general equilibrium e ects of (improving) public
TRANSCRIPT
General Equilibrium Effects of (Improving) PublicEmployment Programs: Experimental Evidence from India∗
Karthik Muralidharan†
UC San DiegoPaul Niehaus‡
UC San DiegoSandip Sukhtankar§
University of Virginia
April 21, 2017
Abstract
Public employment programs play a large role in many developing countries’ anti-povertystrategies, but their impact on poverty reduction will depend on both direct program effectsas well as indirect effects through changes induced in market wages and employment. Weestimate both of these effects, exploiting a large-scale experiment that randomized the roll-outof a substantial improvement in the implementation of India’s public employment scheme. Wefind that this reform increased the overall earnings of low-income households by 13.3%, andreduced an income-based measure of poverty by 18.1% despite no increase in fiscal outlays onthe program. Income gains were overwhelmingly driven by higher private-sector earnings (90%)as opposed to program earnings (10%). Conservatively estimated, private sector low-skilledwages increased 6.1%, while days without paid work fell 7.1%. Migration, land utilizationand prices were unaffected. Our results highlight the importance of accounting for generalequilibrium effects in evaluating social programs, and also illustrate the feasibility of usinglarge-scale experiments to study them.
JEL codes: D50, D73, H53, J38, J43, O18
Keywords: public programs, general equilibrium effects, rural labor markets, NREGA, employ-ment guarantee, India
∗We thank Abhijit Banerjee, Gordon Dahl, Taryn Dinkelman, Gordon Hanson, Supreet Kaur, Aprajit Mahajan,Edward Miguel, and several seminar participants for comments and suggestions. We are grateful to officials ofthe Government of Andhra Pradesh, including Reddy Subrahmanyam, Koppula Raju, Shamsher Singh Rawat,Raghunandan Rao, G Vijaya Laxmi, AVV Prasad, Kuberan Selvaraj, Sanju, Kalyan Rao, and Madhavi Rani; aswell as Gulzar Natarajan for their continuous support of the Andhra Pradesh Smartcard Study. We are also gratefulto officials of the Unique Identification Authority of India (UIDAI) including Nandan Nilekani, Ram Sevak Sharma,and R Srikar for their support. We thank Tata Consultancy Services (TCS) and Ravi Marri, Ramanna, and ShubraDixit for their help in providing us with administrative data. This paper would not have been possible withoutthe continuous efforts and inputs of the J-PAL/IPA project team including Kshitij Batra, Prathap Kasina, PialiMukhopadhyay, Michael Kaiser, Frances Lu, Raghu Kishore Nekanti, Matt Pecenco, Surili Sheth, and PratibhaShrestha. Finally, we thank the Omidyar Network – especially Jayant Sinha, CV Madhukar, Surya Mantha, andSonny Bardhan – for the financial support that made this study possible.†UC San Diego, JPAL, NBER, and BREAD. [email protected].‡UC San Diego, JPAL, NBER, and BREAD. [email protected].§University of Virginia, JPAL, and BREAD. [email protected].
1 Introduction
Public works programs, in which the government provides daily-wage jobs to those who
seek them, are among the most common anti-poverty programs in developing countries.
The economic rationale for such programs (over direct income support for the poor) include
self-targeting through work requirements, public asset creation, and establishing a more
effective wage floor by having the government be an employer of last resort.1 An important
contemporary variant is the National Rural Employment Guarantee Scheme (NREGS) in
India. Launched in 2006, the scheme is the world’s largest workfare program, with 600
million rural residents eligible to participate and a fiscal allocation of 0.8% of India’s GDP.
A program of this scale and ambition raises several fundamental questions for research
and policy. First, how effective is it in reducing poverty? In particular, while direct income
support programs would typically reduce poverty, the general equilibrium effects of public-
works programs on the larger rural economy could amplify or attenuate the direct gains in
wage income for the poor.2 Second, what is the relative contribution of direct income gains
from the program and indirect income changes outside the program? Third, what are the
impacts on broader labor market outcomes such as wages, employment, and migration?
Given the importance of NREGS, a growing literature has tried to answer the questions
above, but the evidence to date has been hampered by two limitations. The first is identifi-
cation, with the results being quite sensitive to the methods used and studies often reaching
opposite conclusions depending on the identification strategy used (see Sukhtankar (forth-
coming) and section 2.1.2 below). Second, implementation of the scheme has proven so varied
that interpreting the existing evidence is difficult because heterogeneity in results could also
reflect variation in the extent of effective NREGS implemented in practice (that is not easy
to measure precisely). For example, the most-cited study of wage impacts finds them only
in states coded broadly as having implemented the program well (Imbert and Papp, 2015).
In this paper we aim to provide credible estimates of the anti-poverty impact of public
works programs by combining exogenous experimental variation, a demonstrable first-stage
impact on program implementation, and units of randomization that are large enough to
capture general equilibrium effects. Specifically, we worked with the Government of the
Indian state of Andhra Pradesh,3 to randomize the order in which 157 sub-districts (with an
1Work-fare programs may also be politically more palatable to tax-payers than unconditional doles. Suchprograms have a long history, with recorded instances from at least the 18th century in India, the publicworks conducted in the US by the Work Projects Administration during the Depression-era in the 1930s,and more modern “Food-for-Work” programs across Sub-Saharan Africa and Asia.
2A practical way of differentiating partial and general equilibrium effects (which we follow) is to definepartial equilibrium effects as those estimated at constant prices, and general equilibrium effects as thosewhich incorporate the effects of interventions on market prices.
3The original state of AP (with a population of 85 million) was divided into two states on June 2, 2014.
1
average population of 62,500 each) introduced a new technology (biometric Smartcards) for
making payments in NREGS. In prior work, we show that the new technology significantly
improved the performance of NREGS on several key dimensions: it reduced leakage or
diversion of funds, reduced delays between working and getting paid, reduced the time
required to collect payments, and increased access to work, without changing the fiscal
outlays on the program (Muralidharan et al., 2016). Thus, the Smartcard intervention
brought NREGS implementation in AP closer to what its architects intended (Khera, 2011).4
Evaluating the impact of improving NREGS implementation (as we do here) is not the
same as evaluating the impact of rolling out the program itself. Yet, given well-documented
implementation challenges in NREGS including poor access to work, high rates of leakage,
and long delays in receiving payments (Mehrotra, 2008; Imbert and Papp, 2011; Khera,
2011; Niehaus and Sukhtankar, 2013b), a significant improvement in implementation quality
is likely to result in a meaningful increase in a measure of effective NREGS.5 Further, since
significant improvements in program performance were achieved without increasing the fiscal
outlay on NREGS, our results on poverty impacts are more likely to reflect the structure of
NREGS rather than simply reflecting additional fiscal transfers to treated areas.
We report six main sets of results. First, we find large increases in household income
in areas where Smartcards were rolled out. We find this result both when using our own
survey data as well as when using data from the Socio-Economic and Caste Census (SECC), a
household census conducted by the national government independently of our activities. The
SECC collected coarse data by income categories of the highest earner in the household; we
find that the Smartcards intervention made it 4.8% (3.9 percentage point reduction on a base
proportion of 81.2%) more likely that this earner moves out of the lowest income category.
Using our survey data on income, we find a 13.3% (Rs. 9,594) increase in household income
in treated areas, which corresponds to a 18.1% reduction in an income-based measure of
poverty (a 5.2 percentage point reduction on a base poverty rate of 28.7%).
Second, we find that the vast majority of income gains are attributable to indirect market
effects rather than direct increases in NREGS income from the improved program imple-
mentation. For NREGS beneficiaries, increases in program income accounted for only 10%
Since this division took place after our study, we use the term AP to refer to the original undivided state.4Smartcards were also used to make payments for rural social security pensions and reduced leakage here
as well, but these are unlikely to have affected broader labor markets (see Section 2.2).5One natural interpretation is to consider the randomized roll-out of the Smartcards intervention to be
an instrument for an endogenous variable that we may call “effective NREGS.” However, given the manydimensions on which NREGS implementation quality can vary (ease of access to work, availability of workon demand, payment delays and inconvenience, and leakage) it is difficult to construct a single-dimensionalsummary statistic of effective NREGS that can be instrumented for. Our estimates are therefore bestinterpreted as the reduced form impact of improving NREGS implementation on multiple dimensions.
2
of the increases in total income, with the remaining 90% attributable to increases in private
sector earnings. Thus, the general equilibrium impacts of NREGS through the open market
appear to be a much more important driver of its impact on poverty reduction than the
direct income provided by the program.
Third, we find that improving the performance of NREGS led to a significant increase in
private market wages. Market wages for NREGS workers rose by 6.1% in treated areas, with
a similar 5.7% increase in reported reservation wages, suggesting that an improved NREGS
likely improved workers’ bargaining power by improving their outside option. We find no
evidence of corresponding changes in consumer goods prices, implying that the earnings and
wage effects we find are real and not purely nominal gains.
Fourth, we find little evidence of distortionary effects on factor allocation. Despite higher
wages in treated areas, we find a significant 7.1% reduction in the number of days idle or
without paid work, with (insignificant) increases in the number of days of both NREGS and
private sector employment. We find no impacts on migration or on measures of land use.
Fifth, we find evidence that households used the increased income to purchase productive
assets. We find no increase in consumption expenditure, but find an 8.3% increase in the
fraction of households reporting land ownership. Using the 2012 livestock census conducted
by the Govt. of India (also collected independent of our study) we find a significant increase
in the total livestock in treated mandals. Households in treated mandals also had higher
outstanding loans suggesting reduced credit constraints for acquiring productive assets.
Finally, we find evidence that the labor market effects of treatment “spill over” into ge-
ographically proximate markets. We estimate spillovers by exploiting the fact that the
randomization design generated variation in both a mandal’s own treatment status and that
of it’s neighbors. Spillover effects are consistent in sign with the direct effects of treatment,
implying that the latter likely underestimate the “total treatment effect” a mandal would
experience if all mandals were treated. We develop methods to estimate this total effect,
and obtain estimates that are significant and substantially larger than our more conservative,
unadjusted estimates, with multiples ranging from 1.5x to 5x depending on the outcome.
An improved NREGS is likely to affect wages, employment, and income through several
channels including increased labor market competition, direct productivity effects of NREGS
asset creation, indirect productivity effects through improved credit access and purchase of
productive assets, local aggregate demand externalities from more money in the hands of
the poor, and changes in investment behavior in response to better wage insurance. In
addition to taking place simultaneously during our 2-year study period, these channels are
likely to interact with each other, which makes “isolating” any one mechanism an implausible
exercise. Thus, our experiment lets us credibly estimate the composite “general equilbrium”
3
effect of an improvement in effective NREGS on a broad set of outcomes (income, poverty,
wages, employment), but is not designed to isolate specific mechanisms of impact.6
Our first contribution is to the growing literature on the impact of public works pro-
grams on rural labor markets and economies (Imbert and Papp, 2015; Beegle et al., 2015;
Sukhtankar, forthcoming). We present experimentally-identified estimates of improving the
implementation of NREGS with units of randomization large enough to capture general
equilibrium effects and find that doing so led to a significant increase in incomes for the poor
and a reduction in poverty, even without spending additional funds on the program. We
also find that the majority of the impact on income and poverty was due to indirect market
effects of a better implemented NREGS rather than direct increases in NREGS income.
Second, these results highlight the importance of accounting for general equilibrium effects
in program evaluation (Acemoglu, 2010). Ignoring these effects (say by randomizing access to
the program at an individual worker level) would have led to a substantial underestimate of
the impact of improving NREGS implementation on poverty reduction. Even analyzing our
own data while maintaining the assumption of stable unit treatment values (i.e. no spillovers
across mandals) would arguably lead us to substantially understate market impacts. On an
optimistic note, however, our study demonstrates the feasibility of conducting randomized
experiments with units of randomization that are large enough, and specifications nuanced
enough, to capture such general equilibrium effects (Cunha et al., 2013; Muralidharan and
Niehaus, 2016).
Third, our results contribute to the general literature on rural labor markets in developing
countries (Rosenzweig, 1978; Jayachandran, 2006), as well as the more specific literature
on the impacts of minimum wages in developing countries (e.g. Dinkelman and Ranchhod
(2012)). Commentators on NREGS have argued that it could not possibly have led to mean-
ingful impacts on rural wages and poverty because the days worked on NREGS constitute
only a small share (under 4%) of total rural employment (Bhalla, 2013). Our results suggest
that this argument is incomplete, as much larger shares of the workforce have jobcards (38%)
and actively participate (23%) in the program, and for these workers the very existence of
a well-implemented public works programs can raise market wages as it increases their bar-
gaining power over wages by providing a more credible outside option (Dreze and Sen, 1991;
Basu et al., 2009).
Fourth, our results highlight the importance of implementation quality as a first-order
consideration in program effectiveness and in the interpretation of program evaluations (es-
6We discuss the extent to which our data are consistent with several individual mechanisms in Section 4.Overall, we do not find evidence for any individual channel other than labor market competition, but we alsocannot rule out the possibility that each channel did contribute to the overall effects on wages, employment,and income.
4
pecially in developing countries). Strikingly, our estimates of the private market wage im-
pacts of improving NREGS implementation are of a similar magnitude as reported by the
most credible estimates to date of the impact of rolling out NREGS itself (Imbert and Papp,
2015). More generally, programs are not just an “intervention”, but an intervention and an
implementation protocol and investing in improved implementation may often yield greater
improvements in the effective presence of a program than increased fiscal outlays on the
program itself.7
Finally, we contribute to the literature on political economy of development. Jayachandran
(2006) shows that landlords typically benefit at the cost of workers from the wage volatility
induced by productivity shocks and may be hurt by programs like NREGS that provide
wage insurance to the rural poor. Consistent with this, Anderson et al. (2015) have argued,
that “a primary reason... for landlords to control governance is to thwart implementation
of centrally mandated initiatives that would raise wages at the village level.” Our results
showing that improving NREGS implementation substantially raised market wages suggest
that landlords may have been made worse off by the reform, and may partly explain the
widely documented resistance by landlords to NREGS (Khera, 2011).
The rest of the paper is organized as follows. Section 2 describes the context, including
NREGS and prior research, and the Smartcard intervention. Section 3 describes the research
design, data, and estimation. Section 4 presents our main results on income, wages, and
employment, and discusses mechanisms. Section 5 examines spillover effects, while Section
6 concludes.
2 Context and intervention
2.1 National Rural Employment Guarantee Scheme
The NREGS is the world’s largest public employment program, making any household living
in rural India (i.e. 11% of the world’s population) eligible for guaranteed paid employment of
up to 100 days. It is one of the country’s flagship social protection programs, and the Indian
government spends roughly 8% of its budget (∼ 0.75% of GDP) on it. The program has broad
coverage; 65.7% of rural households in Andhra Pradesh have at least one jobcard, which
is the primary enrollment document for the program. Workers can theoretically demand
7For instance Niehaus and Sukhtankar (2013b) show that an increase in the official NREGS wage andcorresponding increase in fiscal outlays had no impact on the actual wages received by workers. This presentsa striking contrast with our results in this paper finding significant increases in wages despite no increase infiscal outlay. In a similar vein, Muralidharan et al. (2014) show that reducing teacher absence by increasingmonitoring would be ten times more cost-effective at reducing effective student-teacher ratios (net of teacherabsence) in Indian public schools than the default policy of hiring more teachers.
5
employment at any time, and the government is obligated to provide it or pay unemployment
benefits (though these are rare in practice).
Work done on the program involves manual labor compensated at statutory piece rates.
The physical nature of the work is meant to induce self-targeting. NREGS projects are
proposed by village governance bodies (Gram Panchayat) and approved by mandal (sub-
district) offices. These projects typically involve public infrastructure improvement such as
irrigation or water conservation works, minor road construction, and clearance of land for
agricultural use.
The NREGS suffers from a number of known implementation issues including rationing,
leakage, and problems with the payment process. Although the program is meant to be
demand driven, rationing is common, and work mainly takes place in the slack labor demand
season (Dutta et al., 2012; Muralidharan et al., 2016). Corruption is also common, with theft
from the labor budget taking the form of over-invoicing the government for work not done or
paying worker less than statutory wage rates for completed work (Niehaus and Sukhtankar,
2013a,b). The payment process itself is slow and unreliable, with the norm being payment
delays of over a month, uncertainty over payment dates, and lost wages as a result of time-
consuming collection procedures (Muralidharan et al., 2016; Pai, 2013).
2.1.1 Potential aggregate impacts of NREGS
In theory, employment guarantee schemes such as the NREGS are expected to affect equilib-
rium in private labor markets (Dreze and Sen, 1991; Murgai and Ravallion, 2005). A truly
guaranteed public-sector job puts upward pressure on private sector wages by improving
workers’ outside options. As Dutta et al. (2012) puts it,
“...by linking the wage rate for such work to the statutory minimum wage rate, and
guaranteeing work at that wage rate, [an employment guarantee] is essentially a
means of enforcing that minimum wage rate on all casual work, including that not
covered by the scheme. Indeed, the existence of such a program can radically alter
the bargaining power of poor men and women in the labor market... by increasing
the reservation wage...”
Depending on the structure of labor markets, increased wages may crowd out private
sector employment, perhaps reducing efficiency.
In addition to this competitive effect, NREGS could affect the rural economy through the
channels of public infrastructure, aggregate demand, and the relaxation of credit constraints.
The public goods that NREGS projects create - such as irrigation canals and roads - could
increase productivity, possibly mitigating negative impacts on efficiency. Given the size
6
of the program, it could also have effects through increased aggregate demand as workers’
disposable income increases, given the presence of agglomeration economies or barriers to
trade (for internal barriers to trade in India see Atkin (2013)). Finally, increased income
may relax credit constraints and thereby increase output.
Given the implementation issues discussed in the previous section, it is unclear whether
any of these effects are actually witnessed in practice. For example, Niehaus and Sukhtankar
(2013b) point out that because of corruption by officials who steal worker’s wages, NREGS
does not serve as an enforcement mechanism for minimum wages in the private sector, but
rather functions as a price-taker.
2.1.2 Prior evidence on NREGS impact
The impact of the NREGS on labor markets, poverty, and the rural economy has been hotly
debated since inception.8 Supporters claim it has transformed the rural countryside by
increasing incomes and wages; creating useful rural infrastructure such as roads and canals;
and reduced distress migration (Khera, 2011). Detractors claim that funding is simply
captured by middlemen and wasted; that it couldn’t possibly affect the rural economy since
it is a small part of rural employment (“how can small tail wag a very very large dog?”); or
that even if it increases rural wages and reduces poverty, that this is at the cost of crowding
out more efficient private employment, in rural areas and cities (Bhalla, 2013). The debate is
still politically salient, as the current national government has been accused of attempting to
let the program slide into irrelevance by defunding it,9 while a group of prominent academics
signed a letter asking it to not do so.
The problem with this heated debate is that the debate-to-evidence ratio is high, as
credibly identifying causal impacts of the program is difficult. There was no evaluation
prior to the program’s launch or built in to program rollout, while the selection of districts
for initial deployment was politicized (Gupta, 2006; Chowdhury, 2014). The vast majority
of empirical work estimating impacts of NREGS thus uses one of two empirical strategies,
both relying on the fact that there was a phase-in period for program implementation: it
was implemented first in a group of 200 districts starting in February 2006, followed by a
second group of 130 in April 2007, while the remaining rural districts entered the program in
April 2008. This allows for a difference-in-differences or regression discontinuity approach,
by comparing districts in Phase I with those in Phase II and III (or those in I and II with
III, etc).
8See Sukhtankar (forthcoming) for a review of the literature on the impacts of NREGS.9See, for example, the following article on the most recent apparent attempt: http://thewire.in/
75795/mnrega-centre-funds-whatsapp/, accessed November 3, 2016.
7
These approaches are non-experimental, and as such rely on strong assumptions in order
to identify causal impacts of NREGS. While this problem is well understood - and some of
the studies may well satisfy the necessary assumptions - there is another less appreciated
problem with these types of comparisons. Given that NREGS suffered from a number of well-
documented “teething problems” - for example, two years after the start major issues with
basic awareness, payment delivery, and monitoring were still to be worked out (Mehrotra,
2008) - any estimated impacts are less likely to be informative about steady state effects.
Possibly the most consistent evidence comes from estimated impacts on labor markets,
with three papers (Imbert and Papp, 2015; Berg et al., 2012; Azam, 2012) using a similar
difference-in-differences approach but different datasets estimating that the NREGS rollout
may have raised rural unskilled wages by as much as 4-5%. Yet these papers disagree
on the timing of the effects; while Imbert and Papp (2015) suggest that wage effects are
concentrated in the slack labor season, Berg et al. (2012) find that they are driven by
peak labor seasons. Meanwhile Zimmermann (2015) finds no average effects on wages using
a regression discontinuity approach. Effects on potential crowd-out are also inconclusive:
Imbert and Papp (2015) find 1.5% decrease in private employment concentrated in the slack
season, while Zimmermann (2015) finds a 3.5% reduction in private employment for men,
year-round.
The estimated effects on other outcomes present an even more conflicting picture. Given
the potential for labor market effects to spill over into schooling, a number of papers have
examined educational outcomes. Mani et al. (2014) find that educational outcomes improved
as a result of NREGS; Shah and Steinberg (2015) find that they worsened; while Islam
and Sivasankaran (2015) find mixed effects. Given that the program was targeted towards
underdeveloped areas suffering from civil violence related to the leftist Naxalite or Maoist
insurgency, some papers have examined effects on violence. While Khanna and Zimmerman
(2014) find that such violence increased, Dasgupta et al. (2015) find the opposite.
While no doubt there is variation across these papers in the samples used, as well as in
the risk of bias, the conflicting results do underscore the difficulty of relying exclusively on
non-experimental analysis. Yet experimental estimates have also been difficult to obtain, in
part since capturing general equilibrium effects would require randomization at large scales.
2.2 Smartcards
In an attempt to address problems with implementation, the Government of Andhra Pradesh
(GoAP) introduced a new payments technology based on electronic transfers of benefits
to beneficiary bank accounts and biometric authentication of beneficiaries prior to benefit
8
withdrawal. This technology - which we collectively refer to as “Smartcards” - had two
major components. First, it changed the last-mile payments provider from the post office to
a private Technology Service Provider / Customer Service Provider, along with the flow of
payments to these providers. Second, it changed the authentication technology from paper
documents and ink stamps to a Smartcard and digital biometric check.
The intervention had two major goals. First, it aimed to reduce leakage from the NREGS
labor budget in the form of under-payment and over-reporting. Second, it targeted improve-
ments in the payment experience, in particular delays in NREGS wage payments. More
details on the functioning of the intervention and the changes that it introduced are avail-
able in Muralidharan et al. (2016); for this paper what is relevant is that the intervention
dramatically improved the implementation of NREGS, which we describe briefly in section
2.3.
Note that the Smartcards intervention affected both NREGS as well as the Social Security
Pension (SSP) program. In this paper, we focus on effects coming from improvements to
NREGS, as it is unlikely that improvements to SSP affected the labor market or the broader
rural economy for two reasons. First, the scale and scope of SSP is fairly narrow: only 7%
of rural households are eligible, as it is restricted to those who are Below the Poverty Line
(BPL) and either widowed, disabled, elderly, or had a (selected) displaced occupation. It is
meant to complement NREGS for those unable to work, and the most prominent benefit level
of Rs. 200 per month is small (about $3, or less than two days earnings for a manual laborer).
Second, the improvements from the introduction of Smartcards were less pronounced than
those in NREGS: there were no significant improvements in the payments process, while
reductions in leakage only amounted to Rs. 12 per household.10
2.3 Effects on program performance
In Muralidharan et al. (2016), we show that Smartcards significantly improved the func-
tioning of NREGS in AP on multiple dimensions. Two years after the intervention began
in treatment mandals, the NREGS payments process got faster (29%), less time-consuming
10Of course, SSP households could be affected by changes in wages and employment, and many SSPhouseholds are also NREGS jobcard holders and work on NREGS. Indeed, prior versions of this paper pooledNREGS and SSP samples for analysis in order to take advantage of additional sample size. While resultswith pooled samples are almost exactly the same as the results that we present here, pooling the samplescreates a complicated weighting issue. Samples for NREGS and SSP surveys were drawn independently,with the NREGS sample designed to be representative of all NREGS jobcard-holders while over-weightingrecent workers, while the SSP sample was drawn to be representative of all SSP beneficiaries. It is not clearwhat the pooled sample is representative of. Since NREGS jobcard holders represent 65.7% of the ruralpopulation, while SSP beneficiaries who are not NREGS jobcard holders comprise only 2.2% of the ruralpopulation, it is easiest to simply ignore the SSP sample rather than present all our results with robustnesschecks for multiple weighting schemes. These results are available on request.
9
(20%), and more predictable (39%). Additionally, households earned more through working
on NREGS (24%), and there was a substantial 12.7 percentage point reduction (∼ 41%) in
leakage. Moreover, access did not suffer: both perceived access and actual participation in
the program increased (17%). We find little evidence of heterogenous impacts, as treatment
distributions first order stochastically dominate control distributions for all outcomes on
which there was a significant mean impact. Reflecting this, users were strongly in favor of
Smartcards, with 90% of households preferring it to the status quo, and only 3% opposed.
The improvements in implementation reflect intent-to-treat (ITT) estimates, which is
important since implementation was far from complete. Logistical problems were to be
expected in an intervention of this scale, and two years after implementation the proportion
of payments in treated areas made using Smartcards had plateaud to 50%. It is important to
note that these estimates do not reflect “teething” problems of Smartcards, since Smartcards
had been implemented in other districts in AP for four years prior to their introduction in
our study districts. The estimates reflect steady state, medium run impacts that are net of
management, political economy, and other challenges.
A result that is important to the interpretation of general equilibrium effects of Smartcards
is that there was no increase in NREGS expenditure by the government. Thus unlike the
introduction of NREGS itself, no new money flowed into treatment areas. Any increases in
earnings were due to a reduction in leakage corresponding to a redistribution from corrupt
officials to workers. This is the main significant difference if one wishes to compare the effect
of Smartcards to an idealized effect of “NREGS itself,” although one could potentially use
the mean level of program earnings in the control group as an indicator of what the effect
of the program itself may have been on this dimension.
Other mechanisms via which the reform affected rural economies are similar to those
that one might expect from the introduction of a well-implemented NREGS. First, the
improvement in payments logistics such as timeliness of payments and the increase in earnings
on NREGS made the program a more viable outside option to private sector employment,
and thus led to an increase in competitive pressure in the labor market. Since there was
additional participation in NREGS, there was also an increase in the amount of rural work
done and rural public goods created. The increases in NREGS earning could also have
relaxed credit constraints on participants.
10
3 Research design
3.1 Randomization
We summarize the randomization design here, and refer the reader to Muralidharan et
al. (2016) for further details. The experiment was conducted in eight districts11 with a
combined rural population of around 19 million in the erstwhile state of Andhra Pradesh
(now split into two states: Andhra Pradesh and Telangana). As part of a Memorandum of
Understanding with JPAL-South Asia, GoAP agreed to randomize the order in which the
Smartcard system was rolled out across mandals (sub-districts). We randomly assigned 296
mandals - with average population of approximately 62,500 - to treatment (112), control (45),
and a “buffer” group (139); Figure 1 shows a map showing the geographical spread and size
of these units. We created the buffer group to ensure that we could conduct endline surveys
before deployment began in control mandals, and restricted survey work to treatment and
control mandals. We stratified randomization by district and by a principal component of
mandal socio-economic characteristics.
We examine balance in Tables A.1 and A.2. The former (reproduced from Muralidharan et
al. (2016)) simply shows balance on variables used as part of stratification, as well as broader
mandal characteristics from the census. Treatment and control mandals are well balanced,
with two out of 22 variables significant. The latter shows balance on the outcomes that are
our primary interest in this paper, as well as key socio-economic household characteristics
from our baseline survey (see below). Here, four out of 34 variables are significantly different
at the 10% level at least, which is slightly more than one might expect. Where feasible, we
also test for sensitivity of the results to chance imbalances by controlling for village level
baseline mean values of the outcomes.
3.2 Data
3.2.1 Socio-Economic and Caste Census
Our primary data source is the Socio-Economic and Caste Census (SECC), an independent
nation-wide census for which surveys in Andhra Pradesh were conducted during 2012, our
endline year. The primary goal of the SECC was to enable governments to rank households
by socio-economic status in order to determine which were “Below the Poverty Line” (BPL)
11The 8 study districts are similar to AP’s remaining 13 non-urban districts on major socioeconomicindicators, including proportion rural, scheduled caste, literate, and agricultural laborers; and represent allthree historically distinct socio-cultural regions. See the online appendix to Muralidharan et al. (2016) fordetails.
11
and thereby eligible for various benefits. A secondary (and controversial) goal was to capture
data on caste, which the regular decennial census does not collect. The survey collected data
on income categories for the household member with the highest income (less than Rs. 5000,
between Rs. 5000-10,000, and greater than Rs. 10,000), the main source of this income,
household landholdings (including amount of irrigated and non-irrigated land) and caste,
and the highest education level completed for each member of the household.
The SECC was conducted using the layout maps and lists of houses prepared during the
conduct of the 2011 Census. Enumerators were assigned to cover the households listed in
each block, and were also instructed to attempt to interview homeless populations. The
total number of households in our SECC sample, including treatment and control mandals,
is slightly more than 1.8 million.
3.2.2 Original survey data
We complement the broad coverage of the SECC data with original and much more detailed
surveys of a smaller sample of households. Specifically, we conducted surveys of a represen-
tative sample of NREGS jobcard holders and SSP beneficiaries during August to October of
2010 (baseline) and 2012 (endline). Surveys covered both respondents’ participation in and
experience with these programs, and also their earnings, expenditure, assets and liabilities
more generally. Within earnings, we asked detailed questions about household members’ la-
bor market participation, wages, reservation wages, and earnings during the month of June
(the period of peak NREGS participation in Andhra Pradesh).
Full details of the sampling procedure used are in Muralidharan et al. (2016). In brief,
we drew a sample of NREGS jobcard holders that over-weighted those who had recently
participated in the program according to official records. In Andhra Pradesh, 65.7% of rural
households have a jobcard according to our calculations based on the National Sample Survey
Round 68 in 2011-12. We sampled a panel of villages and repeated cross-sections of the full
concurrent NREGS sampling frame. The sample included 880 villages, with 6 households
in each village. This yielded us 5,278 households at endline, of which we have survey data
on 4,943 households; of the remaining, 200 were ghost households, while we were unable to
survey or confirm existence of 135 (corresponding numbers for baseline are 5,244, 4,646, 68
and 530 respectively).
3.2.3 District Statistical Handbook data
We use District Statistical Handbooks (DSH) published by the Andhra Pradesh Directorate
of Economics and Statistics, a branch of the Central Ministry of Agriculture and Farmers
12
Welfare, to obtain additional data on land use and irrigation – including details by season –
and on employment in industry. DSH are published every year and are not to be confused
with the District Census Handbooks which contains district tables from the Census of India.
Land coverage data presented in the DSH is officially provided by the Office of the Surveyor
General of India. Ideally, land coverage data is obtained from so-called village papers pre-
pared by village accountants. These village papers contain information on land cover that
varies (area sown or fallow) while forest and mountainous areas is recorded centrally. For
cases in which no village papers are maintained, “ad-hoc estimates of classification of area
are derived to complete the coverage.”12
3.2.4 National Sample Survey data
We use unit cost data from Round 68 (2011-2012) of the National Sample Survey (NSS)
published by the Ministry of Statistics and Programme Implementation. The NSS contains
detailed household × item-level data on a sample that is representative at the state and
sector level (rural and urban). The goal of the 68th Round was to obtain estimates on
key household consumer expenditure, employment, and unemployment indicators for States
and UTs. The data provides a comprehensive picture of household-level consumption and
expenditure for over 300 goods and services in categories including food, fuel, clothing, rent
and other fees or services over mixed reference periods varying from a week to a year.
3.2.5 Census data
We use spatial data from the 2001 Indian Census which contains comprehensive geographic
information at the village-level. For each census village, the data contains both a point and
a polygon of geographic coordinates (latitude, longitude) to define the precise location and
borders of the village.
3.3 Estimation strategy
We begin by reporting simple comparisons of outcomes in treatment and control mandals
(i.e. intent-to-treat estimates). Our base regression specification includes district fixed
effects and the first principal component of a vector of mandal characteristics used to stratify
randomization (PCmd), with standard errors clustered at the mandal level:
Yimd = α + βTreatedmd + δDistrictd + λPCmd + εimd (1)
12Information from http://eands.dacnet.nic.in/, accessed March 22, 2016.
13
where Yimd is an outcome for household or individual i in mandal m and district d, and
Treatedmd is an indicator for a treatment group mandal. In some cases we use non-linear
analogues to this model to handle categorical data (e.g. probit, ordered probit). When using
our survey data, we also report specifications that include the baseline GP-level mean of the
dependent variable, Y0
pmd, when available in order to increase precision and assess sensitivity
to any randomization imbalances:
Yipmd = α + βTreatedmd + γY0
pmd + δDistrictd + λPCmd + εipmd (2)
where p indexes panchayats or GPs. Note that we easily reject γ = 1 in all cases and
therefore do not report difference-in-differences estimates.
If outcomes for a given unit (household, GP, etc.) depend only on that unit’s own treat-
ment status, then Equation 1 identifies a well-defined treatment effect. If Smartcards do have
general equilibrium effects, however, then these effects need not be confined to the treated
units. Upward pressure on wages in treated mandals, for example, might affect wages in
nearby areas of control mandals. In the presence of such spillovers, (1) underestimates the
“total” effect of treatment, conceptualized as the difference between average outcomes when
all units are treated and that when no units are treated. Estimating this “total” effect
is complex enough that we defer it to Section 5, and simply note for now that our initial
estimates are likely to be conservative.
Regressions using SECC data are unweighted. Regressions with survey samples are weighted
by inverse sampling probabilities to be representative of the universe of jobcard-holders.
When using survey data on financial outcomes we trim the top 0.5% of observations in both
treatment and control groups to remove outliers, but our results are robust to including
them.
4 Results
4.1 Effects on earnings and poverty
Figure 2 compares the distributions of SECC income categories in treatment and control
mandals. Note that these are raw data without district fixed effects conditioned out, and
so could theoretically reflect cross-district differences in treatment probability as well as
treatment effects. With that caveat in mind, the treatment distribution evidently first-order
stochastically dominates the control, with 4.0 percentage points fewer households in the
lowest category (less than Rs. 5,000), 2.7 percentage points more households in the middle
category (Rs. 5,000 to 10,000), and 1.4 percentage points more in the highest category
14
(greater than Rs. 10,000). Overall, these estimates imply that Smartcards moved 53,626
households out of the lowest income category and into a higher one.
Table 1a reports statistical tests, conditioning on district effects to obtain experimental
estimates of impact. Given the categorical nature of our outcome measure, we report logistic
regressions for each category individually (showing marginal effects) and also an ordered
logistic regression for the combined categories. We find that treatment significantly increased
the log-odds ratio of being in a higher income category, with results strongly significant at
the 1% level. As a sanity check we also confirm that these estimates are unaltered when
we control for demographic characteristics (age of household head, caste, literacy). This
suggests that the Smartcards randomization was indeed orthogonal to other determinants of
earnings.
The SECC income measures let us test for income effects in the entire population of
interest, but have two limitations when it comes to estimating magnitudes. First, much
information is lost through discretization: the 4.0% reduction in the share of households in
the lowest category which we observe, for example, could reflect a small earnings increase for
households that were just below the Rs. 5,000 threshold, or a large impact for households
that were further away from it. Second, because the SECC only captures the earnings of
the top income earner in each household, it is possible that it over- or under-states effects
on overall household earnings.
For a better sense of magnitudes we therefore turn to our survey data. Columns 1 and 2 of
Table 1b reports estimated impacts on overall annual household income, with and without
controls for the mean income in the same village at baseline. In both specifications we
estimate that that treatment increased income by over Rs. 9500. This is a large effect,
equal to 13.3% of the control group mean or 19.6% of the national expenditure-based rural
poverty line for a family of 5 in 2011-12, which was Rs. 48,960 (Government of India, 2013).
Of course, expenditure- and income-based poverty lines may differ and this comparison is
illustrative only. But if these lines were taken as equivalent, we see a 5.2 percentage point
or 18.1% reduction in poverty: this is evident in Figure 3 which plots the empirical CDF of
household earnings for treatment and control groups.
Our survey data are also useful for examining distributional impacts. Estimated earnings
impacts are weakly positive throughout the distribution, but visibly larger at the higher end,
with 55% of the total earnings increase accruing to households above the 80th percentile
(corresponding to annual household income of Rs. 103,000). Of course, a household earning
this much remains poor by any absolute standard; this number is twice the expenditure-based
poverty line, but corresponds to less than $4/day per capita (in PPP exchange rates).
15
4.2 Direct versus indirect effects on earnings
Mechanically, the effects on earnings and poverty we find above must work through some
combination of increases in households’ earnings from the NREGS program itself and in-
creases in their non-program (i.e. private sector) earnings. To examine this decomposition
we use our survey data, which includes measures of six income categories: NREGS, agricul-
tural labor income, other physical labor income, income from own farm, income from own
business, and miscellaneous income (which includes all remaining sources, including salaried
income). In the control group, the average household earns roughly 1/3 of its income from
wage labor, primarily in agriculture; 1/3 from self-employment activities, also primarily in
agriculture; and the remaining 1/3 from salaried employment and public programs, with the
latter making up a relatively small share.
Columns 3-9 of Table 1b report treatment effects on various income categories separately.
Strikingly, effects on NREGS earnings are only a small proportion of overall income gains,
accounting for only 10.4% of the overall increase.13 Instead the primary driver of increased
earnings is an increase in paid labor, both in in the agricultural and non-agricultural sectors.
Effects on own farm earnings are positive but insignificant.
The distribution of impacts is also consistent with private-sector earnings being the main
driver. Table A.5 examines how treatment effects vary with a household’s ability to perform
manual labor, proxied by the fraction of the household that is eligible for the SSP program
(which is reserved for the elderly, widows, and disabled people). Impacts on earnings – and
especially labor earnings – are concentrated in households not headed by widows and with
few members eligible for SSP.
4.3 Effects on private labor markets
We next unpack impacts on private sector earnings, examining how wages and labor quan-
tities were affected.
To examine wage effects we use our survey data, as the SECC does not include wage
information. We define the dependent variable as the average daily wage rate earned on
private-sector work reported by respondents who did any private-sector work. We report
results for the full sample of workers and also check that results are robust to restricting the
sample to adults aged 18-65. (We report additional robustness checks in Section 4.6 below).
13The observant reader may note that the (marginally) insignificant effect on program earnings hereappears to contrast with the significant effect in Muralidharan et al. (2016). The two estimates are for twodistinct measures of NREGS earnings: the measure in Muralidharan et al. (2016) comes from questions abouta specific 6-week study period shortly before surveys were conducted and asked to the worker themselves,while the measure here is from an aggregate, annual recall question posed to the head of the household, andis thus much less precise. The point estimates are nevertheless economically similar.
16
We estimate a significant increase of Rs. 7.9 in private sector wages (Table 2, Column 2).
This is a large effect, equal to 6.2% of the control group mean. In fact, it is slightly larger
than the highest estimates of the wage impacts of the rollout of the NREGS itself reported
by Imbert and Papp (2015).
We then examine how labor market participation was affected by this large wage increase.
We classify days spent during the month of June by adults (ages 18-65) into three categories:
time spent working on the NREGS, time spent working on the private sector, including self-
employment, and time spent idle / on leisure.
We find a significant decrease of 1.2 days per month in days spent idle, equal to 7% of the
control group mean (Table 3, Columns 5 & 6). This time appears to have been reallocated
across both NREGS work and private sector work, which increase by roughly 0.9 days and
0.5 days per month, respectively (though these changes are not individually significant)
(Columns 1-4). In Figure 4 we plot the full distribution of private sector days worked for
treatment and control mandals separately; this shows that the increase in private-sector work
is spread fairly evenly through the distribution. Moreover, it is not the case that private
sector was too low to drop further; 51% of our sample reported doing at least some private
sector work in June (50% in control and 52% in treatment), and there is no evidence of a
reduction in the right tail of the distribution.14
The fact that private sector employment does not fall with higher wages is intriguing,
suggesting some degree of monopsony power in labor markets. A long tradition of research
on rural labor markets has argued that large land-owners hold substantial market power. Our
estimate is imprecise, however, and a 95% confidence interval for the wage elasticity of labor
demand is (−0.44, 0.8) which includes negative values, consistent with perfect competition.
In particular, it includes the estimate of −0.31 reported by Imbert and Papp (2015). In a
statistical sense we thus rule out only competitive markets with a wage elasticity of labor
demand larger than (minus) 0.44.
Given that NREGS peaks in the late spring and early summer, one natural question is
whether the wage effects we observe in June are seasonal, or persist throughout the year.
While our household survey data on individual labor spells cover only the month of June,
we also have data from village leaders on “going rates” throughout the year. With one
observation per village-month power is limited, but these data do suggest persistent wage
differences between treatment and control mandals (Figure A.1). These could imply that
14The pattern of impacts we find on labor supply is at least potentially consistent with results in Imbertand Papp (2015). They estimate a 1-for-1 reduction in “private sector employment” as NREGS employmentincreases, but their measure of private sector employment (based on NSS data) does not distinguish domesticwork and self-employment from wage employment for others. Of course, they also study impacts on a differentpopulation.
17
the NREGS still provides an meaningful outside option even during peak labor demand
season, or alternatively could reflect nominal wage rigidity Kaur (2015) and/or labor tying
(Bardhan, 1983; Mukherjee and Ray, 1995).15
Finally, we examine impacts on labor allocation through migration. Our survey asked
two questions about migration for each family member: whether or not they spent any days
working outside of the village in the last year, and if so how many such days. Table 4
reports effects on each measure. We estimate a small and statistically insignificant increase
in migration on both the extensive and intensive margins, inconsistent with the prevailing
view that the NREGS reduces migration to cities. As our migration questions may fail to
capture permanent migration, we also examine impacts on household size and again find no
significant difference.16
4.4 Effects on consumer goods prices
One potential caveat to the earnings results above is that they show impacts on nominal,
and not real, earnings. Given that Smartcards affected local factor (i.e. labor) prices, they
could also have affected the prices of local final goods, and thus the overall price level facing
consumers, if the markets we study are not sufficiently well-integrated into larger product
markets.
To test for impacts on consumer goods prices we use data from the 68th round of the Na-
tional Sample Survey. The survey collected data expenditure and number of units purchased
for a wide variety of goods; we define unit costs as the ratio of these thing. We restrict the
analysis to goods that have precise measures of unit quantities (e.g. kilogram or liter) and
drop goods that likely vary a great deal in quality (e.g. clothes and shoes). We then test
for price impacts in two ways. First, we define a price index Pvd equal to the the price of
purchasing the mean bundle of goods in the control group, evaluated at local village prices,
following Deaton and Tarozzi (2000):
Pvd =n∑
c=1
qcdpcv (3)
Here qcd is the estimated average number of units of commodity c in panchayats in control
areas of district d, and pcv is the median unit cost of commodity c in village v. Conceptually,
15Note that given this persistence, our point estimates for wage, labor and earnings effects are internallyconsistent: the 11.3% increase in private sector earnings is almost exactly equal to the sum of the 6.2%increase in wages and (insignificant) 5% increase in employment.
16In estimating poverty and labor market effects we also examined interactions between treatment andvarious individual and village characteristics. Generally speaking we find few significant predictors of het-erogeneity, and therefore omit these results (available on request).
18
treatment effects on this quantity can be thought of as analogous to the “compensating
variation” that would be necessary to enable households to continue purchasing their old
bundle of goods at the (potentially) new prices.
The set of goods for which non-zero quantities is purchased varies widely across house-
holds and, to a lesser extent, across villages. To ensure that we are picking up effects on
prices (rather than compositional effects on the basket of goods purchased), we initially re-
strict attention to goods purchased at least once in every village in our sample. The major
drawback of this approach is that it excludes roughly 40% of the expenditure in our sample.
We therefore also present a complementary set of results in which we calculate (3) using
all available data. In addition, we report results using (the log of) unit cost defined at
the household-commodity level as the dependent variable and including all available data.
While these later specifications potentially blur together effects on prices with effects on the
composition of expenditure, they do not drop any information.
Regardless of method, we find little evidence of impacts on price levels (Table 5). The
point estimates are small and insignificant, and when we use the full information available
are also precise enough to rule out effects as large as those we found earlier for wages. These
results suggest that the treated areas are sufficiently well-integrated into product markets
that higher local wages did not affect prices, and can thus be interpreted as real gains for
workers (and real losses for employers).17
4.5 Channels of impact
Our experiment was not designed to tease apart the mechanisms through which improving
the NREGS might affect local markets and earnings. Nonetheless, in this section we assess
what the data can tell us about those mechanisms.
We begin with the hypothesis that a (better-run) employment guarantee puts competitive
pressure on labor markets, driving up wages and earnings. Theoretical models emphasize
this mechanism (Ravallion, 1987; Basu et al., 2009), and it has motivated earlier work on
NREGS wage impacts (e.g. Imbert and Papp (2015)). A key differentiating implication of
this hypothesis is that wages rise not because labor demand increases (as would be the case
if labor become more productive, or aggregate demand rose) but rather because workers
demand higher wages (i.e. the labor demand curve shifts up).
We are able to test this prediction using information on reservation wages that we elicited
in our survey. Specifically, we asked respondents if in the month of June they would have
17We also find little evidence of impacts on household expenditure (Table 7), which one would expect ifeither (i) households interpreted their earnings increase as permanent, or (ii) local prices increased. Thatsaid, confidence intervals around these estimates are wide.
19
been “willing to work for someone else for a daily wage of Rs. X,” where X started at Rs.
20 (15% of average wage) and increased in Rs. 5 increments until the respondent agreed.
Among respondents who worked, 98% reported reservation wages below or equal to the wages
they actually earned, suggesting that they correctly understood the question (Table A.3).
We find that treatment did in fact significantly increase workers’ reservation wages, by
approximately Rs. 5.5, or 5.7% of the control group mean (Table 2, Columns 5-8). This
implies that having a better alternative to wage labor must be at least part of the explanation
for higher private-sector earnings. It does not necessarily imply that the better alternative is
NREGS employment, of course, but (as we discuss below) we do not see evidence of higher
returns to other activities we measure, including self-employment in either farm or non-farm
activities. Moreover, the fact that estimated effects on reservation wages and actual wages
are nearly identical suggests that the labor market competition effect is strong enough to
explain the entire wage effect.18
The most obvious alternative hypothesis is that an improved NREGS created additional
assets – roads, irrigation facilities, soil conservation structures, etc. – which increased pro-
ductivity. We find no evidence of such an effect, however: the number and composition of
projects implemented did not change in treatment areas relative to control (Table 6).
Another a priori plausible explanation is that the reform increased cash flow into treated
areas, stimulating local economic activity and driving up wages and earnings (Krugman,
1991). In practice, however, we find no effect of the intervention on the amount of money
disbursed by the NREGS in administrative data (see Muralidharan et al. (2016)), and no
significant increase in any measure of household expenditure in our survey data (Table 7). We
do see some evidence of improved credit access, with higher borrowing in treated areas (Table
8), but as we will see below cannot connect this with evidence of higher factor utilization or
earnings.
Among various other potential mechanisms – including productivity shocks, aggregate
demand shocks, and expansionary credit access, among others – one common differentiating
feature is that each should increase the utilization of some factor(s) of production in addition
to labor. For example, if an improved NREGS leads to the creation of roads that raise revenue
productivity in a village, then it would directly increase demand for both labor and land,
and might thus increase cultivation of marginal land. We therefore look to see whether the
reform affected utilization of any factors other than labor.
We find little impact for the factors we are able to observe. For land, Table 9 shows no
significant effect on the amount of land under cultivation (% area sown or % area fallow) or
18With the caveat that economically meaningful differences remain statistically possible; a 90% confidenceinterval for this difference in the combined sample is Rs. (-2.78, 7.44).
20
on the total area irrigated.19 The implied confidence intervals let us rule out effects larger
than 4 percentage points in all cases.
A similar argument can be made regarding patterns of earnings: if the reform directly
increased revenue productivity then we might expect to see impacts on both employment and
own-account earnings. For example, better market access would increase the profitability
of both large plantations and small owner-farmed plots; better credit access would allow
beneficiaries to acquire more capital and increase their own-account earnings. Yet we see
no significant impacts on earnings from self-employment, with gains concentrated in labor
income categories (Table 1b).
Overall, we find strong evidence consistent with the labor market competition hypothesis,
and little support for alternative interpretations. Yet each data point has its own caveats,
and so we do not view them as collectively dispositive. We leave it to the judicious reader
to decide his or her own degree of certainty.
4.6 Robustness & other concerns
In this section we describe the main robustness checks of our results on income and wages.
The estimated income effects in Table 1b are robust to a number of checks. Since the SECC
data are categorical, we have used logit and ordered logit models for estimation. The results
are robust to using probits or linear probability models instead (results available on request).
Our survey data on income contains possible outliers. As a robustness check, we exclude
observations at the top 0.5% in treatment and control. This does not change the results:
the estimates are slightly smaller but remain significant at the 1% level (Table A.4).
Our wage results are also robust to alternative choices of sample. We top-censor wages
to account for outliers. Although noisier, results including these observations are similar
(Table A.6a). The main results include data on anyone in the household who reports wages.
Restricting the sample to only those of working age (18-65) again does not qualitatively
affect results (Table A.6b). Next, dropping the small number of observations who report
wages but zero actual employment again does not matter (Table A.6c).
Given that we observe wage realizations only for those who work, a potential concern is
that the effects we estimate are driven by changes in who reports work (or wages) and not
by changes in the distribution of wage offers in the market. We test for such selection effects
as follows. First, we confirm that essentially all respondents (99%) who reported working
also reported the wages they earned, and that non-response is the same across treatment
19In earlier drafts we presented evidence that the intervention increased vegetative cover in the month ofMay as measured based on satellite imagery using the Enhanced Vegetation Index (EVI). When we examinedyear-round impacts on EVI, however, we do not find a robust pattern. Results available upon request.
21
and control. (First row of Table A.3). Second, we check that the probability of reporting
any work is not significantly different between treatment and control groups ( Table A.3).
Third, we check composition and find that treatment did not affect composition of those
reporting in Table A.7. Finally, as we saw above treatment also increased reservation wages,
which we observe for nearly the entire sample (89%) of working-age adults.
5 Spatial spillovers
Given results thus far, it seems plausible that implementing Smartcards in one mandal may
have affected outcomes in other neighboring mandals – through labor market competition,
for example. We turn now to testing for such effects and to estimating “total” treatment
effects that account for them.
As with any such spatial problem, outcomes in each GP could in principle be an arbitrary
function of the treatment status of all the other GPs, and no feasible experiment could
identify these functions nonparametrically. We therefore model spillovers as a simple function
of the fraction of neighboring GPs treated within various radii R. Specifically, we define NRp
the fraction of GPs within a radius R of panchayat p which were assigned to treatment.
Figure 5 illustrates the construction of this measure.
One might hope that the random assignment of mandals to treatment and control arms
ensures that the neighborhood measure NRp is also “as good as” randomly assigned – but
this is not the case. To see this, consider constructing the measure for GPs within a treated
mandal: on average, GPs closer to the center of mandal will have higher values of NRp
(because more of their neighbors are from the same mandal), while those closer to the border
will have lower values (because more of their neighbors are from neighboring mandals). The
opposite pattern will hold in control mandals. Thus, we cannot interpret a coefficient on NRp
as solely a measure of spillover effects unless we are willing to make the (strong) assumption
that the direct effects of treatment are unrelated to location.
To address this issue, we construct a second measure NRp defined as the fraction of treated
GPs within a given radius R and within a different mandal. This has the advantage of
being exogenous conditional on own treatment status (with the one disadvantage that it
is not defined for all GPs at smaller values of R). We use this measure both to test for
the existence of spillovers (where we are interested in testing rather than point estimation)
and as an instrument for estimating the effects of NRp , which we view as the “structural”
parameter.
We construct our measures of neighborhood treatment intensity at radii of 10, 15, 20, 25,
and 30 kilometers. Figure A.2 plots smoothed kernel density estimates by treatment status
22
for several of these measures to illustrate that there is meaningful overlap in the distributions
for control and treatment group. Tables A.8 and A.9 report tests showing that our outcomes
of interest are balanced with respect to these measures at baseline.
5.1 Testing for spillovers
To test for the existence of spillovers, we estimate
Yipmd = α + βNRp + δDistrictd + λPCmd + εimd (4)
separately for the treatment and control groups. This approach allows for the possibility
that neighborhood effects differ depending on one’s own treatment status. We also estimate
a variant that pools both treatment and control groups (and adds an indicator for own
treatment status), which imposes equality of the slope coefficient β across groups. In either
case we interpret rejection of the null β = 0 as evidence of spillover effects.
We find fairly consistent evidence of spillover effects on market wages, consistent in sign
with the direct effects we estimated above (Table 10, Columns 1-5). The effects are strongest
for households in control mandals, where we estimate a significant relationship at all radii
greater than 10km; for those in treatment mandals the estimates are smaller and significant
for three out of five radii, but uniformly positive. For days spent on unpaid work or idle,
the estimated effects are all negative and economically meaningful, as above, and significant
except at smaller radii when we split the sample (Table 10, Columns 6-10). In particular,
when we pool the samples we estimates significant spillover effects at all radii (except at R
= 10 for days spent on unpaid work or idle).
Note that sample sizes increase as we increase the neighborhood radius R, since at larger
values of R a larger share of panchayats have at least one neighbor within distance R and
in another mandal; we use between 90% and 100% of the available data depending on
specification. Effect sizes and t-statistics are for the most part larger, however, for larger
values of R, suggesting that excluding sample is not biasing us towards rejecting the null.20
5.2 Estimating total treatment effects
To estimate the total effect of treatment, we wish to estimate
Yipmd = α + βTTm + βNNRp + βTNTm ·NR
p + δDistrictd + λPCmd + εimd (5)
20Alternatively, we can construct tests using 100% of the data if we are willing to make the (strong)assumption that NR
p is exogenous. We obtain directionally similar results when doing so (Table A.10), butthe estimates are less precise.
23
In the context of this model, we define the total effect of treatment as
β = βT + βN + βTN (6)
Intuitively, this is the difference in expected outcomes between a unit in a treated mandal
with 100% of its neighborhood treated, and that for a unit in a control mandal with 0% of
its neighborhood treated. Notice that this is an partially extrapolative exercise, as much of
our sample is not in either of these conditions. Nevertheless, we are interested in estimating
it since it is this “total” effect one would ideally use for determining policy.
Since NRp is potentially endogenous, we instrument for it in (5) using NR
p (and for it’s
interaction with t using the corresponding interaction). Not surprisingly, we estimate a strong
first-stage relationship between the two, with a minimum F -statistic across specifications
reported here of F = 115. For completeness we also report OLS estimates of (5) in Table
A.11; these are qualitatively similar to the results we present here, and mostly significant,
but less precise.
After adjusting for spillovers, we estimate effects on all of our main outcomes – wages,
labor, and earnings – which are (i) significant for most or all values of R, (ii) consistent in sign
with those we estimate in simple binary treatment specifications, and (iii) meaningful larger.
Table 11 summarizes these results; the estimate of interest is the total treatment effect. We
begin in panel (a) with wage outcomes. Depending on choice of R, the estimated effect on
realized wages is 14-20% of the control mean, and uniformly significant. The estimated effect
on reservation wages is 8-10% of the control mean, significant except for large values of R.
All told, these estimates suggest large general equilibrium effects of Smartcards on labor
markets.
In panel (b) we examine impacts on labor allocation. As above, we estimate that treat-
ment significantly increased days of work done in the private sector (for all R > 10km), and
significantly reduced the number of days spend idle or doing unpaid work. As one would
expect, the effects are again larger than more conservative estimates based on binary T/C
classification. Finally, in panel (c) we document the same pattern for total earnings: the esti-
mated total treatment is significant (except at R = 30km) and larger than the conservative,
unadjusted estimate.
5.3 Implications
Evidence on the extent of spatial spillovers helps to characterize the degree of spatial inte-
gration in rural labor markets in India. This is useful for interpreting past work on these
markets, as much of this research has been constrained by data availability to treat districts
24
as if they were separate labor markets (e.g. Jayachandran (2006), Imbert and Papp (2015),
Kaur (2015)). While this is understood to be an approximation, little is known about how
reasonable it is. Our data seem reasonably consistent with it: the average rural district in
AP had an area of approximately 10,000 square kilometers (equivalent to an idealized 100km
× 100km square), while we find evidence of spillovers at radii of up to 20-30km.21 If any-
thing, we suspect that treating districts as unitary markets under-estimates the importance
of within-district market segmentation.
6 Conclusion
This paper examines the impact of a major reform to a large public works program - the
National Rural Employment Guarantee Scheme - in India. Such large programs often have
general equilibrium impacts, which are difficult to capture non-experimentally or through
experiments where the scale of the randomized unit is small. We take advantage of an un-
usually large-scale intervention that introduced biometric “Smartcards” to make payments
to beneficiaries of the NREGS. In previous work we find that Smartcards significantly im-
proved the implementation of NREGS. Here we examine the corresponding effects of this
improvement on beneficiaries livelihoods and rural labor markets.
We find large increases in income, using not only our representative survey data but also
an independent and concurrent census conducted by the government. We also find that
the indirect effects of the reform are an order of magnitude larger than the direct effect on
NREGS earnings. These indirect effects appear to be driven by effects on private sector
labor markets, particularly increased wages. They also “spill over” into neighboring villages,
further increasing wages and earnings. Finally, we do not find evidence of factor allocation
distortions related to NREGS.
While we estimate the effects of improving NREGS implementation, one might also won-
der how our estimates compare to those from a hypothetical comparison between a “well-
implemented NREGS” and “no NREGS.” Our conjecture is that the effects would be broadly
comparable, but with larger income effects. The Smartcards reform increased the labor-
market appeal of the NREGS and increased participation in its projects, but did not in-
crease the flow of funds into treated areas. In contrast, the NREGS per se clearly represents
a significant transfer of funds from urban to rural areas.
21For context, note that 20km is a 4 hour walk at the average human walking speed of 5km/hour, androughly 7 times the width of an average village (2.9km). NREGS rules, meanwhile, stipulate that employmentshould be provided within 5km of the beneficiary’s home. While individual workers need not travel 20km forthese spillover effects to be observed at that distance, it is in fact plausible that they might do so, as mostuse bicycles for transport. Nearly 80% of households in our sample own at least one bicycle.
25
References
Acemoglu, Daron, “Theory, General Equilibrium, and Political Economy in DevelopmentEconomics,” Journal of Economic Perspectives, 2010, 24 (3), 17–32.
Anderson, Siwan, Patrick Francois, and Ashok Kotwal, “Clientilism in Indian Vil-lages,” American Economic Review, 2015, 105 (6), 1780–1816.
Atkin, David, “Trade, Tastes, and Nutrition in India,” The American Economic Review,2013, 103 (5).
Azam, Mehtabul, “The Impact of Indian Job Guarantee Scheme on Labor Market Out-comes: Evidence from a Natural Experiment,” Working Paper 6548, IZA 2012.
Bardhan, Pranab K., “Labor-Tying in a Poor Agrarian Economy: A Theoretical andEmpirical Analysis,” The Quarterly Journal of Economics, 1983, 98 (3), 501–514.
Basu, Arnab K., Nancy H. Chau, and Ravi Kanbur, “A Theory of EmploymentGuarantees: Contestability, Credibility and Distributional Concerns,” Journal of PublicEconomics, April 2009, 93 (3-4), 482–497.
Beegle, Kathleen, Emanuela Galasso, and Jessica Goldberg, “Direct and IndirectEffects of Malawi’s Public Works Program on Food Security,” Technical Report, Universityof Maryland 2015.
Berg, Erlend, Sambit Bhattacharyya, Rajasekhar Durgam, and Manjula Ra-machandra, “Can Rural Public Works Affect Agricultural Wages? Evidence from In-dia,” CSAE Working Paper Series 2012-05, Centre for the Study of African Economies,University of Oxford 2012.
Bhalla, Surjit, “The Unimportance of NREGA,” The Indian Express, July 24 2013.
Chowdhury, Anirvan, “Poverty Alleviation or Political Calculation? Implementing IndiaısRural Employment Guarantee Scheme,” Technical Report, Georgetown University 2014.
Cunha, Jesse, Giacomo DeGiorgi, and Seema Jayachandran, “The Price Effects ofCash Versus In-Kind Transfers,” Technical Report, Northwestern University 2013.
Dasgupta, Aditya, Kishore Gawande, and Devesh Kapur, “(When) Do Anti-povertyPrograms Reduce Violence? Indiaıs Rural Employment Guarantee andMaoist Conflict,”Technical Report, Harvard University 2015.
Deaton, Angus and Alessandro Tarozzi, “Prices and poverty in India,” 2000.
Dinkelman, Taryn and Vimal Ranchhod, “Evidence on the impact of minimum wagelaws in an informal sector: Domestic workers in South Africa,” Journal of DevelopmentEconomics, 2012, 99 (1), 27 – 45.
Dreze, Jean and Amartya Sen, Hunger and Public Action number 9780198283652. In‘OUP Catalogue.’, Oxford University Press, 1991.
26
Dutta, Puja, Rinku Murgai, Martin Ravallion, and Dominique van de Walle,“Does India’s Employment Guarantee Scheme Guarantee Employment?,” Policy ResearchWorking Paper Series 6003, World Bank 2012.
Gupta, S., “Were District Choices for NFFWP Appropriate?,” Journal of Indian School ofPolitical Economy, 2006, 18 (4), 641–648.
Imbert, Clement and John Papp, “Estimating leakages in Indiaıs employment guar-antee,” in Reetika Khera, ed., The Battle for Employment Guarantee, Oxford UniversityPress, 2011.
and , “Labor Market Effects of Social Programs: Evidence from India’s EmploymentGuarantee,” American Economic Journal: Applied Economics, 2015, 7 (2), 233–263.
Islam, Mahnaz and Anita Sivasankaran, “How does Child Labor respond to changesin Adult Work Opportunities? Evidence from NREGA,” Technical Report, Harvard Uni-versity 2015.
Jayachandran, Seema, “Selling Labor Low: Wage Responses to Productivity Shocks inDeveloping Countries,” Journal of Political Economy, 2006, 114 (3), pp. 538–575.
Kaur, Supreet, “Nominal wage rigidity in village labor markets,” NBER Working PaperSeries 20770, National Bureau of Economic Research, Inc 2015.
Khanna, Gaurav and Laura Zimmerman, “Guns and Butter? Fighting Violence withthe Promise of Development,” Technical Report, University of Michigan 2014.
Khera, Reetika, The Battle for Employment Guarantee, Oxford University Press, 2011.
Krugman, Paul, “Increasing Returns and Economic Geography,” Journal of Political Econ-omy, June 1991, 99 (3), 483–99.
Mani, Shubha, Jere Behrman, Shaikh Ghalab, and Prudhvikar Reddy, “Impact ofthe NREGS on Schooling and Intellectual Human Capital,” Technical Report, Universityof Pennsylvania 2014.
Mehrotra, Santosh, “NREG Two Years on: Where Do We Go from Here?,” Economicand Political Weekly, 2008, 43 (31).
Mukherjee, Anindita and Debraj Ray, “Labor tying,” Journal of Development Eco-nomics, 1995, 47 (2), 207 – 239.
Muralidharan, Karthik and Paul Niehaus, “Experimentation at Scale,” Technical Re-port, University of California San Diego 2016.
, Jishnu Das, Alaka Holla, and Aakash Mohpal, “The Fiscal Cost of Weak Gover-nance: Evidence from Teacher Absence in India,” Working Paper 20299, National Bureauof Economic Research 2014.
27
, Paul Niehaus, and Sandip Sukhtankar, “Building State Capacity: Evidence fromBiometric Smartcards in India,” American Economic Review, 2016, 106 (10), 2895–2929.
Murgai, Rinku and Martin Ravallion, “Is a guaranteed living wage a good anti-povertypolicy?,” Policy Research Working Paper Series 3640, The World Bank June 2005.
Niehaus, Paul and Sandip Sukhtankar, “Corruption Dynamics: The Golden GooseEffect,” American Economic Journal: Economic Policy, 2013, 5.
and , “The Marginal Rate of Corruption in Public Programs: Evidence from India,”Journal of Public Economics, 2013, 104, 52 – 64.
of India, Planning Commission Government, “Press Notes on Poverty Estimates,2011-12,” Technical Report 2013.
Pai, Sandeep, “Delayed NREGA payments drive workers to suicide,” Hindustan Times,December 29 2013.
Ravallion, Martin, “Market Responses to Anti-Hunger Policies: Effects on Wages, Prices,and Employment,” Technical Report November 1987. World Institute for DevelopmentEconomics Research WP28.
Rosenzweig, Mark R., “Rural Wages, Labor Supply, and Land Reform: A Theoreticaland Empirical Analysis,” The American Economic Review, 1978, 68 (5), 847–861.
Shah, Manisha and Bryce Millett Steinberg, “Workfare and Human Capital Invest-ment: Evidence from India,” Technical Report, University of California, Los Angeles 2015.
Sukhtankar, Sandip, “India’s National Rural Employment Guarantee Scheme: What DoWe Really Know about the World’s Largest Workfare Program?,” India Policy Forum,forthcoming.
Zimmermann, Laura, “Why Guarantee Employment? Evidence from a Large IndianPublic-Works Program,” Working Paper, University of Georgia April 2015.
28
Table 1: Income
(a) SECC data
Lowest bracket Middle bracket Highest bracketIncome bracket
3 levels
(1) (2) (3) (4) (5) (6) (7) (8)
Treatment -.041∗∗∗ -.039∗∗∗ .026∗∗ .025∗∗ .014∗∗ .012∗∗ -.041∗∗∗ -.000088∗∗∗
(.014) (.014) (.011) (.011) (.0065) (.0061) (.014) (.000018)District FE Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared 0.01 0.03 0.01 0.02 0.02 0.04 0.01 0.02Control Mean .83 .83 .13 .13 .038 .038N of cases 1.8M 1.8M 1.8M 1.8M 1.8M 1.8M 1.8M 1.8MEstimator Logit Logit Logit Logit Logit Logit Ordered logit Ordered logitControl Variables No Yes No Yes No Yes No Yes
(b) Survey data
Total income NREGA Ag. labor Other labor Farm Business Misc
(1) (2) (3) (4) (5) (6) (7) (8)
Treatment 10308∗∗ 9594∗∗ 905 3675∗∗ 4471∗∗∗ 1738 -773 293(4638) (4642) (589) (1485) (1585) (2704) (1359) (2437)
BL GP Mean .059(.075)
District FE Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .03 .03 .02 .04 .03 .01 .01 .01Control Mean 71935 71935 4743 14784 9315 21708 6620 14765N. of cases 4932 4898 4931 4932 4932 4932 4932 4932
This table shows treatment effects on various measures of household income. Panel a) uses data from the Socioeconomic andCaste Census (SECC), which reports income categories of the highest earner in the household (HH): the “Lowest bracket”corresponds to earning < Rs. 5000/month, “Middle bracket” to earning between Rs. 5000 & 10000/month, and“Highestbracket” to earning > Rs. 10000/month. The table reports marginal effects, or changes in the predicted probability ofbeing in the respective income bracket (columns 1-6) resulting from a change in a binary treatment indicator from 0 to1. In columns 7-8, we show the marginal effects on the predicted probability of being in the lowest income category. Forcolumns where there “Control Variables” is Yes, the following control variables are included: the age of the head of HH, anindicator for whether the head of HH is illiterate, indicator for whether a HH belongs to Scheduled Castes/Tribes. Panel b)shows treatment effects on types of income using annualized HH data from the endline HH survey for the NREGS sample.“BL GP Mean” is the GP mean of the dependent variable at baseline. “Total Income” is total annualized HH income (inRs.). “NREGS” is the earnings from NREGS. “Ag labor” captures income from agricultural work for someone else, while“Other labor” is income from physical labor for someone else. “Farm” combines income from a HHs’ own land and animalhusbandry, while“Business” captures income from self-employment or a HH’s own business. “Other” is the sum of HHincome not captured by the other categories. Note that the income categories were not as precisely measured at baselineso we cannot include the respective lag of the dependent variable. All regressions include the first principal component of avector of mandal characteristics used to stratify randomization as a control variable. Standard errors clustered at mandallevel are in parentheses. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
29
Table 2: Wages
Wage realizations (Rs.) Reservation wage (Rs.)
(1) (2) (3) (4)
Treatment 6.6∗ 7.8∗∗ 4.9∗ 5.5∗
(3.6) (3.6) (2.9) (2.8)
BL GP Mean .15∗∗∗ .12∗∗∗
(.053) (.043)
District FE Yes Yes Yes Yes
Adj R-squared .07 .07 .05 .05Control Mean 128 128 97 97N. of cases 7304 7090 12905 12791
This table shows treatment effects on wage outcomes from the private labor market in June using survey data. The outcome
“Wage realizations” in columns 1-2 is the average daily wage (in Rs.) an individual received while working for someone else
in June 2012. The outcome “Reservation wage” in columns 3-4 is an individual’s reservation wage (in Rs.) at which he or
she would have been willing to work for someone else in June 2012. The outcome is based on an a question in which the
surveyor asked the respondent whether he or she would be willing to work for Rs. 20 and increased this amount in increments
of Rs. 5 until the respondent answered affirmatively. Observations in the top .5% percentile of the respective wage outcome
in treatment and control are excluded in all regressions. “BL GP Mean” is the Gram Panchayat mean of the dependent
variable at baseline (May 31 to July 4, 2010). All regressions include the first principal component of a vector of mandal
characteristics used to stratify randomization as control variable. Standard errors clustered at mandal level in parentheses.
Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
30
Table 3: Employment
Days workedon NREGS
Days workedprivate sector
Days unpaid/idle
(1) (2) (3) (4) (5) (6)
Treatment .95 .88 .44 .53 -1.2∗∗ -1.2∗∗
(.66) (.64) (.57) (.56) (.59) (.59)
BL GP Mean .14∗∗∗ .22∗∗∗ .16∗∗∗
(.043) (.068) (.052)
District FE Yes Yes Yes Yes Yes Yes
Adj R-squared .09 .10 .01 .02 .06 .07Control Mean 8.2 8.2 7.9 7.9 17 17N. of cases 10504 10431 14514 14429 14163 14078
This table analyzes different labor supply outcomes for June using survey data. “Days worked on NREGS” is the number of
days an individual worked on NREGS during June 2012. “Days worked private sector” is the number of days an individual
worked for somebody else in June 2012. Finally, “Days unpaid/idle” is the sum of days an individual did unpaid work or
stayed idle in June 2012. “BL GP Mean” is the Gram Panchayat mean of the dependent variable at baseline. All regressions
include the first principal component of a vector of mandal characteristics used to stratify randomization as a control variable.
Standard errors clustered at mandal level are in parentheses. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05,∗∗∗p < 0.01. Note that observation count varies between columns due to differences in non-response rates between their
corresponding survey questions. An test of non-response rates by treatment status is shown in Table A.3.
31
Table 4: Migration
Did migrate? Days migrated Hhd sizeMigration common
in May?
(1) (2) (3) (4) (5) (6) (7) (8)
Treatment .024 .022 1.1 .75 .059 .054 .047 .049(.017) (.018) (4.9) (5.1) (.1) (.1) (.055) (.038)
BL GP Mean .093 .3 .044(.09) (.19) (.048)
Migration previously common .54∗∗∗
(.044)
District FE Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .03 .03 .01 .02 .02 .02 .12 .45Control Mean .075 .075 16 16 4.3 4.3 .21 .21N. of cases 4907 4873 4943 4909 4943 4909 809 808Level Hhd Hhd Hhd Hhd Hhd Hhd GP GP
This table illustrates treatment effects on various measures of migration using survey data as well as from a separately
conducted village survey. In columns 1 and 2, the outcome is an indicator for whether any household member stayed away
from home for the purpose of work during the last year. Last year refers to the respective time period from the point of
the endline survey (May 28 to July 15, 2012). In columns 3 and 4, the outcome is sum of all days any household member
stayed away from home for work, while in columns 5 and 6 the number of household members is the dependent variable.
“BL GP Mean” is the Gram Panchayat mean of the dependent variable at baseline. In columns 7-8, the outcome is an
indicator for whether it was common for workers to migrate out of the village in search of work during the month of May
ever since the implementation of NREGS. “Migration common prior to NREGS” is an indicator for whether the same type
of migration during the same time was common prior to the start of NREGS. Note that “prior to NREGS” does not refer to
the Smartcards intervention but rather to the rollout of the entire employment guarantee scheme. All regressions include the
first principal component of a vector of mandal characteristics used to stratify randomization as a control variable. Standard
errors clustered at mandal level are in parentheses. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
32
Table 5: Price levels
Log of Price Index Log of Individual Prices
(1) (2) (3)
Treatment .0039 .0047 -.014(.065) (.024) (.012)
District FE Yes Yes Yes
Item FE No No Yes
Adjusted R-squared 0.98 1.00 0.95N. of cases 60 60 18242Level Village Village Item x HouseholdDropped Goods? Yes No
In this table, we use NSS data on household consumption and prices to test for impacts on price levels for a subset of our
study sample. Columns 1 and 2 show analysis of a village-level price index constructed from NSS data. The outcome variable
is the log of the price index. Column 3 shows analysis on observed commodity prices at the household level. The outcome
is the log of prices. Standard errors clustered at mandal level are in parentheses. Statistical significance is denoted as:∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
33
Tab
le6:
Impac
tson
NR
EG
Spro
ject
counts
&ty
pes
#of
dis
tinct
pro
ject
s#
day
ssp
ent
wor
kin
gon
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Tot
alC
onst
r.Ir
rig.
Lan
ddev
.R
oads
Tot
alC
onst
r.Ir
rig.
Lan
ddev
.R
oads
Tre
atm
ent
-1.2
.11
.099
-1.2
∗61
-10
25-1
1916
1(2
.9)
(.44
)(.
31)
(2.7
)(.
12)
(441
)(1
03)
(247
)(4
35)
(112
)
BL
GP
Mea
n.6
1∗∗∗
.23∗
∗∗.0
67∗∗
∗1.
3∗∗∗
.099
∗∗∗
.4∗∗
∗.0
68∗
.23∗
∗∗.3
6∗∗∗
.11
(.07
5)(.
089)
(.02
1)(.
23)
(.02
6)(.
039)
(.03
9)(.
055)
(.06
3)(.
07)
Dis
tric
tF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
es
Adj.
R-s
quar
ed.2
4.1
7.1
1.2
.13
.35
.3.4
7.2
4.1
1C
ontr
olM
ean
322.
81.
816
.51
6,53
949
21,
770
2,60
632
9N
.of
case
s2,
837
2,83
72,
837
2,83
72,
837
2,89
92,
837
2,83
72,
837
2,83
7
Th
ista
ble
anal
yze
sw
het
her
trea
tmen
tim
pac
ted
the
crea
tion
of
pro
du
ctiv
ity-e
nh
an
cin
gass
ets
thro
ugh
the
typ
eof
NR
EG
Sp
roje
cts
imp
lem
ente
dat
the
GP
-lev
el
usi
ng
NR
EG
Sm
ust
erro
lld
ata.
Th
eou
tcom
esin
colu
mn
s1-5
are
cou
nts
of
un
iqu
ep
roje
cts
inG
Ps
as
iden
tifi
edby
thei
rpro
ject
iden
tifi
cati
on
nu
mb
ers
inth
eN
RE
GS
mu
ster
roll
dat
a.T
he
rele
vant
per
iod
isth
een
dli
ne
stud
yp
erio
d(M
ay28
toJu
ly15,
2012).
Th
eca
tegori
esin
colu
mn
s2-5
(an
dals
oin
6-1
0)
are
base
don
manu
al
mat
chin
gof
pro
ject
titl
esto
any
ofth
efo
llow
ing
cate
gori
es:
con
stru
ctio
n,
irri
gati
on
,la
nd
dev
elop
men
t,ro
ad
s,p
lanta
tion
work
,d
esil
tin
gan
doth
erp
roje
cts
(wit
hth
e
latt
erth
ree
omit
ted
from
the
tab
le).
Inco
lum
ns
6-10
,th
eou
tcom
eva
riab
leis
the
sum
of
day
sw
ork
edw
ith
ina
GP
inth
ere
spec
tive
cate
gory
.T
he
“B
LG
PM
ean
”
isco
nst
ruct
edin
the
sam
ew
ayw
ith
the
refe
ren
ceb
ein
gth
eb
ase
lin
est
ud
yp
erio
d(M
ay31
toJu
ly4,
2010).
All
regre
ssio
ns
incl
ud
eth
efi
rst
pri
nci
pal
com
pon
ent
ofa
vect
orof
man
dal
char
acte
rist
ics
use
dto
stra
tify
ran
dom
izati
on
as
aco
ntr
ol
vari
ab
le.
Sta
nd
ard
erro
rscl
ust
ered
at
man
dal
leve
lare
inp
are
nth
eses
.S
tati
stic
al
sign
ifica
nce
isd
enot
edas
:∗ p
<0.
10,∗∗p<
0.0
5,∗∗∗ p
<0.
01.
34
Table 7: Expenditure
Short-term Expenditure Longer-term expenditureMonthly Per
Capita Expenditure
(1) (2) (3) (4) (5)
Treatment -108 -428 -24 -646 6653(1029) (1033) (3239) (3227) (12579)
BL GP Mean .051∗∗ -.003(.02) (.006)
District FE Yes Yes Yes Yes Yes
Adj R-squared .01 .02 .01 .01 .03Control Mean 18915 18915 38878 38878 189430N. of cases 4943 4909 4943 4909 479Recall period 1 month 1 month 1 year 1 year 1 monthSurvey NREGA NREGA NREGA NREGA NSS
This table analyzes different categories of household expenditure using data from our household survey and the NSS. The
outcome in columns 1 & 2 “Short-term expenditure” is the sum of spending on items such as produce, other food items,
beverages, fuel, entertainment, personal care items or rent measured from our survey data. The time frame for this category
is one month, which is also the time period that was referred to in the survey. The outcome in columns 3 & 4 “Longer-term
expenditure” comprises medical and social (e.g. weddings, funerals) expenses as well as tuition fees and durable goods. In the
survey, people were asked to indicate their spending on these items during the last year. “BL GP Mean” is the Gram Panchayat
mean of the dependent variable at baseline. The outcome in column 5 “Monthly Per Capita Expenditure” is measures at the
household-level in the NSS survey. Note that the households from the NSS data are not the same as our sample households.
All regressions from our survey data include the first principal component of a vector of mandal characteristics used to stratify
randomization as a control variable. Standard errors clustered at mandal level are in parentheses. Statistical significance is
denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
35
Table 8: Savings, assets and loans
Total savings (Rs.) Total loans (Rs.) Owns land (%)
(1) (2) (3) (4) (5) (6)
Treatment 1064 1120 11210∗∗ 11077∗∗ .056∗∗ .049∗∗
(859) (877) (4741) (4801) (.024) (.024)
BL GP Mean .027 .038 .21∗∗∗
(.071) (.039) (.042)
District FE Yes Yes Yes Yes Yes Yes
Adj R-squared .00 .00 .01 .01 .01 .03Control Mean 2966 2966 68108 68108 .59 .59N. of cases 4916 4882 4943 4909 4921 4887
This table analyzes household assets using survey data. “Total savings (Rs.)” is defined as the total amount of a household’s
current cash savings (in Rs), including money kept in bank accounts or Self-Help Groups. “Total loans (Rs.)” is the total
principal of the household’s five largest active loans (in Rs). “Owns land (%)” is an indicator for whether a household
reports to own any land. “BL GP Mean” is the Gram Panchayat mean of the dependent variable at baseline. All regressions
include the first principal component of a vector of mandal characteristics used to stratify randomization as a control variable.
Standard errors clustered at mandal level are in parentheses. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05,∗∗∗p < 0.01.
36
Tab
le9:
Lan
duti
liza
tion
and
irri
gati
on
Irri
gate
dla
nd
(ac.
)T
otal
land
(ac.
)T
otal
fallow
s(%
)N
on-a
gri.
use
(%)
Net
area
sow
n(%
)N
etar
eair
riga
ted
(%)
(1)
(2)
(3)
(4)
(5)
(6)
Tre
atm
ent
-5.4
-6.7
-.74
-.83
1.1
.001
8(5
.3)
(6)
(1.1
)(1
.5)
(1.7
)(.
0084
)
BL
GP
Mea
n.0
074
.48∗
∗∗.4
9∗∗∗
.91∗
∗∗
(.00
6)(.
14)
(.06
4)(.
039)
Dis
tric
tF
EY
esY
esY
esY
esY
esY
es
Adju
sted
R-s
quar
ed0.
000.
000.
620.
620.
880.
83C
ontr
olM
ean
7.1
1111
9.1
28.1
8N
.of
case
s1,
835,
191
1,83
5,19
015
415
415
415
4L
evel
Hhd
Hhd
Man
dal
Man
dal
Man
dal
Man
dal
Dat
aso
urc
eSE
CC
SE
CC
DSH
DSH
DSH
DSH
Th
ista
ble
anal
yze
sla
nd
owner
ship
,la
nd
uti
liza
tion
and
irri
gati
on
usi
ng
data
from
the
Soci
oec
onom
ican
dC
ast
eC
ensu
s(S
EC
C)
an
dth
ean
nu
al
Dis
tric
tS
tati
stic
al
Han
db
ook
s(D
SH
)20
12-2
013
(200
9-20
10fo
rth
ela
gged
dep
end
ent
vari
ab
le“
BL
GP
Mea
n”)
for
the
eight
stud
yd
istr
icts
.“Ir
rigate
dla
nd
(ac.
)”is
the
am
ou
nt
of
lan
din
acre
sow
ned
wit
has
sure
dir
riga
tion
for
two
crop
s.“T
ota
lla
nd
(ac.
)”is
the
tota
lam
ou
nt
of
lan
dow
ned
,in
clu
din
gb
oth
irri
gate
dan
du
nir
rigate
dla
nd
.“T
ota
l
fall
ows”
isth
eto
tal
area
wh
ich
aton
ep
oint
was
take
nu
por
cou
ldb
eta
ken
up
for
cult
ivati
on
bu
tis
curr
entl
yle
ftfa
llow
.T
his
isth
esu
mof
“cu
rren
tfa
llow
s”(c
rop
ped
area
wh
ich
iske
pt
fall
owin
the
curr
ent
yea
r),
“oth
erfa
llow
s”(l
an
dw
hic
his
has
bee
nle
ftfa
llow
for
more
than
1ye
ar
bu
tle
ssth
an
5ye
ars
)an
d“cu
ltu
rab
lew
ast
e”
(lan
dav
aila
ble
wh
ich
has
bee
nle
ftfa
llow
for
the
mor
eth
an
5ye
ars
bu
tw
ou
ldb
eav
ailab
lefo
rcu
ltiv
ati
on
).“N
on
agri
.u
seare
a”
isth
eare
aocc
up
ied
by
bu
ild
ings,
road
s,ra
ilw
ays
oru
nd
erw
ater
.“N
etar
easo
wn
”is
tota
lare
aso
wn
wit
hcr
op
san
dorc
hard
sw
her
eare
ath
at
isso
wn
more
than
on
ceis
cou
nte
don
lyon
ce.
“N
etare
a
irri
gate
d”
isth
eto
tal
area
irri
gate
dth
rou
ghan
yso
urc
e.T
he
qu
anti
ties
inco
lum
ns
3-6
are
inp
erce
nta
ge
of
tota
lm
an
dal
are
a.
Note
that
the
nu
mb
erof
ob
serv
ati
on
is15
4(n
ot15
7-
the
nu
mb
erof
stu
dy
man
dal
s)d
ue
toin
com
ple
ted
ata
pu
bli
shed
inth
eD
SH
sof
thre
em
an
dals
.In
the
colu
mn
1-2
regre
ssio
ns
that
use
SE
CC
data
,
the
foll
owin
gco
ntr
olva
riab
les
are
incl
ud
ed:
the
age
ofth
eh
ead
of
HH
,an
ind
icato
rfo
rw
het
her
the
hea
dof
HH
isil
lite
rate
,in
dic
ato
rfo
rw
het
her
aH
Hb
elon
gs
toS
ched
ule
dC
aste
s/T
rib
es.
All
regr
essi
ons
incl
ud
eth
efi
rst
pri
nci
pal
com
pon
ent
of
ave
ctor
of
mand
al
chara
cter
isti
csu
sed
tost
rati
fyra
nd
om
izati
on
as
aco
ntr
ol
vari
able
.R
obu
stst
and
ard
erro
rsar
ein
par
enth
eses
,an
dare
clu
ster
edat
the
man
dal
leve
lfo
rre
gre
ssio
ns
run
at
the
hou
seh
old
leve
l.S
tati
stic
al
signifi
can
ceis
den
ote
d
as:∗ p
<0.1
0,∗∗p<
0.05
,∗∗∗ p
<0.
01.
37
Table 10: Testing for existence of spatial spillovers
(a) Control
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 7.8 17∗∗ 22∗∗ 21∗∗ 24∗∗ -.76 -1.5 -1.7 -3.4∗ -5.4∗∗
(6.6) (6.9) (8.6) (8.6) (11) (1.2) (1.4) (1.8) (2) (2.2)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .06 .05 .05 .05 .05 .08 .07 .07 .07 .07Treatment Mean 128 128 128 128 128 17 17 17 17 17N. of cases 1850 2057 2063 2063 2063 3701 4076 4095 4095 4095% of sample 90 100 100 100 100 90 100 100 100 100
(b) Treatment
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 12∗∗∗ 11∗ 14∗ 14 15 -1.1 -2.3∗∗ -3.1∗∗ -3.5∗∗ -4.3∗∗
(4.4) (5.9) (7.5) (10) (12) (.81) (1.1) (1.3) (1.5) (1.8)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .08 .08 .08 .08 .08 .07 .07 .07 .07 .07Control Mean 134 134 134 134 134 16 16 16 16 16N. of cases 4710 4992 5129 5182 5206 9021 9613 9882 9969 10000% of sample 90 96 99 100 100 90 96 99 100 100
(c) Pooled
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 9.9∗∗∗ 12∗∗ 13∗∗ 13∗ 16∗ -1.1 -2∗∗ -2.5∗∗ -3.3∗∗ -4.2∗∗∗
(3.6) (4.8) (6.2) (7.8) (9.5) (.72) (.91) (1.1) (1.3) (1.5)Treatment 8.5∗∗ 8.3∗∗ 8∗∗ 7.4∗∗ 7.3∗∗ -1.5∗∗ -1.5∗∗ -1.5∗∗ -1.5∗∗∗ -1.5∗∗∗
(3.5) (3.5) (3.5) (3.6) (3.6) (.61) (.58) (.58) (.58) (.57)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .07 .07 .07 .07 .07 .07 .07 .07 .07 .07Control Mean 128 128 128 128 128 17 17 17 17 17N. of cases 6560 7049 7192 7245 7269 12722 13689 13977 14064 14095% of sample 90 97 99 100 100 90 97 99 100 100
This table show the impact of NRp on treatment effects at radius 10, 15, 20, 25, and 30 kilometers on private wages and days
worked using survey datas. All analysis was conducted separately for (a) control and (b) treatment, and then for (c) the
pooled sample. In each panel, the outcome “Wage realizations” in columns 1-5 is the average daily wage (in Rs.) an individual
received while working for someone else in June 2012 (endline). The outcome “Days worked private sector” in columns 6-10
is the number of days an individual worked for somebody else in June 2012 (endline). The “% GPs treated within R km”
is NRp , or the ratio of the number of GPs in treatment mandals within a given radius R km over the total GPs within wave
1, 2 or 3 mandals. Note that wave 2 mandals are included in the denominator, and that same-mandal GPs are excluded in
both the denominator and numerator. The entire GP sample used in randomization are included. Standard errors clustered
at mandal level are in parentheses. All regressions include the first principal component of a vector of mandal characteristics
used to stratify randomization as a control variable. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
Note that the variation in observation counts is due to the construction of the spatial exposure. Observations with no value
for NRp at a given radius R will not be included for the respective regression.
38
Table 11: Estimating total treatment effects including spillovers (IV)
(a) Wage
Wage realizations (Rs.) Reservation wage (Rs.)
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
Treatment -10 4.2 4.9 5.6 7.6 1.3 6.1 5.4 4.8 2(9.3) (9) (9.2) (9.4) (10) (6.7) (6.2) (6.3) (6.9) (7.2)
% GPs treated within R km 13 31∗ 26 21 27 9.6 12 8.4 5 -.39(16) (17) (18) (15) (17) (14) (12) (13) (12) (11)
% GPs treated within R km X Treatment 16 -13 -7.2 -4.6 -9 -2.5 -8.2 -3.9 -1.1 6.2(20) (21) (22) (21) (23) (16) (15) (16) (16) (16)
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Total Treatment Effect 18∗∗∗ 23∗∗∗ 24∗∗∗ 22∗∗ 26∗∗ 8.5∗∗ 9.8∗∗ 9.9∗ 8.7 7.8SE (5.6) (6.6) (7.8) (9.1) (11) (4) (4.6) (5.4) (6.5) (7.4)F-stat for % GPs treated within R km 355 205 259 340 311 344 205 267 367 321F-stat for % GPs treated within R km X Treatment 224 156 165 171 115 226 165 186 187 118Control Mean 128 128 128 128 128 97 97 97 97 97N. of cases 6560 7049 7192 7245 7269 11614 12498 12732 12818 12852
(b) Labor
Days worked private sector Days idle/unpaid
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
Treatment .31 -.27 .037 .27 .48 .33 .79 .68 -.25 -.86(1.6) (1.6) (1.7) (1.7) (1.8) (1.6) (1.5) (1.5) (1.4) (1.5)
% GPs treated within R km 3.5 3.1 3.4 4.1 5.1 -2 -2.7 -2.5 -4 -5.8∗∗
(3.1) (2.9) (3) (3.1) (3.4) (3.3) (3) (3) (2.8) (2.8)% GPs treated within R km X Treatment -2.2 -.53 -.64 -.82 -1.2 -.98 -1.8 -2.5 -.86 .48
(3.6) (3.6) (3.8) (4) (4.3) (3.7) (3.6) (3.5) (3.4) (3.4)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Total Treatment Effect 1.6 2.3∗ 2.8∗ 3.5∗∗ 4.4∗∗ -2.7∗∗∗ -3.7∗∗∗ -4.3∗∗∗ -5.1∗∗∗ -6.2∗∗∗
SE (1.1) (1.3) (1.5) (1.7) (1.9) (1) (1.3) (1.5) (1.6) (1.8)F-stat for % GPs treated within R km 362 206 242 362 360 365 202 236 360 367F-stat for % GPs treated within R km X Treatment 247 158 146 166 136 246 156 143 168 144Control Mean 7.9 7.9 7.9 7.9 7.9 17 17 17 17 17N. of cases 13008 13995 14300 14397 14441 12722 13689 13977 14064 14095
(c) Total Income
Total Income
R = 10 R = 15 R = 20 R = 25 R = 30
Treatment 6269 18794 19742 10544 5341(11543) (11846) (12616) (11412) (11225)
% GPs treated within R km 28562 34327 34456 15459 2569(30417) (27451) (30555) (24440) (19364)
% GPs treated within R km X Treatment -16074 -33250 -31296 -4914 10254(33624) (31336) (34779) (29362) (25663)
District FE Yes Yes Yes Yes Yes
Total Treatment Effect 18757∗∗∗ 19872∗∗ 22903∗∗ 21089∗ 18164SE (7140) (8944) (11560) (11973) (12448)F-stat for % GPs treated within R km 330 216 255 361 361F-stat for % GPs treated within R km X Treatment 225 166 159 177 150Control Mean 71935 71935 71935 71935 71935N. of cases 4420 4767 4864 4892 4903
This table provides estimates for total treatment effects using comprehensive spatial exposure NRp while instrumenting for
exogenous spatial exposure NRp for (a) wage outcomes, (b) labor outcomes, and (c) total income. Each structural equation
contains a treatment indicator, NRp , and an interaction between NR
p and treatment. The NRp is the ratio of the number of
GPs in treatment mandals within a given radius R km over the total GPs within wave 1, 2 or 3 mandals. The instruments
used in the first stage are NRp and the interaction between NR
p and treatment. The total treatment effect estimate reported
in the second section of the table is the sum of the coefficients for those three variables. All regressions also include the first
principal component of a vector of mandal characteristics used to stratify randomization as a control variable. Statistical
significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
39
Andhra PradeshStudy Districts and MandalsGroup
Treatment
Wave 2 & non-study
Control
Figure 1: Study districts with treatment and control mandals
This map (reproduced from Muralidharan et al. (2016)) shows the 8 study districts - Adilabad, Anantapur, Kadapa, Kham-
mam, Kurnool, Nalgonda, Nellore, and Vizianagaram - and the assignment of mandals (sub-districts) within those districts
to one of four study conditions. Mandals were randomly assigned to one of three waves: 112 to wave 1 (treatment), 139
to wave 2, and 45 to wave 3 (control). Wave 2 was created as a buffer to maximize the time between program rollout in
treatment and control waves; our study did not collect data on these mandals. A “non-study” mandal is a mandal that
did not enter the randomization process because the Smartcards initiative had already started in those mandals (109 out of
405). Randomization was stratified by district and by a principal component of mandal characteristics including population,
literacy, proportion of Scheduled Caste and Tribe, NREGS jobcards, NREGS peak employment rate, proportion of SSP
disability recipients, and proportion of other SSP pension recipients.
40
Figure 2: Effects on income: SECC
Less than Rs. 5000
Between Rs. 5000 and Rs. 10000
More than Rs.10000
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Control Treatment
The figure shows the proportions of households in each of the three income brackets in the Socioeconomic and Caste Census
(SECC) 2011 (enumeration started in late June 2011) by treatment and control households. The error bars indicate a 95%
confidence interval around each proportion. The standard error for every category is < 0.001, so these intervals are not visible
on the scale of the figure.
41
Figure 3: Annualized per capital income
0.00
0.25
0.50
0.75
1.00
0 20000 40000 60000 80000Annualized Per Capita Income (in Rps.)
Cum
ulat
ive
dens
ity
Control Treatment
This figure shows an empirical cdf of total annualized per capita income by household for treatment and control groups using
data from the endline household survey. Annualized per capita income was calculated by dividing the total annual household
income by number of household members. The vertical line indicates the annualized per capita poverty line (860 Rs. per
month or 10,320 Rs. per year).
Figure 4: Private sector work in June
0.00
0.25
0.50
0.75
1.00
0 10 20 30Days of private sector paid work
Cum
ulat
ive
dens
ity
Control Treatment
This figure shows an empirical cdf of the number of days an individual worked for someone else during June 2012, using data
from the endline household survey. The dashed lines indicate in-sample means (not weighted by sampling probabilities) in
treatment and control, respectively.
42
Figure 5: Constructing measures of exposure to spatial spillovers
(a) Control
R
p
(b) Treatment
R
p
This figure illustrates the construction of measures of spatial exposure to treatment for a given panchayat p (denoted by
the black X symbol) and radius R in a control mandal (a) and a treatment mandal (b). Dark (light) blue dots represent
treatment (control) panchayats; black lines represent mandal borders . As in the text, Nrp is the fraction of GPs within a
radius R of panchayat p which were assigned to treatment. Nrp is the fraction of GPs within a radius R of panchayat p and
within a different mandal which were assigned to treatment. In Panel (a) these measures are Nrp = 1
4 and NRp = 1
2 , while in
Panel (b) they are Nrp = 3
4 and NRp = 1
2 .
43
Table A.1: Baseline balance in administrative data
Treatment Control Difference p−value
(1) (2) (3) (4)
Official records from GoAP in 2010
% population working .53 .52 .0062 .47% male .51 .51 .00023 .82% literate .45 .45 .0043 .65% SC .19 .19 .0025 .81% ST .1 .12 -.016 .42Jobcards per capita .54 .55 -.0098 .63Pensions per capita .12 .12 .0015 .69% old age pensions .48 .49 -.012 .11% weaver pensions .0088 .011 -.0018 .63% disabled pensions .1 .1 .0012 .72% widow pensions .21 .2 .013∗∗ .039
2011 census rural totals
Population 45,580 45,758 -221 .91% population under age 6 .11 .11 -.00075 .65% agricultural laborers .23 .23 -.0049 .59% female agricultural laborers .12 .12 -.0032 .52% marginal agricultural laborers .071 .063 .0081 .14
2001 census village directory
# primary schools per village 2.9 3.2 -.28 .3% village with medical facility .67 .71 -.035 .37% villages with tap water .59 .6 -.007 .88% villages with banking facility .12 .16 -.034∗∗ .021% villages with paved road access .8 .81 -.0082 .82Avg. village size in acres 3,392 3,727 -336 .35
This table compares official data on baseline characteristics across treatment and control (T & C) mandals. Column 3 reports
the difference in T & C means, while column 4 reports the p-value on the treatment indicator from simple regressions of
the outcome with district fixed effects as the only controls. A “jobcard” is a household level official enrollment document
for the NREGS program. “SC” (“ST”) refers to Scheduled Castes (Tribes). “Old age”, “weaver”, “disabled” and “widow”
are different eligibility groups within the SSP administration. “Working” is defined as the participation in any economically
productive activity with or without compensation, wages or profit. “Main” workers are defined as those who engaged in
any economically productive work for more than 183 days in a year. “Marginal” workers are those for whom the period
they engaged in economically productive work does not exceed 182 days. The last set of variables is taken from 2001 census
village directory which records information about various facilities within a census village (the census level of observation).
“# primary schools per village” and “Avg. village size in acres” are simple mandal averages (others are simple percentages)
of the respective variable. Sampling weights are not needed since all villages within a mandal are used. Note that we did
not have this information available for the 2011 census and hence used 2001 census data. Statistical significance is denoted
as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01
44
Table A.2: Baseline balance in survey data
Treatment Control Difference p-value
(1) (2) (3) (4)
Hhd members 4.8 4.8 .022 .89BPL .98 .98 .0042 .73Scheduled caste .22 .25 -.027 .35Scheduled tribe .12 .11 .0071 .81Literacy .42 .42 .0015 .93Annual income 41,482 42,791 -1,290 .52Total annual expenditure 687,128 657,228 26,116 .37Short-term expenditure 52,946 51,086 1,574 .45Longer-term expenditure 51,947 44,390 7,162 .45Pay to work/enroll .011 .0095 .00099 .82Pay to collect .058 .036 .023 .13Ghost Hhd .012 .0096 .0019 .75Time to collect 156 169 -7.5 .62Owns land .65 .6 .058 .06∗
Total savings 5,863 5,620 3.7 1.00Accessible (in 48h) savings 800 898 -105 .68Total loans 62,065 57,878 5,176 .32Owns business .21 .16 .048 .02∗∗
Number of vehicles .11 .12 -.014 .49Average payment delay 28 23 .036 .99Payment delay deviation 11 8.8 -.52 .72Official amount 172 162 15 .45Survey amount 177 189 -10 .65Leakage -5.1 -27 25 .15NREGS availability .47 .56 -.1 .02∗∗
Hhd doing NREGS work .43 .42 .0067 .85NREGS days worked, June 8.3 8 .33 .65Private sector days worked, June 4.8 5.3 -.49 .15Days unpaid/idle, June 22 22 .29 .47Average daily wage private sector, June 96 98 -3.7 .34Daily reservation wage, June 70 76 -6.8 .03∗∗
NREGS hourly wage, June 13 14 -1.3 .13NREGS overreporting .15 .17 -.015 .55# addi. days hhd wanted NREGS work 15 16 -.8 .67
This table compares baseline characteristics across treatment and control mandals from our survey data. Columns 3 reports
the difference in treatment and control means, while columns 4 reports the p-value on the treatment indicator from a simple
regressions of the outcome with district fixed effects as the only controls. “BPL” is an indicator for households below
the poverty line. “Accessible (in 48h) savings” is the amount of savings a household could access within 48h. “NREGS
availability” is an indicator for whether a household believes that anybody in the village could get work on NREGS when
they want it. Standard errors are clustered at the mandal level. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05,∗∗∗p < 0.01.
45
Table A.3: Non-response and response composition rates by treatment status
(a) Full sample
Treatment Control Difference p-value N
Wage realizations (Rs.) .013 .011 .0018 .59 7417Reservation wage (Rs.) .4 .39 .0072 .64 21435Days worked private sector .33 .3 .031∗∗ .038 21435Days unpaid .36 .34 .021 .11 21435Days idle .35 .33 .02 .12 21435Days unpaid/idle .34 .33 .019 .13 21435Days worked > 0 .52 .49 .028 .21 14512Avg. wage ≥ reservation wage .98 .99 -.0029 .57 7286
(b) People of working age (18-65)
Treatment Control Difference p-value N
Wage realizations (Rs.) .013 .012 .0014 .68 7101Reservation wage (Rs.) .15 .15 -.002 .92 14425Days worked private sector .085 .082 .0034 .63 14425Days unpaid .098 .097 .0016 .86 14425Days idle .088 .088 -.000085 .99 14425Days unpaid/idle .086 .087 -.00095 .89 14425Days worked > 0 .54 .52 .016 .44 13210Avg. wage ≥ reservation wage .98 .99 -.0025 .62 6973
This table analyzes response rates to key questions regarding labor market outcomes. Columns 1 & 2 show the proportion
of missing answers to the respective question in treatment and control. Column 3 reports the regression-adjusted treatment
difference between treatment and control from a linear regression with the first principal component of a vector of mandal
characteristics used to stratify randomization and district fixed effects as the only control variables. Column 4 reports the
p-value of a two-sided t-test with the null-hypothesis being that the difference (Column 3) is equal to 0. Column 5 reports
the number of individuals who ought to have answered the question. “Wage realizations” is the average daily wage (in Rs.)
an individual received while working for someone else in June 2012. “Reservation wage” is an individual’s reservation wage
(in Rs.) at which he or she would have been willing to work for someone else in June 2012. The outcome is based on an a
question in which the surveyor asked the respondent whether he or she would be willing to work for Rs. 20 and increased
this amount in increments of Rs. 5 until the respondent answered affirmatively. “Days worked private sector” is the number
of days an individual worked for somebody else in June 2012. “Days idle” and “Days unpaid” is the number of days an
individual stayed idle or did unpaid work in June 2012. “Days unpaid/idle” is the sum of the latter two variables. Note that
the base group for “Wage realizations” is the set of individuals who reported a strictly positive number of days worked for
someone else. Similarly, the base group for “Days worked > 0” is the set of individuals that reported non-missing values for
days worked for someone else. Panel b) restricts the sample to individuals of age between 18 and 65 years. Standard errors
clustered at mandal level are in parentheses. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
46
Table A.4: Income (Survey Data), with censoring
Total income NREGA Ag. labor Other labor Farm Business Misc
(1) (2) (3) (4) (5) (6) (7) (8)
Treatment 9511∗∗ 8761∗∗ 905 3675∗∗ 4471∗∗∗ 1738 -773 293(3723) (3722) (589) (1485) (1585) (2704) (1359) (2437)
BL GP Mean .025(.071)
District FE Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .04 .04 .02 .04 .03 .01 .01 .01Control Mean 69122 69122 4743 14784 9315 21708 6620 14765N. of cases 4908 4874 4931 4932 4932 4932 4932 4932
In this table, we perform a robustness check for Table 1b, which shows treatment effects on various types of income using
annualized household data from the household survey. We censor observations in the top .5% percentile of total annualized
income in treatment and control households. “BL GP Mean” is the Gram Panchayat mean of the dependent variable at
baseline. “NREGS” is the earnings from NREGS. “Ag labor” captures income from agricultural work for someone else, while
“Other labor” is income from physical labor for someone else. “Farm” combines income from a households’ own land and
animal husbandry, while“Business” captures income from self-employment or through a household’s own business. “Other”
is the sum of household income not captured by any of the other categories. Note that the income categories were not as
precisely measured at baseline which is why we cannot include the respective lag of the dependent variable “ BL GP Mean”.
All regressions include the first principal component of a vector of mandal characteristics used to stratify randomization
as a control variable. Standard errors clustered at mandal level are in parentheses. Statistical significance is denoted as:∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
47
Tab
leA
.5:
Het
erog
enei
tyin
inco
me
gain
sby
hou
sehol
dch
arac
teri
stic
s
Tot
alin
com
eSSP
Ag.
lab
orO
ther
lab
orF
arm
Busi
nes
sM
isc
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
Tre
atm
ent
1243
6∗∗
1115
2∗∗
287
402
4528
∗∗∗
4335
∗∗∗
5206
∗∗∗
5306
∗∗∗
3096
1423
-694
-128
1-2
213
08(5
776)
(498
0)(6
58)
(487
)(1
592)
(148
1)(1
610)
(176
1)(3
362)
(299
3)(1
611)
(160
5)(3
264)
(251
0)
Hhd
frac
tion
elig
ible
for
SSP
-377
08∗∗
∗24
81∗∗
-605
3∗∗∗
-682
1∗∗∗
-130
72∗∗
∗-5
679∗
∗∗-1
1624
∗∗
(735
1)(1
009)
(196
8)(2
124)
(498
3)(1
817)
(445
9)
Hhd
frac
tion
elig
ible
for
SSP
XT
reat
men
t-1
4538
3234
-461
2∗-3
977
-735
4-4
3617
75(1
0185
)(3
679)
(256
4)(3
537)
(628
1)(2
326)
(577
0)
Hea
dof
hhd
isw
idow
-178
21∗∗
∗-5
426
72-1
334
-149
95∗∗
∗-5
853∗
∗∗-3
12(6
372)
(657
)(2
645)
(222
8)(2
907)
(172
6)(3
690)
Hea
dof
hhd
isw
idow
XT
reat
men
t-8
804
3375
-481
6-6
076∗
1010
3157
-302
8(8
533)
(261
1)(3
755)
(318
1)(3
930)
(216
0)(4
363)
Dis
tric
tF
EY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
esY
es
BL
GP
Mea
nY
esY
esN
oN
oN
oN
oN
oN
oN
oN
oN
oN
oN
oN
oA
dj
R-s
quar
ed.0
57.0
42.0
28.0
21.0
48.0
38.0
31.0
28.0
23.0
22.0
11.0
093
.01
.006
9C
ontr
olM
ean
7193
571
935
6079
6079
1478
414
784
9315
9315
2170
821
708
6620
6620
1342
813
428
N.
ofca
ses
4898
4836
4932
4870
4932
4870
4932
4870
4932
4870
4932
4870
4932
4870
Inth
ista
ble
we
anal
yze
wh
eth
era
hou
seh
old
s’s
pot
enti
alab
ilit
yto
per
form
manu
al
lab
or
aff
ects
the
ou
tcom
esof
the
inte
rven
tion
.W
eco
nsi
der
vari
ou
sty
pes
of
inco
me
usi
ng
annu
aliz
edh
ouse
hol
dd
ata
from
the
end
lin
eh
ouse
hold
surv
eyfo
rth
eN
RE
GS
sam
ple
.“B
LG
PM
ean”
isth
eG
ram
Pan
chay
at
mea
nof
the
dep
end
ent
vari
ab
le
atb
asel
ine
(du
eto
avai
lab
ilit
yon
lyin
clu
ded
inco
lum
ns
1-2
).“N
RE
GS”
are
earn
ings
from
the
NR
EG
Sp
rogra
m.
“A
g.
lab
or”
cap
ture
sin
com
efr
om
agri
cult
ura
lw
ork
for
som
eon
eel
se,
wh
ile
“Oth
erla
bor
”is
inco
me
from
physi
cal
lab
or
for
som
eon
eel
se.
“F
arm
”co
mb
ines
inco
me
from
ah
ou
seh
old
s’ow
nla
nd
an
dan
imal
hu
sban
dry
,
wh
ile“
Bu
sin
ess”
cap
ture
sin
com
efr
omse
lf-e
mp
loym
ent
or
thro
ugh
ah
ou
seh
old
’sow
nb
usi
nes
s.“O
ther
”is
the
sum
of
hou
seh
old
inco
me
not
cap
ture
dby
any
of
the
oth
erca
tego
ries
.In
pan
ela)
,“H
hd
frac
tion
elig
ible
for
SS
P”
isth
efr
act
ion
of
hou
seh
old
mem
ber
sw
ho
iden
tify
as
elig
ible
for
SS
P,
thou
gh
they
may
not
act
uall
y
rece
ive
pen
sion
.“H
hd
frac
tion
elig
ible
for
SS
Px
Tre
atm
ent”
an
d“H
ead
of
hh
dis
wid
owx
Tre
atm
ent”
are
inte
ract
ion
term
sco
nst
ruct
edby
mult
iply
ing
the
resp
ecti
ve
vari
able
wit
hth
eb
inar
ytr
eatm
ent
ind
icat
or.
Inp
anel
b),
“D
isab
led
”is
an
indic
ato
rfo
rw
het
her
this
sam
ple
dp
ensi
on
erin
this
hou
seh
old
isel
igib
lefo
rd
isab
led
per
son
sp
ensi
ons
(Rs.
500
per
mon
that
the
tim
eof
the
stu
dy).
“D
isab
led
or
OA
P”
ind
icate
sh
ou
seh
old
sin
wh
ich
the
sam
ple
dp
ensi
on
eris
elig
ible
for
dis
ab
led
per
son
s
or
old
age
pen
sion
s(a
tth
eti
me
ofth
est
ud
y,R
s.20
0fo
rp
eop
leof
age
65-7
9an
dR
s.500
for
peo
ple
ab
ove
79
per
month
).A
gain
,b
oth
thes
eva
riab
les
are
als
o
incl
ud
edas
ali
nea
rin
tera
ctio
nw
ith
the
trea
tmen
tin
dic
ato
r.N
ote
that
pen
sion
sch
eme
was
iden
tifi
edfr
om
offi
cial
rost
ers
rath
erth
an
from
the
hou
seh
old
surv
ey.
All
regr
essi
ons
incl
ud
eth
efi
rst
pri
nci
pal
com
pon
ent
ofa
vect
or
of
man
dal
chara
cter
isti
csu
sed
tost
rati
fyra
nd
om
izati
on
as
aco
ntr
ol
vari
ab
le.
Sta
nd
ard
erro
rscl
ust
ered
atm
and
alle
vel
are
inp
aren
thes
es.
Sta
tist
ical
sign
ifica
nce
isd
enote
das:∗ p
<0.
10,∗∗p<
0.05,∗∗∗ p
<0.
01.
48
Table A.6: Robustness checks for private sector wage outcomes
(a) Includes Wage Outliers
Wage realizations (Rs.) Reservation wage (Rs.)
(1) (2) (3) (4)
Treatment 5.6 6.8∗ 5 5.6∗
(4.1) (4.1) (3.3) (3.2)BL GP Mean .15∗∗∗ .091∗∗
(.054) (.039)District FE Yes Yes Yes Yes
Adj R-squared .05 .05 .03 .03Control Mean 131 131 99 99N. of cases 7326 7112 12955 12841
(b) Restricts sample to age 18-65
Wage realizations (Rs.) Reservation wage (Rs.)
(1) (2) (3) (4)
Treatment 6.6∗ 7.9∗∗ 5∗ 5.7∗
(3.7) (3.8) (3) (2.9)BL GP Mean .16∗∗∗ .1∗∗∗
(.049) (.033)District FE Yes Yes Yes Yes
Adj R-squared .07 .07 .05 .06Control Mean 129 129 98 98N. of cases 6989 6782 12227 12124
(c) Excludes respondents who did not work in June
Wage realizations (Rs.) Reservation wage (Rs.)
(1) (2) (3) (4)
Treatment 6.4∗ 7.6∗∗ 4.7 5.4∗
(3.6) (3.6) (2.9) (2.8)BL GP Mean .16∗∗∗ .1∗∗∗
(.048) (.033)District FE Yes Yes Yes Yes
Adj R-squared .07 .07 .05 .05Control Mean 128 128 97 97N. of cases 7256 7043 12859 12745
In this table, we perform robustness checks for Table 2, which shows treatment effects on wage outcomes from the private
labor market using survey data . In Panel a), we include observations in the top .5% percentile of the respective wage
outcome in treatment and control are included in all regressions. In Panel b), the sample is restricted to respondents in age
18 to 65 and exclude observations in the top .5% percentile of the respective wage outcome in treatment and control for all
regressions. In Panel c), we drop observations from survey respondents who have did not work in the month of June and
exclude observations in the top .5% percentile of the respective wage outcome in treatment and control for all regressions.
The outcome in columns 1-4 is the average daily wage (in Rs.) an individual received while working for someone else in
June 2012. In columns 5-8, the outcome is an individual’s reservation wage (in Rs.) at which he or she would have been
willing to work for someone else in June 2012. The outcome is based on an a question in which the surveyor asked the
respondent whether he or she would be willing to work for Rs. 20 and increased this amount in increments of Rs. 5 until
the respondent answered affirmatively. “BL GP Mean” is the Gram Panchayat mean of the dependent variable at baseline
(May 31 to July 4, 2010). All regressions include the first principal component of a vector of mandal characteristics used to
stratify randomization as control variable. Standard errors clustered at mandal level in parentheses. Statistical significance
is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
49
Table A.7: Predictors of differential non-response and response composition
Missing response toDays worked
> 0Wage Realizations> Res. wage
(1) (2) (3) (4) (5) (6)Wage realizations Reservation wage Days worked Days idle/unpaid
Member is female -.0051 -.0032 -.0016 .0069 -.022 .0069(.0047) (.017) (.015) (.015) (.021) (.0063)
Above median hhd income -.0047 .018 .033∗ .011 .05 -.0045(.0055) (.017) (.019) (.016) (.033) (.0094)
Hhd is ST, SC or OBC .023 .022 .031 .012 -.0042 -.011(.016) (.03) (.025) (.025) (.045) (.012)
Hhd below BPL -.012 .024 .045 .022 .091∗∗ -.0029(.012) (.033) (.031) (.029) (.043) (.0084)
Any hhd member can read .024∗∗ -.012 .018 -.0056 .013 .0069(.011) (.023) (.021) (.019) (.04) (.017)
Head of hhd is widow -.0017 .013 .012 .011 -.022 -.0071(.0069) (.028) (.024) (.021) (.035) (.014)
Carded GP .0031 .0054 .019 .0062 .034∗ -.0038(.0036) (.013) (.014) (.011) (.019) (.0056)
District FE Yes Yes Yes Yes Yes Yes
Control Mean .011 .39 .3 .33 .49 .99Avg. number of cases 7385 21349 21349 21349 14456 7255
This table analyzes interaction effects between household or GP characteristics and treatment status for individual non-
response and strictly-positive response rates in private labor market outcomes. In columns 1-4, the outcome in a binary
indicator for whether an a survey response is missing when it should not. “Wage realizations” is the average daily wage (in
Rs.) an individual received while working for someone else in June 2012. “Reservation wage” is an individual’s reservation
wage (in Rs.) at which he or she would have been willing to work for someone else in June 2012. The outcome is based
on an a question in which the surveyor asked the respondent whether he or she would be willing to work for Rs. 20 and
increased this amount in increments of Rs. 5 until the respondent answered affirmatively. “Days worked private sector” is
the number of days an individual worked for somebody else in June 2012. “Days unpaid/idle” is the number of days an
individual stayed idle or did unpaid work in June 2012. Note that every cell in the regression table reports the coefficient of
an interaction term (except “Carded GP”, see below) of the reported variable with the treatment indicator from a separate
regression that includes the raw respective variable, the treatment indicator as well as a vector of mandal characteristics
used to stratify randomization and district fixed effects as covariates. In columns 5-6, we look examine two types of response
patterns. “Days worked private sector > 0” is an indicator for whether an individual worked in the private sector in June
2012. “Wage realizations > Reservation wage” is an indicator for whether an individual’s reported average daily wage was
great than his/her reservation wage. “Above median hhd income” is an indicator for whether an individual belongs to an
household with total annualized income above the sample median. “Hhd is ST, SC or OBC” indicates household members
belonging to Scheduled Castes/Tribes or Other Backward Castes - historically discriminated against section of the population
- while “Hhd below BPL” indicates individuals from household living below the poverty line. that “CardedGP” is a simple
indicator variable (no interaction effect) for whether a household lives in a GP that has moved to Smartcard-based payment,
which usually happens once 40% of beneficiaries have been issued a card. No interaction effect is included because all carded
GPs are in treatment mandals (by experimental design). Finally, note that each column reports results from 7 different
regressions and there is no single number of observations. This table reports the average number of observations across all
regressions in a column. Standard errors clustered at mandal level are in parentheses. Statistical significance is denoted as:∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
50
Tab
leA
.8:
Bas
elin
ebal
ance
inco
mpre
hen
sive
mea
sure
ofnei
ghb
or’s
trea
tmen
t
Wag
ere
aliz
atio
ns
Res
erva
tion
wag
eD
ays
wor
ked
pri
vate
sect
orD
ays
idle
/unpai
dT
otal
inco
me
(1)
(2)
(3)
(4)
(5)
%G
Ps
trea
ted
wit
hin
10km
-.48
-3.2
-.68
.14
5133
(6.8
)(5
.1)
(.64
)(1
)(4
572)
%G
Ps
trea
ted
wit
hin
20km
1.1
-1.8
-.69
.001
736
83(9
.3)
(6.7
)(.
73)
(1.6
)(5
992)
%G
Ps
trea
ted
wit
hin
30km
-1.2
1-.
32-.
86-3
137
(12)
(7.4
)(.
98)
(2.3
)(7
699)
Dis
tric
tF
EY
esY
esY
esY
esY
es(1
)(2
)(3
)(4
)(5
)C
ontr
olM
ean
104
785.
122
4473
2L
evel
Indiv
.In
div
.In
div
.In
div
.H
hd
Inth
ista
ble
we
anal
yze
bas
elin
eb
alan
ceof
key
outc
omes
wit
hre
spec
tto
NR p
for
GP
sin
trea
tmen
tm
an
dals
.E
ach
cell
show
sth
ere
spec
tive
coeffi
cien
tfr
om
ase
para
te
regr
essi
onw
her
eth
eou
tcom
eis
give
nby
the
colu
mn
hea
der
.“W
age
reali
zati
on
s”th
eav
erage
daily
wage
(in
Rs.
)an
ind
ivid
ual
rece
ived
wh
ile
work
ing
for
som
eon
e
else
inJu
ne
2012
.“R
eser
vati
onw
age”
isan
ind
ivid
ual
’sre
serv
ati
on
wage
(in
Rs.
)at
wh
ich
he
or
she
wou
ldh
ave
bee
nw
illi
ng
tow
ork
for
som
eon
eel
sein
Ju
ne
2012.
Th
eou
tcom
eis
bas
edon
ana
qu
esti
onin
whic
hth
esu
rvey
or
ask
edth
ere
spon
den
tw
het
her
he
or
she
wou
ldb
ew
illi
ng
tow
ork
for
Rs.
20
an
din
crea
sed
this
am
ou
nt
inin
crem
ents
ofR
s.5
unti
lth
ere
spon
den
tan
swer
edaffi
rmati
vely
.“D
ays
work
edp
riva
tese
ctor”
isth
enu
mb
erof
day
san
ind
ivid
ual
work
edfo
rso
meb
od
yel
sein
Ju
ne
2012
.“T
otal
inco
me”
isto
tal
annu
aliz
edh
ouse
hol
din
com
e,w
her
eth
eto
p.5
%of
ob
serv
ati
ons
are
sep
ara
tely
trim
med
intr
eatm
ent
an
dco
ntr
ol.
Th
e“%
GP
s
trea
ted
wit
hin
x”
isN
R p,
orth
era
tio
ofth
enu
mb
erof
GP
sin
trea
tmen
tm
an
dals
wit
hin
radiu
sx
km
over
the
tota
lG
Ps
wit
hin
wav
e1,
2or
3m
an
dals
.N
ote
that
wav
e
2m
and
als
are
incl
ud
edin
the
den
omin
ator
,an
dth
atsa
me-
man
dal
GP
sare
incl
ud
edin
both
the
den
om
inato
ran
dnu
mer
ato
r.N
ote
that
each
cell
show
sa
sep
ara
te
regr
essi
onof
the
outc
ome
wit
hth
e“%
GP
str
eate
dw
ith
inx”
an
dd
istr
ict
fixed
effec
tsas
the
on
lyco
vari
ate
s.F
inall
y,n
ote
that
each
colu
mn
rep
ort
sre
sult
sfr
om
7
diff
eren
tre
gres
sion
san
dth
ere
isn
osi
ngl
enu
mb
erof
obse
rvati
on
s.T
his
tab
lere
port
sth
eav
erage
nu
mb
erof
ob
serv
ati
on
sacr
oss
all
regre
ssio
ns
ina
colu
mn
.S
tand
ard
erro
rscl
ust
ered
atm
and
alle
vel
are
inpar
enth
eses
.S
tati
stic
al
sign
ifica
nce
isd
enote
das:∗ p
<0.
10,∗∗p<
0.05,∗∗∗ p
<0.
01.
51
Tab
leA
.9:
Bas
elin
ebal
ance
inex
per
imen
tal
mea
sure
ofnei
ghb
or’s
trea
tmen
t
Wag
ere
aliz
atio
ns
Res
erva
tion
wag
eD
ays
wor
ked
pri
vate
sect
orD
ays
idle
/unpai
dT
otal
inco
me
(1)
(2)
(3)
(4)
(5)
%G
Ps
trea
ted
wit
hin
10km
3.3
1.4
-.27
.34
-377
(4.1
)(3
.3)
(.42
)(.
53)
(287
2)%
GP
str
eate
dw
ithin
20km
22.
4-.
42.1
921
77(6
.6)
(4.8
)(.
64)
(.95
)(4
826)
%G
Ps
trea
ted
wit
hin
30km
1.3
5.6
.026
-.54
-323
2(9
.6)
(6.1
)(.
89)
(1.7
)(6
738)
Dis
tric
tF
EY
esY
esY
esY
esY
es(1
)(2
)(3
)(4
)(5
)C
ontr
olM
ean
104
785.
122
4473
2L
evel
Indiv
.In
div
.In
div
.In
div
.H
hd
Inth
ista
ble
we
anal
yze
bas
elin
eb
alan
ceof
key
outc
omes
wit
hre
spec
tto
NR p
for
GP
sin
trea
tmen
tm
an
dals
.E
ach
cell
show
sth
ere
spec
tive
coeffi
cien
tfr
om
ase
para
te
regr
essi
onw
her
eth
eou
tcom
eis
give
nby
the
colu
mn
hea
der
.“W
age
reali
zati
on
s”th
eav
erage
daily
wage
(in
Rs.
)an
ind
ivid
ual
rece
ived
wh
ile
work
ing
for
som
eon
e
else
inJu
ne
2012
.“R
eser
vati
onw
age”
isan
ind
ivid
ual
’sre
serv
ati
on
wage
(in
Rs.
)at
wh
ich
he
or
she
wou
ldh
ave
bee
nw
illi
ng
tow
ork
for
som
eon
eel
sein
Ju
ne
2012.
Th
eou
tcom
eis
bas
edon
ana
qu
esti
onin
whic
hth
esu
rvey
or
ask
edth
ere
spon
den
tw
het
her
he
or
she
wou
ldb
ew
illi
ng
tow
ork
for
Rs.
20
an
din
crea
sed
this
am
ou
nt
inin
crem
ents
ofR
s.5
unti
lth
ere
spon
den
tan
swer
edaffi
rmati
vely
.“D
ays
work
edp
riva
tese
ctor”
isth
enu
mb
erof
day
san
ind
ivid
ual
work
edfo
rso
meb
od
yel
sein
Ju
ne
2012
.“T
otal
inco
me”
isto
tal
annu
aliz
edh
ouse
hol
din
com
e,w
her
eth
eto
p.5
%of
ob
serv
ati
ons
are
sep
ara
tely
trim
med
intr
eatm
ent
an
dco
ntr
ol.
Th
e“%
GP
s
trea
ted
wit
hin
x”
isN
R p,
orth
era
tio
ofth
enu
mb
erof
GP
sin
trea
tmen
tm
an
dals
wit
hin
rad
ius
xkm
over
the
tota
lG
Ps
wit
hin
wav
e1,
2or
3m
an
dals
.N
ote
that
wav
e2
man
dal
sar
ein
clu
ded
inth
ed
enom
inat
or,
and
that
sam
e-m
an
dal
GP
sare
excl
ud
edin
both
the
den
om
inato
ran
dnu
mer
ato
r.N
ote
that
each
cell
show
sa
sep
arat
ere
gres
sion
ofth
eou
tcom
ew
ithN
R pan
dd
istr
ict
fixed
effec
tsas
the
only
cova
riate
s.F
inall
y,n
ote
that
each
colu
mn
rep
ort
sre
sult
sfr
om
7d
iffer
ent
regre
ssio
ns
and
ther
eis
no
sin
gle
nu
mb
erof
obse
rvat
ion
s.T
his
tab
lere
port
sth
eav
erage
nu
mb
erof
ob
serv
ati
on
sacr
oss
all
regre
ssio
ns
ina
colu
mn
.S
tan
dard
erro
rscl
ust
ered
at
man
dal
level
are
inp
aren
thes
es.
Sta
tist
ical
sign
ifica
nce
isd
enote
das:∗ p
<0.
10,∗∗p<
0.05,∗∗∗ p
<0.
01.
52
Table A.10: Testing for existence of spatial spillovers (comprehensive neighborhoods)
(a) Control
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 14 18 24∗∗ 24∗ 26 1.1 .025 -.94 -3.1 -5.7∗
(16) (15) (11) (13) (16) (2.3) (2.6) (3) (2.9) (2.9)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .05 .05 .05 .05 .05 .07 .06 .07 .07 .07Treatment Mean 128 128 128 128 128 17 17 17 17 17N. of cases 2063 2063 2063 2063 2063 4095 4095 4095 4095 4095% of sample 100 100 100 100 100 100 100 100 100 100
(b) Treatment
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 11 15 11 8.3 6.6 -1.3 -1.8 -3∗ -3.6∗ -4.4∗∗
(8.5) (9.9) (12) (13) (14) (1.3) (1.6) (1.8) (2) (2.2)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .08 .08 .08 .08 .08 .07 .07 .07 .07 .07Control Mean 134 134 134 134 134 16 16 16 16 16N. of cases 5206 5206 5206 5206 5206 10000 10000 10000 10000 10000% of sample 100 100 100 100 100 100 100 100 100 100
(c) Pooled
Wage realizations (Rs.) Days unpaid/idle
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
% GPs treated within R km 9.5 12 10 9.1 10 -.89 -1.5 -2.5 -3.3∗ -4.4∗∗
(7.4) (8.2) (9) (9.8) (11) (1.2) (1.5) (1.6) (1.8) (1.9)Treatment 1.2 1.9 4 5 5.3 -.79 -.74 -.7 -.78 -.79
(5) (4.4) (4.2) (4) (3.8) (.95) (.82) (.72) (.64) (.6)District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj R-squared .07 .07 .07 .07 .07 .07 .07 .07 .07 .07Control Mean 128 128 128 128 128 17 17 17 17 17N. of cases 7269 7269 7269 7269 7269 14095 14095 14095 14095 14095% of sample 100 100 100 100 100 100 100 100 100 100
This table show the impact of comprehensive spatial exposure measure to treatment NRp at radius 10, 15, 20, 25, and 30
kilometers on private wages and days worked from the NREGS household survey. All analysis was conducted separately
for (a) control and (b) treatment, and then for (c) the pooled sample. In each panel, the outcome “Wage realizations” in
columns 1-5 is the average daily wage (in Rs.) an individual received while working for someone else in June 2012 (endline).
The outcome “ Days worked private sector” in columns 6-10 is the number of days an individual worked for somebody else
in June 2012 (endline). The “% GPs treated within R km” is the ratio of the number of GPs in treatment mandals within
a given radius R km over the total GPs within wave 1, 2 or 3 mandals. Note that wave 2 mandals are included in the
denominator, and that same-mandal GPs are included in both the denominator and numerator. The entire GP sample used
in randomization are included. All regressions include the first principal component of a vector of mandal characteristics
used to stratify randomization as a control variable. Standard errors clustered at mandal level are in parentheses. Statistical
significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.
53
Table A.11: Estimating total treatment effects including spillovers (OLS)
(a) Wage
Wage realizations (Rs.) Reservation wage (Rs.)
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
Treatment .69 .75 5.2 8.1 11 5.3 5.3 6.3 6.1 4.4(8.1) (8.5) (8.4) (8.6) (8.9) (5.2) (5.6) (6.4) (6.7) (7)
% GPs treated within R km 8.4 10 13 15 21 18 14 12 6.8 .59(16) (15) (15) (15) (17) (12) (14) (16) (14) (12)
% GPs treated within R km X Treatment 1.5 3.2 -3.3 -7.9 -15 -14 -8.6 -7.4 -4.4 .94(18) (18) (19) (20) (21) (14) (16) (18) (17) (17)
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Total Treatment Effect 11∗ 14∗∗ 15∗ 15∗ 17∗ 9.5∗∗∗ 11∗∗ 11∗ 8.4 5.9SE (5.6) (6.9) (7.5) (8.5) (9.8) (3.1) (4.3) (5.6) (6.2) (6.8)Control Mean 128 128 128 128 128 97 97 97 97 97N. of cases 7269 7269 7269 7269 7269 12852 12852 12852 12852 12852
(b) Labor
Days worked private sector Days idle/unpaid
R = 10 R = 15 R = 20 R = 25 R = 30 R = 10 R = 15 R = 20 R = 25 R = 30
Treatment .26 .56 .42 .59 .82 -.087 -.16 .0037 -.52 -1(1.1) (1.3) (1.5) (1.6) (1.7) (1.2) (1.4) (1.5) (1.5) (1.5)
% GPs treated within R km .92 1.6 2.1 3 4.4 .65 -.38 -1.2 -2.8 -4.8(2.2) (2.6) (3.1) (3.1) (3.3) (2.6) (3.1) (3.4) (3.2) (3.1)
% GPs treated within R km X Treatment -.42 -1.1 -.79 -1.2 -1.8 -2.1 -1.5 -1.8 -.66 .59(2.5) (3) (3.6) (3.8) (4) (2.9) (3.4) (3.8) (3.6) (3.5)
District FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Total Treatment Effect .76 1.1 1.7 2.5 3.5∗ -1.5∗ -2.1∗ -3∗∗ -4∗∗ -5.2∗∗∗
SE (.8) (1.1) (1.4) (1.6) (1.8) (.78) (1.1) (1.4) (1.6) (1.8)Control Mean 7.9 7.9 7.9 7.9 7.9 17 17 17 17 17N. of cases 14441 14441 14441 14441 14441 14095 14095 14095 14095 14095
(c) Total Income
Total Income
R = 10 R = 15 R = 20 R = 25 R = 30
Treatment 11529 13964∗ 17419∗ 12176 6428(7745) (8422) (10497) (10493) (10288)
% GPs treated within R km 27026 29317 31946 17864 5747(17527) (19081) (27257) (23996) (19520)
% GPs treated within R km X Treatment -21689 -22300 -26050 -8955 7178(19453) (21346) (30000) (27636) (24076)
District FE Yes Yes Yes Yes Yes
Total Treatment Effect 16867∗∗∗ 20981∗∗∗ 23315∗∗ 21085∗∗ 19353∗
SE (5118) (6614) (9528) (10663) (10887)Control Mean 71935 71935 71935 71935 71935N. of cases 4903 4903 4903 4903 4903
This table provides estimates for total treatment effects using the comprehensive spatial exposure measure NRp for (a) wage
outcomes, (b) labor outcomes, and (c) total income. Each specification contains a treatment indicator, NRp , and an interaction
between NRp and treatment. Recall NR
p is the ratio of the number of GPs in treatment mandals within a given radius R
km over the total GPs within wave 1, 2 or 3 mandals. The total treatment effect estimate reported in the second section
of the table is the sum of the coefficients for those three variables. “Wage realizations” is the average daily wage (in Rs.)
an individual received while working for someone else in June 2012. “Reservation Wage” is an individual’s reservation wage
(in Rs.) at which he or she would have been willing to work for someone else in June 2012 . The latter outcome is based
on an a question in which the surveyor asked the respondent whether he or she would be willing to work for Rs. 20 and
increased this amount in increments of Rs. 5 until the respondent answered affirmatively. “Days worked private sector” is the
number of days an individual worked for somebody else in June 2012. “Days unpaid/idle” is the sum of days an individual
did unpaid work or stayed idle in June 2012. “Total Income” is total annualized household income (in Rs). All regressions
also include the first principal component of a vector of mandal characteristics used to stratify randomization as a control
variable. Statistical significance is denoted as: ∗p < 0.10, ∗∗p < 0.05, ∗∗∗p < 0.01.54
Figure A.1: Changes in wages by month and treatment status
0
20
40
60
Janu
ary
Feb
ruar
y
Mar
ch
Apr
il
May
June
July
Aug
ust
Sep
tem
ber
Oct
ober
Nov
embe
r
Dec
embe
r
Month
Wag
e
Control Treatment
This figure shows mean changes in agricultural wages between baseline and endline, by month and treatment status, weighted
by (inverse) GP sampling probability. The data, which is at the village-level, comes from surveys administered to prominent
figures in each village. Standard errors are clustered at the mandal level.
55
Figure A.2: Density of spatial measures of treatment exposure
(a) Exposure to Treatment: Including Same Mandal GPs
10km 20km 30km
0
1
2
3
0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00Fraction neighboring GPs treated
Den
sity
Control Treatment
(b) Exogenous Exposure to Treatment: Excluding Same Mandal GPs
10km 20km 30km
0.0
0.5
1.0
1.5
2.0
0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00Fraction neighboring GPs treated
Den
sity
Control Treatment
These figures show smoothed kernel density estimates of each spatial measure of exposure to treatment. The only GPs
included in these density calculations are surveyed GPs. Panel a) shows the distribution of exogenous spatial exposure to
treatment at a given distance for survey GPs. Panel b) shows the distribution of spatial exposure to treatment at a given
distance for survey GPs. The analysis was conducted at distance 10 km, 20 km, and 30 km. The spatial exposure measure is
the ratio of the number of GPs in treatment mandals within radius x km over the total GPs within wave 1, 2 or 3 mandals.
Note that wave 2 mandals are included in the denominator, and that same-mandal GPs are included in both the denominator
and numerator. The exogenous spatial exposure measure is the ratio of the number of GPs in treatment mandals within
radius x km over the total GPs within wave 1, 2 or 3 mandals. Note that wave 2 mandals are included in the denominator,
and that same-mandal GPs are excluded in both the denominator and numerator.
56