This article presents results on the participation of rural workers in the National Rural Employment Guarantee programme based on a pilot survey of three villages in Udaipur district in Rajasthan. Its focus is on participation in the nreg programme of different socio-economic groups and the determinants of the participation of these groups. It is found that the mean participation was 59 days and that targeting was satisfactory. The performance of the programme has been far from dismal.
INSIGHTmarch 15, 2008 EPW Economic & Political Weekly44We are grateful to ARC-AusAID Linkage grant LP0775444 for financial support, Raj Bhatia for help with the computations and EPW for helpful comments on an earlier draft. The usual disclaimer applies. The fieldwork, and data processing and analy-sis were carried out by Raj Bhatia in consulta-tion with the authors.Raghbendra Jha (r.jha@anu.edu.au) is at the Australian National University, Canberra; Raghav Gaiha at the University of Delhi; and Shylashri Shankar at the Centre for Policy Research, New Delhi. Reviewing the National Rural Employment Guarantee Programme Raghbendra Jha, Raghav Gaiha, Shylashri ShankarThis article presents results on the participation of rural workers in the National Rural Employment Guarantee programme based on a pilot survey of three villages in Udaipur district in Rajasthan. Its focus is on participation in the nreg programme of different socio-economic groups and the determinants of the participation of these groups. It is found that the mean participation was 59 days and that targeting was satisfactory. The performance of the programme has been far from dismal. There has been a spate of comments – mostly critical – following an au-dit of the National Rural Employ-ment Guarantee (henceforth,NREG) pro-gramme by the Comptroller and Auditor General of India [CAG 2007]. This audit has revealed several weaknesses of this anti-poverty programme as well as huge leakages. For example, a bare 3.2 per cent of registered households in 200 of India’s poorest districts managed to get the guar-anteed 100 days of employment in a year.1 The average employment provided was 18 days per needy household. Another as-sessment [Biswas 2007] draws attention to the unevenness in its implementation. Emphasising that while a total estimated expenditure of $ 4.5 billion was expected to generate two billion days of employ-ment, the actual was about one billion, and the benefits varied across different states. In Uttar Pradesh, the most popu-lous state, large segments of the rural population were ignorant of the scheme. By contrast, Rajasthan was among the top performers – the average employment per participating household was 77 days of work. The share of wages was 73 per cent. The small north-eastern state of Tripura performed well too, as the average number of days of employment per rural family was 87 days. Somewhat surprisingly, Kerala – a state with a superb record of human development – was at the bottom. In fact, only one of the southern and western states (Andhra Pradesh, Gujarat, Karnataka, Kerala, Maharashtra and Tamil Nadu) – Karnataka – generated more than 10 days of employment per rural family during 2006-07, while the eastern and northern states performed better.Some encouraging features of this scheme include (i) a high share of female employment (about 40 per cent nationally rising to 81 per cent in Tamil Nadu, and a low of 12 per cent in Himachal Pradesh); (ii) 20 districts spent more than $ 25 million on this scheme, and the benefits are reflected in greater economic security, higher farm wages, lower migration, and building of infrastructure. However, no general conclusions can be drawn about the accuracy of targeting and prompt dis-bursal of wages. Two examples suffice. In Chhattisgarh, 95 per cent of wages were paid to the actual workers, while in east-ern Jharkhand the corresponding share was barely 15 per cent.2 Other failures relate to distribution of job cards – large numbers of needy households are in the queue – the selection, design and exe-cution of projects, resulting in huge leak-ages.3 More specifically, Dreze (2007) highlights a quiet sabotage of the trans-parency safeguards inNREG in western Orissa. In a survey of 30 worksites, the investigators found evidence that a con-tractor was involved in some ways. What is worse is that the job card does not have a column for “wages paid”. Even the number of days worked is hard to verify as the names of the labourer and worksite have been replaced by numerical codes. Yet Dreze (2007) and Roy et al (2008), among others remain optimistic about the potential of the programme, mainly be-cause the awareness of employment as an entitlement has grown.1 ObjectiveThe present analysis is a part of a larger project designed to assess the cost-effec-tiveness of social safety nets in three Indian states, viz, Rajasthan, Andhra Pradesh and Maharashtra. TheNREG programme is operative in six districts of Rajasthan. Our sampling strategy is as follows. Since considerable reduction in the sampling error can be achieved by increasing the number of sample districts without sub-stantially increasing the overall sample size, we have selected 50 per cent of the total districts as the first stage units from the total number of districts covered in theNREG scheme in the state. It is often advantageous to select sampling units with unequal probabilities, which reduces sampling errors. Thus, it is proposed to
INSIGHTEconomic & Political Weekly EPW march 15, 200845select districts with probability propor-tional to size sampling at the first stage, size being the rural population/households as reported in the Census of 2001. The first set of results given below is based on a pilot survey of three villages (Dhundiya, Karanpur and Prithvisingh Ji Ka Khera) in Udaipur district, Rajas-than.4 The total number of households interviewed in December 2007 was 340. Here, the focus is on participation inNREG of different socio-economic groups and the determinants of the participation of these groups. 2 Methodology and Results First, a set of cross-tabulations are given to identify the correlates of participation inNREG. As these tabulations contain ave-rages, two econometric exercises are carried out to assess their relative importance. These involve a probit analysis of partici-pation in the NREG programme and a tobit analysis of the duration of the participation. The participation variables include a number of household specific characteristics such as caste/ethnic affiliations – i e, whether a member of scheduled castes (SC) or sched-uled tribes (ST) or “others” – educational attainment, land-owned, number of male and female adults in the household, and occupational status.5 The details of the model used and how the marginal effect of each of these factors on participation is computed are reported in Appendix 2 (p48). We present our results in two broad cat-egories. First, in our cross tabulations we report on statistics on participation in the NREG programme. Second, we model the participation of workers in the NREG pro-gramme. We report our results under these headings. Cross-TabulationsIn the cross-tabulations, an attempt is made to identify some correlates of par-ticipation and duration of participation in NREG. This is depicted in Table 1. Out of 340 households, one-third par-ticipated inNREG (Y). A vast majority of the participants belonged to “others” (about 90 per cent) and the remaining were equally divided among the SC and ST. Within each caste/ethnic group, the highest proportion of participants was among the ST, followed by “others”. The self-employed in agricultural households accounted for about 46 per cent of the participants, followed by “other labour” households. Within each occupation, the proportion of participants was, however, highest among “other labour”, followed by the self-employed in agriculture. Also, as land continues to be an important asset in rural areas, it is not surprising that the bulk of the participants (about 80 per cent) belonged to three lowest ranges of land-owned. The share of participants was highest among the (nearly) landless (about 52 per cent), followed by each of the three higher land categories. Further, about 42 per cent of the participating households had five or more members, and a little over one-fifth were small (comprising one to three members). However, the share of participants was highest among the latter (about 43 per cent).Contrary to the findings of CAG and “others”, the share of participating house-holds that worked for 90 days or more in 2007 was a little over one-fifth. About 39 per cent worked for 50 to 90 days. So a large majority worked for a fairly long duration. In fact, the mean number of days worked was high – about 59 days in the last year.Some basic characteristics of participa-tion inNREG in these three villages are given in Table 2. In Table 2, the first entry in the Nhead-ed-row indicates the number of responses (227) listing zero days. Also shown are association of participation inNREG by ethnic group. All those who worked for 90 days or more belonged to “others”. Among theSC andST, one-third or more worked for 51 to 90 days, and the majority worked for fewer days (between one and 50 days). Thus, while most groups had access to employment under the NREG, SC andST seem to have benefited relatively less. Table 2 also reports on participation inNREG by occupational category. The variation in duration of participation across occupations is striking too. All agricultural labour households worked in the range of one to 50 days while the majority of other labour participating households worked in the ranges of 51 to 90 and greater than 90 days. The majority of the self-employed in agriculture also worked in these high ranges. Among the self-employed in non-agriculture, the majority worked in the lowest range. This implies that agricultural labour and self-employed in non-agriculture relied on theNREG programme to supple-ment their incomes whereas workers in the other labour and self-employed in agriculture categories usedNREG as the mainstay of their incomes. Table 1: Frequency of Participation in NREG Characteristics N Y TotalParticipation by caste/ethnic group OT 205 102 307 SC 18 6 24 ST 4 5 9 Total 227 113 340Participation by occupation AL 4 1 5 OL 28 46 74 OT 16 3 19 SA 126 52 178 SN 53 11 64 Total 227 113 340Participation by land-owned (ha) 0 to 0.1 28 30 58 0.1 to 0.75 56 28 84 0.75 to 1.5 75 33 108 1.5 to 2.5 39 17 56 > 2.5 29 5 34 Total 227 113 340 Participation by household size 1 to 3 members 33 25 58 4 to 5 members 87 41 128 > 5 members 107 47 154 Total 227 113 340For definition see Appendix 1.Table 2: Characteristics of Duration of Participation in NREG 0 day 1 to 50 days 51 to 90 days > 90 days TotalDuration of participation (aggregate numbers) N 227 0 0 0 227 Y 0 46 44 23 113 Total 227 46 44 23 340Duration of participation by caste/ethnic group OT 205 39 40 23 307SC 184 2 0 24ST 4 3 2 0 9 Total 227 46 44 23 340Duration of participation in NREG by occupation AL 4 1 0 0 5 OL 28 13 16 17 74OT 163 0 0 19 SA 126 22 25 5 178SN 53 7 3 1 64 Total 227 46 44 23 340Duration of participation in NREG by land-owned (ha) 0 to 0.1 28 7 10 13 58 0.1 to 0.75 56 14 8 6 84 0.75 to 1.5 75 15 16 2 108 1.5 to 2.5 39 5 10 2 56> 2.5 295 0 0 34 Total 227 46 44 23 340
102 6 5 205 18 4
INSIGHTEconomic & Political Weekly EPW march 15, 200847replaced by land-owned dummies – 0 to 0.1 ha (omitted group), 0.1 to 0.75 ha, 0.75 to 1.5 ha, 1.5 to 2.5 ha, and larger than 2.5 ha. All land dummies except that for the highest land-owned group have signifi-cant positive coefficients, implying higher probabilities of participation relative to the (nearly) landless. The probability of participation decreases with the number of adult males and females but increases with household size. The village dummies have effects similar to those in the earlier specifications. The marginal effects for the specifica-tion used in Table 5 allow us to assess the relative importance of various determi-nants of participation. As may be noted from Table 6, the highest marginal effect among the land-owned dummies is asso-ciated with the third dummy (i e, house-holds owning land between 0.75 and 1.5 ha), followed by the next higher range of land owned. The negative effect of number of adult females is larger (in absolute value) than that of adult males while that of household size is relatively small. Between the village dummies, the (absolute) effect of the third is larger. Tobit results on the determinants of duration of participation are obtained by combining the (predicted) probabilities of participation and other household and village characteristics. These results are reported in Table 7. The greater the prob-ability of participation, the longer is the duration of participation in NREG. All land-owned dummies have significant negative coefficients, implying lower durations of participation relative to that of the (nearly) landless. The larger the number of adult males and females, the longer is the duration of participation. Household size, however, has a negative effect on the number of days of participa-tion. The duration is higher in the second village and lower in the third, relative to that in the omitted village. The overall specification is validated by the chi-square specification test. 3 ConclusionsAlthough based on the evidence from three villages in one district in Rajasthan, the analysis in this paper suggests that the targeting accuracy of the NREG programme was far from dismal. First, nearly one-third of the households participated in this scheme. Second, large segments of highly disadvantaged groups such as the ST, the landless and labour households participated in it. Third, about one-fifth of the households worked for about 100 days during 2007. Further, the landless and labour households participated for long durations. Table 5: Determinants of Participation in NREG (3)Probit Regression Number of obs = 340 LRchi2(9) = 128.53 Prob > chi2 = 0.0000Log likelihood = -151.91997 Pseudo R2 = 0.2973Participant Coef Std Err z P>|z| [95% ConfInterval]_Iland_g_2 .5825464 .330061 1.76 0.078 -.0643613 1.229454_Iland_g_3 .758315 .3151825 2.41 0.016 .1405686 1.376061_Iland_g_4 .6700685 .3508152 1.91 0.056 -.0175166 1.357654_Iland_g_5 .2354679 .4000868 0.59 0.556 -.5486878 1.019624 a_m -.2107726 .1344599 -1.57 0.117 -.4743092 .0527641 a_f -.2811701 .1539963 -1.83 0.068 -.5829974 .0206572 hhsize .0881745 .0543562 1.62 0.105 -.0183617 .1947108_Ivillage_2 -.6778212.1765229-3.840.000-1.0238-.3318427_Ivillage_3 2.462004.39619546.210.0001.6854763.238533_cons -.6842052 .3430846-1.99 0.046 -1.356639 -.0117719Table 6: Determinants of Participation in NREG (Marginal Effects)Probit Regression, Reporting Marginal Effects Number of obs = 340 LRchi2(9) = 128.53 Prob > chi2 = 0.0000Log likelihood = -151.91997 Pseudo R2 = 0.2973Partic~t dF/dx Std Err Z P>|z| x-bar [ 95% C I ]_Iland~2* .218754 .1262998 1.76 0.078 .247059 -.028789.466297_Iland~3* .2815091 .1167218 2.41 0.016 .317647 .052739 .51028_Iland~4* .2556942 .1358099 1.91 0.056 .164706 -.010488 .521877_Iland~5* .087793 .1536823 0.59 0.556 . 1 -.213419.389005 a_m -.0757644 .0483398 -1.57 0.117 1.62647 -.170509 .01898 a_f -.1010695 .0553256 -1.83 0.068 1.7 -.209506 .007367hhsize .0316952 .01956911.62 0.105 5.50882 -.006659.07005_Ivill~2* -.2345264 .0572666-3.84 0.000.426471 -.346767 -.122286_Ivill~3* .7379032 .04819626.21 0.000.135294.64344.832366 obs P .3323529 pred P .3240018 (at x-bar)(*) dF/dx is for discrete change of dummy variable from 0 to 1.Z and P>|z| correspond to the test of the underlying coefficient being 0.Table 7: Determinants of Duration of Participation in NREGTobit Regression Number of obs = 340 LRchi2(10) = 170.17 Prob > chi2 = 0.0000Log likelihood = -689.42865 Pseudo R2 = 0.1099n_days Coef Std Err t P>|t| [95% ConfInterval]pp 400.242694.81154.220.000213.7315586.7538_Iland_g_2 -26.361514.07078 -1.87 0.062 -54.04123 1.318235_Iland_g_3 -37.703819.08307-1.98 0.049 -75.24361 -.1639841_Iland_g_4 -30.9998917.66301-1.76 0.080-65.746193.746412_Iland_g_5 -6.731264 16.20129-0.42 0.678 -38.602125.13957 a_m 15.85228 7.009551 2.26 0.024 2.063236 29.64131 a_f 18.04148 9.011511 2.00 0.046 .3142311 35.76873 hhsize -5.490571 2.923558 -1.88 0.061 -11.24173 .2605906_Ivillage_2 46.1189220.178162.290.0236.42487985.81297_Ivillage_3 -166.441662.33432-2.670.008-289.0644-43.81888_cons -149.7518 33.88132-4.42 0.000-216.4024 -83.10121/sigma 47.764253.6489540.5861154.94238Obs summary: 227 left-censored observations at n_days<=0. 113 uncensored observations. 0 right-censored observations.
INSIGHTmarch 15, 2008 EPW Economic & Political Weekly44We are grateful to ARC-AusAID Linkage grant LP0775444 for financial support, Raj Bhatia for help with the computations and EPW for helpful comments on an earlier draft. The usual disclaimer applies. The fieldwork, and data processing and analy-sis were carried out by Raj Bhatia in consulta-tion with the authors.Raghbendra Jha (r.jha@anu.edu.au) is at the Australian National University, Canberra; Raghav Gaiha at the University of Delhi; and Shylashri Shankar at the Centre for Policy Research, New Delhi. Reviewing the National Rural Employment Guarantee Programme Raghbendra Jha, Raghav Gaiha, Shylashri ShankarThis article presents results on the participation of rural workers in the National Rural Employment Guarantee programme based on a pilot survey of three villages in Udaipur district in Rajasthan. Its focus is on participation in the nreg programme of different socio-economic groups and the determinants of the participation of these groups. It is found that the mean participation was 59 days and that targeting was satisfactory. The performance of the programme has been far from dismal. There has been a spate of comments – mostly critical – following an au-dit of the National Rural Employ-ment Guarantee (henceforth,NREG) pro-gramme by the Comptroller and Auditor General of India [CAG 2007]. This audit has revealed several weaknesses of this anti-poverty programme as well as huge leakages. For example, a bare 3.2 per cent of registered households in 200 of India’s poorest districts managed to get the guar-anteed 100 days of employment in a year.1 The average employment provided was 18 days per needy household. Another as-sessment [Biswas 2007] draws attention to the unevenness in its implementation. Emphasising that while a total estimated expenditure of $ 4.5 billion was expected to generate two billion days of employ-ment, the actual was about one billion, and the benefits varied across different states. In Uttar Pradesh, the most popu-lous state, large segments of the rural population were ignorant of the scheme. By contrast, Rajasthan was among the top performers – the average employment per participating household was 77 days of work. The share of wages was 73 per cent. The small north-eastern state of Tripura performed well too, as the average number of days of employment per rural family was 87 days. Somewhat surprisingly, Kerala – a state with a superb record of human development – was at the bottom. In fact, only one of the southern and western states (Andhra Pradesh, Gujarat, Karnataka, Kerala, Maharashtra and Tamil Nadu) – Karnataka – generated more than 10 days of employment per rural family during 2006-07, while the eastern and northern states performed better.Some encouraging features of this scheme include (i) a high share of female employment (about 40 per cent nationally rising to 81 per cent in Tamil Nadu, and a low of 12 per cent in Himachal Pradesh); (ii) 20 districts spent more than $ 25 million on this scheme, and the benefits are reflected in greater economic security, higher farm wages, lower migration, and building of infrastructure. However, no general conclusions can be drawn about the accuracy of targeting and prompt dis-bursal of wages. Two examples suffice. In Chhattisgarh, 95 per cent of wages were paid to the actual workers, while in east-ern Jharkhand the corresponding share was barely 15 per cent.2 Other failures relate to distribution of job cards – large numbers of needy households are in the queue – the selection, design and exe-cution of projects, resulting in huge leak-ages.3 More specifically, Dreze (2007) highlights a quiet sabotage of the trans-parency safeguards inNREG in western Orissa. In a survey of 30 worksites, the investigators found evidence that a con-tractor was involved in some ways. What is worse is that the job card does not have a column for “wages paid”. Even the number of days worked is hard to verify as the names of the labourer and worksite have been replaced by numerical codes. Yet Dreze (2007) and Roy et al (2008), among others remain optimistic about the potential of the programme, mainly be-cause the awareness of employment as an entitlement has grown.1 ObjectiveThe present analysis is a part of a larger project designed to assess the cost-effec-tiveness of social safety nets in three Indian states, viz, Rajasthan, Andhra Pradesh and Maharashtra. TheNREG programme is operative in six districts of Rajasthan. Our sampling strategy is as follows. Since considerable reduction in the sampling error can be achieved by increasing the number of sample districts without sub-stantially increasing the overall sample size, we have selected 50 per cent of the total districts as the first stage units from the total number of districts covered in theNREG scheme in the state. It is often advantageous to select sampling units with unequal probabilities, which reduces sampling errors. Thus, it is proposed to 
INSIGHTEconomic & Political Weekly EPW march 15, 200845select districts with probability propor-tional to size sampling at the first stage, size being the rural population/households as reported in the Census of 2001. The first set of results given below is based on a pilot survey of three villages (Dhundiya, Karanpur and Prithvisingh Ji Ka Khera) in Udaipur district, Rajas-than.4 The total number of households interviewed in December 2007 was 340. Here, the focus is on participation inNREG of different socio-economic groups and the determinants of the participation of these groups. 2 Methodology and Results First, a set of cross-tabulations are given to identify the correlates of participation inNREG. As these tabulations contain ave-rages, two econometric exercises are carried out to assess their relative importance. These involve a probit analysis of partici-pation in the NREG programme and a tobit analysis of the duration of the participation. The participation variables include a number of household specific characteristics such as caste/ethnic affiliations – i e, whether a member of scheduled castes (SC) or sched-uled tribes (ST) or “others” – educational attainment, land-owned, number of male and female adults in the household, and occupational status.5 The details of the model used and how the marginal effect of each of these factors on participation is computed are reported in Appendix 2 (p48). We present our results in two broad cat-egories. First, in our cross tabulations we report on statistics on participation in the NREG programme. Second, we model the participation of workers in the NREG pro-gramme. We report our results under these headings. Cross-TabulationsIn the cross-tabulations, an attempt is made to identify some correlates of par-ticipation and duration of participation in NREG. This is depicted in Table 1. Out of 340 households, one-third par-ticipated inNREG (Y). A vast majority of the participants belonged to “others” (about 90 per cent) and the remaining were equally divided among the SC and ST. Within each caste/ethnic group, the highest proportion of participants was among the ST, followed by “others”. The self-employed in agricultural households accounted for about 46 per cent of the participants, followed by “other labour” households. Within each occupation, the proportion of participants was, however, highest among “other labour”, followed by the self-employed in agriculture. Also, as land continues to be an important asset in rural areas, it is not surprising that the bulk of the participants (about 80 per cent) belonged to three lowest ranges of land-owned. The share of participants was highest among the (nearly) landless (about 52 per cent), followed by each of the three higher land categories. Further, about 42 per cent of the participating households had five or more members, and a little over one-fifth were small (comprising one to three members). However, the share of participants was highest among the latter (about 43 per cent).Contrary to the findings of CAG and “others”, the share of participating house-holds that worked for 90 days or more in 2007 was a little over one-fifth. About 39 per cent worked for 50 to 90 days. So a large majority worked for a fairly long duration. In fact, the mean number of days worked was high – about 59 days in the last year.Some basic characteristics of participa-tion inNREG in these three villages are given in Table 2. In Table 2, the first entry in the Nhead-ed-row indicates the number of responses (227) listing zero days. Also shown are association of participation inNREG by ethnic group. All those who worked for 90 days or more belonged to “others”. Among theSC andST, one-third or more worked for 51 to 90 days, and the majority worked for fewer days (between one and 50 days). Thus, while most groups had access to employment under the NREG, SC andST seem to have benefited relatively less. Table 2 also reports on participation inNREG by occupational category. The variation in duration of participation across occupations is striking too. All agricultural labour households worked in the range of one to 50 days while the majority of other labour participating households worked in the ranges of 51 to 90 and greater than 90 days. The majority of the self-employed in agriculture also worked in these high ranges. Among the self-employed in non-agriculture, the majority worked in the lowest range. This implies that agricultural labour and self-employed in non-agriculture relied on theNREG programme to supple-ment their incomes whereas workers in the other labour and self-employed in agriculture categories usedNREG as the mainstay of their incomes. Table 1: Frequency of Participation in NREG Characteristics N Y TotalParticipation by caste/ethnic group OT 205 102 307 SC 18 6 24 ST 4 5 9 Total 227 113 340Participation by occupation AL 4 1 5 OL 28 46 74 OT 16 3 19 SA 126 52 178 SN 53 11 64 Total 227 113 340Participation by land-owned (ha) 0 to 0.1 28 30 58 0.1 to 0.75 56 28 84 0.75 to 1.5 75 33 108 1.5 to 2.5 39 17 56 > 2.5 29 5 34 Total 227 113 340 Participation by household size 1 to 3 members 33 25 58 4 to 5 members 87 41 128 > 5 members 107 47 154 Total 227 113 340For definition see Appendix 1.Table 2: Characteristics of Duration of Participation in NREG 0 day 1 to 50 days 51 to 90 days > 90 days TotalDuration of participation (aggregate numbers) N 227 0 0 0 227 Y 0 46 44 23 113 Total 227 46 44 23 340Duration of participation by caste/ethnic group OT 205 39 40 23 307SC 184 2 0 24ST 4 3 2 0 9 Total 227 46 44 23 340Duration of participation in NREG by occupation AL 4 1 0 0 5 OL 28 13 16 17 74OT 163 0 0 19 SA 126 22 25 5 178SN 53 7 3 1 64 Total 227 46 44 23 340Duration of participation in NREG by land-owned (ha) 0 to 0.1 28 7 10 13 58 0.1 to 0.75 56 14 8 6 84 0.75 to 1.5 75 15 16 2 108 1.5 to 2.5 39 5 10 2 56> 2.5 295 0 0 34 Total 227 46 44 23 340
102 6 5 205 18 4 
INSIGHTEconomic & Political Weekly EPW march 15, 200847replaced by land-owned dummies – 0 to 0.1 ha (omitted group), 0.1 to 0.75 ha, 0.75 to 1.5 ha, 1.5 to 2.5 ha, and larger than 2.5 ha. All land dummies except that for the highest land-owned group have signifi-cant positive coefficients, implying higher probabilities of participation relative to the (nearly) landless. The probability of participation decreases with the number of adult males and females but increases with household size. The village dummies have effects similar to those in the earlier specifications. The marginal effects for the specifica-tion used in Table 5 allow us to assess the relative importance of various determi-nants of participation. As may be noted from Table 6, the highest marginal effect among the land-owned dummies is asso-ciated with the third dummy (i e, house-holds owning land between 0.75 and 1.5 ha), followed by the next higher range of land owned. The negative effect of number of adult females is larger (in absolute value) than that of adult males while that of household size is relatively small. Between the village dummies, the (absolute) effect of the third is larger. Tobit results on the determinants of duration of participation are obtained by combining the (predicted) probabilities of participation and other household and village characteristics. These results are reported in Table 7. The greater the prob-ability of participation, the longer is the duration of participation in NREG. All land-owned dummies have significant negative coefficients, implying lower durations of participation relative to that of the (nearly) landless. The larger the number of adult males and females, the longer is the duration of participation. Household size, however, has a negative effect on the number of days of participa-tion. The duration is higher in the second village and lower in the third, relative to that in the omitted village. The overall specification is validated by the chi-square specification test. 3 ConclusionsAlthough based on the evidence from three villages in one district in Rajasthan, the analysis in this paper suggests that the targeting accuracy of the NREG programme was far from dismal. First, nearly one-third of the households participated in this scheme. Second, large segments of highly disadvantaged groups such as the ST, the landless and labour households participated in it. Third, about one-fifth of the households worked for about 100 days during 2007. Further, the landless and labour households participated for long durations. Table 5: Determinants of Participation in NREG (3)Probit Regression Number of obs = 340 LRchi2(9) = 128.53 Prob > chi2 = 0.0000Log likelihood = -151.91997 Pseudo R2 = 0.2973Participant Coef Std Err z P>|z| [95% ConfInterval]_Iland_g_2 .5825464 .330061 1.76 0.078 -.0643613 1.229454_Iland_g_3 .758315 .3151825 2.41 0.016 .1405686 1.376061_Iland_g_4 .6700685 .3508152 1.91 0.056 -.0175166 1.357654_Iland_g_5 .2354679 .4000868 0.59 0.556 -.5486878 1.019624 a_m -.2107726 .1344599 -1.57 0.117 -.4743092 .0527641 a_f -.2811701 .1539963 -1.83 0.068 -.5829974 .0206572 hhsize .0881745 .0543562 1.62 0.105 -.0183617 .1947108_Ivillage_2 -.6778212.1765229-3.840.000-1.0238-.3318427_Ivillage_3 2.462004.39619546.210.0001.6854763.238533_cons -.6842052 .3430846-1.99 0.046 -1.356639 -.0117719Table 6: Determinants of Participation in NREG (Marginal Effects)Probit Regression, Reporting Marginal Effects Number of obs = 340 LRchi2(9) = 128.53 Prob > chi2 = 0.0000Log likelihood = -151.91997 Pseudo R2 = 0.2973Partic~t dF/dx Std Err Z P>|z| x-bar [ 95% C I ]_Iland~2* .218754 .1262998 1.76 0.078 .247059 -.028789.466297_Iland~3* .2815091 .1167218 2.41 0.016 .317647 .052739 .51028_Iland~4* .2556942 .1358099 1.91 0.056 .164706 -.010488 .521877_Iland~5* .087793 .1536823 0.59 0.556 . 1 -.213419.389005 a_m -.0757644 .0483398 -1.57 0.117 1.62647 -.170509 .01898 a_f -.1010695 .0553256 -1.83 0.068 1.7 -.209506 .007367hhsize .0316952 .01956911.62 0.105 5.50882 -.006659.07005_Ivill~2* -.2345264 .0572666-3.84 0.000.426471 -.346767 -.122286_Ivill~3* .7379032 .04819626.21 0.000.135294.64344.832366 obs P .3323529 pred P .3240018 (at x-bar)(*) dF/dx is for discrete change of dummy variable from 0 to 1.Z and P>|z| correspond to the test of the underlying coefficient being 0.Table 7: Determinants of Duration of Participation in NREGTobit Regression Number of obs = 340 LRchi2(10) = 170.17 Prob > chi2 = 0.0000Log likelihood = -689.42865 Pseudo R2 = 0.1099n_days Coef Std Err t P>|t| [95% ConfInterval]pp 400.242694.81154.220.000213.7315586.7538_Iland_g_2 -26.361514.07078 -1.87 0.062 -54.04123 1.318235_Iland_g_3 -37.703819.08307-1.98 0.049 -75.24361 -.1639841_Iland_g_4 -30.9998917.66301-1.76 0.080-65.746193.746412_Iland_g_5 -6.731264 16.20129-0.42 0.678 -38.602125.13957 a_m 15.85228 7.009551 2.26 0.024 2.063236 29.64131 a_f 18.04148 9.011511 2.00 0.046 .3142311 35.76873 hhsize -5.490571 2.923558 -1.88 0.061 -11.24173 .2605906_Ivillage_2 46.1189220.178162.290.0236.42487985.81297_Ivillage_3 -166.441662.33432-2.670.008-289.0644-43.81888_cons -149.7518 33.88132-4.42 0.000-216.4024 -83.10121/sigma 47.764253.6489540.5861154.94238Obs summary: 227 left-censored observations at n_days<=0. 113 uncensored observations. 0 right-censored observations.
