WPS7361 Policy Research Working Paper 7361 Electoral Rules and Political Selection Theory and Evidence from a Field Experiment in Afghanistan Andrew Beath Fotini Christia Georgy Egorov Ruben Enikolopov East Asia and the Pacific Region Office of the Chief Economist July 2015 Policy Research Working Paper 7361 Abstract Voters commonly face a choice between competent can- single-member districts to prefer candidates with polarized didates and those with policy preferences similar to their policy positions over more competent candidates. Results own. This paper explores how electoral rules, such as from a unique field experiment in Afghanistan are con- district magnitude, mediate this trade-off and affect the sistent with these predictions. Specifically, representatives composition of representative bodies and policy outcomes. elected in elections with a single multi-member district are The paper shows formally that anticipation of bargain- better educated and exhibit less extreme policy preferences. ing over policy causes voters in elections with multiple This paper is a product of the Office of the Chief Economist, East Asia and the Pacific Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at abeath@worldbank.org, cfotini@mit.edu, g-egorov@kellogg.northwestern.edu, and ruben.enikolopov@ upf.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Electoral Rules and Political Selection: Theory and Evidence from a Field Experiment in Afghanistan Andrew Beathy Fotini Christiaz Georgy Egorovx Ruben Enikolopov{ Keywords: electoral rules, political selection, district magnitude, quality of politicians, com- petence, polarization, local public good, legislative bargaining, eld experiment, Afghanistan. JEL Codes: D72, D78. We thank Ernesto Dal Bó, Hülya Eraslan, Horacio Larreguy, Tommaso Nannicini, Mattias Polborn, Mohamed Saleh, and participants of the NBER Summer Institute Political Economy Public Finance meeting, Priorat Workshop on Bargaining and Politics, Political Economy in the Chicago Area conference Elections and Electoral Institutions conference in Toulouse, and seminar participants at Paris Empirical Political Economics Seminar, London School of Economics, and University of Warwick for helpful comments. The authors would like to acknowledge the generous cooperation and assistance provided by H.E. Wais Barmak, Tariq Ismati, and Abdul Rahman Ayubi of the Ministry of Rural Rehabilitation of the Government of the Islamic Republic of Afghanistan (MRRD); Ehsan Zia; sta¤ of AfghanAid, C.H.A., InterCooperation, IRC, NPO/RRRAA, Oxfam UK, and People-in-Need; and Philippe Dongier, Susanne Holste, Qazi Azmat Isa, Zishan Karim, Norman Piccioni, and Mio Takada of the World Bank. We thank Anna Belonog and Denis Shishkin for excellent research assistance and Hamidullah Gharibzada, Shahim Kabuli, and Maiwand Siddiqi for excellent oversight of data collection. Data collection was supported by MRRD through the National Solidarity Programme. Additional …nancial and logistical support for the study was provided by the Food and Agriculture Organisation of the United Nations and the World Bank. Fotini Christia acknowledges support from ARO MURI award No. W911NF-12-1-0509. Ruben Enikolopov acknowledges …nancial support from the Ministry of Education and Science of the Russian Federation, grant No. 14.U04.31.0002 and Deutsche Bank Membership at IAS. Andrew Beath was a consultant for the World Bank during the data collection process and a sta¤ member of the World Bank thereafter. y O¢ ce of the Chief Economist for East Asia and the Paci…c, World Bank. abeath@worldbank.org z Department of Political Science, Massachusetts Institute of Technology. cfotini@mit.edu x Kellogg School of Management, Northwestern University, and NBER. g-egorov@kellogg.northwestern.edu { ICREA– IPEG, UPF, Barcelona GSE, and New Economic School. ruben.enikolopov@upf.edu 1 Introduction Ideally, elections for representative bodies result in the selection of those candidates that both perfectly represent the preferences of the electorate and design e¤ective policies. In the real world, however, candidates with the qualities necessary to design e¤ective policies are scarce. As a result, voters face a trade-o¤ between voting for candidates with policy positions closest to their own and candidates who are intrinsically more capable of designing e¤ective policies. How voters choose between these two kinds of candidates impacts the composition of representative bodies and, by extension, the representativeness and quality of public policy. While the trade-o¤ between well- quali…ed candidates and candidates with agreeable policy positions is familiar to almost all who have voted, there is less appreciation of the role potentially served by electoral rules in mediating this trade-o¤ and thereby producing di¤erences in the composition of representative bodies and in the quality of public policy. This paper seeks to identify the e¤ects of di¤erences in electoral systems –and district magni- tude, in particular – on voter behavior, political selection, and policy outcomes. While numerous studies have posited links between electoral rules and policy outcomes (see Cox, 1997; Norris, 2004; Persson and Tabellini, 2000, 2005), explorations of the intermediate e¤ects of voter incentives, polit- ical selection, and the qualities of elected representatives have been limited.1 Moreover, di¢ culties in identifying exogenous sources of variation in electoral systems have generally precluded causal inferences. This paper overcomes these two issues, making a theoretical and empirical contribution to the understanding of the e¤ects of di¤erences in electoral rules and policy outcomes. Speci…cally, the paper combines models of voter choice and legislative bargaining to generate predictions of how di¤erences in district magnitude a¤ect political selection and policy outcomes. These predictions are then tested using data from a unique experiment that induced randomized variation in rules for electing representatives to local development councils across 250 villages in Afghanistan. The paper focuses on two alternate and commonly used electoral rules: the district election system and the at-large election system. District elections, which are used to elect the U.S. House of Representatives and the British House of Commons, are ‘ single-winner systems’ whereby suc- cessful candidates must attain at least a plurality of votes cast in their assigned electorate. District elections thus necessitate the division of the area to be represented into multiple geographic con- stituencies, with voters con…ned to casting votes for candidates competing to represent the voter’ s constituency. At-large elections systems, which are a common means by which parliaments are elected in continental Europe, broaden the scope of the constituency over which candidates are elected. As a result, multiple candidates are elected by each constituency. In the extreme, an at-large election may contain a single constituency spanning the area represented by the particular body (e.g., an entire country in the case of a national parliament). The theoretical model developed in the paper examines how these two di¤erent electoral rules a¤ect the composition of a body of local representatives. The model considers a linear village with a uniform distribution of villagers. The location of a villager’s home corresponds to her ideal point 1 The importance of political selection, though emphasized as equally important as political accountability as early as the Federalist papers (Madison, 1788), has been largely “neglected” (see Besley, 2005, p. 44). Theoretical work by Myerson (1993) and recent empirical work on the quality of politicans by Hirano and Snyder (2014) provide a notable exception. 2 for the location of a public good (e.g., a drinking well). The location of the public good is to be decided by a two-member local body, elected by village residents, through a bargaining process. Under at-large elections, each resident has two votes and can vote for any two candidates.2 Under district elections, the village is split into two geographically-de…ned districts and each citizen can only vote for a candidate residing in the same district. The number of residents capable of making e¤ective policy decisions is assumed to be small and voters thereby trade o¤ candidates’ policy preferences with candidates’ competence.3 The model predicts that in district elections, voters prefer candidates with more extreme policy preferences over more competent candidates, as they expect such candidates to achieve a better outcome in the bargaining game with candidates elected from other districts. In at-large elections, these strategic considerations are less pronounced, so voters are more willing to elect competent candidates. The di¤erence between the two electoral systems is magni…ed in villages with more heterogeneous policy preferences. The predictions of the theoretical model are tested with data from a …eld experiment covering 250 villages across Afghanistan. The experiment induced randomized variation in the electoral rules governing the composition of community development councils that were mandated to make decisions on the type and location of a package of local public goods. Of the 250 villages, 125 were randomly selected to compose development councils by district elections. Per this procedure, each village was divided into several single-member districts, with candidates elected from each district separately and exclusively by villagers residing in that same district. The other 125 villages were assigned to at-large elections. Per this procedure, each village constituted one multi-member district, with villagers facing no restriction on which candidates in the village they could vote for. Development council members were accordingly elected based on the number of votes garnered across the whole village. Under both electoral rules, all villagers were automatically considered candidates. The results of the …eld experiment are consistent with the predictions of the theoretical model. Speci…cally, we …nd that at-large elections result in the election of more competent representatives, as proxied by their level of education.4 The e¤ect is strong in heterogeneous villages (as measured by the divergence of villagers’ex-ante policy preferences, the geographic size of villages, and ethnic composition) and is absent in homogenous villages. We also …nd evidence that district elections result in the election of representatives with more biased preferences over the location of local public goods, as proxied by the location of their houses. We show that the results are not consistent with a number of alternative explanations, such as restrictions on the number of quali…ed candidates that can be elected from the same district or di¤erences in incumbency advantage. This paper contributes to an extensive literature analyzing the e¤ects of electoral systems in 2 This is in line with Cox (1984), who was the …rst to study double member districts formally. The key di¤erence is that in Cox (1984), voters vote sincerely and candidates choose positions strategically. In this paper, voters are strategic, but candidates’positions are …xed and tied to their location of residence. 3 In a richer model, voters would face other trade-o¤s as well. For example, van Weelden (2013) demonstrates a trade-o¤ between aggregating preferences (choosing politicians with preferences of the median voter) and providing them with incentives to exert e¤ort. There, the median voter prefers politicians with biased preferences, because they may be e¤ectively punished without hurting the median voter by electing their polar opposites. Our focus is purely on political selection. 4 Education as a proxy for the quality of politicians has been previously used in Besley, Pande, and Rao (2005), Galasso and Nannicini (2011), Gagliarducci and Nannicini (2013), Folke, Persson, and Rickne (2014). 3 general and district magnitude in particular. Previous studies have observed that the number of candidates increases with district magnitude (Duverger, 1956; Cox, 1997; Norris, 2004) and that proportional representation, which is characterized by high district magnitude, is more favorable to minorities (Lijphart, 2004), although this e¤ect depends on the geographic concentration of minorities (Moser, 2008), their social status (Moser and Scheiner, 2012), as well as the size of the minority (Trebbi, Aghion, and Alesina, 2007). Studies generally …nd that single-member district systems result in better representation of geographically-concentrated interests, with high threshold proportional representation systems favoring geographically-dispersed interests (Ferree, Powell, and Scheiner, 2013).5 The determinants of the capability of elected representatives and government agents have in- vited extensive theoretical investigation (McKelvey and Reizman, 1992; Banks and Sundaram, 1998; Aragones and Palfrey, 2004; Caselli and Morelli, 2004; Mattozzi and Merlo, 2007; Ace- moglu, Egorov, and Sonin, 2010; Egorov and Sonin, 2011). Empirical investigations have further established links between more capable elected representatives and higher intra-party competition (Besley, Folke, Persson, and Rickne, 2013; Folke, Persson, and Rickne, 2014), higher inter-party competition (Banerjee and Pande, 2007), higher wages (Ferraz and Finan, 2011; Dal Bó, Finan, and Rossi, 2013; Gagliarducci and Nannicini, 2013), smaller budgets (Brollo, Nannicini, Perotti, and Tabellini, 2013), and more democratic institutions (Besley and Reynal-Querol, 2011). However, only Myerson (1993) considers how the quality of elected representatives is a¤ected by the type of electoral system, arguing that, when voting is strategic, small district magnitude increases the barriers to candidate entry, which has a negative e¤ect on the capability of canditates.6 To the best of our knowledge, this paper is the …rst to provide experimental evidence on the causal e¤ects of electoral rules. However, as with much other experimental work, internal validity is achieved within a speci…c setting, which raises the issue of the results’external validity. A number of factors, though, suggest that this study’ s conclusions have broader applicability. First, the trade- o¤ described in the theoretical model is general, so the predictions of the model are applicable to other representative bodies, local, regional, and national, across the world.7 Second, the absence of pre-election politicking (such as candidate selection, primaries, or electoral campaigns) enables identi…cation of the e¤ects of electoral rules on political selection and the exploration of underlying mechanisms in a precise manner that would not be possible in a more complex political system. 5 Persson and Tabellini (1999, 2000) and Persson, Tabellini, and Trebbi (2003) argue that proportional systems with high district magnitude lead to higher levels of political rent extraction. Such systems also favor bigger governments and higher levels of redistribution (Iversen and Soskice, 2006; Persson and Tabellini, 2004). Other works that look at the e¤ect of electoral rules on the composition of government spending include Lizzeri and Persico (2001) and Milesi-Ferretti, Perotti and Rostagno (2002). 6 See also Adams (1996), who considers the e¤ect of a constitutional change in Illinois in 1980 on business friendli- ness of state legislators, and Hirano and Snyder (2015) who look at the e¤ect of primaries on the quality of candidates. 7 For example, in the US context, our model would suggest that the electoral success of the Tea Party movement in Republican primaries and general elections in 2010-2012 need not imply that voters became biased to the right or more polarized. Instead, following the global …nancial crisis, de…cit spending increased, and the salience of the taxation-de…cit-spending issue increased. In this context, voters who are only moderately averse to de…cit spending may have decided to elect representatives who are extremely averse to increasing the de…cit, anticipating that their more extreme position will lead, in equilibrium, to a more favorable compromise, such as lower de…cits. Though always present, this e¤ect became more pronounced as the salience of the issue increased. This interpretation is consistent with …ndings in Ansolabehere, Rodden, and Snyder (2006), which suggests that, in the past three decades, growing polarization of politicians has occurred despite stable electorate preferences. 4 Third, in contrast to many …eld experiments conducted in a single region of a country, the 250 villages that form the sample are drawn from …ve diverse regions of Afghanistan that span numerous ethnicities, levels of social capital, economic structures, and openness to democratic practices. Finally, …eld experiments in Afghanistan have already been used in the literature to study corruption (Callen and Long, forthcoming), determinants of risk preferences (Callen, Sprenger, Isaqzadeh, and Long, 2014), the e¤ect of school construction (Burde and Linden, 2013), and women’ s empowerment (Beath, Christia and Enikolopov, 2013). More generally, this paper is part of a growing literature that studies di¤erent countries of the world in detail in order to derive general implications.8 The rest of the paper is organized as follows: Section 2 presents the experimental design; Section 3 describes the theoretical model; Section 4 formulates empirical predictions from the model; Section 5 describes the data; Section 6 details the empirical results; Section 7 discusses the empirical and theoretical …ndings; and Section 8 concludes. The paper contains three web appendices, available at http://www.kellogg.northwestern.edu/faculty/egorov/ftp/ERPS-Appendices.pdf: Appendix A presents extensions of the baseline model and explores the robustness of its predictions; Appendix B lays out all the proofs; and Appendix C o¤ers additional empirical results. 2 Experimental Design We examine the e¤ect of electoral rules on the quality of elected representatives using a …eld exper- iment which randomized variation in the method of council elections in 250 villages in Afghanistan. This intervention was part of an impact evaluation of the National Solidarity Program (NSP) that randomized assignment of not only electoral rules, but also project selection procedures (Beath, Christia and Enikolopov, 2013b) and the program itself (Beath, Christia, and Enikolopov, 2012, 2013a). This section provides further details on NSP (subsection 2.1), describes the variation in electoral rules induced across the 250 villages (subsection 2.3), details the sample and randomization procedures (subsection 2.4), and discusses the timing of the intervention and the data collection process (subsection 2.5). 2.1 National Solidarity Program The National Solidarity Program (NSP) was devised in 2002 by the Government of Afghanistan to deliver services and infrastructure to the country’ s rural population and build representative institutions for village governance. NSP has been implemented in over 32,000 villages in all of Afghanistan’ s 34 provinces and has disbursed over $1.1 billion, making it the largest development program in Afghanistan. The program is structured around two interventions: (i) the creation of an elected Community Development Council (hereafter, “council” ); and (ii) the disbursement of block grants to councils for implementation of village projects. The program is executed by the Ministry of Rural Rehabilitation and Development, facilitated by contracted NGOs, and funded by bilateral and multilateral donors. 8 Such works include Olken (2007), Barron and Olken (2009), and Martinez-Bravo (2014a) on Indonesia; Ferraz and Finan (2008, 2011) on Brazil; Dell (2010) on Peru; Casey, Glennerster, and Miguel (2012) and Acemoglu, Reed, and Robinson (2014) on Sierra Leone, among others. 5 NSP mandates the creation of gender-balanced village development councils through a secret- ballot, universal su¤rage election.9 Once councils are formed, NSP disburses block grants valued at $200 per household, up to a village maximum of $60,000, to fund local development projects,10 with villages required to contribute at least 10 percent of project costs, which they largely do in the form of labor. Projects are selected by the council in consultation with the village community and are ordinarily focused on either the construction or rehabilitation of infrastructure, such as drinking water facilities, irrigation canals, roads and bridges, or electrical generators; or the provision of human capital development, such as training and literacy courses. Overall, the main task of the elected council members is to guide the choice of development projects and then oversee project implementation. In particular, they are responsible for compiling the list of candidate projects;11 preparing budgets for the proposed projects for the approval of the central NSP o¢ ce in Kabul; and implementing the projects (including selecting the relevant contractors and designating the laborers that will work on each project). NSP aspires to provide repeat block grants to participating villages, although villages receive no …rm guarantees of when – or if –they will receive these. The process for conducting follow-up elections for the council is also uncertain. Per NSP rules, villages are supposed to hold re-elections for council positions every four years, but since follow-up elections are not facilitated, it is unclear whether these actually occur.12 2.2 Local Governance in Afghanistan Afghanistan’ s central government has historically lacked the resources to exercise local control or provide public goods in many parts of the country. As a result, local communities developed cus- tomary structures of governance and accountability (Bar…eld, 1984). The foundation of governance in rural Afghanistan is the local jirga or shura, a participatory council that has traditionally man- aged local public goods and adjudicated disputes (Nojumi, Mazurana and Stites, 2004). Shura or jirga members tend to be the elders of the village (Rahmani, 2006), although membership is ordinarily not …xed. Shuras or jirgas generally convene when there is an issue to resolve and reach decisions based on consensus (Boesen, 2004). In addition to shuras or jirgas, villages ordinarily have a headman (termed a malik, arbab, or qariyadar ) –usually a large landowner –who serves as a liaison between the village and the central government (Kakar, 2005). Traditional leadership in rural Afghanistan consists almost exclusively of males, as the principle of purdah - which stipulates that women should generally be hidden from public observation - precludes female involvement in communal gatherings and local governance. The elected development councils established by NSP di¤er from customary governance insti- 9 Note that this is the …rst time that the population is participating in a local election. Prior to that, villages in Afghanistan had only customary local governance structures (Beath, Christia, and Enikolopov, 2013c) 10 The average block grant in the villages included in the sample was approximately $31,000. 11 The projects were selected either through a secret-ballot referendum or at a village meeting. In either case, the village council was responsible for preparing a list of proposed projects and for implementing the selected project (see Beath, Christia and Enikolopov, 2013b, for more information). The project selection procedure assigned to a village was known to the NGOs, but we do not know whether this information was transmitted to the villagers prior to the elections. We check and con…rm that there are no statistically signi…cant di¤erences on the e¤ect of electoral rules between villages with di¤erent project selection procedures. 12 No such re-elections had occurred by early 2012, when the data collection for this evaluation was completed. 6 tutions in the mode of selection and the respective accountability structure. While the elected development councils are composed by a secret ballot, universal su¤rage election, the position of headman is ordinarily inherited or otherwise derived on account of land holdings or other forms of economic authority. Although there is no formal assignment of local governance functions to elected development councils apart from managing NSP-funded projects, their authority in select- ing, implementing, and managing these projects provides them with control over what is for many villages an unprecedented volume of resources. Thus, the elected development council exists in parallel with customary governance structures, but as an institution vested with substantial de facto and de jure authority. 2.3 Electoral Rules Every village resident, whether male or female, aged eighteen or older, who has lived in the village for at least one year was eligible to vote and be elected to the council. NSP rules required that at least 60 percent of eligible voters must cast votes in the election for it to be valid. All eligible villagers were considered candidates and people voted by writing-in the name(s) of their preferred candidates.13 Villagers interested in being elected to the council were prohibited from campaigning in any way for the position. The council had to contain an equal number of male and female mem- bers, with the total council size varying by village size. All villages in the sample were segmented into geographically contiguous districts containing between 5 and 25 families, with each district having its own polling station. A village map with districts and enclosed dwellings was displayed in a public area in the village. Council elections were conducted by secret-ballot according to one of two sets of electoral rules that di¤er primarily in district magnitude:14 District Election: Voters were restricted to casting a ballot for a single candidate, who had to reside in the same district as the voter. In each district, the one male and the one female with the largest number of votes were elected to the council as representatives of their district. Thus, this method represents a single-member, simple plurality election with multiple districts (Cox, 1997), similar to the …rst-past-the-post system. At-large Election: Under this method, voters could cast their ballot for anyone residing in the village. The men and women receiving the most votes across the village were elected as council members. Voters could cast ballots for a maximum of three di¤erent people, who were not ranked.15 The at-large election method represents a multi-member election under a plurality rule with a single district and multiple non-transferable votes. The two main di¤erences from district elections are: (i) higher district magnitude (multiple elected members instead of one) and (iii) number of votes cast (three instead of one). 13 Illiterate villagers could ask NGO workers to write the name for them. This arrangement could potentially violate the secrecy of the vote. However, the results of election monitoring (see below) indicate that 95 percent of voters were con…dent that their vote would remain secret. 14 A detailed guide on the procedures is available at: http://www.nsp-ie.org/sti.html 15 This means the system allows for plumping, but not cummulation (Cox, 1997). Participating NGOs requested the casting of up to three votes in at-large elections as they considered it a high probability that if villagers were accorded only one vote in at-large elections, the number of candidates receiving votes would be fewer than the number of council seats, thereby necessitating multiple rounds of voting which would not be logistically feasible. 7 In all villages, council elections were organized and administered by “social organizers”employed by the NGO contracted to facilitate NSP in the region. Monitoring results from a randomly selected set of 65 villages that held district elections and 66 villages that held at-large elections, including data from the monitors’ 784 polling station reports and interviews administered to 1,675 male voters, indicate that election procedures were professionally executed by the contracted NGOs and that villagers exhibited a good understanding of the di¤erent electoral rules.16 Monitoring results thus con…rm high levels of compliance with the assigned treatment status. 2.4 Sample and Randomization Electoral rules were randomized across 250 villages that formed the treatment group for the random- ized impact evaluation of NSP and were assigned to receive NSP. The villages are evenly split across ten administrative districts in northern, northeastern, eastern, central, and western Afghanistan (see Figure 1). Despite the necessary exclusion of southern areas from the sample due to security concerns, the 10 districts are broadly representative of Afghanistan’ s ethnolinguistic diversity, with …ve predominantly Tajik districts, four predominantly Pashtun districts, one predominantly Hazara district, and two districts with signi…cant populations of Uzbek and Turkmen minorities. The average population in our sample of villages is roughly one thousand people (see Table 1). There is notable variation in the geographic size of villages, with quite a few villages spanning several kilometers. The average distance between the house of a randomly selected survey respondent and the center of the village is about 400 meters, with a standard deviation of more than one kilometer. About 25 percent of villages are ethnically mixed, with the rest being exclusively Pashtun, Tajik, or Hazara (as well as one Turkmen village). The average level of education in the sampled villages is very low, with more than seventy percent of adult male villagers having no formal education and only four percent having …nished high school. An average household consists of about ten people, of which about …ve are children under the age of …fteen. The sample villages are also very poor: only forty-…ve percent of respondents indicate that they never or rarely have problems supplying food for their families.17 Random assignment of electoral rules was made concurrently with the assignment of project selection procedures. Speci…cally, 25 treatment villages in each district were paired to minimize di¤erences in background characteristics within each pair (leaving one village unpaired) and then matched in pairs of pairs to form quadruples.18 Unpaired villages across districts were also grouped into two quadruples (leaving two villages unmatched). Each village within the quadruple (and the two unmatched villages) was then randomly assigned one of the four combinations of council election rules and project selection procedures. This assignment procedure, which essentially represents randomization with strati…cation by 16 A detailed description of the monitoring results can be found at: http://www.nsp-ie.org/reports/CDCE-MR.pdf 17 An assessment of the demographic and economic characteristics of the sample villages reveals few substantive di¤erences with those of a random sample of Afghan villages surveyed by the 2007-08 National Risk and Vulnerability Assessment. See Beath, Christia and Enikolopov (2013d) for more details. 18 These characteristics include village size (based on data collected by Afghanistan’ s Central Statistics O¢ ce) and a set of geographic variables (distance to river, distance to major road, altitude, and average slope). Pairs of pairs were formed by performing the same matching procedure treating each pair as a single village with background characteristics that equal the average of the respective characteristics of the two villages in a pair. 8 Figure 1: Ten sample districts. quadruples, ensures that each village in the sample had an equal probability of being assigned to each of the two electoral rules and that this assignment was orthogonal to the assignment of project selection procedures. To account for strati…cation at the randomization stage, we include quadruple …xed e¤ects in the empirical analysis (Bruhn and McKenzie, 2009). The randomization resulted in a well-balanced set of villages. Table 1 presents a comparison between the two groups of villages with regard to a number of pre-intervention characteristics, which shows that the di¤erences between the two groups never exceed 13 percent of a standard deviation. 2.5 Phasing of Intervention and Data Collection The baseline survey was administered in September 2007, prior to the randomization of election procedures. Council elections occurred between October 2007 and May 2008. Elections in 131 out of 250 villages were monitored by agents of the research team. Project selection occurred between November 2007 and August 2008 and project implementation occurred between April 2008 and September 2011. A …rst follow-up survey (midline) was conducted between May and October of 2009. A second follow-up survey (endline) was conducted between May and October of 2011, at which time 99 percent of projects funded by NSP were complete. 3 Theory In this Section, we introduce a simple model of elections where competent candidates are rare to study voters’choice between electing the most competent candidate versus voting for someone whose political positions they share. The model is deliberately simple (e.g., we make some extreme assumptions such as perfect political segregation and that only one candidate in the village is competent), but in Appendix A, we consider multiple extensions on how electoral rules can a¤ect the competence and political positions of elected candidates and show that our main …ndings are robust. Some features of the model, such as absence of entry costs for candidates or the absence of parties, were inspired by the speci…c setting of the experiment; we discuss the generality of these 9 assumptions in Section 8 (Conclusion). 3.1 Setup The society consists of a continuum of individuals distributed uniformly on a compact S = [ B; B ]. The policy space that these agents care about coincides with the set S .19 We assume that if policy p 2 S is enacted, then an individual with bliss point b 2 S gets the baseline utility u (p; b), which we assume, for simplicity, to be quadratic: u (p; b) = k (p b)2 , where k > 0 measures the importance of the policy issue to the society. In addition to di¤erent bliss points, individuals in the society di¤er by their competence, or education, which may be high or low: a 2 f0; hg, with h > 0. In other words, each citizen i is characterized by a pair (ai ; bi ), where the …rst component is his competence (a for ability ), and the second is his location (b for bliss point ). To study the trade-o¤ between policy position and competence, we assume that almost all individuals are incompetent (ai = 0), except for a …nite number N randomly picked ones, who have high competence ai = h. The results are most transparent when N = 1, which we assume for the rest of the paper; in Appendix A (Subsection A1) we show that the results go through if we allow N to be any number. We assume that the types of all individuals are known to all other individuals.20 We also make a technical assumption that, for any b 2 [ B; B ], there is a citizen i with (ai ; bi ) = (0; b); this assumption that there is an incompetent citizen for any policy position ensures existence of an equilibrium. Policy p is chosen and implemented by a governing body (henceforth “council” ), which is elected by the citizens and from the citizens. We assume that the council consists of two elected individuals (again, this assumption is relaxed in Appendix A in Subsection A4), and both must agree on a policy for it to be chosen. We also assume that the competence of council members increases the quality of policy implementation regardless of the policy. If the two council members have types (al ; bl ) and (ar ; br ) and implement policy p, then individual i will get utility: wi (al ; ar ; p) = al + ar + u (p; bi ) = al + ar k (p bi )2 . To simplify exposition, we assume that council members pick a policy that maximizes their joint utility: p = bl + br 2 . Notice that this policy will be the outcome of a bargaining game with alternating o¤ers (Rubinstein, 1982) or a legislative bargaining game with random recognition (Baron and Ferejohn, 1989; Banks and Duggan, 2000) in the limit where o¤ers are made very frequently.21 We 19 The model could represent a situation where citizens care only about the location of a public good, such as a school or a water well. However, the results of the model naturally extend to a much more general set of environments where an individual’ s preferences are correlated with geographic location. 20 Given the context of the experiment, it is natural to assume that location of villagers’ dwellings and education are observable to fellow villagers. But the theory is applicable in environments where political positions and/or competence are not readily observable by voters, as long as candidates are able to signal their political positions or competence prior to elections. 21 If the o¤ers are not made frequently, p = bl + 2 br is still the expected outcome of the game, but there is some variance, which will result in disutility for the citizens. We consider an explicit bargaining game in Appendix A 10 consider a bargaining game explicitly when we generalize the game to incorporate councils with more than two members and show that the results are robust to alternative assumptions regarding the bargaining process (see Appendix A). Here, slightly abusing notation, we assume that having a council with members (al ; bl ) and (ar ; br ) yields utility: 2 bl + br bl + br wi (al ; bl ; ar ; br ) = al + ar + u ; bi = al + ar k bi . 2 2 We compare two electoral procedures: at-large elections and district elections. In district elections, the society is divided into two districts: left district L, containing individuals with bi < 0, and right district R, containing individuals with bi 0. This division is made according to the location of an individual’ s residence, so that the two districts also di¤er by the policy preferences of their inhabitants.22 Each individual casts a vote for one of the citizens living in his/her district, that is, every individual living in the district is considered a candidate.23 Then, in each district, the individual who received the largest share of votes is elected, and in the case of a draw, a random person among those who received the most votes is chosen. In at-large elections, the entire society comprises a single district, and each individual casts two votes for two (di¤erent) citizens. The two candidates who received the most votes are elected.24 This setup assumes a homogenous population with no gender di¤erentiation; in Appendix A (Subsection A7), we show that the predictions of the model hold if the council is formed from two populations (e.g. men and women) that have the same geographic distribution, which is more in line with the experiment. The strategy of each voter in district elections is therefore (i), the identity of the individual in his district for whom he casts his vote (since only members of the same district may be elected, b (i) 2 [ B; 0) if bi 2 [ B; 0) and b (i) 2 [0; B ] if bi 2 [0; B ]). The strategy of each voter in at-large elections is (i) = ( 1 (i) ; 2 (i)), which corresponds to the (unordered) pair of individuals for whom he votes. All voting decisions are made simultaneously, which gives rise to a coordination problem. We make the following re…nement: De…nition 1 Voting strategies f i g in case of district elections or f i g in case of at-large elections constitute an equilibrium if, for any electoral district (i.e., L or R in the …rst case, or the entire (Subsection A2). 22 The assumption that the two districts perfectly segregate the inhabitants by political preferences is stark, but simpli…es the exposition considerably (and not unrealistic in the context of the experiment). The substantive insights of the model hold as long as the correlation between geographical location and political preferences is above some threshold. The details are available from the authors upon request. 23 Our model of elections falls into the category of citizen-candidate models with costless entry (Besley and Coate, 1998; see also Osborne and Slivinski, 1996, and Besley and Coate, 1997). 24 We will prove that in at-large elections there is a Condorcet winner, that is, a pair of citizens such that there is no other pair that a majority would want to elect. The voting model we selected ensures that the Condorcet winner pair is elected. The assumption that voters cannot cast both their votes for the same candidate is important. If voters could cast both votes for the same candidate, there may be a continuum of election outcomes even if coalitional deviations are allowed, as in De…nition 1 (we show that in Appendix B). We would run into the same problem if each citizen had only one vote. However, if citizens could cast more than two votes, or if they could cast votes for pairs of candidates rather than individuals, the same Condorcet winner pair would always be elected. This is an interesting observation per se: because citizens’preferences in at-large elections are de…ned over pairs of politicians, voting procedures that facilitate extracting this information are more successful in aggregating these preferences. (In district elections, each district has a standard majority voting rule and for any choice by the other district it elects the Condorcet winner.) 11 society S in the second), there is no subset of voters X in this district who would strictly improve the utility of all voters in X by choosing di¤ erent voting strategies. In other words, we re…ne the (otherwise huge) set of Nash equilibria by allowing for deviations by coalitions of voters, but only within a district. For at-large elections, our equilibrium concept coincides with the Strong Nash. 3.2 Analysis Analysis of the game is greatly simpli…ed by the fact that the median voter theorem applies in both at-large and district elections. In district elections, each of the two districts L and R will elect the council member most favored by the median voter in that district, holding the decision of the other district …xed (denote these median voters by mL and mR , respectively). In at-large elections, the median voter of the entire society, mS , will elect the pair of candidates that he/she likes best. These individuals (or this pair of individuals) will, in fact, be the Condorcet winners in their respective districts. These results hold because individual preferences exhibit the single-crossing property: if a citizen i prefers policy p1 to p2 < p1 , then so does a citizen j with bj > bi . The fact that one of the citizens is competent (denote his policy preference by q , so his type is (h; q )) makes the argument just marginally more involved. The proof of Proposition 1 (see Appendix B) …lls in the details. Proposition 1 In both district elections and at-large elections, equilibria exist, and the types of elected politicians are uniquely determined for almost all realizations of q . Moreover: 1. In district elections, the district without the competent citizen elects the most biased individual (with bi = B ), and the district with the competent citizen (h; q ) elects either this citizen or the most biased individual (with bi = B ). 2. In at-large elections, the two elected citizens are the most competent individual (h; q ) and a citizen with the opposite political preferences (0; q ). Proposition 1 implies that the equilibrium concept we use (De…nition 1) is su¢ ciently strong to pick a (generically) unique equilibrium. To build an intuition for which types of citizens get elected, consider at-large elections …rst. The median voter in the whole district, mS , has the bliss point bmS = 0 and his ideal outcome is to elect two council members who negotiate and implement his ideal policy 0, while at the same time making sure that one of the two is competent. This is feasible: he can achieve this ideal outcome by having the competent citizen (h; q ) and his political antipode (0; q ) elected. Next, consider district elections. The reason to elect the most competent citizen if he lives in the district is clear, but what is the rationale to elect the most biased individual? To answer this question, suppose that district L elects a citizen of type (al ; bl ) and consider the best response of the median voter of district R, mR . His ideal policy is B 2 , and if he elects a resident with type (ar ; br ), he would get utility: 2 bl + br B wmR (al ; bl ; ar ; br ) = al + ar k . (1) 2 2 12 The right-hand side of (1) is strictly increasing in br for br B , because bl 0, and thus it reaches its maximum for br = B . In other words, holding competence …xed, the median voter of district R prefers the most biased candidate, and this is true regardless of voting strategies of the citizens in the left district.25 We thus see the following strategic delegation e¤ect: even though the median voter mR likes policies which are close to mR , he/she prefers to elect a biased citizen because the latter would negotiate a better policy; this is similar to delegating bargaining to a more committed type.26 The same e¤ect causes the median voter in the left district, mL , to favor a candidate with bl = B . Of course, it is also possible that the most competent candidate will be chosen over the most biased, and the next proposition tells us exactly when this happens. Proposition 2 In district elections, both districts elect the most biased and incompetent candidates if: r h jq j < q ^ 2B 4 + B2, (2) k where q is the ideal point of the competent citizen. If (2) does not hold, then one district elects the most biased of its residents, and the other elects the competent citizen. The competent citizen is more likely to be elected if: (i) the society is more homogenous, i.e. less polarized in their preferences (B is lower); (ii) competence is more pronounced (h is higher); (iii) policy matters less relative to competence (k is lower). The district without a competent citizen is bound to elect the most biased council member, (0; B ) or (0; B ). The median voter of the other district faces a trade-o¤ between electing the most biased citizen and the competent citizen. Thus, he is more likely to choose competence over policy if competence is more important (h is high and k is low) or if the competent citizen is also biased (q is close to B ). Interestingly, polarization hurts the chances of the competent citizen, and the reason is that high polarization makes the median voter more sensitive to the political preferences of the council member he elects. We can now compare the expected outcomes of at-large elections with those of district elections. Ex ante, the identity of the competent individual is not known, but in expectation the following proposition holds: Proposition 3 In at-large elections, as compared to district elections: h 1. The expected competence of an elected council member is higher (strictly higher if k < 3 2 4B ) and this di¤ erence is increasing in B and k ; 2. The expected polarization (distance between preferences of a council member and the society’ s median voter, normalized by dividing by B ) is strictly lower, and this di¤ erence is increasing in B and k ; 25 This preference for the most biased candidate would not necessarily hold if the distribution of individuals were non-uniform, for example, in the case of non-bounded support. However, the tendency to elect a relatively biased candidate would remain. We maintain the assumption of a uniform distribution for expositional purposes. 26 This e¤ect is similar to strategic polarization in the case of a divided government in Alesina and Rosenthal (2000) and in the case of party competition in Ortuño-Ortín (1997). 13 Figure 2: Types of council members (competent in red/thick, incompetent in blue/thin) as a function of the competent citizen’s ideal point. 3. There is no correlation between preferences and competence of council members in at-large elections, while in district elections, competence and distance from the median voter are neg- atively correlated. These results follow from Propositions 1 and 2 and are shown on Figure 2, which depicts the outcome of elections (types of council members) for di¤erent realizations of q . In at-large elections, the competent individual is always elected, and this is not true in district elections provided that ^ > 0. Interestingly, if h is high or k is low or B is low, the competent individual will be elected q in both cases, and the di¤erence between the two types of elections disappears. The polarization result is easy to see from the following consideration: the two council members elected in at-large elections are as far from the median as the competent one, while in district elections, one or both districts elect individuals who are further from the median than the competent individual. In ^ > 0, then in district elections, the most moderate types will never be elected. Finally, addition, if q in at-large elections, there is no correlation between preferences and the competence of a council member: as one can see from Figure 2, any council member with any political bliss point is equally likely to be competent or incompetent. In contrast, in district elections, the most biased council members are likely to be incompetent, and any council member with a more moderate ideal point is likely to be competent.27 Let us consider the e¤ect of electing a competent candidate on political polarization. At …rst glance, the question seems moot: in at-large elections, a competent candidate is always elected, while in district elections, the other, non-competent council member has extreme bias. Consider, however, the possibility that a village has no competent candidates. In the case of district elections, such a village would elect the two most biased candidates for the same strategic reason as before. 27 The ability of the median voter to choose both council members at once help him achieve the …rst best, but it does not drive the results in Proposition 3, as we show in Appendix A (Subsection A2 shows that at-large elections lead to a worse outcome than district elections if o¤ers are made infrequently and Subsection A6 shows that at-large elections result in more competent council members even if one member is elected at a time. The results are entirely driven by the bargaining over policies by the council members, which is anticipated by voters at the time of elections. 14 In the case of at-large elections, there are multiple equilibria (any two candidates with opposite biases may be elected); however, if one restricts attention to equilibria which are the limits of ones in a game where council members play an explicit alternating bargaining game as in Section A2 in Appendix B (a plausible re…nement), only the equilibrium where two politicians located in the center, with b = 0, will be elected. Formally, we have the following proposition: Proposition 4 In district elections, electing a competent council member in one district does not a¤ ect the political bias of the council member elected in another district. In at-large elections, electing a competent council member increases the political bias of the second council member, provided that the equilibrium that is the limit of equilibria in games with an explicit bargaining between council members is played. Apart from the empirical predictions about competence and polarization, our model has clear welfare implications: Proposition 5 In at-large elections, compared to district elections, the expected utility of every individual is higher, and thus social welfare is higher. In light of Proposition 3, it is not surprising that social welfare is higher, in expectation, in at-large elections. It is more striking that the expected utility of every single individual is higher in at-large elections, provided that the expectation is taken before the location of the competent citizen becomes known. The intuition, however, is simple: the expected policy is Ep = 0 under both proce- dures, and by moving from at-large to district elections, the society makes the policy outcome less certain and runs the risk of electing two incompetent citizens; both e¤ects hurt every citizen equally. 4 Empirical Predictions The theoretical model generates several empirical predictions, which we test using data from the …eld experiment. It should be emphasized that only empirical results for the …rst prediction were obtained before the model was formulated. All the remaining empirical predictions were …rst ob- tained from the model and only then tested empirically. We have tried to identify all the theoretical predictions the data at hand allow us to test. Based on the …rst statement of Proposition 3, we formulate the following empirical prediction: 1. The competence of elected candidates is higher in at-large elections than in district elections. Following Besley, Pande, and Rao (2005), Galasso and Nannicini (2011), Gagliarducci and Nannicini (2013), and Folke, Persson, and Rickne (2014), we use educational attainment as a measure of candidate competence, since there is evidence that the leader’s educational level has a positive e¤ect on governance outcomes. In particular, better educated public o¢ cials are less likely to use power opportunistically (Besley, Pande, and Rao, 2005), are more likely to promote higher economic growth (Besley, Montalvo, and Reynal-Querol, 2010; Congleton and Zhang, 2013), and 15 are more likely to provide higher quality public goods (Martinez-Bravo, 2014b). We also check the robustness of our results to using occupation as a measure of candidate competence.28 The …rst part of Proposition 3 also asserts that the e¤ect of the electoral system on the compe- tence of politicians is stronger in communities with more diverse preferences (higher B ) or where the choice of policy is relatively more important than the politicians’competence (higher k ), whereas 3 2 for communities with su¢ ciently homogenous preferences (where 4 B 0. Under district elections, an average of 7 percent of male council members …nished high school (see Table 2), whereas under at-large elections, the proportion is higher by 4 percentage points (the di¤erence is statistically signi…cant at the 1 percent level; see column 1 in Table 3). Although modest in absolute magnitude, this e¤ect constitutes a 57 percent increase in the share of male council members who …nished high school. Given the overall low level of education of council members and the very low level of education and literacy in rural Afghanistan, the e¤ect is substantively large. Thus, the results lend con…rmation to the …rst empirical prediction of the model. Due to the very low levels of variation in the competence of elected female representatives (which in turn are caused by the very low levels of female education in Afghanistan), the empirical analysis of the e¤ects of electoral rules on education levels focuses on male council members. The- oretical analysis shows that the main predictions of the model regarding male candidates hold in signi…cant. 38 The latter three variables were winsorized at the 95th percentile and logged. 39 The …rst two measures are available for female and male respondents separately. The latter two variables were winsorized at the 95th percentile and logged. 40 Indicators (i) and (iv) are available for female and male respondents separately. 20 the presence of a group with no such variation, but also predicts no e¤ect of the electoral system on the competence of female candidates (see Subsection A7 of Appendix A).41 Consistent with this prediction, the broadening of the sample to include both male and female council members does not substantively change the …nding that at-large elections enhance the competence of elected representatives, but the e¤ect is fully driven by male council members, with no e¤ect for female council members (see Table C1 in Appendix C).42 The second empirical prediction posits that the e¤ect of at-large elections on the competence of elected council members is higher in more heterogeneous villages, while the third empirical prediction posits that the e¤ect of at-large elections is insigni…cant in homogenous villages. To test the second and third empirical predictions of the model, we use the following OLS model: Educationvi = + ALv + ALv Heterogenv + Heterogenv + q + "vi ; (4) where Heterogenv is a measure of heterogeneity of village v and all other variables are the same as in equation (3). The second empirical prediction posits that > 0 and the third one that = 0 (as long as the measure of heterogeneity is normalized to zero in homogenous villages). Consistent with the second empirical prediction, we …nd that the e¤ect of variation in elec- toral rules is signi…cantly stronger in more heterogeneous villages for all measures of heterogeneity (columns 2– 4 in Table 3). Consistent with the third empirical prediction, we …nd no signi…cant e¤ect of variation in electoral rules on the competence of elected candidates in homogenous villages. The results also provide some evidence that increasing heterogeneity is associated with lower com- petence of elected candidates in villages with district elections. Although this result is not based on an exogenous source of variation, it provides evidence that is consistent with Proposition 2 and thus with the hypothesized mechanism for the e¤ect. The …rst two empirical predictions hold –but with smaller coe¢ cients –if we de…ne competence as the completion of middle school rather than high school (see Table C2 in Appendix C), but do not hold if we de…ne it as the completion of primary school. The …rst empirical prediction also holds if competence is de…ned as holding an occupation other than farming (see Table C3 in Appendix C).43 The results for the second empirical prediction are no longer statistically signi…cant if competence is de…ned by occupation, but the signs of the coe¢ cients are aligned with the predictions of the model. The results in Table 3 are also robust to looking at village averages, rather than at individual council members, which indicates that the results are not driven by a 41 The extension of the model in Subsection A7 of Appendix A also predicts that female council members would have more extreme policy preferences if elected by district elections and that the extremity of such preferences would be higher than that of male council members. Unfortunately, we do not have information on the location of households for female council members, so cannot test this prediction. 42 Another reason to focus on male council members is that in Afghan villages, women are traditionally excluded from community-level decision making and, consequently, female council members are not expected to have a signi…cant e¤ect on project selection and implementation. This is con…rmed by Beath, Christia and Enikolopov (2013b), which …nds that the preferences of even the most important female villagers do not have a signi…cant e¤ect on the choice of projects. However, despite their limited role in project-related decisions, female council member have played an important role in solving women-related issues and their election had an important e¤ect on attitudes toward women (Beath. Christia, and Enikolopov, 2013a). However, these considerations are not related to the trade-o¤ analyzed in this paper. 43 Farming is the occupation characterized by the lowest levels of education. Of council members, 53 percent are farmers. The results also hold if we broaden the de…nition to include keepers of livestock and unskilled laborers. 21 few villages with a high fraction of educated council members.44 The fourth empirical prediction posits that district elections will lead to the election of represen- tatives that hold more extreme policy preferences. To test this prediction, we use the following OLS model by estimating a model similar to (3) with the outcome variable provided by the logarithm of the distance between the house of an elected council member and the village center: Distancevi = + ALv + AverageDistancev + q + "vi ; (5) where Distancevi is the logarithm of the distance between the house of council member i and the center of the village, AverageDistancev is the average distance between houses of baseline survey respondents in village v (which re‡ ects the size of the village), and other variables are the same as in the previous equations. The fourth empirical prediction posits that < 0. Consistent with the fourth empirical prediction, we …nd that the distance between the homes of elected o¢ cials and the center of the respective village is smaller in at-large elections (see columns 1– 2 in Table 4).45 The magnitude of the e¤ect is sizable, with the distance being approximately 25 percent smaller in villages with at-large elections. The …fth empirical prediction posits that if a competent candidate is elected, the distance be- tween the homes of other elected representatives and the center of the respective village is smaller in at-large elections, whereas there is no such relationship in district elections. We test this prediction by estimating the following model on the subsample of elected candidates who are not competent: Distancevi = + ALv + ALv Competent + DSv Competentv + AverageDistancev + q + "vi ; (6) where Competentv is a dummy variable that equals one if at least one competent candidate is elected in village v and zero otherwise, DSv is a dummy variable for district elections, and all other variables are the same as in the previous equations. The empirical prediction posits that > 0 and = 0. Since the supply of competent candidates depends on the educational level of the village population, we also control for the educational attainment of baseline survey respondents. The …ndings are consistent with the …fth empirical prediction. The distance between the homes of elected representatives and the center of the respective village is indeed signi…cantly higher in villages that have elected educated representatives (see columns 3– 5 in Table 4). The e¤ect in at-large elections is highly statistically signi…cant and is robust to controlling for village size and villagers’educational attainment. The e¤ect is large in magnitude, suggesting that, controlling for the village’s geographic size, the distance to the homes of elected representatives in such villages was lower by one half.46 In district elections, the distance to the homes of elected representatives is higher in villages that elected educated representatives, but the e¤ect loses statistical signi…cance once we control for the village’ s geographic size. Thus, the results provide support for the …fth prediction of the model. The sixth empirical prediction posits that, in district elections, there will be a negative corre- 44 These results are available upon request. 45 Not surprisingly, the distance between the homes of elected o¢ cials and village centers is larger in villages with greater distance between the houses of baseline respondents. 46 The number is obtained by taking the exponent of the coe¢ cient in column 4 of Table 4. 22 lation between a representative’ s competence and the extremity of his/her policy preferences. We test this prediction by estimating the following model: Educationvi = + ' ALv + ALv Distancevi + Distancevi + q + "vi ; (7) where all the variables are the same as in the previous equations. The empirical prediction posits that < 0 and + = 0. The …ndings indicate that, in district elections, there is a negative correlation between a repre- sentative’ s education and the distance between his house and the center of the village, although it becomes statistically signi…cant only when controlling for village size (see Table 5). In addition, the di¤erence between the e¤ects of distance in villages with at-large and district elections is positive and statistically signi…cant, con…rming that > 0. We also …nd that, in at-large elections, the correlation between a representative’ s educational attainment and the distance between the repre- sentative’ s house and the center of the village is positive (rather than zero, as implied by the model), but it loses statistical signi…cance once we control for the geographic size of the village.47 Thus, the results provide support for the sixth empirical prediction, although only in the speci…cation that controls for the village’ s geograpic size. The seventh empirical prediction posits that the speed of project implementation will be higher in at-large elections. To test this hypothesis, we estimate the following OLS model: P rogressjv = + ALv + q + "jv ; (8) where P rogressjv is either time to project start or time to project completion and all other variables are the same as in (3). According to the seventh empirical prediction > 0. The results indicate that, in villages with at-large elections, project implementation started earlier and, as a result, was completed faster (see Table 6). The results hold if we control for the number, cost, and type of projects and thus do not appear to be driven by an indirect e¤ect of electoral rules on the type of projects selected.48 The magnitude of the e¤ect is substantial, with projects starting and …nishing more than a month earlier in villages with at-large elections. A comparison of the distribution of dates for project commencement and completion by type of elections (Figures C1 and C2 in Appendix C) indicates that the di¤erence in the speed of project completion re‡ ects a shift in the distribution, rather than the e¤ect of a few outliers. The results are thus consistent with the seventh empirical prediction. The eighth and …nal empirical prediction posits that the e¤ectiveness of development projects, improvements in general economic welfare, and increases in support for local leaders and democratic processes will be higher under at-large elections. To test this prediction, we estimate the following model: Yvit = + 1 ALv 1t + 2 ALv 2t + q 1t + q 2t + "vit , 47 We also …nd that, for the random sample of villagers in the baseline survey, there is no signi…cant relationship between their education and the distance from their houses to the center of their village. 48 There is some evidence that the number of projects is smaller and the cost of projects is higher if councils are populated by more educated representatives. This potentially suggests that more competent candidates are able to implement more complex projects and, in turn, may explain why there is no statistically signi…cant e¤ect of electoral rules on the duration of projects. 23 where Yvit is the outcome of interest for individual i in village v at time t 2 f1; 2g which corresponds to the midline and endline surveys respectively; 1t and 2t are dummies for the midline and endline surveys; all other variables are the same as in (3). Thus, the coe¢ cients 1 and 2 measure the e¤ect of at-large elections on the outcomes of interest at the time of the midline and endline surveys, respectively. Standard errors are clustered at the village level. Estimates using data from the endline survey, which was administered approximately 4 years after the election of the village developments councils and 1– 2 years following the completion of NSP-funded projects, indicate that villages assigned at-large elections exhibited lower levels of diarrhea among children (see Panel A of Table 7) and thus had water sources more likely to be free of contaminants.49 We also observe that at-large elections induced a marginally signi…cant increase in revenue derived from the harvest prior to the endline, which may re‡ ect more successful implementation of irrigation projects. The signs of coe¢ cients for other indicators of the e¤ectiveness of development projects (reported water quality, hours of electricity available, yield of the most recent harvest) are consistent with at-large elections producing better project-related outcomes, although none of these attain conventional levels of statistical signi…cance. Overall, there is weak evidence that the quality of implemented development projects was higher in villages assigned to at-large elections and no evidence that the faster implementation of projects in at-large villages came at the sacri…ce of project quality. Estimates using data from the midline survey, which was administered approximately 2 years after council elections, indicate that female respondents in villages assigned at-large elections were more likely to express positive perceptions of the economic situation and were more likely to attribute positive economic changes to local leaders (see Panel B of Table 7). However, there is no such e¤ect for male respondents at midline or endline or for female respondents at endline. Data from the endline survey also indicates that there is a marginally signi…cant positive e¤ect of at-large elections on household expenditures. The e¤ect of at-large elections on household income at endline is also positive, although not statistically signi…cant. Overall, there is weak evidence that at-large elections improve general economic welfare. Estimates using data from the endline survey indicate that, in villages assigned at-large elections, male respondents were less likely to express a desire to change decisions made by village leaders and were more supportive of using democratic procedures to select the President of Afghanistan (see Panel C of Table 7). Data from the midline survey indicate that male villagers in at-large villages were more likely to support the election of the village headman, but that the e¤ect decreases in magnitude and loses statistical signi…cance at endline. The results for other indicators are not statistically signi…cant, although the signs of all coe¢ cients for measures of support among male villagers for local leaders and democratic processes are consistent with the hypothesis that at-large elections improve villager attitudes. Overall, there is weak evidence that at-large elections resulted in improved attitudes toward village leaders and democratic processes. On the whole, estimates using measures of villager welfare and opinion from the midline and endline surveys are weakly consistent with the eighth empirical prediction. However, as the data 49 Note that not all villages implemented drinking water projects. However, restricting the sample only to villages that implemented drinking water projects would potentially confound the estimation of the e¤ects of electoral rules given the endogeneity of project choice. 24 used to produce these estimates are subject to the imprecision associated with household surveys and describe processes and outcomes over which representatives of the village development councils potentially have limited in‡ uence, these estimates are likely attenuated by measurement error. At a minimum, the results show that the faster speed by which projects in villages assigned to at-large elections are implemented does not come at the cost of lesser quality projects. 7 Discussion According to our theoretical model, electoral rules a¤ect the competence of elected representatives by changing voter incentives to support candidates with more extreme policy preferences over more quali…ed candidates. The results from the …eld experiment are consistent with the predictions of the theoretical model. In particular, we …nd that the competence of elected representatives is higher in villages assigned to at-large elections and that this di¤erence is higher in more heterogeneous villages. There is also evidence that, in more heterogeneous villages, district elections lead to the election of representatives with more extreme policy preferences and lower levels of competence. There are several possible alternative explanations for the positive e¤ect of at-large elections on the quality of candidates. First, the e¤ect may be a mechanical product of the geographic restrictions imposed by district elections on candidate selection, which preclude the election of multiple competent candidates residing in the same electoral district. The viability of this expla- nation is explored by examining the distribution of elected council members across districts. As prescribed, villages with district elections had exactly one male representative elected from each district. Although at-large elections imposed no formal restrictions on the geographic distribution of representatives, the distribution was similar to that in villages with district elections. Speci…- cally, in villages with at-large elections, 93 percent of districts had a person residing in that district elected to the council. Only 37 out of 125 at-large villages had at least one district which did not have a resident council member and, of these, 25 villages had only one district that did not have a resident representative. Thus, the e¤ect of at-large elections on the probability of a district not having a resident council member was very small. To further test the viability of the explanation, we exclude from the sample representatives from those districts that, in at-large elections, had more than one representative elected to the council.50 Thus, we look only at the quality of candidates for whom the geographical restriction was not bind- ing. While the results are to be interpreted with caution given the endogeneity of the restriction, we …nd that the sample restriction has no e¤ect on estimates obtained by the benchmark speci…cation (Table C4 in Appendix C). Finally, the restriction on candidates’ residence in district elections should matter more in smaller villages. However, empirical results indicate that the e¤ect of elec- toral rules on the quality of candidates is stronger in larger, rather than smaller, villages. Overall, these results indicate that the e¤ect of electoral rules on the quality of elected representatives is not a¤ected by the geographic restriction on representatives imposed by district elections. Another alternative explanation is that the larger district magnitude associated with at-large elections makes it harder for pre-existing elites to coordinate voting, which in turn reduces 50 Ideally, we would exclude districts with more than one candidate of high quality, but since all villagers are considered candidates and we do not have a full census of the villages, we cannot implement this strategy. 25 their electoral advantage. Assuming that members of pre-existing elites are less quali…ed than potential challengers, this would increase the quality of elected representatives. The viability of this explanation is explored by examining the share of council members who were members of the de facto elite, for which data was collected from villagers in the baseline survey administered prior to the council elections.51 The results indicate that electoral rules have no signi…cant e¤ect on the share of council members that also belong to the group of pre-existing elites (see Table C5 in Appendix C). In fact, one measure provides a marginally signi…cant positive e¤ect of at-large elections on the share of council members who were named as members of the pre-existing elite. Similar results are obtained if we look at the proportion of pre-existing elite members who were subsequently elected to the council, with no signi…cant di¤erence between villages that used alternative electoral rules. Overall, the results indicate that electoral rules have no signi…cant e¤ect on the electoral advantage of pre-existing elites. Yet another potential explanation is that educational attainment serves as a proxy for candidate competence if voters do not have precise information. In district elections, the size of the districts is smaller and voters are likely to have better information about candidates, so they do not need to rely on proxies such as educational attainment. In at-large elections, by contrast, the larger size of districts leads to greater reliance on educational attainment as a proxy for candidate compe- tence. While this interpretation could potentially explain the observation that at-large elections are associated with more educated representatives, it does a poor job of explaining why the results are stronger in more heterogeneous villages and why the educational attainment of elected repre- sentatives is on average lower in more heterogeneous villages. Speci…cally, in more heterogeneous villages, voters are likely to have less information on candidates, so according to this explanation voters should rely on formal education more. In addition, this explanation cannot account for the results on the locations of the elected representatives’homes. Di¤erences in the competence of elected representatives may also re‡ ect di¤erences in barriers to entry, which are higher in electoral systems with small district magnitudes (Myerson, 1993). However, this explanation is not relevant in our context, as all citizens are considered candidates and there are no barriers to entry. The observed correlation between at-large elections and representatives’competence may also be driven by the advantage given by at-large elections to competent candidates for reasons other than voting decisions by citizens. For instance, if candidates garner support through rallies (public speeches) and bribes (vote-buying) during the campaign and public speeches are more important in larger districts due to economies of scale, competent candidates will be advantaged in at-large elections if they have a comparative advantage in public speaking (and a comparative disadvantage at vote-buying).52 This explanation, while attractive theoretically, cannot explain the results of the …eld experiment, as political campaigning was forbidden and election monitoring indicated no evidence of vote-buying. In addition, this story does not imply any correspondence between electoral rules and the location of representatives’ houses, which contradicts the experimental results. 51 The share of council members that were members of the pre-existing elite, even if we use the most inclusive de…nition of pre-existing elites, is 39 percent, which indicates that pre-existing elites do not dominate the elected council. 52 This explanation relies on a strong assumption that higher quality candidates are relatively better at speaking and not at bribing, which may or may not be true. 26 8 Conclusion This paper examines the e¤ect of electoral rules on voting behavior and by extension on the com- position of representative bodies and the quality of policy. The paper speci…cally compares two alternative electoral rules: district elections, with multiple single-member districts, and at-large elections, with a single multi-member district. Employing a theoretical model which builds on a citizen-candidate model with free entry of candidates and which assumes that voters value both competence and preference representation in electoral candidates, the paper posits that elections with a single multi-member district will result in the election of more competent representatives, with this e¤ect becoming stronger in more heterogeneous communities. Empirical results from a …eld experiment conducted across 250 villages in Afghanistan are consistent with these predictions. Ancillary evidence suggests that these results are driven by strategic voting behavior informed by a voter’ s anticipation of policy bargaining and not by alternative mechanisms such as incumbency advantage or comparative advantages in political campaigning. The paper’ s theoretical model imposes a series of simplifying assumptions. For instance, in order to enable a focus on the choice of voters over an exogenous set of candidates, the model assumes that candidates face zero entry costs.53 In addition, the model assumes only a portion of candidates are competent to hold o¢ ce.54 Furthermore, the experiment used to test the model” s predictions was conducted in a special institutional context that di¤ered from those in which elections are ordinarily held, particularly in the absence of political parties, campaigns, or canvassing; the limited role of long-term considerations for candidates; and the limited mandate of the representative body. These assumptions and contextual peculiarities notwithstanding, the results of the paper provide a number of general implications. First, the paper establishes the general result that electoral rules, by structuring the strategic incentives facing voters, condition voting behavior. The theoretical model demonstrates that ra- tional voters will behave di¤erently according to the district magnitude of the election in which they are voting and the results of the experiment align with how voters should behave given the incentives presented to them by electoral rules. Second, the paper shows that strategic voting results in the selection of candidates with relatively extreme policy preferences. The theoretical model suggests that voters anticipate bargaining over policy and seek to elect candidates with relatively extreme policy preferences in order to ensure a policy outcome that is more likely to be favorable to the voter’ s speci…c interests. The results of the experiment con…rm that voters engage in such strategic behavior. Third, the paper demonstrates that voter preferences for candidates with relatively extreme policy preferences increase with heterogeneity over policy preferences of voters across electoral districts. That is, in cases where there is high dispersion of policy preferences among voters, voters anticipate other voters’ more extreme preferences and the deleterious consequences of the achievement of their policy ideal point, resulting in a preference for candidates with more extreme policy preferences. 53 The endogeneity of candidate selection is addressed in the existing literature (Myerson, 1993). 54 The assumption of a scarcity of competent individuals is consistent with previous work positing negative selection of candidates (Caselli and Morelli, 2004; Mattozzi and Merlo, 2007). Our …nding that strategic voting reduces the election of competent individuals suggests an additional mechanism for negative selection. 27 Fourth, the paper shows that elections with multiple single-member districts cause voters to focus on geographically-correlated aspects of policy preferences, such as the location of local public goods, and to elect candidates that can be expected to best represent the preferences of the district. In contrast, elections with single multi-member districts reduce the salience for voters of geographically-correlated policy preferences and cause voters to focus on other candidate characteristics. Although the model assumes full segregation in terms of political preferences across political districts, qualitatively the results hold as long as political preferences are geographically-correlated within electoral districts (see also Footnote 22). Empirical evidence suggests that political preferences are indeed highly geographically correlated in many countries around the world (Rodden, 2010).55 Fifth and …nally, the theoretical model and empirical …ndings support the general implication that single multi-member district elections better enable the realization of voters’full set of prefer- ences, while multiple single-member district elections unduly magnify the importance of preferences over geographically-correlated policies. A limitation of our results, however, is the focus on policy aspects of representation. In addition to the task of bargaining over policy and facilitating policy implementation, elected representatives are also often tasked with providing constituents with various intangible services, such as assisting them in the navigation of bureaucratic processes and advocating on behalf of individual constituents or groups of constituents. It is certainly possible and, indeed, probable that the structure of direct representation provided by multiple single-member district elections may be more conducive to the provision of such services, as compared to the less direct systems of representation provided by single multi-member elections. To the extent to which citizens derive utility from receiving such services and to the extent to which there is ambiguity in the e¤ects of the two election systems considered here on the provision of such services, the results of the paper cannot speak to welfare implications of the two election systems that extend beyond the policy function of elected representatives. An important area for future research is thus to extend the analysis to consider these aspects of representation. In addition, future research can shed light on how the mechanisms outlined in this paper play out in settings with endogenous candidate entry by candidates, repeated elections, active political campaigns, and political parties playing the role of intermediaries. 55 Empirical evidence suggests that political preferences are highly geographically correlated in many countries around the world (Rodden, 2010). 28 References Acemoglu, Daron, Georgy Egorov, and Konstantin Sonin (2010) “Political Selection and Persistence of Bad Governments,” Quarterly Journal of Economics, 125(4), 1511-1575. Acemoglu, Daron, Tristan Reed, and James A. Robinson (2014) “Chiefs: Elite Control of Civil society and Economic Development in Sierra Leone.” Journal of Political Economy, 122(2): 319- 368. Adams, Greg (1996) “Legislative E¤ects of Single-Member vs. Multi-Member Districts,” American Journal of Political Science, 40(1), 129-144. Alesina, Alberto, Reza Baqir, and William Easterly (1999) “Public Goods and Ethnic Divi- sions,” Quarterly Journal of Economics, 114(4): 1243-1284. Alesina, Alberto, and Howard Rosenthal (2000) “Polarized Platforms and Moderate Policies with Checks and Balances.” Journal of Public Economics, 75(1), 1-20. Angrist, Joshua D., and Jï¿œrn-Ste¤en Pischke (2008) Mostly Harmless Econometrics: An Em- piricist’s Companion. Princeton University Press. Ansolabehere, Stephen, Jonathan Rodden, and James M. Snyder (2006) “Purple America,” Journal of Economic Perspectives, 20(2): 97-118. Aragones, Enriqueta, and Thomas Palfrey (2004) The E¤ect of Candidate Quality on Electoral Equilibrium: An Experimental Study, American Political Science Review 98 (1): 77-90. Austen-Smith, David, and Je¤rey S. Banks (2005) Positive Political Theory II Strategy and Struc- ture, The University of Michigan Press, Ann Arbor. Banerjee, Abhijit, and Rohini Pande (2007) “Parochial Politics: Ethnic Preferences and Politician Corruption,” CEPR Discussion Paper No. DP6381. Banks, Je¤rey S., and John Duggan (2000) “A Bargaining Model of Collective Choice,” American Political Science Review, 94(1): 73-88. Banks, Je¤rey S., and Rangarajan K. Sundaram (1998) “Optimal Retention in Principal/Agent Models,” Journal of Economic Theory, 82(2): 293-323. Bar…eld, Thomas Je¤erson (1984) Weak Links on a Rusty Chain: Structural Weaknesses in Afghanistan’ s Provincial Government Administration. Berkeley, CA: Institute of International Studies, University of California, Berkeley. Baron, David. P., and Ferejohn, John.A. (1989) “Bargaining in legislatures,” American Political Science Review, 83(4): 1181-1206. Barron, Patrick, and Benjamin Olken (2009) “The Simple Economics of Extortion: Evidence from Trucking in Aceh,” Journal of Political Economy 117 (3):. 417-252, Beath, Andrew, Fotini Christia and Ruben Enikolopov (2012) “Winning Hearts and Minds through Development Aid: Evidence from a Field Experiment in Afghanistan,” MIT Working Paper. Beath, Andrew, Fotini Christia and Ruben Enikolopov (2013a). “Empowering Women: Evidence from a Field Experiment in Afghanistan,” American Political Science Review, 107(3): 540-557. Beath, Andrew, Fotini Christia and Ruben Enikolopov (2013b) “Direct Democracy and Resource Allocation: Experimental Evidence from Afghanistan,” MIT Working Paper. Beath, Andrew, Fotini Christia and Ruben Enikolopov (2013c) “Do Elected Councils Improve Governance? Experimental Evidence on Local Institutions in Afghanistan,” MIT Working Paper. 29 Besley, Timothy (2005) “Political Selection,” Journal of Economic Perspectives, 19(3): 43-60. Besley, Timothy, Rohini Pande, and Vijayendra Rao (2005) “Participatory Democracy in Action: Survey Evidence from India,” Journal of the European Economics Association Papers and Proceed- ings, 3(2-3): 648-657. Besley, Timothy and Stephen Coate (1997) “An Economic Model of Representative Democracy,” Quarterly Journal of Economics, 112(1), 85-114. Besley, Timothy and Stephen Coate (1998) “Sources of Ine¢ ciency in a Representative Democracy: A Dynamic Analysis,” American Economic Review, 88(1): 139-56. Besley, Timothy, Olle Folke, Torsten Persson, and Johanna Rickne (2013) “Gender quotas and the crisis of the mediocre man: Theory and evidence from Sweden.” Working paper. Besley, Timothy, Jose Montalvo and Marta Reynal-Querol (2010) “Do educated leaders matter?” Economic Journal, 121, 205-227. Besley, Timothy, Rohini Pande, and Vijayendra Rao (2005) “Political Selection and the Quality of Government: Evidence from South India.” London School of Economics and Political Science, Working Paper. Besley, Timothy, and Marta Reynal-Querol (2011) “Do democracies select more educated leaders?” American Political Science Review, 105(3). Boesen Inger W. (2004) “From Subjects to Citizens: Local Participation in the National Solidarity Programme,” AREU Working Paper Series. Brollo, Fernanda, Tommaso Nannicini, Roberto Perotti, and Guido Tabellini (2013) “The Political Resource Curse,” American Economic Review, 103(5): 1759-96. Bruhn, Miriam, and David McKenzie (2009) “In Pursuit of Balance: Randomization in Practice in Development Field Experiments,” American Economic Journal: Applied Economics, 1(4): 200-232. Burde, Dana, and Leigh L. Linden (2013) “Bringing Education to Afghan Girls: A Randomized Controlled Trial of Village-Based Schools.“ American Economic Journal: Applied Economics, 5(3): 27-40. Callen, Michael and James D. Long (forthcoming) “Institutional Corruption and Election Fraud: Evidence from a Field Experiment in Afghanistan,” American Economic Review. Callen, Michael, Charles Sprenger, Mohammad Isaqzadeh, and James D. Long (2014) “Violence and Risk Preference: Experimental Evidence from Afghanistan.” American Economic Review, 104(1): 123-48. Caselli, Francesco, and Massimo Morelli (2004) “Bad Politicians,” Journal of Public Economics, 88(3-4), 759-782. Casey, Katherine, Rachel Glennerster, and Edward Miguel (2012) “Reshaping Institutions: Evi- dence on Aid Impacts Using a Preanalysis Plan,” Quarterly Journal of Economics, 127 (4): 1755- 1812. Congleton, Roger D., and Yongjing Zhang (2013) “Is it all about competence? The human capital of US presidents and economic performance.“ Constitutional Political Economy 24(2): 108-124. Cox, Gary W. (1984) “Electoral Equilibrium in Double Member Districts,” Public Choice, 44: 443-451. Cox, Gary W. (1997) Making Votes Count: Strategic Coordination in the World’ s Electoral Systems. New York: Cambridge University Press. 30 Dal Bó, Ernesto, Frederico Finan, and Martï¿œn A. Rossi (2013) “Strengthening State Capabilities: The Role of Financial Incentives in the Call to Public Service.“ Quarterly Journal of Economics 128(3): 1169-1218. Dell, Melissa (2012) “The persistent e¤ects of Peru’s mining mita.“ Econometrica 78 (6): 1863-1903. Duverger, Maurice (1954) Political Parties: Their Organization and Activity in the Modern State. Wiley, New York. Egorov, Georgy, and Konstantin Sonin (2011) “Dictators and Their Viziers: Endogenizing the Loyalty-Competence Trade-o¤,” Journal of the European Economic Association, 9(5): 903-930. Ferraz, Claudio and Frederico Finan (2008) “Exposing Corrupt Politicians: The E¤ect of Brazil’ s Publicly Released Audits on Electoral Outcomes,” Quarterly Journal of Economics, 123(2): 703- 745. Ferraz, Claudio and Frederico Finan (2011a) “Electoral Accountability and Corruption in Local Governments: Evidence from Audit Reports,” American Economic Review, 101(4): 1274-1311. Ferraz, Claudio and Frederico Finan (2011b) “Motivating Politicians: The Impacts of Monetary Incentives on Quality and Performance,” Working paper. Ferree, Karen E., Bingham G. Powell, Jr., and Ethan Scheiner (2013) “How Context Shapes the E¤ects of Electoral Rules,” in APSA Report of the Task Force on Electoral Rules and Democratic Governance. Folke, Olle, Torsten Persson, and Johanna Rickne (2014) “Preferential Voting, Accountability and Promotions into Political Power: Evidence from Sweden.” Working paper. Gagliarducci, Stefano and Tommaso Nannicini (2013) “Do Better Paid Politicians Perform Better? Disentangling Incentives from Selection,” Journal of the European Economic Association, 11, 369- 398. Galasso, Vincenzo, and Tommaso Nannicini (2011) American Political Science Review, “Competing on good politicians.“ 105(1): 79-99. Hirano, Shigeo and James M. Snyder Jr. (2014), “Primary Elections and the Quality of Elected O¢ cials,” Quarterly Journal of Political Science, 9 (4): 473-500 Iversen, Torben, and David Soskice (2006) “Electoral Institutions and the Politics of Coalitions: Why Some Democracies Redistribute More than Others,” American Political Science Review, 100(2), 165-181. Kakar, Palwasha (2005) “Fine-Tuning the NSP: Discussions of Problem‘ s and Solutions with Fa- cilitating Partners,” AREU Working Paper. Lijphart, Arend (2004) “Constitutional Design for Divided Societies,” Journal of Democracy, 15: 96-109. Lizzeri, Alessandro, and Nicola Persico (2001) “The Provision of Public Goods under Alternative Electoral Incentives,” American Economic Review, 91: 225-245. Madison, James (1788, 1961). The Federalist Papers: A Collection of Essays in Support of the Constitution of the United States. New York: Doubleday. Martinez-Bravo, Monica (2014a) “The Role of Local O¢ cials in New Democracies: Evidence from Indonesia.“ American Economic Review, 104 (4): 1244-1287. Martinez-Bravo, Monica (2014b) “Educate to Lead? The E¤ect of Politicians’Education on Public Good Provision: Evidence from Java,” working paper. 31 Mattozzi, Andrea, and Antonio Merlo (2007) “Mediocracy,” NBER Working Paper 12920. Mattozzi, Andrea, and Antonio Merlo (2008) “Political Careers or Career Politicians?” Journal of Public Economics, 92: 597-608. Mattozzi, Andrea, and Erik Snowberg (2014) “The Right Type of Legislator,” mimeo. Milesi-Ferretti, Gian Maria, Roberto Perotti and Massimo Rostagno (2002) “Electoral Systems and the Composition of Public Spending.” Quarterly Journal of Economics, 117, 609-657. McKelvey, Richard, and Raymond Reizman (1992) “Seniority in Legislatures,” American Political Science Review, 86, 597-608. Moser, Robert G. (2008) “Electoral Systems and the Representation of Ethnic Minorities: Evidence from Russia,” Comparative Politics, 40(3): 273-292. Moser, Robert G., and Ethan Scheiner (2012) Electoral Systems and Political Context: How the E¤ ects of Rules Vary Across New and Established Democracies. New York: Cambridge University Press. Myerson, Roger (1993), “E¤ectiveness of Electoral Systems for Reducing Government Corruption: A Game-Theoretic Analysis,” Games and Economic Behavior, 5: 118-132. Nojumi, Neamat, Dyan E. Mazurana, and Elizabeth Stites (2004) Afghanistan’ s Systems of Justice: Formal, Traditional, and Customary. Working Paper. Norris, Pippa (2004) Electoral Engineering: Voting Rules and Political Behavior. New York: Cam- bridge University Press Olken, Benjamin (2007) “Monitoring Corruption: Evidence from a Field Experiment in Indonesia,” Journal of Political Economy, 115 (2): 200-249 Ortuño-Ortín, Ignacio (1997) “A Spatial Model of Political Competition and Proportional Repre- sentation.” Social Choice and Welfare 14(3), 427-438. Osborne, Martin, and Al Slivinski (1996) “A Model of Political Competition with Citizen- Candidates, Quarterly Journal of Economics, 111(1), 65-96. Pande, Rohini (2013) “Can informed voters enforce better governance? Experiments in low-income democracies,” Annual Review of Economics 3(1): 215-237. Persson, Torsten, and Guido Tabellini (1999) “The Size and Scope of Government: Comparative Politics with Rational Politicians, 1998 Alfred Marshall Lecture,” European Economic Review 43, 699-735. Persson, Torsten, and Guido Tabellini (2000) Political economics: Explaining economic policy. Cambridge, MA: MIT Press. Persson, Torsten, and Guido Tabellini (2004) “Constitutional Rules and Fiscal Policy Outcomes,” American Economic Review, 94, 25-45. Persson, Torsten, and Guido Tabellini (2005) The Economic E¤ ects of Constitutions. Cambridge, MA: MIT Press. Persson, Torsten, Guido Tabellini, and Francesco Trebbi (2003) “Electoral rules and corruption,” Journal of the European Economic Association, 1(4), 958-989. Rahmani, Ahmad Idrees (2006) The Role of Religious Institutions in Community Governance Af- fairs: How are Communities Governed Beyond the District Level? Budapest, Hungary: Open Society Institute, Central European University Center for Policy Studies. 32 Rodden, Jonathan (2010) “The Geographic Distribution of Political Preferences,” Annual Review of Political Science, 13: 297-340. Rubinstein, Ariel (1982) “Perfect equilibrium in a bargaining model,” Econometrica 50(1): 97-109. Trebbi, Francesco, Philippe Aghion & Alberto Alesina (2007) “Electoral Rules and Minority Rep- resentation in U.S. Cities,” Quarterly Journal of Economics 128(1): 325-358. Van Weelden, Richard (2013) “Candidates, Credibility, and Re-election Incentives,” Review of Economic Studies, 80(4): 1622-1651. 33 Table 1. Balance of Pre-treatment Covariates Standard Deviation District Villages At-large Villages (Full Sample) (Full Sample) Standardized Observation Difference Mean in Mean in Mean Number of Households in Village 118.43 115.71 2264 121.70 115.08 0.06 Household Members 9.79 5.00 2374 9.57 10.02 0.09 Household Members Under 15 Years 4.58 2.76 2374 4.50 4.66 0.06 Distance in Meters from Respondent’s House to 402 1011 2078 351 455 0.10 Village Center Primary Source of Household Income is 0.66 0.47 2360 0.67 0.64 0.07 Agriculture Age of Male Head-of-Household Respondent 43.81 13.30 2336 43.97 43.65 0.02 Male Head-of-Household Respondent Has No 0.71 0.45 2387 0.73 0.69 0.09 Formal Education Male Head-of-Household Respondent Finished 0.07 0.26 2387 0.07 0.08 0.02 Middle School Male Head-of-Household Respondent Finished 0.04 0.20 2387 0.04 0.05 0.03 High School First Language of Male Head-of-Household 0.70 0.46 2387 0.72 0.69 0.05 Respondent is Dari Village is Ethnically Mixed 0.24 0.43 250 0.21 0.26 0.13 Household Never or Rarely Faces Food Shortages 0.45 0.50 2387 0.43 0.47 0.09 Household’s Main Source of Drinking Water is 0.27 0.44 2387 0.28 0.26 0.03 Unprotected Spring Household has Access to Electricity 0.15 0.35 2387 0.14 0.15 0.04 Household has a Mobile Phone 0.18 0.38 2387 0.19 0.17 0.04 Household has a Radio 0.75 0.43 2387 0.74 0.76 0.05 Household Expenditure on Food in Last 30 Days 3561 1982 2340 3524 3600 0.04 (Afghani) Household Received Loan in Past 12 Months 0.47 0.50 2387 0.48 0.46 0.05 Most Preferred Project of Male Respondents is 0.29 0.46 2387 0.30 0.28 0.05 Drinking Water Most Preferred Project of Male Respondents is 0.13 0.33 2387 0.11 0.15 0.11 Irrigation Most Preferred Project of Male Respondents is 0.06 0.24 2387 0.06 0.06 0.02 Electricity Most Preferred Project of Male Respondents is 0.15 0.36 2387 0.16 0.14 0.04 Road or Bridge Male Head-of-Household Respondent Attends 0.32 0.47 2387 0.33 0.31 0.05 Shura Meetings Note: Standardized difference reflects difference in means between district and at-large villages, divided by the standard deviation of district villages. 34 Table 2. Summary Statistics for Outcome Variables Mean Standard Error Obs. Mean Standard Error Obs. District Elections At-Large Elections Male Council Member Finished High School (Percentage) 6.98 25.50 1,031 10.46 30.62 985 Male Council Member Finished Middle School (Percentage) 15.42 36.13 1,031 18.27 38.67 985 Male Council Member Finished Primary School (Percentage) 31.81 46.60 1,031 33.50 47.22 985 Female Council Member Finished High School (Percentage) 1.08 10.36 1,015 0.51 7.13 980 Female Council Member Finished Middle School (Percentage) 3.25 17.74 1,015 2.35 15.15 980 Female Council Member Finished Primary School (Percentage) 10.34 30.47 1,015 8.88 28.46 980 Distance between Residences of Council Members and Village 502.28 1155.18 509 295.70 338.59 494 Center (Meters) Days between Council Election and Project Start 413.30 180.88 252 375.36 155.21 226 Days between Council Election and Project End 710.67 181.04 252 682.70 172.54 226 35 Table 3. Effect of Electoral Rules on Council Members' Competence Percent of Male Council Members who Finished High School (1) (2) (3) (4) 3.96*** 0.07 1.95 -0.90 At-Large Elections [1.32] [1.80] [1.23] [1.55] Fractionalized Project Preferences 7.97*** * At-Large Elections [3.01] -3.10 Fractionalized Project Preferences [1.99] Ethnically Mixed Village 7.96** * At-Large Elections [3.21] -3.27 Ethnically Mixed Village [2.21] Geographically Large Village 9.96*** * At-Large Elections [3.01] -3.36* Geographically Large Village [1.96] Quadruple Fixed Effects Yes Yes Yes Yes Observations 2,016 2,016 2,016 2,016 R-squared 0.18 0.19 0.19 0.19 Note: The unit of observation is council member. The dependent variable is a dummy variable that equals 100 if a council member finished high school and zero otherwise. Standard errors clustered at the village level in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 36 Table 4. Effect of Electoral Rules on Council Member Location Natural Log of Distance between Residences of Council Members and Village Center Council Members Who Did Not Finish High Full Sample School (1) (2) (3) (4) (5) -0.28*** -0.25*** -0.55*** -0.41*** -0.41*** At-Large Elections [0.11] [0.07] [0.16] [0.12] [0.12] At-Large Elections * At Least One Council Member 0.68*** 0.39** 0.39** Finished High School [0.20] [0.19] [0.19] District Elections * At Least One Council Member 0.54** 0.29 0.27 Finished High School [0.22] [0.18] [0.18] Natural Log of Median Distance between Residences of 0.72*** 0.67*** 0.67*** Villagers [0.05] [0.06] [0.06] Share of Baseline Survey Respondents Who Finished 0.38 High school [0.58] Quadruple Fixed Effects Yes Yes Yes Yes Yes Observations 1,003 1,003 770 770 770 R-squared 0.192 0.346 0.208 0.337 0.338 Note: The unit of observation is council member. The sample includes only male council members. Standard errors clustered at the village level in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 37 Table 5. Location of Council Member Residence and Educational Attainment Council Member Finished High School (1) (2) Natural Log of Distance between Council Member's Residence and -1.21 -2.51** Village Center [1.18] [1.24] At-Large Election * Natural Log of Distance between Council 5.13** 5.43** Member's Residence and Village Center [2.27] [2.24] 7.53*** 7.69*** At-Large Elections [2.45] [2.47] 4.28** Natural Log of Median Distance between Residences of Villagers [1.70] Quadruple Fixed Effects Yes Yes p-Value for Effect of Distance in At-Large Villages 0.045 0.150 Observations 857 857 R-squared 0.21 0.21 Note: The unit of observation is council member. The dependent variable is a dummy that equals 100 if a council member finished high school and zero otherwise. Natural Log of Distance between a Council Member's Residence and Village Center is demeaned. Standard errors clustered at the village level in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%. 38 Table 6. Effect of Electoral Rules on on the Speed of Project Implementation Days Between Elections and Project Start Days Between Elections and Project Completion (1) (2) (3) (4) (5) (6) (7) (8) -40.05** -42.91** -39.25** -37.40** -33.94* -35.23* -32.13* -32.74* At-Large Elections [16.29] [16.85] [16.79] [16.99] [18.11] [18.40] [18.43] [18.38] 18.07* 10.45 15.29 17.77 Number of Projects in Village [10.72] [11.65] [10.86] [11.24] Natural Log of Project Budget (US -35.13*** 11.45 Dollars) [13.34] [13.35] Type of Project Fixed Effects No Yes Yes Yes No Yes Yes Yes Quadruple Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes Observations 478 478 478 478 478 478 478 478 R-squared 0.34 0.35 0.36 0.38 0.38 0.40 0.40 0.40 Note: The unit of observation is project. The dependent variable in (1)-(4) measures the number of days between the council election and the start of a 39 project. The dependent variable in (5)-(8) measures the number of days between the start of the project and the day the project was finished. Standard errors clustered at the village level in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. Table 7. Effect of Electoral Rules on Economic and Political Outcomes. At-Large At-Large Standard Standard Dependent Variable: Elections * Elections * Obs. R-squared Error Error Midline Endline Panel A. Benefits Delivered by Projects (1) Seasons in Past Year Water Was of Poor Quality -0.074 [0.092] -0.056 [0.091] 3,806 0.14 (2) Child Suffered Diarrhea in Past 2 Weeks -0.057*** [0.019] 2,465 0.06 (3) Hours of Electricity in Past Month (winsorized and logged) 0.141 [0.177] 0.210 [0.189] 4,464 0.42 (4) Yield of Most Recent Harvest (tons, winsorized and logged) 0.003 [0.031] 0.048 [0.031] 2,954 0.42 (5) Revenue from Most Recent Harvest (winsorized and logged) 0.171 [0.165] 0.246* [0.134] 2,926 0.40 Panel B. General Economic Outcomes (6) Economic Situation Has Improved in Past Year (Female Respondents) 0.045* [0.026] -0.016 [0.019] 3,997 0.15 (7) Economic Situation Has Improved in Past Year (Male Respondents) -0.008 [0.024] -0.012 [0.018] 4,493 0.19 Respondent Attributes Positive Economic Change to Local Leaders or Council (8) 0.013** [0.006] 2,135 0.04 (Female Respondents) Respondent Attributes Positive Economic Change to Local Leaders or Council 40 (9) -0.003 [0.008] 2,355 0.08 (Male Respondents) (10) Income Earned in Past Year (winsorized and logged) -0.015 [0.030] 0.005 [0.033] 4,439 0.20 (11) Annualized Consumption (winsorized and logged) 0.030 [0.035] 0.067* [0.034] 4,172 0.17 Panel C. Political Outcomes Desired Change in Decision of Influential Villagers in Past Year (Female (12) 0.003 [0.007] 0.004 [0.008] 3,980 0.07 Respondents) Desired Change in Decision of Influential Villagers in Past Year (Male (13) 0.009 [0.009] -0.022** [0.009] 4,495 0.05 Respondents) (14) Complaint about Project -0.023 [0.031] 1,716 0.21 (15) Complaint about Corruption or Nepotism Related to Projects -0.008 [0.018] 1,722 0.16 (16) Prefers Election to Select Headman (Female Respondents) -0.010 [0.028] -0.020 [0.031] 3,578 0.12 (17) Prefers Election to Select Headman (Male Respondents) 0.051** [0.025] 0.016 [0.022] 4,440 0.23 (18) Prefers Selection of President by Secret Ballot Election (Male Respondents) 0.025** [0.012] 2,112 0.07 Note: Each row represents a separate regression. Measures are based on answers of villagers to surveys conducted approximately two years (midline) and four years (endline) after council elections. Some of the questions were asked only during one of the surveys, so the coefficients for the other survey is not available. Standard errors clustered at the village level in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%.