an essential aspect of the social organization

5The main castes were further divided into about 3,000 castes and 25,000 sub-castes, each based on their

specific occupation

6Macdonell (1914) emphasizes that without a historical examination of the caste system in contemporary
India, one cannot gain a comprehensive understanding of the institution itself in India. For a more detailed
account of the caste system’s history, please see Macdonell (1914); Dumont (1980)

87

8

8

7, with strict rules governing social interaction, marriage, and even occupation (Macdonell,

1914).

This hierarchical nature of the caste system, where individuals were assigned a specific

caste at birth, determined their social standing and opportunities in life. Those born into

higher castes enjoyed privileges and access to resources, while those in lower castes faced

discrimination, social exclusion, and limited opportunities for upward mobility. The caste

system influenced the division of labor, with each caste being associated with specific occu-

pations and skills. The division of labor facilitated economic specialization and contributed

to various sectors of the economy. However, it also led to the perpetuation of occupational

stereotypes and restricted social mobility (Deshpande, 2011, 2010; Dumont, 1980). The

lack of social mobility resulted in economic disparities and inequalities, with lower-caste

individuals often experiencing poverty and limited access to resources and economic oppor-

tunities. With an amendment to the constitution, the Government of India (GoI) has made

an effort to address the social and economic inequities perpetuated by the caste system. Af-

firmative action policies, such as reservations in education and government jobs, have been

implemented to provide opportunities for historically disadvantaged castes. However, the

caste system continues to influence social and economic dynamics in India (Deshpande and

Newman, 2007; Munshi, 2019). And, political mobilization along caste lines, especially

during elections, remains a common phenomenon (Blakeslee et al., 2013).

3.2.2 The Role of Caste Identity in the Indian Economy

Scholars in the fields of sociology and psychology have extensively explored the influ-

ence of identity on human behavior (Sharma, 1984; Vaid, 2014). However, the formalization

of identity in economics began with the seminal work of Akerlof and Kranton (2000). More

recently, economists have shown increasing interest in measuring the impact of identity on

7The main castes were further divided into about 3,000 castes and 25,000 sub-castes, each based on their

specific occupation (BBC 2019)

88

9

8

various economic and non-economic outcomes. Munshi (2019) emphasizes the significance

of caste in shaping everyday economic and non-economic decision-making among Indians.

Caste identity plays a crucial role in determining an individual’s social and economic status,

as well as influencing the overall social structure and norms within a group or society.

India has made significant economic and social progress by integrating lower-caste

groups through affirmative action. However, barriers continue to limit occupational mobil-

ity, as caste-based occupations abound (Das, 2013; Deshpande, 2011; Oh, 2019). A large

proportion of low-caste households remain employed in menial services (Deshpande, 2018;

Hoff, Kshetramade, and Fehr, 2011; Iversen et al., 2016). This low mobility could be due

to the historical oppression of, and social stigma associated with the social identity of, a

particular caste (Hoff, Kshetramade, and Fehr, 2011).

For example, Hoff and Pandey (2006) conducted an experiment in Uttar Pradesh, India,

and found that public disclosure about the caste impedes the ability to solve the puzzles if

the caste has been historically discriminated against, pointing towards the persistent effect

of discrimination. Similar results are found in an experimental study in the Indian state

of Orissa by Oh (2019). Specifically, Oh (2019) finds that caste identity is a significant

determinant of individuals’ decision whether to accept a job offer, as it reduces participation

in a job-offer experiment by 23 percentage points when the job offer is associated with other

caste.

In a related study, Iyer, Khanna, and Varshney (2013) found that caste is related

to entrepreneurship in India and that lower-caste groups are severely underrepresented in

entrepreneurship in India despite progressive policies.

Caste also plays an important role in the Indian marriage market. More than 80 percent

of marriages in India are arranged, and 90 percent of marriages are within the same caste. In

addition, caste reservations granted to individuals may disproportionately favor men from

lower-caste groups over their female counterparts. For example, due to affirmative action,

government agencies differ in their hiring of men and women belonging to the OBC caste

89

0

9

group. Because men are more likely to be hired, they have greater household bargaining

power in decisions regarding the number of children or the spacing between children (Desai,

2016).

Barriers between caste groups create inefficiencies in Indian villages and affect the

overall productivity and surplus of villages. For example, Anderson (2011) finds that lower-

caste farmers with land in villages dominated by higher castes have lower productivity than

farmers in villages dominated by lower castes. The productivity difference could reflect

the fact that trade in irrigation water is limited across castes in villages, making caste an

important cause of market inefficiency.

3.2.3 Public Works Program

The National Rural Employment Guarantee Scheme (NREGS) is a flagship workfare

program in India. It was enacted by the Indian government in 2005 and implemented in

20068. The program was rolled out in a phase-wise manner, initially covering India’s 200

poorest districts in February 2006, in phase 1. Gradually it expanded to the next poorest

130 districts in April 2007 in phase 2. And finally, in April 2008, the remaining districts

were added to the program.9

To work under the program, one must apply for a job card, which serves as the basis

for work. All job-card holders are eligible to work under the program. Eligible households

can apply for work and receive employment within 15 days of submitting an application.

If work is not provided within 15 days’ time, the applicant is entitled to an unemployment

allowance. The wages paid under the scheme are determined by the state governments but

8India’s national workfare program, NREGS, is the world’s largest cash-for-work program. India spent
about $9.98 billion on NREGS in 2020, double the $4.8 billion spent in 2015. And the program employed 87
million people in 2020 alone (Economic Times, October 22, 2020).

9In the 2001 census, there were 593 districts. In the 2011 census, that number increased to 640. The
Indian Planning Commission selected districts for the program based on poverty rates calculated using data
from 1993–94 National Sample Survey and the 2001 census. The Planning Commission of India created a
development index based on the poverty headcount ratio, agricultural wages, and productivity and population
share of lower-caste groups such as SC and ST.

90

1

9

Figure 3.1 Distribution of Workers in NREGS

must not be less than the minimum wage rates prescribed by the government. The wage

under NREGS is equal for men and women. The program also stipulates that one-third of

all jobs be reserved for women to increase their participation. Most of the work performed

under the program is low-skilled. However, people of all skill levels and caste groups can

apply for jobs.

The scheme focuses on labor-intensive works that contribute to rural development and

builds productive assets, such as roads and water-harvesting structures (Deininger and Liu,

2013). Over the years, the NREGS has witnessed several improvements and modifications

to enhance its effectiveness. These include the use of technology for better implementation

and increased transparency through social audits. Local NGOs are expected to audit the

program, report it to district panchayats on the audit, and make the report public. Wages are

supposed to be paid directly to the applicant’s bank account, and if the applicant does not

have an account, they are encouraged to open one under NREGS to minimize corruption

(Muralidharan, Niehaus, and Sukhtankar, 2016).

A significant body of work has studied the effects of NREGS. The NREGS has had a

91

2

9

significant impact on rural India. It has provided a safety net for the poor (Bose, 2017; Rao,

2019), reduced rural-urban distressed migration(Imbert and Papp, 2020), and increased the

income and purchasing power of rural households (Cook and Shah, 2022). The scheme

has also contributed to empowering women and marginalized communities by ensuring

their participation in the workforce and decision-making processes (Azam, 2011; Desai,

2018). NREGS increased private and public wages for unskilled workers in the dry season

(Berg et al., 2018; Imbert and Papp, 2015), helps diversify crop income (Gehrke, 2019),

and improves health and education outcomes (Dasgupta, 2017; Shah and Steinberg, 2019).

Studies examining the effects of gender identity on village panchayat heads have yielded

mixed findings. While having a female panchayat head enhances the demand for work under

NREGS (Bose and Das, 2018; Deininger, Nagarajan, and Singh, 2020), there is evidence

suggesting inefficiencies and leakages in the program within village councils reserved

for women (Afridi, Iversen, and Sharan, 2017). Notably, the efficiency of public services

improves as female leaders accumulate more experience (Afridi, Iversen, and Sharan, 2017).

Nonetheless, the available evidence remains quite limited regarding the impact of village

leaders’ caste identity and its implications for NREGS outcomes.

3.3 Data

The main data comes from the Socio-Economic Profile of India (SEPRI), which is

a follow-up to the Rural Economics and Demographic Survey (REDS)10. The SEPRI was

enumerated for all households in 192 villages across 13 states from 2014 to 2016. The survey

had two components: village and household. It utilizes the complete set of questions from

the REDS village module, along with additional information on NREGS-related activities.

10REDS builds on the Additional Rural Income Survey, a panel survey first conducted in 1969 and 1971
that provides comprehensive information about more than 4,200 households in 259 villages in 17 major
Indian states; the households are meant to be representative of India’s rural population. In 1981–82, surviving
households, any descendants from the original sample who resided in the same village, and a random sample
of new households were surveyed (Vashishtha, 1989). A similar procedure was employed in subsequent survey
rounds, resulting in samples of 7,474 and 8,659 households in 1998–99 and 2007–8 across 242 villages in 17
major states.

92

3

9

The village-level module collects information on every political candidate who participated

in the past two panchayat elections for the position of village council head. Additionally,

the village-level module also gathers detailed information on political candidates running

for the office of village council members.

In contrast to the village module, the SEPRI

survey utilizes only a subset of questions from the REDS but provides detailed information

on NREGS activities for each member11.

The SEPRI data are particularly suited for my analysis since the survey collects detailed

household- and individual-level information as well as information on the characteristics of

each political candidate who ran for the office of village council head and memberships in the

last two elections. In particular, the information on political candidates includes their name,

education, gender; whether a candidate is from the SC, ST, OBC, or Forward caste group;

vote received, and whether the candidate ran for the village-council position in past elections,

etc. The household component of the survey includes a standard consumption and income

module, employment, religion, social group, and jati. Additionally, the household module

provides information on detailed NREGS activities for each worker. The information on

NREGS includes seasonal allocation and demand for jobs, possession of job cards by

households, household preferences for more work, total wages earned through NREGS, and

details on transparency and accountability.

As all households in the REDS villages were enumerated, we were able to match the

surnames of the village council heads to households and individuals, allowing us to construct

a measure of political connection. Figure 3.2 show the distribution of the population of

age 18 and above with the same subcaste (jati) as the village council head by caste group

in the data. We restrict the sample to households with job cards since job cards determine

11Because of conflicts, no data were collected in Jammu and Kashmir in 1998/99 and 2007/08 survey
rounds (11 villages). For the same reason, no data were collected in Assam in 1981/82 and 2007/08 rounds of
surveys (8 villages). The 242 villages for which data were collected in all three rounds are distributed across
15 states. These states became 17 states after the formation of the new states of Chhattisgarh and Jharkhand
in 2000, which split from Madhya Pradesh and Bihar.

93

4

9

Figure 3.2 Share of Same Caste Population as Village Head

eligibility to work in NREGS.

3.3.1 Summary Statistics

Table 3.1 presents the descriptive statistics of the household-level variables in panel A

and the individual-level statistics in panel B. The data suggest that 27 percent of households

have a job card. Of those who have job cards, only 24 percent worked in NREGS; that

subset worked an average of 38 days for NREGS in a year.

The data indicates that about 29 and 49 percent of households belong to the SC/ST and

OBC caste groups, respectively, while 22 percent of the households in the dataset are from

the Other (Forward) caste group, which is consistent with the findings of the 2011 census.

Among all households, around 56 percent own some agricultural land (Table C.2). There

is no significant statistical difference in land ownership between individuals possessing a

job card and those without one. However, landownership among the SC/ST caste group

is 10 percentage points lower compared to the Forward caste group. On average, each

household in the dataset possesses about 2.6 acres of land. There is a significant difference

in landholding sizes between the SC/ST and Forward caste groups, with the former having

94

Table 3.1 Summary Statistics

HHs with Job card

SC-ST

OBC

OTHER

Total

Variable

Female head
Head’s age
Head’s education
Share of hhs
. . . with no education
. . . up to 5th grade
. . . up to high school
. . . college or above
Married
Household Size
Prime male
Prime female
Children
Own Any land
. . . Size of land
Jobcard
#Obs

Age
Female
Education
Formally Applied
Worked in NREGS
# of Days worked
Received allowance
Wanted to work more
Attended GS meet-
ings
#Obs

All

0.128
50.028
4.779

0.399
0.172
0.379
0.044
0.823
4.561
1.671
1.565
1.103
0.562
2.625
0.271
79646

40.081
0.491
5.58
0.115
0.064
37.059
0.008
0.032
0.181

0.124
48.653
3.267

0.504
0.205
0.260
0.015
0.838
4.481
1.639
1.545
1.121
0.505
1.839

Panel A
0.110
49.490
4.009

0.414
0.239
0.321
0.018
0.854
4.529
1.664
1.587
1.085
0.655
2.353

8959

9806

39.527
0.489
4.223
0.346
0.247
34.412
0.022
0.139
0.190

Panel B
40.129
0.494
4.894
0.325
0.263
42.694
0.020
0.111
0.197

0.106
49.392
4.587

0.358
0.220
0.392
0.023
0.858
4.142
1.603
1.471
0.870
0.517
1.853

2817

40.184
0.479
5.512
0.322
0.168
28.146
0.017
0.094
0.217

249923

25991
Notes: Author’s own calculation from survey data

28327

8208

5

9

0.115
49.130
3.776

0.444
0.223
0.305
0.017
0.848
4.459
1.646
1.554
1.072
0.575
2.116
1.000
21582

39.886
0.490
4.696
0.333
0.244
37.894
0.020
0.120
0.197

62526

95

6

9

an average of two acres and the latter having an average of three acres per household.

Importantly, over 50 percent of households have either no formal education or have received

education only up to grade 5. This is true for both household heads and members of the

households.12

The percentage of job-card holders is calculated based on the total number of households

in the sample. There is considerable variation in job-card ownership among different caste

groups. For instance, 33 percent of job card holders belong to the SC caste group, 9 percent

are from the ST group, 45 percent are OBC, and 13 percent are from other caste groups.

Our primary analysis focuses exclusively on job card holders since possessing a job card is

a prerequisite for employment. Detailed descriptive statistics can be found in Table C.1.

Panel B of Table 3.1 presents summary statistics pertaining to individuals and their

participation in NREGS activities. We find that approximately 11 percent of household

members applied for a job under the NREGS program. While, among households with

job cards, approximately 33 percent of individuals applied for jobs. The average age of

individuals willing to work under NREGS is 40 years. Around 2 percent of household

members applied for a job but were unable to secure one within 15 days, resulting in them

receiving an allowance. The average number of days worked across all caste groups, and

job-card holders ranges from 28 to 43 (Table 3.1).

Approximately 24 percent of individuals holding job cards were engaged in NREGS

employment. Although this number is quantitatively similar for SC/ST and OBC caste

groups, it is 17 percent for the forward caste group. Additionally, 12 percent of job card

holders expressed a desire for more work, indicating the possibility of some form of rationing

at an extensive margin. This disparity becomes even more pronounced when considering

the entire sample of households. There are no significant variations in attendance rates

at village-council meetings among different caste groups or job card holders. Detailed

12In the sample I analyzed, 59 percent of individuals have either no formal education or have received

education only up to grade 5.

96

7

9

descriptive statistics for households with job cards can be found in Table C.2. For the

descriptive statistics of all households, see Table C.2.

3.4 Empirical Strategy

We are interested in estimating the effects of political connections on the employment

status of the beneficiaries. However, identifying the effects of political connections is not

straightforward since leaders are not randomly elected. We capitalize on India’s historical

caste system and construct a measure of connections using the last names of leaders and

beneficiaries, which reflect their caste (jati). There are hundreds of castes within caste

groups. We compare beneficiaries with the same caste (jati) as the village council chairper-

son (pradhan) to those that do not have the same jati as pradhan. We estimate the equation

of the following form.

Y𝑖 𝑗 𝑐𝑣 = 𝛽𝑆𝑎𝑚𝑒𝑖 𝑗 𝑐𝑣 + 𝜃′𝑋𝑖 𝑗 𝑐𝑣 + 𝛼𝑣 + 𝜂𝑐 × 𝛿𝑣,𝑑 + 𝜖𝑖 𝑗 𝑐𝑣

(3.1)

Where, 𝑌 is an outcome variable, i.e., application for the job, whether work and the number

of days worked in NREGS for individual (𝑖) of jati or caste ( 𝑗), caste group (𝑐) in a village

(𝑣). 𝑆𝑎𝑚𝑒 is a measure of political connection that takes on the value one if the village

council chairperson and beneficiaries are of the same jati and zero otherwise.

𝑋 is a vector of household- and individual-level characteristics, including gender; age;

religion, and marital status of the head of the household and individuals. We also include

caste-village-level controls that aggregate household and individual time-invariant variables

by caste group and village to control for differences in caste-group composition within a

village. 𝛼 is the village-level fixed effect that captures village-level differences, 𝜂𝑐 × 𝛿𝑣,𝑑

denotes the caste-by-district fixed effect, and/or the caste-by-village fixed effect captures the

differences in caste composition across districts and villages. 𝜖 is the random error term.

Identification To identify 𝛽, we take advantage of the caste system in India. Caste (jati)

is assigned at birth, and there are numerous castes in a caste group. The role roll-out of

97

8

9

the NREGS program in a district and village is unlikely to affect the caste composition of

villages in the short run. However, the caste (jati) composition of the village can affect the

election results. Therefore, we also conduct a robustness analysis by dropping the villages

with higher shares of a caste since leaders might face little competition in those villages and

may have little incentive to reward voters at the margin.

3.4.1

Instrumental Variable Framework

We can estimate the equation 3.1 using OLS and incorporate village and caste group

dummies to account for unobservable variables at the village level. However, the estimated 𝛽

may still be potentially biased if certain unobservable factors at the village-caste or subcaste

level, such as hierarchy and norms, are correlated with political connections and influence

the election of leadership. To address this potential endogeneity problem, we leverage a

distinctive institutional feature of political reservation in India, introduced through the 73rd

and 74th constitutional amendments.

The 73rd and 74th amendments require that the seats of the local village council heads

be reserved for lower caste groups, such as the Scheduled Caste and Scheduled Tribe13. The

proportion of seats reserved for the Scheduled Caste and Scheduled Tribe is determined

based on their population share. At each panchayat level, reservation is then on a rotation

basis. We leverage this procedure and employ an instrument that takes a value of 1 if

the village panchayat is reserved for the same caste group, and 0 otherwise. Using this

instrument, we estimate the first-stage equation in the following form.

𝑆𝑎𝑚𝑒𝑖 𝑗 𝑐𝑣 = 𝛾RS𝑐,𝑣 + 𝜃′𝑋𝑖 𝑗 𝑐𝑣 + 𝛼𝑣 + 𝜂𝑐 × 𝛿𝑣,𝑑 + 𝜇𝑖 𝑗 𝑐𝑣

(3.2)

Where 𝑆𝑎𝑚𝑒 is a dummy variable that takes the value 1 if the beneficiaries and village

council chairperson share the same caste (jati) as in equation (3.1). 𝑅𝑆 is an indicator

13The amendments also mandated that 1/3 of the total gram panchayat head seats must be reserved for

women, and in some states, this reservation was later extended to 50%.

98

9

9

variable that takes on the value 1 if the village council seats were reserved for the same

caste group as the beneficiaries, and 0 otherwise14. 𝑋 is a vector of controls that includes

household- and individual-level characteristics, as well as caste-by-village-level controls as

specified in equation (3.1). 𝛼 represents the village fixed effect, 𝜂 captures the caste-by-

district and/or caste-by-sub-district fixed effects, and 𝜇 represents the error term.

The validity of the instrument depends on whether it is correlated with the political

connection, i.e., 𝛾 ≠ 0, and is uncorrelated with the error term of the structural equation

3.1. The first-stage estimates reject the null of 𝛾 = 0 satisfying the relevance condition

for the instrument. And for the exclusion restriction, We argue that the instrumental

variable satisfies exclusion restriction since the 73rd and 74th suggest that village councils

(panchayats) must be reserved for lower caste groups on a rotational basis and the reserved

seats must be proportional to the state population share of the lower caste group

3.5 Results

In this section, we present the main findings on the relationship between political

connections and employment in NREGS. Table 3.2 presents the results of our main analysis.

Our main sample of analysis is restricted to job-card holders as it is a prerequisite for working

in NREGS. We also present the analysis of political connections and job card holding rate

in table C.6. Columns 1 and 3 of Table 3.2 present the coefficient of political connections

using the OLS, and columns 2 and 4 present estimates using 2SLS. Furthermore, we provide

different versions of the specifications with alternative fixed effects in Table C.4 using the

OLS method and in Table C.5 using the 2SLS method.

In Panel A of Table 3.2, we present our findings on the impact of having a village council

chairperson (pradhan) from the same caste (jati) on the application for work in the National

Rural Employment Guarantee Scheme (NREGS). Our OLS estimate suggests that having a

14See Gille (2018) which utilized a similar instrument. For a more detailed review of the caste system and

its impact on the economy, refer to Munshi (2019).

99

0

0

1

village council chairperson from the same jati increases the probability of applying for work

in NREGS by 4.2 percentage points (column 2), indicating a 13 percent increase relative to

the mean application rate. Estimated effects are slightly smaller in size if we don’t include

the district by caste fixed effects that allow us to control for cross-district differences in cast

composition that might be correlated with the caste of candidates running for election in

the village council.

In Panels B and C of Table 3.2, we present the estimated coefficients pertaining to the

likelihood of employment in NREGS (extensive margin) and the number of days worked

in the NREGS (intensive margin) when the village council chairperson belongs to the

same caste (jati). In Columns 1 and 3, we provide the OLS estimates. Overall, we find a

statistically significant relationship between the political connection measured as having the

same caste village council chairperson as beneficiaries and the likelihood of employment in

the NREGS. This is consistent with findings on the effects of political connection in other

contexts (Choi, Penciakova, and Saffie, 2021). The results are robust to the inclusion and

exclusion of controls, fixed effects, and different employment measures.

The Individuals who share the same caste as the chairperson experience a noticeable

increase of 3 percentage points in the probability of working in NREGS, which corresponds

to a 12 percent increase from the mean. This finding highlights the substantial influence of

caste dynamics on employment opportunities within the program. Furthermore, the effect of

having the same jati chairperson on the number of days worked in NREGS is 11 percentage

points, representing an 11 percent increase from the mean of NREGS days (Panel C). The

effect sizes across all three measures, including the likelihood of application, working, and

the number of days worked, are strikingly similar. This increase in employment in NREGS

when the village council belongs to the same jati highlights the influential role of caste

dynamics in shaping employment patterns within the NREGS program in India.

2SLS Estimates As discussed in section 3.4, unobservables at the caste level, such as hier-

100

1

0

1

Table 3.2 Political Connections and Its Effects on Employment

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

Village FEs
District by caste FEs
Controls
Estimation methods

(1)

(2)

(3)

(4)

0.038**
(0.019)

62,526
0.217
0.333
0.471

0.023**
(0.011)

62,526
0.260
0.244
0.430

Panel A: Applied for Work
0.042***
0.083**
(0.016)
(0.034)

62,526
0.067
0.333
0.471
16.57

62,526
0.209
0.333
0.471

Panel B: Worked

0.080**
(0.035)

62,526
0.067
0.333
0.471
18.74

0.048*
(0.027)

62,526
0.081
0.244
0.430
16.57

0.030***
(0.010)

0.054**
(0.027)

62,526
0.256
0.244
0.430

62,526
0.080
0.244
0.430
18.74

Panel C: Days

0.083**
(0.042)

0.205**
(0.082)

0.110***
(0.039)

0.201**
(0.083)

62,526
0.263
0.982
1.776

Y
N
Y
OLS

62,526
0.081
0.982
1.776
16.57

Y
N
Y
2SLS

62,526
0.259
0.982
1.776

Y
Y
Y
OLS

62,526
0.080
0.982
1.776
18.74

Y
Y
Y
2SLS

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results
from a separate regression. Dependent variables Applied” and Worked” are dummy variables that
take the value of 1 if a household member applied for a job and worked in NREGS, respectively,
and 0 otherwise. “Days” is the number of days that an individual worked in NREGS. Controls
include household-level variables such as head of the household’s age, sex, marital status, religion,
individual-level variables such as gender, age, education, and caste group by village-level variables
such as average age, education, sex, etc. Standard errors are clustered at village council and
caste-group levels. Robust standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05,
∗ 𝑝 < 0.10

101

2

0

1

archy and influence in a village, might be correlated with a measure of political connections.

This might result in bias in our OLS estimates. To correct this endogeneity bias, we use

an instrumental variable (IV) framework where the instrument is a dummy if the village

council is reserved for the same caste group. We argue that the instrumental variable satis-

fies exclusion restriction since the 73rd and 74th suggest that village councils (panchayats)

must be reserved for lower caste groups on a rotational basis and is proportional to the

state population share of the lower caste group. The first-stage coefficients are reported

in Panel D of Table C.5. The first stage coefficients are statistically significant across all

specifications, with F stat ranging from 16 to 19, indicating the relevance of the instrument.

We report the results of the 2sls estimates from various specifications in Table C.5.

However, our preferred specification is shown in columns 2 and 4 of Table 3.2, which

correspond to columns 2 and 4 of Table C.5.

In Panel A, column 4, we find that the

likelihood of applying to work in NREGS increased by 8 percentage points when the village

council chairperson belongs to the same caste. This is a 24 percent increase from the mean.

Similarly, we find that the likelihood of working in NREGS increased by 5.4 percentage

points, and NREGS days increased by 20 percentage points, which are equivalent to 22

percent and 20 percent increases from the mean, respectively (Table 3.2, Panel B and C).

2SLS estimates are almost twice as large as OLS estimates across all measures, including

application to work in NREGS, if got a work, and number of days worked. These differences

in coefficient can partially be justified by the fact that NREGS is a self-targeting poverty-

alleviating workfare program that requires manual labor, and the low caste group population

is more likely to work coupled with the nature of instrumental variable that requires village

councils to be reserved for the low caste group.

3.5.1 Role of the Village Council Members

In this subsection, we test whether the strength of political connections affects the

likelihood of obtaining public employment. As defined earlier, a direct connection, i.e.,

102

3

0

1

Table 3.3 Village Council Members of 𝑆𝑎𝑚𝑒 Caste (Jati) and Employment

(1)

(2)

(3)

𝑆𝑎𝑚𝑒 Caste VC Member

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste VC Member

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste VC Member

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Village FE
District-by-caste FE
Controls

0.001
(0.014)

62,524
0.171
0.333
0.471

0.016
(0.011)

62,526
0.207
0.244
0.430

0.056
(0.043)

62,526
0.210
0.982
1.776
Y
N
N

Panel A: Applied for Work
-0.001
(0.013)

62,526
0.217
0.333
0.471

Panel B: Worked
0.015
(0.010)

62,526
0.260
0.244
0.430

Panel C: Days
0.052
(0.041)

62,526
0.263
0.982
1.776
Y
Y
Y

-0.006
(0.012)

62,526
0.209
0.333
0.471

0.016*
(0.009)

62,526
0.256
0.244
0.430

0.056
(0.038)

62,526
0.259
0.982
1.776
Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council chief
share the same surname and belong to the same caste (jati). Each column reports results from a separate
regression. Dependent variables Applied” and Worked” are dummy variables that take the value of 1
if a household member applied for a job and worked in NREGS, respectively, and 0 otherwise. “Days”
is the number of days that an individual worked in NREGS. Controls include household-level variables
such as head of the household’s age, sex, marital status, religion, individual-level variables such as
gender, age, education, and caste group by village-level variables such as average age, education, sex,
etc. Standard errors are clustered at village council and caste-group levels. Robust standard errors are
reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

103

4

0

1

shared identity, with a village council leader is considered a strong connection, since

the village council chairperson has discretionary power over the allocation of resources,

including NREGS jobs. However, beneficiaries may also share the same caste as council

members, which we consider to be weaker connections as village council members act as

advisors but have little authority15. Based on this distinction, we construct a measure of

political connection using the jati or the caste of the village council members and whether

they are the same as the beneficiaries.

Using this measure, we estimate a variant of equation 3.1 that includes a measure of

political connection based on the same caste as village council members and report the

results in Table 3.3. We find no effect of having a village council member from the same

caste on the application for NREGS work. This holds across all specifications. However, we

find a small and marginally significant increase in the probability of working in NREGS if

a village council member belongs to the same caste. The increase is 1.6 percentage points,

which corresponds to an increase of 6.5 percent from the mean, much smaller than the

effect size estimated for the chairperson (Panel B of Table 3.3). The results for the number

of days worked in NREGS are reported in panel C of table 3.3. Although coefficients are

positive for the number of days, they are not precisely estimated.

Furthermore, since we find positive effects of connections with village-council members,

a natural question is whether these members act as substitutes for or complements to

the chairperson. To test this, we interact with the dummy for having a village-council

chairperson from the same caste as a dummy for having a member of the council from the

same caste. If the members complement the village-council head, then we should expect the

interaction term to be positive and statistically significant. We report the results in the Panel

15A village council in India typically consists of 5–15 members, depending on the size of the village. Village
council members may also play a role in allocating job cards, and working days since, in making decisions
about public programs, the village-council chairperson must take the council members into consideration.
For example, in NREGS, the types of projects and the number of jobs are determined by the village-council
chairperson in consultation with village council members and other stakeholders such as self-help groups.

104

B of Table C.8: for all outcome variables, although coefficients are positive, the interaction

term is not statistically different from zero.

Table 3.4 Village Council Head of 𝑆𝑎𝑚𝑒 Caste Group and Employment

(1)

(2)

(3)

𝑆𝑎𝑚𝑒 Caste Group Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste Group Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste Group Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Village FE
District-by-caste FE
Controls

Panel A: Applied for Work
0.007
(0.011)

-0.001
(0.011)

0.007
(0.015)

5

0

1

62,526
0.151
0.333
0.471

0.017
(0.011)

62,526
0.195
0.244
0.430

0.057
(0.044)

62,526
0.199
0.982
1.776
Y
N
N

62,526
0.200
0.333
0.471

Panel B: Worked
0.017**
(0.009)

62,526
0.251
0.244
0.430

Panel C: Days
0.054
(0.034)

62,526
0.254
0.982
1.776
Y
Y
N

62,526
0.209
0.333
0.471

0.009
(0.009)

62,526
0.256
0.244
0.430

0.025
(0.034)

62,526
0.259
0.982
1.776
Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable and takes on the value of 1 if both beneficiary and
village council head (chairperson) are from the same caste group such as SC, ST, OBC,
and Others. Each column reports results from a separate regression. Dependent variables
Applied” and Worked” are dummy variables that take the value of 1 if a household member
applied for a job and worked in NREGS, respectively, and 0 otherwise. “Days” is the number
of days that an individual worked in NREGS. Controls include household-level variables
such as head of the household’s age, sex, marital status, religion, individual-level variables
such as gender, age, education, and caste group by village-level variables such as average age,
education, sex, etc. Standard errors are clustered at village council and caste-group levels.
Robust standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

105

6

0

1

3.5.2 Same Caste Group

Another way to test the strength of political connections in our setting is to construct a

measure of connections based on caste group instead of caste (jati). The caste group is a

much larger group, and its members are likely to belong to different castes. Caste groups

are defined as SC, ST, OBC, and Others.

We construct the measure of political connection based on these groups and construct

a dummy if the village council chairperson is of the same caste group as the beneficiaries.

We then run a similar regression as in equation 3.1.

Results are reported in Table 3.4. In Panel A of 3.4, we report the estimated effect on

the probability of applying for NREGS work and find no effects of having a village council

head from the same caste group. In Panel B, when we do not include caste by village level

controls, there is a small and marginally significant effect of having a chairperson from

the same caste group, but it becomes zero once we include caste by village level controls.

We also find no effects on the number of days worked in NREGS. Although coefficients

for the number of days are positive, none of the coefficients reported in Table 3.4, Panel C

are significantly different from zero, suggesting that political connections based on caste

groups are weaker than those based on caste or jati

3.6 Robustness Checks and Mechanism

In this section, we present various robustness checks. We present the main analysis

with different sample restrictions and also a falsification test with placebo treatment, i.e., a

measure of connections constructed based on similar-sounding first names instead of last

names and jati.

3.6.1 Sample Restrictions

In our first robustness check, we dropped the villages that have the largest share of sub-

caste. One of the primary concerns we had was that if a subcaste dominated a village, the

106

7

0

1

chairperson of the village council might not face competition, especially if the caste shapes

the political dynamics in the elections, and might be less inclined to distribute benefits to

individuals from the same caste.

Table 3.5 Effects of Political Connections by Sample Restrictions

VARIABLES

𝑆𝑎𝑚𝑒 Caste Chairperson

(1)

(2)

Applied Worked

(3)
Days

(4)

(5)

Applied Worked

(6)
Days

Panel A: OLS Estimates

0.038*
(0.020)

0.028**
(0.011)

0.085**
(0.034)

0.040**
(0.017)

0.026***
(0.010)

0.078**
(0.031)

Observations
R-squared

59,937
0.233

59,937
0.273

59,937
0.277

59,937
0.223

59,937
0.268

59,937
0.273

𝑆𝑎𝑚𝑒 Caste Chairperson

Panel B: 2SLS Estimates

0.070*
(0.037)

0.052**
(0.026)

0.194**
(0.080)

0.082**
(0.038)

0.066**
(0.030)

0.192**
(0.082)

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

59,937
0.075
0.336
0.472
15.86

Village FE
Caste by State FE
Controls

Y
N
Y

59,937
0.087
0.248
0.432
15.86

Y
N
Y

59,937
0.087
0.823
1.493
15.86

Y
N
Y

59,937
0.074
0.336
0.472
16.68

Y
Y
Y

59,937
0.086
0.248
0.432
16.68

Y
Y
Y

59,937
0.087
0.823
1.493
16.68

Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results
from a separate regression. Dependent variables Applied” and Worked” are dummy variables that
take the value of 1 if a household member applied for a job and worked in NREGS, respectively,
and 0 otherwise. “Days” is the number of days that an individual worked in NREGS. Controls
include household-level variables such as head of the household’s age, sex, marital status, religion,
individual-level variables such as gender, age, education, and caste group by village-level variables
such as average age, education, sex, etc. Standard errors are clustered at village council and caste-
group levels. Robust standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

However, in situations where a village comprises multiple castes (jati), candidates could

potentially face competition during elections and are more likely to reward voters of the

same caste in order to gain their votes (Munshi and Rosenzweig, 2013). To ensure that our

results are not driven by the largest caste share in the village, we dropped those villages.

This leaves us with a sample of villages with multiple castes within a caste group. We

report the results in Table 3.5. In panel A of Table 3.5, we present OLS estimates and 2SLS

107

8

0

1

estimates in panel B. We find the results presented in the main analysis are not driven by

villages with the largest share of castes.

3.6.2 Falsification Test

Since our measure of political connection is based on shared jati or surname, we

conduct a falsification test with a placebo treatment to ensure that our results are indeed

driven by shared jati. To conduct a falsification test, we create a placebo measure of political

connection. The placebo connection is constructed using probabilities matching of similar

sounding first names only. We also do not include any restrictions, such as matching within

villages or states. First names do not typically reveal an individual’s caste or occupation in

India or elsewhere, so this test serves as a falsification test against our main results.

Table 3.6 Placebo Regression of Political Connection’s Effects

VARIABLES

Applied Worked

(1)

(2)

(3)
Days

(5)

(6)

Applied Worked

(7)
Days

𝑆𝑎𝑚𝑒 (Placebo)

-0.006
(0.004)

0.003
(0.004)

0.011
(0.014)

-0.005
(0.004)

0.000
(0.004)

-0.000
(0.013)

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Village FE
Controls
Caste by State FE

62,524
0.217
0.333
0.471

Y
Y
N

62,524
0.207
0.244
0.430

Y
Y
N

62,524
0.209
0.812
1.487

Y
Y
N

62,526
0.209
0.333
0.471

Y
Y
Y

62,526
0.256
0.244
0.430

Y
Y
Y

62,526
0.258
0.812
1.487

Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒(𝑃𝑙𝑎𝑐𝑒𝑏𝑜) is an indicator variable that takes on the value of 1 if a local and the
village-council chief have similar-sounding first names. Each column reports results from a
separate regression. Dependent variables Applied” and Worked” are dummy variables that take
the value of 1 if a household member applied for a job and worked in NREGS, respectively,
and 0 otherwise. “Days” is the number of days that an individual worked in NREGS. Controls
include household-level variables such as head of the household’s age, sex, marital status,
religion, individual-level variables such as gender, age, education, and caste group by village-
level variables such as average age, education, sex, etc. Standard errors are clustered at village
council and caste-group levels. Robust standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01,
∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

We run a similar regression as in 3.1 but replace our original measure of political

connection with a placebo measure of political connection. The results are reported in

108

9

0

1

Table 3.6. As expected, we do not reject the null and indeed find evidence in favor of our

main results. This suggests that caste identity and political connections play an important

role in public works programs and that political leaders may be systematically allocating

jobs to their own groups.

3.6.3 Mechanism

The main results show a positive association between political connections and job

allocations in NREGS. This could be because local leaders aim to win votes by favoring

individuals of the same caste, a supply-side channel, or because individuals are more likely

to reach out to elected leaders when they share the same caste identity, a demand-side

channel. It is important to distinguish between the demand and supply channels, but doing

so can be challenging in this context.

To elucidate the mechanism behind the main effects, we analyze information on local

meetings in the village council and whether village residents approached the chairperson

for any problems. If increased confidence serves as the dominant channel and the rise in

participation in NREGS is mediated by it, we should observe a positive and significant

correlation between having the same jati chairperson and attendance at meetings, as well as

approaching the leader. Similarly, we anticipate a similar correlation between political con-

nections and individuals’ contact with local decision-makers to address their own problems.

The corresponding results are presented in Table 3.7. Panel A of Table 3.7 presents the es-

timated coefficient from the OLS, while Panel B reports the coefficient estimated via 2SLS.

The first two columns in Table 3.7 display the coefficient of the same caste chairperson on

village council meeting attendance. Our findings reveal no significant effect of caste-based

political connections on attendance at village council meetings. Additionally, we find no

effects of the same caste (jati) council chairperson on whether an individual approached

the chairperson for a problem. The results are consistent for both OLS and 2SLS estimates,

suggesting that the demand channel operating through self-confidence when having the

109

0

1

1

Table 3.7 Local Panchayat Meetings and Leaders Accessability

VARIABLES

Any

Total

leaders

problems

(1)

(2)

GS meetings

(3)
Approched

(6)
Solved local

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

Village FE
Caste by Dist FE
Controls

0.015
(0.012)

60,208
0.144

-0.068
(0.057)

60,208
0.106
0.199
0.400
12.62

Y
Y
Y

Panel A: OLS
0.001
(0.012)

0.001
(0.054)

60,208
0.048

60,208
0.111

Panel B : 2SLS
-0.024
(0.038)

-0.253
(0.171)

60,208
0.024
0.362
1.430
12.62

Y
Y
Y

60,208
0.067
0.126
0.333
12.62

Y
Y
Y

0.013
(0.008)

60,208
0.107

0.015
(0.019)

60,208
0.028
0.0435
0.204
12.62

Y
Y
Y

Note: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results from
a separate regression. Controls include household-level variables such as head of the household’s
age, sex, marital status, religion, individual-level variables such as gender, age, education, and
caste group by village-level variables such as average age, education, sex, etc. Standard errors
are clustered at village council and caste-group levels. Robust standard errors are reported in
parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

same caste village council chairperson might not be the dominant one.

However, if patronage politics are at play in the village election, we should expect local

leaders to allocate employment to individuals who may support the voter as a reward for

their backing during the election (Chandra, 2007). Although we lack data on individuals’

support for the local leader of the same jati in the election, we do have information for each

political candidate who ran for election in the last two panchayats and whether they received

caste support during the election. We construct a dataset with candidate and election year

information and run a regression of the following form:

110

1

1

1

Table 3.8 Likelihood of Winning and Caste Support in Elections

VARIABLES

Female

Education

Same Caste Support

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Year FEs
State by Year FEs
Controls
State FEs
District FFs
Village FEs

(1)
All

(2)
All

(3)
All

(4)
SC/ST

(5)
SC/ST

(6)
SC/ST

-0.072***
(0.017)
0.033***
(0.011)
0.074**
(0.030)

-0.057***
(0.018)
0.039***
(0.012)
0.098***
(0.035)

-0.074***
(0.022)
0.049***
(0.012)
0.103**
(0.042)

1,432
0.300
0.255
0.436

Y
Y
Y
Y
N
N

1,432
0.362
0.255
0.436

Y
Y
Y
N
Y
N

1,432
0.395
0.255
0.436

Y
Y
Y
N
N
Y

-0.011
(0.017)
-0.012
(0.008)
0.044**
(0.020)

1,432
0.169
0.0977
0.297

-0.022
(0.020)
-0.009
(0.009)
0.068***
(0.023)

-0.028
(0.025)
-0.006
(0.010)
0.074***
(0.027)

1,432
0.290
0.0977
0.297

1,432
0.358
0.0978
0.297

Y
Y
Y
Y
N
N

Y
Y
Y
N
Y
N

Y
Y
Y
N
N
Y

Note: Dependent variable is a dummy and takes on value one if the candidate won the local
election and zero otherwise. Observations are candidates by election years. Same caste support is
a dummy and takes on value 1 if the candidate in the election received support from the same caste
zero otherwise. Controls include caste groups, religion, support of the political party, and wealthy
people. Standard errors are clustered at the village council level. Robust standard errors are reported
in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

𝑊𝑖𝑛𝑖𝑣𝑡 = 𝛼0 + 𝛼1𝐶𝑆𝑖𝑣𝑡 + 𝛽′𝑋𝑖𝑣𝑡 + 𝛼𝑣,𝑑,𝑠 + 𝜏𝑡 + 𝜖𝑖𝑣𝑡

(3.3)

𝑊𝑖𝑛 is a dummy if the candidate is a winner, 𝛼1 measures the effects of having the

same caste support in the election on the likelihood of winning the election, 𝑋 is a vector of

controls that includes candidate characteristics. 𝛼 is a fixed effect for either village, districts,

or state, 𝜏 is year-fixed effects as well as state-by-year fixed effects, and 𝜖 is an error term.

The estimation results using equation 3.3 are reported in Table 3.8. Across all speci-

fications, we find a significant association between same-caste support and the probability

of winning. The first three columns report the results for all candidates, while columns

4-6 present results for SC and ST candidates. We find that caste plays a significant role in

determining electoral success, not only for all candidates but also for lower caste groups

111

2

1

1

such as SC/ST. Based on these observations, we propose that caste is an important factor in

local elections, and local leaders systematically allocate resources to fellow caste members

as a reward for their votes. Considering the combined results from Table 3.8 and the main

results from 3.2, we find substantial evidence of preferential targeting in NREGS at the jati

level.

3.7 Conclusion

We study the impact of identity-based political connections on job applications, employ-

ment probability, and workdays in NREGS. We measure political connections by comparing

the surnames of local residents with those of local leaders (village council chairperson). The

results show that the probability of working in NREGS increases by 3-5.4 percentage points

if the village head is of the same caste as the applicant. This corresponds to a 12-20 percent

increase relative to the mean at the extensive margin. The effect of political connections on

the number of days worked in NREGS is 11-20 percentage points higher for workers who

share the same caste as the village-council chair. This represents an 11-20 percent increase

relative to the mean worked day.

Our results suggest that the effects of political connections are more pronounced among

lower-caste groups, indicating strategic targeting and favoritism within these groups. When

we exclude surnames and consider only caste groups (SC, ST, OBC, and Forward) of

residents and village leaders, we find no evidence of targeting. These results are consistent

with findings on affirmative action but also reveal significant heterogeneity at the jati level

in the strategic targeting of public goods. Overall, the results are in line with the literature

on patronage politics, which can generate inequality and misallocation in the labor market.

Providing better information to all residents about the program might address this issue and

could help to mitigate patronage politics.

112

3

1

1

REFERENCES

Afridi, Farzana, Vegard Iversen, and MR Sharan. 2017. “Women political leaders, cor-
ruption, and learning: Evidence from a large public program in India.” Economic
Development and Cultural Change 66 (1):1–30.

Akerlof, George A and Rachel E Kranton. 2000. “Economics and identity.” The quarterly

journal of economics 115 (3):715–753.

Anderson, Siwan. 2011. “Caste as an Impediment to Trade.” American Economic Journal:

Applied Economics 3 (1):239–63.

Azam, Mehtabul. 2011. “The impact of Indian job guarantee scheme on labor market

outcomes: Evidence from a natural experiment.” Available at SSRN 1941959 .

Banerjee, Abhijit and Rohini Somanathan. 2007. “The political economy of public goods:

Some evidence from India.” Journal of development Economics 82 (2):287–314.

Bardhan, Pranab and Dilip Mookherjee. 2005. “Decentralizing antipoverty program deliv-

ery in developing countries.” Journal of public economics 89 (4):675–704.

Bardhan, Pranab K and Dilip Mookherjee. 2000. “Capture and governance at local and

national levels.” American economic review 90 (2):135–139.

Beaman, Lori, Raghabendra Chattopadhyay, Esther Duflo, Rohini Pande, and Petia
Topalova. 2009. “Powerful women: does exposure reduce bias?” The Quarterly journal
of economics 124 (4):1497–1540.

Beaman, Lori, Esther Duflo, Rohini Pande, and Petia Topalova. 2012. “Female leadership
raises aspirations and educational attainment for girls: A policy experiment in India.”
science 335 (6068):582–586.

Berg, Erlend, Sambit Bhattacharyya, D Rajasekhar, and R Manjula. 2018. “Can public
works increase equilibrium wages? Evidence from India’s National Rural Employment
Guarantee.” World Development 103:239–254.

Besley, Timothy and Anne Case. 1995. “Does electoral accountability affect economic
policy choices? Evidence from gubernatorial term limits.” The Quarterly Journal of
Economics 110 (3):769–798.

Besley, Timothy, Rohini Pande, Lupin Rahman, and Vijayendra Rao. 2004. “The politics
of public good provision: Evidence from Indian local governments.” Journal of the
European Economic Association 2 (2-3):416–426.

113

4

1

1

Blakeslee, Jennifer Bussell, Kanchan Chandra, Simon Chauchard, Natasha Chichilnisky-I-
Ieal, Miriam Golden, Don Green, Rajeev Gowda, Macartan Humphreys, Lakshmi Iyer,
Francesca Jensenius et al. 2013. “Ethnic Quotas and Political Mobilization: Caste,
Parties, and Distribution in Indian Village Councils.” American Political Science Review
107 (1).

Bose, Nayana. 2017. “Raising consumption through India’s national rural employment

guarantee scheme.” World Development 96:245–263.

Bose, Nayana and Shreyasee Das. 2018. “Political reservation for women and delivery of

public works program.” Review of Development Economics 22 (1):203–219.

Caeyers, Bet and Stefan Dercon. 2012. “Political connections and social networks in targeted
transfer programs: Evidence from rural Ethiopia.” Economic Development and Cultural
Change 60 (4):639–675.

Cassan, Guilhem. 2015. “Identity-based policies and identity manipulation: Evidence from

colonial Punjab.” American Economic Journal: Economic Policy 7 (4):103–31.

———. 2019. “Affirmative action, education and gender: Evidence from India.” Journal

of Development Economics 136:51–70.

Chandra, Kanchan. 2007. Why ethnic parties succeed: Patronage and ethnic head counts

in India. Cambridge University Press.

Chattopadhyay, Raghabendra and Esther Duflo. 2004. “Women as policy makers: Evidence

from a randomized policy experiment in India.” Econometrica 72 (5):1409–1443.

Chau, Nancy H, Yanyan Liu, and Vidhya Soundararajan. 2018. Political activism as a
determinant of clientelistic transfers: Evidence from an Indian public works program,
vol. 1700. Intl Food Policy Res Inst.

Choi, Joonkyu, Veronika Penciakova, and Felipe Saffie. 2021. “Political connections,

allocation of stimulus spending, and the jobs multiplier.” .

Clots-Figueras, Irma. 2012. “Are female leaders good for education? Evidence from India.”

American Economic Journal: Applied Economics 4 (1):212–44.

Colonnelli, Emanuele, Mounu Prem, and Edoardo Teso. 2020. “Patronage and selection in

public sector organizations.” American Economic Review 110 (10):3071–99.

Cook, C Justin and Manisha Shah. 2022. “Aggregate effects from public works: Evidence

from india.” The Review of Economics and Statistics 104 (4):797–806.

114

5

1

1

Das, Maitreyi Bordia. 2013. “Exclusion and discrimination in the labor market.” World

Development Report .

Dasgupta, Aparajita. 2017. “Can the major public works policy buffer negative shocks in
early childhood? Evidence from Andhra Pradesh, India.” Economic Development and
Cultural Change 65 (4):767–804.

Deininger, Klaus and Yanyan Liu. 2013. Welfare and Poverty Impacts of India S National
Rural Employment Guarantee Scheme: Evidence from Andhra Pradesh, vol. 1289. Intl
Food Policy Res Inst.

Deininger, Klaus, Hari K Nagarajan, and Sudhir K Singh. 2020. “Women’s political
leadership and economic empowerment: Evidence from public works in India.” Journal
of comparative economics 48 (2):277–291.

Desai, Kunjal. 2016. “A rmative Action, Caste and the Marriage Market.” .

Desai, Sonalde. 2018. “Do public works programs increase women’s economic empower-

ment?: evidence from rural India.” .

Deshpande, Ashwini. 2011. The grammar of caste: Economic discrimination in contem-

porary India. Oxford University Press.

———. 2018. “Caste, class and social mobility: The critical role of high school years.”

Tech. rep., mimeo.

———. 2019. “Double jeopardy? Stigma of identity and affirmative action.” The Review

of Black Political Economy 46 (1):38–64.

Deshpande, Ashwini and Katherine Newman. 2007. “Where the path leads: The role
of caste in post-university employment expectations.” Economic and Political Weekly
:4133–4140.

Deshpande, Manali S. 2010. “History of the Indian caste system and its impact on India

today.” .

Dumont, Luois. 1980. The caste system and its implications. University of Chicago Press

Chicago.

Fafchamps, Marcel and Julien Labonne. 2017. “Do politicians’ relatives get better jobs?
evidence from municipal elections.” The Journal of Law, Economics, and Organization
33 (2):268–300.

115

6

1

1

Gagliarducci, Stefano and Marco Manacorda. 2020. “Politics in the family: Nepotism and
the hiring decisions of italian firms.” American Economic Journal: Applied Economics
12 (2):67–95.

Gaikwad, Nikhar. 2022. “Identity Politics and Economic Policy.” Unpublished Manuscript.

Columbia University .

Gehrke, Esther. 2019. “An employment guarantee as risk insurance? Assessing the effects
of the NREGS on agricultural production decisions.” The World Bank Economic Review
33 (2):413–435.

Gille, Véronique. 2018. “Applying for social programs in India: Roles of local politics and
caste networks in affirmative action.” Journal of Comparative Economics 46 (2):436–456.

Hoff, Karla, Mayuresh Kshetramade, and Ernst Fehr. 2011. “Caste and punishment: The
legacy of caste culture in norm enforcement.” The economic journal 121 (556):F449–
F475.

Hoff, Karla and Priyanka Pandey. 2006. “Discrimination, social identity, and durable

inequalities.” American Economic Review 96 (2):206–211.

Hsieh, Chang-Tai, Erik Hurst, Charles I Jones, and Peter J Klenow. 2019. “The allocation

of talent and us economic growth.” Econometrica 87 (5):1439–1474.

Imbert, Clement and John Papp. 2015. “Labor market effects of social programs: Evidence
from india’s employment guarantee.” American Economic Journal: Applied Economics
7 (2):233–63.

Imbert, Clément and John Papp. 2020. “Costs and benefits of rural-urban migration:

Evidence from India.” Journal of Development Economics 146:102473.

Iversen, Vegard, Anirudh Krishna, Kunal Sen et al. 2016. “Rags to riches? Intergenerational

occupational mobility in India.” Tech. rep., GDI, The University of Manchester.

Iyer, Lakshmi, Tarun Khanna, and Ashutosh Varshney. 2013. “Caste and entrepreneurship

in India.” Economic and Political Weekly :52–60.

Jaspal, Rusi. 2011. “Caste, social stigma and identity processes.” Psychology and Devel-

oping Societies 23 (1):27–62.

Kumar, Hemanshu, Rohini Somanathan et al. 2017. “Caste connections and government

transfers: The Mahadalits of Bihar.” Tech. rep.

116

Macdonell, Arthur A. 1914. “The early history of caste.” The American Historical Review

19 (2):230–244.

Markussen, Thomas and Finn Tarp. 2014. “Political connections and land-related investment

in rural Vietnam.” Journal of Development Economics 110:291–302.

Moser, Christine. 2008. “Poverty reduction, patronage, or vote buying? the allocation of
public goods and the 2001 election in Madagascar.” Economic Development and Cultural
Change 57 (1):137–162.

Munshi, Kaivan. 2019. “Caste and the Indian economy.” Journal of Economic Literature

57 (4):781–834.

Munshi, Kaivan and Mark Rosenzweig. 2013. “Networks, commitment, and competence:

caste in Indian local politics.” Tech. rep., National Bureau of Economic Research.

7

1

1

———. 2015. “Insiders and outsiders: local ethnic politics and public goods provision.”

Tech. rep., National Bureau of Economic Research.

Muralidharan, Karthik, Paul Niehaus, and Sandip Sukhtankar. 2016. “Building state ca-
pacity: Evidence from biometric smartcards in India.” American Economic Review
106 (10):2895–2929.

Oh, Suanna. 2019. “Does Identity Affect Labor Supply?” Job Marker Paper, Columbia

University .

Persson, Torsten, Gerard Roland, and Guido Tabellini. 2000. “Comparative politics and

public finance.” Journal of political Economy 108 (6):1121–1161.

Rao, Gautam. 2019. “Familiarity does not breed contempt: Generosity, discrimination, and

diversity in Delhi schools.” American Economic Review 109 (3):774–809.

Sankaran, Sindhuja, Maciek Sekerdej, and Ulrich Von Hecker. 2017. “The role of In-
dian caste identity and caste inconsistent norms on status representation.” Frontiers in
psychology 8:487.

Schmutte, Ian. 2016. “how do social networks affect labor markets?” IZA World of Labor .

Shah, Manisha and Bryce Millett Steinberg. 2019. “Workfare and human capital investment:

Evidence from India.” Journal of Human Resources :1117–9201R2.

Sharma, KL. 1984. “Caste and class in India: Some conceptual problems.” Sociological

Bulletin 33 (1-2):1–28.

117

Sheahan, Megan, Yanyan Liu, Christopher B Barrett, and Sudha Narayanan. 2016. Pref-
erential resource spending under an employment guarantee: the political economy of
MGNREGS in Andhra Pradesh. The World Bank.

Vaid, Divya. 2014. “Caste in Contemporary India: Flexibility and Persistence.” Annual

Review of Sociology 40:391–410.

8

1

1

118

9

1

1

APPENDIX

Figure 3.3 Same Caste Population by Village

119

0

2

1

Figure 3.4 Same Caste Population by Age

120

1

2

1

Table C.1 Household Characteristics by Job Card Holding

Total

Jobcard

Yes

No

Variable

Mean

SD

Mean

SD

Mean

SD

Female head
Head’s age
Head’s education
Share of hhs
. . . with no education
. . . up to 5th grade
. . . up to high school
. . . college or above
Married
Household size
Prime male
Prime female
Children
Hindu
Muslim
SC
ST
OBC
OC
Own any land
. . . Size of land
job card

0.128
50.028
4.779

0.334
13.843
4.613

0.115
49.130
3.776

0.319
12.812
4.077

0.132
50.361
5.152

0.339
14.192
4.743

0.399
0.172
0.379
0.044
0.823
4.561
1.671
1.565
1.103
0.886
0.093
0.216
0.071
0.493
0.221
0.562
2.625
0.271

0.490
0.377
0.485
0.205
0.381
2.442
1.109
0.956
1.375
0.318
0.290
0.411
0.257
0.500
0.415
0.496
3.703
0.444

0.444
0.223
0.305
0.017
0.848
4.459
1.646
1.554
1.072
0.906
0.077
0.326
0.089
0.454
0.131
0.575
2.116
1.000

0.497
0.416
0.460
0.131
0.359
2.201
1.038
0.883
1.312
0.292
0.267
0.469
0.285
0.498
0.337
0.494
2.620
0.000

0.383
0.153
0.407
0.054
0.814
4.598
1.680
1.569
1.115
0.878
0.099
0.175
0.064
0.507
0.254
0.557
2.816
0.000

0.486
0.360
0.491
0.226
0.389
2.525
1.134
0.982
1.398
0.327
0.298
0.380
0.245
0.500
0.435
0.497
4.019
0.000

# Obs.

79,646

21582

58,064

Notes: Author’s own calculation from survey data

121

Table C.2 Summary Statistics by Caste Groups for Job Card Holders

Female head
Head’s age
Head’s education
Share of hhs
. . . with no education
. . . up to 5th grade
. . . up to high school
. . . college or above
Married
Household size
Prime male
Prime female
Children
Own any land
. . . Size of land
#Obs

Age
Female
Education
Formally applied
Worked in NREGS
# of Days worked
Received allowance
Wanted to work more
Attended GS meetings

Total

SC-ST

OBC

OTHER

Mean

SD

Mean

SD

Mean

SD

Mean

SD

Panel A: Household

0.115
49.130
3.776

0.319
12.812
4.077

0.124
48.653
3.267

0.329
12.887
3.969

0.110
49.490
4.009

0.313
12.812
4.098

0.106
49.392
4.587

0.308
12.526
4.137

2

2

1

0.444
0.223
0.305
0.017
0.848
4.459
1.646
1.554
1.072
0.575
2.116
21582

0.497
0.416
0.460
0.131
0.359
2.201
1.038
0.883
1.312
0.494
2.620

0.500
0.404
0.439
0.123
0.369
2.181
1.037
0.879
1.325
0.500
2.608

0.504
0.205
0.260
0.015
0.838
4.481
1.639
1.545
1.121
0.505
1.839
8959
Panel B: NREGS & Individuals

0.414
0.239
0.321
0.018
0.854
4.529
1.664
1.587
1.085
0.655
2.353
9806

0.493
0.426
0.467
0.132
0.353
2.315
1.063
0.915
1.359
0.475
2.654

15.651
0.500
4.538
0.471
0.430
29.836
0.141
0.325
0.398

39.886
0.490
4.696
0.333
0.244
37.894
0.020
0.120
0.197
62526

39.527
0.489
4.223
0.346
0.247
34.412
0.022
0.139
0.190
25991
Notes: Author’s own calculation from survey data

40.129
0.494
4.894
0.325
0.263
42.694
0.020
0.111
0.197
28327

15.521
0.500
4.496
0.476
0.431
25.982
0.145
0.346
0.392

15.786
0.500
4.561
0.468
0.440
33.138
0.141
0.314
0.398

0.358
0.220
0.392
0.023
0.858
4.142
1.603
1.471
0.870
0.517
1.853
2817

40.184
0.479
5.512
0.322
0.168
28.146
0.017
0.094
0.217
8208

0.479
0.415
0.488
0.149
0.349
1.801
0.949
0.776
1.061
0.500
2.395

15.573
0.500
4.429
0.467
0.373
22.205
0.129
0.292
0.412

122

3

2

1

Table C.3 Summary Statistics by Caste Groups for All Households

Variable

Mean

SD

Mean

SD

Mean

SD

Mean

SD

Total

SC-ST

OBC

OTHER

Female head
Head’s age
Head’s education
Share of hhs
. . . with no education
. . . up to 5th grade
. . . up to high school
. . . college or above
Married
Household size
Prime male
Prime female
Children
Own any land
. . . Size of land
Jobcard
#Obs

Age
Female
Education
Formally applied
Worked in NREGS
# of Days worked
Received allowance
Wanted to work more
Attended GS meetings
#Obs

Panel A: Household

0.128
50.028
4.779

0.334
13.843
4.613

0.131
48.736
3.969

0.338
13.738
4.426

0.126
50.146
4.748

0.332
13.859
4.585

0.127
51.442
5.903

0.333
13.792
4.682

0.399
0.172
0.379
0.044
0.823
4.561
1.671
1.565
1.103
0.562
2.625
0.271
79646

0.49
0.377
0.485
0.205
0.381
2.442
1.109
0.956
1.375
0.496
3.703
0.444

0.47
0.175
0.313
0.033
0.815
4.487
1.633
1.519
1.142
0.468
1.933
0.392
22826

0.499
0.38
0.464
0.177
0.388
2.297
1.078
0.913
1.353
0.499
2.606
0.488

0.402
0.174
0.38
0.04
0.826
4.674
1.696
1.594
1.16
0.615
2.765
0.25
39256

0.49
0.379
0.485
0.197
0.379
2.542
1.131
0.979
1.432
0.487
3.895
0.433

Panel B: NREGS & Individuals

16.139
0.5
4.895
0.319
0.245
29.85
0.088
0.176
0.385

40.081
0.491
5.58
0.115
0.064
37.059
0.008
0.032
0.181
249923
Notes: Author’s own calculation from survey data

40.066
0.495
5.501
0.103
0.062
41.507
0.008
0.027
0.173
126341

39.418
0.488
4.757
0.161
0.099
33.685
0.01
0.055
0.179
68398

15.765
0.5
4.779
0.368
0.298
26.248
0.101
0.228
0.383

16.189
0.5
4.887
0.305
0.242
33.01
0.09
0.161
0.379

0.303
0.162
0.464
0.067
0.828
4.403
1.663
1.559
0.925
0.564
2.95
0.16
17564

40.936
0.487
6.783
0.083
0.026
28.528
0.004
0.015
0.202
55184

0.46
0.369
0.499
0.251
0.378
2.384
1.1
0.958
1.254
0.496
4.026
0.367

16.438
0.5
4.822
0.275
0.159
22.76
0.06
0.122
0.401

123

4

2

1

Table C.4 Political Connections and Its Effects on Employment

(1)

(2)

(3)

(4)

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Village FE
District-by-caste FE
Controls

0.040*
(0.021)

62,526
0.172
0.333
0.471

0.025**
(0.013)

62,526
0.207
0.244
0.430

0.091*
(0.053)

62,526
0.210
0.982
1.776
Y
N
N

Panel A: Applied for Work
0.038**
(0.019)

0.044**
(0.018)

62,526
0.217
0.333
0.471

62,526
0.162
0.333
0.471

Panel B: Worked

0.023**
(0.011)

62,526
0.260
0.244
0.430

0.029**
(0.011)

62,526
0.201
0.244
0.430

Panel C: Days

0.083**
(0.042)

0.105**
(0.046)

62,526
0.263
0.982
1.776
Y
N
Y

62,526
0.205
0.982
1.776
Y
Y
N

0.042***
(0.016)

62,526
0.209
0.333
0.471

0.030***
(0.010)

62,526
0.256
0.244
0.430

0.110***
(0.039)

62,526
0.259
0.982
1.776
Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results
from a separate regression. Dependent variables Applied” and Worked” are dummy variables that
take the value of 1 if a household member applied for a job and worked in NREGS, respectively,
and 0 otherwise. “Days” is the number of days that an individual worked in NREGS. Controls
include household-level variables such as head of the household’s age, sex, marital status, religion,
individual-level variables such as gender, age, education, and caste group by village-level variables
such as average age, education, sex, etc. Standard errors are clustered at village council and caste-
group levels. Robust standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

124

5

2

1

Table C.5 IV Estimates of Political Connections Impacts

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

𝑆𝑎𝑚𝑒 Caste Chairperson

(1)

(2)

(3)

(4)

0.091*
(0.052)

62,526
0.066
0.333
0.471
15.84

0.077**
(0.035)

62,526
0.080
0.244
0.430
15.84

Panel A: Applied for Work

0.083**
(0.034)

0.085**
(0.034)

0.080**
(0.035)

62,526
0.067
0.333
0.471
16.57

62,526
0.067
0.333
0.471
18.25

62,526
0.067
0.333
0.471
18.74

Panel B: Worked

0.048*
(0.027)

62,526
0.081
0.244
0.430
16.57

0.060**
(0.027)

0.054**
(0.027)

62,526
0.079
0.244
0.430
18.25

62,526
0.080
0.244
0.430
18.74

Panel C: Days

0.292**
(0.115)

0.205**
(0.082)

0.225***
(0.083)

0.201**
(0.083)

Observations
R-squared
Dep. Var Mean
Dep. Var Std
Sanderson-Windmeijer F stat

62,526
0.080
0.982
1.776
15.84

62,526
0.081
0.982
1.776
16.57

62,526
0.079
0.982
1.776
18.25

62,526
0.080
0.982
1.776
18.74

Same Caste Reserved

Observations
R-squared

Village FE
District-by-caste FE
Controls

Panel D: First Stage

0.106***
(0.028)
62,526
0.487

0.108***
(0.024)
62,526
0.490

0.132***
(0.032)
62,526
0.511

0.137***
(0.029)
62,526
0.513

Y
N
N

Y
N
Y

Y
Y
N

Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results
from a separate regression. Dependent variables Applied” and Worked” are dummy variables that
take the value of 1 if a household member applied for a job and worked in NREGS, respectively,
and 0 otherwise. “Days” is the number of days that an individual worked in NREGS. Controls
include household-level variables such as head of the household’s age, sex, marital status, religion,
individual-level variables such as gender, age, education, and caste group by village-level variables
such as average age, education, sex, etc. Standard errors are clustered at village council and caste-
∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05,
group levels. Robust standard errors are reported in parentheses.
∗ 𝑝 < 0.10

125

6

2

1

Table C.6 Political Connections and Jobcard

(1)

(2)

(3)

(4)

VARIABLES

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Village FEs
Caste by districts FEs
Controls

0.011
(0.022)

79,646
0.356
0.271
0.444

0.159**
(0.065)

79,646
0.024
0.271
0.444

Y
N
N

Panel A: OLS Estimates

0.020
(0.020)

79,646
0.387
0.271
0.444

0.019
(0.024)

79,646
0.388
0.271
0.444

Panel B: 2SLS Estimates
0.099*
(0.055)

0.072
(0.052)

79,646
0.046
0.271
0.444

Y
N
Y

79,646
0.019
0.271
0.444

Y
Y
N

0.027
(0.024)

79,646
0.402
0.271
0.444

0.093*
(0.054)

79,646
0.021
0.271
0.444

Y
Y
Y

Notes:𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief share the same surname and belong to the same caste (jati). Each column reports results
from a separate regression. The dependent variable is a dummy variable that takes the value of 1
if a household has a job card and zero otherwise. Controls include household-level variables such
as head of the household’s age, sex, marital status, religion, individual-level variables such as
gender, age, education, and caste group by village-level variables such as average age, education,
sex, etc. Standard errors are clustered at village council and caste-group levels. Robust standard
errors are reported in parentheses ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

126

7

2

1

Table C.7 Caste Heterogeneity in Effects of Political Connections

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared

𝑆𝑎𝑚𝑒 Caste Chairperson

Observations
R-squared

Village FsE
District-by-caste FEs
Controls

(1)
Applied

(2)
Worked

(4)
Days

0.060***
(0.014)
25,911
0.252

0.026**
(0.013)
28,327
0.210

0.087***
(0.030)
8,208
0.224

Y
Y
Y

Panel A: SC-ST
0.029**
(0.013)
25,911
0.283

Panel B: OBC
0.025**
(0.012)
28,327
0.249

Panel C: Other
-0.015
(0.023)
8,208
0.164

Y
Y
Y

0.103**
(0.050)
25,911
0.284

0.098**
(0.047)
28,327
0.250

-0.061
(0.092)
8,208
0.170

Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-
council chief share the same surname and belong to the same caste (jati). Each column
reports results from a separate regression. Dependent variables Applied” and Worked” are
dummy variables that take the value of 1 if a household member applied for a job and
worked in NREGS, respectively, and 0 otherwise. “Days” is the number of days that an
individual worked in NREGS. Controls include household-level variables such as head of
the household’s age, sex, marital status, religion, individual-level variables such as gender,
age, education, and caste group by village-level variables such as average age, education,
sex, etc. Standard errors are clustered at village council and caste-group levels. Robust
standard errors are reported in parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

127

8

2

1

Table C.8 Do Village Council Members and Chairperson Have Differential Effects?

VARIABLES

𝑆𝑎𝑚𝑒 Caste Village Council Member

𝑆𝑎𝑚𝑒 Caste Village Council Chairperson

Observations
R-squared

𝑆𝑎𝑚𝑒 Caste Village Council Member

𝑆𝑎𝑚𝑒 Caste Village Council Chairperson

𝑆𝑎𝑚𝑒 Caste Member × Chairperson

Observations
R-squared
Village FEs
Districts by Caste FEs
Controls

(1)
Applied

-0.006
(0.012)
0.042***
(0.016)
62,526
0.209

0.000
(0.013)
0.031
(0.021)
0.024
(0.041)
62,526
0.217
Y
Y
Y

(2)
Worked

Panel A
0.016*
(0.009)
0.029***
(0.010)
62,526
0.256

Panel B
0.015
(0.010)
0.028**
(0.012)
0.017
(0.025)
62,526
0.260
Y
Y
Y

(3)
Days

0.044
(0.031)
0.088***
(0.032)
62,526
0.258

0.039
(0.035)
0.089**
(0.040)
0.022
(0.084)
62,526
0.261
Y
Y
Y

Notes: 𝑆𝑎𝑚𝑒 is an indicator variable that takes on the value of 1 if a local and the village-council
chief as well as village council members share the same surname and belong to the same caste
(jati). Each column reports results from a separate regression. Dependent variables Applied” and
Worked” are dummy variables that take the value of 1 if a household member applied for a job and
worked in NREGS, respectively, and 0 otherwise. “Days” is the number of days that an individual
worked in NREGS. Controls include household-level variables such as head of the household’s
age, sex, marital status, religion, individual-level variables such as gender, age, education, and
caste group by village-level variables such as average age, education, sex, etc. Standard errors
are clustered at village council and caste-group levels. Robust standard errors are reported in
parentheses. ∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

128

9

2

1

Table C.9 Different Caste (Jati) but Same Caste Group

Different Caste (Jati) Same Caste Group

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Different Caste (Jati) Same Caste Group

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Different Caste (Jati) Same Caste Group

Observations
R-squared
Dep. Var Mean
Dep. Var Std

Village FEs
Caste by districts FEs
Controls

(1)

(2)

(3)

(4)

Panel A: Applied for Work

0.001
(0.014)

62,526
0.151
0.333
0.471

0.011
(0.010)

62,526
0.195
0.244
0.430

0.030
(0.035)

62,526
0.199
0.982
1.776

Y
N
N

0.004
(0.010)

62,526
0.200
0.333
0.471

-0.002
(0.011)

62,526
0.207
0.333
0.471

Panel B: Worked
0.004
0.013
(0.011)
(0.008)

62,526
0.251
0.244
0.430

62,526
0.254
0.244
0.430

Panel C: Days

0.032
(0.027)

62,526
0.253
0.982
1.776

Y
N
Y

0.009
(0.034)

62,526
0.256
0.982
1.776

Y
Y
N

-0.016*
(0.009)

62,526
0.209
0.333
0.471

-0.007
(0.007)

62,526
0.256
0.244
0.430

-0.027
(0.024)

62,526
0.258
0.982
1.776

Y
Y
Y

Notes: Each column reports results from a separate regression. Dependent variables Applied” and
Worked” are dummy variables that take the value of 1 if a household member applied for a job and
worked in NREGS, respectively, and 0 otherwise. “Days” is the number of days that an individual
worked in NREGS. Controls include household-level variables such as head of the household’s
age, sex, marital status, religion, individual-level variables such as gender, age, education, and
caste group by village-level variables such as average age, education, sex, etc. Standard errors are
clustered at village council and caste-group levels. Robust standard errors are reported in parentheses.
∗∗∗ 𝑝 < 0.01, ∗∗ 𝑝 < 0.05, ∗ 𝑝 < 0.10

129