Policy Research Working Paper 9140 Assessing the Longer Term Impact of Community-Driven Development Programs Evidence from a Field Experiment in the Democratic Republic of Congo Eric Mvukiyehe Peter van der Windt Development Economics Development Impact Evaluation Group January 2020 Policy Research Working Paper 9140 Abstract Community-driven development programs are a popular team returned to these villages in 2015, eight years after model for service delivery and socioeconomic development, the onset of the program. The study finds evidence of the especially in countries reeling from civil strife. Despite their physical endurance of infrastructure built by the program. popularity, the evidence on their impact is mixed at best. However, it finds no evidence that the program had an Most studies thus far are based on data collected during, impact on other dimensions of service provision, health, or shortly after, program implementation. Communi- education, economic welfare, women’s empowerment, gov- ty-driven development’s theory of change, however, allows ernance, and social cohesion. These findings suggest that, for a longer time frame for program exposure to produce although community-driven development programs may impact. This study examines the longer term impact of effectively deliver public infrastructure, longer term impacts a randomized community-driven development program on economic development and social transformation appear implemented in 1,250 villages in Eastern Democratic to be limited. Republic of Congo between 2007 and 2012. The study This paper is a product of the Development Impact Evaluation Group, Development Economics. 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://www.worldbank.org/prwp. The authors may be contacted at emvukiyehe@worldbank.org and petervanderwindt@nyu.edu. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Assessing the Longer Term Impact of Community-Driven Development Programs: Evidence from a Field Experiment in the Democratic Republic of Congo* Eric Mvukiyehe (World Bank Research Department) Peter van der Windt (New York University – Abu Dhabi) Keywords: Community Driven Development, Service delivery, Social cohesion, Economic development, Field experiment, Congo. JEL codes: C93, D02, P48 * We thank Marie-France Guimond, Macartan Humphreys, Ann Laudati, Guillaume Labrecque, Helen Poulsen, Dana Olds, Daniel Rogger, Sarah Furrer and Maarten Voors. We thank our 94 enumerators, Alain Manda, Félix Toshibangu, Fils Likuta, Justin Kalombo, Christian Kalombo, Eustache Kuliumbwa, Summer Lindsey, Patrick Milabyo and the late Jean Paul Zibika for in-country work. Thanks to Han Il Chang, Laura Moreno and Nausheen Khan for research assistance. We acknowledge financial support from DFID, the World Bank and NYU Abu Dhabi. This study has been pre-registered at: http://egap.org/registration/1773 (deviations can be found in Appendix H). IRB approval from Wageningen University. Computational reproducibility verified by DIME Analytics. Introduction Community Driven Development (CDD) – a development approach that gives control of decisions and resources to community groups (Dongier et al, 2002) – has become a popular model for channeling foreign aid into local development (Mansuri et al. 2013). The typical CDD program consists of two components. First, communities obtain block grants for local infrastructure projects. Second, communities are responsible for project selection and implementation. This second component often comes with additional activities and requirements to promote democratic decision-making and inclusion of marginalized groups in project implementation and management. Proponents of the CDD approach suggest that the combination of both components not only leads to better targeted and more sustainable investments in infrastructure, but also has the potential to improve other outcomes such as economic welfare, women’s empowerment, social cohesion and good governance. Over the past two decades, there has been a growing number of randomized evaluations of CDD programs in a variety of countries, most notably in Afghanistan (Beath et al. 2013, 2016), the Democratic Republic of Congo (Humphreys et al. 2019), Liberia (Fearon et al. 2015), Sierra Leone (Casey et al. 2013) and Sudan (Avdeenko et al. 2015). The accumulated evidence suggests that CDD programs can effectively deliver on local public infrastructure (Casey 2018). However, there is mixed evidence of their impact on economic welfare and little evidence on social outcomes like governance and social cohesion.2 These studies have one thing in common: they measure impacts during or shortly after the CDD program. Figure 1 presents the start and end dates of the above mentioned CDD programs, and their related study’s data collection period. It shows that data collection, on average, starts around 3.2 years after program onset, and 7 months before the end of the CDD program.3 To date, we thus know little about the longer term impact of CDD programs. A better understanding of the longer term impact of CDD programs, however, is important for two reasons. First, there is an unsettled debate about whether the null effects of CDD programs thus far are because of theory or evaluation timing. That is, the average evaluation timeline in existing studies may be too short to pick up effects. Certain impacts by their nature may take longer 2 See Wong (2012), King and Samii (2014) and Casey (2018) for a summary of accumulated evidence. 3 Program and data collection start and end dates are based on the published article, the donor report, or communication with the authors. For the calculations we exclude Casey et al. (2019) to which we return later. 2 to appear, such as governance spillover from program to village arenas or education learning outcomes (Wong 2012). Figure 1. CDD Program Overview Note: Overview of studies related to randomized CDD programs. Dotted (black) lines indicate CDD program. Solid (red) lines indicate data collection. CDD program name in parenthesis next to country name. Second, a major purported benefit of the CDD approach is the longer term sustainability of infrastructure. Proponents claim that community control over planning decisions and investment resources results in better use and maintenance of the CDD‐produced infrastructure (e.g. Dongier 2002). CDD programs are therefore claimed to be more sustainable compared to when investment decisions are made by actors outside the community. To date, however, there is little evidence to support (or reject) this claim. In response, in 2015, we returned to the Democratic Republic of Congo and built on the randomization of a major CDD program, “Tuungane”. The program was implemented from 2007 onwards, and its short-term effects were studied by Humphreys et al. (2019) who collected data in 2010 and 2011. We focus on the Tuungane program for several reasons. First, the program was well implemented with high levels of exposure and compliance (see Humphreys et al. (2012) for details). Second, the program has many elements in common with other CDD programs. Third, 5 Tuungane was implemented with a variation in treatment, which allows us to explore, in addition to the program’s overall effect, the impact of having women in leadership positions. Fourth, Congo provides a good setting because CDD programs often take place in the context of weak state capacity. We collected outcome data in 735 of a targeted 781 villages, which makes this one of the largest CDD field experiments to date. As a comparison, Fearon et al. (2009) examine 83 villages, Casey et al. (2013, 2019) survey 236 villages, Beath et al. (2013, 2015) study 217 villages, while Avdeenko and Gilligan (2015) investigate 24 communities. This study builds on data from village chief and household surveys and a carefully-designed facility audit to explore the longer term impacts of the Tuungane program on eight outcome families that we pre-registered prior to data collection: service provision (in the health and education sector), health, education, economic welfare, women’s empowerment, governance, intra-village and inter-village social cohesion.4 We find evidence for the physical endurance of infrastructure built by the program. Treatment villages are served by primary schools and hospitals that have higher quality infrastructure, and their hospitals are better stocked. However, we find no evidence that the program had an impact on other dimensions of service provision, health, education, economic welfare, women’s empowerment, governance, and social cohesion. These findings are broadly consistent with findings from Casey et. al (2019); the only other study, we know of, that investigates the longer term impacts of CDD programs. Thus, while CDD programs appear to effectively deliver public infrastructure in the short and longer term, their impact on economic development and social transformation appears to be limited, even in the longer term. The remainder of this paper proceeds as follows. Section 2 discusses the Tuungane program and the experimental design. Sections 3 and 4 discuss the data collection and present results, respectively. Section 5 concludes. Field Experiment in Congo The field experiment was designed around the Tuungane program, which was implemented between July 2007 and December 2012 in 1,250 villages throughout Eastern Congo (Appendix A). With, on average, 1,424 inhabitants per village, the program reached a beneficiary population of approximately 1.8 million people. 4 For the conceptual frameworks of how CDD may affect each outcome, we refer the reader to the earlier 5 mentioned studies. In 2006, prior to program rollout, villages were randomly assigned to the program. The process ran as follows. The implementing partner aggregated all villages into 560 village clusters. Clusters were, in turn, aggregated into 83 lottery bin areas, which largely corresponded to Chiefdoms. Next, half of the clusters in each lottery bin were selected for treatment using public lotteries. This approach improves balance between treatment and control by geographic features, including remoteness, poverty, institutions, and social composition. In total, 280 village clusters and 1,250 constituent villages were selected for treatment. The remaining villages were assigned to the control status. The program was implemented in two phases: a village-level phase and a subsequent village cluster level phase. At the village level, local election teams were established and trained to mobilize and guide village populations. The idea was to ensure a good understanding of the program as well as the subsequent election for newly created village-level management committees. These ten-member strong committees were required to contain five men and five women. Next, these committees, in consultation with the population, decided how to allocate an envelope of $3,000 for a maximum of two projects. The proposed project(s) was then voted on by the whole village. In the two years following project selection, the committees were responsible for project implementation, and were held accountable by village populations via regularly scheduled town hall meetings. Subsequently, program activities took place at the village cluster level. A new village cluster-level committee was formed by selecting members from the constitutive village-level committees, again ensuring that half of the members were women. Each village cluster received a block grant of $50,000 to $70,000 (depending on population size) to implement infrastructure projects that were chosen by the inhabitants of the constituent villages via an election. Next, the cluster committees were responsible for project implementation and were held accountable by cluster populations. In total, 2,335 village-level and 315 village cluster-level projects were undertaken. Appendix B gives an overview of implemented projects. The majority of these projects took place in the education and health sectors. The program was implemented with a variation in treatment related to the gender composition of the management committees. Specifically, in 43 (not randomly selected) lottery 5 bins, half of the village clusters were randomly selected to enter a “gender parity lottery”.5 Among these, half of the village clusters (74 clusters and all 325 villages in those clusters) were selected to be free to choose the gender composition of their management committees. In the other clusters (75 village clusters, 337 villages), the program was implemented as normal; i.e. with an obligatory equal number of men and women in the committee. This design feature allows us to learn about the impact of having women in leadership positions. The short-term impact of the program was assessed by Humphreys et al. (2019), who collected data between October 2010 and October 2011; after the onset of village-level projects, but well before the end of the village cluster-level projects. Appendix C illustrates the timing of the village and village cluster phases of the program, and the short-run data collection. This study leverages the same design as Humphreys et al. (2019) but returns to the villages in 2015, eight years after program onset and three years after all program activities have concluded. Outcomes, Data and Empirical Strategy This study focusses on eight outcome families: service provision, health, education, economic welfare, women’s empowerment, governance, intra-village cohesion, and inter-village cohesion.6 We collected data between June and December 2015, targeting 781 villages in the provinces of Haut Katanga, South Kivu and Tanganyika.7 Data were collected from four sources. In each village, surveys were conducted with the village chief and a randomly selected adult from five randomly selected households per village. In addition, as part of the household survey, in each household with children of school-going age (between 6 and 11 years old), one child was randomly selected for a brief exam. Finally, enumerators visited the village’s primary school and health facility.8 At each facility, enumerators undertook three activities. They conducted a carefully designed audit to assess the quality of the infrastructure and the presence of materials and 5 In total, 149 village clusters (661 villages) entered the lottery. 6 Note that the initial goal of the Tuungane program was improvements in governance, social cohesion and economic welfare (Humphreys et al., 2012). 7 Humphreys et al. (2019) targeted two randomly selected villages in each of the 560 clusters (280 treatment, 280 control). In total, 816 villages out of the targeted 1,120 villages were visited. There are no differences in attrition by treatment condition (Humphreys et al. 2012). In 2015, we targeted the same villages as visited for Humphreys et al. (2012), excluding the Maniema province for logistical reasons. These villages sum to 781. 8 We are interested in service provision as experienced by villagers. We therefore visited the main primary school and health facility for each village, thus not necessarily the Tuungane-built facilities in treatment areas. 6 equipment. They undertook interviews with users of the facility, visiting a randomly selected ongoing class in each school for close observation and interviewing a randomly selected patient in each health facility. Finally, they interviewed the director responsible for each facility. In total, data were collected from 3,379 households in 735 villages, 610 primary schools, 504 health facilities, and 1,496 children’s exams were conducted.9 These data provide us with 171 outcome variables (definition and summary information can be found in Appendix E). To avoid over-rejection of the null hypothesis due to multiple inference (Anderson 2008), we committed, in advance, to a mean effects approach. That is, we reduced the effective number of tests we conduct by combining the individual measures into eight family outcomes (Kling et al. 2007), of which the individual components were pre-registered. In Appendix F, we show that the randomization procedure was successful in ensuring substantive balance across treatment arms. This study's empirical strategy is therefore straightforward. We compare mean outcomes between treatment and control areas, and – for those areas that participated in the parity lottery – between gender parity and non-parity areas. These analyses provide unbiased estimates of the average treatment effect (Rubin 1974). Specifically, we estimate (for individual-level outcomes) an equation of the following form: yijk = β0 + βiTj + vk + εi (1) Where i indicates the individual, j indicates the village, and k the lottery block, and T is the treatment status (i.e., assignment to the Tuungane program). We use lottery bin fixed effects to control for average differences in observable or unobservable predictors across lottery bins, and we cluster our standard errors at the village cluster level. Results Panel (a) of Figure 2 presents results for the eight outcome families. Appendix G provides results for the outcome measures individually.10 Each circle is a point estimate and bars (ticks) present 95% (90%) confidence intervals. The dependent variables are standardized. We report sample 9 Further details related to the data sources and attrition can be found in Appendix D. Attrition took place for a number of reasons, including inaccessibility of some regions for security reasons, as well as the loss, damage, and theft of tablets. Rates of attrition of these sources are balanced across treatment groups. 10 To create the family indices we took the following decisions (similar to Humphreys et al., 2012). We do not impute data for missing observations. When a unit has individual outcomes missing, the family index is constructed based on the remaining non-missing observations. Finally, when a family index is based on individual outcomes from different levels, we conduct the analysis at the village level. Results related to the outcome families mirror those from the individual outcomes. 7 average treatment effects, ignoring minor differences in sampling individuals in differently sized households and differently sized villages within clusters. Service Provision Did the CDD program deliver and maintain local public goods, and improve the local population’s access to, and the quality of, services provided? We find no evidence that Tuungane had a positive impact on average levels of service provision in the health and education sector. The magnitude of 0.05 standard deviation (SD) is small and not significant. In panel (b) of Figure 2, we break up the results by sector and the seven dimensions that make up service provision: infrastructure quality, capacity, availability of material and equipment, staff quality, administration quality, facility-community interactions and the costs and use of health facilities. We find that eight years after program onset, the infrastructure quality of hospitals and school buildings is significantly higher in treatment areas. The quality of hospital buildings is 0.16SD higher in Tuungane areas, which is largely driven by higher quality floors, higher quality walls, and a better rating of facility quality by villagers in the hospital’s catchment area (see Appendix G). We find similar results in the education sector. School buildings in treatment villages score 0.19SD higher than those in control villages, largely driven by higher quality floors and roofs, the presence of windows and the villagers’ rating. We also find that health facilities in Tuungane areas are significantly better stocked (0.22SD).11 We do not find evidence that schools in Tuungane areas are better stocked. We also find no evidence that the Tuungane program improved the capacity, staff quality, administration quality, facility-community interactions and the costs and use of health facilities. Health The health situation in Congo is dire. For example, in control communities, 15% of respondents mention that, within the household, a child younger than five years old passed away during the previous year (a number similarly found in the DHS 2013). Panel (a) of Figure 2 shows that there is no evidence that Tuungane improved health outcomes. The program had no impact on under- 11 That is, enumerators calculated the number of antibiotics, anti-malaria and anti-inflammatory tablets present in the hospital’s stockroom. 8 five mortality, nor did it decrease the incidence of household members falling severely ill or passing away, as reported by respondents. Figure 2. Main Results and Service Provision Note: Bars (ticks) indicate 95% (90%) confidence intervals. Fixed effects at lottery bin level. Errors clustered at the randomization unit level. Education The program also had no longer term impact on education outcomes. Households in program villages do not score better on measures related to school attendance. In addition, we find no differences in children’s exam scores. Specifically, in households with children of school-going age, we randomly select one child and ask two questions about mathematics, French and science. The six questions used were informed by the national curricula for primary schools, and dependent on the child’s age.12 Enumerators first asked the question in French (the official language of education) and repeated the question in the local language if the child had difficulties understanding the question. On average, children answered 1.94 (2.58) out of the six French- instructed (local language-instructed) questions correctly. These numbers, however, are similar in treatment and control villages. 12 Because many children do not attend school, the six questions differed by the child’s age, not their current grade. 9 Economic Welfare We explore the impact of Tuungane on economic welfare across a wide set of indicators. Enumerators recorded the material from which the roof and wall of the respondent’s house was constructed. In addition, they asked the respondent about the household’s asset ownership across 23 items. We also collected detailed information on household spending across ten categories in the month preceding the survey, and household income the preceding week. Figure 2’s panel (a) shows that households in Tuungane areas are, in fact, 0.08SD worse off than in control areas; a result that is statistically significant (p<0.05) but mainly driven by one indicator: lower quality roofs in treatment areas. Women’s Empowerment Many elements of the Tuungane program emphasized women’s empowerment. The village and village cluster management committees were, by default, gender balanced. Committee member trainings were conducted by the implementing partner and, among others, focused on the needs for women’s participation. Furthermore, efforts were undertaken to sensitize village populations to the need of women’s inclusion in committee elections and project choice. We collected a wide set of indicators to measure women’s empowerment, including the respondent’s opinion about a statement related to women’s rights, a combined measure of the respondent’s opinion about eight statements related to domestic violence towards women, the presence of women’s association in the community, the proportion of girls to boys who are currently going to school, who have never been to school, and the share of members of the local development committee that are women. We find no evidence that Tuungane improved the role of women in society.13 Across the six outcome measures, only two tend positively (opinion about domestic violence and committee membership), but are not statistically significant. Governance We followed Humphreys et al. (2012) and separate out governance across five dimensions: participation (the extent to which villagers are willing and able to be part of public decision making), accountability (the willingness and ability of community members to sanction leaders 13 These results reflect the short run findings in Van der Windt (2018). We obtain similar results when we restrict the analyses to the villages with gender parity. 10 for poor performance, and the willingness of leaders to respond to citizen requests), transparency (accessibility of information related to public decision making), efficiency (the ability to organize in order to achieve ends), and capture (the extent to which benefits of public projects are broadly distributed). Panel (a) of Figure 2 shows no evidence that the program had an impact on overall levels of good governance. The magnitude of 0.04SD is small and not significant. In panel (a) of Figure 3 we show the result for each governance dimension. We find that Tuungane had a positive impact on villagers’ participation in decision making (0.18SD, p<0.01), a result largely driven by villagers' participation in village meetings that took place during the six months preceding the survey and in the 2011 elections. Estimated effects for the other four participation measures are not statistically significant. We also find no evidence that Tuungane improved levels of accountability, transparency, efficiency and capture. Figure 3. Governance and Gender Parity Results Note: Bars (ticks) indicate 95% (90%) confidence intervals. Fixed effects at lottery bin level. Errors clustered at the randomization unit level. Panel (b) limits to only the 190 villages that participated in the gender parity lottery. Intra-village Cohesion We also examine Tuungane’s impact on social cohesion within the village. We make use of a large set of measures to measure intra-village cohesion: individuals’ opinion about divisions in the community and about whether voluntary projects have taken place in the village, and the village 11 chief’s opinion about the use of community resources, the presence of a village development committee, and the existence of associations in the village. In addition, we conducted behavioral games to gauge respondents’ trust towards a random fellow villager and the village chief. Across all nine individual measures, no effect is positive and significant. Inter-village Cohesion The village cluster component of the Tuungane program accounted for a considerably larger part of program expenditures than the village component (village-level projects received $3,000 in funding, while cluster-level projects received $50,000 to $70,000). As part of this program phase, multiple villages had to work together. We thus also explore cohesion across villages. We find no evidence that there are fewer cleavages across villages or more cooperation with other villagers, either by community organizations or the village chief, in program areas. Respondents also played a trust game towards a random villager from a neighboring village. Contributions are similar across treatment conditions. Panel (a) of Figure 2 shows that the magnitude on the family measure is almost zero. Impact of Gender Parity The Tuungane program, by default, created management committees that consisted of the same number of men and women. In a random subset of villages, this parity requirement was dropped. This design feature allows us to learn about the causal impact of having women in leadership positions. Program documents corroborate that there are significantly fewer women on village committees in Tuungane areas where gender parity was not mandated (3.1 women) compared committees where it was (4.7 women).14 Panel (b) in Figure 3 shows the impact of having more women on the committee, focusing solely on those villages that were part of the parity lottery. We find that across the eight family outcomes, there is no evidence of positive impact. This result is not driven by low statistical power; although the confidence intervals are larger (because only a subset of villages entered the gender parity lottery), the magnitudes are small and only two of the eight estimates are positive. See Van der Windt et al. (2018) for details, and short term results that are similar as those reported here. 14 There are no program data about the composition of the village cluster committees. 12 Robustness The null results may reflect a weak treatment effect, or weaknesses in the research design. Humphreys et al. (2019), however, show that in the short run there is no evidence for spillovers, differential social desirability biases, or low statistical power. One may also be worried that the outcome measures are insufficiently refined to detect subtle differences between treatment and control communities. Some measures are certainly better than others. However, one of this study’s strengths is the diversity of individual measures for each outcome family, often employing multiple data collection approaches – for example, employing direct observations by our enumerators, survey self-reports, and behavioral measures – and the fact that they produce very similar results. Discussion and Conclusion What is the longer term impact of CDD programs? To answer this question, we collect data in 735 villages in Eastern Congo eight years after the onset of a large, randomized CDD program. The data suggest that treatment villages are served by primary schools and hospitals that have higher quality infrastructure, and that their hospitals are better stocked. In contrast, we find no evidence that the CDD program had an impact on other dimensions of service provision, health, education, economic welfare, women’s empowerment, governance, or intra and inter-village social cohesion. We know of only one other study that explores the longer term impact of a randomized CDD program. Casey et al. (2019) collect data in Sierra Leone about 12 years after the inception of the GoBifo CDD program (Figure 1). They find positive effects on project implementation, the quality of local public services infrastructure, and economic welfare, outcomes they collectively term “hardware” effects. In contrast, they find no sustained impact on measures related to institutional or social change, what they term “software” effects.15 The authors conclude that their data provide evidence for the durability of CDD’s material benefits, including the physical endurance of infrastructure built. We complement Casey et al. (2019) and add to our understanding of the longer term impact of CDD programs in three ways. The first contribution relates to measurement and scope. Due to 15 Note that when the nine individual dimensions (collective action, inclusion, local authority, trust, groups and networks, access to information, participation in governance, crime and conflict and political and social attitudes) that constitute the software family are combined, the data suggest an impact (0.07SD, p<0.01). 13 research budget restrictions, Casey et al. (2019) only make use of surveys conducted with community leaders and a limited number of measures based on direct observation of public infrastructure. In this study, we do the same but also collect data from randomly selected households, including children’s test scores, and from the users and directors of the public infrastructure. One benefit of these additional data sources is that it provides a richer set of measures per outcome family. Another benefit is that these data allow us to explore a larger set of outcome families, such as additional dimensions of service provision, education, health, and women’s empowerment; outcomes that are central to many CDD interventions. Second, the size of the experiment we study here is much larger. Casey et al. (2019) employ data from 236 villages (113 treatment, 113 control). In contrast, our measurement strategy builds on data from 735 villages (367 treatment, 368 control), significantly increasing statistical precision and decreasing the possibility for Type-II errors. Finally, we contribute through replication. Similar to Sierra Leone, the Democratic Republic of Congo scores badly when it comes to development outcomes. However, Congo is different in important ways, including the proliferation of armed groups and continuing conflict and violence, which may affect the effectiveness of a CDD program. As such, this study contributes to generalizable knowledge by understanding the impact of a similar program in a different context, specifically a fragile and conflict setting. In sum, we find very similar results as those reported in Casey et al. (2019). We find no evidence for software results, despite the change in context, the additional precision, and the wider set of software-related outcomes. In addition, we find positive effects on some hardware-related outcomes. Specifically, the positive effect on the stock and quality of local public infrastructure (0.23SD, p<0.01) as reported in Casey et al. (2019), is very similar to the positive effect on the infrastructure quality of health facilities (0.16SD, p<0.01) and primary schools (0.19SD, p<0.01) that we report in this study.16 The similarity in findings related to the durability of material benefits of CDD programs is encouraging. Furthermore, the longer term positive effects on hardware but not on software-related outcomes largely mirror the results found in short-run studies (Casey, 2018). Improving local infrastructure in some of the most challenging environments, and at times in the context of 16 Note that results differ when it comes to economic welfare. Casey et al (2019) find a positive effect (0.24SD, p<0.01), while we find a negative effect (-0.08SD, p<0.05). 14 crippling state incapacity is a worthy achievement of CDD programs in and of itself. Despite this, we remain hesitant to claim that CDD is an effective strategy to obtain these results. Over its four year period, the Tuungane program spent $46 million. A large share of this funding was used for facilitation and indirect costs, with only around $16 million, 35% of total program costs, going directly towards infrastructure. On the one hand, CDD’s software-related activities may be essential to safeguard the initial financial investment over time. On the other hand, these same results may have obtained without these additional activities. The (short and longer term) studies that exist to date, however, only compare CDD programs to a control condition, and thus cannot directly test these claims. Future research could focus on disentangling the relative contribution of hardware-related activities from the contribution of software-related activities. Future studies could also attempt to directly compare CDD to other alternatives, such as more traditional, centrally-led programs or unconditional cash transfer programs. Finally, as is evident from the scarcity of longer term studies of CDD programs, external validity is limited, suggesting the need to accumulate more evidence from different contexts. 15 References • Anderson, M. L. (2008). Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects. Journal of the American Statistical Association, 103(484), 1481–1495. • Avdeenko, A. & Gilligan, M. J. 2015. International Interventions to Build Social Capital: Evidence from a Field Experiment in Sudan. American Political Science Review, 109(3): 427-449. • Beath, A., Christia, F., & Enikolopov, R. (2013). Empowering Women through Development Aid: Evidence from a Field Experiment in Afghanistan. American Political Science Review, 107(3), 540–557. • Beath, A., Christia, F., & Enikolopov, R. (2015). The National Solidarity Programme: Assessing the Effects of Community-Driven Development in Afghanistan. International Peacekeeping, 22(4), 302–320. • Bouguen, A., Huang, Y., Kremer, M., & Miguel, E. (2018). Using RCTs to Estimate Long- Run Impacts in Development Economics. NBER Working Paper, 25356, 1–65. • Casey, K., Glennerster, R. & Miguel, E. 2012. Reshaping Institutions: Evidence on Aid Impacts using a Preanalysis Plan. Quarterly Journal of Economics, 127(4), 1755-1812. • Casey, K. 2018. Radical Decentralization: Does Community-Driven Development Work? Annual Review of Economics, 10, 139-163. • Casey, K., Glennerster, R., Miguel, E. & Voors, M. 2019. Skill versus Voice in Local Development. Working paper, Stanford GSB. • Fearon, J. D., Humphreys, M. & Weinstein, J. M. 2009. Can Development Aid Contribute to Social Cohesion after Civil War? Evidence from a Field Experiment in Post-Conflict Liberia. American Economic Review, 99(2), 287-291. • Fearon, J. D., Humphreys, M., & Weinstein, J. M. 2015. How Does Development Assistance Affect Collective Action Capacity? Results from a Field Experiment in Post- Conflict Liberia. American Political Science Review, 109(3), 450–469. • Humphreys, M., Sanchez de la Sierra, R., & Van der Windt, P. (2012). Social and Economic Impacts of Tuungane: Final Report on the Effects of a Community Driven Reconstruction Program in Eastern Democratic Republic of Congo. 16 • Humphreys, M., Sanchez de la Sierra, R. & Van der Windt, P. 2019. Exporting Democratic Practices: Evidence from a Village Governance Intervention in Eastern Congo". Journal of Development Economics, 140, 279–301. • Kling, J. R., Liebman, J. B. & Katz, L. F. 2007. Experimental Analysis of Neighborhood Effects. Econometrica, 75(1), 83–119. • Mansuri, G. & Rao, V. 2013. Localizing Development: Does Participation Work? World Bank Policy Report. Washington DC. • Rubin, D. B. (1974). Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies. Journal of Educational Psychology, 66(5), 688–701. • Van der Windt, P., Humphreys, M. & Sanchez de la Sierra, R. 2018. Gender Quotas in Development Programming: Null Results from a Field Experiment in Congo". Journal of Development Economics, 133, 326-345. • Van der Windt, P. 2018. Can Development Aid Empower Women? Evidence from a Field Experiment in the Congo". Journal of Politics, 80(3), 1039-1044. 17 Appendix Assessing the Long-run Impact of CDD Programs: Evidence from a Field Experiment in the Congo Eric Mvukiyehe (World Bank) Peter van der Windt (New York University – Abu Dhabi) January 28, 2019 Content: • Appendix A: Map Research Area • Appendix B: Summary of Village and Village Cluster Projects • Appendix C: Timing of Intervention and Short Run Data Collection • Appendix D: Data Sources and Attrition • Appendix E: Variable Definitions and Summary Information • Appendix F: Balance across Treatment Conditions • Appendix G: Results by Individual Measures • Appendix H: Deviations from the Pre-Analysis Plan 1 Appendix A: Map Research Area Figure A1. Map Research Area (a) Congo (b) Tuungane (c) Survey Villages Note: The maps display the location of the study. Panel (a) highlights the provinces (from north to south) of Maniema, South Kivu, Tanganyika, and Haut Katanga. Panel (b) shows Tuungane treatment (black circles) and control (gray squares) villages. Panel (c) plots the survey villages. Data were collected in South Kivu, Tanganyika, and Haut Katanga. 2 Appendix B: Summary of Village and Village Cluster Projects Table A1 and Table A2 give an overview of all the village and village cluster level projects implemented as part of the Tuungane program, respectively. Information is based on 2012 tracking data from the implementing partner. Table A1. Village Level Projects Project type # % Bridge 86 4% Classroom 872 37% Common room 67 3% Health facility 176 8% Health facility equipment 81 3% Latrines 13 1% Market 41 2% Mill 139 6% Mosquito nets 20 1% Other construction 8 0% Other purchase 30 1% Purchase agricultural tools 54 2% Purchase animals 33 1% Purchase doors, windows, benches, etc 76 3% Purchase seeds 13 1% Route 77 3% School material 215 9% Water source 334 14% Total 2,335 100% Note: “Other construction” includes a bus stop, electricity and a morgue. “Other purchases” include projects like an oil press, community radio, satellite dish, field for pygmies, sewing machine and brick press. 3 Table A2. Village Cluster Level Projects Project type # % Bridge 6 2% Electricity 2 1% Health facility 51 16% Health facility equipment 20 6% Market 7 2% Route 9 3% School 131 42% School equipment 67 21% Watsan (latrines, wells, etc.) 18 6% Other 4 1% Total 315 100% Note: “Other” includes soil study and topographic study. 4 Appendix C: Timing of Intervention and Short Run Data Collection Figure A2. Timing of Intervention and Short Run Data Collection Note: Thin black lines indicate length of the Tuungane CDD program per chiefdom. Thick line indicates the village level phase. Shorter, red lines indicate the period of data collection in that chiefdom. Source: Humphreys et al. (2019). 5 Appendix D: Data Sources and Attrition In this section, we discuss attrition in more detail and show that it is unrelated to both the Tuungane and the gender parity treatment status. Table A3 gives, for each data source, an overview of the number of targeted observations, the number of observations actually used for analysis, and possible imbalances between treatment conditions. We discuss each data source in turn. Villages Visited Tuungane was implemented between 2007 and 2012 in 280 clusters comprising 1,250 villages across four provinces of the Democratic Republic of Congo: South Kivu, Maniema, Tanganyika and Haut Katanga. In 2010 and 2011, Humphreys et al. (2019) targeted to collect data from two randomly selected villages in each of the 560 clusters (280 treatment, 280 control). In total, 816 villages out of the targeted 1,120 villages were visited. There are no differences in attrition by treatment condition (Humphreys et al. 2019). In 2015, we targeted the same villages as visited Humphreys et al. (2019), excluding the Maniema province for logistical reasons. Specifically, we targeted 781 villages (286 in Haut Katanga, 208 in Tanganyika, 287 in South Kivu). In total, 735 of the 781 villages (94%) were visited. Attrition took place because of inaccessibility of villages. Table A3 shows that almost the same number of villages are missing in treatment and control communities. Related to the gender parity treatment, a total of 190 villages out of the 781 participated in the gender parity lottery. We again find no differences in attrition by treatment status. Village Chief Survey Among the 735 villages that we visited, we conducted a survey with the village chief in 714 villages (97%). Among the 180 villages that participated in the gender parity treatment and were visited, a total of 176 village chief surveys were conducted. We find no differences in attrition by treatment status. Household Survey Per village, we targeted five randomly selected households. Within households we randomly selected the (adult) respondent in such a way as to ensure that each gender was represented equally within the sample. Given that 735 villages were visited, the study should make use of 3,675 6 household surveys. In total, we collected data from 3,379 households (92%). These numbers are 814 out of 900 (90%) for those villages that participated in the gender parity lottery. Again Table A3 suggests no differences by treatment status. Table A3. Attrition by Treatment Status Data Source Target Collected Missing Missing Beta (se) control treatment Tuungane treatment Villages visited 781 735 22 24 -0.005 (0.017) Village chief survey 735 714 11 10 0.003 (0.012) Household survey 3,675 3,379 149 147 0.0004 (0.009) Children’s exam 3,379 1,496 966 917 0.024 (0.017) Primary school 735 610 67 58 0.0231 (0.027) Health facility 735 504 114 117 -0.010 (0.034) Gender parity treatment Villages visited 190 180 6 4 0.0177 (0.032) Village chief survey 180 176 2 2 -0.001 (0.022) Household survey 900 814 42 44 -0.008 (0.019) Children’s exam 814 338 236 240 -0.041 (0.034) Primary school 180 147 16 17 -0.019 (0.058) Health facility 180 112 33 35 -0.039 (0.072) Note: Table presents number of targeted observations, number of observations used for analyses, and difference between both across treatment condition. Standard errors clustered at the village cluster level. ∗ p ≤ 0.10, ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01. Children’s Exam In those households with children of school-going age (between 6 and 11 years old), we randomly selected one child for a brief exam. In 1,496 of the 3,379 household surveys was the respondent part of a household with a child of school going age, and was the respondent willing to give permission for the exam. This was 338 out of 814 for those household surveys in villages that were part of the gender parity lottery. Again, we find no differences in exams conducted across treatment conditions. 7 Primary School and Health Facility The final data source is the infrastructure survey. We are interested in service provision as experienced by the inhabitants of villages. We thus visited the primary school and health facility within a five-kilometer radius for each village, thus not necessarily the Tuungane-built facilities in treatment areas. Specifically, upon arrival in the village, the survey teams were tasked to visit the village chief to explain the data collection exercise and obtain approval. During this meeting, surveyors also asked the village chiefs about the name and location of the primary school and the health infrastructure that are used by the community. After obtaining this information, both facilities were visited for the infrastructure surveys. We instructed our surveyors not to visit the school or health facility if the facility was located more than five kilometers (about one hour walking distance) away. We thus do not measure the difference in quality provided between a Tuungane facility and a control facility. We compare the quality of nearby service provision for villagers in Tuungane and control areas. In other words, we are thus not interested in whether a Tuungane facility is better than a not-Tuungane facility. This study is interested in whether service provision has improved for people that live in Tuungane areas, compared that those that do not. In total, 610 primary schools and 504 heath facilities were visited. Given that we visited 735 villages, this amounts to 83% and 68%, respectively. These numbers are 147 and 112 out of 180 for those villages that participated in the gender parity lottery. Table A3 shows that there are no differences by Tuungane and gender parity treatment status. This is an important result in and of itself insofar that the nearby presence of hospitals and schools is an indicator for accessibility of service provision. In sum, the data presented in this section suggest that the probability of attrition is similar across treatment conditions. Although unlikely, we acknowledge that those villages lost in treatment and control conditions may be different. 8 Appendix E: Variable Definitions and Summary Information Table A4 gives summary information of all 171 individual outcome variables used in this study. Table A4. Summary Information # Family Subfamily Outcome Description Q Mean Sd. Min. Max. Obs. 1 Hospital Building Floor quality Binary. From the following list: mud, straw, wood/ bamboo, metal plates, concrete/ ES42 0.85 0.35 0 1 502 cement, tiles, plastic, stone, backed bricks, cardboard, other. Floors made of mud, wood, plastic, and cardboard are low quality, the rest are high quality. 2 Hospital Building Wall quality Binary. From the following list: mud, straw, wood/ bamboo, metal plates, concrete/ ES41 0.87 0.33 0 1 462 cement, tiles, plastic, stone, backed bricks, cardboard, other. Walls made of cement and baked bricks are high quality, the rest are low quality. 3 Hospital Building Infrastructure Continuous 0-1. Simple average of the presence of: consultation room, treatment room, ES19- 0.67 0.23 0 1 502 laboratory, observation room, pharmacy, maternity or delivery room, waiting room, ES32 nurse office, trash can in all rooms, incinerator, garbage hole, placenta hole, working latrines, showers. 4 Hospital Building Clean floor Binary. The floor is clean. ES35 0.74 0.44 0 1 498 5 Hospital Building Clean wall Binary. The wall is clean. ES36 0.65 0.48 0 1 499 6 Hospital Building Rate building Binary. Rate building quality. Options: 0) bad, 1) average, 2) good. Respondents that Q101a 0.54 0.5 0 1 3205 reply good equal one, zero otherwise. 7 Hospital Building Toilets Binary. The building has toilets that work and are clean? Q106 0.79 0.41 0 1 3166 8 Hospital Capacity # Providers present Continuous. Number of health care providers present. ES37 3.07 2.9 0 26 501 9 Hospital Capacity # Beds Continuous. Number of beds present. ES18 9.1 10.13 0 95 496 10 Hospital Capacity Wait personnel Continuous. Minutes wait before being seen by qualified person. ES45 6 9.91 0 120 414 11 Hospital Capacity Wait treatment Continuous. Minutes wait before being treated. ES47 5.95 9.37 0 120 410 12 Hospital Capacity # Nurses Number of nurses employed by the facility. ES66 3.21 2.1 0 17 498 13 Hospital Capacity # Doctors Number of doctors employed by the facility. ES67 0.31 0.81 0 8 471 14 Hospital Capacity Treatments Continuous (0-6). From the following list, how many can be treated at the facility: ES64 4.96 1.29 1 6 492 diarrhea, wound, infection of respiratory tract, delivery, dermatosis, and high blood pressure. 15 Hospital Capacity Rate capacity Binary. Rate building capacity. Options: 0) bad, 1) average, 2) good. Respondents that Q101b 0.46 0.5 0 1 3181 reply good equal one, zero otherwise. 16 Hospital Capacity Wait time Continuous. The expected time (in minutes) before seeing the doctor or nurse? Q114 9.9 23.37 0 720 2909 17 Hospital Material # Antibiotics Continuous. Number of antibiotic tablets present ES38 1436.05 1957.18 0 10000 454 18 Hospital Material # Malaria tablets Continuous. Number of malaria tablets present ES39 962.92 1449.74 0 10000 459 19 Hospital Material # Anti-inflammatory Continuous. Number of anti-inflammatory tablets present ES40 1013.29 1616.36 0 10000 449 tablets 20 Hospital Material Rate material Binary. Rate availability and quality of equipment. Options: 0) bad, 1) average, 2) Q101c 0.42 0.49 0 1 3107 good. Respondents that reply good equal one, zero otherwise. 21 Hospital Staff Doctor nurse ratio Continuous. Number of doctors over number of nurses. ES66, 0.1 0.26 0 2.33 466 ES67 22 Hospital Staff Director education Continuous. Years of education by director. ES56 13.4 2.89 0 19 510 1 23 Hospital Staff Director medical Binary. Director studied medicines. ES57 0.9 0.29 0 1 502 studies 24 Hospital Staff Rate care Binary. Rate quality of care. Options: 0) bad, 1) average, 2) good. Respondents that Q101d 0.58 0.49 0 1 3148 reply good equal one, zero otherwise. 25 Hospital Staff Rate health provider Binary. Rate quality of health care providers. Options: 0) bad, 1) average, 2) good. Q101e 0.63 0.48 0 1 3110 Respondents that reply good equal one, zero otherwise. 26 Hospital Staff Presence health Binary. The doctor or nurse is always present on time? Q98 0.95 0.21 0 1 3157 provider 27 Hospital Admin Director present Binary. The director is present. ES50 0.79 0.41 0 1 505 28 Hospital Admin Patient register Binary. Presence (and verification) of patient register ES81 0.94 0.23 0 1 496 29 Hospital Admin Staff register Binary. Presence (and verification) of staff register ES82 0.8 0.4 0 1 493 30 Hospital Admin Stock register Binary. Presence (and verification) of stock register ES83 0.79 0.41 0 1 477 31 Hospital Admin Cash book Binary. Presence (and verification) of cash book ES84 0.74 0.44 0 1 481 32 Hospital Admin Rate administration Binary. Rate quality of administration. Options: 0) bad, 1) average, 2) good. Q101f 0.59 0.49 0 1 2908 Respondents that reply good equal one, zero otherwise. 33 Hospital Community # Comm. meetings Continuous. How many meetings were held with the community during last school ES91 4.2 5.25 0 48 415 year? 34 Hospital Community Contr. in kind Binary. During last year, did [village name] contribute in kind? ES70 0.11 0.32 0 1 453 35 Hospital Community Contr. in $ Binary. During last year, did [village name] contribute in cash? ES71 0.04 0.19 0 1 450 36 Hospital Community Know CODESA Continuous. How many members of CODESA do you know [list]? ES86 9 6.64 0 30 462 37 Hospital Community CODESA meetings Continuous. How many meetings were held with CODESA during last school year? ES87 10.55 7.59 0 48 459 38 Hospital Community Director present Continuous. How many of these meetings did you personally attend? ES88 8.16 6.91 0 48 472 39 Hospital Community Rate interaction Binary. Rate interaction of facility with the community. Options: 0) bad, 1) average, 2) Q101g 0.63 0.48 0 1 3092 good. Respondents that reply good equal one, zero otherwise. 40 Hospital Community Contr. in kind Binary. During last year, did you contribute in kind? Q112 0.04 0.2 0 1 2950 41 Hospital Community Contr. in $ Binary. During last year, did you contribute in cash? Q113 0.03 0.16 0 1 2871 42 Hospital Costs Open Binary. Health center is open. ES14 0.98 0.15 0 1 505 43 Hospital Costs $ Paid Continuous. How much does your treatment cost? In US dollars. ES49 11.39 31 0 277.78 376 44 Hospital Costs # Patients now Continuous. Number of patients at the moment. ES69 4.4 7.69 0 73 480 45 Hospital Costs # Patients last month Continuous. Number of patients during last month. ES68 175.17 183.13 0 943 472 46 Hospital Costs Cost index Continuous. The $ price for a visit, the $ price for a consultation, the $ price for a health ES58- 0.05 0.64 -0.81 3.63 486 card, and the $ price for an overnight stay. Each variable is standardized. We then ES61 average across the four. 47 Hospital Costs Rate cost Binary. Rate costs. Options: 0) bad, 1) average, 2) good. Respondents that reply good Q101h 0.28 0.45 0 1 3114 equal one, zero otherwise. 48 Hospital Costs Cost index Continuous. The $ price for a consultation, the $ price for a health card, and the $ price Q108- 0.02 0.9 -0.61 26.79 2852 for an overnight stay. Each variable is standardized. We then average across the three. Q111 49 Hospital Costs # Visits Continuous. In the last year, how many times did somebody in your household visit the Q92 3.86 8.64 0 300 3184 facility. 50 School Building Floor quality Same as above EE35 0.55 0.5 0 1 550 51 School Building Wall quality Same as above EE37 0.78 0.42 0 1 481 52 School Building Roof quality Binary. From the following list: mud, straw, wood/ bamboo, metal plates, concrete/ EE34 0.82 0.38 0 1 551 cement, tiles, plastic, stone, backed bricks, cardboard, other. Roofs made of metal plates, concrete/cement, tiles and backed bricks are high quality. 53 School Building Windows Binary. Presence of windows with glass. EE32 0.31 0.46 0 1 553 54 School Building Toilets Binary. Presence of functional toilet EE36 0.68 0.47 0 1 554 2 55 School Building Rate building Same as above Q130a 0.41 0.49 0 1 3154 56 School Building Toilets Same as above Q135 0.58 0.49 0 1 3182 57 School Capacity # Classrooms Continuous. Number of classrooms EE31 6.92 3.35 0 26 554 58 School Capacity Classroom size Continuous. Classroom size in square meters EE26 27.79 20.56 0 75 153 59 School Capacity Highest class Binary. Highest degree: elementary, middle, terminal. Response is terminal equal one, EE54 0.86 0.35 0 1 526 zero otherwise. 60 School Capacity # Teachers Continuous. Number of teacher employed EE53 7.25 2.86 0 15 495 61 School Capacity # Students reg. Continuous. Number of students registered EE55 246.85 161.55 0 930 522 62 School Capacity Teacher student ratio Continuous. Number of teachers employed divided by number of students registered EE53, 0.04 0.03 0.01 0.3 475 EE55 63 School Capacity Rate capacity Same as above Q101b 0.44 0.5 0 1 3103 64 School Capacity Classrooms large Binary. Do you consider the classrooms large enough? Q133 0.74 0.44 0 1 2918 65 School Material Blackboard Binary. Blackboard present EE24 0.96 0.19 0 1 154 66 School Material # Benches Continuous. Number of seats EE25 9.84 9.38 0 45 157 67 School Material Prop. books Continuous (0-1). Proportion of students with study books. EE20, 0.19 0.36 0 1 148 EE22 68 School Material Prop. notebooks Continuous (0-1). Proportion of students with notebooks. EE20, 0.68 0.43 0 1 143 EE23 69 School Material Teacher book Binary. Teacher has study book. EE27 0.73 0.45 0 1 143 70 School Material Teacher prep. Binary. Teacher has workbook. EE28 0.84 0.37 0 1 145 71 School Material Teacher list Binary. Teacher has attendance list. EE29 0.82 0.38 0 1 145 72 School Material Rate material Same as above Q130c 0.36 0.48 0 1 2765 73 School Staff Teacher present Binary. Teacher is present. EE19 0.96 0.2 0 1 150 74 School Staff Studied pedagogy Binary. Teacher is studied pedagogy EE30 0.99 0.12 0 1 147 75 School Staff Director education Continuous. Years of education by director. EE45 10.84 3.56 0 17 555 76 School Staff Director pedagogy Binary. Director studied pedagogy EE46 0.95 0.21 0 1 516 77 School Staff Rate teachers Binary. Rate quality of teachers. Options: 0) bad, 1) average, 2) good. Respondents that Q130d 0.59 0.49 0 1 2797 reply good equal one, zero otherwise. 78 School Staff Teacher absence Binary. Are the teachers often absent? Q128 0.26 0.44 0 1 2537 79 School Staff Teacher punctual Binary. Are the teachers punctual? Q129 0.92 0.27 0 1 2551 80 School Staff Teacher qualified Binary. Are the teachers qualified? Q131 0.91 0.29 0 1 2490 81 School Staff Teacher rigorous Binary. Are the teachers rigorous? Q132 0.88 0.32 0 1 2402 82 School Admin Director present Same as above EE39 0.66 0.47 0 1 546 83 School Admin Staff register Binary. Presence (and verification) of staff register EE61 0.9 0.3 0 1 465 84 School Admin National program Binary. Presence (and verification) of national curriculum EE60 0.66 0.48 0 1 384 85 School Admin Rate director Same as above Q130e 0.54 0.5 0 1 2671 86 School Community # Comm. meetings Same as above EE72 2.91 2 0 12 502 87 School Community Contr. in kind Same as above EE58 0.27 0.45 0 1 484 88 School Community Contrib. in $ Same as above EE59 0.19 0.39 0 1 473 89 School Community Know COPA Continuous. How many members of COPA do you know [list]? EE67 5.59 2.36 0 18 521 90 School Community COPA meetings Continuous. How many meetings were held with COPA during last school year? EE68 4.64 3.45 0 27 505 91 School Community Director present Same as above EE69 4.1 3.52 0 28 515 92 School Community Rate interaction Same as above Q130f 0.6 0.49 0 1 2899 93 School Community Contr. in kind Same as above Q140 0.08 0.28 0 1 2867 94 School Community Contr. in $ Same as above Q141 0.06 0.24 0 1 2818 95 School Costs Open Same as above EE13 0.3 0.46 0 1 519 96 School Costs Boys Continuous. Number of boys present in class EE20 18.78 10.01 0 49 147 97 School Costs Girls Continuous. Number of girls present in class EE56 16.05 9.79 1 50 147 3 98 School Costs Students pres. Continuous. On average, how many students are present per day. EE47 218.22 153.12 0 910 520 99 School Costs School fee ($) Continuous. Monthly school fee per child. In US dollars. EE48 2.07 1.54 0 10.56 527 100 School Costs Fee ($) Continuous. Operating fee per child per trimester. In US dollars. EE48 0.87 1.42 0 11.11 418 101 School Costs Rate costs Same as above Q130g 0.33 0.47 0 1 2875 102 School Costs Cost ($) Continuous. Since the start of this school year, how much has the household spent on Q121 81.87 108.39 0 1111.11 2247 the education of the children of this household (6-12 years). This includes tuition, manuals, uniforms, transportation and other fees. In US dollars. 103 School Costs School fee ($) Continuous. Monthly school fee per child. In US dollars. Q137 2.27 1.5 0 11.11 2616 104 School Costs Fee ($) Continuous. Operating fee per child per trimester. In US dollars. Q138 1.87 1.98 0 11.11 1677 105 Health Medical care Binary. In last 12 months, somebody in household fell ill enough to require medical Q88 0.76 0.43 0 1 3372 care? 106 Health U5 mortality Binary. In last 12 months, did a child younger than 5 years old in the household die due Q89 0.12 0.33 0 1 3343 to illness? 107 Health Death head Binary. In the last 12 months, did the head of the household pass away? Q60a 0.02 0.15 0 1 3377 108 Health Death other Binary. In the last 12 months, did somebody else in the household pass away? Q60b 0.16 0.37 0 1 3378 109 Health Sick head Binary. In the last 12 months, was the head of the household severely ill? Q60c 0.29 0.46 0 1 3374 110 Health Sick other Binary. In the last 12 months, was somebody else in the household severely ill? Q60d 0.59 0.49 0 1 3372 111 Education Attendance Continuous. How many daughters have gone to school uninterrupted (since age 6) Q115 1.04 1.19 0 8 2894 daughters 112 Education Attendance sons Continuous. How many sons have gone to school uninterrupted (since age 6) Q115 1.34 1.39 0 9 2919 113 Education Never attended Continuous. How many daughters have never attended school Q118 0.39 0.85 0 9 2787 (daughters) 114 Education Never attended Continuous. How many sons have never attended school Q118 0.36 0.91 0 20 2777 (sons) 115 Education Grade (French) Continuous (0-6). Correct responses by child to six question related to mathematics, EX11 1.93 1.7 0 6 1259 French and science (2 questions each). In French. 116 Education Grade (local) Continuous (0-6). Correct responses by child to six question related to mathematics, EX11 0.87 1.24 0 5 406 French and science (2 questions each). Questions related to mathematics and science are repeated in the local language if incorrect in French. 117 Welfare Roof quality Same as above Q39 0.37 0.48 0 1 3371 118 Welfare Wall quality Same as above Q40 0.08 0.27 0 1 3336 119 Welfare Assets Continuous. Simple average of the number of items owned across the following assets: Q37 1.42 0.83 0.09 8.92 3379 basin, beds, jerry cans, bikes, boats, boxes, buckets, cabinets, chairs, cows, goats, hoes, lamps, mattress, motor, pans, phone, photo camera, pigs, poultry, radio, rooms, straw mattress. 120 Welfare Consumption ($) Continuous. Aggregation of household spending during the preceding 30 days in the Q54 81.29 113.39 0 1461.11 3377 following areas: food, medicine, leisure, clothes, alcohol, cigarettes, seeds, household equipment, small works, large works. In US dollars. 121 Welfare Earnings ($) Continuous. Total household income in the last seven days. In US dollars. Q74 12.57 34.72 0 666.67 2919 122 Women Women rights Binary. Opinion about the following statement: “In this village, women should have the Q237 0.55 0.5 0 1 3359 same rights and obligations as men.” Options: disagree, no opinion, agree. Respondents that reply agree. 123 Women Hit women Continuous (0-8). “Sometimes a husband is upset or angry because of certain things his Q241 2.58 2.39 0 8 3364 wife does. In your opinion, is it justified for a husband to beat or beat his wife in the following situations: 1) if she goes out without telling him, 2) if she refuses to have sex with him, 3) if she neglects children, 4) if she burns the food, 5) if she quarrels with him, 6) if she is unfaithful, 7) if she demands the use of contraceptive methods, 8) if she drinks alcohol.” Simple sum across the eight variables. 124 Women Women association Binary. Is a women association active in the village? Q183d 0.21 0.41 0 1 3204 4 125 Women Daughter school Continuous (0-1). Share of girls, among all household children, that go to school. Q115 0.43 0.33 0 1 2162 attendance 126 Women Daughter never to Continuous (0-1). Share of girls, among all household children, that have never been to Q118 0.53 0.37 0 1 806 school school. 127 Women Women committee Continuous (0-1). Proportion of members of the local development committee that are EC105d, 0.35 0.18 0 0.8 210 members women. EC105e 128 Governance Participation Present meeting Binary. In the last six months, did you participate in a village meeting? Q199a 0.45 0.5 0 1 3349 129 Governance Participation Voluntary Continuous (0-6). In the last six months, did you contribute (time, money or labor) to: Q194 1.33 0.96 0 6 1393 contribution construction and maintenance of primary schools or health infrastructure, construction or maintenance of roads, construction or maintenance of wells, organization of security patrols, maintenance of a church or mosque, construction of a market. Simple summation. Conditional on one of those projects taking place. 130 Governance Participation Voted 2011 Binary. Did you vote in the 2011 elections? Q218 0.93 0.26 0 1 3368 131 Governance Participation Election meeting Binary. Did you participate in a rally/ election campaign during the last election? Q219 0.36 0.48 0 1 3350 132 Governance Participation Right to participate Binary. Opinion about the following statement: “Everyone should have the right to Q232 0.64 0.48 0 1 3347 participate in the political and economic decisions, even if they do not master all the aspects of the problem in question” Options: disagree, no opinion, agree. Respondents that reply agree. 133 Governance Participation Interaction Continuous (0-11). In the last six months, which of 11 activities has the chief EC205 2.35 2.3 0 9 712 undertaken: 1) contact the police or judiciary for problems related to the village, 2) contact the military for problems related to the village, 3) contact provincial government for problems related to the village, 4) contact national government for problems related to the village, 5) contact local, decentralized government entities (ETDs) for problems related to the village, 6) contact the chief of the grouping or chiefdom for problems related to the village , 7) contact MONUSCO to ask to initiate a village project, 8) contact an international NGO to ask to initiate a village project, 9) contact the national assembly member that represents the village, 10) contact armed groups, 11) contact CODESA/ COPA to discuss a development project related to the village. Simple summation. 134 Governance Accountability Interaction Continuous (0-7). In the last six months, how many accountability-related activities Q199c-i 0.76 1.19 0 7 3371 have you undertaken from the following activities: 1) meet the village chief to raise an issue, 2) meet a member of a village management committee to raise an issue, 3) contact the police or the judiciary about some problems you had, 4) meet or contact other state officials about some problems you had, 5) meet representatives of MONUSCO or NGOs to raise an issue, 6) participate in a demonstration or a peaceful protest march, 7) meet with influential individuals, but without authority recognized by the state (e.g. armed groups). Simple summation. 135 Governance Accountability Local committee Continuous (1-8). Does the local committees (COPA and CODESA) undertake the Q200a-h 3.86 3.23 0 8 2620 following activities: 1) inform the public about its actions, 2) inform the population of resource management, 3) inform the community about the performance of providers and the quality of services, 4) allow people like you to participate, 5) be consulted before making decisions, 6) ensure that local resources are used for public purposes and not for private interests, 7) conduct advocacy with the state authorities on community needs, 8) inform state authorities about the performance of providers and the quality of services. Simple summation. 136 Governance Accountability Chief informs Binary. When it comes to making important decisions, the leader takes care to inform Q211 0.71 0.45 0 1 3253 the population about why the decisions were made? 137 Governance Accountability Other bodies Binary. If a village member is not satisfied with the leaders' decisions, are there any Q212 0.5 0.5 0 1 2898 other bodies that can influence the decisions? 5 138 Governance Accountability Influence leaders Binary. Are you of the opinion that you can influence your leaders? Q228 0.17 0.38 0 1 3018 139 Governance Accountability Verify leaders Binary. Opinion about the following statement: “As citizens we have the duty to check Q233 0.26 0.44 0 1 3343 regularly and to question the actions of our provincial political leaders and nationals.” Options: disagree, no opinion, agree. Respondents that reply agree equal one, zero otherwise. 140 Governance Accountability Local committee Same as above EC206a- 2.63 2.1 0 5 601 d,g 141 Governance Transparency Accept school Binary. Each fifth (randomly selected) respondent is asked whether they are willing to Q269 0.83 0.38 0 1 299 seek information about the revenues received in the last period for the school or the hospital (randomly selected). Respondents are offered $1 as compensation for attempting to retrieve the information and an additional dollar upon success. Outcome equals one when the respondent is willing to collect information from the school. 142 Governance Transparency Accept health Binary. See above. Willingness to collect information from the hospital. Q271 0.78 0.41 0 1 228 143 Governance Transparency Knowledge Continuous (0-6). Respondent knows the name of: 1) the Prime Minister of the Congo, Q242 2.37 1.71 0 6 3184 2) the member of the National Assembly who represents the community, 3) the largest party in the National Assembly, 4) the governor of the province, 5) the head of their territory, and 6) the leader of their grouping. Simple summation. 144 Governance Transparency Verify chief Binary. Opinion about the following statement: “As inhabitants of the village, we have Q235 0.74 0.44 0 1 3353 the duty to check regularly and question the actions of our village chief” Options: disagree, no opinion, agree. Respondents that reply agree equal one, zero otherwise. 145 Governance Efficiency Approached state Binary. In the last six months, did members of this village approach the state to ask Q196 0.06 0.23 0 1 3024 them to initiate projects for the village? 146 Governance Efficiency Successful state Binary. In the last six months, did members of this village successfully approach the Q197 0.02 0.12 0 1 3024 state to ask them to initiate projects for the village? 147 Governance Efficiency Approached NGO Binary. In the last six months, did members of this village approach and NGO to ask Q176 0.04 0.19 0 1 3087 them to initiate projects for the village? 148 Governance Efficiency Successful NGO Binary. In the last six months, did members of this village successfully approach and Q180 0.02 0.12 0 1 3084 NGO to ask them to initiate projects for the village? 149 Governance Capture Committee exist Continuous (0-9). Presence of the following committees: 1) water/ sanitation, 2) roads Q183A-I 2.04 1.66 0 8 3343 and erosions, 3) health (CODESA), 4) education/ school (COPA), 5) farming or agriculture, 6) protection or security, 7) conflict resolution, 8) development general, and 9) other. Simple summation. 150 Governance Capture Committee elected Continuous (0-1). Proportion of committees in the village that are democratically Q185 0.83 0.29 0 1 2504 elected. 151 Governance Capture # Associations Continuous (1-11). Presence of the following association: 1) an association affiliated to Q207 1.39 1.79 0 11 3332 the church/ mosque, 2) a peasant association, 3) an association of the elderly, 4) an association of women, 5) a youth organization, 6) an association of former combatants / militia 7) an association for savings and credit, 8) an association to support a certain politician or political party, 9) a human rights association, 10) a cultural association / ethnic, and 11 ) other. Simple summation. 152 Governance Capture Association elected Continuous (0-1). Proportion of associations in the village that are democratically Q198 0.76 0.34 0 1 1652 elected. 153 Governance Capture Collected tax Binary. In the last thirty days, did the village chief collect taxes from you? Q207 0.05 0.22 0 1 3300 154 Governance Capture Committee funds Binary. If the village received $1000 for its development, to whom should the Q198 0.26 0.44 0 1 3201 responsibility to manage this amount to be sure the money is really used for the wellbeing of the village: village chief, development committee, NGO, national government in Kinshasa, provincial government, other. Response is development committee equal one, zero otherwise. 6 155 Intra-village Cleavages Continuous (0-10). It is sometimes difficult for the inhabitants of a village to work Q186 1.06 1.3 0 8 3167 together because of the differences that exist between them. What cleavages exist in this village? The cleavages were not prompted. We have a list from which they could check the following cleavages: 1) between the rich and poor, 2) between men and women, 3) between the young and the elderly, 4) between indigenous and newcomers, 5) between the different religions, 6) between the tribes or ethnic groups, 7) between civilians and ex-combatants/ militia, 8) between pastoralists and farmers, 9) between people of different political parties, and 10) between educated and uneducated. Simple summation. 156 Intra-village Trust village Continuous (0-1000). Contribution in trust game to another randomly selected villager. Q277 384.06 247.39 0 1000 1344 member Specifically, each participant played a standard trust game four times. Each time, they received 1,000 Congolese Francs (around $1, or a day’s wage). The amount send to the receiver would be tripled, and the receiver would subsequently decide how much to return. What was different each time was the receiver type. Participants played with four of the following eight possible receivers: a village member or the village chief (to measure intra-village cohesion), a village member of a neighboring village (to measure inter-village cohesion), and five other potential receivers (not further used in this study). The four receivers and their order was randomly assigned. One of the four games was randomly selected for payout. 157 Intra-village Trust village chief Continuous (0-1000). Contribution in trust game to the village chief. See above for Q277 420.72 263.99 0 1000 1304 game details. 158 Intra-village Voluntary projects Continuous (0-6). In the last six months, which of the following voluntary project take Q192 0.7 1.02 0 6 3352 place in the village: 1) construction and maintenance of primary schools or health infrastructure, 2) construction or maintenance of roads, 3) construction or maintenance of wells, 4) organization of security patrols, 5) maintenance of a church or mosque, 6) construction of a market? Simple summation. 159 Intra-village Community Continuous (0-8). The community is capable of independently determining the rules of EC121b- 1.29 1.36 0 7 465 ownership access and use for a number of community resources: 1) arable land, 2) community EC128b forest, 3) pasture, 4) water (lake, rivers), 5) mineral mine, 6) quarry for stone/ sand, 7) hunting reserve, and 8) other. Simple summation. 160 Intra-village Development Binary. Development committee exist in the village? EC105a 0.39 0.49 0 1 704 committee 161 Intra-village Committee Binary. Development committee undertakes activities that benefits the whole EC105b 0.92 0.27 0 1 272 population community. Conditional on development committee existence. 162 Intra-village Committee Binary. How often does the development committee meet per month? Conditional on EC105c 1.69 1.31 0 8 252 frequency development committee existence. 163 Intra-village # Associations Same as above EC105a- 3.07 2.2 0 13 712 EC120a 164 Inter-village Cleavages other Continuous (0-10). It is sometimes difficult for the inhabitants of a village to work Q187 0.96 1.26 0 8 3124 village together because of the differences that exist between them. What cleavages exist between members of this village and those in neighboring villages? The cleavages were prompted. We have a list from which they could check the following cleavages: 1) between the rich and poor, 2) between men and women, 3) between the young and the elderly, 4) between indigenous and newcomers, 5) between the different religions, 6) between the tribes or ethnic groups, 7) between civilians and ex-combatants/ militia, 8) between pastoralists and farmers, 9) between people of different political parties, and 10) between educated and uneducated. Simple summation. 165 Inter-village Trust other village Continuous (0-1000). Contribution in trust game to a (randomly selected) villager of a Q277 367.32 239.66 0 1000 1358 neighboring village. See above for game details. 7 166 Inter-village Committee other Continuous (0-9). Existing committees that work together with other villages: 1) water/ Q182 1.36 1.32 0 7 2405 sanitation, 2) roads and erosions, 3) health (CODESA), 4) education/ school (COPA), 5) farming or agriculture, 6) protection or security, 7) conflict resolution, 8) development general, and 9) other. Simple summation. 167 Inter-village Projects other Continuous (0-6). Voluntary projects undertaken with other villagers: 1) construction Q195 0.45 0.46 0 1 1385 and maintenance of primary schools or health infrastructure, 2) construction or maintenance of roads, 3) construction or maintenance of wells, 4) organization of security patrols, 5) maintenance of a church or mosque, 6) construction of a market. Simple average. 168 Inter-village Associations other Continuous (0-11). Existing associations that undertake activities with other villages: 1) EC105g- 0.66 0.37 0 1 628 an association affiliated to the church/ mosque, 2) a peasant association, 3) an EC120g association of the elderly, 4) an association of women, 5) a youth organization, 6) an association of former combatants / militia 7) an association for savings and credit, 8) an association to support a certain politician or political party, 9) a human rights association, 10) a cultural association / ethnic, and 11 ) other. Simple average. 169 Inter-village Resources other Continuous (0-8). Community resources held jointly with the other villages in this EC121j- 0.52 0.45 0 1 457 territory: 1) arable land, 2) community forest, 3) pasture, 4) water (lake, rivers), 5) EC128j mineral mine, 6) quarry for stone/ sand, 7) hunting reserve, and 8) other. Simple average. 170 Inter-village Managed conflict Binary. In the last three months, chief has managed conflict between his/her village and EC179 0.24 0.43 0 1 655 a neighboring village. 171 Inter-village Hosted other Binary. In the last three months, chief has hosted the chief of a neighboring village. EC179 0.16 0.36 0 1 655 Note: Variable definitions and summary information. 8 Appendix F: Balance across Treatment Conditions The analyses in this paper rely on randomization, which guarantees that treatment and control areas are similar in expectation. In practice, however, it is possible for them to differ simply by virtue of unlucky draws. To test this, we compare Tuungane treatment and control areas, and – for those areas that partook in the parity lottery – gender parity and control areas. Because we do not have baseline data for the villages, we make use of the data collected in 2010 and 2011 by Humphreys et al. (2019) in 816 randomly selected villages. We limit ourselves to pre-treatment information and variables that do not change due to the treatment. We analyze the following variables. Distance from a set of (nearest) points of importance that are unlikely to change due to the program: mine, post office, and the kingdom headquarters. Distance data (measured in hours of walking) are based on individual responses, mean aggregated to the village level. Note that in a large number of villages, individuals do not know the distance to these locations, and thus we have fewer observations than the 816 visited villages. We also use data on ethnic and religious composition of the village, measured as the probability that two individuals, selected at random from the village, will be of different ethnicities or religious groups. These data come from a survey conducted with village chiefs. In total, 773 of the 816 village chiefs were interviewed. Data were also collected about the characteristics of the previous chief: his year of birth, and whether he was democratically elected. Note that many chiefs responded with “Don’t know”, resulting in considerably fewer than 773 observations for the balance test. We also collected data on the principal economic activities undertaken in the village (as a percentage of the village population). In addition, data were collected on the presence of infrastructure in the village in 2006: wells, schools, clinics, churches and meeting halls. We use data from the chief on the number of IDPs, returned-IDPs, refugees and repatriated refugees that entered the village in 2006. Again, many chiefs responded with “Don’t know”. Finally, at the individual level we analyze gender and age. The data were obtained from the respondent about all the other individuals (both adults and children) in the household. Table A5 lists the mean and standard deviation for each variable for the Tuungane and control areas. We also test the difference between both, based on simple OLS regressions. Table A6 presents the same information comparing villages with and without the gender parity requirement, restricted only to those villages that partook in the gender parity lottery. 9 The results suggest that there are no consistent differences across treatment groups, which is what is to be expected given the random assignment. Table A5. Balance Tuungane and Control Variable Q Tuungane Sd. Control Sd. Diff. (Se.) N Distance mine QE13 20.98 42.46 25.03 59.73 -4.05 (3.86) 723 Distance police post QE13 3.61 6.11 3.69 6.14 -0.08 (0.44) 777 Distance kingdom HQ QE13 8.99 19.6 8.68 11.76 0.31 (1.16) 771 Ethnic heterogeneity CQ13 0.32 0.27 0.33 0.28 -0.02 (0.02) 728 Religious heterogeneity CQ14 0.52 0.19 0.52 0.2 0 (0.01) 724 Birth year former chief CQ45 1936.43 19.62 1938.53 21.13 -2.1 (2.19) 347 Former chief democratic CQ48 0.16 0.37 0.17 0.38 -0.01 (0.03) 653 Share in agriculture CQ15 71.17 20.5 71.61 20.49 -0.43 (1.52) 724 Share in herding CQ15 9.43 10.47 9.29 9.52 0.13 (0.77) 669 Share in commerce CQ15 3.8 8.67 3.98 7.4 -0.19 (0.62) 669 Share in fishing CQ15 6.13 12.76 6.05 12.2 0.08 (0.96) 673 Share in industrial CQ15 0.13 0.97 0.17 1.42 -0.04 (0.09) 665 Share in mining CQ15 3.29 8.71 3.5 8.32 -0.2 (0.66) 667 Share in other CQ15 4.44 8.04 4.21 7.9 0.22 (0.63) 646 Share in other services CQ15 4.41 4.66 4.13 4.55 0.28 (0.36) 665 Wells in 2006 CQ23 0.91 1.83 1.4 3.29 -0.49** (0.2) 705 Schools in 2006 CQ24 3.5 5.04 3.51 4.48 -0.01 (0.36) 713 Clinics in 2006 CQ25 0.32 0.84 0.29 0.51 0.03 (0.05) 718 Churches in 2006 CQ26 2.18 2.5 2.55 2.67 -0.37* (0.19) 716 Halls in 2006 CQ27 0.04 0.3 0.04 0.21 0.01 (0.02) 709 IDPs in 2006 CQ136 2.8 12.64 4.86 20.07 -2.06 (1.47) 533 IDPs returned in 2006 CQ137 5.4 24.45 3.91 16.04 1.49 (1.8) 518 Refugees in 2006 CQ138 0.57 3.78 0.97 7.29 -0.41 (0.5) 557 Refugees repatriated in 2006 CQ139 0.76 12.13 0.25 3.21 0.52 (0.72) 575 Share of male respondents QF7 0.5 0.5 0.5 0.5 0 (0.01) 23567 Average age respondents QF9 20.12 16.82 20.17 17.02 -0.06 (0.23) 22536 Note: Question number responds to 2010-2011 survey (Humphreys et al. 2012). Tests of difference based on simple OLS regressions. * p<0.10, ** p<0.05, *** p<0.01 (two-tailed). 10 Table A6. Balance Gender Parity and Control Variable Q Tuungane Sd. Control Sd. Diff. (Se.) N Distance mine QE13 11.8 18.93 15.59 33.25 -3.79 (4.11) 169 Distance police post QE13 2.91 4.34 3.36 4.74 -0.45 (0.67) 184 Distance kingdom HQ QE13 7.63 10.32 9.93 10.31 -2.3 (1.53) 183 Ethnic heterogeneity CQ13 0.33 0.27 0.33 0.27 0 (0.04) 171 Religious heterogeneity CQ14 0.53 0.18 0.54 0.15 -0.02 (0.03) 171 Birth year former chief CQ45 1937.83 20.82 1935.52 20.84 2.31 (5.14) 66 Former chief democratic CQ48 0.16 0.37 0.16 0.37 0 (0.06) 150 Share in agriculture CQ15 75.79 17.34 71.61 19.79 4.18 (2.85) 170 Share in herding CQ15 8.48 8.34 9.74 13.34 -1.26 (1.8) 153 Share in commerce CQ15 5.83 14.21 2.41 4.85 3.42** (1.7) 156 Share in fishing CQ15 2.96 8.13 5.64 13.2 -2.68 (1.75) 155 Share in industrial CQ15 0.19 1.27 0 0 0.19 (0.14) 155 Share in mining CQ15 2.01 4.7 2.51 7.09 -0.49 (0.97) 156 Share in other CQ15 3.7 4.41 6.02 10.17 -2.31* (1.26) 157 Share in other services CQ15 5.47 5.11 5.06 4.49 0.41 (0.77) 157 Wells in 2006 CQ23 1.07 2.11 0.88 1.93 0.19 (0.31) 168 Schools in 2006 CQ24 2.37 4.02 2.96 4 -0.6 (0.62) 169 Clinics in 2006 CQ25 0.49 1.37 0.28 0.48 0.21 (0.16) 170 Churches in 2006 CQ26 1.62 1.76 1.68 1.75 -0.06 (0.27) 169 Halls in 2006 CQ27 0.02 0.15 0.07 0.46 -0.05 (0.05) 169 IDPs in 2006 CQ136 3.3 9.51 2.93 12.53 0.37 (2.11) 114 IDPs returned in 2006 CQ137 2.36 12.52 7.79 28.52 -5.43 (4.26) 110 Refugees in 2006 CQ138 0 0 1.1 5.24 -1.1 (0.7) 119 Refugees repatriated in 2006 CQ139 0 0 0.05 0.38 -0.05 (0.05) 122 Share of male respondents QF7 0.5 0.5 0.5 0.5 -3.79 (0.01) 5457 Average age respondents QF9 20.49 17.21 19.81 16.84 -0.45 (0.47) 5198 Note: Question number responds to 2010-2011 survey (Humphreys et al. 2012). Tests of difference based on simple OLS regressions. * p<0.10, ** p<0.05, *** p<0.01 (two-tailed). Data only from the 190 villages that participated in the gender parity lottery. 11 Appendix G: Results by Individual Measures Table A7 provides results for the 171 outcome measures individually to provide a sense of their magnitude and economic significance. Table A7. Results by Individual Outcome # Family Subfamily Outcome Control Tuungane (Se.) N 1 Hospital Building Floor quality 0.82 0.08** 0.04 499 2 Hospital Building Wall quality 0.85 0.06* 0.03 459 3 Hospital Building Infrastructure 0.67 0.01 0.02 500 4 Hospital Building Clean floor 0.75 0.01 0.04 496 5 Hospital Building Clean wall 0.64 0.04 0.05 497 6 Hospital Building Rate building 0.52 0.04* 0.02 3205 7 Hospital Building Toilets 0.77 0.05** 0.02 3166 8 Hospital Capacity # Providers present 3.01 -0.01 0.28 499 9 Hospital Capacity # Beds 8.46 1.38 0.85 494 10 Hospital Capacity Wait personnel 5.2 1.88 1.21 413 11 Hospital Capacity Wait treatment 5.66 0.41 1.02 409 12 Hospital Capacity # Nurses 3.19 -0.05 0.21 496 13 Hospital Capacity # Doctors 0.3 0.01 0.07 469 14 Hospital Capacity Treatments 4.95 -0.04 0.12 490 15 Hospital Capacity Rate capacity 0.45 0.04* 0.02 3181 16 Hospital Capacity Wait time 9.74 0.36 1.07 2909 17 Hospital Material # Antibiotics 1293.38 354.14* 199.77 454 18 Hospital Material # Malaria tablets 781.8 325.52** 151.58 458 19 Hospital Material # Anti-inflammatory tablets 868.85 237.61 171.9 448 20 Hospital Material Rate material 0.41 0.03 0.02 3107 21 Hospital Staff Doctor nurse ratio 0.11 -0.03 0.02 464 22 Hospital Staff Director education 13.5 -0.17 0.27 507 23 Hospital Staff Director medical studies 0.9 -0.01 0.03 500 24 Hospital Staff Rate care 0.58 0.03 0.02 3148 25 Hospital Staff Rate health provider 0.62 0.02 0.02 3110 26 Hospital Staff Presence health provider 0.95 0 0.01 3157 27 Hospital Admin Director present 0.81 -0.01 0.04 502 28 Hospital Admin Patient register 0.93 0.04* 0.02 494 29 Hospital Admin Staff register 0.82 -0.01 0.04 491 30 Hospital Admin Stock register 0.77 0.07* 0.04 475 31 Hospital Admin Cash book 0.72 0.02 0.04 479 32 Hospital Admin Rate administration 0.57 0.05** 0.02 2908 33 Hospital Community # Comm. meetings 4.46 -0.11 0.49 414 34 Hospital Community Contr. in kind 0.1 0.03 0.03 451 12 35 Hospital Community Contr. in $ 0.04 0 0.02 448 36 Hospital Community Know CODESA 9.45 -1.07* 0.63 460 37 Hospital Community CODESA meetings 10.96 -0.95 0.78 458 38 Hospital Community Director present 8.82 -1.17 0.72 471 39 Hospital Community Rate interaction 0.63 0.02 0.02 3092 40 Hospital Community Contr. in kind 0.05 0 0.01 2950 41 Hospital Community Contr. in $ 0.03 0 0.01 2871 42 Hospital Costs Open 0.98 0 0.01 502 43 Hospital Costs $ Paid 10.75 1.44 2.79 374 44 Hospital Costs # Patients now 4.04 0.58 0.69 478 45 Hospital Costs # Patients last month 191.33 -33.07** 16.4 470 46 Hospital Costs Cost index 0.01 0.08 0.05 484 47 Hospital Costs Rate cost 0.28 0.01 0.02 3114 48 Hospital Costs Cost index 0.01 0.04 0.04 2852 49 Hospital Costs # Visits 3.88 -0.03 0.31 3184 50 School Building Floor quality 0.5 0.07* 0.04 494 51 School Building Wall quality 0.74 0.04 0.04 435 52 School Building Roof quality 0.75 0.07** 0.03 495 53 School Building Windows 0.26 0.1** 0.04 497 54 School Building Toilets 0.64 0.05 0.04 498 55 School Building Rate building 0.38 0.06*** 0.02 3154 56 School Building Toilets 0.57 0.03 0.02 3182 57 School Capacity # Classrooms 7.06 -0.47* 0.27 498 58 School Capacity Classroom size 29.03 -2.46 2.71 149 59 School Capacity Highest class 0.86 -0.02 0.03 477 60 School Capacity # Teachers 7.21 -0.03 0.25 451 61 School Capacity # Students reg. 262.69 -35.71** 14.88 473 62 School Capacity Teacher student ratio 0.04 0 0 437 63 School Capacity Rate capacity 0.42 0.04* 0.02 3103 64 School Capacity Classrooms large 0.7 0.07*** 0.02 2918 65 School Material Blackboard 0.96 -0.01 0.03 150 66 School Material # Benches 8.44 2.41** 1.2 153 67 School Material Prop. books 0.22 -0.07 0.06 144 68 School Material Prop. notebooks 0.68 0 0.07 139 69 School Material Teacher book 0.72 -0.04 0.09 139 70 School Material Teacher prep. 0.79 0.02 0.07 141 71 School Material Teacher list 0.75 0.09 0.08 141 72 School Material Rate material 0.35 0.01 0.02 2765 73 School Staff Teacher present 0.94 0.03 0.02 146 74 School Staff Studied pedagogy 0.97 0.02 0.02 143 75 School Staff Director education 10.78 0 0.24 499 76 School Staff Director pedagogy 0.95 0.03 0.02 469 77 School Staff Rate teachers 0.59 0.02 0.02 2797 78 School Staff Teacher absence 0.25 0 0.02 2537 13 79 School Staff Teacher punctual 0.92 0 0.01 2551 80 School Staff Teacher qualified 0.91 0 0.01 2490 81 School Staff Teacher rigorous 0.88 0.01 0.02 2402 82 School Admin Director present 0.67 0 0.04 490 83 School Admin Staff register 0.88 0.03 0.03 426 84 School Admin National program 0.63 0.04 0.05 357 85 School Admin Rate director 0.54 0.02 0.02 2671 86 School Community # Comm. meetings 3.11 -0.36** 0.18 457 87 School Community Contr. in kind 0.28 -0.02 0.04 440 88 School Community Contrib. in $ 0.19 -0.02 0.04 432 89 School Community Know COPA 5.69 -0.18 0.21 470 90 School Community COPA meetings 5.02 -0.78** 0.33 456 91 School Community Director present 4.4 -0.69** 0.34 465 92 School Community Rate interaction 0.6 0.03 0.02 2899 93 School Community Contr. in kind 0.08 0.01 0.01 2867 94 School Community Contr. in $ 0.06 -0.01 0.01 2818 95 School Costs Open 0.32 -0.01 0.04 465 96 School Costs Boys 17.65 3.73** 1.82 143 97 School Costs Girls 15.47 1.58 1.54 143 98 School Costs Students pres. 232.16 -32.98** 13.34 473 99 School Costs School fee ($) 2.11 -0.26* 0.13 477 100 School Costs Fee ($) 0.84 0.01 0.15 381 101 School Costs Rate costs 0.33 0.01 0.02 2875 102 School Costs Cost ($) 87.18 -9.63* 5.23 2247 103 School Costs School fee ($) 2.38 -0.21*** 0.07 2616 104 School Costs Fee ($) 1.77 0.2* 0.11 1677 105 Health Medical care 0.79 -0.03 0.04 434 106 Health U5 mortality 0.14 -0.04 0.03 431 107 Health Death head 0.03 0 0.02 434 108 Health Death other 0.2 0.01 0.04 434 109 Health Sick head 0.34 -0.04 0.04 434 110 Health Sick other 0.58 0.04 0.05 432 111 Education Attendance daughters 1.07 -0.03 0.05 2894 112 Education Attendance sons 1.37 -0.03 0.05 2919 113 Education Never attended (daughters) 0.39 -0.01 0.03 2787 114 Education Never attended (sons) 0.35 0.01 0.04 2777 115 Education Grade (French) 1.98 -0.12 0.1 1259 116 Education Grade (local) 0.92 -0.07 0.11 406 117 Welfare Roof quality 0.38 -0.04** 0.02 3371 118 Welfare Wall quality 0.09 -0.01 0.01 3336 119 Welfare Assets 1.45 -0.04 0.03 3379 120 Welfare Consumption ($) 84.93 -5.69 4.24 3377 121 Welfare Earnings ($) 12.27 0.62 1.29 2919 122 Women Women rights 0.57 -0.02 0.02 3359 14 123 Women Hit women 2.65 -0.13 0.08 3364 124 Women Women association 0.21 0 0.02 3204 125 Women Daughter school attendance 0.44 -0.01 0.01 2162 126 Women Daughter never to school 0.53 0 0.03 806 127 Women Women committee members 0.33 0.03 0.03 210 128 Governance Participation Present meeting 0.42 0.05*** 0.02 3349 129 Governance Participation Voluntary contribution 1.33 0.04 0.05 1393 130 Governance Participation Voted 2011 0.92 0.03*** 0.01 3368 131 Governance Participation Election meeting 0.35 0.02 0.02 3350 132 Governance Participation Right to participate 0.63 0 0.02 3347 133 Governance Participation Interaction 2.37 -0.06 0.14 712 134 Governance Accountability Interaction 0.74 0.04 0.04 3371 135 Governance Accountability Local committee 3.86 0.21 0.13 2620 136 Governance Accountability Chief informs 0.71 0.02 0.02 3253 137 Governance Accountability Other bodies 0.49 0.02 0.02 2898 138 Governance Accountability Influence leaders 0.17 0.01 0.01 3018 139 Governance Accountability Verify leaders 0.27 -0.01 0.01 3343 140 Governance Accountability Local committee 2.66 -0.06 0.15 601 141 Governance Transparency Accept school 0.81 0 0.05 299 142 Governance Transparency Accept health 0.81 -0.05 0.07 228 143 Governance Transparency Knowledge 2.39 -0.02 0.06 3184 144 Governance Transparency Verify chief 0.74 -0.01 0.01 3353 145 Governance Efficiency Approached state 0.06 0 0.01 3024 146 Governance Efficiency Successful state 0.02 0 0 3024 147 Governance Efficiency Approached NGO 0.04 0 0.01 3087 148 Governance Efficiency Successful NGO 0.01 0 0 3084 149 Governance Capture Committee exist 2.01 0.06 0.07 3343 150 Governance Capture Committee elected 0.82 0 0.01 2504 151 Governance Capture # Associations 1.4 -0.04 0.07 3332 152 Governance Capture Association elected 0.76 -0.01 0.02 1652 153 Governance Capture Collected tax 0.06 0 0.01 3300 154 Governance Capture Committee funds 0.26 0 0.02 3201 155 Intra-village Cleavages 1.09 -0.07 0.05 3167 156 Intra-village Trust village member 389.67 -2.03 13.74 1344 157 Intra-village Trust village chief 431.89 -14.48 13.23 1304 158 Intra-village Voluntary projects 0.72 -0.02 0.04 3352 159 Intra-village Community ownership 1.26 0.02 0.11 465 160 Intra-village Development committee 0.4 -0.02 0.03 704 161 Intra-village Committee population 0.94 -0.02 0.03 272 162 Intra-village Committee frequency 1.62 0.21 0.17 252 163 Intra-village # Associations 2.99 0.12 0.16 712 164 Inter-village Cleavages other village 0.97 -0.02 0.05 3124 165 Inter-village Trust other village 378.66 -16.13 12.85 1358 166 Inter-village Committee other 1.36 0.02 0.05 2405 15 167 Inter-village Projects other 0.46 0 0.02 1385 168 Inter-village Associations other 0.65 0.03 0.03 628 169 Inter-village Resources other 0.49 0.05 0.03 457 170 Inter-village Managed conflict 0.27 -0.04 0.03 655 171 Inter-village Hosted other 0.15 0.03 0.03 655 Note: We report sample average treatment effects. Regressions use randomization block fixed effects. Standard errors clustered at the cluster level. ∗ p ≤ 0.10, ∗∗ p ≤ 0.05, ∗∗∗ p ≤ 0.01. 16 Appendix H: Deviations from the Pre-Analysis Plan This study was preregistered at the EGAP registry (ID: [Redacted]) on [Redacted]. The registration took place prior to researcher access to outcome data. In this section, we discuss deviations from the pre-analysis plan. First, the family outcomes governance, women empowerment, intra-village cohesion and inter village cohesion were pre-registered as secondary outcomes. In this manuscript, however, we present them together with the original main outcomes. Second, a number of individual outcomes were originally preregistered but were not included. For the building quality dimensions of service provision related to the primary school and the hospital, electricity (EE38, ES34) and running water (EE32, ES33) were not included because of a lack of variation (almost no facility has running water or electricity). Related to governance’s participation dimension, we did not include the village chief’s opinions about the decision making process in the village (EC194-196), as these are particularly prone to social desirability biases. Related to governance’s transparency dimension, we excluded information related to bribes (Q107, Q123, Q136) because of a lack of variation (few people say they pay bribes). Finally, related to the women empowerment family, for sensitivity reasons, we excluded survey questions about respondents’ opinions related to rape (Q301-305). Including these individual measures, however, does not change this study’s findings. Finally, we originally suggested to conduct subgroup analysis across a wide set of different characteristics: improvements in service provision, gender parity, type of community and type of project. We also suggested to explore unintended consequences of the program, specifically whether the program increased prices in treatment communities. We did not include these additional analyses to avoid making the manuscript unwieldy. 17