Journal of Asian Economics 20 (2009) 456464

Contents lists available at ScienceDirect

Journal of Asian Economics

Capture of anti-poverty programs: An analysis of the National Rural

Employment Guarantee Program in India
Raghbendra Jha a,*, Sambit Bhattacharyya a, Raghav Gaiha b, Shylashri Shankar c
a

ASARC, Arndt-Corden Division of Economics, The Australian National University, Canberra, Australia
University of Delhi, New Delhi, India
c
Centre for Policy Research, New Delhi, India
b

A R T I C L E I N F O

A B S T R A C T

Article history:
Received 14 October 2008
Received in revised form 6 March 2009
Accepted 13 March 2009

Using pooled household level data for the Indian states of Rajasthan and Andhra Pradesh
we nd that the size of landholdings is a negative predictor of participation in the
National Rural Employment Guarantee Program (NREGP). In state level analysis this
pattern survives in Rajasthan but reverses in Andhra Pradesh where we notice a positive
relationship. This paper examines whether this sign reversal in Andhra Pradesh is
indicative of program capture in Andhra Pradesh and better targeting in Rajasthan. We
compare land inequality, ratio of NREG and slack season agricultural wage rates, political
interference, and geographical remoteness across the two states and conclude that
program capture may be an issue in Andhra Pradesh, largely because of these reasons. We
also nd evidence of complementarity between NREGP and the Public Distribution
System (PDS), implying that the real income transfer through food subsidy needs
supplementation.
2009 Elsevier Inc. All rights reserved.

JEL classication:
I38
O12
Keywords:
Capture
Poverty
India

1. Introduction
One of the major challenges in delivering benets of anti-poverty programs to the poor is the threat of program capture by
the non-poor. The non-poor can exercise their economic power and contribute to the campaign funds of the major political
parties in exchange for preferential treatment in welfare services. This represents the phenomenon of capture. In theory,
capture can take place at any level of government.1 The empirical evidence however remains inconclusive as studies suggest
that the relative vulnerability of different levels of government to capture may very well be context specic (Gaiha, Kaushik,
& Kulkarni, 1998, 2000; Tendler, 1997).2

We gratefully acknowledge nancial support from Australian Research CouncilAusAID Linkage grant LP0775444.
* Corresponding author at: Australia South Asia Research Centre, Research School of Pacic & Asian Studies, Australian National University, H.C. Coombs
Building (09), Canberra, ACT 0200, Australia. Tel.: +61 2 6125 2683; fax: +61 2 6125 0443.
E-mail addresses: r.jha@anu.edu.au (R. Jha), sambit.bhattacharyya@anu.edu.au (S. Bhattacharyya), rdg@bol.net.in (R. Gaiha), shylashris@gmail.com
(S. Shankar).
1
Seabright (1996) argues that the likelihood of capture is less at the local level as the accountability is higher in a locally homogeneous community.
Alderman (2002) also nds evidence using Albanian data that local knowledge facilitates effective targeting of the poor. However, Galasso and Ravallion
(2005) disagree and they argue that the assumption of locally homogeneous community is not tenable in a developing country context as distributional
conicts are rife at the local level.
2
Bardhan and Mookherjee (2000, 2005) build on earlier theoretical contributions by Baron (1994) and Grossman and Helpman (1996). The other papers
cited above are empirical in nature. Also see Hartmann and Boyce (1983) and Un Nabi (1999) for anecdotal evidence on capture.
1049-0078/$ see front matter 2009 Elsevier Inc. All rights reserved.
doi:10.1016/j.asieco.2009.03.003

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

457

In a related literature Lanjouw and Ravallion (1999), Ravallion (2000), Gaiha (2000a), Alderman (2002), Galasso and
Ravallion (2005), and others empirically evaluate the likelihood of program capture by the non-poor and the issue of
targeting in general.3 The ndings are inconclusive as they nd targeting effectiveness to be context specic and varying
widely across programs, levels of government, countries, and regions. Lanjouw and Ravallion (1999) also analyse the timing
of program capture and nd that the non-poor enjoy the majority of the benets early on but the share of benets to the poor
increases over time. Gaiha (2000a), however, observes the reverse in a study of the Employment Guarantee Scheme in the
Indian state of Maharashtra where, over time, the program experienced capture by the non-poorespecially because of a
sharp hike in the wage rate.4
In this paper, we examine the extent of capture using household level data on participation in the National Rural
Employment Guarantee Program (NREGP) in the states of Andhra Pradesh (AP) and Rajasthan in India. The NREGP was
instituted pursuant to the National Rural Employment Guarantee Act (NREGA) of 2005. The choice of the states is deliberate
and has the following advantages. First, it allows us to compare the extent of capture between AP a high growth high
poverty elasticity state and Rajasthana low growth low poverty elasticity state (Besley et al., 2005). Second, it also allows
us to compare the extent of capture in a northern state (Rajasthan) with that in a southern state (AP). Our results show that in
a combined sample the size of landholdings is a negative predictor of participation in NREGP. This pattern survives in
Rajasthan. However, the pattern reverses in AP where we notice a positive relationship. This may be due to program capture
in AP and its relative absence in Rajasthan. We posit that differences in land inequality, ratio of NREGP and agricultural wage
rates, and geographical remoteness across the two states may help explain program capture in AP. The sample villages in AP
have greater inequality in terms of landholdings,5 are more geographically remote, and experience more political
interference compared to the sample villages in Rajasthan. We also nd evidence of complementarity between NREGP and
the Public Distribution System (PDS), implying that inadequacy of benets from the PDS induces the poor to seek benets
from other anti-poverty programs (e.g., NREGP).6 This result stands in contrast to that of Galasso and Ravallion (2005) who
report evidence of substitution between programs in Bangladesh. They notice that the poor receive relatively less benet
from Food-for-Education in villages in which alternative transfers to the poor are already in place.
The remainder of the paper is organized as follows. In Section 2, we discuss the background on NREGP. Section 3
introduces the data and the estimation approach. Section 4 reports our results and Section 5 concludes.
2. Background on NREGP
The National Rural Employment Guarantee Act (NREGA) of 2005 is perhaps the most signicant social policy initiative in
India in the last decade. The NREGA states that,
[its main objective is] to provide enhancement of livelihood security of the households in rural areas of the country by
providing at least 100 days of guaranteed wage employment to every household in unskilled manual work. (Ministry
of Law & Justice, 2005)
This commitment is clearly a landmark event in the history of rural development policies in India. During 200607,
NREGP involved an expenditure of Rs 88 billion and generated a little less than one billion person days of employment
(Ministry of Rural Development, Government of India7). The NREGPs performance is also crucial to the success of the
Millennium Development Goal of halving global poverty by 2015 (compared to 1990 levels) as rapid reduction in poverty in
India will have an important bearing on the global poverty numbers. Recent gures show that poverty in India has declined,
albeit slowly, over the period 199305 (Himanshu, 2007).8 However, the challenge is to sustain and improve this trend.
Perhaps it is fair to say that a considerable amount depends on the success of NREGP.
There is often a presumption that participation in NREGP is considerably lower than expected and therefore the extension
of the program to cover all of rural India with effect from 1 April 2008, is not likely to make a signicant dent in rural poverty.
As awareness of NREGP has grown, the number of participants has risen sharplyespecially those who completed 100 days
of employment. The total employment generated under the NREGP is clearly much larger than earlier employment
programs, for example, Sampoorna Grameen Rozgar Yojana (SGRY) and National Food for Work Program (NFFWP). These
programs together generated 748 million person days in 200203 and 856 million in 200304 (Mehrotra, 2008). Under the
NREGP, the gure was 905 million in 200607 for only 200 districts and 1437 million in 200708, in part reecting the
expansion to 330 districts and better awareness. In fact, the number who completed 100 days of work rose from 2.1 million

See Conning and Kevane (2000) for a survey of these results.

In a recent contribution, Scandizzo et al. (in press) use a dynamic optimization model taking into account a xed wage rate and certainty of employment
under the EGS and a stochastic wage rate under other activities. Under such conditions, using ICRISAT panel data for selected villages in Maharashtra, it is
demonstrated that the higher the EGS wage the greater is its attractiveness to relatively skilled and afuent workers, and for those already in it to continue.
5
Galasso and Ravallion (2005) report similar results in the context of Bangladeshs Food-for-Education (FFE) Program. They nd targeting improves with
low levels of land inequality. Baland and Platteau (1997), Dayton-Johnson (2000), and La Ferrara (2002) also look at the effect of inequality on participation
in local groups and the management of common property resources.
6
PDS is a system of public distribution of food grains established in 1951 by the Government of India to promote food security and poverty alleviation.
7
http://nrega.nic.in/states/nregampr.asp (accessed on March 4, 2009).
8
See also Gaiha (2008) for a further corroboration, based on the new World Bank poverty estimates.
4

458

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

Table 1
Descriptive statistics.
Variable

Number of obs.

Mean

Standard deviation

Minimum

Maximum

Participation in NREG (NREGi)

Agricultural land holdings (LNDOWNi) (bighas, 1 bigha = 0.16 ha)
PDS participationfood grain (FGPDSi)
PDS participationsugar (SPDSi)
PDS participationkerosene (KPDSi)
Total household size (TOTHSi)

942
942
941
941
941
942

0.60
2.83
0.69
0.68
0.93
4.42

0.49
4.52
0.46
0.47
0.25
2.03

0
0
0
0
0
1

1
50
1
1
1
17

Source: Authors calculations.

(10% of all participating households) in 200607 to 3.5 million (11% of all households) in 200708. Chhattisgarh, Madhya
Pradesh (MP), AP, and Rajasthan all poor states have the highest number of households completing 100 days of work
(Mehrotra, 2008).
A controversial aspect of NREGP is the provision of employment on demand at the minimum wage rate. As argued
elsewhere, given that the slack season agricultural wage rates are typically way below the minimum wage rates, there is a
strong incentive for the (relatively) afuent to masquerade as poor. As a result, the self-selection mechanism of workfare is
weakened and the poor are crowded out.9
3. Data and estimation approach
The present analysis draws upon data drawn from the states of Rajasthan and AP to assess the effectiveness of social
safety nets, in particular NREGP and PDS. These data were collected as part of a larger study of the impact of the PDS and
NREGP in three Indian states: Andhra Pradesh, Rajasthan and Maharashtra. The analysis reported below is based on
(preliminary) data10 obtained from Udaipur district in Rajasthan and Chittoor district in AP.
First, a list of NREGP districts was compiled for each state. From these districts, three were selected on the basis of
probability proportional to size (in this case, rural population as reported in the 2001 Census). In the next step, three villages
were randomly selected from each district, followed by a random selection of households. The villages considered for the
present analysis are Dhundiya, Karanpur and Prithvisingh Ji Ka Khera in Udaipur district (Rajasthan) and Kaligiri,
Obulattapale and Reddivaripalle in Chittoor district (AP). The total number of households interviewed in December 2007 was
942 with 340 from Rajasthan and 602 from AP. The data includes information on caste, occupation, landholdings, household
size, NREG participation, type of ration card, and PDS participation.11 Table 1 presents summary statistics for the key
variables of interest. Tables 2 and 3 provide further details of participation in NREGP in the sample villages of Rajasthan and
Andhra Pradesh.
An important point is that the data used for this analysis are not representative of the two states in question. Rather, the
villages selected are representative of each district. The complete data for all sample districts in each state will be
representative of each sample state. Although a more extensive coverage would have been more representative, our
presumption is that both within and between variation in participation in these major anti-poverty interventions in the three
states selected have considerable policy relevance.
To analyse the extent of capture we estimate the following binary response model
NERGi a bLNDOWNi g PDSi Xi F ei

(1)

where NERGi = 1 if household i is a participant in NREGP and 0 otherwise, LNDOWNi is the total landholdings of household
i measured in hectares, PDSi = 1 if household i is a participant in PDS and 0 otherwise, Xi is a vector of all other
determinants of participation in NREGP including caste xed effects, occupation xed effects, and total household size,
and ei is the random error term. The impact of changes in the independent variables on the probability of NREGP
participation is estimated by assuming a logistic distribution. Therefore, the coefcients b, g and F indicate the impact of
a change in the corresponding independent variable on the natural log of odds of participation in NREG.12 We are
interested in the sign and estimate of the coefcients b and g. If b < 0 then the odds of participation in NREGP decline with
size of landholdings, which will lead us to conclude that the project is reasonably well targeted and relatively free from

See, for example, Gaiha (2000a).

In a companion paper Jha, Gaiha and Shankar (2008) present summary statistics and some results on Rajasthan (but do not deal with program capture).
These results are preliminary in the sense that these are based on data on households collected from these villages but not on surveys of the households
themselves, which will be conducted in due course. That survey uses more detailed questionnaires and would facilitate a richer analysis of the targeting and
the transfer and stabilization benets of NREGP to households but is unlikely to alter the conclusions of the present paper. We are grateful to the editor-inchief for raising this issue.
11
NREG participation is measured using the question are you a beneciary of NREGP? PDS participation is measured using the questions whether the
household draws food grain from PDS, whether the household draws sugar from PDS, whether the household draws kerosene from PDS?
12
Odds are dened as the ratio of probabilities of participation and non-participation in NREGP.
10

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

459

Table 2
Frequency of participation in NREGP by caste/ethnic group, occupation, land-owned and household size in Rajasthan and Andhra Pradesh.
Rajasthan
No

Andhra Pradesh
Yes

Total

No

Yes

Total

Participation by caste/ethnic group

OT
205
SC
18
ST
4

102
6
5

307
24
9

201
78
14

169
99
41

370
177
55

Total

227

113

340

293

309

602

Participation by occupation
AL
4
OL
28
OT
16
SA
126
SN
53

1
46
3
52
11

5
74
19
178
64

135
20
34
95
9

227
0
15
65
2

362
20
49
160
11

227

113

340

293

309

602

Participation landowned (ha)

00.1
28
0.10.75
56
0.751.5
75
1.52.5
39
>2
29

30
28
33
17
5

58
84
108
56
34

120
70
70
25
8

154
82
66
5
2

274
152
136
30
10

227

113

340

293

309

602

Participation by household size

13
33
45
87
>5
107

25
41
47

58
128
154

137
120
36

120
153
36

257
273
72

113

340

293

309

602

Total

Total

Total

227

Source: Authors calculations.

capture by the non-poor. However, if we observe b > 0 then the reverse would be true. The sign and estimate of the
coefcient g would indicate complementarity or substitutability between the two programs. A positive sign would
indicate the former and a negative sign the latter.
An issue is whether real income transfer through a food subsidy or through a form of (guaranteed employment) is
adequate in itself or requires supplementation. In the case of substitutes, it could be argued that the transfer is adequate
either way while in the case of complements it is not adequate.
Obtaining consistent estimates of b and g requires overcoming the following challenges. First, it can very well be the
case that the estimates of b and g are driven by unobserved heterogeneity (culture, or some other factors) or observed
xed effects such as caste and religion which inuences landownership, participation in PDS, and participation in
NREGP. To tackle this potential source of bias we control for caste and occupation xed effects.13 Second, the estimates
may also suffer from endogeneity bias. NREGP participation may inuence participation in PDS rather than the other
way round. If this is the case the coefcient estimates will be biased away from zero. However, the chances of this
occurring may be low as PDS is an old program (established in 1951) and NREG is a new program (established in 2005).
Nevertheless, we formally test for exogeneity of PDS participation using a Hausman style test and we fail to reject the
null.14 As a test of robustness, we also use predicted probabilities of PDS participation as an instrument and estimate the
model using maximum likelihood methods.15 The estimates are very similar to our preferred model and are not
reported here because of space constraints.

13
Caste xed effects are four caste dummies indicating backward caste, schedule cast, schedule tribe, and other caste. Others is the omitted category.
Occupation xed effects are four occupation dummies indicating agricultural labor, self-employed agriculture, self-employed non-agriculture, and other
labor with others as the omitted category.
14
The test follows a Hausman style two step procedure. First, it generates the predicted error from running PDS participation on all other exogenous
variables (excluding NREG participation). Then it runs Eq. (1) with the predicted error as an additional control and performs a t-test on it. A test of this nature
is valid since under the null of exogeneity, the distribution of the predicted error plays no role. Therefore there is no need to assume normality which
increases the power of the test (see Grogger, 1990; Wooldridge, 2002, p. 474). We perform this test using both non-standardized and Pearsons standardized
residuals. The results are qualitatively identical. Only results using non-standardized residuals are reported here to save space.
15
The predicted probabilities are generated by running PDS participation on all other exogenous factors (caste, occupation, landownership, household
size, type of ration card) except NREGP that may inuence PDS participation.

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

460

Table 3
Characteristics of duration of participation in NREGP in Rajasthan and Andhra Pradesh.
Rajasthan
0 days

Andhra Pradesh
>90 days

Total

0 days

150 days

5190 days

>90 days

Total

Duration of participation (aggregate numbers)

N
227
0
0
Y
0
46
44

0
23

227
113

293
0

0
179

0
87

0
43

293
309

Total

23

340

293

179

87

43

602

Duration of participation by caste/ethnic groups

OT
205
39
40
SC
18
4
2
ST
4
3
2

23
0
0

307
24
9

201
78
14

97
54
28

41
39
7

31
6
6

370
177
55

Total

23

340

293

179

87

43

602

Duration of participation in NREGP by occupation

AL
4
1
0
OL
28
13
16
OT
16
3
0
SA
126
22
25
SN
53
7
3

0
17
0
5
1

5
74
19
178
64

135
20
34
95
9

131
0
11
36
1

65
0
4
17
1

31
0
0
12
0

362
20
49
160
11

Total

23

340

293

179

87

43

602

Duration of participation in NREGP by landowned (ha)

00.1
28
7
10
0.10.75
56
14
8
0.751.5
75
15
16
1.52.5
39
5
10
>2.5
29
5
0

13
6
2
2
0

58
84
108
56
34

120
70
70
25
8

88
49
37
4
1

43
23
19
1
1

23
10
10
0
0

274
152
136
30
10

Total

23

340

293

179

87

43

602

227

227

227

227

150 days

46

46

46

46

5190 days

44

44

44

44

Source: Authors calculations.

4. Evidence
Table 4 presents regression estimates of Eq. (1) using a pooled sample of AP and Rajasthan. Column 1 reports estimates when
foodgrains measure of PDS participation is used. This is our measure of choice as it is less likely that the non-poor would draw
foodgrains from PDS because of their low quality.16 Hence this measure is our best hope of capturing the participation of the
poor in PDS. We nd that agricultural landholding is negatively related to the odds of NREGP participation. A one standard
deviation increase in landholdings (4.5 ha) reduces the odds of NREGP participation by 1.3-fold. To put this into perspective, the
difference in marginal effects of a 4.5 ha increase in landholdings between a landless respondent and a respondent with 20 ha
landholdings is approximately 100% with a larger effect for the landless. This indicates that NREGP as a program is reasonably
well targeted and there is little evidence of capture in the pooled sample. We nd evidence that PDS and NREGP participation are
complementary with the former positively impacting the latter. The magnitude of the effect is also larger than the effect of
landholdings with an impact coefcient17 of 1.8-fold. This is in contrast to Galasso and Ravallion (2005) who nd evidence of
substitutability between anti-poverty programs in Bangladesh. We also test for the exogeneity of PDS and fail to reject the null.
The model predicts approximately 75% of the outcomes correctly. In column 2 we use the sugar measure of PDS participation.
This may not be the best measure of PDS participation as many non-poor households choose to draw sugar from PDS in order to
retain their ration card (which is also an important ID) active. Nevertheless, the negative coefcient on landholdings survives,
with the magnitude increasing marginally. We also nd evidence of complementarity between NREGP and PDS. In column 3, we
replace sugar measure with kerosene measure of PDS. These results-especially the (robust) complementarity between NREGP
and PDS (foodgrains)throw new light on the inadequacy of income transfers from the PDS and the need to supplement it with
additional income from NREGP.18
Table 5 reports estimates of the model using the Rajasthan sample. We notice that landholdings continue to negatively
inuence the odds of participation in NREGP. Furthermore, the impact coefcient marginally increases to 1.4-fold when the
foodgrain measure of PDS is used (see column 1).19 There is also evidence of complementarity between NREGP and PDS

16
However there is a possibility that the non-poor could draw food grains from PDS and use them to pay for the services of their servants. Our measure
does not rule out this possibility.
17
Impact coefcient is the effect of one standard deviation increase in PDS participation on the odds of NREG participation.
18
Results on tests for robustness of the results are available with the corresponding author.
19
This is calculated by using sample standard deviation of landholdings in Rajasthan which is 6.14 ha.

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

461

Table 4
Capture of anti-poverty programs: the case of NREGP in Andhra Pradesh and Rajasthan.
Dependent variable
Participation in NREG (NREGi)

Independent variables
Agricultural land holdings (LNDOWNi)
PDS participationfood grain (FGPDSi)
PDS participationsugar (SPDSi)
PDS participationkerosene (KPDSi)
Other controls
Total household size
Caste xed effects
Occupation xed effects
Ration card dummies
Number of observations
Percent correctly predicted
Log-likelihood value
Pseudo R2
Exogeneity test (p-value)

Joint participation in NREG and PDS

(1) Logit
(MLE)

(2) Logit
(MLE)

(3) Logit
(MLE)

(4) Food
grain logit
(MLE)

(5) Sugar
logit (MLE)

(6) Kerosene
logit (MLE)

0.06*** (0.0247)
1.24*** (0.2018)

0.07*** (0.0251)

0.11*** (0.0273)

0.10** (0.0446)

0.09** (0.0442)

0.06** (0.0267)

Yes
Yes
Yes
Yes
939
82.8
327.46
0.49

Yes
Yes
Yes
Yes
939
83.2
328.62
0.49

Yes
Yes
Yes
Yes
939
78.5
443.53
0.31

1.27*** (0.2009)
0.55* (0.3296)

Yes
Yes
Yes
No
940
75.2
476.98
0.25
0.68

Yes
Yes
Yes
No
940
75.4
475.18
0.25
0.98

Yes
Yes
Yes
No
940
72.9
496.65
0.21
0.16

Notes: Figures in the parentheses are cluster standard errors and they are robust to arbitrary heteroskedasticity and arbitrary intra-group correlation. All the
regressions reported above are carried out with an intercept. Caste xed effects are four caste dummies indicating backward caste, schedule cast, schedule
tribe, and other caste. Others are the omitted category. Occupation xed effects are four occupation dummies indicating agricultural labor, self-employed
agriculture, self-employed non-agriculture, and other labor with others as the omitted category. Ration card dummies are BPL, APL, and Annapurna with no
card being the omitted category. The exogeneity test follows a Hausman style two step procedure. First, it generates the predicted error from running PDS
participation on all other exogenous variables (excluding NREG participation). Then it runs Eq. (1) with the predicted error as an additional control and
performs a t-test on it.
Source: Authors calculations.
*
Signicance level of 10% against a two sided alternative.
**
Signicance level of 5% against a two sided alternative.
***
Signicance level of 1% against a two sided alternative.

Table 5
Capture of anti-poverty programs: the case of NREGP in Rajasthan.
Dependent variable
Participation in NREG (NREGi)

Independent variables
Agricultural land holdings (LNDOWNi)
PDS participationfood grain (FGPDSi)
PDS participationsugar (SPDSi)
PDS participationkerosene (KPDSi)
Other controls
Total household size
Caste xed effects
Occupation xed effects
Ration card dummies
Number of observations
Percent correctly predicted
Log-likelihood value
Pseudo R2
Exogeneity test (p-value)

Joint participation in NREG and PDS

(1) Logit
(MLE)

(2) Logit
(MLE)

(3) Logit
(MLE)

(4) Food
grain logit
(MLE)

(5) Sugar
logit (MLE)

(6) Kerosene
logit (MLE)

0.05* (0.0307)
1.15*** (0.2910)

0.05* (0.0308)

0.06** (0.0294)

0.02 (0.0529)

0.02 (0.0542)

0.05* (0.0284)

Yes
Yes
Yes
Yes
337
87.2
77.43
0.51

Yes
Yes
Yes
Yes
337
87.5
76.30
0.52

Yes
Yes
Yes
Yes
337
72.1
177.48
0.17

1.16*** (0.2849)
1.94 (3.012)

Yes
Yes
Yes
No
338
71.6
180.98
0.15
0.89

Yes
Yes
Yes
No
338
71.3
180.70
0.16
0.57

Yes
Yes
Yes
No
337
73.0
189.41
0.11
0.72

Notes: Figures in the parentheses are cluster standard errors and they are robust to arbitrary heteroskedasticity and arbitrary intra-group correlation. All the
regressions reported above are carried out with an intercept. Caste xed effects are four caste dummies indicating backward caste, schedule cast, schedule
tribe, and other caste. Others are the omitted category. Occupation xed effects are four occupation dummies indicating agricultural labor, self-employed
agriculture, self-employed non-agriculture, and other labor with others as the omitted category. Ration card dummies are BPL, APL, and Annapurna with no
card being the omitted category. The exogeneity test follows a Hausman style two step procedure. First, it generates the predicted error from running PDS
participation on all other exogenous variables (excluding NREGP participation). Then it runs Eq. (1) with the predicted error as an additional control and
performs a t-test on it.
*
Signicance level of 10% against a two sided alternative.
**
Signicance level of 5% against a two sided alternative.
***
Signicance level of 1% against a two sided alternative.

462

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

Table 6
Capture of anti-poverty programs: the case of NREGP in Andhra Pradesh.
Dependent variable
Participation in NREG (NREGi)

Independent variables
Agricultural land holdings (LNDOWNi)
PDS participationfood grain (FGPDSi)
PDS participationsugar (SPDSi)
PDS participationkerosene (KPDSi)
Other controls
Total household size
Caste xed effects
Occupation xed effects
Ration card dummies
Number of observations
Percent correctly predicted
Log-likelihood value
Pseudo R2
Exogeneity test (p-value)

Joint participation in NREG and PDS

(1) Logit
(MLE)

(2) Logit
(MLE)

(3) Logit
(MLE)

(4) Food
grain logit
(MLE)

(5) Sugar
logit (MLE)

(6) Kerosene
logit (MLE)

0.22** (0.0884)
1.11*** (0.3273)

0.21** (0.0882)

0.23** (0.0889)

0.11 (0.0971)

0.10 (0.0973)

0.12 (0.0993)

Yes
Yes
Yes
Yes
602
84.2
218.95
0.39

Yes
Yes
Yes
Yes
602
83.6
222.22
0.39

Yes
Yes
Yes
Yes
602
84.6
219.52
0.39

1.19*** (0.3103)
1.20*** (0.3249)

Yes
Yes
Yes
No
602
81.2
252.51
0.25
0.89

Yes
Yes
Yes
No
602
81.6
250.77
0.25
0.45

Yes
Yes
Yes
No
602
81.1
251.75
0.25
0.56

Notes: Figures in the parentheses are cluster standard errors and they are robust to arbitrary heteroskedasticity and arbitrary intra-group correlation. All the
regressions reported above are carried out with an intercept. Caste xed effects are four caste dummies indicating backward caste, schedule cast, schedule
tribe, and other caste. Others are the omitted category. Occupation xed effects are four occupation dummies indicating agricultural labor, self-employed
agriculture, self-employed non-agriculture, and other labor with others as the omitted category. Ration card dummies are BPL, APL, and Annapurna with no
card being the omitted category. The exogeneity test follows a Hausman style two step procedure. First, it generates the predicted error from running PDS
participation on all other exogenous variables (excluding NREG participation). Then it runs Eq. (1) with the predicted error as an additional control and
performs a t-test on it.
*
Signicance level of 10% against a two sided alternative.
**
Signicance level of 5% against a two sided alternative.
***
Signicance level of 1% against a two sided alternative.

participation in Rajasthan (columns 1 and 2), conrming inadequacy of food subsidy implicit in the PDS. Overall, Rajasthan
seems to conform to the trends of the pooled sample. NREGP in Rajasthan is reasonably well targeted in the sense that
expansion of this intervention would benet the (relatively) poor/landless more than others.
These trends, however, reverse in AP. The results are reported in Table 6. We nd a positive relationship between
landholdings and NREGP participation. One standard deviation (1.1 ha) increase in landholdings increases the odds of NREGP
participation by 1.3-fold. This is indicative of program capture in AP. Given that AP is a high growth and high poverty
elasticity state (Besley et al., 2005), this result is somewhat puzzling. In fact one would expect Rajasthan to show this pattern
since it is a low growth and low poverty elasticity state. However, the complementarity between PDS and NREGP
participation survives in AP as well as indicated by the rst three columns of Table 6. The answer to this puzzle may rest with
differences in inequality of land distribution, ratio of NREGP and agricultural wage rates, distribution of political power in the
village, geographical remoteness of a village, and access to information. We analyse these factors in turn.
First, high level of land inequality at the village level may translate into higher incidence of capture because individuals
with larger landholdings are observed to have more inuence over village-level decision-making (Bardhan & Mookherjee,
2000; Galasso & Ravallion, 2005). The land Gini and coefcient of variation for Rajasthan (0.50 and 1.05, respectively) are
lower than corresponding values for AP (0.66 and 1.47, respectively) indicating that land is more unequally distributed in AP
than Rajasthan. This may explain why NREGP is better targeted in Rajasthan as compared to AP and also experiences less
capture. This nding is consistent with Galasso and Ravallion (2005) who also report that more unequal villages in
Bangladesh are worse at targeting the poor.
Another signicant contributory factor is the magnitude of the ratio of NREGP to agricultural wage rates. As shown in
Scandizzo, Gaiha, and Imai (in press) the higher the Employment Guarantee Scheme (EGS) wage rate to agricultural wage
rate, the greater will be the share of (relatively) afuent participants, and capture. Some estimates show that the NREGP
wage rate was more than twice as high in Andhra Pradesh, as compared to market wage for male agricultural laborers (Rs 80
in NREGP and about Rs 37 for male agricultural laborers), as against well below two in Rajasthan (Rs 73 in NREGP and about
Rs 44 for male agricultural laborers) (Mehrotra, 2008). Going by this evidence alone, it is arguable that both the deterrent and
screening arguments favouring self-selection of the poor would be undermined in AP (Gaiha, 2000a, 2000b).
Second, if a social policy intervention is neutral to the distribution of power in the village, redistribution is less likely to
experience capture since the village elite may not feel challenged. The reverse may occur when the intervention alters the
distribution of power. Our qualitative data in Rajasthan shows that the introduction of NREGP did not alter the distribution of
power within the village, but altered the distribution of power between the village head (Sarpanch), the block head
(Pradhan) and the constituency head (MLA). The qualitative survey also shows that within the village power is still

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

463

concentrated in the hands of the upper castes and landed who are historically the power centers.20 Therefore, this is
consistent with our result that NREGP is better targeted in Rajasthan. In AP too, the NREGP did not alter the existing
distribution of power. However, there are differences in the involvement of politicians in the NREGP across the two states.
This forms part of our explanation for greater capture in AP. But there is some subtlety to the argument.
Both Rajasthan and AP have similar competitive two party set-ups (the BJP and the Congress are the two main political
parties in Rajasthan while the Congress and the Telgu Desam Party are the two main political parties in AP).21 However, there
is one signicant difference across the two states in our qualitative data. Local politicians in AP are considerably more
involved in NREGP than they are in Rajasthan. They use social policy programs as an opportunity for advancement for their
followers and encourage capture.22 As one Block Development Ofcer (BDO) in AP put it, the political interference in the
appointments process makes it very difcult to terminate corrupt Field Assistants after their selection. This forms part of our
explanation for why NREGP is experiencing capture in AP.
Third, the likelihood of program capture is relatively high in more isolated villages (Galasso & Ravallion, 2005). Our
sample villages in Rajasthan are about 30 min to an hour away from Udaipurone of the major cities in Rajasthan. This could
be a reason for low capture in Rajasthan. Interviews with landowners and NREGP beneciaries in the villages also support
this explanation. Our NREGP beneciary respondents indicated that men preferred to work in the city rather than dig earth in
the heat in the village. They sent their wives to work on NREGP projects instead. Interviews with large landowners (those
owning 2.44.8 ha) in Karanpur (one of the sample villages in Rajasthan) revealed that these landowners could not get
enough unskilled and skilled labor and had to hike up the market wages for both sets of workers. This is in stark contrast to
the sample villages in AP, which are remote.
Fourth, lower informational constraints are expected to produce lower capture of benets by the non-poor.
Informational constraints could occur because of cultural reasons and/or because of elite capture of the decisionmaking process. The Indian parliament enacted a Right to Information Act (RTI) in 2005, which allows a citizen to ask for
and get information on the activities of state ofcials. We take the use of RTI as a proxy for lower informational
constraints. Our qualitative data shows that the usage of RTI to get information about NREGP is almost non-existent
in Rajasthan even though the campaign for RTI was spearheaded by an organization (MKSS) in Rajasthan. In AP, on
the other hand, in one of the sample villages, an individual led a petition under RTI and received information on
NREGP. One person in Reddivaripalli village used the RTI to request information on the muster rolls and material
payments made in NREGP projects. An enquiry was launched and concluded that the payments had reached the
right persons. Our qualitative interviews with the NREGP beneciaries in both states indicate that the use or non-use
of RTI is not a good proxy for determining the extent of informational constraints. In all the six villages in our
sample, the beneciaries knew the amount and the work that they were entitled to do under NREGP. The beneciaries
said that they found such information from other villagers and the sarpanch (chairperson of village Panchayat).23
Hence, there does not appear to be much variation across sample villages in the two states in respect to access to
information.
5. Concluding remarks
In this paper we examined the extent of capture using household level data on NREGP participation in the Indian
states of AP and Rajasthan. Our results show that in a combined sample the size of landholdings is a negative predictor
of participation in NREGP. This pattern survives in Rajasthan. However, the pattern reverses in AP where we notice a
positive relationship. This indicates that the program is experiencing capture in AP whereas it is reasonably well
targeted in Rajasthan. We explain this difference as arising due to a number of factors including the varied level of land
inequality in the two states, with AP more unequal than Rajasthan, as also the higher ratio of NREGP to agricultural
wage rate. Other contributory factors include the geographical remoteness of the villages in AP, and relatively greater
political interference in NREGP at the village level in AP. We also nd evidence of complementarity between NREGA and
the PDS, in the pooled sample as well as in the samples of the individual states, implying the inadequacy of food subsidy
in PDS.
The results contribute to a growing literature on capture of anti-poverty programs and support the theoretical
observation that the extent of capture is context specic. The ndings vary across region, country, and also the nature of
programs. Therefore, there is a need for more empirical research on what context leads to what outcome in terms of capture.

20

For a detailed account of the qualitative survey in Rajasthan and AP, see Shankar (2008).
Hence the argument of Kohli (2001) that highly competitive political systems are less likely to experience capture is not relevant. Further, our
qualitative analysis shows that in all six villages surveyed in Rajasthan and AP the main electoral competition was between the candidates sponsored by the
two main competing parties in the state. Hence, the argument of Chhibber and Nooruddin (2004) that politicians who work through political parties
engaged in two party competition are more likely to provide public goods than those who have to make decisions within the context of multi-party
environment, is also not relevant to our discussion.
22
This is also consistent with the qualitative evidence presented by Powis (2007).
23
Two observations are pertinent: one is that awareness of NREGP has grown rapidly but the supporting administrative machinery and record keeping
continue to be weak; and the second is that conversion of piece wage rates into minimum wage rates is far from straightforward (Ambasta, Shankar, & Shan,
2008; Dreze, Khera, & Siddharth, 2007; Mathur, 2007; Mehrotra, 2008).
21

464

R. Jha et al. / Journal of Asian Economics 20 (2009) 456464

Acknowledgements
R. Bhatia and P.K. Roy helped with the design and implementation of household surveys. We would also like to
acknowledge our debt to an anonymous reviewer for helpful and constructive suggestions on an earlier version and the
editor-in-chief for helpful comments and encouragement. The usual caveat applies.
Appendix A
Participation in NREG (NREGi): NREGi = 1 if household i is a participant in NREG and 0 otherwise.
Agricultural Land Holdings (LNDOWNi): Measured in hectares.
PDS participationfood grain (FGPDSi): FGPDSi = 1 if household i draws food grain from PDS and 0 otherwise.
PDS participationsugar (SPDSi): SPDSi = 1 if household i draws sugar from PDS and 0 otherwise.
PDS participationkerosene (KPDSi): KPDSi = 1 if household i draws kerosene from PDS and 0 otherwise.
Caste xed effects: Caste dummies signifying backward caste, schedule cast, schedule tribe, and other caste with others being
the omitted category.
Occupation xed effects: Occupation dummies signifying agricultural labor, self-employed non-agriculture, self-employed
agriculture, and other labor with others being the omitted category.
Ration card dummies: Ration card dummies are below poverty line (BPL) card, above poverty line (APL) card, and Annapurna
with no card being the omitted category.
Nnon-participant in NREGP; Yparticipant in NREGP; SCscheduled caste; STscheduled tribe; OTOthers; AL
agricultural labor; OLother labor; OTOthers; SAself-employed in agriculture; SNself-employed in non-agriculture.
References
Alderman, H. (2002). Do local ofcials know something we dont? Decentralization of targeted transfers in Albania. Journal of Public Economics, 83(3), 375
404.
Ambasta, P., Shankar, P. S. V., & Shah, M. (2008). Two years of NREGA: The road ahead. Economic and Political Weekly, 43(08), 4150.
Baland, J. M., & Platteau, J. (1997). Wealth inequality and efciency in the commons. I. The unregulated case. Oxford Economic Papers, 49, 451482.
Bardhan, P., & Mookherjee, D. (2000). Capture and governance at local and national levels. American Economic Review, 90(2), 135139.
Bardhan, P., & Mookherjee, D. (2005). Decentralizing antipoverty program delivery in developing countries. Journal of Public Economics, 89, 675704.
Baron, D. (1994). Electoral competition with informed and uninformed voters. American Political Science Review, 88, 3347.
Besley, T., Burgess, R., & Esteve-Volart, B. (2005). Operationalising pro-poor growth: India case study. Department of Economics, London School of Economics.
(mimeo).
Chhibber, P., & Nooruddin, I. (2004). Do party systems count? The number of parties and government performance in the Indian states. Comparative Political
Studies, 37(2), 152187.
Conning, J., & Kevane, M. (2000). Community based targeting mechanism for social safety nets. Department of Economics, Williams College. (mimeo).
Dayton-Johnson, J. (2000). Determinants of collective action on the local commons: A model with evidence from Mexico. Journal of Development Economics, 62,
181208.
Dreze, J., R. Khera and Siddharth (2007) NREGA in Orissa: Ten Loopholes and the Silver Lining (mimeo).
Gaiha, R. (2000a). On the targeting of the employment guarantee scheme in the Indian state of Maharashtra. Economics of Planning, 33, 203219.
Gaiha, R. (2000b). Do anti-poverty programs reach the rural poor in India? Oxford Development Studies, 28(1), 7195.
Gaiha, R. (2008). Poverty of statistics. Outlook, 15 September.
Gaiha, R., Kaushik, P., & Kulkarni, V. (1998). Jawahar Rozgar Yojana, Panchayats, and the rural poor. Asian Survey, 38, 928949.
Gaiha, R., Kaushik, P., & Kulkarni, V. (2000). Participation or empowerment of the rural poor: The case of the Panchayats in India. In V. Damodaran & M. UnnithanKumar (Eds.), Post colonial India. Delhi: Manohar.
Galasso, E., & Ravallion, M. (2005). Decentralized targeting of an antipoverty program. Journal of Public Economics, 89, 705727.
Grogger, J. (1990). A simple test for exogeneity in probit, logit, and poisson regression models. Economics Letters, 33, 329332.
Grossman, G., & Helpman, E. (1996). Electoral competition and special interest politics. Review of Economic Studies, 63, 265286.
Hartmann, B., & Boyce, J. (1983). A quiet violence: Voices from a Bangladesh village. London: Zed Press.
Himanshu. (2007). Recent trends in poverty and inequality: Some preliminary results. Economic and Political Weekly, 42, 497508.
Jha, R., Gaiha, R., & Shankar, S. (2008). Reviewing the National Rural Employment Guarantee Programme. Economic and Political Weekly, 43(10), 4448.
Kohli, A. (2001). The success of Indias democracy. Cambridge: Cambridge University Press.
La Ferrara, E. (2002). Inequality and group participation: Theory and evidence from rural Tanzania. Journal of Public Economics, 85, 235273.
Lanjouw, P., & Ravallion, M. (1999). Benet incidence, public spending reforms, and the timing of program capture. World Bank Economic Review, 13, 257
273.
Mathur, L. (2007). Employment guarantee: Progress so far. Economic and Political Weekly, 42(52), 1720.
Mehrotra, S. (2008). NREG two years on: Where do we go from here? Economic and Political Weekly, 43(31), 2735.
Ministry of Law and Justice. (2005). National Rural Employment Guarantee Act 2005. The Gazette of India (7 September 2005). New Delhi: Government of India Press.
Powis, B. (2007). Systems of capture: Reassessing the threat of local elites, social development papers. South Asia Series Paper No. 109, December.
Ravallion, M. (2000). Monitoring targeting performance when decentralized allocations to the poor are unobserved. World Bank Economic Review, 14, 331
345.
Scandizzo, P., Gaiha, R., & Imai, K. (in press). Option values, switches and wages: An analysis of the employment guarantee scheme in India. Review of Development
Economics.
Seabright, P. (1996). Accountability and decentralisation in government: An incomplete contracts model. European Economic Review, 40, 6189.
Shankar, S. (2008). The politics of capture: NREG in AP and Rajasthan. Unpublished manuscript.
Tendler, J. (1997). Good government in the tropics. Baltimore: Johns Hopkins University Press.
Un Nabi, R. (1999). Consultation with the poor. Participatory poverty assessment in Bangladesh, Bangladesh: NGO Working Group on the World Bank.
Wooldridge, J. (2002). Econometric analysis of cross section and panel data. Cambridge: MIT Press.