WPS6021 Policy Research Working Paper 6021 Why Quality Matters Rebuilding Trustworthy Local Government in Post-Conflict Sierra Leone Audrey Sacks Marco Larizza The World Bank Poverty Reduction and Economic Management Network Public Sector Governance Group & Africa Region Public Sector Reform and Capacity Building Unit April 2012 Policy Research Working Paper 6021 Abstract A broad consensus has emerged among practitioners government performance on citizens’ beliefs about and researchers that failure to build accountable and the trustworthiness of local government authorities. legitimate institutions is a critical risk factor associated To test the hypothesis, it uses multilevel models to with vicious circles of repeated violence. Despite this exploit variation over time and within and across sub- consensus, very few studies have tested the extent to national units in Sierra Leone. The results suggest that which local government performance and decentralized improvements in the quality of decentralized service service provision shape citizens’ beliefs toward political delivery, as well as perceptions of local councillors’ authorities. This paper contributes to fill this gap honesty, are positively associated with perceptions of local by examining the antecedents of trustworthy local government officials as trustworthy political authorities. government authorities in a post-conflict and fragile These findings speak to the possibility that local service setting, Sierra Leone. Taking advantage of a unique provision can play a role in shaping the relationships longitudinal survey, the National Public Services, it between citizens and the state and in overcoming the root examines the impact of sub-national variation in local causes of fragility and conflict. This paper is a product of the Public Sector Governance Group, Poverty Reduction and Economic Management Network; and the Public Sector Reform and Capacity Building Unit, Africa Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at asacks@worldbank.org and mlarizza@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Why Quality Matters: Rebuilding Trustworthy Local Government in Post-Conflict Sierra Leone∗ Audrey Sacks†and Marco Larizza‡ Keywords: Sierra Leone, Trust, Legitimacy, Decentralization JEL: H1,H7, D73, D74 Sector Board: PSM ∗ This paper is part of a larger research project on “Political Decentralization, Electoral Accountability and Government Legitimacy� managed by the Public Sector Governance Group, Poverty Reduction and Economic Management Network and the Public Sector Reform and Capacity Building unit, Africa region. We wish to thank the Institutional Reform and Capacity Building Project (IRCBP) staff and the Abdul Latif Jameel Poverty Action Lab at the Massachusetts Institute of Technology for granting access to the survey data used for the analysis. The authors are also greatly indebted to Brendan Glynn, Margaux Janin Hall, Stuti Khemani, Jens Kromann Kristensen, Vivek Srivastava, Erik Wibbels and participants of Duke’s Political Economy Workshop for their valuable feedback and suggestions. † asacks@worldbank.org. Leadership and Governance Practice, World Bank Institute ‡ mlarizza@worldbank.org Public Sector Reform and Capacity Building, Africa region, the World Bank. Visiting Fellow, Institute for Democracy and Conflict Resolution University of Essex, UK 1 1 Introduction A broad consensus has emerged among practitioners and researchers that failure to build accountable and legitimate institutions is a critical risk factor associated with vicious circles of repeated violence (World Bank 2011). When disenchantment with poor governance and lack of improvement in service provision is pervasive, a negative feedback cycle might begin with citizens less willing to comply with government demands, laws and regulations. Over time, the inability of the state to deliver basic public services often presents opportunities for the emergence of rebel groups and other non-state actors that provide a credible and legitimate alternative to the state, eventually leading to civil war and the collapse of state authority (Bates, 2008a). This legacy of distrust and discontent poses an additional burden in a post-conflict and fragile environment, in which demands upon the state’s administrative apparatus are high as the government is under pressure to showing quick peace dividends across different groups within society (Call and Wyeth, 2008; World Bank, 2011). Decentralization reforms and local service provision have often been advocated as critical components of the peace-building and state-building process, and pursued as a tool for mitigating social tensions, rebuilding trustworthy state authorities and consequently, enhancing the prospects for political stability. Often backed by the �nancial and technical support of the international community, these decentralization reforms aiming to devolve basic service delivery functions have been designed to strengthen the decision-making power of local governments and empower the citizens of local communities with a greater policy-making and oversight role. If the performance of newly empowered local authorities matches the expectations of citizens, these reforms are expected to have a positive effect on citizens’ beliefs and holds “great potential to...prevent widespread disillusionment with new policies from turning into a rejection of the entire democratic process... [and] boost legitimacy by making government more responsive to citizen needs� (Diamond, 1999, pp. 124-125) The alleged bene�ts of decentralized public service provision have been largely debated in the academic literature, with special attention paid to its potential conflict-mitigation strategies in the context of fragile and ethnically diverse polities (Brancati, 2006; Horowitz, 1985; Treisman, 2007). Few studies test the extent to which local government performance and decentralized service provision shape citizens’ beliefs toward political authorities and enhance the prospects of political legitimacy. This paper contributes to �ll this gap by examining the antecedents of trustworthy local government authorities in a post-conflict and fragile setting, Sierra Leone, using longitudinal household data that was collected one year after an ambitious program of decentralization reform was launched and two times thereafter. Our argument is that the quality of decentralized public services, as well as the competence and perceived honesty of local government authorities, shape citizens’ beliefs towards local political authorities. In other work (Sacks and Larizza, 2011), we examine how government trustworthiness translates into behavioral legitimacy, as reflected by quasi- voluntary compliance with state demands. Together, these studies complement other ongoing research examining the development of procedural justice and its role in generating beliefs about state legitimacy in rural and peri-urban Sierra Leone (Manning, 2008, 2009). 2 To test our hypothesis, we take advantage of a unique longitudinal survey on the Na- tional Public Services (NPS) covering over 6,000 households throughout Sierra Leone and use multilevel models to exploit variation over time and within and across sub-national units in Sierra Leone. Our analysis shows that the quality of decentralized service delivery is posi- tively associated with views of local government authorities as trustworthy. We also �nd that the perceived honesty of councillors is positively correlated with councillors’ trustworthiness and that this �nding is stronger in more remote communities. These results are robust to controls for alternative determinants of trust including ethnicity, the history of war violence, NGO service provision, councillors’ outreach efforts, social capital, and political party affili- ation. Our �ndings speak to the possibility that decentralization reforms aiming to improve local service provision will help to overcome long-lasting patterns of popular discontent and distrust toward political authorities and to enhance the prospects of political stability in Sierra Leone. The paper is organized as follows: Section 2 outlines the theoretical framework and reviews the empirical literature on the links between government performance, administrative competence and honesty, and trustworthy government. Section 3 describes the advantages of the Sierra Leone case-study while Section 4 describe our data, measures and methods. In Section 5, we analyze the sources of beliefs about the trustworthiness of local government officials and discuss the results. Finally, in Section 6, we consider the relevance and policy implications of our �ndings for the ongoing efforts to support local governments and build accountable institutions in post-conflict and fragile settings. 2 Theoretical Framework 2.1 Trustworthy Government: Sources and Consequences The majority of governments, especially those in transitional and post-conflict societies, are struggling to establish regimes of widespread compliance based on legitimacy and not solely upon coercion. A government perceived as legitimate can expect widespread public coopera- tion for such voluntary acts as voting, adherence to public health recommendations (Lieber- man, 2007) and participation in community problem solving, as well as quasi-voluntary compliance with taxes (Levi and Sacks, 2009). Legitimating beliefs, or the obligation to defer to government (Weber, 1968), rests, in part, on evaluations that government is ob- jectively meeting the normative criteria a citizen possesses about what government should be doing. One of the most important antecedent conditions of value-based legitimacy is the trustworthiness of government. A belief that government is trustworthy reflects judg- ments that authorities are motivated to deliver on their promises and to do what is right for the people that they serve. The trustworthiness of political authorities is also enhanced by positive assessments of government’s performance in delivering essential public goods and services and its administrative competence (Cook, Hardin and Levi, 2005; Levi, 1988; Roth- stein, 2005). The other important antecedent is the extent to which government upholds procedural justice (Levi, Sacks and Tyler, 2009; Tyler, 2006). 3 An evaluation of government as trustworthy reflects the judgment that authorities are motivated to ful�ll their promises and do what is right for the societies they serve. When citizens conflate liking a leader with assessments of trustworthiness, their trust of leaders will be driven, in part, by leaders’ popularity. Survey evidence from the U.S. suggests trust in national government often reflects shifts in presidential popularity (Levi and Stoker, 2000). Charisma has often proved a basis for citizens’ evaluations of the trustworthiness of lead- ers, some of whom are trustworthy and others of whom are not. Identifying what drives a leader’s decisions or when or how her personal motives have changed is extremely difficult. Thus, the best objective measure of the trustworthiness of the motivations of government leaders is citizens’ assessment of the quality of institutions in which political authorities are embedded. In countries where citizens assess institutions as effectively selecting and con- straining government personnel and as delivering the services they need, citizens’ evaluations of personal motivations are less important than their judgment that the institutions work (Hardin, 2002; Levi, 1988). When citizens are con�dent that government has the competence to produce promised services, they are more likely to see government agents as trustworthy. A government’s ad- ministrative competence has two attributes: honesty, and the capacity to implement rules and regulations. Citizens are more likely to view government agents as trustworthy when they perceive that government can ensure that revenues will produce the services they ex- pect and government promises. High levels of corruption can undermine this perception, particularly when the corruption takes the form of elected officials and government agents pocketing the revenues rather than spending them on public goods. Corruption reduces the resources government can spend on the poor through social programs and transfers resources generally from the poor to the wealthy (Rothstein and Uslaner, 2005; Tanzi, 1998). Cor- rupt governments are also left with less money to spend on public officials’ salaries creating incentives for government agents to accept bribes and prey upon citizens rather than serve them (Bates, 2001, 2008b). Patronage or other factors that undermine the meritocratic selection of government agents may erode citizen con�dence, particularly if the officeholders are poor at their jobs or if the effect is to reduce agents’ incentives to implement the law among certain segments of the population (Brewer, 1989; Seligson, 2002). By establishing meritocratic mechanisms for the selection and retention of quali�ed officials, bureaucracies reduce predatory activities and increase the capacity of bureaucracies to implement social policies that bene�t a large portion of the population. States with competent and honest bureaucracies are better able to implement their government’s objectives and, should they choose, provide higher levels of goods provision (Ziblatt, 2008). According to Weber (1968, p. 905), the purpose of a state is to provide citizens with collective goods including: the enactment of law; the protection of personal safety and public order; the cultivation of hygiene, social welfare and cultural interests; and, the organized armed protection against outside attack. Theories of state-building emphasize that citizens direct their loyalty to the level of government that is providing them with welfare-enhancing goods (Barkey, 1994; Hechter, 2000). State-building and the expansion of direct rule in 4 pre-modern Europe were characterized by the central government wresting the obligation to provide public goods from local authorities. As direct rule proceeds, individuals become dependent on the central government for their access to protection, dispute resolution, health care, education and other welfare-enhancing goods. By providing these goods, central rulers increase the dependence of their subjects at the expense of local authorities (Hechter, 2000, p. 60). One possible basis for the perception that the central government is meeting its obliga- tions to citizens is provision of those public goods the population requires to ensure at least a minimal level of social welfare, especially security of life and property. Some public goods, such as education or publicly provided health care, may matter more for certain groups or classes than for others. For example, everyone requires sanitation, but some may be satis�ed with sewage management while others expect garbage collection, street-cleaning, and flush toilets (Levi and Sacks, 2009). Expectations about service delivery appear to affect satisfaction with service delivery. Citizens may primarily perceive and reward relative improvements or sanction deteriora- tions in services, rather than react to the absolute level of service quality they receive. Government plays an important role in forming expectations of what it can deliver and then persuading citizens that the government is in fact meeting those expectations. Fjeldstad, Katera and Ngelewa (November 2008) �nd substantial differences in perceptions of the level of improvement in service delivery among localities in Tanzania. When states fail to provide adequate services and protections for their population, citizens are likely to view state agents as untrustworthy and may withdraw their support from the state. Over time, the inability of the state to deliver basic public services often leads to the emergence of rebel groups and other non-state actors that provide a credible and legitimate alternative to the state, eventually leading to war and state failure (Bates, 2008b). When those in power lack the resources to buy the loyalty of influential groups, including officers or regional elites, the state begins to fall apart as it loses its monopoly over the means of coercion (Azam and Mesnard, 2002; Bates, 2008a). The more dependent citizens are on non- state actors for services, the greater the authority non-state actors are able to exercise. If the cost of exclusion is high and members risk losing not only welfare goods but also their social network or religious capital, they are more likely to comply with demands to participate in acts of violence and war (Hechter, 1987). A growing body of evidence is consistent with this argument and suggests that rebel groups continue to flourish where they are able to provide the goods that citizens are not receiving from their own government like education, health care, poverty programs, property rights, �nancial services, and police protection (Iannaccone and Berman, 2006, p. 119). 2.2 Alternative Explanations We examine alternative drivers of citizens’ beliefs about councillors trustworthiness includ- ing non-state actor service provision, councillors’ outreach efforts and ethnicity. In many developing countries and especially post-conflict states like Sierra Leone, donors and non- state actors including NGOs and religious organizations supply citizens with essential goods 5 and services. Non-state actors often deliver services directly to communities as a means to bypass weak and corrupt central governments. Several scholars suggest that when citizens credit donors, NGOs and other non-state actors for service provision, they are less likely to view the government as legitimate and are less likely to quasi-voluntarily comply with government laws and regulations. Moore (2007) also argues that non-state actor service pro- vision is likely to undermine citizen/state relationships by creating allegiances to non-state providers. Similarly, Lake (2010, p. 277) argues that “When the United States posts signs at new public works projects highlighting its contributions to the local population, or hands out school supplies to children, it diverts support likely accorded by the people from the very state it is trying to help become legitimate.� Alternatively, when citizens view their govern- ment as essential to leveraging and regulating external resources, non-state service provision might strengthen citizens’ beliefs about their government’s legitimacy. In some contexts, local councillors may be able to successfully claim credit for donor and non-state actor re- sources even if they are not actively trying to attract and manage donor and non-state actor resources. As the number of service providers increases, it becomes increasingly difficult for citizens to know exactly to whom to attribute service delivery. The literature on clarity of respon- sibility in the context of voting behavior offers some insights on the problem of attribution (Anderson, 2000; Duch and Stevenson, 2008; Samuels, 2004; Tavits, 2007). The ability of voters to sanction their incumbent governments depends on whether citizens can identify and assign responsibility to the correct government actor. Low clarity of responsibility makes it difficult for citizens to gather information about those actually responsible for government policies and outcomes (Powell, 2000). Multiple tiers of government make it harder, not eas- ier, to know to whom to attribute service delivery. In federal states, responsibilities are often shared across levels rather than neatly assigned to one level or another, thereby obscuring actual policy-making responsibility (Treisman, 2007, p. 13). Sub-national governments are �nanced by a variety of sources, some of which are opaque, including intergovernmental transfers from taxes raised at other levels of government (Wibbels, 2006). As a result, cit- izens have a difficult time assigning responsibility for policy outcomes to the correct agent and consequently take informational shortcuts that lead them to attribute responsibility to the incorrect actor (Anderson, 2007; Leon, 2010). Sacks (2011) introduces a theoretical framework that seeks to explain the factors that influence citizens’ attribution judgments. She argues that attribution judgments are, in part, a function of the extent to which government and non-state actors advertise their contri- butions to citizens. NGOs, churches and donor organizations advertise their contributions to attract funding — donations, grants and contracts (Ahmed and Green, 1999; Hankin- son, 2000; Lake, 2010). The more competition there is for funding, the greater the level of advertising we are likely to observe. The amount of resources which a government invests in advertising its contributions to citizens is, in part, a function of the extent of electoral competition (Bates, 1983; Chandra, 2004; Lake and Baum, 2001; Posner, 2005). Whether citizens evaluate their local councillors as trustworthy may also depend on councillors’ outreach efforts. When councillors spend time meeting with their constituents, 6 listening to their complaints and responding with respect and genuine concern, constituents are likely to develop a positive view of their councillors and more generally, state institutions (Cook, Hardin and Levi, 2005; Lipsky, 1980). A large literature on politics in Africa suggests that politicians reach out to their constituents, at least in part, by distributing targeted material goods (Posner, 2005; Wantchekon, 2003). Empirical research also suggests that households which have greater access to information are more likely to receive targeted bene�ts from government programs (Keefer and Khemani, 2012, p. 5). The more informed citizens are about government programs, the more likely they are to try and hold government accountable for delivering these bene�ts. The expectation that informed citizens will push for government accountability motivates governments to deliver more bene�ts to citizens with greater media access (Besley and Burgess, 2002). Citizens access to information about local council efforts can impact their perceptions of councillors either positively or negatively depending on the extent to which the news is critical of government. A large literature argues and using experimental and survey evidence, shows that indi- viduals tend to view their co-ethnics as more trustworthy (Barr, 2004; Eifert, Miguel and N, 2010; Fearon and Laitin, 1996; Gambetta, 1988; Habyarimana et al., 2007, 2009; Horowitz, 1985). Using experimental techniques, Habyarimana et al. (2009) reviewed and tested three possible mechanisms that might explain why individuals are more likely to trust their co- ethnics. Ethnicity is also likely to affect the quality of service provision. When politicians can make credible pre-electoral commitments to only a narrow section of the electorate, such as their co-ethnics, governments tend to prefer to pursue clientelistic policies resulting in lower public goods provision (Keefer and Vlaicu, 2008). In addition to these three alternative hypotheses, we also examine whether beliefs about councillors’ trustworthiness are associ- ated with a household’s past exposure to war violence, their participation in community associations and their demographic characteristics. 3 Case-Selection: Sierra Leone For several reasons, the case of post-conflict Sierra Leone provides a justi�able context to test the antecedents of trustworthy government. The protracted nature of the decentralization process has yielded sub-national variation on several key dimensions of interest. In 2004, an ambitious program of decentralization reforms was launched by the national government with signi�cant support from the donor community. Since then, two rounds of nation-wide local elections were completed, while a simple system of inter-governmental �scal transfer was set up enabling local councils to carry out the service delivery functions devolved to them and to meet their administrative costs (Larizza and Glynn, 2011). However, this process of organizational reconstructing is occurring unevenly across time and across sub-national units. As a result, there is signi�cant variation in our key independent variables including the administrative capacity of councils and the ability of councils to delivery services. We exploit the sub-national variation and test our hypothesis using a particularly rich set of survey data that has been collected in Sierra Leone to monitor the implementation of the decentralization reform program. 7 The choice of Sierra Leone as a case-study is also motivated by historical and policy con- siderations. Sierra Leone’s post-colonial history offers a compelling and dramatic example of the vicious links between poor government performance, popular distrust toward state au- thorities and violence. Over time, the looting of state resources by predatory elites combined with the institutional weakness of the state apparatus contributed to the collapse of public services. These circumstances quickly catalyzed armed groups’ recruitment efforts, leading to escalation of violence and civil war (Humphreys and Weinstein, 2008; Kpundeh, 1995; Reno, 1995; Richards, 1996). Understanding whether the Government of Sierra Leone’s and donors’ focus on strengthening local service provision has contributed to overcoming long- lasting patterns of distrust by driving positive changes in citizens’ beliefs toward political authority is a critical policy input. Focusing on Sierra Leone can contribute to informing ongoing debates on the best available strategies to support sub-national governments and maximize resource allocations in post-conflict states. Given the large amount of government and donor resources channeled to sub-national governments, the policy relevance of such an analysis is important. Drawing on these data, we now turn to empirically test the ef- fects of local government performance on citizens’ beliefs about the trustworthiness of local government authorities. 4 Data, Measures and Methods 4.1 Data To test our hypotheses and control for alternative explanations, we rely on four sources of data (i) the 2005, 2007 and 2008 National Public Service (NPS) household surveys and the 2007 and 2008 community surveys; (ii) the 2004 Sierra Leone Census of population and Households; (iii) district level data on �scal transfers from 2006 and 2007 (disaggregated by sector); and (iv) the 2006 and 2008 Comprehensive Local Government Performance Assess- ment System (CLoGPAS). The NPS surveys are designed to be nationally representative. The country was strati�ed by local council areas. Enumeration areas (EA) were then selected in each local council according to the size of the population living in that local council area, with over-sampling of small local council areas.1 An EA is roughly equivalent in size to a village in rural areas (80 to 100 households) and a neighborhood in urban areas (100 to 120 households). Within each enumeration area, ten households were selected from the 2004 census books and surveyed. Surveyors interviewed an adult member of the household, usually the head or spouse, alternating between interviewing male and female respondents.2 In each of the three rounds, 2005, 2007 and 2008, about 6,300 household members were interviewed. The three 1 In rural areas, an EA is about equal to a medium sized village or a small village and surrounding hamlets. There are about 9,600 EAs in Sierra Leone, with an average population of 483. Sixty-three percent of rural EAs contain only one locality, or village, and 90 percent consist of three localities or less (Glennerster, Miguel and Rothenberg, 2010, pp. 6-7). 2 In 2008, the �nal sample was 47 percent female and 53 percent male. The 2007 survey shows equivalent �gures. 8 rounds of surveys covered the same enumeration areas and interviewed the same respondent or household, if possible.3 Community-level questionnaires were also conducted in 2007 and 2008 NPS.4 In our analysis, we aggregate the community-level data to the enumeration area, which we then merged with the NPS household-level data at the enumeration area level. Unfortunately, the questions on the 2007 and 2008 surveys only include a limited number of questions that were included in the 2005 surveys. As a result, the emphasis of this paper is on the analysis of the 2007 and 2008 rounds of survey data, which includes a richer set of measures of service delivery, administrative competence and honesty.5 4.2 Measures 4.2.1 Dependent Variable There is a large literature that models generalized trust, or how much citizens trust the government. Survey questions that probe respondents about their trust in government are problematic for several reasons. First, there is an issue of the extent to which the same questions capture comparable attitudes in different segments of the population or across time. Second, trust is usually quite domain-speci�c. As Cook, Hardin and Levi (2005, p. 164) argue, trust varies with the task, context and people involved. Consequently, we do not know the domain which respondents have in mind when answering questions about trust. If a respondent answers that they trust a president, we do not know with what tasks the respondent trusts the president. It is highly unlikely that an individual would trust a president to handle all tasks in every possible context. To deal with this problem, we model beliefs about the trustworthiness of local government officials rather than a vague notion of trust in government as an institution. Speci�cally, respondents were asked: “In your opinion, do you believe local councillors or do you have to be careful dealing with them?� By capturing interpersonal trust of speci�c government officials, these indicators are certainly preferably to generalized trust questions.6 Also, the longitudinal nature of our dataset provides us with the ability to capture comparable beliefs 3 From 2005 to 2007, 68 percent percent of the respondents from 2005 were re-interviewed, and in another 20 percent of cases, a different person from the same household was interviewed. From 2007 to 2008, 64 percent of the respondents from 2007 were re-interviewed, and in another 19 percent of cases, a different person from the same household was interviewed. In other cases, the household living at the same address was interviewed or a new household was selected from the census books (IRCBP Evaluation Unit, 2010). 4 One or two community modules were completed in each enumeration area, depending on whether the area consisted of multiple small villages (IRCBP Evaluation Unit, 2010). 5 For the next version of the paper, we plan to re-estimate the models using Amelia to impute missing data. For this version, we omitted the responses on our dependent variable that were either missing or coded as “don’t know�, “not enough info� or “neutral�. Only 6.48 percent of the responses fall into one of these three categories. We model responses on our independent variables that our coded as “don’t know� or “missing� (Honaker, King and Blackwell, November 4, 2010) 6 Despite its advantages, our indicator still suffers from the problem that it is not domain speci�c. While methodologically stronger, a domain-speci�c question would have created additional complications in ad- ministering the surveys and translating the questionnaires into local languages. 9 in the same segments of the population across time. The maps7 in Figure 1 show the mean percentage of respondents for each of Sierra Leone’s 165 chiefdoms who view local government councillors as trustworthy. The number of chiefdoms in which the average percentage of respondents who view local government councillors as trustworthy grew from three in 2007 to twelve in 2008. 4.2.2 Explanatory Variables Local Government Service Provision: Health and Education The 2004 Local Government Act devolved to the 19 local councils (LC) a range of sectoral �scal and administrative functions and responsibilities related to basic service delivery. While the LGA established a formal deadline (2008) for the devolution of functional responsibilities to LCs, in practice, the devolution has been slower than expected, with varying degrees of responsiveness from central Ministries and government agencies. As of 2011, 34 out of 80 functions remain to be devolved in accordance with the law (Larizza and Glynn, 2011, pp. 14-15). In the case of health care and the registration of births and deaths, the functions have been clearly devolved to LCs since 2005. Community leaders and household respondents were asked a set of questions about the quality of health care. The community survey includes a question about whether the health clinic in the community has a functioning refrigerator, which is a proxy for the quality of health infrastructure. As the maps in Figure 2 show, the percentage of health clinics with a functioning refrigerator grew between 2007 and 2008. We include an indicator of the type of health facility in the communities — whether the health facility is run by the government, a private group or a mission or NGO. We include a measure of the level of access to the health clinic or hospital; the survey asked community leaders if individuals can get treatment at the clinic or hospital even if they only have a little money. We also include an individual-level measure of the quality of health care. Respondents were asked: In general, are you satis�ed with 1) the clinic/hospital building and equipment, 2) the staff at the clinic/hospital and 3) with the care you receive at the clinic/hospital? We summed the number of items to which the the respondent answered “yes� and created an index that captures satisfaction with health care delivery. In 2005, the registration of births and deaths was devolved to the local council. If anyone in a respondent’s household had given birth in the past year, the interviewer asked whether the newborn received a birth certi�cate. If anyone in a respondent’s household passed away within the past two years, the respondent was asked if the household received a death certi�cate. 7 Our Chiefdom shape�les rely on Chiefdom boundaries established for the 2004 Census of Household and Population. In 2006, the Electoral Commission reviewed the division of Sierra Leone into constituencies, in accordance with the article 38 of 1991 Constitution which states that such review has to take place at intervals of not less than �ve and not more than seven years (Government of Sierra Leone, 1991). We believe that the Chiefdom boundaries might have been modi�ed as result of this review, and Chiefdom numbers reduced to 149. We are in the process of validating this hypothesis by checking official Parliamentary Acts and getting access to updated Chiefdom shape�les. 10 Although education was, in principle, supposed to be devolved to local councils in 2005, the actual devolution of this sectoral function has been delayed due to signi�cant resistance from the Ministry of Education. The �rst direct transfers to local council accounts for education were not released until 2007 (Casey, 2009). Given this delay, we expect the association between the quality of education and local councillors’ trustworthiness to be weaker than correlation between the quality of health care and councillors’ trustworthiness. We also expect the effect of education on councillors’ trustworthiness to be stronger in 2008 than in 2007. The community survey asked respondents whether there was a primary school in their community. If there was a primary school, respondents were asked who runs the school: the government; an NGO or mission; the community; or a private group. They were then asked a series of questions about the quality of the school’s infrastructure including the type of roof over the school and whether there is a latrine at the school. Respondents were also asked several questions about access to education and perceptions of its quality. If respondents have a child that goes to primary school, they were asked whether the school provides textbooks for free. The survey probed respondents about their level of satisfaction with the school: In general, are you satis�ed with 1) the school building and materials; 2) the teacher; and, 3) what your children are learning in school? These questions were coded either “yes� or “no�. We summed the number of items to which the the respondent answered “yes� and created an index that captures satisfaction with schools. We also include a measure of how long it takes for respondents’ children to travel to school. Local Councils’ Honesty, Competence and Capacity Respondents were asked the following hypothetical question that taps their perceptions of the honesty of local government officials: a) If the local council “was given 500 million Leones to complete a project in this area, do you believe they would spend all the money doing a good job on the project or would they cut some of the money?� The 2006 and 2008 Comprehensive Local Government Performance Assessment System (CLoGPAS), which was carried out across the 19 Local Districts, includes measures of coun- cils’ administrative capacity (Decentralization Secretariat, 2007). CloGPAS includes a range of questions measuring local government capacity to meet seven ‘minimum conditions’ (MCs) capturing major functional and thematic areas. More precisely, the MCs examine functional capacities of the local councils in terms of their preparedness to take over devolved functions and deliver services at the acceptable standards and assess/review the compliance of local councils with existing laws and regulations that guide the decentralization process. In this paper, we use the CloGPAS aggregate score for MCs. As additional measure of government (�scal) capacity, we also include the total amount of central government sectoral transfers to each local council for 2006 and 2007. 11 4.3 Control Variables Financial Resources (Sectoral Grants) In Sierra Leone, the grant transfer system provides largely for tied sectoral grants where the vertical envelope is �xed for each year. This implies that the quality of decentralized service delivery is likely to be, in large part, a function of the amount of funding the council receives from the central government for performing its mandated functions. For instance, as early as 2006, one quarter of the national health budget was transferred to local councils (Zhou, 2009). Accordingly, we control for the log of the amount of sectoral transfers to district councils for 2006 and 2007. We include the following itemized transfers: education, birth and death registrations, and health. We use official data on �scal transfers provided by the Local Government Financial Department, Ministry of Finance and Economic Development, Sierra Leone (Local Government Finance Department (LGFD), Ministry of Finance, 2011). Socio-Demographic Variables We control for standard socio-demographic variables that can affect citizens’ evaluations of government including age, gender, education and religion. We include several reasonably good proxies for income, including whether any of member of the household owns the follow- ing items: a bicycle, a mobile phone, a generator, a radio or cassette player, and, a television, and a car, truck or motorcycle. We also include an indicator of whether respondents or a member of their household had malaria within the past year. We include an indicator of the type of material of the roof, walls and floors of respondents’ current residence. We also control for the size of household and marriage status, since these are demographic factors that affect household resources. Living in a remote and marginalized area might affect citizens’ access to public services, as well as the quality and frequency of their interaction with local authorities, and, conse- quently, their beliefs toward these authorities. Thus, we also control for the remoteness of respondents’ residences. We derive this measure from the community-level NPS question- naire. Community leaders were asked about the distance in miles between their locality and Freetown, as well between their locality and the district headquarters. Finally, we control for the population density of enumeration areas, as large population might pose additional pressures on Local Councils.8 Local Government Outreach Activities and Access to Information Unfortunately, the NPS did not ask respondents about the quality of their interaction with councillors. The NPS asked respondents whether they have talked with a local councillor or participated in a meeting organized by the council within the past year. We control for 8 We also calculated the distance between the center of enumeration areas and Freetown. This variable was not signi�cant and did not affect model estimates. We are also in the process of calculating the distance between the center of enumeration areas and district headquarters. We will include this variable in the model, which we will estimate for the next version of this paper. 12 whether respondents are able to correctly name their local councilor, as additional proxy of the frequency of constituents’ interaction with a local councillor. We also include an indicator of whether respondents are aware of any “projects (anywhere in the district / city) the local council is doing or has done?� We also include an indicator of the number of times within the past week that the respondent listened to the radio. We believe this is a good proxy to measure information availability given qualitative evidence that radio is the premier tool for communities to obtain access to information about their local council’s development plans (Larizza and Glynn, 2011; Zhou, 2009). Consequences of the War In a recent study on Sierra Leone, Bellows and Miguel (2009) �nd that individuals whose households directly experienced more intense war violence are more likely to attend commu- nity meetings, are likely to join local political and community groups, and are more likely to vote. This study suggests that war has had positive long-lasting effects on citizens’ participa- tion in collective action. We control for community’s exposure to war violence. Respondents were asked the following three questions about their experiences with war: (1) “Were you or any member of your household injured or maimed during the conflict? (2) Was your house (or part of your compound) ever burned down during the conflict? (3) Did anyone in your household flee the community because of the conflict?� Following Bellows and Miguel (2009), we created an index from these three variables and then aggregated the data to the enumeration area. As shown in Figure 4 our measure is reliable and consistent with official historical accounts of the geography of civil war in Sierra Leone (Truth and Reconciliation Commission 2004). As indicated, the effects of the war were concentrated in the eastern and southern districts, where a higher percentage of households were made refugees or had houses that burned down during the conflict. Over time, however, the war spread across the country, forcing people to abandon their communities (the Western peninsula was the only area not affected by this massive internal displacement). During the war, the Revolutionary United Front (RUF) successfully targeted traditional authorities (Paramount Chiefs) as well as members of their households (Bellows and Miguel, 2009; Keen, 2005; Richards, 1996). Bellows and Miguel (2009) �nd that traditional authority households are statistically more likely to have experienced violence during the war. Thus, we include a control for whether respondents are a member of a ‘ruling house’. NGO Service Provision Although NGOs provide a large portion of essential goods and services in Sierra Leone, there is no comprehensive data of the locations of NGO and non-state actor projects. The only measure of non-state actor service provision that is available is a question on the NPS that asks community leaders whether an “NGO (not government, Local Council, or MP) has done any projects since the end of the war (Feb 2002) that this community bene�ts from?� We aggregate the number of NGO projects that the community leaders believe have taken place to the enumeration area. Albeit imperfect, the inclusion of this measure allows us to 13 partially control for the possibility that NGO activity might affect citizens’ beliefs toward local authorities. Social Capital Membership in community-level associations, such as religious organizations, unions and farmers’ groups, may affect individuals’ judgements about local councillors’ trustworthiness. Through interacting in voluntary associations, members share information that they can use when evaluating governments (Theiss-Morse and Hibbing, 2005). Others argue that there may be a spill over effect from learning to trust other members in a voluntary associations to learning to trust others in different contexts (Nannestad, 2008, pp. 422-423). Although the theoretical and empirical evidence linking membership in voluntary associations to trust in government is mixed, we control for individuals’ membership in a political party, religious organization and a credit or savings group (i.e. a cooperative).9 Ethnicity and Political Identities Ethnicity in Sierra Leone traditionally plays an important role in shaping citizens’ political identities and their attitudes toward government authorities. At a glance, Sierra Leone looks like a highly fragmented and heterogeneous country. In Easterly and Levine’s list of the most ethno-linguistically fractionalized countries in the world, Sierra Leone ranks �fteenth. This picture is misleading as the three largest ethnic groups (Mende, Temne, and Limba) command roughly 70 percent of the population, while none of the other 10 ethnic groups exceed a 10 percent share of the population (Casey, 2010). Moreover, the distribution of the major ethnic groups is not random and instead highly conterminous with regions. Figure 6 shows the distribution of Sierra Leone’s four largest ethnic groups across enumeration areas. Since independence from British rule, the regional-base distribution provided a strong incentive for political parties to mobilize their constituencies along ethnic lines. These patterns remained resilient in the post-conflict environment with the APC party strongly associated with the Temne and Limba groups in the north and the west, while the SLPP is strongly associated with the Mende, Kissi and Sherbro groups in the south and east (Kandeh, 1992). This rigid ethnic make-up of the two main political parties implies that citizens’ views about local councillors’ trustworthiness might be driven by their ethnic membership and/or political allegiances regardless of the quality of services they receive. Taking into account the salience of ethnic politics in Sierra Leone, we used the NPS self-reported data on ethnicity to identify whether the respondent belongs to the majority ethnic group. Following Glennerster, Miguel and Rothenberg (2010, 6-7), we then use the 2004 Census data to calculate the share of the total population size of an enumeration area that belongs to various ethnic groups. Using these shares, we calculate the standard ethno-linguistic fractionalization index (ELF). Figure 7 also shows that the level of ethnic 9 We did not create an index out of these variables since factor analysis suggests that the questions about involvement in various organizations tap different behavioral concepts. 14 fractionalization across enumeration areas is quite low, with greater fractionalization in the middle areas of the country. To control for respondents’ political identity, we use data from the 2008 NPS survey, which includes a question about respondents’ self reported voting behavior for the national elections. Speci�cally, respondents were asked which party they voted for in the 2007 national elections. The NPS sample is limited to respondents who could verify their claim of voting by producing a voter identi�cation card with the corresponding hole punches made by polling station staff (IRCBP Evaluation Unit, 2010). We code respondents’ political identity as either a supporter of the ruling party, the APC, or as a supporter of an opposition party.10 4.4 Methods The NPS data is multilevel and includes repeated measurements of household and enumer- ation area characteristics. Households are also nested within enumeration areas, which are nested within wards.11 We estimate hierarchical, logistic longitudinal models using the lmer function in R (see Equation 1). Ignoring the multilevel structure of our data can generate a number of statistical problems (Gelman and Hill, 2007; Guo and Zhao, 2000). In addition to correcting for biases in parameter estimates and standard errors, multilevel models offer two additional advantages. First, they also allow us to examine how covariates measured at the enumeration and ward levels influence our outcome variable. For example, we include EA-level measures of distance and war violence. Second, this type of model allows us to test whether slopes are random, e.g., the effect of individual-level measures on our dependent variable differs across enumeration areas (Guo and Zhao, 2000, p. 444). We test, for exam- ple, whether the effect of ethnicity on trust in local government officials varies across EAs. Taking into account the multilevel nature of our data, we estimate random intercepts for the households and enumeration areas and treat our dependent variable as a binary outcome. 5 Results In this section, we present the results of our model on perceived trustworthiness of local government officials. We discuss regression results of trust in local government officials using predicted probabilities and �rst differences (King, 1989). In a multilevel logit equation, 10 We are not completely satis�ed with this control. According to Casey (2010), in Sierra Leone voters tend to cross the party lines for local elections. We therefore believe that a variable indicating whether respondents voted for the winning party/candidate in the 2004 and/or 2008 elections will be more relevant for our purposes. We are in the process of collecting this information, and will include this variable in the next version of the paper. 11 A complication in our data is that 80 percent of wards were redistricted between the 2007 and 2008 rounds of the NPS. There does not appear to be any systematic differences between wards that were re-districted and wards that were not. Nonetheless, we estimated separate models using the 2008 ward boundaries and the 2004 boundaries. Unfortunately, we had to re-create the latter boundaries from other data sources and we could not determine the boundaries for 23 percent of the wards. Because the results were similar for both estimation strategies, we focus our discussion on the results that use the 2008 ward boundaries. 15 ˆ �rst differences are estimates of how much π would change given a particular change in an explanatory variable, while holding the other variables in the equation constant. We hold interval-scaled covariates at their mean levels and categorical variables at their modal levels. We present predicted probabilities and �rst differences with con�dence intervals reflecting the estimation and fundamental uncertainty of the point predictions including the random errors. We graphically present the �rst differences and the uncertainty around these estimates in coefficient plots. In these plots, the �rst differences are reflected as dark circles and the horizontal lines represent the simulated 95% con�dence intervals. 5.1 Model 1: 2007 and 2008 We now turn to the discussion of the results of Model 1, which pools data from 2007 and 2008 (see Table 1).12 Table 2 and Figure 9 show the effect of a change in our explanatory variables from low to high on views about the trustworthiness of local government officials for 2007, 2008 and the change from 2007 to 2008, respectively. To summarize, the quality of decentralized service delivery, especially health care and the registration of deaths, is associated with beliefs about the trustworthiness of local councillors over time. The impact of councillors’ honesty on evaluations of councillors as trustworthy is greatest for respondents who live the farthest away from their district headquarters and in the least populated areas. Moreover, the greater the knowledge of local council projects, the more likely citizens are to view them as trustworthy. Individuals who have interacted with their councillor are more likely to view their councillor as trustworthy. The same is true for respondents, who know the name of their local councillor. We also �nd that the presence of service-delivery NGOs is positively correlated with trustworthy councillors. There is no evidence that councils’ administrative competence is correlated with citizens’ beliefs about councillors’ trustworthiness. Finally, consistent with our expectations, we �nd that the more communities were impacted by violence during the civil war, the more likely community members are to view their elected local government councillors as trustworthy. 5.1.1 Explanatory Variables We now discuss these empirical relationships in turn. Perceptions of the quality of decen- tralized service delivery are positively associated with trustworthy local government officials over time. The higher the quality of health care, the more likely individuals are to view local government officials as trustworthy. Compared to respondents who live in a community with a health clinic that is run by an NGO or a mission, respondents who live in a community with a government-run health facility are more likely to view local government councillors as trustworthy. In 2007 and in 2008, the presence of a government health facility or hospital corresponds to a percentage point increase of 6 and 7 percentage points, respectively, in the 12 We only include the regression results for two model speci�cations: the �rst includes an indicator of satisfaction with health care delivery and local government honesty; the second includes an indicator of satisfaction with education and whether the respondent interacted with a local councillor over the past year. Other regression results are available upon request. 16 predicted probability that community members will view their councillors as trustworthy. There are several potential explanations for this �nding. Dependence on non-state actors for health care may prompt citizens to wonder whether their councillors are trustworthy and are trying to serve their best interests. Likewise, the provision of health care by local governments may strengthen citizens’ con�dence that their local councillors are doing their best to try and improve citizens’ welfare. Alternatively, the quality of health care provision in Sierra Leone may be higher in government clinics than in NGO and mission clinics and citizens may be crediting councillors for the former and blaming councillors for the latter. The NPS does not include questions that would allow us to directly test who citizens credit or blame for public goods provision and whether citizens hold government accountable for non-state service provision.13 Our objective indicator of the quality of health clinic infrastructure, the presence of a functioning refrigerator in a community’s health clinic, is positive and signi�cant and the effect size of this variable is slightly stronger in 2008 than in 2007. Once we include this indicator in the model, the variable, the amount of health transfers to local governments, is no longer signi�cant. Living in a community that has a health clinic with a functioning refrigerator corresponds to a percentage point increase of 6 and 8 in the predicted probability of viewing councillors as trustworthy in 2007 and in 2008, respectively. Although itemized local government budgets (within-sector allocations) are not available to cross-validate this �nding, we interpret these results to suggest that local governments are spending a part of their health budgets on infrastructure and these visible improvements are enhancing citizens’ beliefs about local authorities’ trustworthiness. Our subjective measure of the quality of health care is positively correlated with our dependent variable and is slightly stronger in 2008 than in 2007. The difference in the predicted probability of viewing councillors as trustworthy in 2007 and 2008 is 9.0 and 12.0 percentage points, respectively, higher for respondents who are satis�ed with health care than respondents who are dissatis�ed with health care. We do not �nd any association between a health clinic’s provision of free or low cost treatment and beliefs about councillors’ trustworthiness. Compared to other services, the registration of births and deaths is the most decentralized sector. Even in Kono district, where the local council worked with UNICEF to register 45,000 children (Foster and Glennester, 2009, p. 81), there is no association between a respondent’s experience with registering births and their beliefs about local councillors’ trustworthiness. By comparison, the association between experience with registering a death and a perception of local councillors as trustworthy is positive and signi�cant. In both 2007 and 2008, the difference in the predicted probability of viewing councillors as trustworthy between someone who does not have any experience with registering a family member’s death and someone who has that experience is 7 percentage points.14 13 Survey evidence from Zambia (Sacks 2011), for example, suggests that when dissatis�ed with donor and non-state actor service provision, citizens tend to attribute these services to the government. By comparison, when satis�ed with government services, citizens tend to attribute them to donors and non-state actors. 14 Foster and Glennester (2009, p. 80) report that the least isolated communities saw the greatest improve- ments in the registration of deaths from 2005 to 2008. We test whether the effect of having registered a family member’s death on one’s belief about local councillors’ trustworthiness is dependent on one’s geographical 17 As education has only been very recently devolved, we expected to �nd a weak but positive association between the quality of education and beliefs about councillors’ trust- worthiness. Our results provide evidence for this hypothesis. Although we also expected the effect size of our indicators of educational quality to be larger in 2008 than in 2007, our results do not suggest that this is the case. Respondents whose children attend a pri- mary school run by a government are only slightly more likely to view their councillor as trustworthy than respondents whose children attend a private, NGO or mission school. Whether schools provide free textbooks is positively but weakly correlated with eval- uations of councillors as trustworthy. In both 2007 and 2008, a school’s provision of free textbooks corresponds to a percentage point increase of 3.0 in the predicted probability that a respondent will view their councillors as trustworthy. Our other two objective indicators of the quality of education and its infrastructure (the type of roof over a school) are not sig- ni�cant. Our subjective indicator of the quality of education is signi�cant and the direction of the association with our dependent variable is positive. The difference in the predicted probability of viewing councillors as trustworthy in 2007 is 7.0 percentage points higher for respondents who are satis�ed with education than respondents who are dissatis�ed with education; this �rst difference is 9.0 percentage points in 2008. Turning to the other dimensions of local government performance, perceptions of the honesty of local government councillors are positively associated with trustworthy councillors and the size of this effect is larger in 2008 than in 2007. In 2007 and 2008, a view that local councillors are likely to spend revenue effectively rather than use the funds for private gain is associated with a percentage point increase of 26.0 and 31.0, respectively, in the predicted probability of viewing local councillors as trustworthy. This is largest �rst difference both in absolute and relative terms in our model. Including an interaction between respondents’ perceptions of their local councillor’s hon- esty and their geographic remoteness improves their model �t (see Figure 11). For those respondents who live in their district’s headquarters, a shift from a belief in 2007 that coun- cillors are dishonest to a belief in 2008 that they are honest corresponds to a percentage point increase of 38.6 in the predicted probability of viewing councillors as trustworthy. For those respondents who live 37 kilometers from their district headquarters, the �rst difference is 45.0 percentage points. Including an interaction term for the population density of enu- meration areas and beliefs about local councillors’ honesty also improves the model �t. The direction of the interaction term is negative and signi�cant. The effect of local councillors’ honesty on beliefs about their trustworthiness decreases with the population density of the community. This might suggests that highly populated areas might pose additional pressures to political authorities and require extra-efforts on their side to improve citizens’ attitudes toward them. isolation. The interaction variable is positive but only slightly signi�cant. 18 5.1.2 Alternative Explanations Table 3 show the effect of a change in our control variables from low to high on views about the trustworthiness of local government officials for 2007 and 2008 and the change from 2007 to 2008, respectively. We control for NGO service provision as NGOs are likely to provide a substantial portion of services in Sierra Leone. Our results provide little support for the hypothesis that NGO service provision is likely to undermine citizens’ beliefs about coun- cillors’ trustworthiness. In fact, we �nd that citizens who live in neighborhoods with active NGOs are more likely to believe councillors are trustworthy. Speci�cally, the difference in the predicted probability of trusting local government officials in 2007 between an individual who lives in a neighborhood with few, if any, NGOs and someone who lives in a context with several active NGOs is 3.0 percentage points. The �rst difference is similar for 2008. Interpreting this result is somewhat problematic as we do not know if citizens are crediting their local council for NGO services or if citizens are crediting their local council for helping to leverage NGO resources. We expect that in communities, where NGO branding is less visible or where councillors’ invest heavily in claiming credit for NGO goods, the former explanation is more plausible. In communities, where NGOs heavily brand their services, the latter explanation is more likely. Consistent with our expectations, respondents who live in areas that were particularly affected by the war are more likely to view their local government councillors as trustworthy. Qualitative accounts from recent focus groups discussions carried out in war-affected districts found that the most heavily damaged and isolated districts had given rise to greater beliefs in social cohesion and community action. The collective memory of violence and brutal devastation left by the war can act as a mobilizing force at the local level, increasing citizens’ engagement in setting development priorities and their willingness to promote consensual, rather than confrontational, social attitudes (Larizza and Glynn, 2011). These results provide additional support to a growing body of �eld research which consistently shows that exposure to violence and civil war may increase social capital. For example, Bellows and Miguel (2009, pp. 17-19) �nd that citizens in communities more severely affected by the conflict in Sierra Leone manifested greater level of local political mobilization and greater capacity for collective action. Likewise, Voors et al. (2010) show that people exposed to wartime violence in Burundi were more altruistic than those who were not. Finally, a recent study on Nepal (Gilligan, Pasquale and Samii, 2011) �nds that members of communities with greater exposure to violence during Nepal’s civil war exhibit signi�cantly greater levels of social capital, measured by subjects’ willingness to invest in trust-based transactions and contribute to a collective good. Each of our indicators of councillors’ outreach efforts to constituents are positive and signi�cant. The difference in the predicted probability of viewing councillors as trustworthy in 2008 is 6.0 percentage points higher for respondents who are aware of councillors’ projects compared to respondents who are unaware of any council projects. Individuals who have interacted with a local councillor within the past year are also more likely to view their councillors as trustworthy. The predicted probability of viewing councillors as trustworthy is 6.0 and 7.0 percentage points higher for a respondent who interacted with their councillor 19 in 2007 and in 2008, respectively. We also �nd that the more citizens listen to the radio, the greater their likelihood of trusting their local government councillors. In contrast, higher attendance to the ward committees does not seem to be associated with changes in citizens beliefs. All together, these results speak of the positive effects played by councillors’ outreach efforts in shaping the attitudes of ordinary citizens toward state authorities. They are also consistent with recent qualitative research pointing at the positive impact of outreach radio campaigns in aiding more effective communication and physical interactions with political authorities, eventually generating a stronger sense of ownership of local development process and inclusion among the local communities (Larizza and Glynn, 2011).15 Membership in community associations — a political party, cooperative or a religious group — is positively correlated with views about the trustworthiness of local councillors. The predicted probability of viewing the local council as trustworthy is 9.0, 5.0 and 3.0 percentage points higher for respondents who in 2008 belonged to a political party, a cooper- ative or a religious group, respectively. This is consistent with previous research pointing at a growing level of civic activism at the grassroots level in post-conflict Sierra Leone (Zhou, 2009, pp. 97-99), and suggest that citizens might take advantage of their membership of in- terests groups to collectively mobilize and become active in demanding service improvements and government responsiveness to their needs. In contrast to our expectations, very few of our socio-demographic measures are sig- ni�cantly associated with trustworthy local government councillors. We do not �nd any correlation between our ethnic fractionalization measure and views about the trustworthi- ness of local councillors. We also do not �nd any relationship between whether respondents belong to a minority ethnic group and views about the trustworthiness of local councillors. There is no evidence that respondents who are members of a ruling family are more likely to see local government councillors as trustworthy. Further, there is no association between whether a respondent supports the ruling party, the APC, and their beliefs about local coun- cillors’ trustworthiness. All together, these �nding directly challenges historical accounts of ethnicity and party affiliation as the main driver of citizens socializing practices and be- liefs (Kandeh, 1992; Reno, 1995). They are also consistent with recent analyses suggesting that ethnic and partisan politics might be less relevant for local elections, where develop- ment issues seem to play a more prominent role in shaping the linkages between voters and politicians (Srivastava and Larizza, 2011; Zhou, 2009).16 15 Ward committees are the key formal mechanism (as mandated by the law) to channel citizens views to the councils. It is also the forum to communicate development plans at the village level. In this respect the Ward Committees represents an essential channel for upward and downward communication in Sierra Leone. However, evidence from �eldwork research and focus group discussions consistently found that the frequency of ward committees meetings is usually hampered by mobility constraints across scattered communities as well as by lack of funds. This might explain the low response rates in the NPS, as well as the the absence of an association between indicators of ward committee outreach and beliefs about the perceived trustworthiness of councillors. 16 For example, a household survey carried out by the Institutional Reform and Capacity Building Project (IRCBP) in 2009 (Bonthe and Bombali districts) asked people their main reasons for voting. The survey found that in national elections participants ranked the political party as the most important and develop- ment as a second priority; for local elections the promise of development was the priority (Zhou, 2009). 20 5.2 Robustness Checks We now present the results of several robustness checks. First, we present separation plots as a visual method for assessing the predictive power of the regression results (Greenhill, Ward and Sacks, 2011). The plots allow us to evaluate model �t based upon the models’ ability to assign high-probabilities to actual occurrences of the event of interest (e.g. a perception of local government councillors as trustworthy) and low-probabilities to non-occurrences of the event of interest (e.g. a perception of local government councillors as untrustworthy). The plots separate the events and non-events into two separate plots. In these plots, the dark and light panels correspond to the actual instances of the events and non-events, respectively. The colors correspond to ranges of probabilities; the darkest shade of red corresponds to a probability of 0.9 and higher and the lightest shade of red corresponds to a probability of 0.1 and lower. A perfect model would produce a plot with complete separation between the events and non-events. All the events would be in the upper band of the plot, which would be represented in the darkest shades of red. All the non-events would be in the lower band of the plot, which would be represented in the lightest shades of red. The separation plots presented in Figure 12 suggest that the model performs relatively well in assigning higher probabilities to actual events and lower probabilities to non-events. Second, we check our assumption that citizens in Sierra Leone are correctly attributing health and education services to their local governments. This assumption is somewhat problematic in Sierra Leone, where a diversity of actors including churches, NGOs, traditional leaders are providing services, and where local councils ‘de-facto’ share the responsibility of providing services with other actors including the central government (line ministries) and NGOs. Although the devolution of functions is clear, the hiring of personnel has not yet been devolved. The lack of devolution of personnel could affect local government capacity to perform their devolved functions and might also create some confusion in the accountability relationship between citizens and service providers.17 We test whether satisfaction with health care and education are associated with the trust- worthiness of central government officials. If we �nd an equally strong association between satisfaction with health care and education and trust in central government officials, then our assumption that citizens are correctly attributing service delivery to local government officials is problematic. Therefore, we estimate the same model described in Section 5.1 using trustworthiness of central government officials as the dependent variable. Consistent with our expectations, we do not �nd any positive associations between satisfaction with health care or education and trustworthy central government officials.18 17 The new Decentralization Policy foresees direct employment and control of all locally based personnel devolved to the Councils by 2016. 18 The results show that compared to respondents who are extremely dissatis�ed with health care, respon- dents who are somewhat dissatis�ed with health care are less likely to view their central government officials as trustworthy. This �nding is puzzling and we are considering potential explanations. Another more strin- gent test would be to consider if voting patterns for the local elections are affected by the quality of services provided by local government. If our assumption regarding attribution to local government is correct, we would expect voters to be willing to reward local politicians who deliver while punishing incumbent who under-perform. We plan to explore these empirical relationship as part of our larger research project. 21 Third, we examine whether the results from Section 5.1 vary based on geographic varia- tion of war violence. We estimate separate models for respondents from Kono district and respondents from Western Area Urban, which comprises Freetown. As Sierra Leone’s largest diamond-mining region, looting and �ghting were ubiquitous throughout Kono during the war. On our measure of communities’ exposure to war violence, which ranges from one to four with four indicating the greatest exposure to war violence, the mean for Kono and Western Area Urban is 3.26 and 1.47, respectively. At the start of decentralization, Kono’s infrastructure was in a much worse state than Western Area Urban’s. For example, 69.72 of NPS respondents from Kono reported that the quality of the nearest road was “very poor (rough and unpaved)� and another 25.54 percent rated the quality of the nearest road as “poor (smooth and unpaved).� By comparison, only 29.14 percent and 30.88 percent of NPS respondents from Western Area Urban evaluated the closest road as very poor and poor, respectively. As individuals tend to reward relative, rather than absolute, improvements in the public goods they receive, we expect the effects of local government performance on be- liefs about trustworthiness to be stronger for residents of Kono than for residents of Western Area Urban. The separation plots in Figure 13 suggest that our model performs better for Kono than for the Western Area Urban district. As these plots show, the models only perform moder- ately well in assigning high probabilities to cases of trustworthiness and low probabilities to cases of distrust. The results from Kono and Western Area Urban are only partially consis- tent with our model on the full dataset (see Table 4). Whereas in Kono, the effect of having registered a family member’s death is positively and strongly associated with beliefs about local government officials’ trustworthiness, this variable is not signi�cant in the regression on Western Area Urban. Experience with registering a family member’s death in 2007 corre- sponds to a percentage point increase of 42.7 in the predicted probability that a respondent from Kono will view councillors as trustworthy. Our indicator of the quality of health clinic infrastructure, the presence of a refrigerator in the community’s health clinic, is positive and signi�cant on the Kono subset but not on the Western Area Urban subset. Compared to respondents who attend an NGO or mission clinic, respondents who attend a government clinic are less likely to view their local government representatives as trustworthy. We do not �nd any evidence of an association between the quality of education and our dependent variable in either Kono or Western Area Urban. In both Kono and Freetown, perceptions of local councillor’s honesty has a large relative effect on evaluations of councillors’ honesty. In contrast to our full model, the results from Kono and Western Area Urban point to a relationship between ethnicity and beliefs about local authorities’ trustworthiness. On the one hand, respondents from Kono, who are members of the largest ethnic group in their ward, are more likely to view their local councillor as trustworthy. On the other hand, respondents from Western Area Urban, who are members of the largest ethnic group in their ward, are less likely to view their local councillor as trustworthy. Our indicator of ELF is statistically signi�cant in the regression on Kono but not in the regression on Western Area Urban. In Kono, the higher the ethnic fractionalization of a ward, the more likely community members will view their councillor as trustworthy. In our future research, 22 we plan to examine explanations for the different effects of ethnic diversity on beliefs about government authorities within and across Sierra Leone’s districts. We will address the extent to which the relationship between ethnic diversity and attitudes towards local government authorities is mediated by the competitiveness of local elections. Finally, we did not �nd any association between membership in community associations and attitudes toward local authorities in Kono, where there has been a recent growth of youth CSOs and youth-led protests against industrial mining operations (Fanthorpe and Maconachie, 2010). 6 Policy Implications and Concluding Remarks Based the case of Sierra Leone, this paper offers an empirical contribution to the literature on citizens’ beliefs toward political authorities. It shows that local service provision plays a critical role in shaping citizens’ attitudes toward political authorities. However, our results also suggest the devolution of power per se is not a sufficient condition for building the trustworthiness of local authorities. In other words, citizens’ trust and support is not uncon- ditional. On the contrary, bureaucratic honesty combined with the quality of local service provision is what really matters to citizens. The decentralization of service provision and the role it is playing in shaping the relationships between citizens and local political authorities is a remarkable achievement for a country where the authority and the legitimacy of the state almost collapsed a decade ago. Our �ndings offer pragmatic lessons and policy implications for international develop- ment practitioners engaged in supporting local development in Sierra Leone. First, support to local service delivery should be complemented with continuous �nancial and technical assistance to improve the effective use of public resources, as this remains the single largest predictor of citizens’ beliefs about councillors’ trustworthiness. Second, efforts to improve the transparency of local government’s should prioritize the most remote and least popu- lated areas, as these are the ones where an improvement in transparency is likely to produce greater political capital for the authorities.19 Third, special efforts should be made to support and enhance local councils’ outreach efforts, including efforts to improve the radio coverage across districts. As our analysis suggests, positive interactions between citizens and local authorities are critical to strengthening citizens’ attitudes toward political authorities. A great deal of attention in policy circles and development research is focused on whether the political economy dynamics which led Sierra Leone into poverty and a prolonged civil war have re-appeared since the return of peace in 2002 (Brown et al., 2006; Hanlon, 2004; International Crisis Group Report, 2008). In this paper, we show that a different development path is possible for Sierra Leone. By doing so, we directly challenge the traditional portrait of Sierra Leone’s society as static, underdeveloped and largely patronage-driven. Instead, we 19 In Sierra Leone, this approach would be consistent with recent recommendations to revise the current formula for �scal grant allocations to take into better account equity issues and equalization principles. Under conditions of scarce resources, however, a choice of this kind is also likely to imply difficult trade-offs between political and economic considerations, as economics returns are usually higher in highly populated urban areas (Boex, 2011). 23 suggest that ordinary citizens can be quite vigilant in monitoring government performance at the local level, rewarding better performing councillors with a higher level of trustworthiness. These �ndings are consistent with focus group discussions, which suggested that the local council is contributing to greater civic engagement in policy making and enhancing citizens’ voice in local decision-making in Sierra Leone’s society, thereby providing a viable mechanism to overcome deep-rooted legacies of war (Fanthorpe, Andrew and Sesay, 2011; Larizza and Glynn, 2011; Zhou, 2009). In related studies, we consider the consequences of beliefs about trustworthy government for citizens’ quasi-voluntary compliance with state demands and regulations — participation in town cleaning and road-brushing. We are exploring whether the criteria citizens use to evaluate political authorities, such as the quality of local services, has consequences for politicians electoral strategies. Finally, future studies should test whether these �ndings are applicable across a larger sample of post-conflict and fragile states. 24 7 Tables 25 Table 1: Model 1: Multilevel Logistic Regression on Trustworthiness of Local Government Officials (2007 and 2008) (N=15,841) Speci�cation 1 Speci�cation 2 Variable Coefficient Std. Err. Coefficient Std. Err. Intercept -4.01∗∗∗ 0.59 -1.81∗∗∗ 0.36 ∗∗∗ Year 2008 0.89∗∗∗ 0.08 0.75 0.08 ∗∗∗ Listen to radio every day/past week 0.22∗∗∗ 0.06 0.28 0.06 Distance to district headquarters 0.00 0.00 0.00 0.00 Population density of enumeration area -0.03 0.02 -0.02 0.02 Ethnic minority in ward -0.05 0.05 -0.04 0.05 Ethnic fractionalization -0.19 0.19 -0.13 0.20 NGO projects in community 0.15∗ 0.07 Community exposure to war violence 0.15∗∗ 0.05 Belong to a religious group 0.11∗∗ 0.04 0.10∗ 0.04 Belong to a cooperative 0.20∗∗∗ 0.06 0.21∗∗∗ 0.05 Belong to a political party 0.37∗∗∗ 0.08 0.38∗∗∗ 0.08 Household member had malaria last year 0.13∗∗ 0.04 0.10∗ 0.04 Experience with registering deaths 0.30∗∗ 0.10 0.31∗∗ 0.10 Health clinic in community has working fridge 0.30∗∗∗ 0.07 Clinic in community: district hospital 0.10 0.13 26 Clinic in community: maternal/child post or community post 0.30∗∗ 0.11 Clinic in community: private 0.38∗ 0.15 Satis�ed with health clinics: Level 1 0.04 0.12 Satis�ed with health clinics: Level 2 0.19 0.11 Satis�ed with health clinics: Level 3 0.47∗∗∗ 0.09 School in community provides free textbooks 0.13∗ 0.06 Satis�ed with education: Level 1 0.05 0.15 Satis�ed with education: Level 2 0.23∗ 0.11 Satis�ed with education: Level 3 0.37∗∗∗ 0.11 Local councillors spending money well 1.28∗∗∗ 0.08 Identi�ed local councillor’s name 0.13∗∗ 0.05 Health transfers from central to local government 0.05 0.03 Council’s administrative capacity 0.01 0.00 0.00 0.00 Random Effects Enumeration Areas (N=623) 0.21 0.46 0.22 0.46 Wards (N=154) 0.16 0.40 0.18 0.43 Local Councils (N=19) 0.09 0.29 0.12 0.35 ∗ p < .05; ∗∗ p < .01; ∗∗∗ p < .001 Other socio-demographic variables are included in the model but the results are not shown. Table 2: This table shows the effect of a change in the following explanatory variables from low to high on evaluations of local government councillor trustworthiness. The �rst differences were simulated from three speci�cations of Model 1, which pools the 2007 and 2008 data. Year 2007 Year 2008 2007 to 2008 Variable ˆ D Lower 95% Upper 95% ˆ D Lower 95% Upper 95% ˆ D Lower 95% Upper 95% Local councillors spend money for good of community 27 (vs. for private gain or corruption) 0.26 0.22 0.30 0.31 0.28 0.34 0.48 0.44 0.53 Highly satis�ed with education (vs. highly unsatis�ed) 0.09 0.05 0.13 0.12 0.07 0.16 0.25 0.20 0.31 Highly satis�ed with health (vs. highly unsatis�ed) 0.09 0.06 0.12 0.12 0.07 0.16 0.31 0.26 0.35 Health clinic in community run by government (vs. NGO/mission clinic) 0.06 0.02 0.10 0.07 0.02 0.13 0.27 0.21 0.33 Registered death of a family member (vs. didn’t register) 0.07 0.02 0.12 0.07 0.03 0.13 0.29 0.23 0.35 Health clinic/hospital has a working fridge (vs. no fridge) 0.06 0.03 0.09 0.08 0.04 0.11 0.28 0.23 0.32 Primary school in community run by government (vs. non-state actor) 0.04 0.01 0.06 0.04 0.02 0.07 0.20 0.16 0.24 School provides free textbooks (vs. not free) 0.03 0.01 0.05 0.03 0.01 0.06 0.20 0.15 0.25 Table 3: This table shows the effect of a change in the following explanatory variables from low to high on evaluations of local government councillor trustworthiness. The �rst differences were simulated from three speci�cations of Model 1, which pools the 2007 and 2008 data. Year 2007 Year 2008 2007 to 2008 Variable Dˆ Lower 95% Upper 95% Dˆ Lower 95% Upper 95% Dˆ Lower 95% Upper 95% Community’s exposure to war violence (low to heavy) 0.11 0.03 0.19 0.14 0.04 0.24 0.33 0.24 0.42 28 Belong to a political party 0.09 0.05 0.12 0.09 0.05 0.12 0.30 0.25 0.36 Interacted with a local councillor (vs. no experience) 0.06 0.04 0.08 0.07 0.05 0.09 0.28 0.24 0.32 Aware of local councillor projects (vs. unaware) 0.05 0.03 0.07 0.06 0.03 0.09 0.26 0.22 0.30 Listen to radio every day (vs. no radio in past week) 0.05 0.02 0.08 0.05 0.02 0.08 0.26 0.22 0.31 Belong to a cooperative 0.04 0.02 0.07 0.05 0.02 0.07 0.26 0.21 0.31 Central government health transfers (min. to max.) 0.04 0.00 0.08 0.08 0.00 0.15 0.25 0.18 0.31 No. of NGO projects in community 0.03 0.00 0.06 0.04 0.00 0.07 0.25 0.20 0.29 Identi�ed councillor by name (vs. unaware of councillor’s name) 0.03 0.01 0.04 0.04 0.01 0.07 0.19 0.15 0.23 Belong to a religious group 0.02 0.01 0.04 0.03 0.01 0.04 0.24 0.19 0.28 Table 4: Model 1: Multilevel Logistic Regression on Trustworthiness of Local Government Officials for Kono and Western Area Urban (2007 and 2008) Kono Western Area Urban Variable Coefficient Std. Err. Coefficient Std. Err. Year 2008 -1.37 1.59 0.29 0.15 ∗∗ Listen to radio every day/past week 0.44 0.28 0.54 0.20 Population density of enumeration area -0.18 0.11 -0.03 0.07 Ethnic majority in ward 0.60∗∗ 0.22 -0.26∗ 0.13 Ethnic fractionalization 1.70∗ 0.79 -1.18 0.90 NGO projects in community 0.35 0.28 -0.05 0.35 Community exposure to war violence -0.43 0.29 0.38∗∗ 0.12 Belong to a religious group -0.31 0.17 0.00 0.13 29 Belong to a political party 0.20 0.29 0.62∗ 0.28 Experience with registering deaths 2.16∗∗∗ 0.56 0.53 0.84 Health clinic in community has working fridge 1.01∗∗ 0.38 -0.37 Clinic in community: district hospital -3.33∗∗ 1.04 Clinic in community: maternal/child post or community post -2.80∗∗ 0.98 Clinic in community: private -2.36 1.41 Local councillors spending money well 1.65∗∗∗ 0.35 0.88∗∗∗ 0.24 Random Effects Intercept (Kono=1,080; Western Area Urban=1,660) 0.28 6.07 -1.79 1.13 Enumeration Areas (Kono=46; Western Area Urban=97) 0.28 0.53 0.29 0.54 Wards (Kono=26; Western Area Urban=44) 0.19 0.43 0.18 0.42 ∗ p < .05; ∗∗ p < .01; ∗∗∗ p < .001 Other socio-demographic variables are included in the model but the results are not shown. 8 Figures under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (a) Perceived Trustworthiness of Councillors (2007) (b) Perceived Trustworthiness of Councillors (2008) Figure 1: These maps show the mean percentage of respondents for each of Sierra Leone’s 165 chiefdoms who view local government councillors as trustworthy. The number of chiefdoms in which the average percentage of respondents who view local government councillors as trustworthy grew from three in 2007 to twelve in 2008. Although services are delivered at the ward level, we aggregated the individual-level measures of trust to the chiefdom level. Whereas only a small percentage of wards were included in the NPS, all chiefdoms were included in the NPS except for three. 30 under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (a) Health Clinics with a Refrigerator (2007) (b) Health Clinics with a Refrigerator (2008) Figure 2: These maps show the mean percentage of communities in a chiefdom that have a health clinic with a functioning refrigerator. There is a clear increase in the percentage of communities that have a health clinic with a refrigerator from 2007 to 2008. 31 % of Respondents who Listen to Radio Every Day % of Households who are Aware of Local Council Projects under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 over 60 over 60 (a) Year 2008 (b) Year 2008 % of Respondents who have Interacted with Councillor under 20 20 − 40 40 − 60 over 60 (c) Year 2008 Figure 3: These maps show the mean percentage of respondents who listen to the radio every day, who are aware of their local council’s projects and who have interacted with their local councillor over the past year. Only a very small percentage of respondents, 15 percent, are aware of councillors’ development projects. Councillors’ outreach efforts appear to be more common in the middle of the country, which has greater ethnic fractionalization. 32 under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (a) Household Member Injured or Maimed (b) Household Member Made Refugee under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (c) House Burned Down (d) Household Member Fled Figure 4: These maps show the mean percentage of respondents for each of Sierra Leone’s 165 chiefdoms who were exposed to war violence. 33 under 10 under 10 10 − 20 10 − 20 20 − 30 20 − 30 30 − 40 30 − 40 40 − 50 40 − 50 over 50 over 50 (a) Spend Revenue Effectively (2007) (b) Spend Revenue Effectively (2008) Figure 5: These maps show the mean percentage of respondents for each of Sierra Leone’s 165 chiefdoms who believe that their councillor is likely to spend revenue effectively and for the good of the community rather than use this revenue for private gain. As these maps illustrate, between 2007 and 2008, there was a substantial increase in the percentage of respondents who have con�dence that their councillor is likely to spend revenue effectively. 34 under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (a) Temne (b) Mende under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (c) Limba (d) Kono Figure 6: These maps show the percentage of CEAs who identi�ed as Temne, Mende, Kono or Limba in Sierra Leone’s 2004 National Population and Housing Census. 35 under 0.2 0.2 − 0.4 0.4 − 0.6 0.6 − 0.8 over 0.8 Figure 7: This map shows the Ethnic Linguistic Fractionalization (ELF) index across Sierra Leone’s Census Enumeration Areas. As the map illustrates, the level of fractionalization outside of the Western Peninsula is quite low. Seventy-percent of the NPS’ respondents identify as a member of the dominant ethnic group in their ward. under 20 under 20 20 − 40 20 − 40 40 − 60 40 − 60 60 − 80 60 − 80 over 80 over 80 (a) Support for the APC (b) Support for the Opposition Figure 8: These maps show the average percentage of respondents by chiefdom who voted for the APC and the opposition in the 2007 national elections, according to the 2008 NPS survey. 36 2007 0 0.1 0.2 0.3 0.4 Councillors spend revenue effectively Highly satisfied with education Highly satisfied with health care Health clinic run by government Registered death of a family member Health clinic has a working fridge Primary school run by government School provides free textbooks (a) Year 2007 0 0.1 0.2 2008 0.3 0.4 0 0.1 0.2 0.3 0.4 Local councillors spend revenue effectively Health clinic run by government Highly satisfied with health care Highly satisfied with education Registered death of a family member Health clinic/hospital has a working fridge Primary school run by government School provides free textbooks (b) Year 2008 0 0.1 0.2 0.3 0.4 Figure 9: This �gure shows the �rst differences that are represented by dark circles and the hori- zontal lines represent the simulated 95% con�dence intervals. We calculate the effect of a change in quantities of interest from their minimum value to their maximum values (while holding all other variables at the central tendencies including year) on beliefs about the trustworthiness of local councillors. 37 Change from 2007 to 2008 0 0.1 0.2 0.3 0.4 0.5 0.6 Councillors spend revenue effectively Highly satisfied with education Highly satisfied with health care Health clinic run by government Registered death of a family member Health clinic has a working fridge Primary school run by government School provides free textbooks 0 0.1 of 0.3 0.4 0.5 0.6 Figure 10: This �gure shows the effect 0.2 a change in quantities of interest from their minimum value to their maximum values including year (a change from 2007 to 2008) on beliefs about the trustworthiness of local councillors. In each case, we hold all other variables at their central tendencies. 38 Effect of Local Councillors' Honesty and Geographical Remoteness on Beliefs about Councillors' Trustworthiness 1.0 Honest Dishonest Predicted Probability of Trusting Councillors 0.8 0.6 0.4 0.2 0.0 0 50 100 150 Distance to District HQ (in miles) Figure 11: This �gure shows that the effect of local councillors’ honesty on trustworthy local councillors varies based on respondents’ geographical remoteness. As respondents’ geographically isolation increases, the effect of their perceptions of local councillors’ honesty on theirs views about councillors’ trustworthiness also increases. By comparison, the effect of perceptions of local coun- cillors’ dishonesty on trustworthiness is higher in less geographically remote areas than in more geographically remote areas. 39 y=1 (n=6666) Probabilities: over 0.9 0.8 − 0.9 0.7 − 0.8 0.6 − 0.7 0.5 − 0.6 y=0 (n=8434) 0.4 − 0.5 0.3 − 0.4 0.2 − 0.3 0.1 − 0.2 under 0.1 (a) Model 1: Speci�cation 1 y=1 (n=6666) Probabilities: over 0.9 0.8 − 0.9 0.7 − 0.8 0.6 − 0.7 0.5 − 0.6 y=0 (n=8434) 0.4 − 0.5 0.3 − 0.4 0.2 − 0.3 0.1 − 0.2 under 0.1 (b) Model 1: Speci�cation 2 Figure 12: This plot visualizes the �t of the two speci�cations of Model 1. The width of the darker shades of red on the upper band of the second speci�cation is wider than the width of the darker shades of red on the upper band of the �rst speci�cation. The width of the lighter shades of red on the top band of the second speci�cation is larger than the width of the lighter shades of red on the bottom band of the �rst speci�cation. This suggest that compared to the �rst speci�cation, the second one performs better in assigning high probabilities to actual events (trusting local govern- ment officials) and lower probabilities to actual cases of non-events (not trusting local government officials). (a) Model 1: Kono (b) Model 1: Western Area Urban Figure 13: This plot visualizes the �t of Model 1 that was estimated on a subset of the data: Kono and Western Area Urban. These plots suggest that the models performs better with the former data than the latter. The model performs only moderately well in assigning very high probabilities to events and very low probabilities to non-events. 40 9 Annex (Observation) : Log[Pijklm /(1 − Pijklm )] = β0jklm Xijkl + aijklm (1) (Household) : β0jk = δ00klm C0jklm + b0jklm (2) (EA) : δ00klm = γ0lm C0klm + c0klm (3) (Ward) : γ00lm = α0 + d0lm (4) (District) : α00m = 0 C0m + e0m (5) In this model, i,j, k, l and m index the observation (at each year), household and EA levels. The Greek symbols refer to parameters estimated at particular levels of analysis: β, δ and γ,α and refer to �xed effects at the observation, household, EA, ward and district levels, respectively. The random effects of intercepts at the observation, household, EA, ward and district levels are described by aijk , b0jk and c0k , d0k and e0jk . C0jk refers to household-level variables, such as whether the household lost a household member during the war. C0klm refers to EA-level variables including ethnic fractionalization and the level of war violence the community experienced during the war. C0m refers to district-level variables, such as local councils’ administrative capacity. Likelihood ratio tests of model �t suggests that estimating random intercepts for the enumeration area, wards and districts improves the �t of the model. Including random intercepts for households does not improve the model �t. Thus, we exclude the household level from the estimation of the models that we discuss in the paper. 41 References Ahmed, Ismail I. and Reginald Herbold Green. 1999. “The Heritage of War and State Collapse in Somalia and Somaliland: Local Level Effects, External Interventions and Re- construction.� Third World Quarterly 20(1):113–127. Anderson, Christopher J. 2000. “Economic Voting and Political Context: A Comparative Perspective.� Electoral Studies 19(1):151–170. Anderson, Christopher J. 2007. “The End of Economic Voting? Contingency Dilemmas and the Limits of Democratic Accountability.� Annual Review of Political Science 10:271–296. Azam, Jean-Paul and Alice Mesnard. 2002. “Civil War and the Social Contract.� Public Choice 115(3-4):455–475. Barkey, Karen. 1994. Bandits and bureaucrats: the Ottoman route to state centralization. Ithaca, N.Y.: Cornell University Press. Barr, Abigail. 2004. Kinship, Familiarity, and Trust: An Experimental Investigation. In Foundations of Human Sociality - Economic Experiments and Ethnographic Evidence from Fifteen Small-Scale Societies, ed. Joseph Henrich, Robert Boyd, Samuel Bowles, Colin Camerer, Ernst Fehr and Herbert Gintis. Oxford: Oxford University Press. Bates, Robert H. 1983. Modernization, Ethnic Competition, and the Rationality of Politics in Contemporary Africa. In State versus Ethnic Claims: African Policy Dilemmas, ed. Donald S. Rothchild and Victor A. Olorunsola. Boulder, Co.: Westview Press. Bates, Robert H. 2001. Prosperity and Violence: The Political Economy of Development. New York: Norton. Bates, Robert H. 2008a. “State Failure.� Annual Review of Political Science 11:1–12. Bates, Robert H. 2008b. When Things Fell Apart: State Failure in Late-Century Africa. Cambridge: Cambridge University Press. Bellows, John and Edward Miguel. 2009. “War and local collective action in Sierra Leone.� Journal of Public Economics 93(11-12):1144–1157. Besley, Timothy and Robin Burgess. 2002. “The Political Economy of Government Re- sponsiveness: Theory and Evidence from India.� The Quarterly Journal of Economics 117(4):1415–1451. Boex, Jamie. 2011. “Towards a Fiscal Decentralization Strategy for Sierra Leone: Recom- mendations for a Second-Generation Interngovernmental Fiscal Transfer System.� Mimeo, World Bank. 42 Brancati, Dawn (2006). 2006. “Decentralization: fueling the �re or dampening the flames of ethnic conflict and secession?� International Organization 60(3):651–685. Brewer, John. 1989. The Sinews of Power: War, Money and the English State, 1688-1783. New York: Alfred A. Knopf. Brown, Taylor, Richard Fanthrope, Janet Gardener, Lansana Gberie and M. Gibril Sesay. 2006. Sierra Leone: Drivers of Change. The IDL Group. Call, Charles T. and Vanessa Wyeth. 2008. Building States to Build Peace. Boulder, CO.: Lynne Rienner. Casey, Katerine. 2010. “Crossing Party Lines: The Effects of Information on Redistributive Politics.� Department of Economics, Brown University. Casey, Katherine. 2009. Decentralization in Practice. In Decentralization, Democracy, and Development: Recent Experience from Sierra Leone, ed. Yongmei Zhou. Washington D.C.: World Bank pp. 60–72. Chandra, Kanchan. 2004. Why Ethnic Parties Succeed: Patronage and Ethnic Head Counts in India. Cambridge University Press. Cook, Karen S., Russell Hardin and Margaret Levi. 2005. Cooperation without trust? New York: Russell Sage Foundation. Decentralization Secretariat. 2007. Comprehensive Local Government Performance Assess- ment System (CLoGPAS): 2006 Summary Report. Freetown, Sierra Leone: Government of Sierra Leone. Diamond, Larry. 1999. Developing democracy: Toward consolidation. Baltimore: Johns Hopkins University Press. Duch, Raymond and Randolph T. Stevenson. 2008. The economic vote: How political and economic institutions condition election results. Cambridge: Cambridge University Press. Eifert, Benn, Edward Miguel and Daniel N. 2010. “Political Competition and Ethnic Iden- ti�cation in Africa.� American Journal of Political Science 54(2):494–510. Fanthorpe, Richard, Lavali Andrew and Mohamed Gibril Sesay. 2011. “Decentralization in Sierra Leone: Impact, Constraints and Prospects.� DFID Sierra Leone. Fanthorpe, Richard and Roy Maconachie. 2010. “Beyond the ‘Crisis of Youth’ ? Mining, Farming, and Civil Society in Post-War Sierra Leone.� African Affairs 109(435):251–272. Fearon, James and David Laitin. 1996. “Explaining Interethnic Conflict.� American Political Science Review 90(4):715–735. 43 Fjeldstad, Odd-Helge, Lucas Katera and Erasto Ngelewa. November 2008. “Disparities Exist in Citizens’ Perceptions of Service Delivery by Local Government Authority Authorities in Tanzania.� REPOA Brief: Research on Poverty Alleviation 13. Foster, Elizabeth and Rachel Glennester. 2009. Impact of Decentralization on Public Ser- vices: Evidence to Date. In Decentralization, Democracy, and Development Recent Expe- rience from Sierra Leone, ed. Yongmei Zhou. Washington D.C.: World Bank chapter 5, pp. 73–84. Gambetta, Diego. 1988. Trust: Making and Breaking Cooperative Relations. New York: Basil Blackwell. Gelman, Andrew and Jennifer Hill. 2007. Data analysis using regression and multilevel/hier- archical models. Cambridge, UK: Cambridge University Press. Gilligan, Michael J., Benjamin J. Pasquale and Cyrus D. Samii. 2011. “Civil War and Social Capital: Behavioral-Game Evidence from Nepal.� Available at SSRN: http://ssrn.com/abstract=1911969. Glennerster, Rachel, Edward Miguel and Alexander Rothenberg. 2010. “Collective Action in Diverse Sierra Leone Communities.� NEBR Working Paper 16196. Government of Sierra Leone. 1991. The Constitution of Sierra Leone, Government Printing Department. Freetown, Sierra Leone.: . Greenhill, Brian, Michael Ward and Audrey Sacks. 2011. “The Separation Plot: A New Visual Method for Evaluating the Fit of Binary Models.� American Journal of Political Science 55(4):991–1002. Guo, Guang and Hongxin Zhao. 2000. “Multilevel Modeling for Binary Data.� Annual Review of Sociology 26:441–462. Habyarimana, James, Macartan Humphreys, Daniel N. Posner and Jeremy M. Weinstein. 2007. “Why Does Ethnic Diversity Undermine Public Goods Provision?� American Po- litical Science Review 101(4):709–725. Habyarimana, James, Macartan Humphreys, Daniel Posner and Jeremy M. Weinstein. 2009. Coethnicity and Trust. In Whom Can We Trust?, ed. Karen Cook, Margaret Levi and Russell Hardin. New York: Russell Sage Foundation pp. 42–64. Hankinson, Philippa. 2000. “Brand Orientation in Charity Organisations: Qualitative Re- search into Key Charity Sectors.� International Journal of Nonpro�t and Voluntary Sector Marketing 5(3):207–219. Hanlon, Joseph. 2004. “Is the International Community Helping to Re-create the Precondi- tions for War in Sierra Leone?� The Round Table 94:459–472. 44 Hardin, Russell. 2002. Trust and Trustworthiness. Cambridge: Cambridge University Press. Hechter, Michael. 1987. Principles of Group Solidarity. Berkeley: University of California P. Hechter, Michael. 2000. Containing Nationalism. Oxford; New York: Oxford University Press. Honaker, James, Gary King and Matthew Blackwell. November 4, 2010. “AMELIA II: A Program for Missing Data.� Version 1.2-18. Horowitz, Donald. 1985. Ethnic groups in conflict. Berkeley: University of California Press. Humphreys, Macartan and Jeremy Weinstein. 2008. “Who �ghts? The determinants of participation in civil war.� American Journal of Political Science 52(2):436–455. Iannaccone, Laurence R. and Eli Berman. 2006. “Religious extremism: The good, the bad, and the deadly.� Public Choice 128(1-2):109–129. International Crisis Group Report. 2008. “Sierra Leone: A New Era of Reform.� Africa Report 143. IRCBP Evaluation Unit. 2010. Report of the IRCBP 2008 National Public Service Survey. Freetown, Sierra Leone: Government of Sierra Leone. Kandeh, Jimmy D. 1992. “Politicization of Ethnic Identities in Sierra Leone.� African Studies Review 35:81–99. Keefer, Philip and Razvan Vlaicu. 2008. “Democracy, Credibility, and Clientelism.� Journal of Law, Economics, and Organization 24(2):371–406. Keefer, Philip and Stuti Khemani. 2012. “The Impact of Radio on the Government Distri- bution of Public Health Bene�ts.� World Bank Policy Research Working Paper 5952. Keen, David. 2005. Conflict and Collusion in Sierra Leone. Oxford: James Currey. King, Gary. 1989. Unifying Political Methodology: The Likelihood Theory of Statistical Inference. Cambridge, U.K. and New York: Cambridge University Press. Kpundeh, Sahr. 1995. Politics and Corruption in Africa: A Case Study of Sierra Leone. Lanham: University Press of America. Lake, David. 2010. “The Practice and Theory of U.S. Statebuilding: Loyalty, Legitimacy, and Regime Change.� Journal of Intervention and State-Building 4(3):257–284. Lake, David A. and Matthew Baum. 2001. “The Invisible Hand of Democracy: Political Control and the Provision of Public Services.� Comparative Political Studies 34:587–621. 45 Larizza, Marco and Brendan Glynn. 2011. “Institutions Taking Root: Building State Capac- ity in a Challenging Context - The institution of Local Councils in Sierra Leone.� Mimeo, World Bank. Leon, Sandra. 2010. “Who is responsible for what? Clarity of responsibilities in multilevel states: The case of Spain.� European Journal of Political Research 50(1):80–109. Levi, Margaret. 1988. Of Rule and Revenue. Berkeley: University of California Press. Levi, Margaret and Audrey Sacks. 2009. “Legitimating Beliefs: Concepts and Indicators.� Regulation and Governance 3(4):311–333. Levi, Margaret, Audrey Sacks and Tom R. Tyler. 2009. “Conceptualizing Legitimacy, Mea- suring Legitimating Beliefs.� American Behavioral Scientist 53(3):354–375. Levi, Margaret and Laura Stoker. 2000. “Political Trust and Trustworhiness.� Annual Review of Political Science 3:475–507. Lieberman, Evan S. 2007. “Ethnic Politics, Risk, and Policy-Making: A Cross-National Sta- tistical Analysis of Government Responses to HIV/AIDS.� Comparative Political Studies 40(12):1407–1432. Lipsky, Michael. 1980. Street-Level Bureaucracy: Dilemmas of the Individual in Public Ser- vices. New York: Russell Sage Foundation. Local Government Finance Department (LGFD), Ministry of Finance. 2011. Sectoral grants allocations by Local Councils. Freetown, Sierra Leone: Government of Sierra Leone. Manning, Ryann Elizabeth. 2008. “Challenging Generations: Youths and Elders in Rural and Peri-Urban Sierra Leone.� Justice and Development Working Paper Series 1(2). Justice for the Poor, World Bank. Manning, Ryann Elizabeth. 2009. “The landscape of local authority in Sierra Leone: how �traditional� and �modern� justice and governance systems interact.� Justice and Devel- opment Working Paper Series 1(1). Moore, Mick. 2007. “How does Taxation Affect the Quality of Governance?� Tax Notes International 47(1):79–98. Nannestad, Peter. 2008. “What Have We Learned About Generalized Trust, If Anything?� Annual Review of Political Science 11:413–436. Posner, Daniel N. 2005. Institutions and Ethnic Politics in Africa. New York: Cambridge University Press. Powell, Binham G. 2000. Elections as Instruments of Democracy. New Haven: Yale Univer- sity Press. 46 Reno, William. 1995. Corruption and State Politics in Sierra Leone. New York, New York: Cambridge University Press. Richards, Paul. 1996. Fighting for the rain forest: War, youth and resources in Sierra Leone. Portsmouth, N.H.: Heinemann. Rothstein, Bo. 2005. Social Traps and the Problem of Trust. New York: Cambridge Univer- sity Press. Rothstein, Bo and Eric M. Uslaner. 2005. “All for All: Equality, Corruption, and Social Trust.� World Politics 58(1):41–72. Sacks, Audrey. 2011. “The Political Economy of Attribution Judgments for Public Goods and Service Provision in Zambia.� Department of Sociology, University of Washington. Sacks, Audrey and Marco Larizza. 2011. “Why Quality Matters: a Multilevel Analysis of Decentralization, Local Government Performance and Legitimating Beliefs in Post- Conflict Sierra Leone.� Paper prepared for the American Political Science Association Annual Meetings, 2011. Samuels, David. 2004. “Presidentialism and Accountability for the Economy in Comparative Perspective.� American Political Science Review 98(3):425–436. Seligson, Mitchell A. 2002. “The Impact of Corruption on Regime Legitimacy: A Compar- ative Study of Four Latin American Countries.� Journal of Politics 64(2):408–433. Srivastava, Vivek and Marco Larizza. 2011. Decentralization in Postconflict Sierra Leone: The Genie is out of the Bottle. In Yes Africa Can: success stories from a dynamic conti- nent, ed. Punam Chuhan-Pole and Manka Angwafo. Washington D.C.: World Bank. Tanzi, Vito. 1998. “Corruption around the World: Causes, Consequences, Scope and Cures.� IMF Staff Papers 45. Tavits, Margit. 2007. “Clarity of Responsibility and Corruption.� American Journal of Political Science 51(1):218–229. Theiss-Morse, Elizabeth and John R. Hibbing. 2005. “Citizenship and Civic Engagement.� Annual Review of Political Science 8:227–249. Treisman, Daniel. 2007. The Architecture of Government. Cambridge: Cambridge University Press. Tyler, Tom R. 2006. “Psychological Perspectives on Legitimacy and Legitimation.� Annual Review of Psychology 57:375–400. 47 Voors, Maarten, Eleonora E. Nillesen, Philip Verwimp, Erwin H Bulte, Robert Lensink and Daan P. Van Soest. 2010. “Does Conflict Affect Preferences? Results from Field Experiments in Burundi.� Available at SSRN: http://ssrn.com/abstract=1564289 or doi:10.2139/ssrn.1564289. Wantchekon, Leonard. 2003. “Clientelism and Voting Behavior: Evidence from a Field Experiment in Benin.� World Politics 55:399–422. Weber, Max. 1968. Economy and Society. Berkeley: University of California Press. Wibbels, Erik. 2006. “MADISON IN BAGHDAD? Decentralization and Federalism in Com- parative Politics.� Annual Review of Political Science 9:165–188. World Bank. 2011. World Development Report: Conflict Security and Development. Wash- ington D.C.: World Bank. Zhou, Yongmei. 2009. Decentralization, Democracy and Development: Recent experiences from Sierra Leone. Washington DC: World Bank. Ziblatt, Daniel. 2008. “Why some cities provide more public goods than others: A subna- tional comparison of the provision of public goods in German cities in 1912.� Studies in Comparative International Development 43(3-4):273–289. 48