Collective Action and Community Development: Evidence from Self-Help Groups in Rural India Raj M. Desai and Shareen Joshi In response to the problems of high coordination costs among the poor, efforts are under- way in many countries to organize the poor through “self-help groups” (SHGs)—member- ship-based organizations that aim to promote social cohesion through a mixture of education, access to finance, and linkages to wider development programs. We randomly selected villages in one of the poorest districts in rural India in which to establish SHGs for women. Two years of exposure to these programs increased women’s participation in group savings programs as well as the non-agricultural labor force. Compared to women in control villages, treated women were also more likely to participate in household decisions and engage in civic activities. We find no evidence however, that participation increased income or had a disproportionate impact on women’s socio-economic status. These results are important in light of the recent effort to expand official support to SHGs under India’s National Rural Livelihood Mission. JEL codes: D70, I3, I38, J16, Q1 Collective action by the poor has been shown to strengthen property rights, increase bargaining power in labor markets, improve access to financial markets Raj M. Desai is a professor at the Edmund A. Walsh School of Foreign Service and Department of Government, Georgetown University, Washington, D.C., and a Nonresident Senior Fellow at the Brookings Institution, Washington, D.C. His email address is desair@georgetown.edu. Shareen Joshi (corresponding author) is a professor at the Edmund A. Walsh School of Foreign Service, Georgetown University, Washington, D.C. Her email address is sj244@georgetown.edu. The authors gratefully acknowledge the cooperation of the Self-Employed Women’s Association (SEWA). Research for this article was financed by the Wolfensohn Center for Development and the Development and Governance Initiative at the Brookings Institution, and Georgetown University’s Engaging India Initiative. Surveys were conducted by Social and Rural Research Institute – IMRB, New Delhi. Previous versions of this paper were delivered at the annual meeting of the American Political Science Association and at seminars at Georgetown University, the World Bank, the Indian Statistical Institute, the National Council for Applied Economic Research, and Jindal University. The authors are grateful to Lael Brainard, Kristin Bright, Marc Busch, Jishnu Das, Sonalde Desai, Quy Toan Do, Antrara Dutta, Garance Genicot, Alain de Janvry, Homi Kharas, Johannes Linn, Ghazala Mansuri, Hari Nagarajan, Reema Nanavaty, Anders Olofsga ˚ rd, Vijayendra Rao, Elisabeth Sadoulet, Arunav Sen, Amita Shah, J. P. Singh, E. Somanathan, Rohini Somanathan, James Raymond Vreeland, Dominique Van de Walle, and two anonymous reviewers for comments on earlier drafts, as well as to Carolina Fernandez and Joshua Hermias for invaluable research assistance. All errors and omissions are the authors’ own. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL. 28, NO. 3, pp. 492– 524 doi:10.1093/wber/lht024 Advance Access Publication September 7, 2013 # The Author 2013. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com 492 Desai and Joshi 493 and increase public investments in poor communities (Bardhan 2005; Narayan et al. 2000; Ostrom and Ahn 2009). In light of this evidence, governments, donors, and non-governmental organizations (NGOs) have sought to expand their support to membership groups, cooperatives, producer associations, and other types of organized platforms for collective action in poor communities. Since 1999, the World Bank has disbursed over $50 billion in loans for community-based and community-driven development projects (Mansuri and Rao 2012).1 We examine whether community organizations can lower the barriers to col- lective action in one of the poorest districts in India. We focus on a group for whom coordination constraints are likely to be particularly binding: rural, tribal women. These women face some of the lowest levels of literacy, labor-force par- ticipation, and personal autonomy in the world (King and Mason 2001; Sen 2001; Sen and Dreze 2002).2 Divisions along the lines of religion, class, caste, and tribe have, as with other groups in the Indian polity, hindered the formation of a unified women’s movement (Agnihotri and Mazumdar 1995). Moreover, or- ganizational resources for rural groups remain quite limited (Chhibber 2001). Tribal groups also remain among the most politically excluded in modern India. As with Scheduled Castes (SC), the Scheduled Tribes (ST) have faced historic dis- advantages. But unlike SC groups—which now claim national political parties as well as several high-profile leaders who represent their interests in the wider po- litical system—ST groups have been slower to mobilize or gain political represen- tation (Ambagudia 2011; Banerjee and Somanathan 2007). Given these barriers to collective action, we focus on an increasingly common effort to overcome them: the creation of “self-help groups” (SHGs). In India, SHGs are village-based organizations that focus on building the savings and credit, as well as the social empowerment, of their (mostly female) members (Chen et al. 2007). These groups perform three principal functions: (i) they act as an in- termediary in transactions with the formal financial sector; (ii), they provide a mechanism for alternative (i.e., non-public) service delivery—such as contracting directly for training in agriculture or other vocational skills, healthcare, childcare, and educational services; and (iii) they serve as a platform for broader engagement by members in local civic affairs. The intervention we investigate was facilitated by the Self-Employed Women’s Association (SEWA) in Dungarpur, Rajasthan, where 80 villages were randomly assigned into control and treatment categories. We examine effects using both village- and individual-level treatment vari- ables, that is, both the effect of residing in a SEWA (treatment) village and the effect of being an SHG member in a SEWA village, in order to identify communi- ty outcomes as well as intra-village spillovers from members to non-members. 1. If loan disbursements for projects with “decentralization” components are included, the total is closer to $80 billion. 2. Adult female literacy currently stands at 51 percent for women and 76 percent for men (World Development Indicators, 2012). These numbers are generally lower, and the gender-gap larger, in rural areas. 494 THE WORLD BANK ECONOMIC REVIEW We also attempt to identify heterogeneous impacts with respect to baseline levels of education and landholding. Finally, we attempt to investigate some plausible channels by which SEWAs program benefits are transmitted. Over a period of two years, women in treatment villages were more likely to participate in group programs, acquire greater “personal autonomy” (including greater control over household decision-making), partake in collective action on issues such as water and sanitation, and engage in community affairs, than their counterparts in control villages. We find no evidence that the program’s effects are concentrated among women who were better off at the baseline. Rather, landless women are more likely to save regularly, and increase their cash income as a result of SEWA’s programs, compared to landholders. Although the precise causal mechanisms behind these effects is difficult to measure due to the bundled nature of SEWA’s rural livelihoods programs, we exploit variation in the timing and implementation of specialized modules imple- mented within the broader intervention to identify some plausible channels through which SEWA may have affected certain specific outcomes. We find that exposure to vocational training services as well as financial training services in- creased access to labor and credit markets respectively. Although the evaluation is over a relatively short time horizon, in an excep- tionally poor area, these results nevertheless carry important implications for India’s large-scale antipoverty efforts. Investment support to rural membership- based organizations is currently being expanded through the National Rural Livelihood Mission (NRLM), which envisions mobilizing all rural, poor house- holds into membership-based groups by 2015 (Planning Commission 2011). In Rajasthan as well as other states, the project is being implemented in collabora- tion with a variety of NGOs whose strategy of mobilizing the rural poor resem- bles SEWA’s. This bottom-up approach is being pursued under the assumption that it can be effective in overriding other divisions such as religion, caste, tribe, ethnicity and language in rural India to organize women around the goal of poverty alleviation. The remainder of this paper is organized as follows. The next section reviews related evidence of the effects of SHGs and describes the research setting and in- tervention. The second and third sections examine effects of village-level treat- ment and individual participation in the program, respectively. We then turn to impact heterogeneity and causal mechanisms. The final section concludes. I. THE EMPIRICAL SETTING We describe, below, the “self-help” movement in India, as well as an initiative of the Self-Employed Women’s Association, which was randomly assigned across villages. We then perform some basic tests of randomization before describing the principal outcomes of interest. Desai and Joshi 495 Self-Help Groups in Rural India A typical Indian SHG consists of 10-20 poor women from similar socio- economic backgrounds who meet once a month to pool savings and discuss issues of mutual importance. The pooled fund is then deposited in a group bank account and used to provide credit to women in need. These activities are typical- ly facilitated by NGOs, the government, and in some cases, even the private sector. Facilitators oversee the operations of the group and link women to rural credit institutions, state agricultural produce market committees, and district agencies. SHGs can also contract for additional services such as childcare, extra- curricular programs for school children, and job-training programs. SHGs, finally, perform important social functions: they may serve as a platform to address community issues such as the abuse of women, alcohol, the dowry system, educational quality, and inadequate infrastructure. In 1992, India’s National Bank for Agricultural and Rural Development (NABARD) piloted its “bank-SHG linkage program” by facilitating group lending by rural banks and by providing participating rural banks with low- interest financing and refinancing support. Since then, the SHG linkage program has expanded into one of the world’s largest micro-finance networks. Women’s SHGs, additionally, have been heavily promoted by the Indian government, par- ticularly in the southern states since at least since the 1980s (Basu 2006; Chakrabarti and Ravi 2011; Deshmukh-Ranadive 2004; Galab and Rao 2003; Reddy and Manak 2005). Several large development programs, such as the Integrated Rural Development Program (Swarnjayanti Gram Swarojgar Yojana) and most recently, the NRLM, have targeted these groups. Available evidence on SHGs is mixed, showing a number of positive effects on credit and default risk on the one hand, but less improvement in income or assets. A study of SHGs in Andhra Pradesh finds improvements in consumption and savings mainly for participants of newly-formed groups (Deininger and Liu 2009). In Orissa SHG-members are better able to coordinate in managing common pool resources (Casini and Vandewalle 2011). Diversity within groups with respect to education, landholdings, and family networks affect group stabil- ity and more vulnerable women are most likely to exit from the groups (Baland, Somanathan, and Vandewalle 2008). One of the few randomized-controlled trials finds that regular SHG participants trust and interact with each other more, are more willing to pool risk, and are less likely to default on those loans (Feigenberg, Field, and Pande 2013). Despite the large scale-up in the number of SHGs in India in recent years, the impact on women and communities remains poorly understood. Efforts to measure impact are typically constrained (with a few notable exceptions) by the non-random placement of programs, the non-random assignment of indi- viduals to groups, and wide variations in the methods employed by the organi- zations that facilitate the creation of village SHGs. Using random assignment, we examine impacts of SHGs not only on salient economic and financial 496 THE WORLD BANK ECONOMIC REVIEW outcomes, but also on member’s empowerment and civic participation. These results are particularly noteworthy in the setting in which it is conducted: one of the poorest districts in India where the barriers to collective action are severe. SEWA and the Integrated Rural Livelihoods Program Research was conducted in Dungarpur district of Rajasthan, India, a rural dis- trict of 1.1 million located on the tribal belt between Gujarat and Rajasthan. According to the 2011 Census of India, 65 percent of the population belongs to Scheduled Tribes (STs). In 2005 Dungarpur was selected for the national Backward Districts Initiative (Rashtriya Sam Vikas Yojana). Of 601 districts countrywide, Dungarpur ranks: 524th in adult female literacy, 505th in terms of percentage of the population owning land, 480th in household asset-holdings, and 450th in terms of poverty, i.e., in the bottom quintile on all indices.3 Through the Backward Districts Initiative, block grants were provided by the Indian Planning Commission to state governments that were to use the funds for economic development in the 100 poorest districts. State governments were to prepare district plans for the use of funds—some Rs. 150 million ($3.3 million) per district per year for three years. For Dungarpur (as with the other backward districts in Rajasthan), the district plan emphasized “sector livelihood develop- ment,” or a multi-component program focused on rural unemployment, creation of SHGs, skills training, credit linkages, and the provision of other rural services. To implement these programs, the state government invited the Self-Employed Women’s Association (SEWA), an NGO based in neighboring Gujarat, to imple- ment its program. SEWA began its activities in the district in 2007. Founded as an offshoot of the Textile Labor Association in 1972, it now claims a membership of over 1 million women across 10 Indian states. The organization’s main mission is to help women achieve economic independence through bundled interventions that address simultaneous challenges: skill shortages, limited access to credit and in- surance, weak market linkages and limited public services. It typically provides members with a variety of services that include employment training programs, new sources of credit, subsidized access to new technologies, and access to free child-care services (Bhatt 2006; Chen 1991; Datta 2000). For the Dungarpur district pilot, all registered villages in Dungarpur from the Census of India (2001) were stratified according to: (i) average female literacy; (ii) total village population; and (iii) average household size. From these strata 80 villages were randomly selected, and randomly assigned to the SEWA program (32 villages) or as controls (48 villages). The rollout was implemented in stages. First, all women in a village were invited to become members of SEWA by paying a nominal fee of Rs. 5 3. Authors’ calculations using India’s District Level Household Survey, Round 3 (International Institute for Population Sciences 2008). Desai and Joshi 497 (approximately $0.10).4 Members participated in a full day of basic training programs that were intended to create a sense of unity and direction, and an understanding of SEWA’s objectives. They were then organized into SHGs with an elected leader. All these activities were led by SEWA field organizers: typically local, married women with at least 12 years of education who were highly regarded by the local community. These field workers reported to a SEWA coordinator, who worked from the SEWA office in Dungarpur city. Once SHGs were formed and leaders were elected, participants would meet once a month and set savings targets of Rs. 25-100 ($5-20) per member per session. These were deposited into a savings account at an SHG-linked bank. The group would then lend these funds—for a period and at an interest rate set by the SHG—to members in need of extra cash. Meetings were also used to discuss other issues—details of job training programs, motivational messages, the importance of participating in local government, etc. SHG leaders were trained to run meetings, maintain minutes, manage group accounts, and monitor the group’s activities. All meetings were attended by SEWA field workers, who provided women with information about government schemes/programs and their eligibility for those programs. They also helped with other activities such as recording minutes of the meetings, assisting in necessary activities such as filling out all necessary paperwork at the local bank and/or arbitrating in the event of any dispute between the women. In addition to these activities, SEWA also con- ducted educational programs, job-training programs and employment/income- generation workshops.5 All SEWA programs were always open to all female village residents regardless of SHG membership. None of these services, however, were available to women in control villages. Population density in the Dungarpur area is one of the lowest in India, distances between villages are significant, opportunities for inter-village transport are quite limited and, women’s mobility is severely constrained. Additionally, village residency was a requirement for SHG membership or participation in SEWA pro- grams. We are not aware of the presence of any other NGO in our control villag- es, but it is important to note that the Indian government began a major poverty-alleviation effort in all the villages in our sample during the period of study: the National Rural Employment Guarantee Act (NREGA), a large public works program started in 2005. NREGA came to this area shortly after the inter- vention began, and was popular in both treatment and control areas. There is no 4. Recruitment of members is carried out by making announcements about SEWA at village Panchayat meetings, and/or private meetings with educated and influential members of the village who then spread awareness about SEWA’s programs. 5. As such, SEWA SHGs went beyond the traditional activities of micro-finance groups in three ways. First, SEWA’s groups attempted to promote the personal empowerment of individual members, and better cooperation as a group. Second, SEWA did not establish its own micro-finance programs in villages, but rather, used its SHG-based revolving fund to help households establish credit histories. Third, SEWA SHGs also provided additional skills training to members. 498 THE WORLD BANK ECONOMIC REVIEW indication however, that the programs were selectively targeted in either the treatment or control villages in our study. Baseline and follow-up surveys were conducted in late (November and December) 2007 and during the same months in 2009. These form a pooled cross-section with treatment and control samples. The pooled sample includes a total of 1,410 women who resided in the villages where SEWA programs were in place. 748 of these women were interviewed in the 2007 baseline and 662 inter- viewed in the 2009 follow-up. The sample of control women includes 1,795 women who did not reside in SEWA villages over the two year period, with 855 interviewed in 2007, 940 in 2009. Summary statistics of all variables used in the analysis in this paper, across both treatment and control areas, and both before and after the intervention, are presented in table 1. Tests of Randomization Comparisons of pre-program characteristics are presented in columns 1-3 of table 2. These estimates are constructed from individual-level data. Village-level differences are presented in the online Supplementary Appendix (available at http:// wber.oxfordjournals.org/), table S.1. Estimates in column 3 of table 2 contain the difference in mean outcomes between treatment and control populations prior to the treatment. Estimates are obtained from weighted regressions with robust stan- dard errors clustered at the village-level. Note that the two sets of villages—both before and after the treatment—show no statistically significant differences with respect to demographic and socio-economic variables (panel A) such as women’s literacy level, marital status, labor-force participation caste, and socio-economic characteristics. There is also no evidence that the treatment villages had more SHGs prior to the arrival of SEWA. There are however, some other pre-intervention differences. Women in SEWA villages were less likely to be in the habit of saving prior to the program, were more likely to participate in the agricultural workforce and had higher cash income than their counterparts in control villages. They also had lower levels of participation in family-planning decisions. These estimates, however, disappear when looking at village-level averages, indicating that these initial differences may be driven by a small number of distinctive villages (table S.1). We also present results that control for these, and other, possible factors. We note, additionally, that these unconditional estimates show that residents in SEWA villages experienced a small drop in cash income relative to residents in control villages following SEWA’s intervention. In fact, both SEWA and non-SEWA villages experience drops in income during this period—mainly due to drought conditions and weak monsoons in Rajasthan between 2008 and 2009. As seen below, this effect disappears when fixed effects and additional controls are added. Desai and Joshi 499 T A B L E 1 . Summary Statistics Mean Std. Dev. Min. Max. SEWA village resident 0.450 0.498 0 1 SEWA member 0.115 0.318 0 1 SEWA training-module village resident 0.152 0.358 0 1 SEWA training-module participant 0.013 0.116 0 1 SEWA finance-module village resident 0.374 0.484 0 1 SEWA finance-module participant 0.028 0.165 0 1 Participates in group savings 0.220 0.414 0 1 In the habit of saving 0.198 0.398 0 1 Credit (past 5 years) 0.095 0.294 0 1 Cash savings (log, 3 months) 0.856 2.228 0 10.31 Cash income (log, 3 months) 0.681 2.236 0 11.24 Employed (past 3 months) 0.783 0.412 0 1 Employed (non-farm past 3 months) 0.063 0.242 0 1 Final say: children’s schooling 0.088 0.283 0 1 Final say: medical decisions 0.097 0.296 0 1 Final say: family-planning 0.031 0.173 0 1 Grievance: water 0.245 0.430 0 1 Grievance: roads 0.199 0.400 0 1 Grievance: electricity 0.238 0.426 0 1 Grievance: education/health 0.180 0.385 0 1 Addressed grievance: water 0.212 0.409 0 1 Addressed grievance: roads 0.173 0.378 0 1 Addressed grievance: electricity 0.181 0.385 0 1 Addressed grievance: education/health 0.110 0.314 0 1 Know of Gram Sabha and Gram Panchayat 0.230 0.421 0 1 Engage with Gram Sabha and Gram Panchayat 0.014 0.118 0 1 Know anyone in the village who paid a bribe 0.038 0.192 0 1 Age 37.13 10.002 14 80 Literate 0.183 0.387 0 1 Married 0.945 0.228 0 1 Scheduled tribe 0.726 0.446 0 1 Husband age 40.798 9.990 18 77 Husband literate 0.082 0.275 0 1 Own house 0.849 0.358 0 1 Have own farm 0.877 0.329 0 1 Kutcha house 0.685 0.464 0 1 Household has toilet 0.072 0.259 0 1 Note: N ¼ 3,205. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. Outcomes of Interest The SEWA intervention focused not only on micro-savings and finance, but also on women’s roles in household decision-making and civic engagement. Our eval- uation has been structured to examine the effect of this integrated program, rather than its specific components. We focus, therefore, on three central objectives of SHG participation: (i) inclusion in financial and labor markets; (ii) autonomy of household decision-making; and (iii) local civic and political engagement. 500 THE WORLD BANK ECONOMIC REVIEW T A B L E 2 . Pre- and Post-program Differences Pre-intervention (1) – (3) Post-intervention (4) – (7) (1) (2) (3) (4) (5) (6) (7) SEWA Control SEWA SEWA village village village village Control resident  residents residents Difference residents village residents Difference Post-Intervention (A) Independent variables Age 37.39 36.35 1.044 (0.645) 36.69 37.97 2 1.077* (0.618) Literate 0.184 0.188 2 0.004 (0.037) 0.213 0.186 0.057 (0.039) Married 0.947 0.952 2 0.006 (0.012) 0.923 0.952 2 0.024 (0.015) Scheduled tribe 0.668 0.730 2 0.061 (0.100) 0.725 0.77 2 0.057 (0.078) Husband age 41.06 40.24 0.824 (0.710) 40.50 40.87 2 0.463 (0.613) Husband literate 0.086 0.083 0.003 (0.020) 0.095 0.070 0.023 (0.021) Own house 0.861 0.835 0.026 (0.027) 0.805 0.884 2 0.071** (0.030) Have own farm 0.900 0.891 0.009 (0.040) 0.835 0.874 2 0.027 (0.051) Kutcha house 0.667 0.746 2 0.079 (0.071) 0.642 0.676 2 0.039 (0.060) Household has toilet 0.098 0.081 0.017 (0.036) 0.073 0.045 0.025 (0.028) (B) Outcome variables Participates in group savings 0.132 0.146 2 0.014 (0.030) 0.427 0.199 0.223*** (0.053) 0.237*** (0.028) In the habit of saving 0.155 0.194 2 0.039* (0.023) 0.256 0.188 0.065** (0.030) 0.104*** (0.028) Credit (past 5 years) 0.090 0.076 2 0.013 (0.014) 0.162 0.116 0.015 (0.016) 0.029 (0.020) Cash savings (log, 3 months) 0.069 0.617 2 0.007 (0.935) 1.194 1.037 0.156 (0.221) 0.163 (0.352) Cash income (log, 3 months) 1.379 0.895 0.483* (0.239) 0.340 0.166 0.173*** (0.01) 2 0.310** (0.154) Employed (past 3 months) 0.798 0.768 0.030 (0.039) 0.784 0.783 2 0.000 (0.051) 2 0.030 (0.029) Employed (non-farm, past 3 months) 0.048 0.063 2 0.015 (0.017) 0.091 0.053 0.038 (0.028) 0.053** (0.017) Final say: children’s schooling 0.092 0.087 0.006 (0.023) 0.130 0.055 0.067*** (0.019) 0.061** (0.020) Final say: medical decisions 0.098 0.110 2 0.012 (0.021) 0.131 0.061 0.063*** (0.021) 0.075*** (0.021) Final say: family-planning 0.018 0.055 2 0.036** (0.014) 0.044 0.010 0.032** (0.012) 0.068*** (0.012) Grievance: water 0.183 0.164 0.019 (0.027) 0.421 0.245 0.156*** (0.052) 0.137*** (0.030) Grievance: roads 0.146 0.130 0.016 (0.026) 0.301 0.234 0.052 (0.051) 0.036 (0.028) Grievance: electricity 0.136 0.109 0.028 (0.027) 0.435 0.298 0.112* (0.060) 0.084** (0.029) Grievance: education/health 0.163 0.159 0.004 (0.031) 0.258 0.159 0.082* (0.044) 0.078** (0.027) Addressed grievance: water 0.160 0.139 0.021 (0.025) 0.349 0.220 0.128** (0.050) 0.107*** (0.029) Addressed grievance: roads 0.128 0.103 0.025 (0.025) 0.252 0.215 0.037 (0.046) 0.012 (0.027) Addressed grievance: electricity 0.100 0.083 0.017 (0.021) 0.316 0.236 0.080 (0.048) 0.063* (0.027) Addressed grievance: education/health 0.086 0.091 2 0.006 (0.024) 0.166 0.107 0.059* (0.033) 0.065** (0.022) Know of Gram Sabha and Gram Panchayat 0.215 0.218 2 0.002 (0.035) 0.268 0.226 0.042 (0.049) 0.045 (0.030) Engage with Gram Sabha and Gram Panchayat 0.011 0.019 2 0.008 (0.007) 0.018 0.009 0.010 (0.008) 0.018* (0.008) Know anyone in the village who paid a bribe 0.040 0.055 2 0.015 (0.015) 0.049 0.013 0.036** (0.015) 0.051*** (0.014) N 748 855 662 940 Notes: Columns 1, 2, 4, and 5 contain means for given sub-samples; columns 3 and 6 are differences with standard errors (clustered at the village level) in parentheses. *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,205. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. Desai and Joshi 501 502 THE WORLD BANK ECONOMIC REVIEW Savings, Credit, and Labor Force Participation. In the long-run, participation in SHGs could be expected to increase income, assets and labor force participa- tion rates. In the short-run however, we expect the presence of SEWA programs to increase women’s participation in group programs that are aimed at increasing saving, access to credit, and employment opportunities. We measure participa- tion as a dummy variable that takes value 1 if a woman reports any participation and 0 otherwise. Women in treated villages are also expected to have greater access to credit from the SHG-linked bank, and we code this outcome 1 if the woman borrowed through a formal financial institution, 0 otherwise. With respect to savings, we code this outcome 1 if the women reports that she saves money each month, 0 otherwise. We also look at the log of the last savings amount (in the three months prior to the survey).6 Since SHGs seek to increase female participation in the labor force, we also define two binary labor-participation indicators—relating to the general work- force, and the non-farm sector. Both are coded 1 if a woman is employed general- ly (employed as a casual laborer in agriculture), and 0 otherwise. We also include the log of women’s cash income, earned over the past three months, setting this value to 1 for those who earn no income. Household Decision-Making. If SHG membership raises a household’s current and future income by increasing labor-force participation and returns on savings, we expect the presence of SEWA programs to increase women’s decision-making autonomy within their households. Higher wages also increase the opportunity costs of a woman’s time, lowering the demand for children and raising the likeli- hood of contraceptive use. We thus examine respondent’s involvement in three types of decisions: children’s schooling, medical decisions, and family-planning. We define dummy variables that take value 1 if a woman reports that she is able to make independent decisions in these matters and 0 otherwise. Civic Inclusion and Engagement. We also test the hypothesis that participa- tion in SHGs expands women’s knowledge of authority structures in the village and motivates them to redress grievances about public issues. We measure this in three ways. First, we examine women’s knowledge of where to report grievances regarding five types of public services: water and sanitation, road conditions, electricity supply, education services, and health services. These variables take value 1 if the woman knows where to report a grievance in the village and 0 oth- erwise. We also measure whether she has actually approached authorities to report a complaint and demand improvements in delivery, again using a variable coded 1 if the woman reports that she has reported a grievance at least once in the preceding two years, 0 otherwise. Second, we examine whether women are aware of bribes being collected from villagers by public officials, coded 1 if they personally know someone who has been asked to pay bribes, 0 otherwise. Finally, we also measure women’s 6. Since some values are likely to be 0, we add 1 to all reported savings amounts. Desai and Joshi 503 participation in the main local governmental institutions, the Gram Sabha and Gram Panchayat.7 These are measured by two dummies. The first takes value 1 if the respondent knows of the Gram Sabha and the Gram Panchayat and 0 oth- erwise. The second takes value 1 if the woman has ever engaged with both insti- tutions (by attending meetings and/or interacting with Gram Panchayat members outside of meetings) and 0 otherwise. II. VILLAGE TREATMENT EFFECTS We first measure the impact of SEWA programs on all women who reside in vil- lages where SEWA programs were implemented. We favor this village-based measure of treatment rather than a direct measure of actual participation for three reasons. First, SEWA’s intervention was randomized at the village level and we avoid the problem of estimating the program’s impact exclusively on self- selected participants by focusing instead on individual effects based on village residence. Second, low female mobility causes women’s networks in rural North India to be highly localized and concentrated in their villages of residence (Dyson and Moore 1983; Jeffrey and Jeffrey 1996). Consequently it makes little sense to operationalize treatment at the individual or household levels, since new information introduced into a single village can diffuse along social networks quite quickly, leading to the rapid spread of information and social learning (Munshi 2007). Third, SEWA’s integrated approach is designed to promote intra-village spillovers and change prevailing attitudes of both men and women of communities. Program effects can be estimated as follows: Yi;v;b;t ¼ b0 þ b1 SEWAv þ b2 Post-interventiont þ b3 ðSEWAv  Post-interventiont Þ þ b4 Xi;v þ mb þ 1i;v;b;t where Yi,v,b,t is the outcome of interest for individual i in village v in block (sub-district) b during survey year t. SEWA takes value 1 if the respondent resided in a village selected for SEWA’s program, Post-intervention is a dummy variable that takes value 1 if the household was interviewed after the treatment program, X is a vector of household and village-level control variables, m is a block fixed-effect8, and ei,v,b,t is a standard disturbance. b1 is the pre-program difference, b2 is the trend, i.e., the changes in the outcome in the absence of the treatment, and b3 is the intent-to-treat (ITT) effect. Control variables include 7. The Gram Panchayat is the local governing body of a village or small town in India. The Gram Sabha is composed of all men and women in the village who are above 18 years of age. Meetings of the Gram Sabha are usually convened twice a year to discuss community issues. 8. Blocks, or tehsils are district subdivisions comprising multiple villages. In our sample, villages belong to one of three blocks. We do not include village fixed-effects because we are measuring impact at the village level. 504 THE WORLD BANK ECONOMIC REVIEW the respondent’s age, literacy, marital status, household size, husband’s age and literacy, scheduled-tribe status, and dummies for home/land ownership, kutcha (non-permanent) dwellings, and the presence of a toilet (both being proxies for income and assets that are likely to be unaffected by a two-year intervention). We also include an indicator coded 1 if public-works programs from NREGA were operating within the village during the survey year, on the assumption that presence of public works programs may affect village-level outcomes and may proxy the effectiveness of village-level institutions. Finally, given the subjective nature of many of our dependent variables we include responses by women to questions about the quality of roads to their village on the assumption that this should be invariant across village households. The distribution of responses to this questions in equations including village-fixed effects, should therefore closely proxy individual bias. We use a dummy variable that takes value 1 if she reports that the village roads are either “bad” or “very bad” and 0 otherwise, to correct for individual-specific “systemic” bias. All standard errors are clustered at the village-level. Unconditional Impact We first examine unconditional ITT effects by using a specification with no control variables. The simplest estimates of impact—differences in mean values for the key groups—are presented in columns 4-6 of panel (B) of table 2. Estimates in columns 3 and 6 contain the difference in mean outcomes between treatment and control populations before and after the treatment respectively. Estimates in column 7 present the difference in the differences. Note that two years after the program, individuals in SEWA villages are 24 percentage points more likely to participate in group programs and 10 percentage points more likely to save regularly. They are also more likely to take bank loans and save more per month (as measured by log savings values), but these estimates are not statistically significant at the 10 percent level. There are also differences in employment outcomes: women in villages with SEWA programs report declines in overall employment but increased non- agricultural employment. The declines in overall employment in our study-area are largely driven by the 2008–2009 drought, which reduced the cropped area in this region.9 We find that SEWA members were less hard-hit since they were 5 percentage points more likely to find non-agricultural employment. This effect is also noteworthy in light of the fact that only 6.3 percent of women participate in the non-agricultural labor force (table 1). Employment opportunities are likely to be influenced by the presence of the National Rural Employment Guarantee 9. 2008 and 2009 were years of below-average rainfall in western India and southern Rajasthan was particularly hard-hit. The government established the NREGA program to help address the declines in agricultural income in this area. Desai and Joshi 505 Act (NREGA). While the program appears to have benefitted both areas, we nev- ertheless control for the presence of this program. The results in table 2 also illustrate that SEWA programs strengthened women’s participation in household decision-making. Treated women are 6-8 percentage points more likely to have a say in decisions about children’s schooling, family medical care, and family planning. The impact on family-planning decisions is particularly striking considering that only 3 percent of women report any partici- pation in this decision (table 1) and women in SEWA villages had lower levels of participation in this decision at baseline (column 3, table 2). Women in treatment villages were more likely to know where to report griev- ances related to the failures of public services: these estimates range from 14 per- centage points for water, 3 percentage points for roads and 8 percentage points for electricity, education and health institutions. For the case of water, estimates are significant at the 1 percent level. Treated women were not only more knowl- edgeable about where to report their grievances, but also more likely to take action and actually report a grievance to the concerned authorities.10 These esti- mates are 11 percent points for the case of drinking water, more than 6 percent points for electricity, education and health services, and 1 percent for roads. The results are statistically significant for the case of drinking water, electricity, edu- cation and health services. The result on drinking water is particularly striking; across our entire sample in both periods, only 25 percent of women in the entire sample were aware of where to report grievances about drinking water and only 21 percent of women had ever made the effort to report a grievance to the authorities (table 1). SEWA village residents in 2009 were thus about 50 percent more likely to be aware of where to report some grievances such as drinking water and also take action in the case of poor service delivery. This is a critical difference, given that women in rural Rajasthan are responsible for collecting drinking water and spend consider- able amounts of time on this activity. Women in treatment villages were, finally, modestly more disposed towards local political awareness and participation: they were 5 percentage points more likely to be aware of bribe-payments to local officials. They were also 2 percent- age points more likely to interact with the Gram Sabha and Gram Panchayat. While these estimates of civic-engagement are small, they are nonetheless impor- tant considering the short time-frame of this evaluation. 10. The estimates of reporting a grievance are lower than estimates of “knowing where” to report a grievance. Note that the first may be unrelated to the second. Women can participate in collective action regarding grievances without exact knowledge of appropriate channels for addressing those grievances, because information can be managed by other members in the group, or actions may be taken through non-official channels (e.g., contacting hand-pump contractors directly, complaining to village councilors about public services, etc., rather than registering complaints with the agency responsible for such matters, namely, the sub-district Public Health and Engineering Departments. 506 THE WORLD BANK ECONOMIC REVIEW Conditional Effects Conditional estimates of our specification are presented in table 3. The first four columns contain estimates from a specification that includes block-level fixed effects but excludes all other controls. Columns 5-8 present estimates from the full specification, with controls, but we omit the listing of control variables and present only the coefficients of interest.11 The results are very close to the uncondi- tional estimates discussed above. Women in villages with SEWA programs were 24 percentage points more likely to participate in group-savings programs, 11 per- centage points more likely to be in the habit of saving money, 5-7 percentage points more likely to have a final say in household decisions, 13 percentage points more likely to know where to report a grievance for drinking water and 10 per- centage points more likely to actually report this grievance. The program has no effect on women’s reporting of other types of grievances (roads, electricity or health/education institutions). These findings on water largely confirm other studies of rural India that have documented the salience of this issue for women (Chattopadhyay and Duflo 2004; Joshi 2011). Women who resided in SEWA villages were 5 percentage points more likely to know if anyone in the village had paid a bribe to either gain access to water for farming or to public officials. Two years of exposure to the program also resulted in a slightly higher (2 percent) village-wide likelihood of interaction with the Gram Sabha and Gram Panchayat. An interesting difference between the unconditional estimates and conditional estimates are the coefficients for employment. Conditional estimates suggest that women in SEWA villages were also more 5 percentage points likely to be involved in non-agricultural employment. The effect is significant at the 10 percent level. This is important considering that the monsoon crop in this season had largely failed due to a drought in the district and agricultural incomes had declined, as is seen by the negative and significant coefficient for “Post Intervention” (table 3, columns 3 and 7, row for “Log of Cash Income”). Controlling for the presence of NREGA public works strengthened this coefficient, indicating that labor markets during the period of study were being considerably transformed by NREGA. We cannot rule out the possibly that the transformation occurred at a different pace in treatment and control villages.12 Anecdotal evidence from field-workers as well as local government representatives suggests that the program was highly popular among women from both treatment and control villages and they chose to partici- pate in NREGA public works projects in large numbers. Both self-employment and entrepreneurship, already at very low levels in Dungarpur, fell even further as a result. We return to this issue below. 11. Complete estimates are available from the authors on request. 12. In both 2007 and 2009, we observe no difference in either the intensity of NREGA programs, or the timing of its rollout, between treatment and control villages, but it is possible that the program was rolled out quicker in group of villages. T A B L E 3 . Village Treatment Effects, Unconditional and Conditional Estimates Unconditional Estimates (1) – (4) Conditional Estimates (5) – (8) (1) (2) (3) (4) (5) (6) (7) (8) SEWA village SEWA village resident  Post SEWA village resident  Post SEWA village Intervention resident Post Intervention R2 Intervention resident Post Intervention R2 Participates in 0.238*** (0.052) 2 0.007 (0.030) 0.055* (0.032) 0.080 0.243*** (0.049) 2 0.007 (0.027) 0.076** (0.035) 0.110 group programs In the habit of 0.105** (0.043) 2 0.041* (0.024) 2 0.001 (0.028) 0.011 0.108** (0.043) 2 0.042* (0.025) 0.019 (0.031) 0.042 saving Credit 0.029 (0.038) 2 0.014 (0.022) 0.011 (0.017) 0.004 0.033 (0.037) 2 0.018 (0.021) 0.016 (0.019) 0.024 Cash savings 0.162 (0.229) 2 0.017 (0.122) 0.424*** (0.151) 0.013 0.123 (0.224) 2 0.004 (0.108) 0.362** (0.154) 0.047 (log, 3 months) Cash income 2 0.315 (0.352) 0.492* (0.295) 2 0.738*** (0.194) 0.051 2 0.167 (0.285) 0.365 (0.249) 2 0.509*** (0.153) 0.108 (log, 3 months) Employed (past 2 0.029 (0.056) 0.033 (0.040) 0.015 (0.025) 0.004 2 0.002 (0.040) 0.028 (0.027) 0.038 (0.026) 0.267 3 months) Employed 0.053 (0.033) 2 0.016 (0.017) 2 0.010 (0.021) 0.005 0.051* (0.029) 2 0.022 (0.016) 0.008 (0.021) 0.065 (non-farm past 3 months) Final say: 0.061** (0.026) 0.005 (0.022) 2 0.028** (0.014) 0.009 0.047* (0.024) 0.004 (0.018) 2 0.020 (0.015) 0.163 children’s schooling Final say: medical 0.075*** (0.028) 2 0.013 (0.021) 2 0.047*** (0.017) 0.007 0.066** (0.027) 2 0.018 (0.017) 2 0.029 (0.018) 0.137 Desai and Joshi decisions Final say: 0.068*** (0.017) 2 0.037** (0.014) 2 0.045*** (0.013) 0.012 0.063*** (0.016) 2 0.034*** (0.012) 2 0.050*** (0.015) 0.032 family-planning Grievance: water 0.137** (0.055) 0.017 (0.027) 0.087*** (0.026) 0.046 0.129** (0.052) 0.004 (0.020) 0.109*** (0.027) 0.109 Grievance: roads 0.036 (0.055) 0.012 (0.024) 0.111*** (0.031) 0.031 0.040 (0.055) 0.002 (0.023) 0.155*** (0.034) 0.075 Grievance: 0.084 (0.069) 0.026 (0.027) 0.198*** (0.034) 0.084 0.089 (0.069) 0.014 (0.027) 0.233*** (0.037) 0.110 electricity 507 (Continued ) TABLE 3. Continued 508 Unconditional Estimates (1) – (4) Conditional Estimates (5) – (8) (1) (2) (3) (4) (5) (6) (7) (8) THE WORLD BANK ECONOMIC REVIEW SEWA village SEWA village resident  Post SEWA village resident  Post SEWA village Intervention resident Post Intervention R2 Intervention resident Post Intervention R2 Grievance: 0.003 (0.031) 0.006 (0.030) 0.078 (0.053) 0.009 2 0.008 (0.029) 0.046 (0.031) 0.081 (0.052) 0.051 education/ health Addressed 0.107* (0.055) 0.019 (0.025) 0.082*** (0.023) 0.037 0.100* (0.052) 0.009 (0.022) 0.100*** (0.025) 0.083 grievance: water Addressed 0.012 (0.051) 0.023 (0.022) 0.113*** (0.029) 0.028 0.016 (0.051) 0.014 (0.021) 0.153*** (0.031) 0.058 grievance: roads Addressed 0.062 (0.058) 0.018 (0.022) 0.152*** (0.029) 0.061 0.068 (0.058) 0.013 (0.023) 0.183*** (0.031) 0.077 grievance: electricity Addressed 2 0.006 (0.024) 0.016 (0.024) 0.064 (0.043) 0.009 2 0.011 (0.025) 0.045 (0.028) 0.067 (0.042) 0.033 grievance: education/ health Know of Gram 0.047 (0.055) 2 0.001 (0.033) 0.010 (0.038) 0.013 0.049 (0.054) 2 0.011 (0.031) 0.011 (0.041) 0.072 Sabha and Panchayat Engage with Gram 0.018 (0.011) 2 0.008 (0.007) 2 0.010 (0.007) 0.002 0.018 (0.011) 2 0.009 (0.007) 2 0.007 (0.008) 0.022 Sabha and Panchayat Known anyone 0.051** (0.021) 2 0.015 (0.015) 2 0.041*** (0.011) 0.008 0.051** (0.020) 2 0.017 (0.014) 2 0.035** (0.014) 0.016 who has paid a bribe Notes: Columns 1 – 4 present estimates for specified coefficients by regressing listed outcomes on village-treatment indicators (residence in a SEWA village) plus a constant and block (sub-district) fixed effects. Columns 5 –8 are OLS results with the following, additional controls: age (quadratic), literacy, marital status, caste, husband’s age, husband’s literacy, home ownership, farm ownership, kutcha dwelling, flush toilet, NREGA in village, and bias adjustment, with block (sub-district) fixed effects. Robust standard errors (in parentheses) are clustered at the village level. *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,205. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. Desai and Joshi 509 I I I . RE S U LT S : I N D I V I D U A L PA R T I C I PAT I O N In addition to effects of village-level treatment, we are interested in the effects of individual membership (and non-membership) in an SHG in a SEWA (treatment) village. We examine these treatment effects with the following functional form: Yi;v;b;t ¼g0 þ g1 ðSEWA memberi  Post-interventiont Þ þ g2 ðSEWA village non-memberi  Post-interventiont Þ þ g3 Post-interventiont þ g4 SEWA villagev þ g5 Xi;v;t þ mb þ 1i;v;t where Yi,v,t is the outcome of interest for individual i in village v during survey phase t. SEWA Member takes value 1 if an individual participated in SEWA pro- grams (launched after the baseline survey), SEWA village non-member  Post-intervention takes value 1 if the individual resides in a SEWA village but was not a member of the SHG, X is the same vector of individual and household control variables described in the previous section, m is a village fixed-effect and ei,v,t is a standard disturbance. From this estimation g1 is the effect of participa- tion in a SEWA program by SHG members, g2 is the spill-over effect, g3 is the time effect, and g4 is a measure of pre-program differences between SEWA and non-SEWA villages. As this is a pooled cross-section, there were no SEWA villag- es/members in the baseline survey.13 The central challenge in estimating individual impact is that membership within villages is not randomly assigned. As mentioned earlier, SEWA randomly selected villages, and though membership was open to all women, actual partici- pation could not be randomized. Information about the program was disseminat- ed widely but we cannot discount the possibility of intra-village selection bias. To address this selection problem, we pre-process our data with propensity matching methods, then re-run our parametric analyses weighted by the propen- sity score as a bias-adjustment for matching (Abadie and Imbens 2006). This ensures that SEWA members are as similar as possible to non-members in terms of relevant covariates (Caliendo and Kopeinig 2008). Our propensity score is es- timated with a logit regression of SEWA membership on age, education, marital status, husband’s age, husband’s education, family size and the number of mi- grants in the household. We construct the matched sample using one-to-one matching without replacement.14 13. Since there were no SEWA members in the 2007 survey, “SEWA memberi  Post-interventiont” could simply be written “SEWA memberi.” We use the full interaction term to emphasize the treatment effect. 14. We tried all permutations and combinations of variables in the match equation and performed sensitivity checks for all the results presented in this paper. We found that the size of the matched sample remained within 10 percent of the sample reported here. We also checked the robustness of the results using caliper matching and kernel matching methods, and again found similar sample sizes as well as estimated coefficients. These results are available upon request. 510 THE WORLD BANK ECONOMIC REVIEW Our selection of variables to conduct matching is guided by existing literature as well as SEWA’s field strategies. The inclusion of education and land is motivat- ed by the findings from a variety of studies that have found that educated and wealthier women are better positioned to understand the benefits of participa- tion in community based development programs (Arcand and Fafchamps 2011; Bernard and Spielman 2009). A test of balance is presented in the Supplementary Appendix, table S.2. This table presents summary statistics of key variables for the unmatched and matched samples. The standardized bias is reported as a percentage before and after matching. This estimate is the difference of the sample means in the treated and non-treated (full or matched) sub-samples as a percentage of the square root of the average of the sample variances in the treated and non-treated groups (for- mulae from Rosenbaum and Rubin, 1985). The estimates confirm a significant reduction in bias from the matching procedure: we cannot reject the hypothesis of equality of the characteristics across the treated and non-treated groups. Estimates of the program’s impact, ( g1, g2, and g3) are presented in table 4 (we omit estimates of pre-program differences). Unconditional estimates are pre- sented in columns 1-4 and conditional estimates are presented in columns 5-8. Note that for almost every outcome, we once again find that unconditional and conditional impacts are very similar in magnitude as well as significance. This discussion will focus on conditional estimates. In the matched sample, more than 55 percent of SEWA members participate in group programs, and more than 20 percent of members report that they save regular- ly. These estimates are significant at the 1 percent level. There is no effect of SEWA participation however, on the actual amount saved three months prior to the survey. Here, as in table 3, women report an average income loss during the period under study due to drought-induced agricultural distress in Dungarpur district. However, at least for unconditional affects, we see that SEWA membership (as did SEWA’s presence in the village) provided a “cushion” against these shocks, with SEWA members reporting no significant change in income, and with non- members reporting income loss. With control variables added, there is no diffe- rence between members and non-members. As found in the case of the village-level impact estimates, we once again find that participating women were 7-8 percentage points more likely to be employed outside of agriculture. The effect is significant at the 10 percent level. The esti- mated improvements in women’s bargaining power are also positive and signifi- cant at the 1 percent level. Here, we find that SEWA members as well as non-members experienced benefits: members were 8-12 percentage points more likely to participate in family decisions. The coefficients for non-members are smaller (with the exception of family-planning) but statistically significant. We continue to find that the programs have a strong and significant effect on knowledge of where to express grievances as well as women’s willingness to report their grievances. For all our measures of grievances, participants report that they are 11-20 percentage points more likely to know where to report them. T A B L E 4 . Individual Participation Effects, Unconditional and Conditional Estimates Unconditional Estimates (1) –(4) Conditional Estimates (5) –(8) (1) (2) (3) (4) (5) (6) (7) (8) SEWA village SEWA village SEWA member  non-member  SEWA member  non-member  Post-intervention Post-intervention Post-intervention R2 Post-intervention Post-intervention Post-intervention R2 Participates in 0.546*** (0.070) 0.025 (0.053) 0.038 (0.035) 0.301 0.549*** (0.071) 0.024 (0.052) 0.041 (0.039) 0.316 group programs In the habit of 0.199*** (0.059) 0.044 (0.050) 2 0.008 (0.033) 0.096 0.208*** (0.064) 0.044 (0.050) 0.005 (0.036) 0.121 saving Credit 0.091 (0.057) 0.001 (0.045) 0.013 (0.021) 0.094 0.101* (0.057) 2 0.003 (0.042) 0.001 (0.024) 0.114 Cash savings 0.343 (0.258) 2 0.187 (0.278) 0.396** (0.166) 0.092 0.366 (0.281) 2 0.263 (0.280) 0.303* (0.169) 0.122 (log, 3 months) Cash income 2 0.169 (0.328) 2 0.485* (0.291) 2 0.894*** (0.175) 0.137 0.089 (0.297) 2 0.224 (0.274) 2 0.613*** (0.132) 0.170 (log, 3 months) Employed (past 2 0.051 (0.059) 2 0.075 (0.069) 0.027 (0.029) 0.118 2 0.051 (0.052) 2 0.024 (0.046) 0.061* (0.031) 0.286 3 months) Employed 0.070** (0.030) 0.039 (0.029) 2 0.028* (0.015) 0.132 0.081** (0.031) 0.039 (0.029) 2 0.013 (0.016) 0.173 (non-farm past 3 months) Final say: children’s 0.128*** (0.042) 0.070* (0.039) 2 0.037** (0.018) 0.084 0.119*** (0.040) 0.081** (0.036) 2 0.020 (0.019) 0.252 schooling Final say: medical 0.133*** (0.040) 0.078** (0.036) 2 0.060*** (0.020) 0.085 0.128*** (0.040) 0.088** (0.035) 2 0.047** (0.020) 0.217 decisions Final say: 0.076*** (0.021) 0.096*** (0.023) 2 0.050*** (0.014) 0.090 0.074*** (0.021) 0.089*** (0.021) 2 0.057*** (0.015) 0.106 Desai and Joshi family-planning Grievance: water 0.221*** (0.067) 0.075 (0.062) 0.090*** (0.025) 0.143 0.202*** (0.063) 0.069 (0.059) 0.101*** (0.027) 0.183 Grievance: roads 0.101 (0.076) 0.007 (0.062) 0.108*** (0.032) 0.146 0.091 (0.072) 0.015 (0.063) 0.146*** (0.035) 0.175 Grievance: 0.123* (0.070) 0.015 (0.077) 0.210*** (0.034) 0.190 0.128* (0.074) 0.023 (0.079) 0.232*** (0.036) 0.204 electricity Grievance: 0.131** (0.061) 0.014 (0.053) 0.008 (0.032) 0.121 0.114* (0.058) 0.012 (0.053) 0.027 (0.035) 0.150 education/health 511 (Continued ) TABLE 4. Continued 512 Unconditional Estimates (1) –(4) Conditional Estimates (5) –(8) THE WORLD BANK ECONOMIC REVIEW (1) (2) (3) (4) (5) (6) (7) (8) SEWA village SEWA village SEWA member  non-member  SEWA member  non-member  Post-intervention Post-intervention Post-intervention R2 Post-intervention Post-intervention Post-intervention R2 Addressed 0.151** (0.065) 0.070 (0.057) 0.084*** (0.022) 0.123 0.142** (0.064) 0.065 (0.054) 0.094*** (0.025) 0.155 grievance: water Addressed 0.062 (0.067) 2 0.000 (0.062) 0.110*** (0.030) 0.135 0.052 (0.065) 0.006 (0.062) 0.143*** (0.032) 0.154 grievance: roads Addressed 0.089 (0.066) 2 0.003 (0.065) 0.162*** (0.031) 0.139 0.096 (0.068) 0.005 (0.065) 0.181*** (0.033) 0.149 grievance: electricity Addressed 0.107* (0.054) 0.009 (0.044) 0.018 (0.024) 0.099 0.100* (0.050) 0.010 (0.043) 0.031 (0.030) 0.115 grievance: education/health Know of Gram 0.115* (0.067) 0.049 (0.057) 2 0.001 (0.039) 0.123 0.112* (0.065) 0.050 (0.061) 2 0.005 (0.041) 0.183 Sabha and Panchayat Engage with Gram 0.051*** (0.019) 0.016 (0.010) 2 0.017** (0.008) 0.057 0.052*** (0.019) 0.019* (0.011) 2 0.014** (0.007) 0.086 Sabha and Panchayat Known anyone who 0.067** (0.028) 0.062** (0.030) 2 0.052*** (0.014) 0.082 0.061** (0.028) 0.062* (0.031) 2 0.048** (0.019) 0.090 has paid a bribe Notes: Columns 1 – 4 present estimates for specified coefficients generated by regressing listed outcomes on individual treatment indicators (membership in a SEWA group) plus a constant and village-fixed effects. Columns 5 –8 are OLS results with the following, additional controls: age (quadratic), literacy, marital status, caste, husband’s age, husband’s literacy, home ownership, farm ownership, kutcha dwelling, flush toilet, NREGA in village, and bias adjust- ment, with village-fixed effects. All estimations are weighted by a propensity score, generated by one-to-one matching (logit) on SEWA participation without replacement. Robust standard errors (in parentheses) are clustered at the village level. *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,158. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. Desai and Joshi 513 The coefficients for actually reporting grievances are lower, but are statistically significant in the case of drinking water and considerably larger than the village- level averages reported earlier. It is striking that we see no spillovers for these in- dicators of civic engagement. Collective action remained restricted to SEWA members and did not draw in non-members. We also find evidence that SEWA participation had effects on political partici- pation: estimates from unconditional and conditional regressions confirm that SEWA members were 11 percentage points more likely to know of the local Gram Panchayat and Gram Sabha and were 5 percentage points more likely to engage with these institutions. Finally, we note that SEWA members are also 7 percentage points more likely to be aware of bribing within the village. The result is statistically significant at the 5 percent level. Here too, we again find spillovers: non-members in SEWA vil- lages are also 7 percentage points more likely to be aware of the payment of bribes and the result is significant at the 5 percent level. Robustness Checks We perform several robustness checks of unconditional and conditional esti- mates using a variety of different matching functions as well as matching methods. Table S.3 in the Supplementary Appendix contains estimates of treat- ment effects using “coarsened exact matching” (CEM). CEM-based causal esti- mates have been shown to reduce imbalance, model dependence and estimation error with informative data (Iacus, King, and Porro 2012). Note that we obtain very similar estimates to those that are reported in the paper. In some cases we observe greater impacts than those we report in the main text of the paper. We also check the sensitivity of our results to hidden bias, i.e. bias induced by unobservable factors that could affect participation in the program itself (an indi- vidual’s motivation, prior experience with NGOs, etc). We use Rosenbaum’s bounding approach (Becker and Caliendo 2007; DiPrete and Gangl 2004; Rosenbaum 2002). Rosenbaum bounds calculate upper- and lower-bounds for the average treatment effect on treated individuals in the presence of unobserved het- erogeneity which is assumed to influence participation in the program.15 Results are reported in the Supplementary Appendix, table S.4 for some key variables. I V. V A L I D I T Y AND EXTENSIONS Heterogeneity Next, we explore whether the program had differential impacts across women of different socio-economic groups. Much recent research illustrates that 15. Upper (lower) bounds adjust coefficients downwards for positive (negative) selection, i.e. the possibility that people with the best outcomes selected into (out of ) the program, introducing upward bias into the effects of our program. 514 THE WORLD BANK ECONOMIC REVIEW community-based development projects such as this one may be susceptible to elite capture (Alatas et al. 2013; Bardhan and Mookerjee 2006; Gugerty and Kremer 2008; Mansuri and Rao 2012). In other parts of India, diversity within groups has been shown to have an effect on group performance as well as stabil- ity (Baland, Somanathan, and Vandewalle 2011). To explore this, we interact the main treatment indicators with measures of education and land-ownership. We define two binary indicators—“Illiterate” and “Landless”—that respectively take value 1 if a woman is illiterate and 0 oth- erwise, and value 1 if a woman’s household owns no land and 0 otherwise. Results are presented in table 5 (village-level treatment variable) and table 6 (individual-level treatment variable). In table 5, we note that landless women benefitted more from SEWA programs. They are about 16 percentage points likely to participate in group programs or save, have higher cash incomes (despite the drought) but are 17 percent less likely to take action regarding grievances regarding drinking water. Interviews with SEWA field workers and SHG members confirm that there is no selective targeting of women, but the lower par- ticipation of landless women in voicing grievances is likely to be driven by the higher time and information costs faced by these women in the rural economy. The higher cash incomes of landless women may also confirm their selection into the NREGA program. In table 6, we find weaker individual effects of landlessness and illiteracy. In fact, we find that landless SHG members are 13 percentage points less likely to save, presumably because they are borrowing from the group’s internal funds. We also find that landless members are more likely to be employed in the three months prior to the survey. We find no evidence however, that landless or illiterate women were either particularly targeted or discriminated in this program. The results suggest that SEWA programs did not disproportionately benefit the educated or socio-economically wealthy women. This result is similar to recent evidence from other contexts (Alatas et al. 2013; Olken 2010). Mechanisms Our analysis has so far focused on estimating the total impact of a comprehensive package of efforts. Identifying the specific component of this package that bene- fits members is more difficult (Green, Ha, and Bullock. 2010; Imai 2011). This is largely due to program design: SHGs were rapidly established in all treatment vil- lages, but additional modules were not simultaneously rolled out. Our design therefore provides exogenous variation in the application of the SEWA liveli- hoods program across villages, but not on potentially intermediary variables that can affect the outcomes observed. Nevertheless data on the additional modules can be used to examine the extent to which different mechanisms are at work in treatment villages, as well as among SEWA members. We examine two modules that were core parts of SEWA’s broader intervention in Dungarpur under the Backward Districts Initiative, and that were T A B L E 5 . Heterogeneity of Impact, Village Treatment SEWA village SEWA village resident  resident  SEWA village SEWA village SEWA village Post-intervention  Post-intervention  resident  resident  resident  SEWA village Illiterate Landless Post-intervention Illiterate Landless resident Post-intervention R2 Participates in 2 0.016 (0.095) 0.160** (0.063) 0.230** (0.087) 2 0.013 (0.042) 0.005 (0.054) 0.002 (0.047) 0.072** (0.035) 0.114 group programs In the habit of 0.026 (0.067) 0.160** (0.062) 0.061 (0.067) 0.003 (0.050) 2 0.020 (0.056) 2 0.043 (0.054) 0.017 (0.031) 0.045 saving Credit 2 0.040 (0.063) 2 0.044 (0.063) 0.073 (0.073) 2 0.026 (0.041) 0.024 (0.052) 2 0.000 (0.041) 0.015 (0.019) 0.025 Cash savings 2 0.429 (0.321) 0.407 (0.368) 0.403 (0.358) 0.277 (0.256) 0.104 (0.275) 2 0.249 (0.252) 0.351** (0.154) 0.049 (log, 3 months) Cash income 2 0.348 (0.333) 0.749* (0.388) 2 0.007 (0.426) 0.466* (0.280) 2 0.328 (0.382) 0.029 (0.351) 2 0.511*** (0.152) 0.111 (log, 3 months) Employed (past 2 0.072 (0.071) 2 0.046 (0.067) 0.063 (0.067) 2 0.019 (0.041) 2 0.059 (0.054) 0.053 (0.045) 0.039 (0.026) 0.269 3 months) Employed 2 0.073 (0.060) 0.006 (0.029) 0.108 (0.071) 0.008 (0.032) 2 0.005 (0.023) 2 0.028 (0.033) 0.007 (0.021) 0.067 (non-farm past 3 months) Final say: 0.055 (0.049) 0.022 (0.046) 2 0.001 (0.056) 0.001 (0.041) 2 0.047 (0.030) 0.010 (0.043) 2 0.018 (0.015) 0.165 children’s schooling Final say: medical 0.032 (0.047) 0.005 (0.045) 0.039 (0.057) 0.001 (0.045) 2 0.061* (0.036) 2 0.009 (0.044) 2 0.027 (0.018) 0.139 decisions Final say: family- 2 0.004 (0.024) 2 0.012 (0.018) 0.068*** (0.023) 2 0.016 (0.025) 0.004 (0.015) 2 0.022 (0.025) 2 0.051*** (0.015) 0.033 Desai and Joshi planning Grievance: water 0.101 (0.088) 2 0.131 (0.080) 0.068 (0.088) 2 0.073 (0.063) 0.141** (0.062) 0.043 (0.058) 0.109*** (0.027) 0.112 Grievance: roads 0.038 (0.073) 0.007 (0.067) 0.008 (0.089) 2 0.031 (0.070) 0.034 (0.065) 0.022 (0.069) 0.154*** (0.034) 0.075 Grievance: 0.067 (0.083) 2 0.006 (0.053) 0.035 (0.101) 2 0.023 (0.045) 0.013 (0.045) 0.031 (0.055) 0.233*** (0.037) 0.111 electricity Grievance: 2 0.015 (0.060) 0.084 (0.064) 0.079 (0.071) 0.011 (0.053) 0.061 (0.056) 2 0.027 (0.054) 0.043 (0.031) 0.054 education/ 515 health (Continued ) TABLE 5. Continued 516 SEWA village SEWA village resident  resident  SEWA village SEWA village SEWA village THE WORLD BANK ECONOMIC REVIEW Post-intervention  Post-intervention  resident  resident  resident  SEWA village Illiterate Landless Post-intervention Illiterate Landless resident Post-intervention R2 Addressed 0.092 (0.081) 2 0.165** (0.071) 0.053 (0.087) 2 0.052 (0.068) 0.134** (0.064) 0.032 (0.068) 0.100*** (0.025) 0.087 grievance: water Addressed 0.041 (0.058) 2 0.004 (0.066) 2 0.017 (0.070) 2 0.057 (0.068) 0.043 (0.065) 0.054 (0.066) 0.151*** (0.031) 0.059 grievance: roads Addressed 0.071 (0.064) 2 0.033 (0.056) 0.017 (0.078) 2 0.002 (0.044) 0.012 (0.047) 0.013 (0.045) 0.184*** (0.031) 0.078 grievance: electricity Addressed 0.030 (0.070) 0.016 (0.058) 0.041 (0.071) 0.019 (0.056) 0.042 (0.048) 2 0.034 (0.055) 0.045 (0.028) 0.034 grievance: education/ health Know of Gram 2 0.063 (0.087) 0.121 (0.085) 0.080 (0.101) 0.054 (0.055) 2 0.005 (0.057) 2 0.055 (0.064) 0.009 (0.041) 0.074 Sabha and Panchayat Engage with 2 0.037 (0.023) 0.005 (0.025) 0.047* (0.024) 0.025 (0.017) 0.025 (0.021) 2 0.034* (0.018) 2 0.008 (0.008) 0.025 Gram Sabha and Panchayat Known anyone 0.003 (0.031) 0.071* (0.042) 0.038 (0.035) 2 0.006 (0.030) 0.022 (0.020) 2 0.016 (0.032) 2 0.037** (0.014) 0.022 who has paid a bribe Notes: Estimates are for specified coefficients generated by regressing listed outcomes on village-treatment indicators (residence in a SEWA village) along with the following controls: age (quadratic), literacy, marital status, caste, husband’s age, husband’s literacy, home ownership, farm ownership, kutcha dwelling, flush toilet, NREGA in village, and bias adjustment, with village-fixed effects. Robust standard errors (in parentheses) are clustered at the village level. *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,158. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. T A B L E 6 . Heterogeneity of Impact, Individual Participation Non-member  SEWA village SEWA member  SEWA member  resident  Post- Illiterate Landless SEWA member intervention Post-intervention Illiterate Landless R2 Participates in group programs 2 0.038 (0.082) 2 0.100 (0.110) 0.587*** (0.087) 0.025 (0.051) 0.041 (0.039) 2 0.088*** (0.033) 2 0.019 (0.024) 0.317 In the habit of saving 0.062 (0.085) 2 0.136** (0.059) 0.178** (0.071) 0.045 (0.049) 0.006 (0.036) 2 0.116*** (0.041) 2 0.002 (0.030) 0.122 Credit 2 0.098 (0.063) 2 0.052 (0.094) 0.176** (0.076) 2 0.003 (0.042) 0.001 (0.024) 2 0.006 (0.023) 2 0.032 (0.026) 0.117 Cash savings (log, 3 months) 2 0.254 (0.465) 1.076 (0.837) 0.408 (0.466) 2 0.279 (0.281) 0.307* (0.169) 2 0.288 (0.181) 2 0.175 (0.213) 0.124 Cash income (log, 3 months) 2 0.423 (0.325) 0.503 (0.607) 0.308 (0.429) 2 0.239 (0.275) 2 0.608*** (0.132) 2 0.047 (0.195) 0.197 (0.199) 0.170 Employed (past 3 months) 2 0.055 (0.063) 0.244** (0.093) 2 0.041 (0.064) 2 0.026 (0.046) 0.061* (0.031) 2 0.018 (0.025) 2 0.543*** (0.031) 0.290 Employed (non-farm past 3 2 0.042 (0.050) 0.084 (0.113) 0.099* (0.053) 0.038 (0.029) 2 0.012 (0.016) 2 0.034 (0.021) 0.106*** (0.023) 0.174 months) Final say: children’s schooling 0.145* (0.075) 2 0.042 (0.079) 0.014 (0.075) 0.079** (0.036) 2 0.017 (0.019) 0.013 (0.026) 2 0.003 (0.018) 0.253 Final say: medical decisions 0.110 (0.076) 2 0.051 (0.089) 0.048 (0.073) 0.086** (0.036) 2 0.044** (0.020) 0.008 (0.028) 2 0.003 (0.026) 0.215 Final say: family-planning 0.032 (0.037) 0.030 (0.056) 0.046 (0.028) 0.087*** (0.021) 2 0.056*** (0.015) 2 0.001 (0.016) 0.026** (0.012) 0.105 Grievance: water 0.032 (0.086) 2 0.048 (0.080) 0.190** (0.084) 0.072 (0.059) 0.099*** (0.028) 2 0.171*** (0.036) 2 0.012 (0.040) 0.182 Grievance: roads 0.003 (0.088) 2 0.107 (0.119) 0.110 (0.110) 0.020 (0.062) 0.144*** (0.035) 2 0.089* (0.045) 2 0.009 (0.032) 0.172 Grievance: electricity 2 0.001 (0.110) 2 0.171 (0.154) 0.150 (0.116) 0.025 (0.079) 0.232*** (0.036) 2 0.096** (0.040) 2 0.010 (0.029) 0.205 Grievance: education/health 2 0.002 (0.093) 2 0.059 (0.117) 0.126 (0.088) 0.014 (0.053) 0.026 (0.035) 2 0.090** (0.035) 0.031 (0.029) 0.150 Addressed grievance: water 0.090 (0.095) 2 0.058 (0.065) 0.089 (0.096) 0.068 (0.054) 0.093*** (0.025) 2 0.146*** (0.040) 0.010 (0.035) 0.155 Addressed grievance: roads 2 0.019 (0.077) 2 0.043 (0.110) 0.078 (0.095) 0.010 (0.061) 0.140*** (0.032) 2 0.072* (0.041) 2 0.033 (0.032) 0.151 Addressed grievance: 0.046 (0.093) 2 0.088 (0.118) 0.073 (0.099) 0.006 (0.064) 0.182*** (0.033) 2 0.058* (0.033) 2 0.013 (0.028) 0.150 electricity Addressed grievance: 0.099 (0.086) 0.015 (0.115) 0.030 (0.065) 0.011 (0.042) 0.031 (0.030) 2 0.058* (0.034) 2 0.005 (0.021) 0.117 education/health Know of Gram Sabha and 2 0.157** (0.070) 2 0.026 (0.077) 0.226*** (0.085) 0.050 (0.060) 2 0.006 (0.041) 2 0.186*** (0.036) 2 0.063** (0.029) 0.186 Panchayat Engage with Gram Sabha and 2 0.060 (0.047) 2 0.001 (0.059) 0.095** (0.046) 0.020* (0.011) 2 0.015** (0.007) 2 0.023* (0.013) 0.002 (0.008) 0.091 Desai and Joshi Panchayat Known anyone who has paid a 0.059** (0.028) 0.006 (0.060) 0.018 (0.034) 0.062* (0.031) 2 0.047** (0.019) 2 0.029 (0.019) 0.031 (0.023) 0.092 bribe Notes: Estimates are for specified coefficients generated by regressing listed outcomes on the individual-treatment indicators (membership in a SEWA group) along with the following controls: age (quadratic), literacy, marital status, caste, husband’s age, husband’s literacy, home ownership, farm ownership, kutcha dwelling, flush toilet, NREGA in village, and bias adjustment, with village-fixed effects. All estimations are weighted by a propensity score, generated by one-to-one matching (logit) on SEWA participation without replacement. Robust standard errors (in parentheses) are clustered at the village level. 517 *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,158. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. 518 THE WORLD BANK ECONOMIC REVIEW implemented in the first year of the intervention.16 The two specialized training modules we consider were offered by SEWA to improve women’s economic op- portunities. First, SEWA ran a series of agricultural workshops covering farming techniques—based on similar workshops used in “training and visit” initiatives that were part of agricultural extension systems in India (Feder, Willett, and Zijp 2001), but targeted towards female farmers17—as well as workshops on craft-making (fabric weaving, embroidery, and handloom operation) or on the manufacture of simple products (washing powder and incense sticks) as well as supplemental training on pricing and packaging. Women who participated in these vocational modules were eligible to receive support from SEWA’s Producer Cooperative (Gram Mahila Haat), which would provide marketing and distribu- tion support to female producer groups.18 Second, SEWA supported financial awareness and capacity-building efforts that focused on helping women manage household finances, and better understand the use of savings accounts, lending operations of rural banks, and micro-enterprise development through SEWA’s own microfinance institution, SEWA Bank. Of the 32 SEWA-treatment villages, the vocational training module was implemented in 19, the financial capacity- building module in 22, while 14 villages received both.19 To examine the effects of these specialized programs, for each we construct two new specialized “treatment” variables: (i) a village-based module indicator coded 1 if the module was implemented in the village, or 0 otherwise; and (ii) an individual-based module indicator, coded 1 if the individual participated in the specific module run in the village, 0 otherwise. In the latter care, the room for spillover effects is potentially quite large given that any training or capacity- building modules run by SEWA were always open to all (female) village residents whether or not they were SEWA members. These narrower treatment variables may be used to identify effects of particular SEWA services on specific outcomes. Results from these tests are presented in table 7. We begin by examining the effect of the vocational-training module on three separate outcomes to which agricultural and vocational training could be directly linked: income (in natural logs), outside employment (whether the re- spondent worked outside the household in the past three months), and outside 16. The NRLM, additionally, envisions grant support for similar modules of SHGs as part of their expansion. 17. Investments in agricultural-support programs—such as agricultural extension—have typically excluded women and have almost exclusively been targeted at men (Danida 2002;Raabe 2008). During India’s green revolution and land reforms, state-led rural development programs were almost exclusively targeted to men, and training offered through the “training and visit” system was primarily aimed at male farmers (Berger, Delancey and Mellencamp 1984;Macklin 1992). 18. SEWA Gram Mahila Haat (SGMH) was established in 1998 to provide marketing and support services to rural producer associations. Among the services offered were a common “branding” of goods (both agricultural and non-agricultural) made by SEWA’s groups, which SGMH could then purchase and resell through SGMH-run retail shops. 19. We ignore the effects of some of the other modules given the low rates of participation: healthcare training, water purification, and childcare services. T A B L E 7 . Mechanisms Vocational Training (1) –(3) Financial-Capacity Building (4) –(6) SEWA modules: (1) (2) (3) (4) (5) (6) Employed Employed (non-farm, Cash savings Outcomes: Cash income (log) (3 months) 3 months) Regular saving Credit utilization (log) (A)Village-level treatment SEWA module village 2 0.014 (0.356) 0.031 (0.043) 0.077** (0.032) 0.077* (0.032) 0.067* (0.040) 0.248 (0.234) resident  Post-intervention SEWA module village resident 0.323 (0.285) 2 0.016 (0.032) 2 0.031** (0.015) 2 0.018 (0.024) 2 0.030 (0.019) 0.074 (0.108) Post-intervention 2 0.596*** (0.151) 0.029 (0.027) 0.007 (0.018) 0.035 (0.031) 0.004 (0.019) 0.315** (0.148) R2 0.108 0.267 0.067 0.040 0.026 0.049 (B) Individual-level treatment SEWA module participant 0.066 (0.399) 2 0.177 (0.114) 0.143* (0.080) 0.349*** (0.091) 0.126* (0.076) 0.316 (0.344) Nonparticipant  SEWA module 2 0.020 (0.337) 2 0.053 (0.054) 0.061 (0.039) 0.090* (0.048) 0.037 (0.043) 0.195 (0.238) village resident  Post-intervention Post-intervention 2 0.674*** (0.163) 0.098*** (0.033) 2 0.004 (0.026) 0.017 (0.040) 0.003 (0.025) 0.260 (0.167) R2 0.195 0.308 0.188 0.112 0.099 0.113 Notes: 19 treatment villages had vocational training employment programs during the study period, and 22 treatment villages had financial capacity- building programs. Estimates are for listed coefficients generated by regressing specified outcomes on village-treatment (residence in a village that implement- ed a SEWA training or finance module—panel A) and individual-treatment ( participation in the SEWA-run training or finance module) along with the Desai and Joshi following controls: age (quadratic), literacy, marital status, caste, husband’s age, husband’s literacy, home ownership, farm ownership, kutcha dwelling, flush toilet, NREGA in village, and bias adjustment. Village-treatment estimations include block (sub-district) fixed effects, while Individual-treatment estimations include village-fixed effects. Panel B estimations are weighted by a propensity score, generated by one-to-one matching (logit) on SEWA participation without replacement. Robust standard errors (in parentheses) are clustered at the village level. *p , 0.10, **p , 0.05, ***p , 0.01. N ¼ 3,158. Source: Authors’ analysis based on data from the Self-Employed Women’s Association. 519 520 THE WORLD BANK ECONOMIC REVIEW non-agricultural employment (whether respondent worked in a non-farm capaci- ty in the past three months). As with previous specifications, in all cases we examine effects at the individual level of living in a village that implemented a SEWA-run vocational training module, as well as of being a participant in the vo- cational training module. Panel A presents results of the village-level treatment while panel B shows results of individual participation. As with previous results, we weight regressions in the second panel by the propensity score, generated from a matching model using one-to-one matching without replacement.20 We do not observe strong effects of vocational training among women who reside in villages where those modules were run, with the exception of non-farm employment. We see no effects, for example, on income earned or outside em- ployment. The incidence of non-farm employment in villages where SEWA would run vocational training modules was 3 percentage points lower than in vil- lages where no SEWA training took place prior to program implementation. However, women who live in villages that received vocational training saw their incomes rise by 8 percentage points compared to their counterparts in villages where no SEWA training was implemented. The effect of individual participation in SEWA’s vocational training modules— as opposed to the effect of residing in a village where SEWA’s vocational training modules were run—is similar: vocational training participants increased their inci- dence of non-farm employment by 14 percentage points compared to non- participants. In this case, moreover, we see evidence of non-farm employment spillovers in that non-members in villages where SEWA’s training programs were run benefited from a 7 percentage points increase in non-farm employment. Turning to financial capacity-building activities, we examine three additional, specific impacts: whether the individual made a deposit into a bank account, whether the individual received a loan through the SHG-bank mechanism, and the total amount saved over the past 6 months. Women who lived in villages where financial capacity-building modules were run were 8 percentage points more likely to save regularly and 7 percentage points more likely to have taken a loan than women in control villages. Individual women who participated in financial capacity-building modules were 35 percentage points more likely to save and 13 percentage points more likely to borrow, than average non-participating counterparts. With the exception of saving—non-participants living in villages where SEWA financial modules were implemented were 9 percentage points more likely to save—we observe no spillover effects of financial capacity building. Finally, we see no effects of finance activities on actual savings. Finally, we present a “placebo test” in the Supplementary Appendix, table S.5, in which we examine the effect of vocational training on financial outcomes, and of financial literacy/capacity-building on employment. Estimates of the 20. As above, our propensity score is estimated with a logit regression of participation in the specific SEWA module on age, education, marital status, husband’s age, husband’s education, family size and the number of migrants in the household. Desai and Joshi 521 village-level treatment ( panel A) find that vocational training has an effect on the regularity of savings among women who reside in the village. This is to be ex- pected, given that the principal inducement to save likely comes from earning a regular wage. Alternatively we also find that financial-literacy training increases non-farm employment among women in these villages, but has no effect on other employment indicators. Note that these effects are without regard to SEWA membership in the village. By contrast, we find no significant placebo effects of individual participation in SEWA modules ( panel B), suggesting that confidence in the mechanism test should be greater for individual participation in SEWA’s programs than for SEWA’s presence in any given village.21 Taken together, this preliminary evidence suggests that the information- provision and training functions played by SHGs were among the channels oper- ating in SEWA villages by which SEWA’s interventions improved employment outcomes and encourage women to participate in the formal financial system. V. C O N C L U S I O N S Large-scale antipoverty strategies have increasingly incorporated small-scale membership organizations in project design as elements of both “pro-poor” em- powerment and as institutional platforms from which local accountability may be demanded. Evidence of the impact of these organizations outside of microfi- nance activities, however, remains scarce. We explore whether collective action can be promoted in communities through the establishment of self-help groups (SHGs), an archetypal village-based membership organization that has plays a critical role in India’s “rural-livelihoods” approach to poverty alleviation. In 2007, the Self-Employed Women’s Association (SEWA) piloted an “integrated rural livelihoods” program in Dungarpur district, Rajasthan, where villages were randomly assigned to treatment or control groups. We find that women who live in villages with SEWA programs or who are members of SEWA’s village-level SHGs report greater participation in group programs, increased control over do- mestic decision-making, greater awareness of where to express grievances about public-services ( particularly drinking water), a willingness to take action on grievances in the case of drinking water, and finally, an increase in satisfaction with the state of these services. We see some evidence that SEWA’s intervention benefited women who were landless at the start of the program more than landholding women. Additional work is needed to uncover the precise mechanisms in operation and their longer- term impact, but we also see evidence that information provision, through SEWA’s specialized vocational and financial capacity-building modules, helped women with respect to non-farm employment and savings accumulation. 21. We cannot discount, for example, the possibility that the placement of modules across villages was not random nor that the mechanisms that influence the behavior of village residents regardless of SHG membership may encompass more than vocational training or financial capacity-building. 522 THE WORLD BANK ECONOMIC REVIEW Donors are investing heavily in developing institutional arrangements to enhance the access of poor, rural households to public services and to improve in local governance by giving the poor, women, and other vulnerable groups greater representation in village-level government. In the absence of effective state institu- tions, NGOs are often seen as policy innovators, as facilitators of critical informa- tion regarding public services, and mechanisms for alternative service delivery. Our evaluation suggests that NGOs can play critical roles in linking unorganized and marginalized populations to state-led antipoverty efforts. SEWA’s main effect appears to be to facilitate the organization of communities, provide them with in- formation, motivate greater intra-group cooperation, and lower the costs of partic- ipating in collective decision-making. It may be that these “indirect,” behavioral effects on program participants outweigh the direct effects on income, consump- tion, and employment, at least in the short run. REFERENCES Abadie, A., and G.W. Imbens. 2006. “Large Sample Properties of Matching Estimators for Average Treatment Effects.” Econometrica 74(1):235 –67. Agnihotri, I., and V. Mazumdar. 1995. “Changing Terms of Political Discourse: Women’s Movement in India, 1970s-1990s.” Economic and Political Weekly 30(22):1869 –78. Alatas, V., et al. 2013. “Does Elite Capture Matter? Local Elites and Targeted Welfare Programs in Indonesia.” NBER Working Paper 18798. National Bureau of Economic Research, Cambridge, MA. Ambagudia, J. 2011. “Scheduled Tribes and the Politics of Inclusion in India.” Asian Social Work and Policy Review 5(1):33– 43. Arcand, J.L., and M. Fafchamps. 2011. “Matching in Community-Based Organizations.” Journal of Development Economics 98(2):203–19. Baland, J.-M., R. Somanathan, and L. Vandewalle. 2008. “Microfinance Lifespans: A Study of Attrition and Exclusion in Self-Help Groups in India.” India Policy Forum 4(1):159–210. ———. 2011. “Socially Disadvantaged Groups and Microfinance in India.” CRED Working Paper 1117. Centre for Research in the Economics of Development, University of Namur, Belgium. Banerjee, A., and R. Somanathan. 2007. “The Political Economy of Public Goods: Some Evidence from India.” Journal of Development Economics 82(2):287– 314. Bardhan, P.K. 2005. Scarcity, Conflicts, and Cooperation: Essays in the Political and Institutional Economics of Development. Cambridge, MA: MIT Press. Bardhan, P., and D. Mookerjee. 2006. “Pro-Poor Targeting and Accountability of Local Governments in West Bengal.” Journal of Development Economics 79(2):303– 27. Basu, P. 2006. Improving Access to Finance for India’s Rural Poor. Washington, DC: World Bank. Becker, S., and M. Caliendo. 2007. “Mhbounds-Sensitivity Analysis for Average Treatment Effects.” IZA Discussion Paper 2542. Forschungsinstitut zur Zukunft der Arbeit, Bonn, Germany. Berger, M., V.H. Delancey, and A. Mellencamp. 1984. Bridging the Gender Gap in Agricultural Extension. Washington, DC: International Center for Research on Women. Bernard, T., and D.J. Spielman. 2009. “Reaching the Rural Poor through Rural Producer Organizations? A Study of Agricultural Marketing Cooperatives in Ethiopia.” Food Policy 34(1):60– 69. Bhatt, E.R. 2006. We Are Poor but So Many: The Story of Self-Employed Women in India. Oxford: Oxford University Press. Desai and Joshi 523 Caliendo, M., and S. Kopeinig. 2008. “Some Practical Guidance for the Implementation of Propensity Score Matching.” Journal of Economic Surveys 22(1):31–72. Casini, P., and L. Vandewalle. 2011. “Public Good Provision in Indian Rural Areas: The Returns to Collective Action by Self-Help Groups.” CRED Working Paper 1119. Centre for Research in the Economics of Development, University of Namur, Belgium. Chakrabarti, R., and S. Ravi. 2011. “At the Crossroads: Microfinance in India.” Money & Finance:125– 48. Chattopadhyay, R., and E. Duflo. 2004. “Women as Policy Makers: Evidence from a Randomized Policy Experiment in India.” Econometrica 72(5):1409 –43. Chen, M. 1991. Coping with Seasonality and Drought. New Delhi: Sage Publications. Chen, M., R. Jhabvala, R. Kanbur, and C. Richards. 2007. Membership-Based Organizations of the Poor. Abingdon: Routledge. Chhibber, P.K. 2001. Democracy without Associations: Transformation of the Party System and Social Cleavages in India. Ann Arbor: Univ of Michigan Press. Danida 2002. Evaluation/Impact Study of Four Training Projects for Farm Women in India. Copenhagen, Denmark: Udenrigsministeriet. Datta, R. 2000. “On Their Own: Development Strategies of the Self-Employed Womens Association (SEWA) in India.” Development 43(4):51–55. Deininger, K., and Y. Liu. 2009. “Longer-Term Economic Impacts of Self-Help Groups in India.” Policy Research Working Paper 4886. World Bank, Washington, DC. Deshmukh-Ranadive, J. 2004. “Women’s Self-Help Groups in Andhra Pradesh: Participatory Poverty Alleviation in Action.” Presented at the Conference on Scaling Up Poverty Reduction, Shanghai, May 2004. DiPrete, T.A., and M. Gangl. 2004. “Assessing Bias in the Estimation of Causal Effects: Rosenbaum Bounds on Matching Estimators and Instrumental Variables Estimation with Imperfect Instruments.” Sociological Methodology 34(1):271 – 310. Dyson, T., and M. Moore. 1983. “On Kinship Structure, Female Autonomy, and Demographic Behavior in India.” Population and Development Review 9(1):35– 60. Feder, G., A. Willett, and W. Zijp. 2001. “Agricultural Extension: Generic Challenges and the Ingredients for Solutions.” Knowledge Generation & Technical Change 19:313–53. Feigenberg, B., E. Field, and R. Pande. 2013. “The Economic Returns to Social Interaction: Experimental Evidence from Microfinance.” Review of Economic Studies, forthcoming. Galab, S., and N.C. Rao. 2003. “Women’s Self-Help Groups, Poverty Alleviation and Empowerment.” Economic and Political Weekly 38(12/13):1274–83. Green, D.P., S.E. Ha, and J.G. Bullock. 2010. “Enough Already About “Black Box” Experiments: Studying Mediation Is More Difficult Than Most Scholars Suppose” The Annals of the American Academy of Political and Social Science 628(1):200– 8. Gugerty, M.K., and M. Kremer. 2008. “Outside Funding and the Dynamics of Participation in Community Associations.” American Journal of Political Science 52(3):585– 602. Iacus, S.M., G. King, and G. Porro. 2012. “Causal Inference without Balance Checking: Coarsened Exact Matching.” Political Analysis 20(1):1– 24. Imai, K., L. Keele, D. Tingley, and T. Yamamoto. 2011. “Unpacking the Black Box of Causality: Learning About Causal Mechanisms from Experimental and Observational Studies.” American Political Science Review 105(4):765– 89. International Institute for Population Sciences. 2008. District Level Household Survey, Third Round. Mumbai, India: IIPS. Jeffrey, P., and R. Jeffrey. 1996. Don’t Marry Me to a Plowman!: Women’s Everyday Lives in Rural North India. Boulder, CO: Westview Press. Joshi, D. 2011. “Caste, Gender and the Rhetoric of Reform in India’s Drinking Water Sector.” Economic and Political Weekly 46(18):57– 63. 524 THE WORLD BANK ECONOMIC REVIEW King, E.M., and A.D. Mason. 2001. Engendering Development: Through Gender Equality in Rights, Resources, and Voice. Washington, DC: World Bank. Macklin, M. 1992. “Agricultural Extension in India.” Technical Paper 190. World Bank, Washington, DC. Mansuri, G., and V. Rao. 2012. “Localizing Development: Does Participation Work?” World Bank Policy Research Report. World Bank, Washington, DC. Munshi, K. 2007. “Information Networks in Dynamic Agrarian Economies.” In T.P. Schultz, and A.S. John, eds., Handbook of Development Economics, vol. 4. Amsterdam, The Netherlands: Elsevier. Narayan, D. et al. 2000. Voices of the Poor. Can Anyone Hear Us?. Oxford: Oxford University Press. Olken, B.A. 2010. “Direct Democracy and Local Public Goods: Evidence from a Field Experiment in Indonesia.” American Political Science Review 104(2):243– 67. Ostrom, E., and T.K. Ahn. 2009. “The Meaning of Social Capital and Its Link to Collective Action.” In G.T. Svendsen, and G.L. Svendsen, eds. Handbook of Social Capital. Northampton, MA: Edward Elgar. Planning Commission, Government of India. 2011. Report of Working Group on National Rural Livelihoods Mission (NRLM). New Delhi: Government of India. Raabe, K. 2008. “Reforming the Agricultural Extension System in India: What Do We Know About What Works Where and Why?” Discussion Paper 00777. International Food Policy Institute, Washington, DC. Reddy, C.S., and S. Manak. 2005. “Self-Help Groups: A Keystone of Microfinance in India-Women Empowerment and Social Security.” Hyderabad, India: Andhra Pradesh Mahila Abhivruddhi Society (APMAS). Rosenbaum, P.R. 2002. Design of Observational Studies. New York: Springer. Sen, A. 2001. “The Many Faces of Gender Inequality.” New Republic:35–39. Sen, A., and J. Dreze. 2002. India: Development and Participation. New Delhi: Oxford University Press.