Credit Groups, Women s Political Engagement and Public Goods Provision

Transcription

1 Credit Groups, Women s Political Engagement and Public Goods Provision Sabyasachi Das Pushkar Maitra Paromita Sanyal Indian Statistical Institute, Delhi Monash University Florida State University July 26, 2016 Abstract We examine whether membership in Self-Help Groups (SHGs), the predominant type of microcredit group in India, affects women s political engagement via attendance in village assemblies, participation in political party activities, and membership in village level political committees. We also examine the influence of women s village assembly attendance on public goods provision by the village council. Using pan-indian survey data and an IV approach, we find that SHG membership increases women s attendance in village assemblies. This improves the gender composition of village assemblies in women s favor, which in turn changes the composition of public goods provided by village councils towards more women preferred ones. Membership of microcredit groups also leads to increased long term political engagement of women. Microcredit groups, therefore, contribute to democratic deepening by indirectly promoting the participation of women in political institutions and processes. Key words: Self-Help Groups, Village Assemblies, Political Engagement, Public Goods, Gender, India JEL Codes: O12, P48, H41 We would like to thank Vegard Iversen, Tarun Jain, Takashi Kurosaki, Dilip Mookherjee, Abhiroop Mukhopadhyay, Saurabh Singhal, Marian Vidal-Fernandez, Sujata Visaria and seminar and conference participants at the Indian Statistical Institute Kolkata, the Jawaharlal Nehru University in Delhi, the University of Sydney, Hong Kong University of Science and Technology, Hitotsubashi University and the Monash-CDES workshop on Sustainable Development for their comments and suggestions. Part of the research was conducted when Pushkar Maitra was visiting the Hitotsubashi Institute for Advanced Studies (HIAS) at Hitotsubashi University. Sabyasachi Das, Economics and Planning Unit, Indian Statistical Institute, New Delhi, , India. s.das@isid.ac.in Pushkar Maitra, Department of Economics, Monash Business School, Monash University, Clayton Campus, VIC 3800, Australia. Pushkar.Maitra@monash.edu Paromita Sanyal, Department of Sociology, Florida State University, Tallahassee, FL, USA. psanyal@fsu.edu 1

2 1 Introduction The world of politics across the globe is heavily gendered. Women s participation in the institutions of democracy at the national level, such as parliaments, continues to be woefully low. 1 In lower level institutions of democracy, i.e., at the provincial, district, municipal, or village levels, women s political representation is low (in unconstrained elections), though cross-country figures on this are harder to obtain. Studies of women s political participation have usually focused on gender-based affirmative action and its consequences for achieving gender equity, for efficacy in public goods provision, and for changing perceptions about women leaders. A large part of the research in this area has been conducted on India, where gender-based affirmative action in policies (particularly in village level local governments) has been in place since the 1990 s. Researchers have examined the performance of female local government leaders in prioritizing public goods and targeting local subsidy programs meant for disadvantaged sections and also whether having a woman as the head of the village council promotes women s attendance in village assemblies (Chattopadhyay and Duflo, 2004b, Besley, Pande, Rahman, and Rao, 2004, Ban and Rao, 2008, Bardhan, Mookherjee, and Parra Torrado, 2010, Beaman, Duflo, Pande, and Topalova, 2010). However, the political participation of ordinary women in local political processes as lay citizens has rarely been studied. In this paper we move away from the dominant approach of studying gender-based affirmative action in politics and, instead, focus on alternative social forces that may help ordinary women participate in grass-root political processes. The motivating idea is that participation in locallevel democratic processes might increase the experience and exposure ordinary women have to politics, and this may lead to increased long-term political engagement for women. In the long run, it may result in a superior selection of women into electoral politics, leading to more effective representation. Also, in countries where crucial public goods are distributed through local political institutions that incorporate models of deliberative and participatory democracy, whether ordinary women participate or not can have important effects on public goods provision. Finally, women s participation in local political processes might transfer onto participation in other consequential spheres, like participatory community development projects that are funded by international development agencies and are targeted at disadvantaged communities in developing 1 For example, in 2015, only 22.7% of the Parliament members in the world were women. 2

3 countries (Mansuri and Rao, 2013). Specifically we look at the role played by microcredit self-help groups (henceforth SHGs) in getting women involved in local political processes. While generally women become members of SHGs in order to improve the financial status of their households, we argue that these groups have an important associational aspect and can potentially promote active citizenship by women. Using survey data collected from 17 major states of India we find that membership of SHGs is strongly and positively associated with increased attendance in village assemblies. The correlation between membership in SHGs and village assembly attendance is stronger in states with worse gender norms compared to states with better gender norms. This suggests that the associationism through SHG membership is more important and has a stronger role to play when norms and institutions are biased against women. To account for potential selection into SHG membership, we use an IV approach. The IV results show that SHG membership leads to increased participation in village assemblies. This, in turn, significantly changes the gender composition of attendees in these village assemblies in favor of women and results in a significant increase in the provision of women preferred public goods. Additionally, we find that membership in SHGs fosters long-term engagement of women in politics, providing pathways for women to become effective leaders. We argue that SHGs contribute to democratic deepening (Heller, 2000, page 487) by indirectly promoting the participation of discriminated groups, such as women, in political institutions and processes. In India, two important developments have taken place in parallel since the nineties. In 1992 (as a part of the 73 rd amendment to the Indian constitution), the decentralization act constitutionally mandated that village assemblies (or gram sabhas or gram sansads) were to be held in all villages across the country at least twice and up to four times a year, and incorporated them into the system of governance. Village assemblies were made responsible for oversight of government funded public works, for providing public inputs on community needs and problems, and for beneficiary selection for government sponsored schemes and subsidies aimed at disadvantaged groups. These assemblies are supposed to be attended by the general public, by elected local government leaders, and by government bureaucrats and staffs from various departments. Overall, village assemblies inaugurated an era of public participation in local politics and in local development planning. Simultaneously, the early nineties also marked the start of a non-commercialized model of mi- 3

4 crocredit: lending through the formation of women s self-help groups or SHGs. These SHGs are microfinance groups organized by non-governmental organizations (NGOs) or by community mobilizers under state-sponsored anti-poverty programs. They can be thought of as village based women s organizations that are facilitated with access to credit as their center-piece. SHGs in India are linked to banks, receive bank loans, and have the potential for becoming a channel for non-public service delivery. 2 SHGs require compulsory participation of group members in regular meetings at the neighborhood level and participation by group leaders in monthly meetings at the cluster level. Members have a compulsory savings requirement that can be used to alleviate the credit needs of the group members through internal lending. Starting in the Southern Indian states, SHGs gradually spread across many states of the country as a result of development NGOs working to promote them and because of states incorporating them as a pillar of their development strategy. Sociological work on SHGs has argued that these groups have inaugurated the opportunity for women to participate in an associational life outside of the sphere of family and kinship. This associational mechanism raises the possibility that women s participation in SHGs may foster their capability for engaging in deliberations in village assemblies (Sanyal, 2009, 2014). Additionally, a significant body of literature suggests there is a relation between associationsim and democracy, that participation in associations promotes active citizenship (Cohen and Rogers, 1992, Gutmann, 1998). Contemporary research on SHGs has shown that they have various effects outside of their intended economic goal. Sanyal (2009) and Sanyal (2014) in a qualitative study of SHGs in West Bengal found that SHGs foster women s agency through the associational rather than the financial mechanism and provide a pathway from credit to collective action and also foster women s civic participation, i.e. their participation in Indian village assemblies. Desai and Joshi (2014) using data from a randomized control trial conducted in Rajasthan found that compared to women in control villages, women in treatment villages are more likely to engage in civic activities. 3 Casini, Vandewalle, and Wahbaj (2015), using survey data from Orissa, argue that SHG members, by undertaking collective action and contributing to the local public good, incentivize 2 For example, group members can contract for collective activities, like providing child care or cooking mid-day meals for schools and also perform important social activities. Microfinance groups studied by Rai and Ravi (2011) in India also provide health insurance in addition to credit. There is supporting evidence from other countries as well. Ugandan microfinance members form associations called Munno Mukabi, which organize social functions such as weddings and burials (Sebstad and Cohen, 2001). In Kenya, about one fourth of the Roscas in Kibera invest in long term projects, health insurance or self-employment schemes (Anderson and Baland, 2002). 3 The experiment was conducted in conjunction with SEWA, one of the leading NGOs in India, in Dungarpur, one of the poorest districts of India. 4

5 local ward members (elected to the village council) to also do the same. 4 Khanna, Kochhar, and Palaniswamy (2015) evaluate the Tamil Nadu Empowerment and Poverty Reduction (Pudhu Vaazhvu) Project (PVP) and find that the program had significant and positive impacts on both women s empowerment and agency, including a higher tendency in reporting issues of local service delivery and women s public safety; a significant increase in propensity to approach the local government to solve these problems; and a significant increase in their participation in village assemblies. Women exposed to the program also report significantly greater agency in key intrahousehold decisions ranging from purchase of durable assets to children s education and their own occupational choices. Datta (2015) using data from JEEViKA, a large scale rural livelihoods program in Bihar, found that the program had significantly increased women s empowerment even though the effect on productive activities and accumulation on assets was weaker than expected. We add to this body of work by examining the impact of SHG membership on women s political engagement and the impact of such participation along different dimensions. Finally since these results are not specific to any particular NGO and not geographically restricted, this paper also contributes towards the external validity of the argument. 2 Background 2.1 Deliberative Democracy in India The Panchayat is a system of local governance in India involving the village (Gram Panchayat), the block (Panchayat Samiti) and the district (Zilla Parishad). The Gram Panchayat or the village council (henceforth GP) is the lowest form of local governance in the country with members of the council being directly elected by villagers. Each GP covers about individuals across one or more villages. These GPs are responsible for the provision and maintenance of local public goods. While the panchayat system has existed in the major provinces of India since the mid 1950 s, this system of governance was not effective until the 1990 s. The 73 rd amendment of the Indian constitution enacted in 1992 established this three tiered system of local governance as universal, which all states of the country were required to implement. The GP was given the primary responsibility of provision of public goods (like building of hand pumps, sanitation 4 All the SHGs studied by Casini, Vandewalle, and Wahbaj (2015) were created by the NGO PRADAN (Professional Assistance for Development Action) in two impoverished districts of Orissa. 5

6 facilities, local roads, public irrigation system) and also given the responsibility of identifying those villagers most in need of government support (for example identifying households that are eligible for the below poverty line or BPL card). While the GP has very little revenue raising capacity, it has some freedom to allocate funds (which comprises of central and state level grants) to what it sees as relevant and important for the local population. What is relevant and important for the villagers is, at least partly, determined through public consultation. Every GP is required by law to organize at least 2 and up to 4 Town Hall meetings (or village assemblies) every year. Any eligible voter within the jurisdiction of the GP can attend these village assemblies, can deliberate on the agenda and raise new issues. In these meetings development plans are ratified, decisions are taken regarding the provision of public resources and infrastructure, budgetary allocations are discussed and approved through deliberation. The village councils also use these meetings to select beneficiaries for government programs targeted towards the disadvantaged households. Key local elected politicians and local bureaucrats use these meetings to announce government funded policies and projects. Villagers are expected to attend these meetings and have a say in determining which policies are ultimately selected and who is selected to receive the different kinds of government funded anti-poverty support. Any specific project that gets approval from the majority of the meeting attendees is noted down for action and used in the preparation of the annual plans of the council. The village council is required by law to get the annual plan approved at the meeting before it sends it up to the next level of government to request for funds. Village assemblies therefore have a critical role to play in this framework of local governance. While village councils are required by law to typically organize at least two such village assemblies during the year and every villager is expected to attend, women s attendance in these meetings have traditionally been very low. This is where Self-help groups could have an important role to play. It is argued that that through membership in these groups women have become aware of their equal right to participate in these forum and the role they can play in determining how government funds are allocated. We classify the different benefits villagers receive from the local government into two broad categories. The first class of benefits are targeted towards households, which we refer to as household benefits. These include provision of housing, drinking water supply, sanitation facility in the 6

7 dwelling and allocation of BPL (below poverty line) card to the household. These are government funded subsidized goods for the household; i.e., publicly funded private goods. The second category includes public goods that benefit the whole community, such as provision of hand pumps in the neighborhood (for drinking water), public sanitation and sewage facilities, provision of roads, medical centers, schools, and investment in irrigation (canals). 5 These are referred to as community level public goods. 2.2 Self-Help groups in India The Self-help group (or SHG) model is the dominant form of microfinance in India, both in terms of outreach and overall loan disbursements. The SHG model evolved in the early 1990 s following the Reserve Bank of India s directives to all nationalized banks in the country to lend to informal groups. In 1992 India s National Bank for Agricultural and Rural Development (NABARD) initiated its bank-shg linkage program and over the last two decades this program has evolved into one of the world s largest microfinance networks. During this period several large development programs like the Swarna Jayanti Swarojgar Yojana (part of the Integrated Rural Development Program) and the National Rural Livelihood Mission have utilized these SHGs. As of March 2015, there are nearly 77 lakh SHGs that are linked with formal financial sector institutions with total loan outstanding to SHGs amounting to more than 51 crores (see NABARD, 2015, page 10). A typical Indian SHG consists of women members from similar socio-economic backgrounds who usually meet once or twice a month to pool their savings and discuss matters that are of interest to them. The pooled savings is deposited into a group bank account and can be used by group members in times of need. They can receive loans from the bank via the bank-linkage program. The groups are typically facilitated by NGOs who also oversee their activities. SHGs can contract for collective activities (like providing child care or cooking mid-day meals for schools). Participation in weekly or bi-weekly group meetings is a crucial part of being a member of a SHG. At these meetings women discuss their loan requirements, repayment obligations and difficulties and other matters of mutual interest. They also have to advocate for their demands, give reasons for their loan needs or repayment problems, and have to make judgements on the demands and 5 We look at these public goods because these are the 6 most demanded ones in the sample, as reported by the survey participants. 7

8 reasons given by other group members. In many parts of India, the gender context within which SHGs operate is adverse for women. The patriarchal nature of society and rules of marriage, like early marriage and village exogamy, combine to restrict women s freedoms and capabilities. In real life this means, girls marry at an adolescent age and move to their husband s village to join a family of strangers. 6 This transition is typically followed by restrictions being placed on newly married women on their mobility, interactions with outsiders, and on their general freedom of association. Consequently, their ability to develop meaningful ties with other women that can benefit them in times of crises or help them mobilize in response to personal troubles and public problems is severely limited. For the most part, their voices go unheard and their presence and participation in public sphere institutions is negligible. It is because of this adverse context that SHGs, which bring women together into a membershipbased collective with an associational life, may have an important role to play. They may have the unintended effect of empowering women, enhancing their social networks, and even generating social capital among women. There is evidence that membership of SHGs can increase collective action by women in support of issues like prevention of domestic violence, abuse of women, alcoholism and dowry related conflicts and increased provision of women centric public goods. Additionally membership of SHGs can generate social capital within the community by internalizing trust, information and willingness to help each other. See for example Sanyal (2009) for qualitative evidence and Field, Feigenberg, and Pande (2013) for experimental evidence from West Bengal in India. 3 Data and Descriptive Statistics We primarily use the Rural Economic and Demographic Survey (REDS), 2006 dataset for our work. The National Council of Applied Economic Research (NCAER) has been carrying out this survey since But only the most recent round asks the household members about their Gram Sabha attendance and participation information and hence is relevant for this exercise. The survey covers about 8,600 households spread across 242 villages spanning 17 major states of India. 7 Even 6 Rosenzweig (1988) shows that in India marriages are typically arranged across villages. 7 The states are Andhra Pradesh, Bihar, Chhattisgarh, Gujarat, Haryana, Himachal Pradesh, Jharkhand, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Orissa, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West 8

9 though the specific survey began in 2006, most of the households were surveyed in The data that we use in this paper is collected using three different surveys. The first is a household survey, which collects information about household assets, demographics, household members education, occupation, voting behaviour, attendance in the last four gram sabhas, benefits received from the GP. The second survey is the village survey and contains data on stock of public goods in the village, street or neighbourhood level construction of public goods under past three Panchayats, information on local elections such as candidate characteristics and vote shares. The last survey is the listing survey and is a census of all households in each of the surveyed villages. Table 1 presents selected descriptive statistics on some of the key variables of interest in our analysis. The average age of the respondents is 38 years, 15% belong to Scheduled Castes (SC) and 17% belong to Scheduled Tribes (ST). 76% of the survey respondents are married and they have on average 5.5 years of schooling. On average households own 3.5 acres of land though this is subject to considerable variation. Overall attendance in village assemblies is a low 13% and there is considerable gender divide in attendance rates in village assemblies: while 10% of women report attending any village assembly in the year prior to the survey, 26% of men do so. 8 The average woman to man ratio of attendees at village assemblies is a low 0.21 indicating that for every woman present at a village assembly there are 5 men present. This is clearly a reflection of societal norms: local politics in India is the man s domain. Attendance by the disadvantaged groups (SC s and ST s) are no different from the sample average. Meeting attendance rates are of course lower compared to voter turn-outs in elections. However participating in a village assembly is likely to be a costlier exercise compared to voting as it involves potential conflicts and arguments. 13% of women were members of a SHG under the previous GP and on average 10% of all household members were members of SHGs: each household thus primarily had one SHG member should they choose to join a SHG. 88% of survey respondents report that a SHG is present in the village. There is no gender difference in the extent of female and male preferred public goods provided by the current GP, with 68% (70%) of respondents reporting that any women (men) preferred public goods are provided under the current GP. To ensure that our results are not the artefact of the dataset that we use in this paper, we ex- Bengal. 8 These attendance rates are consistent with those from other countries with similar institutions (see for example Das, 2016). These attendance rates (by gender) are also consistent with those reported by Banerjee, Deaton, and Duflo (2004) and Besley, Pande, and Rao (2012) using different data sets. 9

10 amine the robustness of these results using two rounds of the India Human Development Survey (IHDS) data. These surveys were conducted in 2005 and This is a nationally representative, multi-topic survey of 41,554 households in 1,503 villages and 971 urban neighbourhoods across India collected by the National Council of Applied Economic Research and the University of Maryland. Around 83% of the households were re-interviewed for the second wave of data, and the response rates were more than 90% for both waves. The survey collected information about health, education, employment, economic status, marriage, fertility, gender relations, social networks, membership in groups and associations (including in SHGs) and political engagement. The information on social networks, membership of groups and associations and political engagement is however not particularly detailed. Therefore we continue with the REDS data for our primary analysis. 4 Reduced Form Analysis: Effects of SHG membership on political engagement of women Our focus in this paper is to examine the effect of participation in SHGs on political engagement (captured by attendance in village assemblies) and ultimately on service provision by the village council. The first step is to examine the link between participation in SHG and political and civic engagement. Our story is one of network formation through participation in SHGs, which empowers women and ultimately encourages them to attend the village assemblies. This kind of social network formation is not immediate and to capture this time dimension we look at the effect of SHG membership under the previous GP (time t 1) on current (period t) attendance in the village assemblies. Our estimating equation is given by: m ihvd,t = γ w1 SHG ihvd,t 1 + X ihvd β w1 + Y hvd β 2 + ζ v + ɛ ihd i W (1) Here m ihvd,t denotes village assembly attendance by woman i in household h in village v in district d at time t; SHG ihvd,t 1 is an indicator of SHG membership of woman i at time t 1 (in the previous GP regime). The regressions include a set of individual and household controls (X ihv and Y hv respectively) and a set of village fixed effects (ζ v ) to control for unobserved village characteristics. The survey asks about attendance in the previous four village assemblies. Using this information we construct two different measures of attendance: the first is the fraction of village assemblies 10

11 attended and the second is an indicator variable whether or not the individual attended any (of the last four) village assemblies. The corresponding regression results are presented in columns 1 and 3 of Panel A of Table 2 for the two different measures of attendance. The results in column 1 show that the fraction of village assemblies attended is 7 percentage points higher for women who are members of SHG; those in column 3 show that the likelihood of attending village assemblies is 8.5 percentage points higher for women who are members of SHG. This is a large effect: on average 13 percent of women report attending any village assembly. Membership of SHGs increases the likelihood of attending any village assembly by almost 66%. unintended. Given the typical focus of SHGs, this effect is possibly 4.1 Selection Bias: First look Of course the selection bias both at the individual and the household level could imply that the correlation between SHG membership and attendance of village assemblies is spurious. At the individual level this selection bias can arise from the fact that women who are more engaged, politically or socially, are likely to both become members of SHG and also attend village assemblies. At the household level again some households might be more engaged socially and politically; therefore in such households we would expect to find greater SHG participation rate and also greater political engagement. We address the potential selection issues in a number of different ways. The first is a naive look at observables. The results presented in Table 3 show that the observable characteristics that effect SHG membership are systematically different from those that effect attendance in village assemblies: the same set of variables never affect both membership of SHGs and attendance in village assemblies and never in the same direction. Our results are therefore unlikely to be driven by selection on observables. This still leaves us with selection on unobservables. Suppose there are some household specific (unobserved) omitted variables that are positively correlated with both SHG participation and attendance in the village assemblies. For example suppose the household is highly pro-social. This is likely to affect both membership of SHGs and also attendance in the village assemblies, 11

12 for both male and female household members. To capture this effect we estimate an extended version of equation (1) on the sample of males and females. We include the number of other women in the household who were members of a SHG in the previous GP (OSHG ihvd,t 1 ). m ihvd,t =γ w1 SHG ihvd,t 1 +γ w2 OSHG ihvd,t 1 + X ihvd β w1 + Y hvd β 2 + ζ v + ɛ ihvd,t i W (2) m jhvd,t = γ m2 OSHG ihvd,t 1 + X jhvd β m1 + Y hvd α 2 + ζ v + ɛ jhvd,t j M (3) Note that in equation (3) OSHG ihvd,t 1 captures the total number of women who were members of any SHG in the previous GP. If our results are driven by household selection, then we expect γ w2 > 0 and γ m2 > 0. The results are presented in columns 2 and 4 of Table 2, Panel A for the sample of women and Panel B for the sample of men. While it is statistically significant only in one case, the estimated values of γ w2 and γ m2 are always positive. Most importantly however, as the results in column 2, Panel B of Table 2 show, an increase in the total number of SHG members in the household is associated with an increase in male attendance in the village assemblies. The effect is also quite strong: an additional SHG member in the household is associated with a 5 percentage point increase in the proportion of village assemblies attended by males. This translates to a 20% increase in the attendance for male members of the household. Given these results we cannot ignore the issue of selection on unobservables. We return to this issue in Section 5, below. 4.2 Additional Issues Before proceeding further we examine several additional issues within the reduced form framework. The first relates to the importance of the group. SHGs and the associated group meetings bring women together and can thus have a significant effect on social cohesion and network formation of women. While SHGs are not the only form of voluntary organization of women, in the context of rural India they are the predominant ones. Approximately 7% of women report themselves to be members of SHGs (see Table 1). Does membership in other groups have a similar effect on attendance in village assemblies? We examine the effect of membership in a credit cooperative in the previous GP and membership of a religious group in the previous GP on attendance in village assemblies. Essentially we re-estimate equation (1) by replacing membership of SHG by 12

13 membership of credit cooperatives and membership of religious groups. These results are presented in Table 4. The results in columns 1 and 3 imply that membership in credit cooperatives in the previous GP is not associated with an increase in the likelihood of attending village assemblies. On the other hand, membership in a religious group in the previous GP is associated with an increase in the likelihood of attending village assemblies. If anything the results in columns 2 and 4 of Table 4 are stronger than the results presented in Panel A of Table 2. It is however important to bear in mind that only 1% of women in our sample report being members of a religious group in the previous GP. The second relates to regional heterogeneity. One reason for SHG membership to be correlated with attendance in the village assemblies is partly related to prevailing gender norms in India, which often imposes restrictions on mobility of women and the extent to which women can interact with the wider community. Here SHGs have an important role to play. For example Sanyal, Rao, and Majumdar (2015) argue that membership in SHGs change cultural norms towards women by giving women access to new networks and increasing knowledge. To examine whether membership in SHGs help women over-ride discriminatory gender norms we examine the association between membership of SHGs (in the previous GP) and attendance in village assemblies in different regions of India, characterized by variation in gender norms and how traditionally women have been viewed in the society. Southern India is characterized by significantly higher gender equality and is very different from the rest of the country in this respect. We estimate equation (2) separately for households residing in the Southern states of the country (Andhra Pradesh, Karnataka, Kerala and Tamil Nadu) and the rest of the country. Varshney (2000) argues that grass roots democracy is more entrenched in South India (compared to the north) and this is marked by greater political participation and higher achievements in terms of social development. On average women in the Southern states report attending 16.5% of the previous 4 village assemblies, compared to the 4.7% reported by women in the non-southern states. 9 Political engagement, as captured by attendance in the village assemblies is therefore significantly higher for both men and women in the Southern states of India. Given that gender norms are not so discriminatory in the Southern states of the country, we expect that the association between SHG membership and attendance in the village assemblies to be weaker in the these states 9 Men in the Southern states report attending 42% of the previous four village assemblies, compared by 22% reported by men in the non-southern states. 13

14 compared to the rest of the country. The results presented in Table 5 columns 1 and 2 show that our hypothesis is indeed validated. The association between SHG membership and attendance in the village assemblies for women is significant only in the non-southern states of the country: in the non-southern states, SHG membership is associated with a 4 percentage point increase in the fraction of village assemblies attended by women. The effect is much smaller in magnitude and not statistically significant in the Southern states. The results presented in columns 3 and 4 show that in the Southern states but not in the rest of the country, there is significant intra-household spillovers: an increase in the number of members in the household who are members of SHG is associated with an increase in male attendance in village assemblies. The third relates to the social network effect of SHGs. One way SHGs have this strong effect on attendance in village assemblies is through their effects on social network formation. As women join SHGs, they become a part of a larger network, which in turn provides these women the self confidence to attend village assemblies. However attending village assemblies is costly (both in terms of time commitment and in terms of the social tensions it might create) and as the size of the network increases those not in the network have greater incentive to free ride. This is likely to be reflected in a discontinuous change in the attendance rate of women after a particular threshold is attained. The question is how do we define the social network. In this paper we use a simple and rather naive measure of social network. The network is defined by the number of SHG members in the neighbourhood (or muhalla). The data at our disposal allow us to identify the village neighbourhoods where the sampled SHG members reside. This information then allows us to construct the social network (by aggregating the number of SHG members in a particular neighbourhood and also provides us with variation in the size of the network. The recent literature on the effect of social network on a variety of different outcomes argues that the size of the network must be sufficiently large, i.e., the size of the network must be greater than a specific threshold (see for example Munshi, 2011, Chay and Munshi, 2015, Munshi and Rosenzweig, 2016). Define n to be the threshold for the social network to be effective and let I( n) = 1 if the number of women in the neighbourhood who are members of SHG (in the previous GP) > n. We estimate 14

15 the following equation, which is an extended version of equation (2) m ihvd,t =γ w1 SHG ihvd,t 1 + γ w2 OSHG ihvd,t 1 +γ w3 I( n) + γ w4 SHG ihdv,t 1 I( n) + γ w5 OSHG ihvd,t 1 I( n) +X ihvd β w1 + Y hvd β 2 + ζ v + ɛ ihvd,t i W (4) We do not know what n is. So we consider three possible measures of n: 2, 6 and 10. The dummy variable I( n) then denotes that the network size is large. The regression results are presented in Table 6, columns 1, 2 and 3 for the three different thresholds. A comparison of the results in these three columns suggests that network effects start playing an effective role when the size of the network is sufficiently large here sufficiently large being defined by 10. The estimated γ w1 is positive and statistically significant implying that when social networks are small (i.e., the number of women in the neighbourhood who are members of SHG in the previous GP n), SHG members are significantly more likely to attend village assemblies compared to women who are not members of any SHG. The estimated γ w3 is negative (though statistically significant only when n = 10) indicating that the village assembly attendance rates of women who are not members of any SHG is significantly lower in neighbourhoods where the size of the social network is large (relative to neighbourhoods where the size of the network is small). The difference-in-difference estimate γ w4 is positive and statistically significant. Therefore relative to women who are not members of SHG, the attendance rates at village assemblies of SHG members is higher in neighbourhoods where the size of the network is large relative to neighbourhoods where the size of the network is small. Interestingly this is not driven by an increase in attendance in village assemblies by SHG members in neighbourhoods with a larger social network; rather it is driven by a reduction in the attendance rate of non-shg members in neighbourhoods with a larger social network. There is no free-riding by SHG members in neighbourhoods with a larger social network. The SHGs appear to be playing an important role of enforcing norms and this is independent of the size of the network. In column 4 of Table 6 we present the results from a variant of equation (4). We continue with the three thresholds at 2, 6 and 10, but define I(3 6) = 1 if the number of SHG members in the neighbourhood is in the range 3 6; I(7 10) = 1 if the number of SHG members in the neighbourhood is in the range 7 10 and finally I(10+) = 1 is defined as before (the number of 15

16 SHG members in the neighbourhood is greater than 10). The estimating equation is given by m ihvd,t =γ w1 SHG ihvd,t 1 + γ w2 OSHG ihvd,t 1 +γ w3 I(3 6) + γ w4 SHG ihdv,t 1 I(3 6) + γ w5 OSHG ihvd,t 1 I(3 6) +γ w6 I(7 10) + γ w7 SHG ihdv,t 1 I(7 10) + γ w8 OSHG ihvd,t 1 I(7 10) +γ w9 I(10+) + γ w10 SHG ihdv,t 1 I(10+) + γ w11 OSHG ihvd,t 1 I(10+) +X ihvd β w1 + Y hvd β 2 + ζ v + ɛ ihvd,t i W (5) The results presented in column 4 of Table 6 are consistent with those presented in the first three columns. The difference-in-difference estimate γ w10 is positive and statistically significant. Therefore relative to women who are not members of SHG, the attendance rates at village assemblies of SHG members is higher in neighbourhoods where the size of the network is large relative to neighbourhoods where the size of the network is small. Finally we also examine the robustness of our results using alternate data. The results presented in Table A1 in the Appendix show the same results as those in Panel A column 3 of Table 2 using the IHDS data. We find that membership of SHGs in 2005 is associated with a 6 percentage point increase in attendance in village assemblies in 2011 by women. This is similar to the effect of 8.5 percentage points that we see in Table 2. The results are therefore broadly comparable using two different nationally representative data sets from the country. 5 IV estimation Results: Effects of SHG membership on political engagement of women The results presented in Section 2.2 show that membership in SHG is associated with an increase in attendance in village assemblies. While it is unlikely that the results are driven by selection on observables, we cannot ignore the possibility that selection on unobservables might be driving some of the results. To provide a causal interpretation of the results, we use an instrumental variable approach. For the IV to work we need one variable that captures the effect of SHG membership for both men and women. Men are not members of SHGs. However as we saw in the regression results presented in Table 2, columns 2 and 4, the total number of SHG members in the household affects attendance 16

17 rates in village assemblies for both men and women (for women it is through the direct effect of SHG membership). regressions: Instead of estimating equations (2) and (3) we estimate the following m ihvd,t =δ w T SHG hvd,t 1 + X ihvd β w1 + Y hvd β 2 + ζ d + ɛ ihvd,t ; i W (6) m jhvd,t =δ m T SHG hvd,t 1 + X jhvd β m1 + Y hvd β 2 + ζ d + ɛ jhvd,t ; j M (7) Here T SHG hvd is the total number of SHG members in household h in village v in district d. The rest of the variables are as defined above. We would expect δ w > δ m. 10 The results are presented in Panel A of Table 7. Columns 1 and 2 present the results for women, those in columns 3 and 4 present the results for men. These results are consistent with those presented in columns 2 and 4 in Table 2. Not surprisingly, an increase in the total number of SHG members in the household has a stronger effect on women s attendance in village assemblies. 11 A valid instrument should affect attendance in village assemblies only through its effects on SHG membership. We use supply side information on the presence of SHGs in the village (under the previous GP regime) as our instrument. This variable is unlikely to have a direct effect on current attendance in village assemblies, except through its effect on SHG membership of women. The only caveat here is that the information on presence of SHG membership is collected through the household survey and not through a village or community survey. 12 This means however that survey responses do not always match within a village. We use two difference measures of existence of SHG in the village. The first is probabilistic existence, captured by the proportion of survey participants who report that a SHG is present and operational in the village. second is an indicator variable which takes the value of 1 if even one survey respondent reports the existence of an SHG in the village. The use of this supply side variable as the instrument is driven by the fact that typically SHGs are not invited into a village, therefore it is unlikely to be correlated with any household level unobserved factor that drives the selection bias that is described in Section Note that henceforth we move to a district fixed effects framework. Table A2 shows that our primary results remain qualitatively (and even quantitatively) similar irrespective of whether we include district, village or neighbourhood fixed effects. 11 It is unlikely that the results are driven by multiple SHG members in the same household. There are only 61 households with 2 SHG members and 4 households with 3 members. 12 Ideally we would need administrative data on the opening of new SHGs across villages. This data is however not available. 13 The community mobilizers that are often credited with facilitating the spread of SHGs and also the enrolment The 17

18 Table A3 in the Appendix presents the first stage regression results when T SHG is regressed on the full set of exogenous variables. Note that the presence of SHGs in the village in the previous GP is positive and statistically significant effect on the total number of SHG members in the household. In Table A4 in the appendix we examine whether village level observables affect the existence of SHG in the village. With the exception of median land holding and median school years in the village none of the additional village level observables have a statistically significant effect on the presence of SHGs in the village. 14 One could however argue that SHGs choose to enter villages that are more likely to be receptive to these SHGs. These villages could be more pro-social or villagers might be socially more engaged. That in turn might be reflected in more villagers choosing to attend village assemblies. To account for this possibility we regress entry of SHGs into the village during the period on the propensity of the villagers to attend village assemblies in period t 1. The REDS data however does not have information on attendance in village assemblies in the previous period. For this purpose we instead use the IHDS data: we regress the entry of SHGs into the village during the period (between the two survey rounds) on the proportion of households reporting in 2005 that at least one member of the household attended a village assembly in the one year prior to the survey. 15 The regression results presented in Table A5 in the Appendix show that the proportion of households reporting in 2005 that at least one member of the household attended a village assembly in the one year prior to the survey does not have a statistically significant impact on the entry of new SHGs into the village. In equations (6) and (7), T SHG hvd (the total number of SHG members in household h in village v in district t under the previous GP regime) is instrumented by P SHG vd, which is the existence of a SHG in village v in district d under the previous GP regime. The IV regression results are presented in Panels B and C of Table 7 for the continuous and dummy instrument respectively. These results show that unlike the reduced form regression results, there of women into SHGs are typically not residents of the village. Often they are outsiders (or agents) hired by the government to perform the task of mobilizing the community. 14 The first stage F-statistics presented in Table A3 are greater than 10 when we use the continuous measure (proportion of households that report that a SHG existed in the village in the previous GP). See columns 1 and 2. It is not so for the dummy instrument. See columns 3 and 4. Also in Table A4, the village level observables do not jointly explain existence of SHG in the village in column 1 (p-value = 0.19) but are in column 2 (p-value = 0.07). Therefore we will focus on the results that use the continuous instrument. The results using the dummy instrument are presented for comparison purposes of the 1419 villages in the sample report having no SHGs in 2005 but having one in

19 is either no impact or there is a negative impact of an increase in the total number of SHG members in the household on the likelihood of men attending village assemblies. The effect on attendance in village assemblies by women is large, positive and highly statistically significant. An increase in the total number of SHG members in the household results in a close to 100% increase in the fraction of village assemblies attended by women. 6 Effect on Provision of Public Goods Additionally we are interested in the effect of increased women s participation in village assemblies on the mix of benefits and public goods that are provided by the village council. We focus on two different kinds of public goods: household level benefits and community level or local public goods. Does increased participation in village assemblies result in the household receiving more household level benefits (through the allocation of government programs)? Once again we are interested in the IV estimates. We use the presence of a SHG in the village in the previous GP (P SHG vd,t 1 ) as the instrument. The corresponding first stage results are presented in Table A6. A priori we cannot hypothesize how increased attendance of women in village assemblies will affect the provision of household level benefits. There is no evidence to suggest that women prefer certain kinds of household level benefits while men prefer some others. Indeed the results presented in Table 8 show that increased attendance in village assemblies has no effect on household level benefits received. Next we turn to the provision of women preferred community level public goods. To do this we examine the IV regression results of changed composition of attendees at village assemblies on the composition of public goods provided. We have already shown that our instrument (P SHG v ) has a direct effect on attendance in village assemblies through increased participation in SHGs. To examine the effects on the composition of village level public goods, we proceed in steps. We first show that the presence of SHGs in the village changes the gender composition of attendees at village assemblies. We then examine whether this change in gender composition of attendees at village assemblies in turn affected the provision of village level public goods. Prior evidence 19