WPS7941 Policy Research Working Paper 7941 Reforming Village-Level Governance via Horizontal Pressure Evidence from an Experiment in Zimbabwe Kate Baldwin Shylock Muyengwa Eric Mvukiyehe Development Research Group Impact Evaluation Team January 2017 Policy Research Working Paper 7941 Abstract How can patrimonial local-level governance be reformed? on the framework governing village leadership, and one in Debates on this topic have focused largely on the possibility which civil society leaders were trained alongside village of reform via pressure from above (superordinate leaders) or heads. The results suggest that horizontal pressure from civil below (citizens). This paper tests whether horizontal pres- society leaders increased village heads’ knowledge of and sures from civil society leaders can reform local governance compliance with regulated procedures, improved their man- in a context where neither of these mechanisms operates agement of issues and raised citizens’ trust in their leadership. effectively. The study analyzes an experimental intervention A quantitative and qualitative analysis of the mechanisms in Zimbabwe intended to reduce abuse of power by village through which the trained civil society leaders had these heads. Analytic leverage comes from the fact that the 270 effects suggests they accomplished reform by directly apply- study villages were randomly assigned to two variants of the ing social pressure on village heads to abide by regulations. intervention, one in which only village heads were trained This paper is a product of the Impact Evaluation Team, Development Research Group. 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 emvukiyehe@worldbank.org. 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 Reforming Village-Level Governance via Horizontal Pressure: Evidence from an Experiment in Zimbabwe Kate Baldwin, Shylock Muyengwa and Eric Mvukiyehe∗ Keywords: Traditional institutions, horizontal accountability, village-level governance, for- eign assistance, Zimbabwe. JEL Classification: H7 ∗ Baldwin is Assistant Professor, Department of Political Science, Yale University. Muyengwa is PhD Candidate, School of Natural Resources and Environment, University of Florida. Mvukiyehe is Economist in the Development Research Group, Impact Evaluation Unit, The World Bank. We are grateful to our im- plementing partners in Zimbabwe for their participation in the experiment and the United States Agency for International Development (USAID) for providing funding through their Evaluating Democracy and Gov- ernance Effectiveness (EDGE) Initiative. Thanks to MacKenzie Saunyama, Lauren Young, Sarah Elven and Srujith Lingala for excellent research assistance and to Target Research for coordinating our surveys. We have received valued feedback from Rob Blair, Gwyneth McClendon, Kristin Michelitch and audiences at the Boston-Area Working Group on African Political Economy (B-WGAPE), Brown University, Columbia University, the International Food Policy Research Institute (IFPRI), Northwestern University, Yale Univer- sity, Washington University in St. Louis, the African Studies Association Annual Meeting and the Midwest Political Science Association Annual Meeting. 1 Introduction In many countries of the world, village-level political leaders remain the main government structures with which citizens interact. However, the quality of village-level governance is very uneven, remaining patrimo- nial and personalistic in many places. For example, village heads in Indonesia, village maliks in Afghanistan and panchayat leaders in India often make decisions based on their own interests and biases, begging the question of how best to reform these structures to reduce cronyism and corruption (Olken, 2010; Murtaza- shvili, 2016; Veron et al., 2006). Much of the debate on this topic has focused on the possibility of reform via pressure from above, in- volving sanctioning by higher-level leaders to whom these leaders report, or pressure from below, especially through popular elections for village-level leaders (Grossman, 2014; Tendler, 1997). But these two mech- anisms for reforming local governments may not always be available or effective, especially when village leaders have high socio-economic status in their community independent of their bureaucratic appointment or electoral legitimacy. Furthermore, although the policy community has favored community monitoring interventions that promote reform from below, investing hundreds of millions of dollars into them (Molina et al., 2013), a growing body of evidence suggests that these interventions have limited effects on governance (Banerjee et al., 2010, 2015; Lieberman, Posner and Tsai, 2014; Olken, 2007). In view of the mixed suc- cess of community monitoring interventions on governance, we consider the effectiveness of an alternative strategy which promotes horizontal pressure from existing civil society leaders. Our theoretical framework builds on a long tradition of emphasizing the importance of civil society leaders for checking the influence of government and improving development outcomes (Putnam, Leonardi and Nanetti, 1994; De Tocqueville, 1956). We use the term civil society leaders broadly to refer to individu- als who have status by virtue of their leadership of a community-based group, regardless of how democratic the purposes or goals of the group are. This non-normative definition follows a tradition of defining civil society as a force that can be mobilized for a variety of ends (Krishna, 2007; Berman, 1997). The effective- ness of horizontal pressures from existing leaders in reforming village-level governance is thus a plausible but empirically uncertain proposition. We test the efficacy of using civil society leaders to hold village-level leaders to account in Zimbabwe, 2 a context where citizens often express grievances against local leaders but top-down and bottom-up mech- anisms for reform are weak. We study the effectiveness of horizontal pressure for reforming village-level governance by examining the differential effects of two variants of a randomized capacity-building inter- vention for village leaders, run by an international NGO in collaboration with local partners in 270 villages. In both arms of the intervention, the village head received training on rules, regulations and best practices pertaining to village-level governance in Zimbabwe. Our analytical leverage comes from the fact that in one arm only, another local civil society leader was trained alongside the village head and also learned the rules, regulations and best practices he (or, in rare instances, she) was supposed to follow in his (or her) decision-making. Our empirical findings reveal that horizontal pressure from civil society leaders can improve the quality of governance at the local level. Specifically, the inclusion of a civil society leader in the training sessions increased village head’s knowledge of and compliance with regulated procedures, improved their manage- ment of local issues and raised trust in their leadership among citizens compared to a control condition in which only the village head was trained. This finding suggests that civil society leaders can serve as a mechanism of change even in the absence of complementary pressures from above or below. Furthermore, we draw on ancillary quantitative and qualitative evidence to probe the mechanisms through which civil society leaders had this effect. The results of these analyses suggest that civil society leaders caused change not through competition with village heads but through direct social pressure and, to some extent, increased human resource capacity. The paper proceeds as follows. Section two provides a brief overview of the empirical literature linking community monitoring interventions to governance reform, highlighting in particular inadequate attention to horizontal pressure, and points to a number of mechanisms linking horizontal pressures to local-level gover- nance outcomes. Section three provides a brief background to village level governance in Zimbabwe, while section four discusses the intervention under study and our identification strategy. Section five discusses our measurement strategy and data collection and section six discusses our main findings. Section seven probes a number of potential mechanisms discussed in the theoretical section and section eight concludes. 3 2 Theoretical framework 2.1 Local-Government Reform Local government around the world is often characterized as highly patrimonial in that local leaders make decisions based on their personal whims, and citizens’ access to resources is conditional on their personal relationships with these leaders (Weber, 1922). The results are cronyism and corruption that harm the interests of the majority of local citizens. Most of the scholarly debates on the best path to reforming patrimonial local governments have focused on the merits of top-down versus bottom-up reform. In favor of top-down reforms, scholars have shown that reform-minded higher-level governments can sometimes be very effective in reducing cronyism and clientelism in local-level politics (Malesky, Nguyen and Tran, 2014; Tendler, 1997). In cases where higher-level political leaders have both the incentive to reform local politics and the capacity to sanction local leaders who do not adhere to reforms, top-down reforms may be very effective. However, in cases where higher-level politicians lack the political will to implement reforms, this path to reform is a non-starter. In addition, even if higher-level politicians have the political will to implement reforms, this mechanism is also dependent on adequate capacity to monitor the actions of local leaders and sufficient control over their career prospects to motivate reform. In cases where the central state is weak and/or local leaders have independent sources of legitimacy based on their own political, economic or social standing within their communities, sanctioning from the higher-ups may be inadequate. Alternatively, reform may be possible via pressures from below. In particular, when local-level leaders are popularly elected, citizens have a tool for sanctioning politicians who engage in corruption or cronyism (Grossman, 2014; Martinez-Bravo et al., 2014). However, in the absence of elections for local leaders, the prospects for reform from below are more nebulous. In some circumstances, citizens may be able to apply non-electoral pressure on local leaders to govern well. For example, Tsai (2007) argues that citizens can apply moral pressure on local leaders when they are embedded in encompassing community networks, and Baldwin (2016) notes that citizens can use economic sanctions against leaders who depend on citizen contributions for their material upkeep. However, in other instances, the power differentials between citizens and local leaders and the collective action dilemma faced by communities in holding them to account are 4 insurmountable. Yet, a third model of reform, via the empowerment of civil society leaders, is also possible. This model is distinct from the bottom-up model in that it depends on local leaders, rather than citizens themselves, to encourage political leaders to reform their behavior. As a result, horizontal pressure has a separate logic and may be effective in improving governance even in situations where bottom-up pressure is weak. 2.2 Community Monitoring Interventions In recent years, international institutions have devoted significant resources to fostering bottom-up reform. In particular, they have invested huge amounts in community monitoring interventions, programs intended to facilitate the ability of communities to pressure policymaking bodies and service delivery organizations without fundamentally increasing citizens’ power to discipline them.1 Information campaigns, scorecards, social audits and capacity building for oversight institutions have all become common in developing coun- tries (Mansuri and Rao, 2012; Molina et al., 2013). On balance, the evidence that community monitoring interventions improve local service provision is mixed. Banerjee, Deaton and Duflo (2004) find that civilian engagement and monitoring in rural Rajasthan was not associated with improved health provision or outcomes. Cilliers et al. (2016) test two monitoring schemes–one in which the headteacher was required to report on teachers’ attendance and another in which they were required to report on teachers’ attendance in exchange for bonus payments–and find that neither intervention was associated with improvements in teachers’ attendance. Similarly, Lieberman, Posner and Tsai (2014) find little evidence that community report cards improved educational outcomes in Kenya. Other studies point to positive effects of community monitoring. For instance, in the education sector in Kenya, Duflo, Hanna and Ryan (2012) find that when communities selected, monitored, and assessed the performance of primary school teachers, educational performance improved. Likewise, in a randomized field experiment in Uganda that involved community-based monitoring of public primary health providers, Bjorkman and Svensson (2009) find that disseminating information about local health facilities and encour- aging civilian engagement/oversight significantly improved the quality of healthcare delivery and health outcomes in treatment areas. A study by Aker, Ksoll et al. (2015) investigated the effects of mobile moni- 1 Molina et al. (2013) reports on the large budgets of a number of NGOs and initiatives in this sector. 5 toring of teacher effort and student outcomes in Niger, finding that the intervention substantially improved students’ performance. The mixed evidence in favor of community monitoring interventions is not surprising from the per- spective of many theories of civic participation. One problem is that many political actions are subject to collective action dilemmas. Each citizen may avoid taking action, preferring that other citizens bear the costs entailed in doing so (Olson, 1965). A second problem is the power differentials between citizens and powerholders. The average citizen may not have access to the ear of the targeted leader, or may not be- lieve he or she has the power to influence them (Gamson, 1968). Indeed, Acemoglu, Reed and Robinson (2014) suggest that these types of power differentials make civil society organizations in Sierra Leone tools of traditional leaders. Especially in contexts in which the targets of community monitoring interventions are not hired or elected by citizens, there are numerous ways the causal chain between increased monitoring by citizens and better behavior by leaders could break down (Lieberman, Posner and Tsai, 2014).2 How then can these obstacles be overcome? One variant on community monitoring that could provide an answer is “horizontal pressure” from established civil society leaders. Instead of providing training and information to citizens with the goal of having them hold targeted leaders to account, organizations could provide training and information to other community leaders. This has not been widely promoted as a mechanism for improving accountability, probably due to reasonable concerns that leaders could have distinct interests from the citizenry as a whole (Olson, 1982) and the potential for collusion among leaders at the expense of citizens (Gottlieb, 2015; Mattingly, 2016). But other considerations weigh in favor of horizontal pressures. In particular, civil society leaders often have actions available to them that most citizens do not. As a result, they may be able to motivate reforms through mechanisms distinct from bottom-up pressure, as elaborated in the next section.3 2 For further discussion of the challenges of transparency initiatives in settings without electoral account- ability, see Kosack and Fung (2014) and Malesky, Shuler and Tran (2012). 3 Interestingly, two recent training interventions for committee members, rather than community mem- bers more generally, show promising results, especially when complemented with increased linkages to key decision makers (Barr et al., 2012; Pradhan et al., 2014). 6 Leadership Competition Empowered Increased Monitoring Improved Civil Society • Direct social pressure Local Governance • Fire alarm triggering Leaders • Community mobilization Outcomes Capacity upgrading Figure 1: Mechanisms through which horizontal pressures could cause reform 2.3 Horizontal Pressures for Reform There are a number of different mechanisms through which the empowerment of civil society leaders could result in local governance reform. Specifically, civil society leaders could cause village leaders to reform their behavior as a result of: (1) leadership competition (2) capacity upgrading or (3) increased monitoring. These channels are displayed in figure 1 and elaborated below. 2.3.1 Leadership competition Competition from other civil society leaders to serve as the most important intermediary between citizens and the outside world may be sufficient to reduce cronyism and corruption by local leaders. Even absent electoral competition for key political posts, market competition may induce better behavior from leaders. When community members have an outside option for helping to fix their problems, established leaders may need to start offering them better terms. A number of empirical studies point to the relevance of this mechanism. For instance, Krishna (2007) finds that competition from a new generation of local “fixers” 7 motivates better governance from established leaders in northern India. However, in contrast, Acemoglu, Reed and Robinson (2014) find that traditional leaders in chiefdoms with more ruling families (i.e., more elite competition) enjoyed less legitimacy and lower levels of social capital than chiefdoms with fewer ruling families. 2.3.2 Capacity upgrading Human capital deficiencies are also frequently a key constraint on public sector performance in many de- veloping countries, especially at the local level where there has traditionally been a shortage of staff skills in specialized areas of governance (World Bank, 2000). Thus, training community leaders to fulfill gover- nance functions may improve outcomes by bridging this capacity gap. If the key constraint on reforming local governance is not the incentives of local political leaders but their capacity, the empowerment of other civil society leaders may provide the skills and resources necessary to enact reforms. By acting as an addi- tional human resource at the local level, civil society leaders may be able to improve local governance. 2.3.3 Increased monitoring Finally, civil society leaders may induce reforms by monitoring village leaders. Increased monitoring by vil- lage leaders may induce changes in behavior three ways, corresponding to three types of sanctions: (i) civil society leaders may apply direct pressure upon political leaders, using their own status vis-a-vis politicians to threaten social sanctions for poor behavior (ii) civil society leaders may serve as fire alarms that allow higher level politicians to sanction bad behavior by local leaders; or (iii) civil society leaders may mobilize the community as a whole to pressure local leaders to reform their actions. Civil society leaders may be able to directly pressure political leaders to change their behavior due to the desire of politicians to remain in good social standing in their eyes. Social incentives and social recognition have been found to be important motivators of public officials in a variety of contexts (Levi and Sachs, 2009; Ashraf, Bandiera and Jack, 2014; Ahmed et al., 2012). Furthermore, compared to citizens, who typically face power differentials and collective action problems in monitoring and sanctioning political leaders, civil society leaders may be better positioned to employ social sanctions (Tendler, 1997, p. 66-69). In addition, civil society leaders may operate as a fire alarm to alert higher level authorities to poor 8 village-level governance (McCubbins and Schwartz, 1984). The idea of fire alarm, which fits with the stan- dard shirking model, presumes that the principal has difficulty observing the agent’s efforts due to costly monitoring. Compared to higher level authorities, local civil society leaders may be able to observe the behaviors of village heads more easily. In contexts in which civil society leaders can communicate trans- gressions to superordinate political leaders who will then sanction bad behavior, this may motivate local leaders to improve their actions.4 In this way, monitoring by civil society leaders may feed into top-down pressures for reform. Finally, civil society leaders can also coordinate sanctions from the community as a whole. In communi- ties facing large collective action dilemmas with regard to both observing transgressions by local leaders and then mobilizing to hold them accountable, civil society leaders can play a critical facilitating role (Popkin, 1979). In this way, civil society leaders can activate bottom-up pressures. It should be noted that most of these mechanisms depend on civil society leaders having incentives to implement reforms. Especially given our non-normative definition of civil society leaders, this is likely to vary among places and among different civil society leaders. In contexts where leaders do not have adequate incentives to push for reforms, they may collude with or be co-opted by government (Mattingly, 2016). However, in many places, the diversity of interests among village elites will create a constituency for reform within this group (Tendler, 1997). Our primary goal in this paper is to test the effectiveness of mobilizing village-level civil society leaders to press for local government reform, assuming that any such effects could occur through either one (or a combination) of the afore-discussed channels. Although we present preliminary evidence on the mech- anisms through which civil society leaders influence local governance, we cannot definitively adjudicate between mechanisms, and indeed, it is quite likely that more than one mechanism is at work. 4 In this regard, community leaders may be playing the same theoretical role as the cheap mobile moni- toring in the Niger experiment by Aker, Ksoll et al. (2015), which was shown to contribute to adult education outcomes. However, the Cilliers et al. (2016) study on locally monitored performance of teachers in Ugan- dan primary schools suggests that monitoring by itself may be inadequate to induce desired behaviors. In their study, effects on teacher attendance were the result of monitoring combined with financial incentives, rather than monitoring alone. 9 3 Village-Level Governance in Zimbabwe Village governance in Zimbabwe has been resistant to pressures from either above or below. Villages are led by village heads; local leaders who allocate land, run village courts and help to broker a wide range of development programs, including food aid distribution. There are an estimated 24,000 village heads across Zimbabwe (Zamchiya, 2011). These leaders remain the most important point of contact for most rural villagers (Ncube, 2011), but they are frequently accused of bias and abuse of power. The possibility for bottom-up reform is limited due to the fact that these leaders are not subject to regular election. The vast majority of these leaders inherit their positions from within their village’s founding family and rule for life.5 Community members have little formal recourse against village heads who govern poorly, although they may be able to apply social or moral pressure under some circumstances. Importantly, Shona village heads have traditionally worked with a handful of senior kinsmen to hear disputes and make decisions, and these councils of advisers historically provided a check on their decision-making.(Bourdillon, 1976, p. 79). The possibility for effective top-down reform has been constrained by central government’s capacity and incentives. At independence in 1980, the Zimbabwean government was committed to reigning in patrimonial village-level governance. However, as the state became more politicized, this may have further complicated the central government’s interactions with village heads.6 At that time, lack of capacity made it difficult for the central government to check the power of village-level leaders. Specifically, in the 1980s, the Zimbabwean government tried to reduce the influence of village-level traditional leaders by stripping them of powers. It set up 6-member elected village-level development com- mittees – VIDCOs – which were intended to check the power of village heads and push forward legal and agricultural reforms. Interestingly, two of the six members of VIDCOs were supposed to be chosen by civil society organizations to represent women and youths (Munro, 1998, p.243). However, in most rural com- 5 Our survey asked the current village head when and why the previously village head was removed from office. The vast majority of village heads ruled until they died or were bed-ridden by illness; only 5 percent of village heads had been removed for reasons related to performance or community divisions. Similarly, 94 percent of the village heads in our sample said they did not fear being demoted by anyone, including community members and government officials. 6 The government of national unity formed following the contentious 2009 elections made little differ- ence to this general trend. 10 munities, the government proved unable to reduce the power of village heads (Alexander, 1996, p.187). The Report of Land Tenure Commission, published in 1994, concluded that village heads remained much more powerful than VIDCOs and continued to perform wide-ranging functions (Ncube, 2011, p.94). Another study noted that where VIDCOs were successful, it was because they served as advisers to rather than rivals to the village head (Munro, 1998, p.270). Since the mid-1990s, the government has focused on trying to regulate village heads, rather than creating parallel institutional structures. Village heads now have official recognition and receive small salaries of $25 per month from the state. They have official duties per Zimbabwean law, including collecting levies and taxes, maintaining an up-to-date population register, and allocating land. There are regulations regarding the procedures through which they are supposed to make decisions and how to treat villagers. However, village heads continue to have much more de facto power than they do under the law, and they frequently do not follow regulated procedures. Indeed, many village heads have retained a measure of independence from both formal legal regulations and informal partisan pressures to support Zimbabwe’s ruling party. Village heads are much more divided in their political loyalties than higher-level traditional leaders, who have been more thoroughly co-opted by state patronage.7 The main complaints against village heads are that they abuse their power by allowing their personal interests and biases to enter into their decision-making. In particular, they do not follow regulations requiring broader consultation before making decisions and they are frequently accused of self-interested behavior and favoritism, especially, in Zimbabwe’s polarized political system, along partisan lines.8 Village heads are often accused of favoring their co-partisans in food distribution and court cases. Importantly, many but not all of these complaints are against village heads affiliated with the ruling party; village heads affiliated 7 An underground pro-democracy organization in Zimbabwe, Sokwanele, attempted to collect compre- hensive information on the political affiliations of traditional leaders in three constituencies in Manicaland that experienced violence during the 2008 elections. In their sample, 89 percent of higher-level traditional leaders (chiefs and headmen) were supporters of the ruling party, 0 percent were supported the main op- position, with the remaining 11 percent considered politically “neutral”; in contrast, 69 percent of village heads were supporters of the ruling party and 31 percent of the main opposition. See Sokwanele (2011). When we asked the village heads in our survey an open-ended question about whether they felt close to any “masangano”, a Shona term for organization that connotes both political and non-political organizations, 44 percent of village heads in our sample said that they felt close to the ruling party, 13 percent said that they felt close to an opposition party and 43 percent did not offer a political response, either because they did not feel politically affiliated or because they self-censored in response to the question. 8 For more on the politicization of all spheres of life in Zimbabwe, see LeBas (2006). 11 with opposition parties are also regularly accused of partisanship.9 An interview with the local church leader in one of our study villages clearly demonstrates the perils of village heads having unchecked power: “[The village head’s] way of dispute management was frightening offenders and people were afraid to bring issues to the dare [village head’s court]. He had advisers at the dare but he hardly consulted them or the villagers when resolving cases. He had the ultimate decision-making powers and used political statements to intimidate people on the dare. The disadvantage of this process was that it was only the village head with all the unregulated power to pass resolutions without the people’s contributions. There were biased resolutions done on political lines.”10 This anecdote describes how village heads can fail to follow procedures that could potentially check their power, and how – in the Zimbabwean context – this is associated with favoritism along partisan lines. To assess how broadly village heads fail to follow regulations intended to increase consultation and curb bias, we consider village heads’ adherence to five regulated procedures in our study control villages. According to Zimbabwean regulations, village heads are supposed to consult other village institutions prior to making certain decisions, including the local resource management council and the local women’s council, they are supposed to include women in decision-making, they are supposed to make their court’s decisions available to the public, and they are not allowed to accept payments from citizens when hearing cases. However, as table 1 indicates, most village heads reported that they did not follow regulated procedures prior to exposure to the intervention at the center of this study. In particular, we observe low levels of consultation with other village institutions and women in particular, and a high tendency to take illegal payments for hearing disputes. Village heads did well only in the area of making court decisions publicly available. We also draw on household survey data from our control villages to consider whether there is systematic evidence that village heads are (or are perceived to be) biased against villagers who share different political opinions in the distribution of food aid and the resolution of court cases. Respondents are coded as having different political opinions from the village head based on their responses to a direct question about this.11 9 See Sokwanele (2011, p. 11). 10 Interview with community leader in village 31729. 11 Specifically, respondents were asked whether they agreed or disagreed with the statement “My own political views are very similar to those of the village head.” Unfortunately, even though the district author- ities initially approved a questionnaire containing this question, we were asked to remove it two-thirds of the 12 Table 1: Village Heads’ Adherence to Regulated Procedures (Control Group) Mean in Control Group N Consultation of resource management council 0.389 126 Consultation of women’s council 0.226 124 Proportion of female dare members 0.262 125 Access to records 0.645 124 No payments 0.336 119 Table reports means of each variable in the villages that did not receive any training in year 1. In table 2, we observe that the amount of bias in favor of the village head’s co-partisans is actually relatively small (and statistically insignificant) in the case of food aid distribution – people who said they had different political views from their village head were only five percentage points less likely to receive food aid than people who said they had the same political views as their village head. 12 However, we find substantial perceived bias in the case of court decisions; people with different political views from their village head were 20 percentage points less likely to believe that most of the village court’s decisions were fair and 8 percentage points less likely to take the most common local disputes – disputes over the destruction of crops by livestock – to them for resolution. Thus, the widespread bias in village heads’ decision-making gives many citizens grievances against them. The question of how to reform village-level governance is tricky because of the low potential for top-down reform and the weak mechanisms in place for achieving change from the bottom up. Instead, the intervention described in the next section draws on the tradition of senior community members providing advice to village heads as a potential mechanism of reform. way through surveying, so we do not have data on this question for one-third of respondents. Fortunately, because the survey was proceeding geographically and we randomized by geographic blocks, the sample is still well balanced across experimental arms. In table 8 of the appendix, we consider how the villages for which we have political data differ from the villages for which we do not. 12 Interestingly, these results are not simply the result of the village head’s personal biases being over- whelmed by biases against opposition supporters. There is even less evidence that supporters of the ruling party are favored in the distribution of food aid in our study villages, as indicated in the results in table 2 of the appendix. Much of the food aid in these villages comes from international organizations, which may explain the limited bias. 13 Table 2: Biases in Village-Level Decision Making (Control Group) Different Political View Same Political View Difference N Received Food Aid Last Year 0.420 0.466 0.046 632 (0.036) (0.024) (0.043) Believe Most Decisions by 0.439 0.634 0.200*** 619 Village Head’s Court Fair (0.036) (0.023) (0.043) Took Case to Village Head’s 0.396 0.472 0.076* 453 Court if Had Dispute (0.032) (0.025) (0.042) The first two columns report means with standard errors below in parentheses. The third column reports the differences between column two and column three, with the standard error of the difference reported in parentheses below. Only villages that did not receive any training in year 1 are included in the calculations. 4 Intervention and Research Design We study the effectiveness of horizontal pressure in reforming village-level governance in Zimbabwe by comparing two variants of a training intervention for village heads. An international NGO working in rural Manicaland noted that many village heads in the region were not following regulations or recognizing best practices when making decisions and they worried that this was both increasing the bias of their decisions and decreasing the legitimacy of their leadership. The concern was that any bias in the village head’s decision-making is problematic both insofar as it violates norms regarding impartiality but also because it decreases their ability to manage village problems. For example, if village heads provide food aid on the basis of partisanship rather than objective need, it reduces their effectiveness in addressing food security issues. If village heads adjudicate cases in a biased manner, it may reduce the effectiveness of their courts in resolving local disputes. More generally, village heads who are more biased and less efficacious in managing village problems may have their legitimacy reduced, thereby leading to a vicious cycle in which citizens’ poor perceptions of them make it difficult for them to resolve village issues (Levi, 1988; Levi and Sachs, 2009; Levi, Sachs and Tyler, 2009). As a result, the international NGO worked with a local NGO to develop a series of training sessions that would inform village heads of the regulations governing their activities and “best practices” in conflict 14 management techniques. The training sessions were divided into two three-day modules, with about a three- month gap in between the two sessions. In one variant of the training session, only the village head attended. But, due to concerns that simply informing village heads of the laws regulating their activities would not be sufficient to change their behavior, there was also a second variant of the training sessions in which another community leader was trained alongside the village head on the rules constraining them. By comparing the effectiveness of these two variants, we can parse the effects of civil society leader involvement from the effects of training the village head. This effect is estimated contingent on the village head also being trained, as it was not deemed feasible to have a third variant of the intervention in which only the community leader was trained. Crucially, both the timing and the variant of the training to which villages were exposed were randomly assigned. Specifically, we randomly selected the villages that would be trained in the first year of the analysis, and those that would not be trained until year 2. Then, within the villages selected for training in year 1, we randomly determined in which villages only the village head would be trained and in which villages both the village head and a community leader would be trained. At the end of the first year of training, the year 2 villages served as the comparison group to determine the effectiveness of each training variant. In total, 270 villages were included in the randomized roll-out of the intervention. The roll-out of the two interventions across villages is described in table 3. Randomization was done by blocks of villages in the same ward and on the same land classification to improve the likelihood of achieving balance across the experimental group. Balance statistics, which are compiled in Table 1 of the appendix, indicate good balance across treatment arms. Compliance with treatment assignment was very high. Fully 96 percent of the village heads assigned to the “village head only” training attended the sessions, 98 percent of the village heads assigned to the “village head plus civil society leader” training attended, and only 1 percent of the village heads assigned to the “no training” variant attended training sessions. Similarly, 97 percent of civil society leaders assigned to the “village head plus civil society leader” training attended sessions, and civil society leaders from just 4 percent of the villages assigned to the “village head only” training and 1 percent of the villages assigned to “no training” attended training sessions. 15 Table 3: Roll-Out of Intervention Number of Villages No training in year 1 136 Training for VH only 69 Training for VH + CL 65 The civil society leaders invited to attend the sessions were required to be from the “non-traditional” sector, and included village health workers, representatives of farmers’ groups, church leaders, represen- tatives of NGOs and school development committee members.13 These same sets of leaders have often been represented on VIDCOs and tasked with implementing reforms via that institution (Munro, 1998, p. 308-314, 340-324). The expectation was that these leaders would provide a constituency for reform because they had less stake in the traditional system than the village head. Indeed, table 4 compares data from our surveys of village heads and our surveys of a representative sample of civil society leaders in these villages to show that – on average – civil society leaders are more pro-reform than village heads.14 Specifically, the top panel of table 4 shows that civil society leaders are often from demographic groups who would benefit from the reform of village-level governance. For example, a majority of civil society leaders in these village are female. In addition, they are younger and less likely to be visible supporters of the ruling party than the village head (as measured by whether they had a sign displayed outside their home during our interviews, which were shortly after the 2013 elections). At the same time, the profile of the average civil society leader suggests low power differentials between them and the village head. They are significantly more educated than village heads, and they report higher levels of income and wealth. These findings suggest civil society leaders often have both an interest in and the ability to check the village head’s power. We also find that 57 percent of all civil society leaders have a family relationship to the village head. 13 The exact breakdown of the civil society leaders who were selected to participate in the sessions is indicated in table 6 of the appendix. 14 We exclude dare members, who were included in the survey but not eligible for the training, from the sample of civil society leaders described in this table. 16 In many contexts this would suggest a favored relationship between these two leaders. However, in the villages we study 50 percent of all households were related to the village head. The bottom panel of table 4 further considers whether civil society leaders can realistically be viewed as a mechanism for checking the power of village heads by comparing the attitudes and legal knowledge of village heads and civil society leaders prior to the intervention (by looking at the statistics for control villages only). Importantly, a slight majority of civil society leaders favored checks on the power of the village head’s court. In addition, civil society leaders express significantly higher levels of support for individual rights than village heads.15 Knowledge of the laws and procedures regulating village governance is approximately equal between village heads and civil society leaders. This evidence suggests that the average civil society leader in these villages has both the incentive and the ability to check the power of village heads. However, it is noteworthy that the NGO depended on the village head to identify civil society leaders living in his or her village, and to select which one to invite. Our expectation was that this would reduce the probably of finding differential effects across the two training variants insofar as village heads should choose to invite civil society leaders over whom they have more influence. In table 6 in the appendix, we compare the traits of the trained civil society leaders to our random sample. We find village heads were more likely to invite female leaders and NGO officers to accompany them and less likely to invite religious leaders; these patterns are interesting insofar as they suggest village heads chose people who likely had an interest in reform but perhaps had weaker independent power bases. 5 Data Collection and Measurement of Key Variables We conducted surveys of village heads, a randomly selected civil society leader, and a random sample of eight households in each of the villages in the study three months following the completion of the training sessions, in late August and early September 2013. We contracted a local survey firm to conduct the survey after training from the research team. We were able to conduct household surveys in each of the 270 villages and to interview the village head in 91 percent of them. We also interviewed a randomly selected civil society leader in 96 percent of villages. Most questions on the questionnaires were closed-ended and retrospectively 15 The differences in support for individual rights and beliefs about whether the dare should have more power are both statistically significant at the 95 percent confidence level. 17 Table 4: Comparing Village Heads to Civil Society Leaders Village Head Civil Society Leader Demographics, Economic Status and Partisanship (All Villages) Female 0.07 0.58 (0.25) (0.49) Average Age (Years) 64.1 51.7 (15.7) (14.2) Sign for Ruling Party 0.30 0.16 (0.46) (0.37) Finished Primary Education 0.69 0.86 (0.46) (0.35) Average Income ($) 959.7 1023.7 (1537.5) (1806.8) Average Cattle Wealth 2.98 4.30 (3.69) (28.45) Related to Village Head 0.57 (0.50) Knowledge and Opinions (Control Villages Only) Dare Should Have More Power 0.73 0.47 (0-1) (0.45) (0.50) Average Support for Individual Rights 2.86 2.98 (1-4) (0.40) (0.41) Average Knowledge of Law 0.66 0.69 (0-1) (0.23) (0.21) Table reports means with standard deviations below in parentheses. The means in the top panel aggregate data from 247 village heads and 196 community leaders across all villages. The means in the bottom panel aggregate data from 128 village heads and 95 community leaders in the control villages. 18 self-reported. In Zimbabwe, individuals’ political opinions are often a very sensitive issue and members of the public are often reluctant to express these openly (Bratton, 2011; LeBas, 2006). As a result, the study was de- signed and implemented to ensure multiple levels of protection for respondents. For example, to protect respondents, we collected identifying information on separate cover sheets that could not be linked back to the main surveys in the field. To maintain confidentiality of village heads and villages as a whole, we used codes for different communities, which were filled out in advance before teams entered the field to avoid sending the codebook into the field. In addition, we used prompts not recorded on the survey instrument and alphabetic and numeric codes to record responses to sensitive questions. We are interested in the effects of the intervention on a sequence of outcomes. First, we study the effectiveness of the intervention in reforming the processes governing decision-making at the village level, making it less based on the personal preferences of the village head and more respectful of legal regulations and individual rights. In addition, we study whether the intervention made village-level leaders fairer and more effective in managing local problems, and whether they increased their legitimacy. To measure local governance reforms, we measure the knowledge of village heads of legal regulations, the support village heads express for individual rights, and their adherence to regulated procedures. We measured knowledge by asking village heads six questions about the regulations governing their activities; the measure used in our analysis is the proportion of questions they answered correctly. We measured the attitudes of village heads toward individual rights by asking them about their level of agreement with statements in support of three specific rights on a four-point scale. Finally, we measure village head’s adherence to recommended procedures by asking them a series of questions about whether they consulted other organizations in the village (specifically, the resource management council and the council of women), the proportion of their court made up of women, whether they took payments for hearing cases and whether they made records of their court decisions publicly available. Due to concerns that those who attended the training sessions may have been more likely to know the desired outcomes to these questions and report that they were following them even if they were not, we also asked community leaders about the procedures followed by their village head. Some community leaders do not have detailed or up-to-date information on the village head’s procedures, so this second measure is less subject to demand effects but more noisy. 19 To measure the fairness and effectiveness of village heads in managing local problems, we consider village-level outcomes in two issue areas where they are typically involved: the distribution of food aid and the handling of land and livestock disputes.16 The most prevalent type of bias that features in village head’s decision-making in Zimbabwe is bias against political opponents. As table 2 showed, individuals who report having different political opinions from their village head are significantly less likely to view the village head’s court to be a neutral mechanism for resolving disputes. In contrast, we do not observe similar divisions of opinion across gender lines or ethnic lines.17 As a result, we measure whether the two interventions reduced bias by looking at their effects on household survey respondents who reported having different political views from their village head.18 We measure bias in food aid distribution by examining the likelihood of people with different political views from the village head receiving food aid, a measure of bias in real outcomes. We examine perceived bias in dispute resolution, measured as the likelihood of people with different political views from the village head viewing most decisions by the village head’s court as fair.19 We also measure the efficacy of the village head in solving problems in both these areas. We measure the effectiveness of village processes in reducing food insecurity by measuring the effect of households’ measured need on their likelihood of receiving maize food aid in the past year, where our measure of need is based on whether respondents in the household survey reported they had enough grain to feed their family after either harvesting it or buying it on the market.20 In villages where the village head is more effective in using available food aid to reduce food insecurity, we expect that the need of a household will be a more 16 Because the involvement of the village head in these issue areas is the result of strategic processes, we cannot exclude the subset of cases in which they are not involved in estimating our effects. 17 These results are reported in tables 3 and 4 of the appendix. 18 For the purposes of this study, we treat difference in reported political views as an exogenous covari- ate; although it is theoretically possible that people could have changed their response to this question in response to the intervention, we have empirically assessed whether individuals who give this response differ across the arms of the experiment, and we have found no evidence that this is the case. These results are reported in table 7 of the appendix. 19 If the village head actually reduces bias against political opponents, individuals who share the village head’s views will likely start to receive less favorable treatment. For this reason, we measure the likelihood of receiving food aid and believing court cases to be fair on the village head’s political opponents only. Note that we cannot identify the effect of the intervention on whether the decisions of the village head’s courts are actually more biased against political opponents; this outcome can only be measured over people who have disputes and also choose to take them to the village head, and the characteristics of this subset of people might be affected by the intervention. 20 We provide further details on this measure in section 6 of the appendix. 20 important factor in determining who receives food aid; therefore we estimate the effect of this variable under different treatment arms. We measure the efficacy of the village head in managing disputes by measuring the likelihood of respondents in their village stating that they had an unsatisfactorily resolved land or crop destruction dispute in the past year. When examining the village head’s efficacy in managing disputes, we estimate the effect of each variant of the intervention on the likelihood of having unsatisfactorily resolved disputes across all respondents. We do not restrict our analysis to respondents who report having disputes in the past year, because the efficiency of the village-level dispute resolution processes could conceivably change whether disputes emerge in the first place. Finally, we measure the amount of legitimacy that citizens accord to village heads, following Tyler (2006, p. 375)’s definition of legitimacy as the belief that institutions or authorities are “appropriate, proper, and just” which thereby leads citizens that hold these beliefs to feel obliged to obey their rules and require- ments. Empirically, we measure these aspects of legitimacy by asking citizens questions related to their evaluations of the village head’s authority and their compliance with village-level institutions. Specifically, to get at the evaluative component, we asked respondents about the trustworthiness of their village head and the quality of their own relationship with the village head. To get at levels of compliance, we asked citizens about the disposition of villagers in general to obey the village head and their own willingness to comply with the village head’s institutions. For the most part, we report outcomes on easily interpretable scales, such as the percentage of test questions answered correctly or the proportion of respondents with unresolved disputes. However, in cases where we combine multiple variables measured on different scales to capture one of the outcomes of interest, we create mean effects indices, following the approach pioneered by Kling, Liebman and Katz (2007). The full list of outcomes we consider, how we measured each outcome, and the survey instruments used to measure them are listed in table 5. The main analysis in this study was pre-registered with Evidence in Governance and Politics (EGAP), and we note any departures from the pre-registered analysis in section 8 of the appendix. 21 Table 5: Outcomes and Measurement Outcome Measurement Scale Survey VH’s knowledge of laws Proportion of questions answered 0-1 VH Survey and regulations correctly about laws and regulations governing village head’s decision- making (i.e. are there are restrictions on the disputes committed within their geographic jurisdiction over which traditional leaders can preside?) VH’s support for Average stated support for three 1-4 VH Survey individual rights individual rights (right to speak in defense at court hearings, inheritence rights, and right to protections from arbitrary authority) VH’s adherence to Index measuring whether the VH Standardized VH Survey procedures adheres to five procedures (consultation index (mean=0, (and CL Survey) of RMC, consultation of women’s council, s.d. = 1) % women on court, no court fees, public court records) Bias in VH’s decisions Proportion of respondents with different 0-1 HH Survey political views who received food aid Proportion of respondents with different 0-1 HH Survey political views who perceived most court decisions to be fair Efficacy of management Effect of need on receipt of food aid OLS regression HH Survey of problems by VH coefficient Proportion of respondents with unresolved 0-1 HH Survey land or livestock disputes Legitimacy of VH Index measuring evaluations of VH Standardized HH Survey (trustworthiness, quality of relationship) index (mean=0, and compliance with VH (perceptions of s.d.=1) disposition to obey within village, compliance with VH’s court) 22 6 Main Results We estimate the effects of horizontal pressure from civil society leaders by comparing the effects of the variant of the intervention that included civil society leaders to the effects of the variant of the intervention that did not include them. In other words, it is the difference in the effects of the two interventions that is the quantity of greatest interest to us. In tables 6, 7, 9 and 10 below, we estimate the effects of each intervention and their differential effects using the following equation: y = β0 + β1 V H + β2 CL + αj + (1) where VH is a dummy variable indicating whether the village head was assigned to be trained in either variant of the intervention, CL is a dummy variable indicating whether the community leader was also assigned to be trained, and αj are strata fixed effects for the strata used in the lottery.21 For outcomes measured at the individual level, we replace y with yi and with i, and the standard errors are clustered at the village level. The effect of the training intervention for the village head alone is β1 , the effect of the training intervention for the village head and a community leader is β1 + β2 , and the additional effect of the civil society leader is β2 . The tables below present the mean outcomes in villages not exposed to any variant of the intervention followed by all three of these effects. Table 6 considers whether the inclusion of civil society leaders led to improvements in the village head’s knowledge, adherence to procedures, and attitudes toward rights. Interestingly, training the village head alone resulted in little change in any of these outcomes. Village heads did not even have improved knowledge of the regulations governing their activities three months after the training sessions had been completed. Village heads reported following only one of five procedures emphasized during the training sessions with significantly greater frequency following the training sessions (the consultation of the resource management council), and the effect of the training on adherence to two procedures (consultation of the women’s council and access to records) was negative. As a result, the overall effect of training the village head alone on our procedural index is close to zero. The training for the village head alone also had little effect on their stated support for individual rights. 21 Table 8 uses a slightly different estimation strategy, described in detail below. 23 Table 6: Changes in Knowledge, Rights and Processes of VH Control Mean Effect of VH Effect of VH CL Effect N (and St. Dev.) Training + CL Training Knowledge test score (0-1) 0.661 -0.004 0.071** 0.075* 245 (0.231) (0.03) (0.035) (0.041) Support for rights (1-4) 2.859 -0.004 -0.014 -0.010 247 (0.402) (0.061) (0.065) (0.075) Procedures (Mean effects index) 0.0 0.106 0.581*** 0.475** 247 (1.0) (0.157) (0.168) (0.194) Consultation of resource management 0.389 0.169** 0.179** 0.010 241 council (0/1) (0.489) (0.077) (0.082) (0.095) Consultation of women’s council (0/1) 0.226 -0.069 0.151** 0.220*** 238 (0.420) (0.066) (0.072) (0.082) Proportion of women on dare (0-1) 0.262 0.025 0.053* 0.027 236 (0.019) (0.027) (0.030) (0.034) Do not charge fee (0/1) 0.336 0.05.8 0.160** 0.102 233 (0.474) (0.075) (0.080) (0.092) Access to records (0/1) 0.645 -0.102 0.005 0.107 239 (0.480) (0.071) (0.076) (0.088) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. 24 In contrast, the variant of the intervention in which a civil society leader was trained alongside the village head significantly improved both the knowledge of the village head and their adherence to procedures. It improved the village head’s score on the knowledge questions by seven percentage points, an effect that is statistically significant at the 95 percent confidence interval. It also resulted in better adherence to four of five procedures; the overall effect of training both the village head and the community leader on our procedural index is large – more than half a standard deviation – and statistically significant at the 99 percent confidence level. However, the village head’s expressed attitudes towards individual rights were not significantly improved by the variant of the intervention that included a civil society leader. Given our interest in the effectiveness of horizontal pressure on village-level reforms, the effect of great- est theoretical interest to us is the differential effect caused by including a community leader in the training. Table 6 shows this resulted in greater knowledge acquisition (an eight percentage point improvement in their knowledge score, which is statistically significant at the 90 percent level). In addition, it caused a large improvement in adherence to procedures (an improvement of half a standard deviation in our index, which is statistically significant at the 95 percent confidence level). The latter finding is unlikely to be the result of demand effects (changes in learning the correct answer, rather than changes in adherence to procedures) because we find largely similar effects (though only significant at the 90 percent level) when using the vari- able constructed from the community leader’s responses, rather than the village head’s responses.22 This indicates that that the training of civil society leaders resulted in village heads both being more aware of regulations and in being more likely to follow them. But did the empowerment of civil society leaders simply improve procedural adherence or did it also change the outcomes of village-level governance? In particular, we are interested in whether horizontal pres- sure from civil society leaders can reduce the ability of village heads to make decisions based on personal biases and thereby improve their efficacy in managing problems in two areas where they are particularly influential: food aid and dispute resolution. Table 7 shows the effects of both variants of the training and the additional effect of training a community leader on the likelihood of individuals with different political views from their village head reporting their household received food aid and perceiving most of the village head’s court decisions to be fair. In both 22 The results are reported in table 10 of the appendix. 25 instances, there is little effect of training the village head alone. The effect of training a community leader alongside a village head is larger, improving the likelihood of receiving food aid by 15 percentage points and the likelihood of perceiving a decision as as fair by 11 percentage points, both effects that are statistically significant at the 90 percent confidence level; however, the additional effect of training a community leader is not quite statistically significant in either instance. Table 7: Bias in Decision Making Control Mean Effect of VH Effect of VH CL Effect N (and St. Dev.) Training + CL Training Whether received food aid given 0.420 0.011 0.145* 0.134 363 different political views (0/1) (0.496) (0.065) (0.078) (0.089) Whether perceived most court decisions 0.439 0.001 0.114* 0.105 360 fair given different political views (0/1) (0.497) (0.071) (0.060) (0.076) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. Next we consider whether horizontal pressure improved the village head’s efficacy in managing food security in his or her village. We measure this by examining how large the effect of households’ need for food is on their likelihood of receiving food across different arms of the intervention. To do this, we estimate interaction effects between each arm of the treatment and households’ need using the following equation: yi = β0 + β1 V H + β2 CL + γ0 needi + γ1 V H ∗ needi + γ2 CL ∗ needi + αj + i (2) where needi is a dummy variable indicating whether the household reported having sufficient food from their own harvest and market purchases, the standard errors are clustered at the village level, and the effects of interest are γ1 , γ1 + γ2 and γ2 . Table 8 shows that need does not play a significantly or substantively larger role in determining who receives food aid in villages where only the village head was trained. It plays a much larger positive role in communities in which both the village head and a civil society leader were trained, although the interaction effect is not statistically significant at conventional levels in this instance 26 either. The additional effect of training a civil society leader on the role of need in determining who receives food aid is positive but statistically insignificant. Table 8: Efficacy of Food Aid Distribution Effect of Need Effect of Need Effect of Need Effect of Need N (Control Group) due to VH Training due to VH + CL Training due to CL Effect of need on -0.014 -0.012 0.081 0.094 2119 receipt of food aid (0.033) (0.060) (0.058) (0.070) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. In table 9, we consider whether the training of civil society leaders improved the village head’s efficacy in managing disputes, estimating the effects of interest using equation 1. Specifically, we examine how likely a household is to have had a land or livestock dispute in the past year that they consider unsatisfactorily resolved; these are the two most common types of civil disputes in this area of Zimbabwe. Interestingly, we find that households are slightly more likely to have an unsatisfactorily resolved land or livestock dispute in villages in which only the village head received training, an effect that is substantively fairly large – a 28 percent increase – but not statistically significant. In contrast, households are less likely to have an unsatisfactorily resolved land or livestock disputes in villages in which both the village head and another civil society leader were trained, an effect that is substantively even larger – a 32 percent decrease – though also statistically insignificant. Together, this means the additional effect of training a community leader on reducing the proportion of households with unresolved disputes is large – a 59 percent decrease – which is statistically significant at the 95 percent confidence level. Taken together, the evidence in tables 7, 8 and 9 suggest that the inclusion of a civil society leader in the training sessions caused real improvements in village-level governance. The variant of the intervention that included a community leader appears to have reduced bias against individuals holding different political views from the village head in both food aid distribution and dispute resolution, and the civil society leader variant of the intervention causes significantly better dispute management than the training variant including only village heads. Although many of the results are on the margins of statistical significance, together they 27 Table 9: Efficacy of Resolution of Disputes Control Mean Effect of VH Effect of VH CL Effect N (and St. Dev.) Training + CL Training Unresolved land or livestock dispute 0.053 0.015 -0.017 -0.032** 1913 (0.224) (0.012) (0.011) (0.014) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. tell a consistent story. The final outcome we consider is the village head’s perceived legitimacy among citizens. The effects of the two variants of the intervention and the added effect of training a civil society leader on households’ views of their village head’s legitimacy are reported in table 10. We report the effects both on our overall index of legitimacy and the four sub-components of this index. Interestingly, we find that the effect of train- ing the village head alone on our overall index of the village head’s legitimacy is negative but statistically insignificant. The negative direction of the effect is driven by the fact that households reported worse rela- tionships with the village head in these villages; training for the village head alone appears to have made citizens as a whole slightly more suspicious of the village head. In contrast, the effect of training the village head and a civil society leader on the overall index of a village head’s legitimacy is positive but statistically insignificant. The additional effect of training a civil society leader on the overall index is positive and statis- tically significant at the 90 percent confidence level. Specifically, the positive effect appears to be operating mainly through the dimension of legitimacy related to trust, suggesting including civil society leaders in training increases trust in the village head. Considering these results as a whole, we conclude that civil society leaders helped advance village-level governance reforms. They increased adherence to regulations designed to increase consultation and reduce corruption, they caused improvements in the outcomes of village-level governance, and they may even have increased the perceived legitimacy of the village head among community members. Given the difficulty of designing interventions that are effective in reforming local governance, these are important findings. 28 Table 10: Perceived Legitimacy of Village Head Mean Control Effect of VH Effect of VH CL Effect N Training + CL Training Legitimacy of VH (Mean effects index) 0.00 -0.070 0.048 0.118* 2154 (1.00) (0.061) (0.056) (0.067) How much trust VH to take right action (1-4) 3.547 -0.047 0.037 0.085* 2130 (0.728) (0.042) (0.043) (0.049) Rating of relationship with VH (0-10) 8.153 -0.288** -0.085 0.202 1975 (2.127) (0.138) (0.122) (0.152) How much think people in the village influenced 3.186 0.015 0.050 0.034 2114 by the VH’s opinions (1-4) (0.817) (0.046) (0.044) (0.051) Whether people say they would take a livestock 0.726 0.005 0.020 0.015 2150 dispute to the VH’s court if they had one (0/1) (0.446) (0.029) (0.025) (0.032) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. However, they beg the question of how civil society leaders had this impact. We turn to this in the next section. 7 Mechanisms The inclusion of civil society leaders in the training sessions for village heads could have resulted in reforms to village governance through a variety of different mechanisms. In this section, we consider the evidence in support of three mechanisms, first outlined in figure 1. In particular, we examine whether civil society leaders caused reform through (1) competition between the two sets of leaders, (2) capacity upgrading or (3) increased monitoring. We probe the plausibility of each of these mechanisms using both quantitative and qualitative evidence. 7.1 Quantitative evidence First, we use our survey data to test observable implications of each mechanism, drawing on evidence from our survey of village heads, households and especially civil society leaders. It is important to note that 29 the civil society leaders interviewed in our surveys are representative of community leaders within these villages, but they are only rarely the leaders who were actually trained alongside the village head. In our analysis, we focus on the effect of the intervention on all the independent civil society leaders included in our sample, thereby capturing the extent of changes to this group of leaders in treated villages.23 The effect of the intervention on civil society leaders as a group is both of theoretical interest in understanding the total horizontal pressures on village heads, and it is an identifiable effect, in the sense that it is possible to measure without knowing which civil society leader would have been selected for training had villages not assigned to the civil society leader treatment arm instead been assigned to it. However, outcomes measured at this level may not be sensitive enough to detect changes that affected only the trained civil society leader.24 The first mechanism through which civil society leaders could encourage reform is by competing with the village head to offer services to villagers, thereby allowing villagers to select among them as service providers and motivating both sets of leaders to offer better terms in order to keep clients. If competition is the mechanism driving village-level reforms, we would expect to see an increase in civil society leaders independently offering services in the same domains as the village head as a result of the training. In table 11, we present data that suggests this did not generally occur. This top row shows other civil society leaders were slightly less likely to be involved in distributing food aid in villages where a civil society leader was trained. The second row shows that, across all treatment arms, household respondents rarely said they would take land or livestock disputes to community leaders other than the village head for resolution, and the training of a civil society leader had little additional effect. The bottom row shows that household respondents were slightly less likely to contact civil society leaders other than the village head for assistance in villages where a civil society leader was trained. Thus, civil society leaders do not appear to have improved local governance by competing with the village head to provide services. The remaining two mechanisms – capacity upgrading and monitoring – are distinguished by the fact that the first involves only changing capacity to implement reforms the village head already favors and the second also involves changing their incentives to implement reform. Both of these mechanisms share two observable implications: the trained civil society leaders should have more information about village-level 23 In our survey of civil society leaders, we also sampled dare members. Because they were not eligible to be the “civil society leader” included in the training, we exclude them from analyses in this section. 24 We estimate there were about 5 eligible civil society leaders per village. 30 Table 11: Leadership Competition Mean Control Effect of VH Effect of VH CL Effect N Training + CL Training Whether community leaders involved in 0.242 -0.008 -0.124 -0.116 195 distributing food aid (0/1) (CL survey) (0.431) (0.074) (0.076) (0.086) Whether HHs would take dispute (land/livestock) 0.025 -0.002 0.003 0.005 2154 to CL for resolution (0/1) (HH survey) (0.156) (0.007) (0.009) (0.008) Whether HHs ask CLs for assistance 0.255 0.006 -0.033 -0.039 2151 (0/1) (HH survey) (0.436) (0.026) (0.029) (0.030) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. regulations and should interact more with village heads as a result of the intervention, in the first instance in order to help them implement reforms and in the second instance to monitor their behavior. Because our data measures effects across civil society leaders as a whole in the study communities, not on the trained civil society leaders specifically, we are underpowered with respect to identifying these effects. Even so, table 12 shows that civil society leaders as a whole appear more informed of village regulations and more likely to interact with the village head in villages subject to the civil society leader intervention. One unique observable implication of the human resource scarcity mechanism is that the effect of the civil society leader intervention should have been particularly large in places that initially had particularly low human resources. In table 13, we re-run the regressions in table 6 splitting the sample by villages with high and low initial human resources as measured in turn by the education level of the village head, the age of the village head and whether the village head’s dare consisted of more than four men. Interestingly, the evidence suggests that, if anything, the effect of the intervention was higher in high human resource communities. This is inconsistent with the human resource mechanism. The monitoring mechanism is distinguished from the human resource mechanism in that it involves civil society leaders changing not just the capacity but the incentives of village heads to implement reform. The ability of civil society leaders to affect change via monitoring depends on two things: (i) their ability to observe and recognize transgressions by the village head, and (ii) their ability to levy sanctions on them. The 31 Table 12: Civil Society Leader’s Information and Information Exchange Mean Control Effect of VH Effect of VH CL Effect Training + CL Training CL’s knowledge test score (0-1) 0.693 -0.018 0.064 0.082* 196 (CL survey) (0.214) (0.038) (0.039) (0.044) No. of times exchange information 1.758 -0.268 0.433 0.701* 189 btwn CL and VH per week (CL survey) (1.991) (0.352) (0.356) (0.407) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. Table 13: Civil Society Leader Effects by Human Resource Levels Knowledge of VH (0-1) Procedural Compliance of VH (st. index) Measure of human resource CL Effect CL Effect Difference CL Effect CL Effect Difference level High HR Low HR High HR Low HR Whether VH completed 0.072 0.076 0.004 0.793*** -0.137 -0.930** primary education (0.050) (0.070) (0.088) (0.238) (0.334) (0.417) Whether VH above 65 0.055 0.099 0.044 0.463* 0.490* 0.027 (0.057) (0.061) (0.086) (0.273) (0.289) (0.408) VH has large court 0.126* 0.028 -0.098 0.722 0.263 -0.459 (0.065) (0.058) (0.088) (0.297) (0.266) (0.401) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. 32 intervention could increase monitoring pressures by increasing either of these variables; if the civil society leader has the ability to sanction the village head, a change to the first variable would be sufficient to induce change, and vice versa. Table 12 has already shown that civil society leaders were better able to identify the regulations gov- erning the village head’s behavior and had more interaction with them as a result of the intervention. Next, we consider whether the intervention also changed the ability of civil society leaders to trigger sanctions against village heads. Civil society leaders could potentially trigger sanctions through three channels (1) by using their own status vis-a-vis the village head to levy direct social pressure (2) by serving as a fire alarm activating sanctions from higher-level leaders or (3) by mobilizing the community as a whole to pressure the village head. In table 14, we consider whether each of these channels was strengthened as a result of the intervention. In the top panel of table 14, we consider whether the civil society leader variant of the intervention increased the status of the civil society leader vis-a-vis the village head and therefore their ability to apply direct social pressure for change. We have two measures of the status of civil society leaders – the likelihood of the village head meeting civil society leaders at the civil society leaders’ homes (a proxy for the status of the leaders compared to the village head) and whether civil society leaders perceived themselves to be respected by all groups in their village (a measure of the status of the leaders in the community as a whole). Again, we measure these changes across civil society leaders as a whole and so we are underpowered to detect changes that affect the trained civil society leader only. Accepting this, the evidence in table 14 provides little support for the claim that the social standing of civil society leaders increased as a result of being included in the training. In fact, the version of the training that excluded civil society leaders appears to have increased the status of the village head vis-a-vis community leaders, and the inclusion of a civil society leader only slightly mitigates this effect. In the middle panel of table 14, we consider whether the training of civil society leaders resulted in the village head feeling more constrained by government and civil society leaders having higher levels of communication with district administrators, the most powerful bureaucrat at the district level. Inconsistent with the fire alarm channel, we find that the village head’s perceptions of their own independence from government did not decrease as a result of civil society leaders’ inclusion, and the amount of information 33 exchanged between civil society leaders and the district administrator significantly decreased when civil society leaders were included in the sessions. Thus, civil society leaders may have been less able to activate sanctions from above as a result of the intervention.25 In the bottom panel of table 14, we consider whether the civil society leader variant of the intervention increased the ability of civil society leaders to mobilize villagers to pressure the village head. If this were the case, we would expect civil society leaders to play a bigger role in organizing community members as a result of the intervention, and we might also expect villagers to have more information about the regula- tions guiding village governance. Inconsistent with this sanctioning channel, the variant of the intervention that included civil society leaders is associated with a substantively large, even if statistically insignificant, reduction in organizational activity on the part of civil society leaders in the village. We also not find any evidence that the civil society leader variant of the intervention increased households’ knowledge of village governance regulations. Thus, the evidence indicates the ability of civil society leaders to identify transgressions by the village head increased as a result of the intervention, which by itself would be sufficient to motivate reformed behavior if civil society leaders were already in a position to trigger sanctions on village heads, whether directly or indirectly. From the analysis above, it appears most likely that civil society leaders applied direct social pressure to change behavior; these leaders typically already had high social standing in their communities, even if we do not detect a measurable increase in status as a result of the intervention, and their interactions with the village head increased as a result of the intervention. In contrast, the likelihood of civil society leaders triggering pressure from above or below appears to have lessened as a result of the intervention. 7.2 Qualitative Research We also collected qualitative data from 10 villages for the purposes of better understanding mechanisms of change. Because our main goal for this research was mechanism analysis, we selected 10 villages for qualitative research that were “on the regression line” in the sense that there was little improvement in 25 We would expect civil society leaders to have higher levels of communication with superordinate polit- ical leaders, at least until a new equilibrium is established, and recall that our survey was only three months after the intervention ended. 34 Table 14: Monitoring and Sanctioning Processes Mean Control Effect of VH Effect of VH CL Effect Training + CL Training Civil Society Leader’s Status - Increased Direct Pressure VH usually meets CL at CL’s home 0.076 -0.075** -0.060 0.015 191 (0/1) (CL survey) (0.267) (0.038) (0.038) (0.043) CL’s perceptions of whether respected 0.685 -0.048 0.001 0.046 193 by all in village (0/1) (CL survey) (0.467) (0.083) (0.085) (0.097) Fire Alarm - Increased Pressure from Above VH’s perceived independence from govt 2.448 0.037 0.081 0.044 244 (1-4) (VH survey) (0.996) (0.149) (0.159) (0.182) No. times per year exchange information 0.287 0.613** -0.243 -0.857** 191 btwn CL and district administrator (CL survey) (0.911) (0.283) (0.301) (0.341) Community Mobilization - Increased Pressure from Below No. community meetings organized 5.323 0.337 -1.511 -1.848 193 by CL (CL survey) (7.643) (1.231) (1.262) (1.430) HH’s knowledge test score (0-1) 0.597 -0.005 0.001 0.006 2151 (HH survey) (0.228) (0.014) (0.013) (0.016) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. 35 governance if only the village head was trained but there was a large improvement in governance if both the village head and the community leader were trained (Lieberman, 2005).26 Once we had identified all of the on-the-line cases, we selected all cases where a positive on-the-line case was paired with a negative on-the-line case in the same geographic strata (defined by wards and land classification). This resulted in the selection of four “village head only” villages and six “village head plus community leader” villages for qualitative research, with these villages falling into four distinct geographic strata. In each of the ten villages, the qualitative research involved conducting interviews and focus groups. First, we interviewed the village head alongside advisers of his or her choosing. Second, we conducted two focus groups, one with a group of about 8 villagers organized by the village head, and one with the random sample of 8 villagers included in the follow-up survey. In addition, in each village, we conducted a one-on-one interview with the civil society leader included in the training sessions (or, in cases where no civil society leader was trained, the civil society leader the village head said he would have liked to have included in the training sessions if he had been given the opportunity). We use the qualitative evidence to consider three aspects of the dynamics of change relevant to ad- judicating the mechanism: (1) Did the civil society leader become a competitor to or integrated into the traditional village governance structure? (2) Did the civil society leaders simply change the capacity of the village head to accomplish reforms they already favored or did they also change their incentives to imple- ment reforms? (3) Did the civil society leader directly pressure the village head to accomplish change, or did he or she levy pressure indirectly by informing higher leaders or mobilizing villagers? The qualitative evidence suggests that the civil society leader who was trained alongside the village head often took on new responsibilities and powers after the training session, but these responsibilities were through existing traditional institutions, not independent of them. In particular, the trained community leaders often became advisers to the village head. Although we did not observe an increase in the status of civil society leaders as a whole vis-a-vis village heads on the cruder measures included in the quantitative survey, the qualitative evidence suggests there were important changes in how the village head viewed the 26 We focused on understanding the differences in the effects of training both the village head and a civil society leader versus only training the village head both because these are the effects of greatest theoretical interest and because the program had been rolled out in the year 2 villages by the time of the qualitative research, making a comparison to the year 1 control villages impossible. 36 trained civil society leader. For example, when we asked village heads to invite their close advisers to meet with us for discussion, 80 percent of the village heads who were trained alongside community leaders subsequently included the trained community leader in this group, while only 25 percent of village heads trained alone invited the person they subsequently said they would choose for inclusion in the training if given the option.27 In addition, we found trained community leaders were often included in the village head’s courts, but that they rarely took on independent activities in the village unless these tasks were “delegated” by the village head;28 for example, trained community leaders were in several instances tasked with adjudicating disputes when the village head was away from the village or otherwise preoccupied.29 This is inconsistent with civil society leaders acting either as competitors to the village head or independent watchdogs, but it is consistent with either the human resources or the social pressure mechanism. The qualitative evidence also suggests civil society leaders changed both the capacity and the incentives of village heads with regards to implementing reform. In interviews with the village heads, they described the civil society leader primarily as reminding and advising them about regulations and best practices, sug- gesting changes in capacity may have been particularly critical. For example, village heads described the inclusion of the community leader as beneficial because ‘I now have an aide who reminds, guides and in- forms me of the best practices to lead the people,” “I am reminded of some crucial issues that I would have forgotten since age is catching up with me,” and “Some issues that I did not understand, she helps clarify to the people.”30 However, village heads might have emphasized the “reminder” role of civil society leaders partly to save face, and interviews with civil society leaders suggest that there was often resistance by village heads to implementing some reforms, especially the requirement that they stop taking payments for hearing cases.31 A number of civil society leaders noted the importance of being persistent in getting village heads to change their behavior, describing how they needed to “argue with” the village head to get him to make 27 There would be a 17 percent chance of getting this distribution of included community leaders across the two arms of the treatment if the treatment effect were zero. Note that one of the village heads did not get the instruction to bring advisers and is not included in these calculations. 28 Interview with village head in village 22909. 29 Interview with village head in village 31729; focus group in village 23510; focus group in village 21202. 30 Interview with village head in village 20515; interview with village head in village 23510; interview with village head in village 21202. 31 Interview with civil society leader in village 20512; interview with civil society leader in village 20515; interview with civil society leader in village 33026 37 changes, that “it took a while for him to do it”, and that he would “go down fighting but ultimately give in.”32 Thus, both increased social pressure to implement reforms and improved human resource capacity to do so appear to be at work. Finally, the qualitative evidence suggests most of the pressure on the village head came directly from the civil society leader, rather than from higher-level leaders or villagers. When trained civil society leaders were asked what they could do if the village head was not making decisions in a proper manner, they all emphasized that they would approach him directly.33 A number of them also noted that if their own inter- vention was not sufficient, they would mobilize other community leaders, especially elders and members of the village court, to put pressure on the village head.34 Only in one instance did a trained civil society leader say that he could escalate the matter to higher-level authority if necessary – he specifically men- tioned higher-level traditional leaders and the police – and no one mentioned organizing the community as a whole.35 Indeed, although some civil society leaders had organized community meetings for the purpose of conveying what they had learned in the training sessions to the community as a whole, citizens in our focus groups said that efforts to inform them about regulations had generally been inadequate.36 Thus, most of the pressure on village heads to change their behavior was the result of direct pressure from civil society leaders and other village elites. Together, this suggests social pressure from civil society leaders themselves was critical in encourag- ing reformed behavior from village heads. Trained civil society leaders appear to have increased both the capacity of village heads to implement reform and given them new incentives to change their behavior. 32 Interview with civil society leader in village 20515; interview with civil society leader in village 31729; interview with civil society leader in village 33026 33 Interview with civil society leader in village 20515; interview with civil society leader in village 21202; interview with civil society leader in village 21507; interview with civil society leader in village 22909; interview with civil society leader in village 23510; interview with civil society leader in village 31729 34 Interview with civil society leader in village 20515; interview with civil society leader in village 20517. 35 Interview with civil society leader in village 31729. 36 Focus group village 20515; focus group village 21507. 38 8 Conclusion This paper has examined the possibilities for using horizontal pressure from civil society leaders to reform patrimonial village-level governance in Zimbabwe. The results suggest that empowering civil society leaders can be an effective way to reform village-level governance. In this particular context, the training of civil society leaders increased village heads’ compliance with procedures designed to increase consultation and reduce corruption, and appears to have improved their management of local issues overall. Our analysis of the mechanisms through which the trained community leaders had these effects suggests they checked power through direct social pressure. Given the challenges of using top-down or bottom-up pressure to reform local governance in many settings, the finding that horizontal social pressure from civil society leaders can be singularly effective in motivating reform is important. On the one hand, these results can be viewed as highly consistent with a long tradition of countervailing elites serving as checks on the power of rulers in Africa and beyond. For example, in pre-colonial Africa, earth priests, queen mothers and traditional councilors played key roles in restraining the power of chiefs (Baldwin, 2016; Mamdani, 1996). On the other hand, the results can be described as counterintuitive insofar as one might expect political leaders to neutralize pressure from small numbers of elites by buying them off (Gottlieb, 2015; Mattingly, 2016). There are a number of reasons why village heads may not have been able to neutralize pressure for reforms from civil society leaders by offering private payoffs in the case studied. First, many of the civil society leaders in these villages had vested interests in the procedural reforms emphasized in the training sessions: they were relatively young, female and lacking in partisan affiliations, and therefore stood to gain personally from reforms to a system that benefited elderly, male supporters of the ruling party. Second, village heads were often less wealthy than civil society leaders, and so they may not have been able to offer private payoffs sufficient to neutralize pressures for procedural changes. The ability of civil society leaders to serve as a constituency for reform is likely to vary. Still, as Tendler (1997) long ago noted, there is often sufficient diversity of interests among community leaders for them to serve as effective checks on one another’s power. In view of this, horizontal pressure from other leaders may frequently be a promising avenue for motivating local governance reforms. 39 References Acemoglu, Daron, Tristan Reed and James Robinson. 2014. “Chiefs: Economic Development and Elite Control of Civil Society in Sierra Leone.” Journal of Political Economy 122:319–368. Ahmed, Nasiruddin, Raj Chetty, Mushfiq Mobarak, Aminur Rahman and Monica Singhal. 2012. “Improving Tax Compliance in Developing Economies: Evidence from Bangladesh.”. Aker, Jenny C, Christopher Ksoll et al. 2015. “Call Me Educated: Evidence from a Mobile Monitoring Experiment in Niger.”. Alexander, J. 1996. Things fall apart, the centre can hold: processes of postwar political change in Zim- babwe’s rural areas. In Society in Zimbabwe’s Liberation War, ed. N. Bhebe and T. Ranger. Oxford: James Currey. Ashraf, Nava, Oriana Bandiera and B Kelsey Jack. 2014. “No margin, no mission? A field experiment on incentives for public service delivery.” Journal of Public Economics 120:1–17. Baldwin, Kate. 2016. The Paradox of Traditional Chiefs in Democratic Africa. New York: Cambridge University Press. Banerjee, Abhijit, Angus Deaton and Esther Duflo. 2004. “Health, health care, and economic development: Wealth, health, and health services in rural Rajasthan.” The American economic review 94(2):326. Banerjee, Abhijit V., Esther Duflo, Daniel Keniston and Nina Singh. 2015. “Improving Police Effective- ness:The Rajasthan Experiment.” MIT Working Paper. Banerjee, Abhijit V., Rukmini Banerji, Esther Duflo, Rachel Glennerster and Stuti Khemani. 2010. “Pitfalls of Participatory Programs: Evidence from a Randomized Evaluation in Education in India.” American Economic Journal: Economic Policy 2:1–30. Barr, Abigail, Lawrence Bategeka, Madina Guloba, Ibrahim Kasirye, Frederick Mugisha, Pieter Serneels and Andrew Zeitlin. 2012. “Management and Motivation in Ugandan Primary Schools: An Impact Eval- uation Report.” Partnership for Economic Policy Working Paper. 40 Berman, Sheri. 1997. “Civil society and the collapse of the Weimar Republic.” World politics 49(03):401– 429. Bjorkman, Martina and Jakob Svensson. 2009. “Power to the People: Evidence from a Randomized Field Experiment on Community-Based Monitoring in Uganda.” The Quarterly Journal of Economics 124:735– 769. Bourdillon, Michael. 1976. The Shona Peoples: An Ethnography of the Contemporary Shona. Harare: Mambo Press. Bratton, Michael. 2011. “Violence, Partnership and Transitional Justice in Zimbabwe.” The Journal of Modern African Studies 49:353–380. Cilliers, Jacobus, Ibrahim Kasirye, Clare Leaver, Pieter Serneels and Andrew Zeitlin. 2016. “Pay for locally monitored performance?”. De Tocqueville, Alexis. 1956. Democracy in America [1835]. na. Duflo, Esther, Rema Hanna and Stephen P Ryan. 2012. “Incentives work: Getting teachers to come to school.” The American Economic Review pp. 1241–1278. Gamson, William A. 1968. Power and discontent. Vol. 124 Dorsey Press Homewood, IL. Gottlieb, Jessica. 2015. “The logic of party collusion in a democracy: Evidence from Mali.” World Politics 67(01):1–36. Grossman, Guy. 2014. “Do Selection Rules Affect Leader Responsiveness? Evidence from Rural Uganda.” Quarterly Journal of Political Science 9(1):1–44. Kling, Jeffrey R., Jeffrey B. Liebman and Lawrence F. Katz. 2007. “Experimental Analysis of Neighborhood Effects.” Econometrica 75:83–119. Kosack, Stephen and Archon Fung. 2014. “Does transparency improve governance?” Annual review of political science 17:65–87. 41 Krishna, Anirudh. 2007. “How Does Social Capital Grow? A Seven-Year Study of Villages in India.” Journal of Politics 69(4):941–956. LeBas, Adrienne. 2006. “Polarization as Craft: Party Formation and State Violence in Zimbabwe.” Com- parative Politics 38:419–438. Levi, Margaret. 1988. Of Rule and Revenue. Berkeley: The University of California Press. Levi, Margaret and Audrey Sachs. 2009. “Legitimating Beliefs: Concepts and Measures.” Regulation and Governance 3:311–333. Levi, Margaret, Audrey Sachs and Tom Tyler. 2009. “Conceptualizing Legitimacy, Measuring Legitimating Beliefs.” American Behavioral Scientist 53:354–375. Lieberman, Evan S. 2005. “Nested Analysis as a Mixed-Method Strategy for Comparative Research.” Amer- ican Political Science Review 99:435–452. Lieberman, Evan S., Daniel N. Posner and Lily L. Tsai. 2014. “Does Information Lead to More Active Citizenship? Evidence from an Education Intervention in Rural Kenya.” World Development 60:69–83. Malesky, Edmund J., Cuong Viet Nguyen and Anh Tran. 2014. “The Impact of Recentralization on Pub- lic Services: A Difference-in-Differences Analysis of the Abolition of Elected Councils in Vietnam.” American Political Science Review 108:144–168. Malesky, Edmund J., Paul Shuler and Anh Tran. 2012. “The Adverse Effects of Sunshine: A Field Ex- periment on Legislative Transparency in an Authoritarian Assembly.” American Political Science Review 106:762–786. Mamdani, Mahmood. 1996. Citizen and Subject: Contemporary Africa and the Legacy of Late Colonialism. Princeton: Princeton University Press. Mansuri, Ghazala and Vijayendra Rao. 2012. Localizing development: does participation work? World Bank Publications. 42 Martinez-Bravo, Monica, Gerard Padro i Miquel, Nancy Qian and Yang Yao. 2014. “Political Reform in China: The Effect of Local Election.” Unpublished paper. Mattingly, Daniel. 2016. “Elite Capture: How Institutions Weaken Political Accountability in China.” World Politics 68:XX–XX. McCubbins, Mathew D. and Thomas Schwartz. 1984. “Congressional Oversight Overlooked: Police Patrols versus Fire Alarms.” Americal Journal of Political Science 28:165–179. Molina, Ezequiel, Ana Pacheco, Leonardo Gasparini and Andres Rius. 2013. “Community Monitoring Interventions to Curb Corruption and Increase Access and Quality in Service Delivery in Low and Middle- income Countries: A Systematic Review.” Campbell Collaboration Protocol. Munro, William A. 1998. The Moral Economy of the State: Conservation, Community Development, and State Making in Zimbabwe. Vol. 68 Ohio University Press. Murtazashvili, Jennifer B. 2016. “Afghanistan: A Vicious Cycle of State Failure.” Governance 29:163–166. Ncube, Godfrey Tabona. 2011. “Crisis of Communal Leadership: Post-Colonial local government reform and administrative conflict with traditional authorities and in the communal areas of Zimbabwe, 1980- 2008.” African Journal of History and Culture 3:89–95. Olken, Benjamin A. 2007. “Monitoring Corruption: Evidence from a Field Experiment in Indonesia.” Journal of Political Economy 115:200–249. Olken, Benjamin A. 2010. “Direct Democracy and Local Public Goods:Evidence from a Field Experiment in Indonesia.” American Political Science Review 104:243–267. Olson, Mancur. 1965. The Logic of Collective Action: Public Goods and the Theory of Groups. Cambridge: Harvard University Press. Olson, Mancur. 1982. The Rise and Decline of Nations: Economic Growth, Staglflation and Social Rigidi- ties. New Haven, CT: Yale University Press. 43 Popkin, Samuel L. 1979. The rational peasant: The political economy of rural society in Vietnam. Univ of California Press. Pradhan, Menno, Daniel Suryadarma, Amanda Beatty, Maisy Wong, Arya Gaduh, Armida Alisjahbana and Rima Prama Artha. 2014. “Improving Educational Quality through Enhancing Community Participa- tion: Results from a Randomized Field Experiment in Indonesia.” American Economic Journal: Applied Economics 6:105–126. Putnam, Robert D, Robert Leonardi and Raffaella Y Nanetti. 1994. Making democracy work: Civic tradi- tions in modern Italy. Princeton university press. Sokwanele. 2011. “The Anatomy of Terror.” Newsletter. Tendler, Judith. 1997. Good Governance in the Tropics. Maryland: Johns Hopkins University Press. Tsai, Lily. 2007. “Solidary Groups, Informal Accountability, and Local Public Goods Provision in Rural China.” American Political Science Review 101:355–372. Tyler, Tom. 2006. “Psychological Perspectives on Legitimacy and Legitimation.” Annual Review of Psy- chology 57:375–400. Veron, Rene, Glyn Williams, Stuart Corbridge and Manoj Srivastava. 2006. “Decentralized Corruption or Corrupt Decentralization? Community Monitoring of Poverty-Alleviation Schemes in Eastern India.” World Development 34:1922–1941. Weber, Max. 1922. Economy and Society: An Outline of Interpretive Sociology. Berkley: University of California Press. World Bank. 2000. Reforming Public Institutions and Strengthening Governance. World Bank. Zamchiya, Phillan. 2011. “A Synopsis of Land and Agrarian Change in Chipinge District, Zimbabwe.” The Journal of Peasant Studies 38:1093–1122. 44 Reforming Village-Level Governance via Horizontal Pressure Appendix , Shylock Muyengwa†and Eric Mvukiyehe‡ Kate Baldwin∗ ∗ Assistant Professor, Department of Political Science, Yale University † PhD Candidate, School of Natural Resources and Environment, University of Florida ‡ Economist, Development Impact Evaluation, The World Bank 1 1 Balance Statistics Table 1: Balance Statistics Variable Training No Training p-value VH Only VH & CL p-value Village characteristics Communal land 0.71 0.71 0.92 0.72 0.69 0.74 No. HHs 163 171 0.61 148 181 0.19 No. community groups 4.5 4.1 0.07 4.4 4.6 0.43 VH characteristics Female VH 0.05 0.09 0.27 0.05 0.05 0.85 VH with prim. education 0.66 0.72 0.30 0.66 0.65 0.99 Age of VH 65 64 0.56 65 64 0.74 Tenure of VH 15 13 0.46 14 16 0.59 VH with ZANU-PF sign 0.28 0.31 0.52 0.28 0.27 0.92 HH characteristics Villagers with prim. education 0.80 0.81 0.70 0.80 0.81 0.49 Age of villagers 43 43 0.43 44 42 0.06 Villagers born in village 0.47 0.49 0.47 0.48 0.46 0.58 HH member in wage labor 0.31 0.30 0.85 0.30 0.31 0.86 Own plots 0.88 0.86 0.05 0.88 0.89 0.62 Newspaper readers 0.20 0.20 0.87 0.22 0.18 0.09 Cattle wealth index 2.1 1.8 0.12 2.3 1.9 0.09 Diff. political views from VH 0.30 0.27 0.30 0.27 0.27 0.97 2 2 Dimensions of Village Head Bias Table 2: Pro-ZANU-PF Biases in Village-Level Decision Making Not ZANU-PF ZANU-PF Difference Received Food Aid Last Year 0.448 0.434 -0.014 (0.023) (0.038) (0.044) Believe Most Decisions by 0.584 0.671 0.137*** Village Head’s Court Fair (0.023) (0.036) (0.036) Took Case to Village Head’s 0.439 0.451 0.012 Court if Had Dispute (0.023) (0.036) (0.044) The first two columns report means with standard errors below in parentheses. The third column reports the differences between column two and column three, with the standard error of the difference reported in parentheses below. Only villages that did not receive any training in year 1 are included in the calculations. Table 3: Gender Biases in Village-Level Decision Making Female Male Difference Received Food Aid Last Year 0.405 0.346 -0.059** (0.021) (0.021) (0.030) Believe Most Decisions by 0.602 0.602 0.000 Village Head’s Court Fair (0.021) (0.021) (0.030) Took Case to Village Head’s 0.441 0.469 0.028 Court if Had Dispute (0.023) (0.021) (0.031) The first two columns report means with standard errors below in parentheses. The third column reports the differences between column two and column three, with the standard error of the difference reported in parentheses below. Only villages that did not receive any training in year 1 are included in the calculations. 3 Table 4: Kinship Biases in Village-Level Decision Making Non-Related Related Difference Received Food Aid Last Year 0.360 0.385 0.025 (0.022) (0.021) (0.030) Believe Most Decisions by 0.550 0.643 0.093*** Village Head’s Court Fair (0.023) (0.021) (0.030) Took Case to Village Head’s 0.431 0.470 0.039 Court if Had Dispute (0.022) (0.021) (0.031) The first two columns report means with standard errors below in parentheses. The third column reports the differences between column two and column three, with the standard error of the difference reported in parentheses below. Only villages that did not receive any training in year 1 are included in the calculations. 3 Disputes Taken to Village Heads Table 5: Village Heads and Dispute Resolution % HHs with % taken to dispute in VH past year Land Disputes 18 74 Crop Destruction 49 77 Theft 27 47 Witchcraft 8 43 Domestic Violence 10 25 Marriage Disputes 14 16 Assault 10 33 4 4 Community Leaders Included in Training Table 6: Comparison of Trained Civil Society Leaders to Sample % Trained % Surveyed Community Leader Type Caregiver/ Village Health Worker 54 33 Church Leader 8 19 NGO Officer/Focal Point 6 0 Farmers Group Representative 9 17 School Development Committee Member 4 2 Secretary/Treasurer/Member of Village Committee 19 15 War Veteran 0 4 Youth Chairperson 0 4 Other 1 6 Community Leader Gender Male 45 52 Female 55 48 5 5 Measuring Political Views We measured whether respondents have different political views from their village heads by asking whether they agreed or disagreed with the statement "My own political views are very similar to those of the vil- lage head” as part of our follow-up survey. This was measured post-treatment and as a result there may be concerns that the treatment changed the number or types of people who disagreed with this statement, especially because the purpose of the intervention was to constrain the ability of the village head to make politically biased decisions. However, in the context of Zimbabwe, where people’s partisan affiliations are generally well known and very polarized (LeBas 2006), we think it unlikely that the intervention would affect perceptions of the village head’s partisanship in the medium-term; rather, the goal was to affect the village head’s behavior towards non-co-partisans. Indeed, the following table shows that the characteristics of the respondents identifying as having different political views from their village head is similar across treatment arms. Of 18 comparisons, just two are statistically significant at the 90 percent confidence level and just one is statistically significant at the 95 percent confidence level; this is consistent with what we would expect to see by chance in the absence of systematic differences between the comparison groups. Particularly important is the fact that respondents who identify as having “different political views” from their village heads do not have significantly different levels of affiliation with ZANU-PF or the major oppo- sition party, the Movement for Democratic Change (MDC), across the treatment arms. Across the different arms of the experiment, between 23 and 26 percent of them identify as ZANU-PF supporters and between 34 and 45 percent of them identify as MDC supporters.1 Unfortunately, even though the district authorities initially approved a questionnaire containing this question, we were asked to remove it two thirds of the way through surveying, so we do not have data on this question for one third of respondents.We randomized the intervention by geographic blocks and were proceeding geographically. The following table considers how the communities in which we were able to ask this question differ from the communities in which we were not able to ask this question. Out of 15 comparisions, we find two differences that are statistically significant at the 95 percent confidence level and one difference that is statistically significant at the 99 percent confidence level, slightly more differences 1 The MDC split into two factions prior to the 2013 election. Respondents who identified with either faction are coded as MDC members. 6 Table 7: Characteristics of Respondents with Difference Political Views from VH by Treatment Arm Variable Training No Training p-value VH Only VH & CL p-value Villagers with prim. education 0.87 0.89 0.64 0.89 0.86 0.53 Age of villagers 41 40 0.42 43 40 0.16 Villagers born in village 0.49 0.52 0.61 0.47 0.52 0.46 HH member in wage labor 0.37 0.31 0.24 0.32 0.42 0.15 Own plots 0.86 0.82 0.28 0.79 0.93 0.01** Newspaper readers 0.18 0.25 0.10 0.17 0.19 0.78 Cattle wealth index 2.2 1.9 0.33 1.8 2.6 0.10* Support for ZANU-PF 0.26 0.23 0.43 0.26 0.26 0.98 Support for MDC 0.42 0.34 0.15 0.45 0.38 0.35 N 645 640 334 311 % of all respondents 0.30 0.27 0.34 0.28 0.27 0.87 than can be attributable to chance. This suggests the sample for which we have political information is different than the sample for which we do not. Specifically, it is made up of smaller, more recently settled villages. 7 Table 8: Differences in Communities with and without Political Data Variable Political Data No Political Data p-value Village characteristics Communal land 0.74 0.66 0.16 No. HHs 185 142 0.01** No. community groups 4.3 4.3 0.97 VH characteristics Female VH 0.06 0.08 0.64 VH with prim. education 0.71 0.66 0.40 Age of VH 65 63 0.40 Tenure of VH 15 12 0.10 VH with ZANU-PF sign 0.33 0.25 0.20 HH characteristics Villagers with prim. education 0.81 0.80 0.50 Age of villagers 43 43 0.81 Villagers born in village 0.51 0.43 0.001*** HH member in wage labor 0.31 0.30 0.77 Own plots 0.86 0.88 0.28 Newspaper readers 0.20 0.21 0.49 Cattle wealth index 1.9 2.0 0.68 8 6 Measuring Need We measure the need of households for food aid in the past year based on their responses to a series of questions: “In the last twelve months, did you harvest enough grain to feed your family without buying it on the market?” “In the last twelve months, did you harvest enough grain to feed your family after buying it on the market?” Households that did not have enough grain to feed their families through their own harvest and their own market purchases are classified as “needy.” Table 9 shows the distribution of needy families across treatment arms. Table 9: Food Aid Need by Treatment Arm Variable Training No Training p-value VH Only VH & CL p-value Needy 0.655 0.644 0.663 0.640 0.689 0.092 9 7 Changes in Process as Reported by Community Leaders Table 10: Changes in Process as Reported by Civil Society Leaders Control Mean Effect of VH Effect of VH CL Effect N (and St. Dev.) Training + CL Training Procedures (Mean effects index) 0.00 -0.022 0.284* 0.306* 257 (1.00) (0.156) (0.156) (0.182) Consultation of resource management council (%) 0.439 0.040 0.003 -0.037 223 (0.498) (0.081) (0.085) (0.096) Consultation of women’s council 0.150 0.007 0.056 0.048 237 (0.359) (0.060) (0.061) (0.071) Proportion of women on dare 0.303 -0.028 0.014 0.042 257 (0.205) (0.030) (0.030) (0.035) Do not charge fee 0.204 0.045 0.125* 0.079 213 (0.405) (0.071) (0.069) (0.081) Access to records 0.600 -0.030 0.051 0.081 234 (0.492) (0.078) (0.076) (0.091) Table displays standard errors in parentheses. *, ** and *** indicate significance at the 90, 95 and 99 percent confidence levels respectively. 10 8 Deviations from Pre-Analysis Plan We registered a pre-analysis plan for our impact evaluation prior to receiving the data. This plan was based on intensive discussions with the implementing partner about the outcomes they expected their programming to achieve on governance by village heads and community-level conflict. In this paper, we focus on the differential effects of the two variants of the intervention (one that included a community leader monitor and one that did not) on governance outcomes; we plan to discuss the effects of the intervention on community- level conflict more broadly in a separate paper. Within the category of “governance”, we registered four outcomes – the village head’s knowledge of the law, the village head’s attitudes towards human rights, the impartiality of the village head, and the village head’s legitimacy. In our paper, we also look at one additional outcome, the village head’s governance procedures. We asked a series of questions about the processes by which the village head makes decisions and manages con- flict. We did not register them because the implementing partner did not emphasize them as a key outcome of their program. This was an oversight on our part, as the training curriculum placed great emphasis on the adoption of inclusive and transparent decision-making processes, and this is a theoretically important inter- mediate step in reducing bias, especially in dispute resolution. The section on decision-making procedures runs from D18-D24 of the village head survey. In all instances in which significant numbers of village heads were not adopting prescribed procedures or best practices prior to the intervention, we consider the effect of the training sessions on the adoption of the procedure. We have also made a few adjustments to our measurement of the registered outcomes in cases where we felt the registered measurements suffered from conceptual problems or were not consistent with the measurement strategies employed by other scholars conducting research in this area. We discuss these adjustments below, while noting that none of these shifts change the statistical significance of the results. As specified in the pre-analysis plan, we are interested mainly in the village head’s decision-making in two areas – reducing food insecurity and dispute resolution. For both areas, we look at both (a) the amount of bias in the village head’s decision-making in the area and (b) the effectiveness of the village head in managing problems. 11 Food Aid Outcomes Bias: Likelihood of people with different political views from the VH receiving food aid Efficiency: Effect of measured need on the receipt of food aid Both of these analyses were pre-registered, but as part of our secondary analysis because we were con- cerned about looking at the effect of the intervention only on certain societal sub-groups. Specifically, we were concerned that the intervention could potentially change the types of people who reported having dif- ferent political views from the village head and the types of people who have measured need before receiving food aid. We have given these results greater weight in this paper because auxiliary analysis suggests the intervention did not change the types of people falling in these sub-groups, and because other recent pa- pers looking at the effects of monitoring and governance interventions on food aid have constructed similar measures by societal sub-groups (Banerjee et al. 2015; Beath et al. 2015). In contrast, we do not discuss one pre-registered analysis, which involved using a list experiment to measure perceptions of bias in food aid distribution over the previous five years. In retrospect, it was not realistic to expect our follow-up study, which happened less than a year after the intervention began, to affect this measure. Dispute Resolution Outcomes Bias: Likelihood of people with different political views from the VH viewing most decisions by the VH’s court as fair Efficiency: Likelihood of people having an unsatisfactorily resolved land or crop destruction dispute In the pre-registered analysis, we proposed looking at the likelihood of all people viewing the VH’s decisions as fair. However, because a reduction in bias against groups that previously experienced discrim- ination should result in a decrease in the favorable treatment accorded to the previously privileged group, we have revised the analysis to focus on perceptions of fairness within the group suffering discrimination. In our pre-analysis plan, we also proposed to look at the resolution of all types of disputes within a village. However, in this paper, we focus only on land and crop disputes because these are the only two types of disputes that a majority of households in Eastern Zimbabwe take to village heads for resolution if they have 12 them, and so they are the two measures of conflict most clearly related to the village head’s governance. The focus on unresolved disputes is in line with strategies of measuring conflict reduction in recent studies in political science (Blattman et al. 2014) and is a clearer measure of the village head’s ability to manage conflict than the number of conflicts, especially in the short to medium-term. The other sections of the analysis closely follow the pre-analysis plan, with the following slight departures: (1) In the section on legitimacy, we initially proposed combining measures from the village head’s survey (or his or her perceived legitimacy) and measures from the household survey. Ultimately, we decided that it was conceptually clearer and theoretically more important to focus on community member’s perceptions of their village head’s legitimacy. (2) Rather than dichotomizing ordered outcome variables, we treat them as continuous outcomes, as is becoming standard practice in econometrics. (3) Although we initially proposed standardizing all measures and before combining them into indices, in cases where many variables to be combined are on the same scale, we have instead sought ways to combine them that are more easily interpretable (i.e. rather than standardizing and then summing and restandardizing variables on the same scale, we have preferred to average or sum them as described in the text). In cases where the variables to be combined are on different scales, we use the originally proposed method. (4) We employ strata fixed effects for the blocks over which the original randomization was done, as we consider this to be best practice. We also note that the analysis of mechanisms in the last section of the paper – based on both qualitative and quantitative analysis – goes well beyond the originally specified analysis. References Banerjee, Abhijit, Rema Hanna, Jordan Kyle, Benjamin Olken and Sudarno Sumarto. 2015. “The Power of Transparency: Information, Identification Cards and Food Subsidy Programs in Indonesia.” NBER Working Paper #20923. Beath, Andrew, Fotini Christia and Ruben Enikolopov. 2015. “Do Elected Councils Improve Gover- nance? Experimental Evidence on Local Institutions in Afghanistan.” 13 Blattman, Christopher, Alexandra Hartman and Robert Blair. 2014. “How to Promote Order and Prop- erty Rights under Weak Rule of Law? An Experiment in Changing Dispute Resolution Behavior through Community Education.” American Political Science Review 108 (1): 100-120. 14