Does political connections and affiliation affect allocation of benefits in the Rural Employment Guarantee Scheme: Evidence from West Bengal, India #

Does political connections and affiliation affect allocation of benefits in the Rural Employment Guarantee Scheme: Evidence from West Bengal, India # Upasak Das * Abstract Decentralization at local level in developing economies can be seen as forces of social change with the power at the hands of the citizens to influence policies according to their needs. However, the problem of political clientelism may be imminent, where public resources are allocated to individuals or specific groups, who are members of the political party locally in power. In this context, using survey data for 540 rural households in the Cooch Behar district of the state of West Bengal, the paper investigates the prevalence of political clientelism in allocating works under Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA), which is implemented and carried out at the local level. It is found that households, which are well connected to the political parties have significantly higher chances of getting work compared to those, who are not connected. Similar results are found for households, who support the local ruling political party in power. Further, it is found that politically well-connected households and ruling party supporters are associated with significantly higher number of days of work. Ethnographic evidences collected from the field corroborate with these findings. As further analysis, the paper explores if political clientelism is stronger in left governed villages and finds that to be the case. The study points to the existence of political clientelism in the implementation of MGNREGA and lays emphasis on the importance of reducing rationing of labour to curb clientelism. Key words: Political Connections, Political support, MGNREGA, West Bengal JEL Codes: I38, J71, J78 # The author would like to thank Dr. Srijit Mishra, Dr. Sudha Narayanan, Dr. Subrata Mukherjee and Dr. Ranjita Chakraborty for ideas and suggestions. * The author is a PhD Scholar at Indira Gandhi Institute of Development Research, Mumbai. (Email: upasak.das@gmail.com) 1 Electronic copy available at: http://ssrn.com/abstract=2262533

I. Introduction Political parties at local level in developing economies can be seen as forces of social change. In the view of participatory democracy, decentralization is significant since it allows citizens to have more power and influence over public decision making and formulation and implementation of policies. It seeks to create a democratic environment and institution for good governance and development at the local level, which would facilitate participation of the citizen in the decision making process. With transfer of power to democratically elected local authorities, it is expected that they would act according to the needs of the people. However, the problem of political clientelism may be imminent, especially in underdeveloped and unequal communities. Public resources are allocated to individuals or specific groups, who are members of the political party locally in power. Party based clientelism may be defined as the strength of the correlation between an individual s status of membership in the political party in power and his/her beneficiary status of public programme. In exchange of political support, local leaders (patrons) extend benefits of some scarce public resources to individuals or specific groups (clients). This is a deviation from the theory, given the predictions of standard voting models, which says political leaders who are concerned with re-election would focus on delivering benefits to swing voters and not the loyalists (Downs 1957). However, political parties need to run political campaigns and rallies to convince the uninformed voters, some of whom may be swayed by the information and messages they receive from campaigns (Grossman and Helpman, 1996). Hence it is important that clients provide labour for campaigns and spread positive feedbacks about the political party at the local level. Apart from these services, it is also important that they vote for the party. Based on a theoretical model, Grossman and Helpman (1996) show that individuals and specific groups and political leaders may trade these campaign contributions and votes for benefits in form of cash transfers or in kinds. From the point of view of citizens, this is acceptable since it provides insurance and means to obtain access to scarce public resources. Prevalence of party based clientelism is fostered by low productivity, poverty and inequality (Robinson and Verdier, 2002). Higher poverty and lower productivity implies higher demand for access of public resources. In this situation, individuals would be more willing to provide the services required by political leaders in exchange of favours and security. Hence, political clientelism is more common in developing countries. 2 Electronic copy available at: http://ssrn.com/abstract=2262533

There are several studies which have pointed to the existence of political clientelism in India (Kumar, 1965; Brass, 1990). Besley et al. (2005) finds that politician households are likely to get the benefits of poverty alleviation programmes than other households. They also reported that households affiliated with the party of the local political leader have higher probability of deriving the benefits of the programme than others. Markussen (2011) finds that, in two out of four states surveyed (Kerala and Tamil Nadu), households, who are members of the locally elected presidents party are more likely to have a Below Poverty Line (BPL) card, controlling for other factors. Coming from India to the eastern state of West Bengal, the Left front Government, in power from the year 1977 to 2011 strongly pursued reconstruction of local governments and democratic decentralization strategy. The Bengal decentralization experiment is often considered as one among the successful experiments that other states would do well to follow (Bardhan and Mookherjee, 2006). However, several studies have pointed out that party based clientelism exists strongly in the state. Bardhan et al. (2008) found evidences that left voters are significantly and positively associated with receipt of benefits from the Gram Panchayat (GP). Sarkar (2006) argues that individuals in West Bengal depend on informal benefits and security from the party in exchange of his vote. Sarkar (2010) develops a theoretical framework to analyse the above phenomenon, which he argues is the cause of unexpected political stability in the state. As discussed, political clientelism may be prevalent for public programmes which are implemented and carried out by the authorities at local level. One of such programme currently in place in India is the Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA), which is a rural public works programme, implemented in India from 2006. One of the unique features of this programme is its universal self-targeting demand based nature, which entitles every household which seek work to be provided with the work from the local authorities. This leaves the programme with no provision for rationing of works, implying there should be no households, who would not get work despite seeking. However, Dutta et al. (2012) finds from a nationally representative sample that 44.4% of the rural households in India did not get work under the programme after seeking. Hence, it is clear that the households have to depend on local authorities for work, after seeking. This may give rise to the practice of party based clientelism, where the political parties might choose households, which are affiliated to the party locally in power. 3

Under this context, the paper investigates the role of local authorities in allocating works under MGNREGA. It seeks to find if households, affiliated to the local ruling political party has higher chance of getting work under the programme. Also, it aims to find if these households get work for higher number of days compared to households, who are supporters of the opposite party. Further, the paper tries to analyse if households, who attend political meetings and rallies and participate in campaigning are more likely to get the benefits of the programme. It is found that households, who support the local ruling party have higher chances of securing work under the programme after seeking for it. Also they get work for higher number of days on average than the ones, who are supporters of opposite party. Of note is the fact that households, who are well connected with the political parties through attendance in meetings and rallies also have higher probability of getting work and are associated with higher number of days of work. As further analysis, the paper then explores if Left Front governed villages suffer more from political patronage than the non-left Front governed villages and finds it to be the case. The results along with some of the ethnographic evidences point out the prevalence of political clientelism in the implementation of MGNREGA and lays emphasis on the importance of reducing rationing of labour to curb clientelism. The structure of the paper is as follows. The next section tells briefly about the programme, MGNREGA and its implementation in the area of field survey which is the Cooch Behar district of West Bengal, India. The next two sections describe the data and variables and the regression strategy used in the analysis respectively. The fifth and sixth section presents the results from the regressions and ethnographic evidences supporting these results respectively. The penultimate section shows some of the further analysis that has been carried. The last section ends with a discussion of the results obtained from the analysis and then gives a conclusion. II. MGNREGA and its implementation in Cooch Behar District Features of the Act The MGNREGA, previously called NREGA 1, was passed unanimously by the Lok Sabha on 23 rd August 2005. It was implemented in 200 rural districts of India initially on 2 nd February 2006 and has been extended to the whole country at present. Under this Act, any 1 NREGA was renamed MGNREGA on 2 nd October 2009 4

adult from a household living in rural areas, willing to do unskilled manual labour at statutory minimum wage is entitled to be employed for at least 100 days a year on public works. One striking feature of the programme is its decentralized nature, where administration and allocation of works are carried out by the elected local authorities of the respective villages. Adult members of a rural household, who are willing to do unskilled manual work would have to apply for registration in the Gram Panchayat (GP) 2. After verification of the place of residence and age of the adult members, the household is issued a job card, which is mandatory for households to work under the programme. An application has to be made to the GP or MGNREGA supervisors if the household want work, indicating the time and duration of work. Against this application, the GP has to provide work to the household within 15 days, failing to which an unemployment allowance has to be paid 3. Implementation of MGNREGA in Cooch Behar district Cooch Behar district is situated in the north eastern part of West Bengal, bounded by the state of Assam in its eastern part and Indo-Bangladesh boundary in the southern part. With the highest proportion of Below Poverty Line (BPL) families (51.81%) and agricultural labourer in the state (59.20%), proper implementation of MGNREGS becomes important for higher welfare 4. Table 1 presents the person days generated in each of the districts of West Bengal over the last 4 years. It is clear that Cooch Behar district has displayed one of the worst performances in terms of implementation. In fact, over the last 3 years, average persondays generated by the district under MGNREGA is lowest in West Bengal. Further, studies have shown that implementation of the programme has been poor in the area in other indicators like denial of work and proper wage to the labourers, providing work to the marginalized sections like Scheduled Castes/ Tribes and women and completed works 5 (Government of West Bengal 2009; Dey 2010). [Table 1 about here] 2 A GP is the primary unit of the three tier structure of the local self-government in the rural parts of India. Each Panchayat consists of few villages. 3 For more information on the programme, please refer Dey et. al (2006) 4 This statistic is according to the BPL Survey conducted in the state by Government of West Bengal in 2005. For more information, please refer http://www.wbprd.nic.in/htmlpage/bpl.aspx 5 The report by the Right to Food Campaign also highlights problems in Cooch Behar district. The URL of the report is available at http://www.indiaenvironmentportal.org.in/reports-documents/report-implementation-nregsstate-its-gaps-grievances-and-demands 5

III. Data and Variables The data used in this paper is collected from a field survey conducted from January to April 2012 in two blocks of Cooch Behar district, the Haldibari and the Cooch-Behar-I block. In terms of person days generated by the programme in 2010-11, Cooch Behar-I was among the best performing block in the district along with Mathabhanga-II and Tufanganj-II blocks. Haldibari block, on the other hand was among the worst performing block with average person days of less than 8.5 days 6. From each of these two blocks, two Gram Panchayats (GP) are chosen randomly. From Haldibari block, Dakhin Bara Haldibari and Devanganj are chosen and from Cooch Behar-I block, Dawaguri GP and Falimari GP are selected. From these four GPs, 556 households are chosen randomly and interviewed. Questions on their socio-economic and demographic characteristics are asked along with information on MGNREGA. Questions like whether the household sought work in 2011 are asked along with information on the number of days of work in the years, 2011, 2010 and 2009, the wages received and the awareness level about the programme are collected. Two interesting information collected from the survey are (i) whether the household head participates in political meetings, rallies and campaigns and (ii) the political party which the household supports at the local level. While the first question is straight forward, the second question could pose some problem if asked directly. To affirm the political support, indirect questions like (i) do you like the policies proposed by the new party that has come to power in the state?; (ii) did you agree with the rule of the incumbent political party that had been in power for the last 34 years? and (iii) do you believe in the stronger political leaders (names of the leaders) of the incumbent and the new party. The respondent generally revealed their political position after discussions for some time with these questions. Yet, 16 of the 556 households surveyed did not reveal any strong preference for the opposition or the ruling party. Hence, these households are dropped from the analysis since one of the main variables of interest is the political support of the household. So the main analysis is done using 540 households. The two primary independent variables of interest in this study are political affiliation and political strength. Political affiliation is categorized as 1 if the household supports the party which is locally in power (at the GP level) and 0 otherwise. For political strength, 3 6 Please refer http://164.100.112.66/netnrega/writereaddata/state_out/empstatusall3208001_local_1011_.html for more details. 6

categories have been created- 0 if the household head do not participate in political party meetings and campaigns; 1 if the household head participates some times and 2 if the household head participates regularly. Outcome Measures-Getting work in MGNREGA and number of days of work Since the main objective is to examine how political affiliation and connection affect households getting benefits from the programme, the outcome of interest is whether the household got work in the year 2011 given that it has sought work and the number of days of work. Information on these has been collected in the survey. Getting work under the programme has been categorised as 1 or 0 if the household got work in 2011 or not respectively. Similarly, whether the household sought work in 2011 has been categorized as 1 or 0. Further, total number of days of work in 2011 for each household is calculated and used in the analysis as a dependent variable. Independent Variables Drawing from the published literature on the models work and taking into account the Indian context, we have incorporated a number of controls. These include socio-economic and demographic characteristics of the household. To characterise the households by social group and religion, we use dummies for caste and Muslim households. The categorical variable for caste is coded into the categories of Scheduled Caste/Scheduled Tribes (SC/ST), Other Backward Castes (OBC), and Upper Castes (UC; taken as reference) which are meaningful representations of the Indian social fabric along caste lines 7. Since the two main religions of the survey region are Hindu and Muslims and the sample does not consist of households, belonging to any other religion, it is also divided into two categories, namely, Hindu (the majority religious group in the Indian population; taken as the reference category) and Muslim (largest group among religious minorities). To capture the economic condition of the household, a number of variables are used. Amount of land cultivated and number of livestock are included because they are one of the best indicators of wealth status in rural areas. House types such as non-cemented, semi- 7 It may be noted that the SCs and STs have suffered from severe social exclusion and discrimination from historical times and lag behind the OCs in the different indicators of welfare. A much more meaningful categorization would have been to divide the other castes (OCs) into Other Backward Classes (OBCs) and Upper Castes but the 1993-94 survey includes the OBCs into Others and therefore making it impossible for us to have a detailed caste classification (Deshpande, 2011). 7

cemented and fully cemented are used as one of the controls to assess the economic condition of the household. A dummy like whether the household possess BPL card is also introduced. Further the main occupation of the household is added to capture the financial health. Apart from these controls, demographic variables like age of the household head and his/her education, the gender, number of kids who do not go to school (below 7 years of age) and number of working individuals in the household are used. To control for the fixed effects at the GP level, GP fixed effects are included. To control sansad fixed effects, standard errors at sansad level are clustered in the regressions. IV. Regression Strategy Getting work This study attempts to explore if political networks and affiliation affect households getting work under the programme, once they seek for it. Since participation in MGNREGA is conditioned only to seeking, taking just the households, who sought work, might lead to the classic case of sample selection bias (Heckman, 1979). The households that decide not to supply labour in works under MGNREGA might do so when their reservation wage is higher than that offered under the programme, which is unobserved. Hence probit regression model, taking only households who sought work might lead to biased estimates. Thus, we proceed with bivariate probit model with sample selection. We estimate the probability of households, working under MGNREGA, fitting a regression model with selection by maximum likelihood (Wynand and Van Praag, 1981). The model is formulated in terms of two equations: i. a selection equation or participation equation a probit regression (binary dependent variable taking a value of 1 if the household sought work, 0 otherwise) to explain the decision of whether to seek work under MGNREGA and ii. a outcome equation- a probit regression to explain whether the household, actually got work, observable only for those, for whom the dependent variable in the above equation equals to 1. More formally, the probit model to estimate the probability of households getting work under the programme assumes that there exists an underlying relationship y' j x j + u1 j = β Outcome equation 8

such that we observe only the binary outcome, work y if y ' j > 0 However, the work y is observed only if seek = j j j y ( z γ + u2 > 0) Selection equation where u 1~ N (0,1), u2 ~ N(0,1) and corr( u 1,u2 ) = ρ Here x j are the independent variables for household, j affecting its probability of getting work, β are the coefficients and u 1 j are the error terms. z j are the independent variables affecting the probability of household, j seeking work. u 2 j are the error terms. N(0,1) represents the standard normal distribution. When ρ 0, standard probit estimations through the outcome equation taking only the households, who sought work would yield biased results. Hence, bivariate probit regression with sample selection is applied, where the inverse mills ratio, calculated from the estimation of selection equation goes into the outcome equation as an independent variable. The extent of sample selection depends on the significance level of this ratio. As required in these two step models, at least one among the independent variables, that are used for estimating the selection equation have to be excluded while estimating the outcome equation (Cameron and Trivedi, 2009). Otherwise, the model is identified by the functional form and the coefficients have no structural interpretations. Hence exclusion restriction demands a variable which influences household to seek work under the programme, but would not influence the probability of those households to get work. In this case, the two exclusion variables chosen are number of kids and number of working individuals in the household. More number of kids in the household may add to the number of dependents, which might add burden on the household earning members. Hence this might induce the adult members to seek work under the programme. Conversely, that might also imply more care needed for these children, which might induce the household not to seek work. Nevertheless, this variable is instrumental in affecting whether the household sought work under the programme. However, when the local authorities select households, who sought work for allocating the works, this variable might not be influential since, in this case, 9

local connections or networks or poverty status might influence the authorities to allocate work. Similarly, it can be argued that number of working members may affect the decision whether to seek work or not but would not influence whether the household would get work. Hence number of kids and working members in the household are chosen as the instruments, which might induce households to seek work, but would not induce local authorities to allocate works. However, it is found that the inverse mills ratio is insignificant implying no sample selection bias. Number of Days of work The second objective of the paper is to explore if political connections and support for the ruling political party at the local level induce household to get work for higher number of days. Number of days of work would be observed only for households, who got work under the programme. However, if the households, who got work are separated and regression analysis is carried out, it might again result in sample selection bias as already discussed (Heckman 1979). Hence, the best model would be to estimate the selection equation, based on the probability of households getting work and then use inverse mills ratio to estimate the number of days of work in the outcome equation. However, households, who got work, would be dependent on the condition that the household sought work under the programme. But, if we consider only the households, who sought work to estimate the selection equation, it may again result in sample selection bias. To get away with the problem, tobit model with sample selection is used. The selection model is a probit estimation of the probability of households seeking work as done in the earlier section and then using the inverse mills ratio to estimate the number of days of work using tobit regression. However, asking the households the number of days it would have worked for the whole year would not reveal correct information. It might not be possible for the respondent to know exactly how many days it would have worked given the labour they put in diversified occupations like agriculture or construction labour all through the year. But, considering field studies across and our field experiences, it may be safely assumed that poorer households would want work for higher number of days than the richer households. The problem lies in the fact that the number of days demanded by households would depend on the number of kids and working individuals in the households. So these two variables cannot be taken as instrument for the regression as in the earlier case. Hence, the dummy variable of whether the household has a job card or not, is taken as the instrument. 10

Households, who would seek work, would be in need for work and hence more likely to possess job card. However, households are eligible to get work only if they possess job card. Hence, number of days of work would be uncorrelated with possessing a job card, implying households which got work for both lesser and higher number of days would possess job card. Hence the selection model is a probit estimation of the probability of households seeking work as done in the earlier section and then introducing the inverse mills ratio to estimate the number of days worked in 2011 using tobit regression, taking all the households who sought work in the regression. In this case, inverse mills ratio comes out to be significant indicating that there would have been sample selection bias had we run the regressions directly without the two step Heckman methodology. V. Results Descriptive Statistics Table 2 lists the variables used in the regressions as well as the descriptive statistics of samples of households (i) who got work (ii) who sought but did not get work and (iii) did not seek for work. It is found that more than 46% of the sample households got work, 28% did not get work despite seeking and the remaining 26% did not seek for work under the programme. The average number of days worked by a household, who got work is about 16 days. [Table 2 about here] Coming to primary variables of interest, it is found that among the households, who got work, more than 39% of them are strongly connected to the local political parties in the sense that they regularly attend party meeting and participate in campaigns during election. However, only 25% of the households, who did not get work after seeking are strongly connected to political party and 48% of them do not attend meeting and campaigns. Further, among the households, that got work, about 70% support the ruling political party at the GP level. In contrast to this, only 32% of the households, who did not get work after seeking are supporters of the ruling party. Of note is the fact that among households, who got work, the ruling political party supporters worked for 16.4 days on average whereas the non-supporters worked for about 15.7 days. Similar results also follow for political connection statistics wherein the strongly connected households worked for 20 days on average whereas the ones, who do not attend meeting and political rallies, work for 14 days. 11

Coming to the statistics on the controls used in the regression analysis, it is found that SC and ST households constitute to 56.4% of the households, that got work and 48.7% of the households that are rationed. 27.6% of the households, which got work are found to be Muslims whereas among the households, which did not seek work, the proportion is just above 12%. As far as main occupation of the household is concerned, 32.4% of the households, who got work are agricultural labourers whereas among the households, who did not get work, the proportion is only 6.4%. This is because of the fact that MGNREGS wages locally was Rs. 130 during the survey period whereas the agricultural wage was around Rs. 140 in peak season and Rs. 110 in lean season for males. However, works in MGNREGS are less strenuous than that of being agricultural labour and the labourers are able to enjoy substantial leisure. One notable finding may be that among households, who got work, about 25% of the heads are illiterate but more than 29% of the heads are educated at least till secondary level. This is intriguing and the reason might be that better educated households are aware of the programme details and features, so they are able to extract its benefits. Hence they get the work after demanding for it. Coming to the statistics on number of days, it is found that SC/ST households work more than the upper caste households but less than the OBC households. Hindu households, on an average have worked for 1 more day than the Muslim households. As expected, agricultural labourer and non-agricultural labourer are the two main occupations of the households, who have got work for the highest number of days. Households with non-cemented houses get work for almost 19 days on average, whereas households with fully cemented houses get work for about 13 days. Households with BPL card get work for about 17 days on average and those without BPL card get about just more than 13 days. Correlation between number of livestocks in the households and number of days of work is negative suggesting that households, with higher number of livestocks get work for lower number of days. Similar results are obtained for land cultivated by the households. Regression Results (i) Getting work: Table 3 presents the probit estimation of the chances of households, seeking work under the programme taking all the households (selection equation). As expected, households with BPL 12

card have higher probability of seeking work compared to the ones who do not have it. Also, households with non-cemented houses have higher chances of seeking for work compared to the households, with cemented houses. Households, with the main occupation other than that of regular wage, have significantly higher chances to seek work compared to the ones engaged in regular jobs. Surprisingly, SC/ST households do not have higher probability of seeking work indicating that upper caste households have also sought work. However, Muslim households tend to seek work more than the Hindu households. Interestingly, it is found that female headed households have lesser chance of seeking work than the male headed counterparts and this is significant at 5% level of significance. Field investigations suggest that the former is more labour constrained than the latter as indicated by many studies (Liu and Barrett 2013, Das 2013). [Table 3 about here] Table 3 also presents the results from probit estimations of households getting work after seeking for it with and without inverse mills ratio (IMR). Since IMR from the selection equation is insignificant, estimations with and without IMR is presented. In the first model, both political connections as well as political support are incorporated. In the second specification, only political connection is included and in the third, only political support is kept in the model. Control variables are included in each of the models. Estimations for the primary variables of interest seem to remain fairly unchanged along with their significance level. It is found that households with strong political connections (participate in political meetings and rallies regularly) have significantly (1% level of significance) higher chances of getting work from the local authorities compared to households which do not participate in the rallies and meetings. Households which participate in these meetings and rallies at times also have higher chances but the results are significant at 10% level of significance only. It is also households, which support the political party locally in power have significantly higher probability of getting work after seeking for it. The results show that political clientelism exists and influences demand based welfare schemes even. Coming to the control variables used in the analysis, it is found that SC/ST households seem to have significantly higher chances of getting work after seek for it than the upper caste households. Similar findings are also obtained for Muslim households. Unlike the case of selection equation, poorer households like the ones with the main occupation as 13

agricultural labour are not found to have significantly higher chances of getting work compared to the regular wage ones except for some specifications. Households with BPL cards are also found not to have significantly higher probability of getting work compared to the non BPL households. Further, it is found that households with non-cemented houses also do not have significantly higher chances of getting work after seeking compared to households with cemented households, which can be considered richer than the households with non-cemented houses. Similarly, variables like number of livestock and land cultivated by the households are not found to be positive and significant. These findings indicate that poorer households do not have significantly higher chance of getting work after seeking. It is also found that households, which worked in the year 2010 have significantly higher chance of getting work in 2011 even. (ii) Number of days of work Table 4 presents the results from tobit regressions for the number of days of work in the year 2011 by households taking those, who sought work. In the first model, both the primary variables of interest are kept- political connections and political support. In the second model, only political connection variables are included and in the third model, only political support is incorporated. Control variables are included in each of the three models. As discussed the selection model is the probit estimations of the probability of households seeking work under the programme. The instrument variable used is the dummy of whether the household has a job card. [Table 4 about here] Results show that political clientelism exists and matters even while allocating the number of days of work to the household. It is found that houesholds, which support the local ruling political party are associated with significantly higher number of days of work than those, which support the opposite party (at 1% level of significance). Further, it is found that households, which regularly attend political meetings and participate in political rallies, get significantly higher number of days to work for than the ones, which do not involve in rallies and meetings (at 1% level of significance). However, no significant difference in the number of days of work is found between households, which partially attend these meetings and rallies and those, who do not attend at all. The results on these variables are robust across all specifications. 14

Coming to the controls, SC/ST households work for higher number of days on average than the upper caste households and this is significant at 1% level. OBC households also work for higher number of days but no significant relationship is found between the two. Muslim households on average are found to work for significantly higher number of days than Hindu households. It is also found that households whose main occupation is agricultural labour are associated with higher number of days of work than households primarily engaged in regular jobs. Interestingly, households with BPL cards do not have significant relationship with the number of days of work. This indicates that either poorer households do not get work for higher number of days or allocation of BPL cards is associated with inclusion and exclusion errors. Field experiences suggest that both of these effects play a role. It has been widely found that households with no land and working as casual labour have got work for only 6 days of work in the whole year. Also, similar households are found to have no BPL cards while households with more than 10 bighas of land possess BPL cards 8. Further, wide literature also points towards the evidence of high exclusion of poor households from the list of BPL card holders (Dreze and Khera, 2010; Swaminathan, 2010). VI. Ethnographic Evidences Interviews and focus group discussions during the field survey revealed information and observations, which closely match with the findings from the econometric exercises, discussed in the earlier sections. Some of these evidences are presented as follows: (i) Mr. ABC (name not revealed) lives in a cemented house at the Devanganj GP of the Haldibari block with his mother. The family members are relatives of the head of panchayat in the village. The household received 65 days of work in the year, 2010 and 18 days in 2011. According to the information retrieved from the MGNREGA website, this household got the highest number of days of work in these two years 9. (ii) Ms. XYZ (name not revealed) lives with her two daughters aged 14 years and 13 years and one son who is of age 11 years. Her husband is suffering from heart disease and is not able to work since one and half years. Hence she works as a domestic help and lives in dilapidated house in the Dakhin Bara Haldibari GP. Two years back when 8 Bigha is the traditional unit of measurement of land. In West Bengal roughly 3 bighas equal 1 acre of land. 9 The information can be retrieved from http://nrega.nic.in/netnrega/home.aspx 15

her husband s health condition was fine, he used to participate actively in the opposite party s meetings and campaigns and hence was strongly connected to politics. However, since he is sick, he can no longer remain connected to the party and Ms. XYZ cannot participate in the meetings due to the daily works. According to her, since her household is historically the opposite party supporter and used to participate in the political rallies and campaigns, she does not get work even after repeatedly seeking for work over the last couple of years. (iii) Mr. MNO from the Falimari GP lives with his wife and mother along with two daughters and a son all of whom are of age below 10 years. He is a casual labour and at times migrates to the city of Siliguri, which is about 100 kms from the village to work as ricksaw puller. His family is the supporter of the political party, who is opposite to the party in power at Falimari GP. However, he cannot remain close to the party through regular attendance in meetings and rallies. According to him, the panchayat authorities have not allocated work to his family. Had he been the supporter of the ruling party in power, his family would have got work and that too for much higher number of days under the programme. Similar experiences have been shared by many other households even who told that households, which supports the local political party in power get work after seeking and for higher number of days of work. VII. Further Analysis Among the two blocks surveyed, the Devanganj and Dakhin Bara Haldibari GPs of the Haldibari block was governed by the Left Front government in 2011 whereas the Dawaguri and the Falimari GPs of the Cooch Behar-I block was governed by the Trinamool Congress (TMC). As further analysis, we check how political connections and alliance affect allocation of benefits in GPs governed by these two parties in the surveyed area. Table 5 presents the estimation results for the probit model as well as tobit regressions as done in the earlier section. In the Left front ruled GPs, invasion of political clientelism is much prominent. It is found that households, supporting the Left government have significantly higher chances of getting work once they seek for it (at 1% level of significance). Also, households, who are close to the political parties and attend meetings and campaigns regularly, have significantly higher probability of getting work. Of note is the fact that the households, who participate in these meeting and rallies at times, also have higher 16

chances of getting work than those, who do not attend these meetings and rallies at all. Further, these politically well-connected households are associated with significantly higher number of days of work than households, who are not politically strong. The Left front supporter households also do get higher number of days for work as compared to the nonsupporter households and this finding is significant at 1% level of significance. [Table 5 about here] Coming to the TMC dominated GPs, estimations for political connections show insignificant findings. It is found that households, which are politically well connected through regular participation in rallies and meetings do not have significant chances of getting work after seeking compared to households, which do not attend these meetings and campaigns. Households, which are partially well connected also do not have significant chances of getting work. As far as the number of days of work is concerned, politically wellconnected households are not significantly associated with higher number of days of work compared to the ones, which are not politically connected. However, it is found that households, who support TMC at the local level have significantly higher probability of getting work compared to households, who do not support the party. These households also are found to have significant association with higher number of days of work than the ones, who support opposite political party at the local level. The estimations in both the cases are found to be significant at 5% level of significance. VIII. Discussion and Conclusion Decentralization at local level in developing economies can be seen as forces of social change since it allows people to have more power through which it can have influence over public decision making and implementation of policies, according to their needs. With the transfer of power to democratically elected local bodies, it is expected that they would work according to the requirements of the citizens at the local level. However, the problem of political clientelism may be imminent, where public resources are allocated to individuals or specific groups, who are members of the political party locally in power. Political parties need to run political campaigns and rallies, which involves spreading positive feedbacks about the party and also vote for them. Hence political parties and individuals may trade campaign contributions for benefits in form of cash transfers or in kinds. In this context, the paper investigates the role of local authorities in allocating works under MGNREGA, which is implemented and carried out at the local level. It seeks to find if households, affiliated to 17

the local ruling political party has higher chance of getting work under the programme. Also, it aims to find if these households get work for higher number of days compared to households, who are supporters of the opposite party. Further, the paper also tries to analyze if households, who attend political meetings and rallies and participate in campaigning are more likely to get the benefits of the programme. As further analysis, the paper then goes on to find if left front governed villages suffer from this problem more than the no left governed villages. Using bivariate probit with sample selection model, it is found that political connections and affiliation of the household do matter in getting work after seeking for it. Households, which attend political meetings and rallies and participate in campaigns have significantly higher chances of getting work compared to those, who do not attend these meetings and rallies. Further, it is found that households who support the local ruling political party also have significantly higher probability of getting works under the programme than households, who support the opposite political party. The paper, also tries to explore if, among the households, who got work, political connections and affiliation matter in getting work for higher number of days. Using tobit model with sample selection, it is found that households who are strongly connected to political parties through participation in meeting and campaigns are associated with significantly higher number of days of work than those, who do not participate in these meetings and rallies. Also, it is found that supporters of the ruling political party at the local level work for significantly higher number of days of work than the non-supporters. The paper contributes to the growing literature on political clientelism and is among the first attempts of how can it affect households getting benefits from MGNREGA. It finds evidences that political connections and affiliation of household play a major role in allocating benefits of the programme. The paper lays stress on the fact that despite being a demand based programme, which gives equal legal rights to work to the rural households, party based clientelism is prevalent. This arises especially when demand for work exceeds supply leading to rationing of labour. In other words, when local authorities have to choose from a pool of workers who have sought work, the problem of selection based on party affiliation and connections may surface. Surveys across India also show high unmet demand in different parts of the country (Gaiha et al. 2010; Das et al 2012). Hence it becomes important for policy makers to nullify negative implications of rationing and lays emphasis 18

on the importance of reducing unmet demand to realise higher positive impacts. In this respect, the recent move by the government to widen the scope of permissible works, which includes watershed and agricultural related works under MGNREGA is welcome (Government of India, 2012). The step is expected to reduce the rationing and hence preferential treatment based on political party alliance and closeness. References Bardhan, P., Mitra, S., Mookherjee, D. and Sarkar, A. (2008) Political participation, clientelism and targeting of local government programs: analysis of survey results from rural West Bengal. Mimeo, Boston University. Besley, T., Pande, R., and Rao, V. (2005) Political selection and the quality of government, in: World Bank, The Political Economy of Gram Panchayats in South India: Results and Policy Conclusions from a Research Project (Washington, DC: World Bank) Brass, P. (1990) The Politics of India since Independence. 2nd edition, Cambridge: Cambridge University Press. Cameron, A.C. and Trivedi, P (2009). Microeconometrics: Methods and Applications, New York: Cambridge University Press. Das, U. (2012) Accuracy of Targeting and Implications of Rationing in Mahatma Gandhi National Rural Employment Guarantee Scheme: Evidence from West Bengal, India. Available at SSRN: http://ssrn.com/abstract=2157089 or http://dx.doi.org/10.2139/ssrn.2157089 Das, U., Singh, A. and Mahanto, N (2012). Awareness about Mahatma Gandhi National Rural Employment Guarantee Act: Some evidence from the northern parts of West Bengal, India, Economics Bulletin, 32(1), pp. 528-537. Deshpande, A. (2011) The Grammar of Caste. New Delhi: Oxford University Press. Dey, S. (2010) Evaluating India s National Rural Employment Guarantee Scheme: The Case of Birbhum District, West Bengal, International Institute of Social Studies, Hague, Netherlands, Working Paper No. 490. 19

Dey, N., Dreze J. and Khera R. (2006) Employment Guarantee: A Primer. New Delhi: National Book Trust. Downs, A. (1957) An Economic Theory of Political Action in a Democracy. Journal of Political Economy, 65(2), pp. 135 150. Dreze, J. and Khera R. (2010) The BPL Census and A Possible Alternative. Economic and Political Weekly, 45(29), pp. 54-63. Dutta, P, Murgai, R., Ravallion, M. and Van De Walle, D. (2012) Does India s Employment Guarantee Scheme Guarantee Employment?, Economic and Political Weekly, 47(16), pp. 55-64. Gaiha, R. Shankar, S. and Jha, R. (2010) Targeting Accuracy of the NREG: Evidence from Rajasthan, Andhra Pradesh and Maharashtra, ASARC Working Paper 2010/03, Australian National University, Australia South Asia Research Centre Government of India (2012). Report of the Committee for Revision of MGNREGA Operational Guidelines, Ministry of Rural Development. Government of West Bengal (2009) Implementation of the National Rural Employment Guarantee Act, 2005 in West Bengal during the year, 2008-09, Panchayat and Rural Development Department, Government of West Bengal. Grossman, G.M. and Helpman, E. (1996) Electoral Competition and Special Interest Politics. Review of Economic Studies, 63(2), pp. 265 286. Heckman, J.J. (1979). Sample Selection Bias as a Specification Error, Econometrica, 47(1), pp. 153-61. Kumar, D. (1965) Land and Caste in South India: Agricultural Labour in the Madras Presidency during the Nineteenth Century, Cambridge: Cambridge University Press. Liu, Y. and Barrett, C.B. (2013) Heterogeneous Pro Poor Targeting in India s Mahatma Gandhi National Rural Employment Guarantee Scheme, Economic and Political Weekly, 48(10), pp. 46-53. 20

Markussen, T. (2011) Inequality and Political Clientelism: Evidence from South India, Journal of Development Studies, 47(11), pp. 1721-38 Robinson, J.A. and Verdier, T. (2002) The Political Economy of Clientelism. CEPR Discussion Paper. Sarkar, A. (2006) Political Economy of West Bengal: A Puzzle and A Hypothesis, Economic and Political Weekly, 41(04), pp. 341-48. Sarkar, A. (2010). On the Political Economy of a Backward Region, Indian Growth and Development Review, 3(2), pp. 122-137. Swaminathan, M. (2010) A Methodology Deeply Flawed. The Hindu, February 5, accessed from http://www.thehindu.com/opinion/lead/article100972.ece (viewed on 10th April, 2013) 21