states; indeed, it is posi-(including wage and non-wage spending) in 2004-05 to 2009-10 using state-specific consumer price indices for agricultural labourers current prices during the 2009-10 and 2010-11 (CPIAL) and per capita consumption expenditures in Schedule 1.0. Poverty rates for J&K tively correlated with the

poverty rate, though only Development website (http:\\nrega.nic.in). To share of person days from MGNREGS administrative data (http:\\nrega.nic.in). Remaining calculate expenditure per capita the authors used columns from authors’ calculations from unit record data of 2009-10 National Sample weakly so (r=0.183). It is

Figure 2: Demand for MGNREGS Work Is Greater in Poorer States

under 20%. There is clearly a large excess demand for work.

A striking observation about the data in Table 1 is that participation rates are only weakly correlated with rural poverty rates across states, as can be seen in Figure 1. If MGNREGS worked the way the Act intended then this weak correlation would be surprising, as one would expect the scheme to be more attractive to poor people, and hence have higher take up in poorer states. The same point holds for public spending on MGNREGS.

Figure 1: Participation Rates in MGNREGS and Incidence of Poverty across States

r=0.50 .8 .7 .6 .5 .4 .3 .2 .1Demand rate: Share of rural households who want work

10 15 20 25 30 35 40 45 50 55 60 Headcount index of rural poverty 2009-10 (%)

Poorer states have greater unmet demand for MGNREGS, as can be seen in Figure 3 (p 58), which plots the share of the rural population that is rationed – i e, the rationing rate times the demand rate – against the poverty rate. Yet there is variation even among poorer states. Some of the poorest states (Bihar, Jharkhand and Orissa) have low participation rates and high levels of unmet demand. This is in contrast to other poor states like Chhattisgarh, Rajasthan, Madhya Pradesh and West Bengal that perform better in providing employment under the scheme. For example, at a

1015 20 25 30 35 40 45 50 55 60 Headcount index of rural poverty 2009-10 (%) similar poverty rate, Chhattisgarh has a participation rate almost

EPW

.r = 0.13 .7 .6 .5 .4 .3 .2 .1 .0Participation rate in MGNREGS (share of rural households)

Figure 3: Poorer States Have Greater Unmet Demand for Work on MGNREGS

Biharr=0.74 Jharkhand Orissa Punjab .40 .35 .30 .25 .20 .15 .10 .05Share of rural households who were rationed Rajasthan Tamil Nadu Himachal Pradesh Chhattisgarh Kerala

10 15 20 25 30 35 40 45 50 55 60 Headcount index of rural poverty 2009-10 (%)

five times that of Bihar. Public spending is also lower in Bihar at roughly one-third of the level in Chhattisgarh.

3 Explaining the Interstate Variation in Participation Rates

In understanding why the MGNREGS is not more active in poorer states, we postulate that being a poor state has two opposing effects on participation. First, there is an indirect effect of greater poverty via a higher demand for MGNREGS work. We saw an indication of this in Figure 2. The regression coefficient of demand for MGNREGS (based on the NSS responses) on the state poverty rate is 0.591 (st. error=0.192), meaning that a 10 percentage point increase in the poverty rate comes with about a 6 percentage point increase in the share of rural households demanding MGNREGS work, on average.

The second (direct) effect is that poorer states tend to have greater unmet demand for work on the scheme. This works in the opposite direction to the indirect effect. We suggest three reasons for this direct effect. First, poorer states will be less able to afford the share of the costs that are borne by the state and local governments. Second, poorer states will tend to have a weaker capacity for administering such a scheme. Third, the poor may well be less empowered in poorer states. As we will see in the next section, both poor and non-poor people have a demand for work on the scheme, though the demand is greater amongst the poor. If poor people tend to have less power to influence local decision-making (reflected in lower awareness of their rights under the Act), then a higher poverty rate will lead the state government to put less weight on the need to accommodate the demand for work on the scheme.9

Both the direct and indirect effects are in evidence when we regress the state-level participation rate on both the demand rate and the rural headcount index of poverty, as in Table 3. There are reasons to be cautious about giving these regressions a causal interpretation. The demand for MGNREGS may well be endogenous to actual employment on the scheme. High levels of employment may stimulate demand, while low levels may create a “discouraged worker effect”, whereby potential workers stop showing interest in the scheme. There is also an endogeneity concern arising from the fact that demand for MGNREGS automatically includes actual employment; measurement error in the latter would thus create a spurious correlation between

58

Table 3: Regressions for Participation Rate in MGNREGS across States of India

Ordinary Least Squares Instrumental Variables Estimate

Constant -0.074 -0.064 (-1.639) (-1.197)

Demand for work 1.109 1.178

(13.798) (9.880)

Headcount index of rural poverty -0.501 -0.619 (-4.922) (-4.581)

R2 0.902 0.894

SEE 0.056

0.058

N 18

18

The dependent variable is the participation rate, defined as the share of the population of rural households who did any work on MGNREGS. Demand for work is the share of rural households saying they want work on the scheme. Headcount index of rural poverty is the percentage of population below the poverty line. The t-ratios in parentheses are based on White standard errors. The IVs were log SDP per capita and its squared value. Sources: Authors’ estimates from NSS (2009-10). See Table 1.

demand and actual employment. One might also question treating the poverty rate as exogenous to work on the scheme; hopefully, the poverty rate will fall with higher participation.

To address these concerns, Table 3 also gives an Instrumental Variables (IV) estimate treating both the demand for work and the headcount index as endogenous. For this purpose we assume that the log of state domestic product (SDP) per capita and its squared value influence both the demand rate and the poverty rate, but do not influence the participation rate independently of these variables.10

We see that both the effects described above of a higher poverty rate are evident in the data, using both estimation methods. The participation rate rises with demand at a given poverty rate, but a higher poverty rate is associated with lower participation at given demand. Of course, a higher poverty rate also entails higher demand for work. On factoring in the effect of differences in the state poverty rate on demand for work we find that the poverty effect operating via demand for work entails that a 10 percentage point increase in the headcount index implies a 6.6 percentage point increase in participation (using the OLS estimate),11 while the effect operating independently of demand entails a 5.0 percentage point drop. The direct effect of a high poverty rate works in the opposite direction to the indirect effect, via demand – so much so that, on balance, we see only a small positive effect of higher poverty on participation rates across states (Figure 1).

To better understand this strong direct effect of poverty, in Dutta et al (2012b) we study more closely the performance of the state with the highest poverty rate, Bihar. Drawing on various (qualitative and quantitative) data sources, including our own special-purpose surveys, we argue that both of the factors identified above – lower capacity in poorer states and lower empowerment of poor people – are at work in Bihar. Furthermore, we argue that changing one alone will not assure that MGNREGS will reach its potential in India’s poorest areas. Effective action on both fronts will be necessary.

4 Is Rationing Undermining the Self-Targeting Mechanism?

By insisting that participants do physically demanding manual work at a low wage rate, workfare schemes such as MGNREGS aim to be self-targeted, in that non-poor people will not want

april 21, 2012 vol xlviI no 16

to participate. The substantial rationing that we have demonstrated above raises the question of how well this self-targeting mechanism works in practice. The fact that there is rationing does not mean that targeting will not be pro-poor. For one thing, the manual work requirement at a low wage rate will still discourage non-poor people from wanting to participate. For another, the local authorities doing the rationing may well favour the poor. The local officials who are deciding who gets work could either enhance or diminish the scheme’s targeting performance. There has been no comprehensive national assessment of targeting performance. The quantitative studies that have been done so far have been based on selected samples and the tests used have often been problematic.12 What does the evidence from the NSS survey for 2009-10 suggest?

Table 4 gives the participation rate, demand rate and the rationing rate by rural household quintiles defined on household consumption per person from the survey.13 As expected, we see that demand for work on the MGNREGS declines with consumption per person. Richer households are less likely to want to do this work, although there is demand even amongst the richest quintile in rural areas. Consistent with the incidence of expressed demand, we also see that the proportion of households who have obtained job-cards declines with consumption per person. But notice that the demand rate is higher than the proportion with job-cards; there are many households who express demand for work who have not obtained job-cards. Table 4: Coverage of MGNREGS across Consumption Quintiles of the Rural Population of India (2009-10)

Quintiles	Participation	Demand	Rationing	Share of	Mean Person	Mean Person
	Rate	Rate	Rate	with HHs	Days amongst	Days among
				a Job-card	Participating	all Rural
					HHs	HHs
Q1 (poorest)	0.335	0.609	0.450	0.465	33.7	11.3
Q2	0.297	0.540	0.450	0.414	36.2	10.7
Q3	0.273	0.507	0.462	0.385	38.3	10.4
Q4	0.226	0.434	0.479	0.329	40.0	9.0
Q5 (richest)	0.138	0.309	0.553	0.218	40.0	5.5
All	0.242	0.462	0.476	0.347	37.4	9.0

The participation rate is the share of rural households working on MGNREGS. The demand rate is the share of rural households who want work on the programme. The rationing rate is the share of those who wanted work who did not get it. Source: Authors’ estimates from NSS (2009-10).

Strikingly, however, across India as a whole, the rationing rate also tends to rise with consumption per person. The locallevel processes of deciding who gets work amongst those who want it entails that poorer households are less likely to be rationed, although the difference is modest. Thus we fi nd that the participation rate declines with consumption even more steeply than the demand rate. Quintile averages lose a lot of detail. A finer representation of the data shows that the participation rate declines rather slowly until one reaches about the 50th percentile of the consumption distribution: households just below the official poverty line are no more likely to participate in MGNREGS than those just above the line. The marked decline in participation rates does not emerge until we get to the upper half of the rural consumption distribution. Although far fewer “rich” rural households participate, there are still some. This could reflect recent shocks, or poor individuals within generally well-off households.

Economic & Political Weekly

EPW

april 21, 2012 vol xlviI no 16

Note that although the rationing rate tends to rise with consumption, this does not imply that more rationing would improve targeting. What the numbers in Table 4 reflect is the rationing process at a given level of participation. When the participation rate rises through a reduction in rationing the self-targeting mechanism will start to play a bigger role. We will see evidence of this when we compare targeting performance across states with very different participation rates.

Also notice that, amongst participants, the days of work received shows a slightly positive gradient with consumption per person. The pro-poor targeting is achieved through both demand for work and the rationing of work, not by the amount of work actually received.

It is of interest to compare targeting performance across states. There are many measures of “targeting performance” in the literature that might be used for this purpose. Ravallion (2009) surveys the various measures and tests their performance in predicting the impacts on poverty of a large antipoverty programme in China, called the Di Bao programme. (This provides cash transfers targeted to those with income below the locally-determined Di Bao poverty lines.) Amongst all standard targeting measures, the one that performed the best (and by a wide margin) in predicting the programme’s impact on poverty was the “targeting differential” (TD), originally proposed by Ravallion (2000). In the present context, this can be defi ned as the difference between the MGNREGS participation rate for the poor and that for the non-poor. In obvious notation:

non-poor

TDi = Pipoor – Pi ...(2)

Here Pi (i=poor, non-poor) is again the participation rate (as defined by equation (1)), but this time differentiated between the poor and non-poor. To interpret the targeting differential, note that when only poor people get help from the programme

Table 5: Targeting Performance of MGNREGS across States

State	Participation	Participation	Targeting	Rationing	Rationing
	Rate for	Rate for	Differential	Rate for	Rate for
	the Poor	the Non-Poor		the Poor	the Non-Poor
Andhra Pradesh	0.513	0.322	0.191	0.215	0.259
Assam	0.233	0.149	0.085	0.523	0.590
Bihar	0.127	0.075	0.052	0.756	0.816
Chhattisgarh	0.571	0.386	0.186	0.260	0.366
Gujarat	0.298	0.185	0.114	0.319	0.490
Haryana	0.106	0.037	0.069	0.701	0.760
Himachal Pradesh	0.510	0.318	0.192	0.173	0.206
Jharkhand	0.237	0.163	0.075	0.613	0.641
Karnataka	0.126	0.065	0.061	0.503	0.703
Kerala	0.116	0.112	0.005	0.535	0.516
Madhya Pradesh	0.534	0.319	0.215	0.297	0.438
Maharashtra	0.096	0.025	0.071	0.738	0.898
Orissa	0.317	0.135	0.182	0.509	0.650
Punjab	0.145	0.035	0.110	0.729	0.872
Rajasthan	0.728	0.579	0.149	0.166	0.150
Tamil Nadu	0.484	0.302	0.182	0.093	0.219
Uttar Pradesh	0.242	0.120	0.122	0.483	0.582
West Bengal	0.559	0.379	0.179	0.282	0.376
All India	0.325	0.210	0.115	0.428	0.463

Households classified into poor or non-poor based on poverty lines for Schedule 10 that would yield the same state-specific poverty rates as estimated from Schedule 1, and reported in Table 1. All-India figures reported in this table include only the states shown. Source: Authors’ calculations from NSS (2009-10).

and all of them are covered, TD = 1, which is the measure’s upper bound; when only the non-poor get the programme and all of them do, TD = -1, its lower bound. This measure is easy to interpret, and it automatically reflects both leakage to the nonpoor and coverage of the poor.

Table 5 (p 59) gives the TD and participation rates for the poor and non-poor.14 Participation rates among the poor vary enormously across states, from a low of 0.10 in Maharashtra to a high of 0.73 in Rajasthan. They also vary a lot among the non-poor. Although participation rates are always higher for the poor, the gap with that for the non-poor is not large. The targeting differential for India as a whole is 0.12. (The TD for China’s Di Bao programme mentioned above was 0.27.) Madhya Pradesh has the highest TD, at 0.22, while Kerala has the lowest, at 0.01.

Table 5 also gives the rationing rates for the poor and nonpoor. Consistent with the all-India results in Table 4, we see that the non-poor are rationed more than the poor in almost all states (the only exceptions are Kerala and Rajasthan).

The TD is determined by how the demand rates and the rationing rates vary between the poor and non-poor. We can do use a simple decomposition method to show how much of the TD is due to each factor:

non-poor) – D(Ri

TDi = (1 – R)(Dipoor – Di poor – Rinon-poor) + residual (“Self-targeting effect”) (“Rationing effect”) ...(2)

Here the bars denote fixed reference values, while Di and Ri are the demand rates and rationing rates for i=poor, non-poor, respectively. TD can thus be interpreted as the “self-targeting effect” (greater demand for work amongst the poor) net of the rationing effect (the extent to which the poor might be rationed more). (Since the decomposition is not exact – given the non-linearity in equation (1) – there is also a residual.)

Applying this decomposition, and using the all-India values for the reference we find that 85.6% of the national TD is attributed to the difference in demand between the poor and non-poor while 13.7% is due to the difference in rationing rates. (The residual is negligible.) There are differences across states, though we find that the demand effect dominates in 17 of the 20 states. So, despite the rationing, the bulk of the propoor targeting is coming through the self-targeting mechanism.

Targeting performance is better in states with higher overall participation rates. Figure 4 plots the two participation rates from Table 5 against the overall participation rate (Table 1). We see that the TD – the gap between the two lines – rises with the overall participation rate, and the two are strongly correlated (r=0.748).

Targeting performance also tends to be worse in the states with higher levels of rationing (the correlation between TD and rationing rate is -0.71). However, this arises because overall participation rates are low in states with higher degrees of rationing. Indeed, once one controls for the participation rate, there is no significant partial correlation between the TD and the rationing rate (the t-statistic is -0.611).15

So we find that higher overall participation rates tend to come with better targeting performance and lower rationing rates. The fact that targeting performance improves as the programme expands makes this an example of what Lanjouw and

Figure 4: The Participation Rate of Poor Households in MGNREGS

(Poor, non-poor)

.0 .1 .2 .3 .4 .5 .6 .7 .8 Participation rate for poor/non-poor Poor Non-Poor

.0 .1 .2 .3 .4 .5 .6 .7 .8

Overall participation rate

Ravallion (1999) call “early capture” by the non-poor, which they showed to be a common feature of access to safety-nets and schooling in India.16 Lanjouw and Ravallion also show in a theoretical model of the political economy of targeted programmes that for programmes with relatively large start-up costs, early capture by the non-poor may be the only politically feasible option (especially when the start-up costs must be fi nanced domestically). So this feature of MGNREGS is possibly not surprising.

Targeting by social groups (castes and tribes) is another dimension of interest. Qualitative studies have suggested that scheduled castes (SCs), scheduled tribes (STs) and women – groups that have traditionally been excluded – have benefi ted disproportionately from the scheme.17 We shall return to discuss participation by women in Section 6. Here we focus on the

Figure 5: Participation Rates for STs, SCs and OBCs in MGNREGS

Participation rate for each group ႷႷႷႷႷႷႷႷႷႷႷႷႷႷႷႷႷScheduled tribe Scheduled caste Other Backward Classes Other ST SC OBC Other

Overall participation rate

scheme’s performance in reaching ST, SC and Other Backward Class (OBC) households.

Table 6 (p 60) gives the participation rates by these groupings. Nationally, 42% and 34% of rural ST and SC households respectively participated. Participation was lower for OBCs at 21% and lowest for all others, at 16%. But there is a wide range across states. For STs, the range is from 6% of households participating in Maharashtra to 82% in Rajasthan, while for SC households it is from 2% to 65%, and for the same states. Figure 5 plots the participation rates against the overall participation rates across states. Similarly to targeting by poverty status, we see that the participation rates for STs, SCs and OBCs rise faster as the overall participation rates rise, suggesting that the targeting of disadvantaged castes improves with programme expansion.

Table 6 also gives the targeting differential for STs, SCs and OBCs together, defined as their (weighted) average participation rate less the participation rate for “others”. This “caste TD” varies from -0.02 in Assam to 0.29 in Chhattisgarh, with a national mean of 0.12, almost identical to the national “poverty TD” in Table 5. Similarly to the poverty TD, the “caste TD” is positively correlated with the overall participation rates (r=0.723).

5 Wages and Rationing on MGNREGS

There have been a number of concerns about the stipulated wage rates for the programme. On the one hand, it is argued that setting scheme wages below the state-mandated rates under the Minimum Wages Act is a violation of the law and tantamount to “forced labour”, a stand that has been recently upheld by the Supreme Court.18 On the other hand, concerns have been raised that the wage rate on the MGNREGS is being set too high, relative to actual casual labour market wages. The concern here is that the scheme will attract workers from market work and so bid up the market wage rate.19 (Of course to supporters of the scheme this is counted as a benefi t.)

What does the evidence suggest? Table 7 gives the average wage rates from the administrative data. These are calculated as total MGNREGS spending on unskilled labour divided by total person days of employment provided.20 Table 7 also reports estimates for average wages in (private) casual labour from the NSS for the same year.21

Economic & Political Weekly

EPW

Table 7: Average Wages on MGNREGS and in Casual Labour 2009-10

Average Wage Rate on Average Casual Wage Rate (Rs/day) MGNREGS (Rs/day) Overall Male Female

Andhra Pradesh 91.9 98.5 115.4 75.7

Assam 87.0 90.1 94.4 74.9

Bihar 97.5 79.4 81.0 65.8

Chhattisgarh 82.3 68.8 70.8 65.5

Gujarat 89.3 83.3 87.3 71.0

Haryana 150.9 139.6 146.1 99.1

Himachal Pradesh 109.5 139.6 141.4 110.2

Jammu and Kashmir 93.3 158.3 157.5 na

Jharkhand 97.7 101.2 103.6 82.2

Karnataka 86.0 84.5 96.9 62.8

Kerala 120.6 206.5 226.6 119.3

Madhya Pradesh 83.7 69.0 74.5 58.1

Maharashtra 94.3 75.2 86.0 58.2

Orissa 105.9 75.6 81.0 59.1

Punjab 123.5 130.4 133.5 91.8

Rajasthan 87.4 125.7 132.3 94.3

Tamil Nadu 71.6 110.8 132.1 72.6

Uttar Pradesh 99.5 94.3 97.0 69.2

Uttarakhand 99.0 118.7 122.1 96.7

West Bengal 90.4 85.3 87.8 65.9

All India 90.2 93.1 101.5 68.9

MGNREGS wage rates estimated as total expenditure on wages (excluding skilled or semi-skilled) divided by total number of person days of employment for FY 2009-10 (April 2009 to March 2010). Casual wages for June 2009 to July 2010 period, based on NSS 66th round survey. “All India” includes smaller states not reported. Note that we do not report the female wage rate for Jammu and Kashmir, as we found it was based on a sample of only seven observations and was not reliable. (The sample estimate of 206.5 was also implausibly high relative to the male wage). Sources: Casual wages from Key Indicators of Employment and Unemployment in India, 2009-10, NSSO, Government of India (June 2011). MGNREGS expenditure and employment data are from the state-wise Monthly Progress Reports (www.nrega.nic.in).

We see that it is not the case that the MGNREGS wage rate is everywhere well above the market wage rate. Indeed, for India as a whole the two wages are quite close. If rural India was one labour market one might conjecture that the scheme has indeed brought the two wage rates into parity. However, rural India is not one labour market, as mobility is clearly imperfect. When we look at the states we see that for half of them the MGNREGS wage rate in 2009-10 is actually lower than the average wage rate for casual labour.

Given the extent of rationing, it does not seem plausible that the scheme would be having a large impact on wages for other casual work, let alone resulting in a higher casual wage than for MGNREGS in half the states. For example, with only 17% of those who wanted work on the scheme in Punjab getting that work, it is hard to believe that the casual (non-PW) wage rate is above the MGNREGS wage rate due to competition for workers.

That said, we do find that the relative wage – defined as the mean wage rate for casual (non-PW) labour divided by the MGNREGS wage – tends to be lower in states with higher levels of unmet demand, as measured by the difference between the demand rate and the participation rate. The correlation coeffi cient between the relative wage and unmet demand is -0.520, which is significant at the 2% level. However, there are two reasons to question whether this really reflects greater tightening of the casual labour market in states where there is less unmet demand for work on the scheme. First, the implied relative wage rate at zero rationing is too high to be believed. Regressing the log of the relative wage rate (the log gives a slightly better fit) on the unmet demand, the intercept is 0.397 (t=3.800), implying that the non-PW wage will be almost 50% higher than the MGNREGS wage rate in the absence of rationing.22 The work is very similar, and there is no obvious reason why such a differential would exist in equilibrium. Possibly the unmet demand is picking up some other factor correlated with it.

Poverty is a plausible candidate, and this suggests a second reason for questioning whether this negative correlation between relative wages and rationing reflects how the casual labour market has responded. The correlation largely vanishes when we control for the poverty rate. A higher poverty rate may be associated with greater landlessness and hence a larger supply of casual labour, bringing down the wage rate. Regressing the relative wage rate on unmet demand and the poverty rate, the effect of unmet demand becomes insignifi cant (prob.=0.16).

6 Rationing and the Participation of Women

Nationally, almost half (48%) of the employment as registered in the administrative data for 2009-10 goes to women.23 This is very high for a country where a minority of women participates in the paid labour force; for example, women’s participation rate in the MGNREGS is about twice their share of other (non-PW) casual wage work.24 The variation across states is striking; between the two extremes, only 7% of the work goes to women in Jammu and Kashmir as compared to 88% in Kerala (Table 1). The female share in MGNREGS work is greater than their share of the work in the casual wage labour market in all states, but the gap tends to be larger in states where women participate less in the casual labour market (Figure 6).

Figure 6: Women in MGNREGS and Casual Wage Labour

r=0.45 Female share work on MGNREGS (%) 100 80 60 40 20 0Kerala Tamil Nadu Rajasthan Line of equality Regression line Jammu and Kashmir

0 5 10 15 20 25 30 35 40 45 50 Female share of work in the casual labour market (%)

Women are less likely to participate (relative to men) in MGNREGS in poorer states. Figure 7 plots the share of person days of employment going to women against the poverty rate. We see a negative correlation (r=-0.47). By contrast, women’s share of casual (non-PW) wage work tends to be slightly higher on average in poorer states, though the difference is not statistically significant (r=0.09). So the scheme is clearly bringing women into the paid workforce, but more so in less poor states.

Why do we see less of the available work going to women in poorer states? A plausible explanation is that there is greater rationing of work in poorer states and that women are rationed more than men.25 Assuming that the effect of being a poorer

62

Figure 7: Share of Work Going to Women

0.0 0.2 0.4 0.6 0.8 1.0 Female share of work Female share of non-PW work Female share of MGNREGS rFemale share of non-WPW work • Female share of MGNREGS

0 5 1015202530354045505560

Headcount index of rural poverty 2009-10 (%)

state on demand for work is the same for men and women, the pattern in Figure 7 suggests that the rationing process is less favourable to women in poorer states.

Do women have equal access to the scheme, when they need it? Again, we cannot give a direct answer from the survey data, but the patterns in the interstate data are suggestive of greater rationing of women. There is a negative correlation between the female share of work and the rationing rate (Figure 8, p 63). It might be conjectured that this correlation actually refl ects differences in the extent of poverty. Women may well be less aware of their rights and less empowered to demand work in poorer states. For example, when other work is scarce, they may get crowded out by men. However, the negative correlation between the female share of work and the rationing rate persists when we control for the poverty rate, as can be seen in Table 8. Table 8: Regressions for the Female Share of Employment in MGNREGS

Full Sample with Headcount Sample with Male and Sample Index Available Female Wages Available

Constant 0.676 0.829 0.193 0.131 0.697

(8.671) (5.341) (0.887) (0.246) (3.922)

Rationing -0.505 -0.419 -0.307 -0.456 -0.469 (-3.758) (-2.898) (-2.904) (-5.108) (-5.972)

Headcount index of rural poverty na -0.515 0.241 na na (-1.538) (1.518)

Female casual non-PW wage (log) na na -0.549 -0.744 na (-2.056) (-3.475)

Male casual non-PW wage (log) na na 0.675 0.806 na

(5.156) (3.556)

Female wage relative to na na na na -0.836 male wage (log) (-3.301)

R2 0.336 0.467 0.811 0.692 0.687

SEE 0.164 0.145 0.093 0.111 0.108

N 20 18 1819 19

The dependent variable is the share of total person days of employment on MGNREGS going to women. The rationing rate is the share of those who wanted work who did not get it. The headcount index is the per cent of population below the poverty line. The t-ratios in parentheses are based on White standard errors.

Gender differences in the opportunities available in the casual labour market can also be expected to influence demand for work. We find that the female market wage rate has a signifi cant negative effect on women’s share of the work provided, while the male wage rate has the opposite effect (Table 8).26 The female wage relative to the male wage is the relevant variable. This suggests that there is an intra-household substitution effect; for

april 21, 2012 vol xlviI no 16

Figure 8: Rationing Rate Trends and Share of Women in the MGNREGS

0 20 40 60 80 100 Female share of person days of employment (%) Tamil Nadu Kerala Jammu and Kashmir r=0.58

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Rationing rate (share of those who wanted work who get it)

example, when casual labour market opportunities are good for men but bad for women this makes it easier for women to get the (limited) number of jobs available on the scheme.27 The wage effect is strong statistically, and greatly increases the explanatory power.28 The negative effect of rationing on women’s access to the scheme also persists when we control for differences in the wages received for private casual work.

7 Conclusions

There has been much public debate about India’s MGNREGS since it was introduced. There have been many media reports and some selective surveys, covering at most a few states and/ or selected districts. This paper used the NSS of 2009-10 to test some of the claims that have been made in past debates using data for all major states of India. We have focused on a distinctive and important feature of MGNREGS: the guarantee of employment at the stipulated wage rates.

We confirm expectations that the demand for work on MGNREGS tends to be higher in poorer states. This appears to reflect the scheme’s built-in “self-targeting” mechanism, whereby non-poor people fi nd work on the scheme less attractive than do poor people.

However, actual participation rates in the scheme are not (as a rule) any higher in poorer states where it is needed the most. The reason for this paradox lies in the differences in the extent to which the employment guarantee is honoured. The answer to the question posed in our title is clearly “no”. Rationing is common, but far more so in some of the poorest states.

We do not find that the local-level processes determining who gets work amongst those who want it are generally skewed against the poor. There are sure to be places where this is happening (and qualitative field reports have provided examples). But it does not appear to stand up as a generalisation. We do find evidence that the poor fare somewhat less well when it comes to the total number of days of work they manage to get on the scheme. However, despite the pervasive rationing we find, it is plain that the scheme is still reaching poor people and also reaching the STs and OBCs.

Participation rates on the scheme are higher for poor people than others. This holds at the official poverty line, but the scheme is also reaching many families just above the offi cial line.

Economic Political Weekly

EPW

It is only at relatively high consumption levels that participation drops off sharply. This should not be interpreted as indicating that well-off families in rural India are turning to MGNREGS. There may well be shocks that are not evident in the household consumption aggregates. And there may be individual needs for help that are not evident in those aggregates.

Targeting performance varies across states. The overall participation rate seems to be an important factor in accounting for these interstate differences in targeting performance, with the scheme being more pro-poor and reaching STs and OBCs more effectively in states with higher overall participation rates.

While the allocation of work through the local-level rationing process is not working against the poor, there are clearly many poor people who are not getting help because the employment guarantee is not in operation almost anywhere (Himachal Pradesh, Rajasthan and Tamil Nadu could be counted as the exceptions, where 80% or more of those who want work got it). And other potential benefi ts of the scheme to poor people are almost certainly undermined by the extensive rationing, notably the empowerment gains and the insurance benefits. The fi rstorder problem for the MGNREGS is the level of unmet demand.

While the scheme is clearly popular with women – who have a participation rate that is double their participation rate in the casual labour market – the rationing process is not favouring them. We also find evidence of a strong effect of relative wages on women’s participation – both wages on the scheme relative to the market wage and the male-female differential in market wages. As one would expect, poor families often choose whether it is the man or the woman who goes to the scheme according to relative wages.

It has been claimed by some observers that the scheme is driving up wages for other work, such as in agriculture; some observers see this as a good thing, others not. For India as a whole, we find that the scheme’s average wage rate was roughly in line with the casual labour market in 2009-10. This might look like a competitive labour market equilibrium, but that view is hard to reconcile with the extensive rationing we find. Interestingly, we do find a significant negative correlation between the extent of rationing and the wage rate in the casual labour market relative to the wage rate on the scheme. Although this is suggestive, on closer inspection we are more inclined to think that other economic factors are at work. Indeed, the correlation largely vanishes when we control for the level of poverty. Poorer states tend to see both more rationing of work on the scheme and lower casual wages – possibly due to a greater supply of labour given the extent of rural landlessness.

Notes

1 On the incentive arguments for workfare schemes see Besley and Coate (1992).

2 The distortions could be due to monopsony power in rural labour markets (Basu et al 2009) or labour-tying (Basu 2011). Nor does the distortion need to be in the rural labour market; it could also be in the urban labour market, generating excess migration to urban areas (Ravallion 1990).

3 The scheme also tried to reduce future poverty by creating useful assets. This is not an issue we address here.

4 The policy choice between limited coverage at a socially acceptable (“living”) wage and wide coverage is studied in Ravallion (1991).

5 The central government bears 90% of all variable costs. This includes wage costs and three-quarters of the non-wage component (working on an assumed 60:40 labour capital ratio). The centre also provides an additional 6% of programme costs to the states to defray the costs of administering the scheme. States are responsible for paying unemployment allowances from their own budget.

6 The scarcest manpower resource locally is the junior engineer or panchayat technical assistant who can prepare technical estimates and draw up engineering plans for the works. Andhra Pradesh (AP) has used information technology to reduce the need for such skilled local staff, by developing standardised and computerised templates for the engineering plans of common types of works. AP is the exception however. Most other states face shortages of such skilled local staff.

7 Data is from the official Government of India website for MGNREGS (http:\\nrega.nic.in).

8 In Dutta et al (2012b) we provide supportive evidence on this point for Bihar, based on our surveys in 2009-10.

9 Note that this second reason for the direct effect of poverty is not consistent with a model of public decision-making based on any standard form of utilitarian calculus. For then one would expect the policy weight on accommodating the demand for work on the scheme to be higher in states with a higher share of poor people who need that work more than the non-poor.

10 The first stage regressions had ample explanatory power, and there was no sign that SDP per capita had an independent effect on MGNREGS participation when added as an additional regressor to the OLS estimate. We cannot test our identifying assumption that SDP per capita and its squared value only affect MGNREGS participation via demand for work and the poverty rate when both the latter are treated as endogenous. However, over-identifi cation test passed when only the demand rate was treated as endogenous. Neither SDP per capita nor its squared value was signifi cant when added to the IV regression treating the demand rate as endogenous.

11 Recall that the regression coefficient of the demand rate on the poverty rate (headcount index) is 0.591. Then the poverty effect operating via the demand rate is 0.591*1.109=0.655.

12 Jha et al (2009) report evidence that households with larger landholdings were more likely to participate in the scheme in AP, though they find evidence of better targeting in Rajasthan. They conclude that the scheme is being “captured” by the non-poor in AP. Note, however, that their regressions control for other variables that may well capture poverty, including occupation and whether the household has a below poverty line (BPL) card. The full regressions are not presented in their paper, but it may well be that having a BPL card (say) is already capturing the pro-poor targeting of MGNREGS, but that the BPL card puts too high a weight on landlessness from the point of view of explaining participation. Then the amount of land may appear to have the wrong sign, even though the scheme is targeting the poor. By contrast, the results of Liu and Deininger (2010) suggest quite pro-poor targeting of MGNREGS in AP. Shariff (2009) reports participation regressions for MGNREGS in selected backward districts of northern states (including some districts in Bihar). Some of the regression coeffi cients also suggest perverse targeting. Shariff is careful in interpreting the results though the same inferential concerns hold as for the study by Jha et al.

References

Basu, Arnab K (2011): “Impact of Rural Employment Guarantee Schemes on Seasonal Labour Markets: Optimum Compensation and Workers’ Welfare”, Journal of Economic Inequality.

Basu, Arnab K, Nancy H Chau and Ravi Kanbur (2009): “A Theory of Employment Guarantees: Contestability, Credibility and Distributional Concerns”, Journal of Public Economics, 93: 482-97.

Besley, Timothy and Stephen Coate (1992): “Workfare vs Welfare: Incentive Arguments for Work Requirements in Poverty Alleviation Programs”, American Economic Review, 82(1): 249-61.

Datt, Gaurav and Martin Ravallion (1994): “Transfer Benefits from Public Works Employment”, Economic Journal, 104: 1346-369.

Drèze, Jean and Reetika Khera (2009): “The Battle for Employment Guarantee”, Frontline, 26(1): 3-16.

Dutta, Puja, Rinku Murgai, Martin Ravallion and Dominique van de Walle (2012a): “Does India’s Employment Guarantee Scheme Guarantee Employment”, Policy Research Working Paper 6003, World Bank, Washington DC, http://econ. worldbank.org/external/default/main?page PK=64165259theSitePK=469372piPK=6416 5421menuPK=64166093entity ID=000158349_ 20120320090137

– (2012b): Rozgar Guarantee? Assessing India’s Biggest Anti-Poverty Program in India’s Poorest State, World Bank, Washington DC, forthcoming.

Imbert, Clément and John Papp (2011): “Equilibrium Distributional Impacts of Government Employment Programs: Evidence from India’s Employment Guarantee”, mimeo, Paris School of Economics.

Jha, Raghbendra, Sambit Bhattacharyya, Raghav Gaiha and Shylashri Shankar (2009): “‘Capture’ of Anti-Poverty Programs: An Analysis of the National Rural Employment Guarantee Program in India”, Journal of Asian Economics 20(4): 456-64.

Khera, Reetika, ed. (2011): The Battle for Employment Guarantee (New Delhi: Oxford University Press).

Lanjouw, Peter and Martin Ravallion (1999): “Benefit Incidence and the Timing of Program Capture”, World Bank Economic Review, 13(2): 257-74.

Liu, Yanyan and Klaus Deininger (2010): “Poverty Impacts of India’s National Rural Employment Guarantee Scheme: Evidence from Andhra Pradesh”, mimeo, World Bank.

Ravallion, Martin (1990): “Market Responses to Anti-Hunger Policies: Wages, Prices, and Employment” in Jean Drèze and Amartya Sen (ed.), The Political Economy of Hunger: Volume 2 (Oxford: Oxford University Press).

Shariff, Abusaleh (2009): “Assessment of Outreach and Benefits of National Rural Employment Guarantee Scheme in India”, Indian Journal of Labour Economics, 52(2): 243-68.

april 21, 2012 vol xlviI no 16