The World Bank Economic Review, 31(1), 2017, 97–128 doi: 10.1093/wber/lhv041 Article Public Good Provision in Indian Rural Areas: The Returns to Collective Action by Microfinance Groups Paolo Casini, Lore Vandewalle, and Zaki Wahhaj Abstract Self-help groups (SHGs) are the most common form of microfinance in India. We provide evidence that SHGs, composed of women only, undertake collective actions for the provision of public goods within village com- munities. Using a theoretical model, we show that an elected official, whose aim is to maximize re-election chances, exerts higher effort in providing public goods when private citizens undertake collective action and coordinate their voluntary contributions towards the same goods. This effect occurs although government and private contributions are assumed to be substitutes in the technology of providing public goods. Using first- hand data on SHGs in India, we test the prediction of the model and show that, in response to collective action by SHGs, local authorities tackle a larger variety of public issues, and are more likely to tackle issues of interest to SHGs. Our findings highlight how the social behavior of SHGs can influence the governance of rural Indian communities. JEL classification: D72, H41, G21, O16 Key words: SHGs, Public Goods Paolo Casini is a researcher at the LICOS Center for Institutions and Economic Performance, KU Leuven, and an economic analyst at the European Commission, DG Internal Market, Industry, Entrepreneurship and SMEs; his email address is paolo.casini@ kuleuven.be. Lore Vandewalle (corresponding author) is an assistant professor at the Graduate Institute of International and Development Studies, Geneva, and a nonresident senior researcher at the University of Oslo; her email address is lore.vandewalle@ graduateinstitute.ch. Zaki Wahhaj is a senior lecturer at the University of Kent; his email address is Z.Wahhaj@kent.ac.uk. This work was supported by Centre de Recherche en Economie du De ´ veloppement (CRED) at the University of Namur; Fonds de la Recherche Scientifique (FNRS) in Belgium; KU Leuven; and the European Research Council [AdG-230290-SSD]. We are grateful to Jean-Marie Baland, Guilhem Cassan, Eliana La Ferrara, Dilip Mookherjee, Jean-Philippe Platteau, Rohini Somanathan, Vincent Somville, and William Pariente ´ for helpful discussions and suggestions and seminar participants at Bocconi University, the BREAD summer school, CES, CRED, Chr. Michelsen Institute, the EEA conference, the FEDEA gender conference, Georg-August- at Go Universit€ ¨ ttingen, the Graduate Institute of International and Development Studies, LICOS, LSE, the NEUDC conference, Universidad de Chile, and the University of Oslo for useful comments. We also thank the PRADAN teams and especially Narendranath for discussions and for support in facilitating data collection and Sanjay Prasad and Amit Kumar for excellent research assistance. This paper is produced as part of the project “Actors, Markets, and Institutions in Developing Countries: A micro- empirical approach” (AMID), a Marie Curie Initial Training Network (ITN) funded by the European Commission under its Seventh Framework Programme. Contract Number 214705 PITN-GA-2008-214705. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. C The Author 2015. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. V All rights reserved. For permissions, please e-mail: journals.permissions@oup.com. 98 Casini, Vandewalle, and Wahhaj Self-help groups (SHGs) are the most common form of microfinance in India. By March 2012, about 103 million households had a member in an SHG (Nabard 2013). Their primary aim is to help the poor to save and borrow. At the outset, SHG members pool their savings to create a common fund and give out small loans to one another. At a later stage, SHGs are able to open savings accounts with commercial banks and apply for loans. Because of some features of their functioning, such as the high frequency of meetings and the mutual trust necessary for their stability, SHGs can impact the lives of their members beyond the mere financial sphere. We study the long-term, nonfinancial impact of an SHG program that focuses on women only. Using first- hand data collected in the Indian state of Odisha, we document how collective actions undertaken by SHGs impact the variety of public goods that the Gram Panchayat—which is the lowest official authority—deals with. In our research area, each Gram Panchayat is divided into several wards. A representative, known as ward member, is elected in each of those wards. He is the official spokesperson of the villagers: his main responsibility is communicating the ward’s problems and needs to the officials in charge at a higher administra- tive level, that is, to officials who are senior to him. The ward member is the only official appointment for these duties. Yet, we find evidence that SHG members undertake collective actions that de facto overtake or comple- ment the work of the ward member. They visit higher officials as well, or intervene directly to solicit a solution for a variety of problems affecting their ward. We show that these collective actions impact the ward member’s choices: it becomes more likely that he starts tackling public goods that SHGs consider important. We believe the contribution of our findings is twofold. First, to the best of our knowledge, ours is the first paper assessing the long-term effect of microfinance groups on social outcomes (up to thirteen years after their creation). Our data suggest that considering such a long time span is crucial since, in our sam- ple, an average SHG undertakes collective actions for the first time after three years of weekly meetings only. Second, our results put forward how microfinance can also be used as a political economy mecha- nism to improve the lives of the poor. In September 2010, we interviewed all SHGs created by the NGO PRADAN (Professional Assistance for Development Action) in the Mayurbhanj and Keonjhar districts of rural Odisha. PRADAN’s SHG program aims at providing financial intermediation and does not have an explicit socio-political agenda.1 We asked the SHG members what kind of problems they had faced in their ward and whether they had tried to solve them. Some groups merely discussed problems during their meetings, but others undertook collective actions to tackle them. We also interviewed the ward members elected in the past 20 years (1992, 1997, 2002, and 2007). Their main focus is on the major responsibilities of the Gram Panchayat: public infrastructure and wel- fare schemes.2 But we provide evidence that the range of problems that ward members take care of is also influenced by observing the collective actions undertaken by SHGs. To explain these observations, we propose a simple theoretical model in which local public goods may be provided through voluntary contributions by community members and effort by an elected offi- cial. The official, whose effort is unobservable, is incentivized by the fact that his chance of future re- election is increasing in the present welfare of community members. We show that if the community members undertake collective action—more precisely, can commit to making an efficient level of contri- bution rather than play the Nash equilibrium—then the official also provides a higher level of effort in equilibrium. 1 In contrast, the Grameen Bank in Bangladesh has a clear social development agenda. Members are required to obey 16 Decisions, which have a clear social connotation. For example, Decision 7 states: “We shall educate our children and en- sure that they can earn to pay for their education.” 2 Welfare schemes are governmental programs aiming at helping disadvantaged parts of the population. Among other pro- grams, it includes Mahatma Gandhi National Rural Employment Guarantee Act (MGNREGA) and Indira Gandhi National Old Age Pension Scheme (IGNOAPS). The World Bank Economic Review 99 Thus, although the contributions by the official and the community members are substitutes in the technology for providing public goods, they behave as strategic complements in equilibrium. The simple intuition behind the result is that collective action by the community increases the marginal effect of the official’s effort on their welfare which, in turn, implies that the official’s optimal level of effort is higher under collective action. In the context of the SHGs set up by PRADAN, the theoretical model has the following implication: to the extent that setting up SHGs made it easier for women in the PRADAN villages to undertake col- lective action regarding public issues that concerned them, the ward member should increase his efforts in addressing the same issues. We test the prediction. The information we collected on ward members allows us to construct ward- level panel data over four elections. To identify the impact of collective actions undertaken by SHGs, we use an instrumental variables approach that exploits the variation in the timing of the creation of SHGs. To identify a causal effect, the creation of SHGs should be uncorrelated with determinants of future pub- lic good provision. After including ward fixed effects, the assumption underlying our approach is that the creation of SHGs is not correlated with pre-existing differences across wards. We assess the plausibil- ity of this assumption using the 1991 Census data and our information on the period before the first SHGs were created. We show that the first villages in which PRADAN created SHGs have socioeco- nomic characteristics comparable to those in which SHGs were created later. Our empirical results confirm the prediction of our model. We find that ward members deal with, on average, 1.5 extra public goods after SHGs start undertaking actions. In particular, they are 29% more likely to deal with alcohol issues, 35% more with forestry issues, and 31% more with school problems, all of which are issues in which SHGs are particularly interested. Our work is related to several strands of the literature. A number of studies look at the social implica- tions of microfinance programs. Feigenberg et al. (2013) provide evidence that the frequency of meetings is a determinant of long-run increases in social interaction. Chowdhury et al. (2004) discuss why, in eval- uating the impact of microfinance programs, nonclient beneficiaries ought to be considered. In India, SHG membership makes socially disadvantaged women more likely to engage in community affairs (Desai and Joshi 2014) and has a positive impact on female empowerment (Deininger and Liu 2009; Datta 2015; Khanna et al. 2015). Four recent papers assess the impact on female decision-making power of microfinance programs that provide loans through group lending. The results are more diverse: Banerjee et al. (2015), Cre ´ pon et al. (2015), and Tarozzi et al. (2015) do not find significant effects on women’s empowerment in India, Morocco, and Ethiopia, respectively.3 But Angelucci et al. (2015) document a strengthening of women’s decision-making power in households in Mexico. These studies focus on the short-term effects, that is, within 24 months after providing access to microfinance.4 In con- trast with these studies, we focus on the long-term effects on a different social outcome. We describe how the social behavior of SHGs can influence the governance of rural Indian communities. Several papers provide evidence that men and women have diverging preferences for some public poli- cies (Lott and Kenny 1999; Edlund and Pande 2002; Edlund et al. 2005). Still, in many countries, wom- en’s preferences hardly find their way into the policy-making agendas. Some governments have imposed political reservations to alter policy choices in favor of women. Chattopadhyay and Duflo (2004) have shown the significant effect of these reforms in India. We add to this literature by exploring an 3 In Morocco and Ethiopia most borrowers are men. For this reason—as noted by the authors in both papers—the limited effect on the empowerment of women is not surprising. 4 Banerjee et al. (2015) resurveyed the households after three years. However, at that time, the control group had access to microfinance as well. The control group had larger loans and was treated for a longer period, but these circumstances “may limit power to detect differences in the social outcomes at the community level” (Banerjee et al. 2015, page 50). 100 Casini, Vandewalle, and Wahhaj alternative channel through which the preferences of women can sway political decisions without resort to overt policy controls. Our theoretical model is related to a long literature, which explores the relation between government provision of, and voluntary contributions towards, public goods going back to Warr (1982), Roberts (1984), and Bergstrom et al. (1986). An important finding in this literature is that increases in govern- ment provision can cause crowding out of private contributions. Our theoretical model shows, in a simi- lar setup, that collective action by private citizens in contributing to a public good can incentivise an elected official to increase effort in providing the same. Thus, private provision and government provi- sion of public goods can be complementary. The remainder of the paper is organized as follows. In section I we describe our data set, the ward structure, and the collective actions undertaken. In section II, we develop a theoretical model on public goods provision by elected officials and citizens. Section III discusses the implications of the theoretical model in the context of women’s SHGs in Indian villages. The empirical analysis, including a test of the theoretical prediction, is carried out in section IV. Section V concludes. I. Background Information Data collection was assisted by our partner, the NGO PRADAN. Its main mission is the improvement of forest-based livelihoods and natural resource management of socioeconomically disadvantaged people. It pioneered the creation of SHGs (consisting entirely of women) as an instrument to achieve its mission (PRADAN 2005). In 2006, all of the PRADAN SHGs created in the Mayurbhanj and Keonjhar districts of Odisha were surveyed, independent of whether the groups were still actively meeting or not. The data set contains of 532 SHGs and 8,589 women who, at some point, belonged to these groups (Baland et al. 2008). In the autumn of 2010, we complemented this data set in two ways. First, we revisited these SHGs to gather information on the collective actions that they undertook. Second, we performed a ward survey to col- lect data on the characteristics and activities of ward members. As PRADAN started working in Odisha in 1998, and as we needed information dating back to the period before the creation of the first SHG, we interviewed the ward members elected in 1992, 1997, 2002, and 2007.5 We asked them to recall the types of issues they dealt with, that is, the type of issues for which they visited a higher official or inter- vened directly in the ward.6 In total, we gathered information on 425 SHGs, and we have complete information on 441 ward members, covering 108 villages and 141 wards.7 Wards are in most cases smaller than villages. On aver- age, there are 1.3 wards per village. Villages and wards coincide for 75% of the villages in our data set. Wards are larger than villages for eight small villages only. These eight villages belong to four different wards. 5 Elections take place every five years. Ward members can be re-elected. 6 To avoid a recall bias—which occurs if ward members elected in 1992 remember less of their interventions than those elected in 2007—we gathered information as follows: first, we conducted focus group discussions in a subset of wards to list (as many as possible) ward problems. Based on this information, we defined the six broad categories that are de- scribed in the supplemental appendix S1 (for example, problems related to a well or a road are both categorized as “pub- lic infrastructure”). Second, when interviewing the ward members, we first asked the type of issues they dealt with as an open question and then we proposed the categories they had not mentioned. 7 We were not able to resurvey 21 villages (62 SHGs) because of social tensions created by a private mining firm (the roads to those villages were blocked). Another 45 groups that no longer meet refused to be resurveyed. 34 of those groups are based in wards where other SHGs are still meeting. Thus, for those we obtained all of the information needed for our analysis. The World Bank Economic Review 101 Ward Structure In rural India, the lowest official authority is the Gram Panchayat. It is composed of 5–15 contiguous vil- lages. The 73rd Amendment Act 1992 of the Constitution of India empowers the State Legislature “to endow the Panchayats with the power and authority necessary to prepare the plans and implement the schemes for economic development and social justice.” The main responsibilities passed onto the Gram Panchayat are managing public infrastructure and identifying villagers who are entitled to welfare schemes (Xaxa 2010). Each Gram Panchayat is divided into wards and is governed by one Sarpanch, a Naib-Sarpanch, and several ward members. One ward member (hereafter WM) is elected in each ward. WMs have the right to access the records of the Gram Panchayat, to question any official about its administration, and to inspect the actions it undertakes. Their main responsibility is informing government officials in charge about the ward’s problems and needs. Apart from the Sarpanch, they can also approach higher authorities at the block or district level. WMs do not control financial means. Therefore, they cannot intervene without the involvement of higher authorities, unless the intervention is costless (Xaxa 2010). In what follows, we use the general label higher official to indicate any government official who is at a higher administrative level than the WM and who is endowed with the financial means and power to solve a particular issue. Although SHGs are created for financial intermediation, we find evidence that members participate in col- lective actions that de facto overtake or complement the work of the WM. They either communicate their concerns about a public issue to a higher official or they intervene directly in their ward. SHGs undertake col- lective actions as a group, with eleven out of an average fifteen members actively involved in the first action.8 Apart from SHGs, there are two other bodies active in solving ward problems: some single individuals and some other groups of villagers. We label the latter as Other Groups. They consist of villagers who meet on average once a month for a specific, nonfinancial reason. Most of them are forest committees (69.2%), that is, groups of people dedicating time and resources to avoid forest exploitation.9 Some of these committees are created by officials of the forest department (35.4%). 26.6% of Other Groups are known as village help clubs. They are formed by young villagers and deal with a wide range of issues related to social violence and public infrastructure. Finally, there are groups formed for cultural activities (3.2%) and farming issues (1.0%). Remarkably, of the ninety-four Other Groups in our data set, only one consists entirely of women. We also surveyed a random sample of Individuals to obtain information on villagers who joined nei- ther an SHG nor an Other Group. Table 1 shows the characteristics of WMs, members of SHGs, members of Other Groups, and Individuals who dealt with ward problems at least once in the columns (1) to (5). The Other Group members differ from SHG members in several respects: they are mainly men, are more educated, and own about 1 acre more of land. SHG members also differ from WMs and Individuals: the latter are bet- ter educated and own more land. Thirty-one percent of the WMs are women, but 78% of them are elected thanks to reservations for women.10 Remarkably, there are few women among the Individuals who dealt with ward problems (2.3%). For this reason, we group men and women in column (5). 8 A first action typically concerns public infrastructure (33.6%), forestry issues (26.1%), and alcohol problems (21.9%). See supplemental appendix S1 for a description of the ward level problems. 9 As most villages are located close to the forest, households depend on it for cooking and as a source of income (e.g., an important source of income is making leaf plates). An increasing population adds more pressure on the forest. To prevent excessive deforestation, villagers formed voluntary forest committees. Later, the forest department started supporting ex- isting committees and created new ones. They provide training and supplies and introduce new forms of sustainable ex- ploitation of the forest. 10 During the period under study, it was imposed that one-third of the seats had to be reserved for women. The reservation of seats is allotted by rotation among the different wards (Xaxa, 2010). 102 Casini, Vandewalle, and Wahhaj Table 1. Characteristics of WMs, Members of SHGs, Members of Other Groups and Individuals Bodies who dealt with ward problems Bodies who never dealt with ward problems Ward members Individuals Female Male SHGs Other Groups Individuals SHGs Other Groups Female Male (1) (2) (3) (4) (5) (6) (7) (8) (9) Number of groups — — 388 91 — 37 3 — — Number of members 138 303 6,299 734 132 567 23 79 765 Women (%) — — 100.0 13.4 2.3 100.0 17.4 — — Education level (years) 5.8 7.4 2.6 7.6 9.0 1.4 7.2 3.3 4.8 Can read and write (%) 76.1 89.4 30.5 83.0 96.2 16.4 87.0 36.7 57.9 Land (acres) 2.1 2.5 1.7 2.6 3.3 1.7 2.6 2.6 1.8 Number of children 2.7 3.0 2.6 2.6 2.8 2.9 2.1 1.9 2.6 Age (years) 40.1 46.9 35.5 41.0 47.7 35.4 34.8 37.0 42.4 Caste category: ST (%) 65.9 74.9 62.9 67.3 64.4 82.5 56.5 77.2 66.5 Caste category: SC (%) 14.5 4.6 9.3 4.3 4.5 1.4 17.4 1.3 6.7 Caste category: OBC (%) 19.6 19.8 26.5 27.9 28.8 15.5 26.1 21.5 26.7 Caste category: FC (%) 0.0 0.7 1.3 0.5 2.3 0.6 0.0 0.0 0.1 Source: Authors’ analysis based on data described in the text. Columns (6) to (9) show the characteristics of bodies who never dealt with ward problems. Both SHG members and male Individuals are less educated than their counterparts who dealt with ward problems. Female Individuals are slightly more educated and own more land than SHG members, which suggests that SHGs are formed by the more disadvantaged part of the female population. Ward Problems Table 2 shows, for each issue, the percentage of WMs and SHGs that tried to solve a problem either by visiting a higher official or by intervening directly.11 A brief explanation of the different problems is given in supplemental appendix S1 (available at http://wber.oxfordjournals.org/). WMs are more likely to deal with public infrastructure and welfare schemes than SHGs. This is not surprising since these are the main responsibilities of the Gram Panchayat. Compared to WMS, SHGs are more likely to deal with alcohol and school problems and forest-related issues. The focus on alcohol issues is in line with the findings of a wide literature on the topic.12 Some SHGs visited higher officials to request the suspension of alcohol licenses. Others intervened directly by organizing anti-alcohol cam- paigns or by trying to dissuade households from producing alcohol. This is quite interesting since anec- dotal evidence suggests that women consider alcohol consumption as a prerogative of men; hence, they 11 SHG members also provided mutual help in about 9% of the SHGs. They provided assistance when a member needed medical help or when a funeral had to be organized. They also intervened when a conflict took place in a member’s household. WMs do not get involved in these activities. Therefore, we do not consider those in the remainder of the paper. 12 The literature shows three main results. First, households realize that alcohol consumption reduces the budget available for primary expenses (Mishra 1999). According to Banerjee and Duflo (2007), alcohol ranks among the first goods that poor families would like to eliminate from their consumption bundle if they had more self-control. Secondly, in India, men are 9.7 times more likely than women to regularly consume alcohol (Neufeld et al. 2005). Finally, there is strong evidence that alcoholism triggers violence against women. Rao (1997) and Panda and Agarwal (2005) show that the risk of wife abuse increases significantly with alcohol consumption. Babu and Kar (2010) find that domestic violence is pervasive in Eastern India, which includes Odisha. They show that alcohol consumption is an important risk factor for physical, psychological, and sexual violence against women. The World Bank Economic Review 103 Table 2. Aim of Collective Actions of WMs and SHGs % that dealt with the issue WMs SHGs (1) (2) Public infrastructure 81.9 53.7 Welfare schemes 65.1 16.5 Alcohol problems 12.9 59.8 School problems 12.5 16.5 Forest issues 33.8 55.1 Other 4.8 3.5 Average number of different issues (conditional on at least one) 2.2 2.3 Number of observations 441 425 Source: Authors’ analysis based on data described in the text. rarely undertake legal actions, even in case of domestic violence or abuse. Indeed, we could not find any woman undertaking an action alone. School problems are mainly related to the provision of free midday meals, sanitation, and teacher quality. The interest in these issues is in line with the common finding that women generally spend more time and resources on family welfare than men.13 Furthermore, in our sur- vey, SHGs are responsible for providing midday meals at schools in 23.2% of the villages. The focus of SHGs on forestry issues is not surprising either, as the livelihood of many households depends on for- estry. Moreover, 29.7% of the SHGs received training from PRADAN to improve their forest-based sources of income. The Evolution of WMs’ Activities We are interested in studying whether the collective actions undertaken by SHGs impact the choices of WMs. Unfortunately, we cannot measure changes in the productivity of WMs, as we only know the SHGs’ perception of how successful he was for each intervention.14 But we do know whether he tackled an issue or not, and this is what we exploit. Table 3 shows the percentage of WMs dealing with each issue in their ward. WMs are classified depending on whether their mandate finished before the first SHG was created in the ward (column (1)) or after (column (2)). These simple descriptive statistics docu- ment a sharp increase for almost all of the problems. This analysis can be refined by taking into account the fact that SHGs do not undertake collective actions from the very start of their existence. SHGs are created for financial intermediation and not for public good provision. Furthermore, PRADAN’s SHG program has no explicit socio-political agenda. Indeed, on average, an SHG undertakes the first collective action only after about three years of weekly meetings.15 If the activities of the WMs are influenced by the collective actions of SHGs, we might observe a change only when SHGs become active. In other words, the mere creation of an SHG might 13 See Anderson and Baland (2002) and Duflo (2012). 14 See table 10 for a qualitative analysis of the WMs’ productivity. 15 Mishra (1999) describes the process leading SHGs to different forms of cooperation as a three-stage evolution over time. In the first stage, group members have a minimum level of awareness and need to shed their prejudices. In the next stage, groups experience pressure from both outside and inside that helps the emergence of a group leader and shapes in- ternal norms. Groups reach the third stage when they agree on their objective. They start functioning as a team, recog- nize common problems (both economic and social), and undertake collective actions. Following this reasoning, we believe that groups deal with elaborated nonfinancial issues only when they reach a minimum level of financial stability. 104 Casini, Vandewalle, and Wahhaj Table 3. Problems WMs Dealt with, before, and after the Creation of SHGs % of WMs dealing with an issue in their ward before the first once SHGs are present SHG was created All before SHGs undertake actions since SHGs undertake actions (1) (2) (3) (4) Public infrastructure 75.2 84.4** 78.3 86.5*** Welfare schemes 37.2 75.6*** 63.9*** 79.8*** Alcohol problems 2.5 16.9*** 1.2 22.4*** School problems 5.0 15.3*** 2.4 19.8*** Forest issues 22.3 38.1*** 16.9 45.6*** Other 3.3 5.3 4.8 5.5 Average number of different issues 1.7 2.4*** 1.8 2.6*** (conditional on at least one) Number of observations 121 320 83 237 Significance of the difference relative to column (1): *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent. Source: Authors’ analysis based on data described in the text. not matter. For this reason, we split the time frame after the creation of SHGs. Table 3 reports the per- centage of WMs who deal with a problem depending on whether his mandate finishes after the creation of the first SHG but before an SHG undertakes collective actions in his ward (column (3)) or after the first SHG does so (column (4)). For most issues, we observe an increment after the creation of the first SHG in the ward, but the increase is statistically significant only after the first SHG undertakes action. Section II provides a simple theoretical model that formalizes why and if WMs have incentives to deal with a different set of public goods as a response to the collective actions of SHGs. II. A Model of Public Goods Provision and Collective Action Imagine a community consisting of n individuals. An individual i derives utility from a private good denoted by xi and a set of public goods y1,..,yk enjoyed by all community members. We represent the utility of an individual i as follows: Y k Ui ðxi ; y1 ; ::; yk Þ ¼ ðxi Þai ð yj Þ b i ; j¼1 where ai ; bi 2 ð0; 1Þ for i ¼ 1; ::; n. The private good is generated according to the production function f ðlxi Þ ¼ xlxi , where lxi is the amount of time devoted to private goods production. (Here, x can represent productivity in home pro- duction or the wage that an individual receives from providing labor in the labor market, which is subse- quently used to purchase private goods). Public good g is generated according to the production function hðlg Þ ¼ hg lg , where lg is the total labor contribution by all community members in the production of pub- lic good g. (Here, hg is an efficiency parameter in public goods production). Each individual has an endowment of 1 unit of labor time that can be used either for the production of private goods or for the provision of public goods. This setup is similar to that of Bergstrom et al. (1986), who investigate how voluntary contributions to a public good in a group of individuals is affected by the distribution of wealth and by the centralised provision of the public good financed by taxes. In the following, we investigate how the efforts of an elected official—who gains from raising the utility of his constituents—in the provision of the public The World Bank Economic Review 105 good is affected by the presence of collective action within the community. We show that, paradoxically, the elected official makes a greater contribution to the provision of public goods when the community members coordinate their efforts in contributing to the same. Subsequently, we look at the case in which different subgroups within the community care about dif- ferent public goods and show that, when one of the groups undertakes collective action with regard to its preferred public good, the elected official shifts his efforts towards the same. The Case of Homogeneous Preferences The idea underlying our theoretical argument can be illustrated with a simple setup where there is a sin- gle public good, and all members of the community have the same preferences regarding private and public goods. We write the utility of person i as Uðxi ; yÞ ¼ ðxi Þa ðyÞb : The community can elect an official whose job will be to address the public good issues faced by the community members. We assume that if he contributes lm units of labor to the provision of the public good, this is equivalent to mlm units of labor contribution by any other community member, where m>0 (His level of effectiveness may be different from that of other community members because he acts in an official capacity). Therefore, we write the total contribution to the public good as Xn L ¼ mlm þ l ; j¼1 yj where lm is the amount of labor provided by the elected official and lyj the contribution by person j. The elected official receives a remuneration xm for his official work. His labor contribution cannot be contracted upon. However, his chances of re-election will depend on how well he has served his con- stituents in the past. His tenure lasts for one period, following which he has to contest elections. His 0 probability of being successful at the elections is given by pð1n UÞ, where U is a vector describing the util- ity achieved by each community member in the preceding period and 1n is a vector of ones of dimension n. We assume that the probability function is linear over the relevant range of utilities to ensure that the official’s incentives are unaffected by the level of utility achieved by community members: 0 00 Assumption 1. p ð:Þ > 0 and p ð:Þ ¼ 0 in the interval [Umin, Umax] where Umin and Umax are, respectively, the minimum and maximum utility that community members can achieve in the game.16 The ward member’s utility in the current period is given by Um ðxm ; yÞ ¼ ðxm Þam ðyÞbm ; where xm denotes the level of consumption of his private good and am ; bm 2 ð0; 1Þ. We assume, for sim- plicity, that his utility gain from re-election is equal to a constant W. The term W may include not only the financial reward from re-election but also the utility derived from the prestige associated with the position and the added benefits of political capital.17 Before the elected official chooses his labor contribution in the provision of the public good, the com- munity decides whether they will undertake “collective action”; that is, whether they will choose their own labor contribution towards the public good collectively or individually. “Collective action” involves 16 To be precise, U min ¼ nðaxÞa ðbhÞb ðn=ðan þ bÞÞaþb and U max ¼ nðaxÞa ðbhÞb ððm þ nÞ=ða þ bÞÞaþb . 17 An alternative approach to modeling the official’s re-election incentive would be to derive his expected utility from being re-elected from a repeated game. While this alternative approach can provide additional insights, it is beyond the scope of the present paper. Our approach would be a reasonable simplification, if the official’s time horizon for utility maxi- mization extends only to the next election. 106 Casini, Vandewalle, and Wahhaj a one-time utility cost of C for each community member (this may be thought of as the cost of organiza- tion, negotiation, setting up an enforcement mechanism, etc.). Specifically, the timing of events within the game is as follows. 1. Members of the community decide whether or not they will undertake collective action. 2. The elected official chooses his labor contribution towards the public good and his own private good. 3. If the members of the community invested in collective action at stage 1, they decide upon their labor contribution collectively; otherwise they make their labor contribution individually. At each stage of the game, the individuals have knowledge of what has happened before. When the com- munity members decide upon their labor contribution “collectively,” we assume that their contributions are efficient and symmetric. When they decide upon their labor contributions individually, we assume that this results in the symmetric Nash equilibrium of the subgame. To predict the outcome in this model, we first compute, for a given contribution by the elected official, the utility level that each community member would achieve in the absence and presence of collective action. Equilibrium in the Absence of Collective Action First, we derive the equilibrium level of contributions within the community in the absence of collective action. Let us denote by lxj and lyj the labor contribution of person j to the private good and public good respectivly. Then, we can write the utility to person i as !b X n Ui ¼ ðxlxi Þa mhlm þ hlyj : j ¼1 In the absence of coordination, each individual will equate marginal utility between the private good and the public good. Thus, we obtain axðxi ÞaÀ1 ðyÞb ¼ bhðxi Þa ðyÞbÀ1 ; (1) P where xi ¼ xð1 À lyi Þ and y ¼ mhlm þ j hlyj . In the case of a symmetric Nash equilibrium where each community member makes a labor contribution of ly to the public good, we have y ¼ mhlm þ nhly and xi ¼ xð1 À ly Þ. Substituting for y and xi in (1) using these expressions, we obtain b À amlm ly ¼ (2) an þ b     mlm þ n mlm þ n ) y ¼ bh and xi ¼ ax : (3) an þ b an þ b So, the utility of each community member is given by   mlm þ n aþb Uðxi ; yÞ ¼ ðaxÞa ðbhÞb : (4) an þ b Equilibrium in the Presence of Collective Action Next, we consider the labor allocation by the community members under collective action. If they imple- ment an efficient contract, their allocation of labor should solve the following optimization problem: !b X n a X n max ðlxj ; lyj Þ kj ðxlxj Þ hlyk þ mhlm j¼1 k¼ 1 for some Pareto weights k1 ; ::; kn . Using the first-order condition, we obtain The World Bank Economic Review 107 X n ki axðxi ÞaÀ1 yb ¼ bh kj ðxj Þa ybÀ1 ; (5) j¼1 P where xi ¼ xð1 À lyj Þ and y ¼ mhlm þ h n j¼1 lyj . In the case of a symmetric contract where the Pareto weights are equal and each community member makes the same labor contribution of ly to the public good, we have y ¼ mhlm þ nhly and xi ¼ x ¼ xð1 À ly Þ. Substituting for x and y in (5) using these expressions, we obtain bn À amlm ly ¼ (6) nða þ bÞ     mlm þ n mlm þ n ) y ¼ bh and x ¼ ax : (7) aþb an þ b n So, the utility of each community member is given by  aþb mlm þ n Uðxi ; yÞ ¼ ðaxÞa ðbnhÞb : (8) an þ b n Why Collective Action Makes a Difference For a given level of contribution to the public good by the elected official, the community members are at least as well-off when they undertake collective action as compared to their utility under the symmet- ric Nash equilibrium. This is true by construction as the contract under collective action is symmetric and efficient. However, whether they are ultimately better-off depends on the labor response of the elected official to such a decision. Using (4) and (8), we can determine how an increase in labor contribution by the elected official towards the public good affects the utility of community members in the presence and absence of collec- tive action. Differentiating (4) and (8) with respect to lm, we obtain     dUi mlm þ n ðaþbÀ1Þ m ¼ ða þ bÞðaxÞa ðbhÞb (9) dlm an þ b an þ b in the absence of collective action;     dUi mlm þ n ðaþbÀ1Þ m ¼ ða þ bÞðaxÞa ðbnhÞb (10) dlm an þ bn an þ bn in the presence of collective action. Comparing the expressions on the right-hand sides of (9) and (10), we can establish the following lemma (proof shown in supplemental appendix S2). Lemma 1. The marginal utility of the elected official’s effort is higher when the community members undertake collective action, as compared to when their labor contributions constitute a Nash Equilibrium if and only if 1=n1Àb > ða þ bÞ=ðan þ bÞ: The condition in lemma 1 is satisfied for a large range of values for a, b, and n. For example, if a ¼ b ¼ 0.5, the condition is satisfied for all n > 1. 108 Casini, Vandewalle, and Wahhaj The Elected Official’s Optimal Choice Using lemma 1, we can investigate how the community’s decision about whether or not to undertake col- lective action affects the elected official’s labor choice at the second stage of the game. The official’s opti- mization problem can be written as follows: max lm Uðxm ; yÞ ¼ ðxm Þam ðyÞbm þ pð10 n UÞW subject to xm ¼ xm þ xð1 À lm Þ, y ¼ nhly þ mhlm and Ui ¼ Ueq for i ¼ 1,..,n. Here Ueq is given by (4) in the absence of collective action and by (8) in the presence of collective action; in addition, ly denotes the equilibrium labor contribution towards the public good by each com- munity member, given by (2) in the absence of collective action and by (6) in the presence of collective 0 action, and U ¼ðU1 ; U2 ; :::; Un Þ. From the first-order condition to the elected official’s optimization problem, we obtain the following, assuming an interior solution: X n dUi am xðxm Þam À1 ðyÞbm ¼ bm mhðxm Þam ðyÞbm À1 þ Wp0 ð:Þ ; (11) i¼1 dlm 0 where p ð:Þ is the derivative of the probability function pð:Þ. This equation is identical to that obtained for ordinary community members except for the last term, which derives from the fact that any effort by the elected official in improving the community’s access to the public good affects his re-election chances. We see from lemma 1 that, for reasonable parameter values, the marginal utility of the elected offi- cial’s effort, dUi =dlm (which affects his re-election chances) is higher when the members of the commun- ity undertake collective action. On the other hand, they also allocate a higher level of effort in the delivery of the public good for any given level of effort applied by the elected official, and this provides the elected official an incentive to free-ride on public good delivery. These two effects go in opposite directions, and so the net effect on the elected official’s effort is ambiguous. If the elected official does not directly care for the public good, that is, bm ¼ 0, then the first effect dominates and he provides higher effort in the delivery of the public good. If bm > 0, the first effect will dominate if W is sufficiently large; that is, if he cares enough about being re-elected and/or the benefits of holding office are sufficiently high. These results are summarized in the following proposition (proof shown in supplemental appendix S2). Proposition 1. If the conditions in lemma 1 and assumption 1 hold and the elected official has no intrinsic preferences regarding the provision of the public good, then his labor contribution towards the public good is higher under collective action. If the official cares directly about the public good, then his labor contribution towards the public good is higher under collective action if W is sufficiently large. Thus, assuming that the conditions in proposition 1 hold, the elected official provides a higher labor contribution towards the public good, if the community undertakes collective action at the first stage of the game. It follows that the community will have a higher level of utility when they undertake collective action. However, their net gain from collective action also depends on the fixed cost C. As the cost may vary across communities, some of them may find it in their interest to form collective action groups while others may not. We discuss this issue in more detail in section III. The Case of Heterogeneous Preferences Thus far, we have assumed that all members of the community have the same preferences vis-a-vis their own private consumption and the provision of the public good. Now, we consider the setting where the community is composed of distinct groups, which differ in terms of their preferences regarding public goods. We consider how the elected official’s decisions are affected when one of the groups organises The World Bank Economic Review 109 and engages in collective action to improve the provision of public goods. We show that the outcome is similar to that obtained in the preceding subsection. For ease of analysis, we assume that there are two groups and two public goods. But our main arguments apply for any number of groups and public goods. We label the two groups in the population as h and f (for “homme” and “femme”), of size nh and nf, respectively, (nh þ nf ¼ n). We label the two public goods as 1 and 2. If individual i is a member of group g 2 fh,fg, his/her preferences are given by the following utility function: Ug ðxi ; y1 ; y2 Þ ¼ ðxi Þa ðy1 Þb1g ðy2 Þb2g ; (12) where xi denotes the level of i’s private consumption and y1 and y2 denote the level of provision of public goods 1 and 2, respectively. We impose the following conditions on the preference parameters: Assumption 2. b1h > b1f and b2h < b2f Assumption 3. b1h þ b2h ¼ b1f þ b2f ¼ b where b 2 ð0; 1Þ Assumption 2 simply means (without loss of generality) that the male group has a stronger preference for public good 1 and the female group has a stronger preference for public good 2. Assumption 3 means that the two groups are alike in terms of their preferences between private goods and total public goods. We impose assumption 3 to abstract away from any differences in behavior that may arise due to differ- ences in preferences for public goods in general. Note that assumptions 2 and 3 allow the preference parameters b2h and b1f to take negative values. In other words, the second good may be a “public bad” for men and the first good may be a “public bad” for women. The preferences of the elected official are defined in a similar manner: Um ðxi ; y1 ; y2 Þ ¼ ðxm Þam ðy1 Þb1m ðy2 Þb2m : As before, he receives a remuneration xm for his official work and his labor contribution cannot be con- tracted upon. He needs to allocate his time between the production of his own private good and the two distinct public goods. We indicate his labor contributions to his own private good and the two public goods by lxm, l1m and l2m, respectively. As before, his chances of re-election depend on how well he has served his constituents in the past. His tenure lasts for one period, following which he has to contest elections. His probability of being suc- 0 0 ^ðkh 1 h Uh þ kf 1 f Uf Þ , where Uh and Uf are vectors describing the util- cessful at the elections is given by p ity levels achieved by, respectively, the male and female community members in the preceding period, 1g is a vector of ones of dimension ng, and kh and kf are positive scalar terms. Thus, the elected official’s chances of re-election depend on a weighted sum of utilities of his constituents with all individuals within each gender group assigned the same weight. As in the case of homogeneous preferences, we assume that the probability function is linear over the relevant interval: min min max max Assumption 4. p ^0 ð:Þ > 0 and p^ 00 ð:Þ ¼ 0 in the interval ½kh nh Uh þ kf nf Uf ; kh nh Uh þ kf nf Uf Š where Ug min and Ug max are, respectively, the minimum and maximim level of utility that community mem- bers of type g 2 fh; f g can achieve in the game.18    18 To be precise, Uhmin ¼ n =ðan þ b ÞÞaþb1h n =ðan þ b ÞÞb2h A , Ufmin ¼ ðnf =ðanf þ b2f ÞÞaþb2f nh = h h 1h f f 2f h  aþb1h  b2h  aþb2f ðanh þ b1h ÞÞb1f Af U ,hmax ¼ ðm þ n Þ=ða þ b Þ ðm þ nf Þ=ða þ b2f Þ Ufmax ¼ ðm þ nf Þ=ða þ b2f Þ Ah and h 1h  b1f ðm þ nh Þ=ða þ b1h Þ Af , where Ah ¼ ðxaÞa ðhb1h Þb1h ðhb2f Þb2h and Af ¼ ðxaÞa ðhb2f Þb2f ðhb1h Þb1f . 110 Casini, Vandewalle, and Wahhaj As in the case of homogeneous preferences, the official’s gain in utility from re-election is equal to a constant, W. Note that the game can potentially have multiple Nash equilibria because both men and women can contribute to each public good. To simplify the analysis, we make an assumption about “separate spheres” of activity for men and women, in the spirit of Lundberg and Pollak (1993). Specifically, we assume that men can only engage in public action with regard to good 1, while women can only engage in public action with regard to good 2. This gender division may be prescribed by social norms so that it is prohibitively costly for women to contribute to public good 1 or for men to contribute to public good 2, regardless of the preferences within each group. The timing of events in the game are as described in the case of homogeneous preferences, but we assume that only the female group has the choice of making the necessary investment for collective action at the first stage of the game. Equilibrium in the Absence of Collective Action First, we consider the case where community members decide on their contribution to the public good in an uncoordinated manner, and the outcome corresponds to a Nash Equilibrium. Recall that the exis- tence of separate spheres implies that individuals in group h can only contribute to public good 1 and individuals in group f can only contribute to public good 2. Then, the optimization problem for an indi- vidual i belonging to group g 2 fh; f g is given by max lxig ;lkig ðxig Þa ðy1 Þb1 g ðy2 Þb2 g (where k ¼ 1 when g ¼ h, and k ¼ 2 when g ¼ f) subject to lxig þ lkig 1, xig ¼ xlxig , y1 ¼ h h Ph i h Pnf i ml1m þ nj¼1 l1jh and y2 ¼ h ml2m þ j¼1 l2jf . We restrict our focus to Nash equlibria where all individuals belonging to the same group make the same labor choices; that is, lxjg ¼ lxg for g ¼ h,f; l1jh ¼ l1h and l2jh ¼ 0; l2jf ¼ l2f and l1jf ¼ 0. Then, from the first-order conditions, we obtain the following optimization conditions: b1h À aml1m an þ aml1m l1h ¼ and lxh ¼ h for men; an h þ b 1 h anh þ b1h b2f À aml2m anf þ aml2m l2f ¼ and lxf ¼ for women: anf þ b2f an f þ b 2 f Note that if aml1m > b1h and aml2m > b2f , then we obtain a corner solution in which the community members do not make any contributions to the public good. Using the labor contributions derived above, we can calculate the utility level achieved by men and women in equilibrium for a given level of labor contribution by the elected official in public goods:   !b2h ml1m þ nh aþb1h ml2m þ nf Uh ðxh ; y1 ; y2 Þ ¼ ðxaÞa ðhb1h Þb1h ðhb2f Þb2h (13) anh þ b1h anf þ b2f !aþb2f   ml2m þ nf ml1m þ nh b1f Uf ðxf ; y1 ; y2 Þ ¼ ðxaÞa ðhb2f Þb2f ðhb1h Þb1f : (14) an f þ b 2 f an h þ b 1 h It should be evident from the expressions that an increase in the elected official’s contribution to either public good affects the welfare of both groups. But, ceteris paribus, women benefit more from contribu- tions to public good 2 (since a þ b2f > b1f ), and men benefit more from contributions to public good 1 (since a þ b1h > b2h ). The World Bank Economic Review 111 Equilibrium in the Presence of Collective Action Next, we consider the case in which female community members coordinate their actions and choose the efficient level of contribution to public good 2 (recall that separate spheres imply that they cannot con- tribute to public good 1). For any given level of labor allocation by men, an efficient contract among the women is given by the solution to the following optimization problem: nf X max ðlxif ;l2if Þ ki ðxif Þa ðy1 Þb1f ðy2 Þb2f ; i¼1 h Pnh hi Pnf i subject to xif ¼ xlxf , y1 ¼ h ml1m þ j¼1 l1jh and y2 ¼ h ml2m þ j¼ 1 l2jf . Here k1 ; :::; knf are Pareto weights. In our analysis, we focus on the symmetric equilibrium, where the utility of each woman is weighted equally and they all make the same contribution to public good 2; that is, l2if ¼ l2f for i ¼ 1; ::; nf . Similarly, all men make the same contribution to public good 1; that is, l1ih ¼ l1h for i ¼ 1; ::; nh . Then, using the first-order condition, we obtain b2f nf À aml2m anf þ aml2m l2f ¼ and lxf ¼ : b 2 f n f þ an f b2f nf þ anf Since collective action enables the group to internalise the externalities provided by their respective contributions to the public good, the level of contribution to the public good is higher. If men do not undertake collective action, their labor contributions are given by the same equations as before (we dis- cuss in the next subsection how collective action by men, exogenously determined, affects the analysis). Using these labor contributions, we can calculate the utility level achieved by men and women in equili- brium for a given level of labor contribution by the elected official in public goods:  aþb1h !b2h ml1m þ nh ml2m þ nf U h ð x h ; y1 ; y2 Þ ¼ ðxaÞa ðhb1h Þb1h ðhb2f Þb2h ; (15) anh þ b1h a þ b2f !aþb2f   ml2m þ nf ml1m þ nh b1f Uf ðxf ; y1 ; y2 Þ ¼ ðxaÞa ðhb2f Þb2f ðhb1h Þb1f : (16) a þ b2f anh þ b1h The presence of collective action by women improves welfare for the male group if b2h > 0 and reduces it if b2h < 0. It improves welfare for the female group, and by a greater extent than for the male group since, by assumption, b2f > 0 and b2f > b2h . Why Collective Action Makes a Difference We now show why and how the presence of collective action affects the incentives of the elected official. Let us suppose that he assigns all individuals of the same gender group the same weight. And, let us denote the weights assigned to male and female community members by km and kf, respectively. Then the elected official’s optimal choice should satisfy the following equation (we derive this condition from the elected official’s optimization problem in the next subsection): @ Uh @ Uf @ Uh @ Uf km nm þ kf nf ¼ km n m þ kf n f @ l1m @ l1m @ l2m @ l2m     (17) @ Uh @ Uh @ Uf @ Uf )n ^m À ¼n^f À ; @ l1m @ l2m @ l2m @ l1m ^m ¼ km nm and n where n ^f ¼ kf nf . In words, equation (17) says that the effect of a shift in a unit of labor from public good 2 to public good 1 on the weighted sum of utilities of male members should be equal and opposite to its effect on the weighted sum of utilities of female members. Using (13)–(16) and (17), 112 Casini, Vandewalle, and Wahhaj we can show that the elected official’s optimal choice must satisfy the following condition (the derivation is shown in the proof of lemma 2):       À1 Uh n ^f am þ b 2 f m b1f m am þ b1h m b2h m ¼ À À ; (18) Uf n ^m ml2m þ nf ml1m þ nh ml1m þ nh ml2m þ nf where Uh and Uf are, respectively, the utility achieved by the male and female community members. The con- ditions in (17) and (18) hold whether or not members of the community are undertaking collective action. For any given combination of l1m and l2m, the ratio Uh =Uf is smaller when women undertake collec- tive action (since such collective action helps the female group more than the male group). Therefore, the combination of l1m and l2m that is optimal in the absence of collective action—that is, satisfies (18)— cannot be optimal in the presence of collective action. Thus, if the total amount of labor allocated to public goods is fixed, the elected official’s optimal allocation of labor involves a larger l2m and a smaller l1m under collective action. In other words, once the women organise, the elected official applies more labor in providing the public good that they care about more. Formally, we have the following result (proof shown in supplemental appendix S2). Lemma 2. Assume that the elected official maximizes a weighted sum of utilities of the elector- ate, applying the same weight to individuals of the same gender, and the total amount of the elected offi- cial’s labor available for the two public goods is fixed. Assume also that men do not undertake collective action in the provision of public good 1. Under assumptions 2 and 3, if women organise to undertake collective action in the provision of public good 2, the elected official shifts labor from public good 1 to public good 2, compared to the situation where they do not. So far we have assumed that the men do not undertake any collective action. A collective action by men contributing to public good 1 would increase the level of provision of this good and, therefore, affect the level of utility achieved by both groups. It is straightforward to extend the analysis above to show that, in such circumstances, a switch by women from a symmetric Nash equilibrium to collective action would still cause the elected official to shift his labor towards public good 2. This is shown in lemma 3 provided in supplemental appendix S2. The Elected Official’s Optimal Choice Using lemma 2, we can determine how the female community members’ decision about whether or not to undertake collective action affects the labor contribution by the elected official. At the second stage of the game, the elected official’s optimization problem can be written as follows: max l1m ;l2m Um ðxm ; y1 ; y2 Þ ¼ ðxm Þa ðy1 Þb1m ðy2 Þb2m þ p ^ðkh nh Uh þ kf nf Uf ÞW; subject to xm ¼ xm þ xlxm , y1 ¼ nhl1y þ mhl1m and y2 ¼ nhl2y þ mhl2m . Here l1y and l2y are the equilibrium level of labor contribution to the public goods—and Uh and Uf are the utility levels achieved—by male and female community members, respectively, as derived earlier. Furthermore, lxm ¼ 1 À l1m À l2m . From the first-order conditions to the elected official’s optimization problem, we obtain, assuming an interior solution, am xðxm Þam À1 ðy1 Þb1m ðy2 Þb2m X nh Xn dUh dUf ¼ b1m mhðxm Þam ðy1 Þb1m À1 ðy2 Þb2m þ W ^0 ð:Þ kh p þW ^0 ð:Þ kf p i ¼1 dl 1 m i¼n þ1 dl 1m (19) h X nh Xn dUh dUf ¼ b2m mhðxm Þam ðy1 Þb1m ðy2 Þb2m À1 þ W ^0 ð:Þ kh p þW ^0 ð:Þ kf p : i ¼1 dl 2 m i¼n þ1 dl 2m h The World Bank Economic Review 113 The introduction of collective action in public good 2 would increase the marginal utility of the elected official’s contribution to that public good. On the other hand, to the extent that the elected official receives utility directly from that public good, it would increase his incentive to free-ride on the contribu- tion made through collective action. These two effects go in opposite directions, and their effect on the elected official’s contribution is ambiguous. If b1m ¼ b2m ¼ 0—that is, the official has no intrinsic preference for the public goods—then the incentive to free-ride is absent. Note that we can obtain the equation in (17) by plugging in b1m ¼ b2m ¼ 0 and rearranging the second equation in (19). In this case, using lemma 2 and lemma 3, we can show that the elected official’s labor contribution to public good 2 is higher if women opt for collective action at the first stage of the game. For the case b1m ; b2m > 0, we can show that the official’s re-election incen- tive outweighs the incentive to free-ride for W sufficiently large. Therefore, when women opt for collec- tive action, he provides a higher labor contribution to public good 2 assuming that he cares sufficiently about being re-elected. Formally, we obtain the following proposition (proof shown in supplemental appendix S2). Proposition 2. Suppose assumptions 2–4 hold and that b2h ! 0. (i) Assume that the elected offi- cial has no intrinsic preference for the public goods. The elected official allocates a higher level of labor in the provision of public good 2 when women organise and take collective action compared to the situa- tion where they do not. (ii) If the elected official has an intrinsic preference for the public goods, that is, b1m ; b2m > 0, then the same result is obtained for W sufficiently large.19 If public good 2 imposes a negative externality on men, that is, b2h < 0, then the official may not increase his labor contribution towards the good in response to collective action by women. This is because collective action by women would not only increase the effect of the official’s labor on the mar- ginal utility of women but also on the marginal disutility of men. But, because of continuity of the func- tions involved, it can be shown that the results of proposition 2 will continue to hold if the negative externality is sufficiently small. It follows from proposition 2 that, if the relevant conditions hold, women will achieve higher utility under collective action (not only because the labor allocation within the group is efficient but also because the official shifts effort towards the public good for which they have a stronger preference). Whether this gain in utility is sufficient for them to make the necessary investment for undertaking col- lective action at the first stage of the game will depend on the size of the fixed cost C, as in the case with homogenous preferences. As the cost may vary across communities, women in some communities may find it in their interest to form collective action groups while others may not. We discuss this issue in more detail in section III. III. Implications of the Theory for SHGs and Collective Action In this section, we discuss the implications of our theoretical results in the context of women’s SHGs that, as we have discussed, engage in various types of collective action to address issues of public concern within their ward. First, it is important to recognise that undertaking collective action involves some fixed costs. The participating individuals have to agree on what action they will undertake, negotiate about the division of labour within the group, and set up a monitoring and enforcement mechanism to minimize 19 The effect of collective action by women on the official’s contribution to public good 1 is ambiguous. Although the offi- cial will tend to shift his available labor towards public good 2 in response to the women’s collective action, it is possible that his contribution to public good 1 will also increase, as the public goods may be complementary in the preferences of the electorate. 114 Casini, Vandewalle, and Wahhaj free-riding. As discussed in section I, there are a variety of groups in the PRADAN villages that under- took collective actions before the NGO began its operations. However, we found only one women-only group, and women play a minor role in the other groups. We argue that the creation of SHGs provides a space for women to interact in a group on a regular basis, build trust, and develop organizational skills, thereby lowering the cost of collective action regarding ward-level issues of interest to women. The crea- tion of these SHGs thus increases the capacity of women in the ward to organise and undertake collective action on “women’s issues” in the subsequent years. v Let us denote by Uc the utility that women in ward v obtain if they undertake collective action, as rep- v resented by equation (16). Similarly, let Un denote the corresponding utility when they do not undertake v v collective action, as represented by equation (14). The gain from undertaking collective action Uc À Un will vary across wards according to the preferences of the WMs and the divergence in preferences of his or her constituents. Let us represent the distribution of gains by the c.d.f. F(.). Following the reasoning in section II, the theoretical model implies that members of the SHG would always experience higher wel- fare when they undertake collective action compared to when they do not; that is, F(0) ¼ 0. Let us denote by C0 the fixed cost per woman for engaging in collective action when they do not have any kind of external organizational support. Given the absence of women-only groups that undertook collective v v action prior to the creation of the SHGs, it seems reasonable to assume that Uc À Un < C0 for each v; that is, FðC0 Þ ¼ 1. The creation of SHGs, we argue, lowered the cost of collective action to Cshg < C0 such that some fraction of the SHGs found it in their interest to organise themselves and undertake collective action; that is, FðCshg Þ < 1. We argue that, in our empirical setting, the WMs have similar strategic interests to those of the “elected official” considered in the theoretical model. The WMs care about being re-elected and, there- fore, have incentives to take action regarding the public issues that his constituents consider to be of high priority. On the other hand, the time he spends addressing the ward’s problems and needs has an oppor- tunity cost in terms of the loss of earnings from alternative income-generating activities. Furthermore, the WM’s strategic incentives depend on whether his electorate—including members of SHGs—are capa- ble of undertaking collective action. As discussed in section I, SHG members undertake two types of action that complement the WM’s own efforts: (i) communicating their concerns about a public issue to a higher official; and (ii) interven- ing directly in the ward. In the empirical analysis, we define the SHG as being active if the survey records either of these undertakings during the relevant time period. It should be noted that, for a number of public goods (for example, public infrastructure, welfare schemes, etc.), resources and inputs provided by higher government authorities are an essential element in their provision. But, in the context of the PRADAN villages, the WMs and citizens provide an impor- tant input for the provision of such goods by identifying problems or ensuring the quality of delivery. The labour contributions by the elected official and individuals that we model would correspond to these types of inputs. We do not model the contributions of the higher authorities explicitly, as our focus is on the strategic interaction between the WMs and citizens. It is important to note that whether or not an SHG is active is an endogenous variable in our setting. The SHG may choose not to engage in collective action if, for example, the WM is doing a satisfactory job of addressing their social concerns. However, the timing of the creation of SHGs provides an exoge- nous source of variation in collective action by SHGs. As discussed in section I, SHGs generally became active with a time-lag following their creation. This is logical given that members of the SHG would have to gain expertise regarding its basic operations before it is able to engage in collective action. Thus, the variation in the timing of SHG creation allows us to estimate how WMs would respond to exogenous changes in the level of collective action among their female constituents. The World Bank Economic Review 115 As mentioned in section I, we do not directly observe the WM’s effort in addressing issues of interest to his or her constituents. But we do have information on the set of issues regarding which the SHGs engaged in collective action and the type of issues that the WMs dealt with (see table 2). Table 3 shows that some issues of concern to the women’s groups, such as “alcohol problems” and “school problems,” had been dealt with by very few WMs before the SHGs were created. Proposition 2 implies that an exog- enous increase in collective action increases the WM’s labour contribution to public issues of interest to women and, therefore, in the present context, increases the probability that the WM tackles such issues. Relatedly, it should be noted that, if there were certain issues of no concern to women’s groups or outside of their sphere of control, then, as per the discussion in section II, an increase in collective action by women may cause the WM to decrease his effort in providing them. But we would not be able to detect the change in our data unless the WM ceased his efforts altogether. This is unlikely to be the case, and we find no evidence of this in the descriptive statistics. Combined with the reasoning in the previous paragraph, this implies that collective action by the SHGs may have caused their WMs to increase the number of public issues they dealt with. Thus, the prediction—based on proposition 2—that we test empirically can be summarized as follows: Prediction 1. An exogenous increase in collective action by women within the ward, due to the creation of an SHG, would increase the probability that the WM tackles public issues that are of interest to the group and would cause the WM to tackle a greater variety of issues. As discussed in section I, some PRADAN villages included other groups, primarily composed of men, that also undertook collective action on a variety of community-level issues. As discussed in section II, our theoretical result—on how an elected official responds to the introduction of collective action by women— holds regardless of whether or not male members of the community undertake collective action. Therefore, prediction 1 is applicable both in communities where there were pre-existing male groups that undertook collective action on ward-related issues and in communities where there were not. IV. Empirical Analysis In this section, we discuss the identification strategy and formally test the prediction suggested by the model. To provide further support to our results, we run a placebo regression. Finally, we present descriptive statistics on the qualitative impact of SHGs’ collective actions. Identification Strategy For each ward, we gathered information on the political activities over the past twenty years (four elec- tions). We asked WMs about the type of issues they dealt with, that is, the issues for which they visited a higher official or intervened directly in the ward. This allows us to construct a panel for 141 wards. Consider the following OLS regression to estimate whether the collective actions undertaken by SHGs have an effect on the activities of WMs: Tijt ¼ a1 þ a2 SHG activeijt þ a3 Xijt þ Ct þ Dj þ eijt ; (20) where Tijt is the total number of different issues that WM i, elected in ward j in year t, deals with. SHG active is a dummy that takes value 1 if an SHG started undertaking collective actions in ward j before or during the mandate of WM i. The dummy takes value 0 if no SHG was present in ward j during the man- date of WM i or if existing SHGs did not yet start undertaking collective actions. Xijt are WM level char- acteristics, including education level, land ownership, the total number of children, age, caste category,20 and a dummy indicating whether the WM is a man. Ct is a set of three dummies controlling 20 Castes are classified in the following categories: ST (scheduled tribe), SC (scheduled caste), and OBC (other backward caste)/FC (forward caste). 116 Casini, Vandewalle, and Wahhaj for the year in which the WM was elected (elected in ’92, ’97, and ’02). The dummies are included to ensure that our variable of interest does not pick up election-year effects, as, for example, the quality of WMs might increase over time. Finally, Dj are ward fixed effects that control for differences in time- invariant unobservables across wards, and eijt is the error term. We cluster the standard errors at the ward level. Our panel is WM specific, so in case a WM is re-elected, we have information for the full period only (we assumed it would be difficult for WMs to provide term-wise information).21 The decision of an SHG to become active is potentially endogenous. OLS underestimates the influ- ence of SHGs on the activities of WMs if SHGs choose not to engage in collective action because the WM is doing a satisfactory job. On the other hand, OLS overestimates the effect if a WM particularly sensitive to women’s issues encourages SHGs to undertake collective actions. As discussed in section III, we use the variation in the timing of the creation of SHGs as an exogenous source of variation in the tim- ing of the first collective action by SHGs. Table 4 overviews the evolution of both the creation of SHGs and their activity over the election periods. The first SHGs were created during the 1997–2001 election period and, by the end of the 2002–06 period, all wards had at least one SHG. As mentioned before, on average, SHGs that undertake collective actions, do so for the first time after three years of weekly meetings. Therefore, it is not surprising that most SHGs start undertaking collective actions during the mandates 2002–06 and 2007–11. We will instrument SHG active by whether or not an SHG was created in the ward during the rele- vant time period. Our approach leads to the following first stage regression: SHG activeijt ¼ b1 þ b2 SHG createdijt þ b3 Xijt þ Ct þ Dj þ fijt ; where SHG created is a dummy that takes value 1 if SHGs were present in ward j during the mandate of WM i. We use the parameters to predict whether SHGs undertake collective actions, and estimate consis- tent estimators for: d ijt þd3 Xijt þ Ct þ Dj þ hijt : Tijt ¼ d1 þ d2 SHG active (21) The assumption underlying this analysis is that the timing of the creation of SHGs is uncorrelated with pre-existing differences in the tendency to intervene for issues across wards, after controlling for time- varying WM characteristics, time-invariant ward characteristics, and a common trend. Next, we assess the plausibility of this assumption. Exogeneity of the Timing of SHG Creation In 1994, PRADAN opened an office in the town Keonjhar and started operating in the poorest block, namely Banspal, which is southwestwards of the town (Figure 1). The initial focus was on agriculture, as villagers owned some land and small but perennial rivulets were available to provide irrigation without the need of major investments. A few years later, in 1998, PRADAN started promoting SHGs to provide extra resources to strengthen livelihoods. In the following years, PRADAN identified the poorest regions in the contiguous areas and expanded its activities. At first, two new offices were opened: one in Suakati, in the Banspal block in 1998 (figure 1), and a second in Turumunga, in the Patna block in 1999 (figure 2). Later, from 2000 onwards, employees based at the Keonjhar town office started activities in the Keonjhar Sadar block, located to the west of Keonjhar (figure 1), and in the adjacent Karanjia block, which is located in the Mayurbhanj district (figure 3). PRADAN’s decision about where to operate was based on the suitability of an area for its agricultural programs. We surveyed all the wards where PRADAN operates in the Keonjhar and Mayurbhanj 21 To test the sensitivity of our results, we also ran regressions in which we included term-wise information. The dependent variable was the same over the different terms, but the independent variables differed. Our results did not change using this different specification. The World Bank Economic Review 117 Table 4. The Creation and Activity of SHGs by Election Period Election mandate: 1992–96 1997–2001 2002–06 2007–11 % of wards in which at least one SHG has been created 0.0 42.6 100.0 100.0 % of wards in which at least one SHG has started 0.0 28.3 68.1 92.9 undertaking collective actions (conditional on an SHG being present) Source: Authors’ analysis based on data described in the text. districts of Odisha. Therefore, we believe that, in terms of our research question, the program placement bias is limited. However, the program timing bias is a concern if the wards in which PRADAN started working earlier within blocks are systematically different with respect to the outcomes of interest. Figures 1, 2, and 3 show the villages in which PRADAN promoted SHGs. We show villages instead of wards, as ward-level maps do not exist. The villages are a good approximation, as 75% of the villages encompass exactly the same territory as wards and only 2.8% of the wards cover several villages. We col- ored the villages according to the mandate during which a first SHG was created in the village: 1997–2001 versus 2002–06.22 We distinguish between these two periods, as they correspond to the election periods during which SHGs were created (see table 4). The maps reveal two important patterns. First, all PRADAN offices in all blocks created SHGs both before and after the 2002 elections. Second, the villages in which SHGs were created earlier and later are neighboring one another. This is the result of PRADAN’s preference for having the same employee operate in the same villages over time: each employee is assigned a cluster of neighboring villages in which he creates SHGs following the pace and the direction he considers appropriate. This policy saves traveling time and limits transportation costs. We address potential employee biases by including ward fixed effects. As the villages in which PRADAN created SHGs before 2002, and those in which SHGs were created later are situated close together, it is likely that they have sim- ilar socioeconomic characteristics. We check whether this is indeed the case in the tables 5 and 6. Table 5 displays data from the 1991 Census of India. The village is the lowest level at which data is available, so we cannot do a ward-level comparison. We could match all but three villages (105 total). The first column provides details on villages in which the first SHG was created during the 1997–2001 mandate, the second column on villages that had a first SHG created during the 2002–06 mandate, and the third column provides the difference. With the exception of power supply, there are no significant differences for a wide range of facilities and public infrastructures. Villages also have, on average, the same size, and similar rates of employment, and literacy. We also used all of the variables in table 5 to predict SHG creation (a dummy taking value 1 (0) if the first SHG was created in 2002–06 (1997–2001)). We cannot reject the joint null hypothesis (p-value ¼ 0.64).23 This result provides an argument in favor of SHG creation not being influenced by the elected WMs. Next, table 6 shows the outcome variables and controls of our main regressions for the period 1992– 96, that is, before the first SHGs were created, at the ward level. Although the number of observations is limited, the difference is small for all variables except for welfare schemes. Furthermore, the sign is not systematically positive or negative.24 22 In case there are several wards in the village, we use the mandate during which the first SHG was created in any of the wards that belong to that village. 23 The regression is available upon request. 24 We excluded the WMs who have been re-elected in 1997 and faced SHGs undertaking collective actions, since for them we do not have disaggregated information. This reduces the number of observations, but allows us to compare informa- tion on WMs before SHGs were active. 118 Casini, Vandewalle, and Wahhaj Figure 1. Keonjhar District: Banspal and Keonjhar Blocks Source: The map was downloaded from http://ceoorissa.nic.in/acmaps.asp (AcNo 24), accessed August 03, 2015. The World Bank Economic Review 119 Figure 2. Keonjhar District: Patna Block Source: The map was downloaded from http://ceoorissa.nic.in/acmaps.asp (AcNo 23), accessed August 03, 2015. Figure 3. Mayurbhanj District: Karanjia Block 120 Source: The map was downloaded from http://ceoorissa.nic.in/acmaps.asp (AcNo 30), accessed August 03, 2015. Casini, Vandewalle, and Wahhaj The World Bank Economic Review 121 Table 5. Village Characteristics Based on the Census of India 1991, by Mandate during which a First SHG was Created in the Village First SHG created in the village in 1997–2001 2002–06 Difference (Std. Dev.) (Std. Dev.) (Std. Err.) Number of households 129.3 112.4 16.9 (83.2) (84.5) (16.5) Literacy rate of women 12.3 15.5 À3.2 (9.5) (10.9) (2.0) Literacy rate of men 33.1 37.6 À4.5 (16.0) (17.2) (3.3) Employment rate of women 18.8 16.3 2.5 (15.8) (16.9) (3.2) Employment rate of men 53.6 51.9 1.7 (5.2) (9.3) (1.5) EDUCATION FACILITIES Number of primary schools 0.87 0.78 0.09 (0.40) (0.46) (0.09) Number of middle schools 0.24 0.14 0.10 (0.43) (0.39) (0.08) Number of high schools 0.13 0.07 0.06 (0.34) (0.31) (0.06) MEDICAL FACILITIES A medical facility is available (% of villages) 15.2 6.8 8.4 (36.3) (25.4) (6.0) DRINKING WATER FACILITIES Well water (% of villages) 73.9 83.1 À9.2 (44.4) (37.8) (8.0) Tank water (% of villages) 45.7 54.2 À8.5 (50.4) (50.2) (9.9) Tube well (% of villages) 30.4 32.2 À1.8 (46.5) (47.1) (9.2) River water (% of villages) 28.3 20.3 8.0 (45.5) (40.6) (8.4) OTHER INFRASTRUCTURE Bus stop (% of villages) 17.4 13.6 3.8 (38.3) (34.5) (7.1) Village accessibility by paved road (% of villages) 28.3 25.4 2.9 (45.5) (43.9) (8.8) Market facility available (% of villages) 13.0 6.8 6.2 (34.1) (25.4) (5.8) Power supply (% of villages) 54.3 71.2 À16.9* (50.4) (45.7) (9.4) Irrigated land (% of land) 5.2 6.1 À0.9 (11.4) (17.6) (3.0) Number of villages 46 59 ***Significant difference at 1 percent, **significant difference at 5 percent, *significant difference at 10 percent. Source: Authors’ analysis based on data from the Census of India, 1991. 122 Casini, Vandewalle, and Wahhaj Table 6. Summary Statistics at Baseline (1992–96), by Mandate during which a first SHG was Created First SHG created in the ward in 1997–2001 2002–06 Difference (Std. Dev.) (Std. Dev.) (Std. Err.) DEPENDENT VARIABLES FOR 1992–96 Number of ward problems addressed by the WMs 1.42 1.36 0.06 (0.87) (1.09) (0.23) WM deals with alcohol issues (%) 0.0 2.6 À2.6 (0.0) (16.0) (2.8) WM deals with school issues (%) 6.1 7.7 À1.6 (24.2) (27.0) (6.1) WM deals with forest issues (%) 18.2 23.1 À4.9 (39.2) (42.7) (9.7) WM deals with public infrastructure (%) 78.8 71.8 7.0 (41.5) (45.6) (10.4) WM deals with welfare schemes (%) 39.4 25.6 13.8 (49.6) (44.2) (11.1) CHARACTERISTICS WMs IN 1992–96 (5 independent variables) Male (%) 87.9 79.5 8.4 (33.1) (40.9) (8.9) Education level (years) 4.9 6.8 À1.9** (3.1) (3.7) (0.8) Land (acres) 3.0 3.2 À0.2 (2.0) (7.8) (1.4) Number of children 3.7 3.1 0.6 (1.8) (1.8) (0.4) Age 58.8 52.3 6.5** (11.1) (12.2) (2.8) Caste category: SC (%) 6.1 7.7 À1.6 (24.2) (27.0) (6.1) Caste category: OBC/FC (%) 12.1 25.6 À13.5 (33.1) (44.2) (9.4) Number of wards 33 39 *** Significant difference at 1 percent, ** significant difference at 5 percent, * significant difference at 10 percent. Source: Authors’ analysis based on data from the Census of India, 1991. Results First, we measure the impact of SHGs on the number of different problems a WM deals with. The results are shown in table 7. Columns (1) and (2) provide the estimates using OLS (equation 20), while columns (3) and (4) use instrumental variables (equation 21). Regressions without WM controls are provided in the odd-numbered columns and those with controls in the even-numbered columns. WMs operating in wards where SHGs have started undertaking collective actions deal with 1.5 extra issues compared to WMs who operate in wards where SHGs are not present or do not undertake actions. The result confirms our prediction: an increase in collective actions within the ward causes the WM to tackle a greater variety of issues. The coefficients are significant, but slightly smaller in size in the OLS regressions. We also find that, overall, male WMs deal with fewer issues. Next, we examine the type of extra issues the WM starts dealing with. To do that, we replace the left hand side variable of the equations (20) and (21) with Tijtk, a dummy equal to one if WM i, elected in ward j in year t, deals with issue k. The independent variables are the same. The results are given in table 8. The odd numbered columns provide the results using OLS, while the even numbered columns use IV. The World Bank Economic Review 123 Table 7. The Effect of Collective Actions by SHGs on the Number of Issues Addressed by the WM Number of issues addressed by the WM OLS 2SLS (1) (2) (3) (4) SHG active 0.89*** 0.91*** 1.48*** 1.54*** (0.16) (0.16) (0.42) (0.42) Male — À0.29** — À0.33*** — (0.12) — (0.13) Education level (years) — 0.03 — 0.03 — (0.02) — (0.02) Land (acres) — À0.04*** — À0.04*** — (0.01) — (0.01) Number of children — À0.02 — À0.01 (0.03) (0.03) Age — 0.05* — 0.05* — (0.03) — (0.03) Squared age — À0.00* — À0.00 — (0.00) — (0.00) Caste category: SC — À0.16 — À0.12 — (0.21) — (0.23) Caste category: OBC/FC — 0.02 — 0.06 — (0.16) — (0.17) Ward fixed effects Yes Yes Yes Yes Election year dummies Yes Yes Yes Yes F-stat first stage — — 31.6 29.1 Observations 441 441 441 441 An observation in the regression model is a WM. The dependent variable is the total number of different issues that the WM deals with during his mandate. The variable “SHG active” is a dummy that takes value one if an SHG started undertaking collective actions in the ward before or during the mandate of the WM. The first two columns provide the results using standard linear regression techniques. In the last two columns we instrument for “SHG active” using the timing of SHG cre- ation in the ward. The uneven columns do not include WM characteristics, while the even columns do so. Standard errors clustered at the ward level are given in parentheses. *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent Source: Authors’ analysis based on data described in the text. Once SHGs undertake collective actions, WMs are 29% more likely to deal with alcohol problems, 31% with school problems, and 35% with forestry issues. WMs are also more likely to start dealing with welfare schemes. Thus, the estimates confirm that WMs start dealing with public goods that are of interest to SHGs (see section I). The bias is similar to the previous regression: The IV coefficients are close to or larger than their OLS counterparts. Interestingly, male WMs are less likely to take care of alcohol issues, school problems, and welfare schemes, but they are more likely to deal with public infrastructure. Placebo Regression In this section, we perform a placebo regression. SHGs become active during the election periods 2002– 06 and 2007–11 in 78% of the wards. Before 2002, there should not be differences between wards where SHGs started undertaking collective actions in 2002–06 as compared to those in which SHGs became active in 2007–11. To test this, we restrict our data set to WMs whose mandate finished before 2002, that is, we keep the election periods 1992–96 and 1997–2001 and run the following differences- in-differences regressions: 124 Casini, Vandewalle, and Wahhaj Table 8. The Effect of Collective Actions by SHGs on the Type of Issues WMs deal With Probability that the WM deals with: Alcohol issues School issues Forest issues Infrastructure Welfare scheme OLS IV OLS IV OLS IV OLS IV OLS IV (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) SHG active 0.30*** 0.29** 0.19*** 0.31** 0.22*** 0.35** 0.04 0.10 0.14* 0.40* (0.06) (0.14) (0.06) (0.14) (0.05) (0.17) (0.07) (0.21) (0.08) (0.23) Male À0.11*** À0.11*** À0.07 À0.08* À0.06 À0.06 0.09* 0.09* À0.14*** À0.16*** (0.04) (0.04) (0.04) (0.04) (0.04) (0.04) (0.05) (0.05) (0.05) (0.06) Education level (years) 0.01** 0.01** À0.00 À0.00 À0.01 À0.01 0.00 0.00 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Land (acres) À0.01 À0.01 À0.01*** À0.01*** À0.01 À0.01 À0.01 À0.01 À0.00 À0.00 (0.01) (0.01) (0.00) (0.00) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Number of children À0.02** À0.02** 0.01 0.01 À0.00 0.00 À0.03 À0.03 0.01 0.01 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.02) (0.02) (0.01) (0.01) Age 0.00 0.00 0.01 0.01 0.00 0.00 0.02* 0.02* 0.02* 0.02 (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Squared age 0.00 0.00 À0.00 À0.00 À0.00 À0.00 À0.00 À0.00* À0.00* À0.00* (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Caste category: SC À0.08 À0.08 0.05 0.06 À0.12 À0.12 À0.10 À0.10 0.11 0.13 (0.06) (0.06) (0.08) (0.08) (0.08) (0.08) (0.09) (0.09) (0.11) (0.11) Caste category: OBC/FC À0.14** À0.14** 0.08* 0.08** À0.03 À0.03 À0.01 À0.01 0.16** 0.18** (0.06) (0.06) (0.04) (0.04) (0.05) (0.05) (0.05) (0.05) (0.07) (0.07) Ward fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Election year dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Observations 441 441 441 441 441 441 441 441 441 441 An observation in the regression model is a WM. The dependent variable is a binary variable that equals one if the WM deals with the issue during his mandate. The variable “SHG active” is a dummy that takes value one if an SHG started undertaking collective actions in the ward before or during the mandate of the WM. The odd numbered columns provide the results using OLS. In the even numbered columns we instrument for “SHG active” using the timing of SHG creation in the ward. Standard errors clustered at the ward level are given in parentheses. *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent. Source: Authors’ analysis based on data described in the text. Tijtjt < 2002 ¼ bActive2002À2006 þ X0ijt c þ dElected1997 þ Dj þ ijt Tijtkjt < 2002 ¼ bActive2002À2006 þ X0ijt c þ dElected1997 þ Dj þ ijtk : Tijt is the total number of different issues that the WM deals with, and Tijtk is a dummy equal to one if WM i, elected in ward j in year t, deals with issue k. The binary variable Active2002–2006 takes value one if SHGs became active in the ward in the election period 2002–06 and zero otherwise. The election-year dummies Ct are replaced by a dummy indicating that the WM is elected in 1997. SHGs became active during the election period 1997–2001 in 12.2% of the wards. We drop these wards in the regressions.25 Table 9 shows that the timing of becoming active (2002–06 versus 2007–11) does not affect the num- ber of issues discussed before 2002: the difference is small (less than one third of an issue) and not signifi- cant. The type of issues discussed are not significantly different either. Therefore, we can conclude that the WM’s public good activities—prior to the first SHGs being active—cannot be predicted by the fact that SHGs will become active in the future. 25 The results do not change if we include those wards. The independent variable “Caste category: SC” is omitted due to perfect collinearity. We do not have results for alcohol issues, as there is not enough variation (only 2% of the WMs dealt with this topic before 2002). The World Bank Economic Review 125 Table 9. Placebo Test Prediction 1 Prediction 2 Number of issues addressed by the WM Probability that the WM deals with: School issues Forest issues Infrastructure Welfare schemes (1) (2) (3) (4) (5) Active2002–2006 À0.28 À0.04 À0.02 À0.15 0.00 (0.20) (0.04) (0.06) (0.13) (0.13) Male À0.09 0.03 À0.01 0.11 À0.22 (0.24) (0.03) (0.03) (0.15) (0.13) Education level (years) 0.05* À0.00 0.01 0.02 0.00 (0.03) (0.00) (0.01) (0.02) (0.02) Land (acres) 0.00 0.00 0.00 À0.00 0.00 (0.01) (0.00) (0.00) (0.01) (0.00) Number of children À0.05 0.00 0.00 À0.08** 0.03 (0.06) (0.00) (0.01) (0.04) (0.03) Age 0.02 0.02 0.01 0.04 À0.05 (0.09) (0.02) (0.01) (0.05) (0.05) Squared age À0.00 À0.00 À0.00 À0.00 0.00 (0.00) (0.00) (0.00) (0.00) (0.00) Caste category: OBC/FC À0.58 À0.08 À0.03 À0.33* À0.13 (0.40) (0.08) (0.04) (0.19) (0.24) Elected in ’97 0.55*** 0.02 0.06 0.29** 0.12 (0.17) (0.02) (0.06) (0.11) (0.11) Ward fixed effects yes yes yes yes yes Observations 142 142 142 142 142 An observation in the regression model is a WM. The data set is restricted to WMs whose mandate ended before 2002. In column (1) the dependent variable is the total number of different issues that the WM deals with during his mandate. In columns (2)-(5) the dependent variable is a binary variable that equals one if the WM deals with the issue during his mandate. The variable “Active2002À2006 ” is a binary variable that equals one if SHGs became active in the ward in the election period 2002–06 and zero otherwise. Standard errors clustered at the ward level are given in parentheses. *** significant at 1 percent, ** significant at 5 percent, * significant at 10 percent Source: Authors’ analysis based on data described in the text. The Qualitative Impact As mentioned in section I, a caveat of our empirical results is that we can focus on the type of public goods only and not on the quality of the work of the WM. To partially overcome this limitation, we asked SHGs how effective their collective actions were, that is, whether they obtained what they requested or, at least, received the promise that a solution was to be provided. The answers to these ques- tions are summarized in column (1) of table 10. SHGs obtained what they requested in 43–60% of the wards where they undertook collective actions, and they received the promise that a solution would be provided in 28–47% of the wards. They received either a solution or promise in 86–92% of the wards (obtained by summing up the percentages).26 Most likely, column (1) overestimates the success of SHGs, as there might be a selection issue: SHGs undertake actions only if they believe they will be successful. We try to correct for selection in the columns (2) and (3). For each type of issue, we asked SHGs if they faced it at least once. In column (2) we show the percentages over all the wards where the problem appeared, independent of whether an SHG undertook action or not. Hence, this is equivalent to treating SHGs that did not undertake actions as not successful. Finally, in column (3), we assume that all of the problems appeared in all of the wards. 26 Examples of successful stories are a replacement of the school teacher, the prohibition of alcohol brewed by outsiders, the reparation of a well, the obtainment of a widow pension or below the poverty line card, etc. 126 Casini, Vandewalle, and Wahhaj Table 10. Outcome of Collective Actions Undertaken by SHGs (1) Wards where SHGs undertook (2) Wards where the (3) All wards collective actions problem appeared Wards Solution Promise Wards Solution Promise Wards Solution Promise (#) (%) (%) (#) (%) (%) (#) (%) (%) Public infrastructure 92 42.5 46.8 135 20.0 23.7 141 17.1 19.7 (34.0) (35.5) (25.0) (29.7) (23.1) (25.9) Welfare schemes 60 59.2 30.5 113 25.5 13.4 141 14.0 7.7 (42.1) (37.6) (37.3) (28.0) (25.0) (19.2) Alcohol problems 66 51.9 39.7 121 20.1 15.7 141 16.4 12.9 (38.1) (36.6) (29.5) (25.5) (27.5) (23.6) School problems 35 58.2 27.5 98 13.6 5.6 141 5.6 2.5 (40.5) (33.2) (27.5) (14.9) (16.7) (10.3) Forest issues 74 59.6 32.3 121 26.8 14.0 141 19.9 10.6 (34.8) (34.4) (32.7) (23.6) (27.7) (18.9) We asked SHG members whether they obtained what they requested thanks to their collective actions (“solution”) or, at least, received the promise that a solution was to be provided (“promise”). The columns (1) show summary statistics for wards where at least one SHG undertook collective actions for the issue. The columns (2) show the percentages over all the wards where the problem appeared (we asked SHGs whether they faced the problem in their ward). Finally, the columns (3) assume that all the problems appeared in all the wards. Standard deviations are given in parentheses. Source: Authors’ analysis based on data described in the text. Column (2) might underestimate the impact of SHGs, and yet the figures are still remarkable. The data suggest that undertaking action for alcohol issues led to a solution in one fifth of the wards where the problem appeared.27 Therefore, despite some obvious limitations, we believe to be providing reliable evidence about the positive impact of collective actions by socially disadvantaged women. V. Conclusions We examine the impact of SHGs’ collective actions on the variety of problems dealt with by local gov- ernments. Using a theoretical model, we show that an elected official, whose aim is to maximize re- election chances, would exert higher effort in providing public goods when private citizens undertake collective action and coordinate their voluntary contributions towards the same. This occurs although government and private contributions are assumed to be perfect substitutes. We test the prediction of the model using firsthand data on SHGs in India. We find that WMs take care of a larger variety of ward issues when SHGs undertake collective actions and that they start dealing with the public goods pre- ferred by SHGs. In particular, WMs are more likely to deal with alcohol, forestry, and school problems. Our results are important in light of the attention given to SHGs in anti-poverty programs in India. For example, the National Rural Livelihood Mission puts forward the creation of groups as a first step in its poverty alleviation program. As shown by Desai and Joshi (2014), SHGs help overcoming the bar- riers to collective actions imposed by traditions and social pressure. We complement their findings by providing evidence that this effort is recognized by local authorities and affects their behavior. An additional and interesting research question is whether the phenomena we observe are related or not to the financial role of SHGs. Potentially, similar outcomes could be attained by different types of groups. However, this should not lead to understatements on the role of microfinance in India. In the 27 This increase is remarkable, as we know that WMs barely dealt with alcohol issues before SHGs were created (see ta- ble 3). The World Bank Economic Review 127 context of our survey region, where the social role of women is restrained, intra-household interactions may play an important role. In this respect, the explicit financial aim of SHGs can make the difference by providing a socially acceptable reason for women to meet regularly. References Anderson, S., and J.-M. Baland. 2002. “The Economics of Roscas and Intrahousehold Resource Allocation.” The Quarterly Journal of Economics 117 (3): 963–95. Angelucci, M., D. Karlan, and J. Zinman. 2015. “Microcredit Impacts: Evidence from a Randomized Microcredit Program Placement Experiment by Compartamos Banco.” American Economic Journal: Applied Economics 7(1): 151–82. Babu, B. V., and S. K. Kar. 2010. “Domestic Violence in Eastern India: Factors Associated with Victimization and Perpetration.” Public Health 124 (3): 136–48. 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. Banerjee, A., and E. Duflo. 2007. “The Economic Lives of the Poor.” The Journal of Economic Perspectives 21 (1): 141–67. Banerjee, A., E. Duflo, R. Glennerster, and C. Kinnan. 2015. “The Miracle of Microfinance? Evidence from a Randomized Evaluation.” American Economic Journal: Applied Economics 7 (1): 22–53. Bergstrom, T., L. Blume, and H. Varian. 1986. “On the Private Provision of Public Goods.” Journal of Public Economics 29 (1): 25–49. Chattopadhyay, R., and E. Duflo. 2004. “Women as Policy Makers: Evidence from a Randomized Policy Experiment in India.” Econometrica 72 (5): 1409–43. Chowdhury, M., P. Mosley, and A. Simanowitz. 2004. “The Social Impact of Microfinance: Introduction.” Journal of International Development 16 (3): 291–300. Cre ´ . 2015. “Estimating the Impact of Microcredit on those who Take it ´ pon, B., F. Devoto, E. Duflo, and W. Pariente Up: Evidence from a Randomized Experiment in Morocco.” American Economic Journal: Applied Economics 7 (1): 123–50. Datta, U. 2015. “Socio-Economic Impacts of JEEViKA: A Large-Scale Self-Help Group Project in Bihar, India.” World Development 68: 1–18. Deininger, K., and Y Liu. 2009. “Economic and Social Impacts of Self-help Groups in India.” World Bank Policy Research Working Paper 4884, Washington, DC. Desai, R. M., and S. Joshi. 2014. “Collective Action and Community Development: Evidence from Self-Help Groups in Rural India.” World Bank Economic Review 28 (3): 492–524. Duflo E. 2012. “Women Empowerment and Economic Development.” Journal of Economic Literature 50 (4): 1051–79. Edlund, L., L. Haider, and R. Pande. 2005. “Unmarried Parenthood and Redistributive Politics.” Journal of the European Economic Association 3 (1): 95–119. Edlund, L., and R. Pande. 2002. “Why Have Women Become Left-Wing? The Political Gender Gap and the Decline in Marriage.” The Quarterly Journal of Economics 117 (3): 917–61. Feigenberg, B., E. Field, and R. Pande. 2013. “The Economic Returns to Social Interaction: Experimental Evidence from Microfinance.” Review of Economic Studies 80 (4): 1459–83. Khanna, M., N. Kochhar, and N. Palaniswamy. 2015. “A Retrospective Impact Evaluation of the Tamil Nadu Empowerment and Poverty Alleviation (Pudhu Vaazhvu) Project.” Journal of Development Studies 1–14. Lott, J. R., and L. W. Kenny. 1999. “Did Women’s Suffrage Change the Size and Scope of Government?” The Journal of Political Economy 107 (6): 1163–98. Lundberg, S., and R. A. Pollak. 1993. “Separate Spheres Bargaining and the Marriage Market.” Journal of Political Economy 101 (6): 988–1010. Mishra, A. K. 1999. “Orissa: Tribal Women against Liquor.” Economic and Political Weekly 34 (19): 1089–91. NABARD. 2013. “Status of Microfinance in India 2011–2012.” Technical Report, National Bank for Agriculture and Rural Development, Mumbai, www.nabard.org. 128 Casini, Vandewalle, and Wahhaj Neufeld, K. J., D. H. Peters, M. Rani, S. Bonu, and R. K. Brooner. 2005. “Regular Use of Alcohol and Tobacco in India and its Association with Age, Gender, and Poverty.” Drug and Alcohol Dependence 77 (3): 283–91. Panda, P., and B. Agarwal. 2005. “Marital Violence, Human Development and Women’s Property Status in India.” World Development 33 (5): 823–50. PRADAN. 2005. “Group Dynamics: Promoting Women’s Self-Help Groups amongst the Rural Poor.” Technical Report, Professional Assistance for Development Action, Delhi. Rao, V. 1997. “Wife-Beating in Rural South India: A Qualitative and Econometric Analysis.” Social Science & Medicine 44 (8): 1169–80. Roberts, R. D. 1984. “A Positive Model of Private Charity and Public transfers. Journal of Political Economy 92 (1): 136–48. Tarozzi, A., J. Desai, and K. Johnson. 2015. “The Impacts of Microcredit: Evidence from Ethiopia.” American Economic Journal: Applied Economics 7 (1): 54–89. Warr, P. G. 1982. “Pareto Optimal Redistribution and Private Charity.” Journal of Public Economics 19 (1): 131–38. Xaxa J. 2010. “Grama Panchayat in Orissa.” Orissa Review.