Cash Transfers Increase Trust in Local Government

How does a locally-managed conditional cash transfer program impact trust in government? On the one hand, delivering monetary benefits and increasing interactions with government officials (elected and appointed) may increase trust. On the other hand, it can be difficult for citizens to know to whom to attribute a program and reward with greater trust. Further, imposing paternalistic conditions, and possibly prompting citizens to experience feelings of social stigma or guilt could reduce trust. This study answers this question by exploiting the randomized introduction of a locally-managed transfer program in Tanzania in 2010, which included popular election of community management committees to run the program. The analysis reveals that cash transfers can significantly increase trust in leaders. This effect is driven by large increases in trust in elected leaders as opposed to appointed bureaucrats. Perceptions of government responsiveness to citizens' concerns and honesty of leaders also rise, and these improvements are largest where there are more village meetings at baseline. One of the central roles of village meetings is to receive and share information with village residents, providing some evidence on the value of a high-information environment for generating trust in government. The study also finds that records from school and health committees are more readily available in treatment villages. Notably, while stated willingness of citizens to participate in community development projects rises, actual participation in projects and the likelihood of voting do not. Overall, the results suggest little reason to worry that local management of a conditional cash transfer program reduces trust in government or the quality of governance?especially in high-information contexts.

How does a locally-managed conditional cash transfer program impact trust in government? On the one hand, delivering monetary benefits and increasing interactions with government officials (elected and appointed) may increase trust. On the other hand, it can be difficult for citizens to know to whom to attribute a program and reward with greater trust. Further, imposing paternalistic conditions, and possibly prompting citizens to experience feelings of social stigma or guilt could reduce trust. This study answers this question by exploiting the randomized introduction of a locally-managed transfer program in Tanzania in 2010, which included popular election of community management committees to run the program. The analysis reveals that cash transfers can significantly increase trust in leaders. This effect is driven by large increases in trust in elected leaders as opposed to appointed bureaucrats. Perceptions of government responsiveness to citizens' concerns and honesty of leaders also rise, and these improvements are largest where there are more village meetings at baseline. One of the central roles of village meetings is to receive and share information with village residents, providing some evidence on the value of a high-information environment for generating trust in government. The study also finds that records from school and health committees are more readily available in treatment villages. Notably, while stated willingness of citizens to participate in community development projects rises, actual participation in projects and the likelihood of voting do not. Overall, the results suggest little reason to worry that local management of a conditional cash transfer program reduces trust in government or the quality of governance-especially in high-information contexts.

Introduction
How does government provision of social protection impact trust in government, and how does the quality of information available to citizens moderate this relationship? The very existence of government is often predicated on its ability to protect citizens' well-being.
Intuitively, government efforts to provide for that well-being should increase trust. A high level of trust in government is critical not only for healthy state-society relations, but also for the economy; for example, Fukuyama (1995) highlights how social trust plays a role equal to that of physical capital in determining economic prosperity. Cash transfer programs, a popular form of social protection, were ubiquitous in Latin America by the mid-2000s (Fiszbein and Schady, 2009), but just emerging in poorer, African countries. They have since exploded in Africa; by the 2010s, the vast majority of African countries had formally discussed, planned, or piloted a cash transfer program of some form (Garcia and Moore, 2012). However, their expansion has coincided with a general decline in levels of trust across Africa, illustrated in Figure (Afrobarometer, 2008(Afrobarometer, , 2012. Existing research shows that citizens selectively reward government for providing social protection. While several studies show that its provision increases voter turnout and support for incumbent politicians (Chen, 2013;Linos, 2013;Layton and Smith, 2015;Marschall et al., 2016), Mettler and Stonecash (2008) find that it may lower the likelihood of voting, and Ellis and Faricy (2011) find that public opinion is unaffected by the level of federal social spending. Zucco (2013) shows that incumbents may benefit from dispensing a cash transfer program, but impacts are short-term and limited to presidential (and not legislative) candidates. And Kosec and Mo (2018) show that the degree to which cash transfers improve confidence in government may be moderated by perceptions of one's relative poverty level. These studies raise important questions about how the information available to citizens may influence whether and how they reward government for social protection.
The very nature of social protection programs makes their net effect on trust in government ambiguous. On the one hand, they deliver monetary and other benefits that should improve livelihoods, and they signal to citizens the value their government places on their welfare (Hunter and Sugiyama, 2014). Social protection programs may also help individuals build stronger social relationships and make them cooperate more (Adato, 2000;Camacho, 2014;Attanasio et al., 2009Attanasio et al., , 2015. And Camacho (2014) shows that they may increase individuals' exposure to and trust in certain public sector institutions. 1 All of these factors may raise trust in government.
On the other hand, for many social protection programs, it is difficult for citizens to know to whom to attribute the program and thus give credit and greater trust. For example, even if local governments are involved in the delivery of a program, citizens may give all of the credit to the central government or to donors. And, as Zucco (2013) notes, it is unclear whether citizens should reward those who initially introduced a social protection program or those currently dispensing it. Further, such programs often impose paternalistic conditions which could sour state-society relations (Freeland, 2007). Participation in them may further carry a social stigma (Mettler and Stonecash, 2008;Chong et al., 2009;Camacho, 2014;Oduro, 2015) or otherwise cause social tensions between beneficia-ries and non-beneficiaries (Adato, 2000;Adato and Roopnaraine, 2004;Cruces and Rovner, 2008;MacAuslan and Riemenschneider, 2011;Ellis, 2012), thus reducing civic engagement and lowering trust in government. Citizens may also fear that the process for allocating social protection is politicized and unfair; this could affect perceptions of government (Bruhn, 1996;Dahlberg and Johansson, 2002;Guo, 2009;Costa, 2011;Brollo and Nannicini, 2012;Aytaç, 2014). This problem may be especially likely in the case of local government management of a program, where elite capture (and thus a failure to effectively target the poor) may be more likely (Mansuri and Rao, 2004).
We contend that a key factor moderating how receipt of social protection affects trust in government is the availability of information-on government decision-making broadlyspeaking, as well as on the program specifically. We empirically examine both the question of how government provision of social protection impacts trust in government, and how information moderates the relationship, in the context of Tanzania's pilot, community-managed conditional cash transfer program. In 2010, the Government of Tanzania randomized 80 study villages into treatment and control groups of 40 villages each, with control villages to receive the program with a 2.5 year delay. The program conditioned receipt of transfers on child enrollment in and attendance at school and on health clinic visits by both children and the elderly. In all 80 study villages, citizens elected a community management committee (CMC) via secret ballot to select beneficiaries and run the program. In the study context, village meetings play a central role in disseminating information to village residents. We define a high information environment to be one in which there was an above-median number of village assembly meetings prior to program implementation (i.e., four or more per year, which is the number the village council is officially required to hold). While village meetings may have other effects than conveying information, and while variation in meetings is not randomly-assigned, we argue that this is a central goal of meetings, and we show that there are few statistically significant correlates of meetings.
In high-information environments, we anticipated at least three differences in how a local community-managed social protection program is carried out: First, the process by which beneficiaries were selected should be clearer and more transparent, reducing suspicions among both beneficiaries and non-beneficiaries that beneficiaries were selected for political, clientelistic, or personal rather than poverty-related reasons. This should reduce feelings of guilt among beneficiaries and envy among non-beneficiaries, engender goodwill within the community, and allay suspicions of government malfeasance. Second, individuals should be more aware of who makes decisions and better understand the duties and obligations of government related to operation of the program and supply of the services on which the program is conditioned-namely, education and health care. This includes knowledge of requirements related to staffing facilities, record-keeping, sharing information with citizens, and responding to citizens' concerns, among others. This information should in theory increase citizens' expectations of government throughout the time the program is in place, and this should in turn put pressure on government (especially elected as opposed to appointed leaders) to live up to these higher expectations. Finally, individuals should better understand how they themselves can influence the program's implementation-specifically, through voting and attendance at village meetings.
We find that cash transfers can significantly increase trust in leaders. This effect is driven by large increases in trust in elected leaders as opposed to appointed bureaucrats. Perceptions of government responsiveness to citizens' concerns and honesty of leaders also rise, and these improvements are largest where there are more village meetings at baseline. One of the central roles of village meetings is to receive and share information with village residents, providing some evidence about the value of a high-information environment for generating trust in government. We also find that records from school and health committees are more readily available in treatment villages. However, the availability of records unrelated to the conditions of the program-including finance and planning committee, and defense and security committee records-did not change, suggesting that improved government recordkeeping may be restricted to items directly related to the program. Notably, while trust among villagers increases and stated willingness to participate in community projects rises, actual participation in community projects and voting remain unchanged. Overall, concerns that local management of a cash transfer program will destroy trust in government or reduce the quality of governance appear unfounded-especially in high-information contexts.
The remainder of the paper is organized as follows: Section 2 provides background information on local governance in Tanzania, as well as the pilot CCT program whose impacts we evaluate. Section 3 describes the evaluation design, data, and outcomes of interest. Section 4 presents our empirical specification and balance tables showing the outcome of our randomization. Section 5 characterizes our main empirical results. Section 6 concludes.

Local government in rural Tanzania and trust
The importance of trust in local leadership is partly driven by the key role local leaders play in community development. Figure 2 illustrates the structure of government in Tanzania.
As of 2012, the country was organized into 30 regions comprised of 169 districts. Each district is comprised of a set of wards which are then further sub-divided into villages.
In Tanzania, each village has a Village Council, 15 to 25 members in size, which is elected by the Village Assembly (all residents at least 18 years old) every five years, one year prior to the general election. At least a quarter of Village Council members seats are reserved for women (Simonen, 2010). The Village Council is overseen by a Village Chairman, who is also elected by the Village Assembly every five years. The Village Chairman coordinates with district officials, manages local development projects, and presides over village meetings. Each village also has a Village Executive Officer (VEO) who is responsible for supervising village developmental activities, maintaining law and order, and often serving as non-voting secretary for the Village Council. The VEO is not elected, but is appointed by the District Executive Director. Decisions about devel-opment planning and budgeting are made at the district level (Venugopal and Yilmaz, 2010).
( Figure 2 here) Despite the importance of local policy makers in development, recent Afrobarometer surveys suggest that respondents in surveyed countries in Africa as a whole, in surveyed countries in East Africa, in Tanzania, as well as in our three study regions in Tanzania have less trust in local government councils than they do in parliament/ national assemblies or in the president (Figure 3). In Tanzania, while more than 80 percent of respondents interviewed during 2005-2016 reported trust in national leadership, the comparable figure for local leadership was under 75 percent (Afrobarometer, 2005(Afrobarometer, , 2008(Afrobarometer, , 2012(Afrobarometer, , 2016. 2 Statistics for our three study regions in Tanzania are similar to those for Tanzania as a whole, though trust in local government is slightly lower and trust in national leadership is slightly higher. Given increasing pressure around the world for governments to decentralize, or devolve authority to lower levels of government (Gadenne and Singhal, 2014), it is increasingly important to understand what drives trust in local government, and how trust might be raised. Our focus on village meetings as a source of information is rooted in their centrality in Tanzania as a means of keeping citizens informed and enabling them to participate in and influence policy making. In rural villages, these meetings and the Village Council are the two central institutions of governance (Venugopal and Yilmaz, 2010). Village meetings should, by law, be called every three months, and the full Village Assembly is invited to attend. 3 The two key figures involved in calling village assembly meetings are the Village Chairperson, who is elected, and the Village Executive Office, who is a bureaucrat appointed by the district. Meetings are advertised both by formal means, such as posting a notice or sending someone from house to house, and by local/ informal means, such as signaling with drums the night before a meeting. Miguel (2004) notes that "village meetings are held for local elections, to discuss development projects, and to disseminate information from higher levels of government (for example, to promote HIV/AIDS awareness), and they serve as the focal point for local politics." Village leaders also report on financial resources gathered from village residents and how those have been used.
The development literature places substantial weight on participatory democracy; for example, Verba et al. (1995) finds that village meetings may improve governance by providing a forum for citizens to scrutinize their elected representatives. Chattopadhyay and Duflo (2004) find that village leaders invest more in infrastructure directly relevant to the needs of their own gender. And Duflo and Topalova (2004) show that reserving a portion of village council head positions for women leads to more drinking water facilities in the village.
In rural Tanzania, the evidence is mixed on the role of village meetings. Citizens often use village meetings to inform leaders about the state of local government performance and complain when it is unsatisfactory (Lawson and Rakner, 2005). However, the Village Assembly does not have legislative or executive powers, and several studies of rural Tanzania have found that the "grassroots" contribution of the Village Assembly to village decisionmaking is minimal and ineffective (Norman and Massoi, 2010).

Pilot CCT program
Tanzania's pilot CCT program, implemented by the Tanzania Social Action Fund (TASAF, a social fund agency of the Tanzanian government), began delivering transfers in January 2010. Its aims were to increase investments in health for young children (ages 0-5) and the elderly (ages 60 and over) and to increase educational investments for children aged 7-15. It While CCT payments were made at the household level, conditions applied at the individual level. 5 Children aged 0-5 had to visit a health clinic at least six times per year (the condition was relaxed for children aged 2-5 to two visits per year starting in 2012), 6 those age 60 or over had to visit at least once per year, and no health conditions applied to others.
Both preventive and curative visits fulfilled the health clinic visit conditions of the program, though visits had to be to a public facility (either a dispensary, health center, or hospital).
There were no further restrictions on the timing of visits, nor on the services to be received. Children aged 7-15 had to enroll in school and maintain an 80 percent attendance record.
TASAF worked with an elected community management committee (CMC) in each village to select beneficiary households. 7 The CMC surveyed the poorest half of households, collecting data on eight household characteristics: roof material, light supply, water supply, type of toilet, ownership of four different assets (vehicle/motorcycle, radio, iron, poultry), number of windows on the house, household size, and number of meals eaten per day. TASAF then carried out a proxy means test to propose a ranking of households by poverty level, for CMC and village leader approval. On average, 23 percent of households became beneficia-ries. 8 This oversight and validation helped promote community buy-in. Following beneficiary selection, CMCs in treatment villages continued to screen potential beneficiaries, communicate program conditions, and transfer funds. The CMCs also played a key role in monitoring conditions; they were responsible for collecting monitoring forms from health clinics and schools, updating records, delivering warnings when conditions were not met, making home visits to stay abreast of developments in beneficiary households, and conducting regular awareness sessions. A year and a half into the program, over 86 percent of beneficiary households reported that a member of the CMC had visited their household since the program began, and only 1.5 percent reported being asked for part of their transfer.
CMCs existed in both treatment and control communities. All communities had, in the past, implemented community development projects with resources from TASAF, and the CMC was responsable for administering those resources. In control villages, the CMC continued to exist, but did not play an active role in administering resources during the time of the conditional cash transfer program.

Evaluation design
We evaluate the impacts of the CCT program using three waves of data on beneficiaries and would-be beneficiaries (no data were collected from those not selected to be beneficiaries). Table 1 presents the chronology of the program and impact evaluation. A baseline survey was carried out during January-May 2009 and payments began in January 2010. A midline survey was conducted during July-September 2011 (18-21 months after transfers began) and an endline survey was conducted during August-October 2012 (31-34 months after transfers began). 9 The baseline survey included 1,764 households (a subset of beneficiary households) comprised of 6,918 individuals. The quantitative data collection was supplemented by two rounds of qualitative data collection (following the midline and endline surveys) employing focus group discussions and in-depth interviews.

Data and outcomes
We collected information on five broad outcomes. First, we wanted to learn about trust in leaders. In each of the three survey rounds, we collected data from the household head 10 on whether leaders can generally be trusted. At endline, we delved further into this outcome by asking whether the head's trust in village leaders had improved over the last three years.
We additionally asked at endline about trust levels on a scale from 1 to 5 for four leaders/ groups: the village executive officer (VEO), the village chairman (VC), the village council, and the community management committee (CMC). Respondents were informed that on this scale, 1 meant trusting to a very small extent and 5 means trusting to a very great extent. From these ratings we constructed dummy variables for trusting the leader/ group to a great or very great extent.
Second, we wanted to learn about perceptions that leaders are doing their jobs correctlythat is, that they are responsive, honest, and hard-working. Our outcomes include a dummy for household heads believing that local government and leaders take citizens' concerns into account "a lot," a dummy for believing that honesty of local government and leaders has improved in the last three years, a dummy for being somewhat or very satisfied with the work of the village council, and for being very satisfied with the work of the village council. We further considered dummies for the individual considering village school and health facilities to be "good or excellent." Third, we wanted to know about the quality of government record-keeping and transparency. These outcomes were only collected at endline. They include dummies for leaders being able to show reports for four types of village committees: finance and planning, defense and security, school, and health; a dummy for the household head reporting that the village council publicly posts information on projects and finances (i.e., how much money the village council receives and spends, and what they accomplish); and another dummy for that information on projects and finances being freely available.
Fourth, we collected data on voting and civic engagement. Specifically, we asked if the individual voted in the 2012 CMC election, and if they voted in the most recent village council election. 11 We also coded a dummy for the head having attended a village assembly or village council meeting within the last 12 months.
Fifth, we looked at outcomes related to community development and involvement of the household head. Specifically, we coded dummies for willingness to contribute time to communal projects that do not directly benefit the household, for committing money to such projects, for having worked with fellow villagers for the benefit of the community within the last year, and for having contributed labor to a community development project. We also coded a dummy for whether the household head indicated participating in more civic groups than they did three years ago. Finally, we used village-level data from all three survey rounds on whether there was a parent association and a health committee operating in the village.

Empirical specification
We carried out follow-up surveys in 2011 and in 2012 to capture both short-term (1.5 years) and medium-term (2.5 years) impacts of the program. Given random assignment to treatment, we recover causal intent-to-treat estimates from the following empirical specification: where i indexes individuals and t indexes the survey round. c it is a trust-related outcome, For outcomes present only in the endline survey, we obtain intent-to-treat estimates from the following empirical specification: where µ d are district fixed effects and W i is a vector of household-and village-level controls including the head's age, age 2 , sex, and education level; dummies for household size, having an improved roof, having an improved toilet, having an improved floor, and having piped water; the 2009 village population; and the first principal component from a principal components analysis (PCA) using information on ownership of 13 household assets at baseline. In treatment villages, 9.0% of households did not receive treatment-likely due to last-minute changes in community prioritization or household refusal. In control villages, 0.6% of households received treatment-likely due to proximity to a treatment village. As a result, our intent-to-treat estimates represent a lower bound on the impact of receiving transfers.

Heterogeneous treatment effects by number of meetings
We estimate the overall impacts of the CCT program as well as its heterogeneous impacts by the baseline (2009) number of village meetings. We divide villages into two types: those with above-median and below-median 2009 meetings. 13 Villages with above-median 2009 meetings held at least four meetings during the year-the number they are legally supposed to hold. In such villages, information could be provided to citizens, and citizens engaged in the policymaking process, at the legally-intended frequency. This provides insight into how one measure of the quality of the information environment could moderate the effects of a cash transfer program on trust and rural governance more broadly.
To properly interpret these heterogeneous treatment effects, it is helpful to test whether having more village meetings (i.e. an above-median number) at baseline proxies for some other village characteristic. For example, do wealthier, more socially cohesive, or smaller villages have more meetings? Table 2 presents village-level regressions using baseline data; column 1 shows that treatment does not predict more meetings. We next show (column 2) that when controlling for 13 additional village characteristics, only one is a statistically significant predictor of having more meetings at the 10 percent level or higher: average household consumption, significant at the 5 percent level. However, we see no relationship between other measures of household welfare and having more meetings, including average household wealth as measured by an asset index, the share of household heads with some education, an average literacy index, and a Gini coefficient of household consumption.
This lack of statistical significance is noteworthy; asset ownership and other measures of socioeconomic status are often used instead of consumption when measuring poverty (Sahn and Stifel, 2000;Morris et al., 2000;Filmer and Pritchett, 2001;Ferguson et al., 2003;Bader et al., 2017). Arguably, asset ownership and formal education are better indicators of long-term wealth than consumption, which fluctuates more rapidly over time.
We also find that three measures of baseline governance quality and social cohesion are uncorrelated with having more meetings: the share of households that trusts leaders, the share that trusts community members, and the share that contributed labor to a community development project within in the last year. Finally, measures of village size and the age profile of the population are uncorrelated with having more meetings. This bolsters our interpretation of villages with more meetings being distinct predominately due to their being high-information environments, as opposed to their being places that naturally have higher levels of welfare, social cohesion, or other characteristics.

Outcome of the randomization
Despite randomization of villages into treatment and control, it is possible that some observable characteristics of treatment villages are significantly different than those of control villages. If this were the case, one would worry that treatment is correlated with observed and unobserved omitted variables. We address this concern in two ways. First, we show that randomization generally led to balance between treatment and control villages. Second, we use household fixed effects to account for baseline imbalances whenever possible.
In Table 3, we examine differences in baseline means between treatment and control groups for an array of demographic and housing characteristics as well as outcome variables.
From Panel A, we see that households in the treatment and control groups are balanced on the head's sex, the presence of an improved roof, access to toilet facilities, and access to piped water-though treatment households are about 6 percentage points less likely to have an improved floor (significant at the 5 percent level). We also find balance on all six of the trust and governance-related outcomes for which we have data from multiple survey rounds (Panel B). Examining village-level characteristics (Panel C), we find balance on a wide array of characteristics: the number of community meetings in 2009, the numbers of total households in 2009, the number of (prospective, in the case of control communities) beneficiary households in 2009, the share of households that generally trust leaders, the share of households that contributed to a community development project (CDP) in the last year, a household asset index, the share of household heads with some education, an average literacy index, logged average household consumption, a gini coefficient of total value of consumption (a measure of inequality), average age, and the numbers of individuals under age 18 and age 18 or greater. The only imbalances we find are for the share of households that trust community members (treatment villages are 6 percentage points more trusting, significant at the 10 percent level) and average household head age (those in treatment villages are 2.6 years younger, on average (significant at the 1 percent level). Our use of household and/or individual fixed effects (depending on the unit of analysis) accounts for these observed imbalances as well as any unobserved, time-invariant household and/or individual characteristics that might be correlated with treatment.
( Table 3 here) For outcomes only available at endline, we show descriptive statistics in Table 4.
Differences across groups at endline obviously do not inform us about the state of balance at baseline. We posit that the broad balance we observe across baseline characteristics -including trust in leaders -is likely to extend to these other measures, but this is ultimately untestable, and we rely on the randomized nature of the program and the control variables shown in Eq.

Attrition
Between baseline and midline, 8.6 percent of households attrited from the sample, and between baseline and endline, 13.2 percent of households attrited. Evans et al. (2017) show that this attrition is not correlated with treatment or with a variety of household covariates interacted with treatment. 14 Overall, we conclude that attrition does not affect the internal validity of our results.

Results
In this section, we present the impacts of Tanzania's CCT program on a variety of trustrelated outcomes. First, we consider the impacts on trust in and satisfaction with village leaders. Second, we examine the impacts on village and household level measures of local government transparency and record-keeping. Third, we consider impacts on voting behavior and civic engagement. Fourth, we examine how treatment affected voting and civic engagement. Fifth, we consider impacts on community development and involvement. Finally, we explore the robustness of our results to corrections for multiple hypothesis testing.

Trust in leaders
In Table 5, we document the impact of the CCT on self-reported trust in village leaders.
At midline (1.5 years after treatment began), treatment is associated with a 5.2 percentage point increase in the share of households reporting that leaders can generally be trusted. At endline (2.5 years after treatment began), this effect is of a similar magnitude but is more statistically significant. These treatment effects represent about a 6.5 percent increase over the baseline mean of 0.81. Furthermore, households in treatment villages at endline were 4 percentage points more likely to report a belief that trust in leaders had improved over the last three years, i.e., since the program began.
( Table 5 here) The coefficients on the time trends in column 1 help us understand from where these effects on trust in leaders stem: an increase in trust in treatment villages, or a decrease in trust in control villages. We see that in control villages, trust in leaders had dropped by a statistically insignificant 0.027 percentage points by midline, and by a statistically significant 0.052 percentage points by endline; this represents a 3.4 percent drop in trust levels in control villages between baseline and midline (i.e. over 1.5 years) and a 6.5 percent drop between baseline and endline (i.e. over 2.5 years). Comparing these estimates to the coefficients on treatment interacted with the midline and endline dummies, we see that trust increased modestly in treatment villages at midline (0.052 -0.027 = 0.025 percentage points), and increased minimally in treatment villages at endline (0.055 -0.052 = 0.003 percentage points). Thus, the CCT-especially by endline-largely prevented the erosion of trust seen in control villages.
Absent a general erosion of trust in Tanzania between baseline and endline, one might worry that reductions in trust in control villages are due to a "sour grapes" effect, whereby those who had hoped to received transfers were disappointed when they ultimately did not.
If this were the case, cash transfers might not actually increase trust at all; rather, learning about a program and then not getting it could sap trust. Our evidence suggests that this did not take place here: Afrobarometer data on Tanzania as a whole surrounding our study period show a trend of reduced trust in all types of leaders during 2008-2012-the president, parliament, and local government-that is similarly-sloped to the decrease in trust in village leaders we observe in control villages during 2009-2012 (Afrobarometer, 2008(Afrobarometer, , 2012. 15 This is consistent with a negative secular trend in trust in leaders in Tanzania that similarly affected our study villages, rather than a "sour grapes" effect due to the CCT program itself. 16 Figure 5 illustrates these trends. Notably, the VEO is the one leader listed who is not locally elected, but rather appointed by the district. Treatment is associated with an 8 percentage point increase in trusting the village council to a great or very great extent and a 26 percentage point increase in trusting the CMC to a great or very great extent. The latter effect is unsurprising, given the role of the CMC in distributing the cash transfers to beneficiary households. Panel B shows that the impact of the program on trust in leaders at endline is entirely concentrated in villages that had relatively more (i.e. an above-median number of) village meetings at baseline. For villages with more meetings, general trust in leaders was 7 percentage points higher and they were 7 percentage points more likely to report an improvement in trust during the course of the program. In those villages, as a result of treatment, we observe significantly higher trust -relative to control villages that also had more village meetings at baseline -in the village chairman (8 percentage points), the village council (8 percentage points), and the CMC (28 percentage points). Importantly, however, there are no impacts, even in villages with above-median meetings, on support for the VEO -an ap-pointed bureaucrat. For villages with fewer meetings, the only statistically significant effect of treatment is on trust of the CMC (again, the distributors of the cash transfers). Further, the point estimate on the effect of treatment for those living in villages with fewer meetings is always smaller compared to its value for those living in villages with more meetings (across all columns)-though this difference is generally not statistically significant given limited statistical power.
A note is warranted on our R-squared values; as column 1 (the one column using all three rounds of data) of Table 5 reveals, the R-squared is only 0.003. This may seem like a "low" value. However, as de Brauw (2015) highlights, the goal of estimating the effect of a program is not to explain variation in outcome variables, but rather to understand how much the outcome variables change, on average, due to the program. He uses Monte Carlo analysis to show that even when treatment has a moderately large effect on an outcome variable (specifically, increasing a standard normal outcome variable by 0.1 standard deviations), the R-squared on the regression is still extremely small. The conclusion is that policy makers hoping to improve average outcomes should worry about the effects of treatment on the average outcome rather than focusing on the R-squared value.
Overall, these results provide evidence that cash transfers can significantly increase trust in leaders-especially trust in elected leaders as opposed to appointed bureaucrats. 18 A natural question is what is the source of this trust. Are leaders perceived as being more responsive? More honest? Do households feel that leaders are actually doing better work, as measured by the perceived quality of publicly-provided goods, like education and healthcare facilities? We turn to these questions in the next sub-section. Table 6 reports the effects of being in the treatment group on satisfaction with the performance of local government. Households receiving treatment are 7 percentage points more likely to report that local government leaders take citizens' concerns into account "a lot" (column 1) and 4 percentage points more likely to say that honesty of local government and leaders has improved over the last three years (column 2). Reinforcing the increased trust in the village council that we saw in Table 5, treatment is associated with a significant increase in being either somewhat (6.5 percentage points) or very satisfied (10 percentage points)

Satisfaction with local government
with the work of the village council (columns 3 and 4). On the other hand, households reported higher perceptions of the quality of schools and health facilities at midline, but by endline, perceptions had returned to close to baseline levels (columns 5 and 6). This may reflect an initial favorable perception due to increased use and awareness of school and health services (and potentially even higher quality public services), but increased exposure can also lead to changes in reference points when one grows accustomed to the services.
( Table 6 here) As with trust in local leaders, these effects are entirely concentrated in villages with more baseline village meetings (Panel B). For those villages, the association between treatment and all 6 outcomes is positive and statistically significant at either the 0.05 or 0.01 level. Households in villages with more baseline village meetings report that local government leaders are 13 percentage points more likely to take citizens' concerns into account a lot (column 1). Households in villages with more baseline village meetings are 21 percentage points more likely to be very satisfied with the work of the village council (column 4). Even the improved perceptions of schools and health facilities persist to endline (columns 5 and 6). At endline, treatment effects for villages with fewer meetings were significantly different from those with more for all but two outcomes.
Trust and satisfaction with government performance are closely linked and likely highly correlated. It is thus noteworthy that we see improvements in both; this raises questions about whether improvements in government performance are actually driving improvements in trust. Our experiment does not allow us to causally test whether or not this mechanism is at work. However, we can test, in a village-level regression, whether treatment predicts a greater correlation between trust in leaders and measures of satisfaction with government (Table 7). We find that the coefficient on treatment is always positively signed when considering as our outcome the correlation coefficient at endline between a dummy for generally trusting leaders and each of four dummies measuring satisfaction with government: considering community health (school) facilities to be good or excellent, being somewhat or very satisfied with the work of the village council, and being very satisfied with its work.
While the coefficient on treatment is never statistically significant at conventional levels, this suggests that at the very least treatment does not significantly reduce the extent to which individuals base trust in leaders on the perception that they are actually doing a satisfactory job governing.

Government record-keeping
Beyond perceptions, there is some indication that treatment villages keep better records (Table 8). Specifically, treatment is associated with a significantly higher likelihood that the village can show reports from the village education committee (24 percentage points) and health commitee (also 24 percentage points) from the two months prior to the the interview (columns 5 and 6). Such reports are important since they make policy making in these domains more transparent and can thus create opportunities for citizens to weigh in on policy decisions. We do not observe the same improvement in record keeping in other areas, such as finance and planning, or defense and security (columns 1 -4). This may well be a consequence of the fact that the cash transfers are conditioned on health and education, and the communities have the responsibility to monitor those conditions. It could also reflect the fact that, due to the program, citizens had a renewed interest in using the services (since using them is a requirement to receive transfers), and thus greater knowledge and interest in government decisions about these services-pressuring government to release information about schools and health care facilities. A positive spillover benefit of the conditions may thus be improved record keeping in these areas.
( Despite measured improvements in the availability of village education and health committee reports as a result of treatment, our regressions say nothing about the quality of those reports or the information they convey. They suggest that committee meetings were held, and that at least some citizens can learn about what is happening in the committees through them, but not how participatory meetings were, what was discussed, how well the reports capture the essence of meetings, or even how available these reports are to citizens and policymakers. As such, these results on government record-keeping should be interpreted with caution.

Voting and civic engagement
The increased trust in village leaders and improved perceptions of their work does not appear to translate to increased political activity. Treatment is not associated with a higher incidence of attending a village assembly or council meeting, nor of voting in the village council election (

Community trust and networks
To this point, we have demonstrated increased trust and confidence in village leaders as a result of the CCT. One potential manifestation of that is through political action, which we do not observe. But another manifestation is through increased community action, as village leaders can play a role in encouraging or facilitating community development projects. Table 10 reports the CCT's effect on household participation in community development.
Households do report an increased willingness to contribute to community projects (columns 1 and 2), whether through time (5 percentage points) or money (6 percentage points), but we observe no evidence that households translate that willingness into action: households in treatment villages are no more likely to have worked on community development projects in the last year (columns 3 and 4). More households in CCT villages do report the existence of a health committee, but not a parent association, despite the fact that neither was near-universal at baseline (columns 6 and 7). Households in treatment villages do report a small but highly significant increase in membership in community groups overall (column 5).
( Table 10 here) Across villages with more or fewer meetings, the increased willingness to contribute is concentrated among villages with fewer meetings (Panel B), but there is still no effect on actual labor contribution to community projects in the last year. The increased reported existence of health committees is observed in both types of villages.

Robustness checks
Because this program tests a wide array of outcomes, several of which were identified after the program was underway, it is important to correct our findings for potential false-positive results resulting from testing multiple hypotheses. One popular method is the Benjamini-Krieger-Yekutieli (BKY) method, which controls for the false discovery rate (Benjamini et al. 2006). In other words, in limits the "expected proportion of rejections are type I errors," or false positives (Anderson 2008). 20 Table 11 shows the results of our analysis. We use two approaches, first grouping the results all together (column 8) and then by table (column 10).
(  (Table 5), the ability of communities to show records (Table 8), and community development and involvement (Table 10). On the whole, our main findingsthat conditional cash transfers increase trust in leaders, improve record keeping in areas related to the transfers, and increase reported willingness to participate in community projects -hold, as well as our important "non-findings," that such increased trust does not appear to translate into voting behavior or community action.
We still demonstrate evidence for our heterogeneous treatment effects result -that the trust impacts of the program are concentrated in villages with more meetings at baseline. Of 16 significant effects for villages with more meetings in the main tables, 4 or 5 of those remain significant, depending on the grouping of hypotheses. But those that remain significant still suggest that the effect on leaders being trustworthy, on trusting the village chairman, and on the existence of a village health committee remain.

Conclusion
This paper provides evidence that, after 2.5 years, Tanzania Overall, our study has at least two key lessons. First, concerns that local management of a cash transfer program will destroy trust in government or reduce the quality of governance appear unfounded-especially in high-information contexts. Second, while the difference in treatment effects according to baseline village meetings cannot be causally attributed to the meetings, we provide some support to the argument that having information available to citizens can improve the impacts of a community-managed program on communal trust.
Unfortunately, there are several questions that we are unable to answer with this survey.
We are unable to clarify the mechanism through which this CCT program operates. The re-sults could be driven by something as simple as a small reduction in poverty (Bastagli et al., 2016) and improved health (Evans et al., 2017). Or perhaps simply being given what is essentially a gift changes one's outlook. Due to the survey design 21 we cannot measure the impact of this CCT program on nonbeneficiaries within a treatment village, and there is some evidence from other contexts of adverse impacts on nonbeneficiaries (Adato, 2000;Adato and Roopnaraine, 2004;Haushofer et al., 2015;Haushofer and Shapiro, 2018). However, we do not observe retaliation towards beneficiaries: the program actually improved safety nets for beneficiaries and we find no evidence of a significant reduction in transfers that beneficiaries receive from other households (Evans et al., 2014).
This paper contributes to a better understanding of some of the less direct effects of CCTs. More research is needed to understand how nonbeneficiaries may be affected, but absent negative effects for that group, trust in community leadership should be added to the list of areas that have been shown to benefit from a modest investment in the neediest members of a community.

Notes
1 Specifically, Peru's Juntos CCT increased trust in institutions related to the conditions of Juntos: the national office in charge of identity registration, the Ministry of Health, and the Ministry of Education.
However, it decreased trust in the institution that handled the complaints related to program targeting.
2 It is worth noting that trust in all levels of government is higher in Tanzania than in either East Africa or Africa as a whole.
3 A recent sample of 29 villages in northern Tanzania suggests significant variation in household attendance of the village assembly. The share of households that had participated in a village assembly in the past 12 months ranged from 8 to 75 percent, and the average was 34 percent.
4 At baseline, villages ranged from 64 to 10,078 households. The average size was 980 households, and the median size was 560 households 5 We lack administrative data on compliance with conditions. However, in each follow-up survey, we asked: "[For your last transfer payment,] did you receive less money than you usually get?" and "What do you think was the reason?" While one may hesitate to admit to non-compliance (for fear of sanction) and while this cannot tell us how many households had at least one payment reduced (it only tells us about the last), this gives some indication of compliance levels. At midline and endline respectively, 1.9 and 3.0 percent of treatment households reported receiving less than usual for a reason related to not meeting conditions. 6 As our endline survey was carried out during August-October 2012, we define compliance with clinic visit conditions at endline for 2-5 year olds as having two or more clinic visits in the last year. 7 CMC elections occurred at village meetings; 10-14 members were elected, with secret ballots. To run, a candidate had to have received financial training and successfully managed a past TASAF-supported project. 12 A Pearson's chi-squared test for baseline values of our main, binary outcome, "believes leaders can generally be trusted," suggests that the data are not homogeneous across the three districts (p-value = 0.000). However, the government identified the three pilot districts for the cash transfer program so as to represent the diverse array of villages present in Tanzania, and was interested in receiving a single answer about the value of the program for households in districts such as these. Further, our overall sample size was carefully selected using power calculations such that we require this sample size to make inference. Pooling data across the survey sites was thus both a) practical, from a policy standpoint, and b) necessary, from a statistical power standpoint.
13 36 of the 80 villages had three or fewer meetings while the remaining 44 had at least four meetings.
14 Specifically, the authors compute F-statistic for the joint significance of treatment, these covariates, and their interactions. 15 Afrobarometer data on Tanzania are not available for 2008.
16 Specifically, the percentage of Tanzanians saying they trust the president, parliament, and the local government council either "somewhat" or "a lot" dropped from 88.6, 84.1, and 75.8 in 2008 to 74.2, 77.0, and 67.5 in 2012, respectively (Afrobarometer, 2008(Afrobarometer, , 2012. In comparison, in our sample the share of respondents saying that they can generally trust leaders in their village declined from 80.5 percent in 2009 to 74.7 percent in 2012-a similarly-sloped reduction during the period. 17 We do not have political knowledge questions like "Who is your VEO?" or "Who is the president?" 18 All of the results from Table 5 are robust to omitting individual fixed effects; indeed, the coefficients in columns 1-6 are uniformly larger in magnitude when we omit them. 19 We did not estimate heterogeneous treatment effects for the village-level specifications (columns 1-6) because there are only 80 observations. 20 Controlling the false discovery rate contrasts with controlling the family-wise error rate, or the "probability of rejecting at least one null hypothesis." FWER is most appropriate for cases where the cost of a false rejection of the null hypothesis has strong policy implications (Anderson 2008).  Source: Afrobarometer (2005Afrobarometer ( , 2008Afrobarometer ( , 2012Afrobarometer ( , 2016 Notes: The bars measure the share of individuals who indicate that they trust politicians at three levels (president, parliament, and local government council) "somewhat" or "a lot" over four different rounds of the Afrobarometer. Responses for a given year are weighted according to the national population and distribution of the sample based on individual selection probabilities (i.e. based on region, gender, urban-rural distribution, and size of household and enumeration area). The list of countries included in Afrobarometer has grown over time. To exclude any composition effects, we only include countries that were present in all rounds. The 18 countries included in the Africa sample are: Benin, Botswana, Cape Verde, Ghana, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique, Namibia, Nigeria, Senegal, South Africa, Tanzania, Uganda, Zambia, and Zimbabwe.

Figure 3: Trust in leadership
Source: Afrobarometer (2005Afrobarometer ( , 2008Afrobarometer ( , 2012Afrobarometer ( , 2016 Notes: The bars measure the share of individuals who indicate that they trust "somewhat" or "a lot" over four different rounds of the Afrobarometer. Responses for a given year are weighted according to the national population and distribution of the sample based on individual selection probabilities (i.e. based on region, gender, urban-rural distribution, and size of household and enumeration area). When averaging across all four rounds, each round is weighted equally. The list of countries included in Afrobarometer has grown over time. To exclude any composition effects, we only include countries that were present in all four rounds.   Source: Afrobarometer (2008Afrobarometer ( , 2012 and authors' calculations based on World Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: The lines measuring the trust in the president, parliament, and the local government council are from Afrobarometer. They represent the share of individuals who indicate that they trust "somewhat" or "a lot". Responses are weighted according to selection probabilities (i.e. based on region, gender, urban-rural distribution, and size of household and enumeration area). Trust in village leaders is a measure from the household survey, and it is measured only in the control communities.      Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: Fewer refers to those residing in villages in the bottom half of the distribution of baseline village meetings per year, while more refers to those in the top half. Column 1 includes household fixed effects. Columns 2 -6 include controls for age, age 2 , sex, and education level of the household head. Also included are dummies for district, household size, having an improved roof, having an improved toilet, having an improved floor, having piped water, logged 2009 village population, and the first principal components from a PCA using information on ownership of 13 household assets at baseline. Standard errors are in parentheses and clustered at the village level. *** indicates p<0.01; ** indicates p<0.05; and * indicates p<0.10.  Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: Fewer refers to those residing in villages in the bottom half of the distribution of baseline village meetings per year, while more refers to those in the top half. Columns 1-4 include controls for age, age 2 , sex, and education level of the household head. Also included are dummies for district, household size, having an improved roof, having an improved toilet, having an improved floor, having piped water, logged 2009 village population, and the first principal components from a PCA using information on ownership of 13 household assets at baseline. Columns 5-6 include household fixed effects. Standard errors are in parentheses and clustered at the village level. *** indicates p<0.01; ** indicates p<0.05; and * indicates p<0.10. Source: Authors' calculations based on World Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: The number of observations is less than 80 because the correlation is missing when there is not within-village variation in both variables (measure of trust and measure of satisfaction). Standard errors are in parentheses and clustered at the village level. *** indicates p<0.01; ** indicates p<0.05; and * indicates p<0.10.  Bank (2012), the endline round of the Tanzania pilot CCT impact evaluation household survey. Notes: Only month and year information (no day of month) is available for the defense/security, school, and health reports. The outcome was coded 1 if the number of days between the endline interview (for which we know exact date) and the meeting was less than 62 days (2 months). The date of the meeting was arbitrarily chosen to be the 15th of the month reported. Fewer refers to those residing in villages in the bottom half of the distribution of baseline village meetings per year, while more refers to those in the top half. Columns 1 -3 includes controls for logged 2009 village population and district controls. Heterogeneous treatment effects were not calculated for these three village-level outcomes. Columns 4 -5 include controls for age, age 2 , sex, and education level of the household head. Also included are dummies for district, household size, having an improved roof, having an improved toilet, having an improved floor, having piped water, logged 2009 village population, and the first principal components from a PCA using information on ownership of 13 household assets at baseline. Standard errors are in parentheses and clustered at the village level. *** indicates p<0.01; ** indicates p<0.05; and * indicates p<0.10.   Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: Columns 1, 2, 3, and 5 include controls for age, age 2 , sex, and education level of the household head. Also included are dummies for district, household size, having an improved roof, having an improved toilet, having an improved floor, having piped water, logged 2009 village population, and the first principal components from a PCA using information on ownership of 13 household assets at baseline. Column 4 includes household fixed effects. Columns 6 and 7 include village fixed effects. Treatment estimates are estimates of the effect of living in a treatment village (intent to treat). Standard errors are in parentheses and clustered at the village level. *** indicates p<0.01; ** indicates p<0.05; and * indicates p<0.10.  Bank (2009Bank ( , 2011Bank ( , 2012, which correspond to the baseline, midline, and endline rounds of the Tanzania pilot CCT impact evaluation household survey, respectively. Notes: Treatment effect estimates with p-values<0.10 displayed. Midline and endline treatment effects are abbreviated T×2011 and T×2012, respectively. More and fewer refer to whether the village has more or fewer village meetings than the median village in the sample. Column refers to the column in which the estimate appears in the original table. BKY stands for Benjamini, Krieger, and Yekutieli q-values-the smallest level at which the the null hypothesis that the coefficient is 0 is rejected. An x in Columns 8 and 10 indicates that the q-value exceeds a standard significance level of 0.10).