WPS7370 Policy Research Working Paper 7370 Promoting Democracy in Fragile States Insights from a Field Experiment in Liberia Eric Mvukiyehe Cyrus Samii Development Research Group Impact Evaluation Team July 2015 Policy Research Working Paper 7370 Abstract A field experiment in rural Liberia is used to study demo- into monthly exchange with United Nations peacekeep- cratic participation in fragile states. Fragile states are marked ers. Civic education workshops increased enthusiasm for by political fragmentation, local patronage systems, and electoral participation, produced a coordinated shift from voter vulnerability. To understand the effects of such con- parochial to national candidates, and increased willing- ditions on democratic expression through elections, the ness to report on manipulation. A program combining experiment introduced new forms of interaction between the two interactions had similar effects. The security rural citizens and third-party actors: (i) civic education and committees had negligible effects. Barriers to political town hall workshops directed by non-governmental organi- information and voter coordination appear to be impor- zations in communities over nine months and (ii) security tant but resolvable problems for elections in fragile states. committees that brought rural community representatives This paper is a product of the Impact Evaluation Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at emvukiyehe@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Promoting Democracy in Fragile States: Insights from a Field Experiment in Liberia Eric Mvukiyehe and Cyrus Samii⇤ ⇤ Mvukiyehe (contact author) is Economist, Development Economics Research Group, Impact Evaluation Unit, The World Bank, 1818 H Street NW, Washington DC 20433 (email: emvukiyehe@worldbank.org). Samii is Assistant Professor, Politics Department, New York University, 19 West 14th Street, New York, NY 10012 (email: cds2083@nyu.edu). The study was generously supported by funds from the Folke Bernadotte Academy (FBA), Sweden, and the New York University Department of Politics. The research comes under Columbia University IRB protocol No. AAAI1171. We thank Andrew Peterson and Madeleine Parish for helpful research assistance. For useful feedback, the authors thank Bernd Beber, Christopher Blattman, An- nette Brown, Kanchan Chandra, Delia Dumitrescu, Oeindrila Dube, Michael Gilligan, Birger Heldt, Macar- tan Humphreys, Joan Ricart-Huguet, Leonard Wantchekon, Sule Yaylaci, Thomas Zeitzoff, and workshop participants at 3ie/IFPRI, EGAP, EPSA 2013, Evaluation 2013, FBA Peacekeeping Working Group, NYU, Princeton, Stanford CDDRL, and UBC. This work does not in any way represent the views of any of the above-named institutions. Promoting Democracy in Fragile States 1 Introduction One way to characterize the “quality” of elections is in terms of the clarity with which they convey voters’ policy preferences (Powell, 2007). In “fragile states,” which are defined by a lack of consolidated political authority operating under the rule of law (North, Wallis and Weingast, 2009), a few factors interfere with election quality in these terms. These include problems in citizens’ access to information, opportunities to coordinate, and vulnerability to intimidation by patrons of political factions. We develop these ideas theoretically and then use a field experiment to study the effects of strategies to improve voter information, coordination, and security on electoral quality in post-war Liberia. Our study is motivated by the fact that citizen-oriented democracy promotion is a major component of external aid to “fragile” states like Liberia, where institutional weakness is either a pre-cursor or product of violent conflict (OECD, 2011). The United States has since 2007 spent about $13 billion per year in official development assistance to fragile and confict-affected states, with about 10% of this going to citizen-oriented governance and democracy promotion programs.1 Based on OECD accounting, about 12% of the $127 billion dollars in 2012 development aid went to governance programming.2 Such programs are based on assumptions about individuals’ desire for democratic expression combined with the belief that “elections with integrity” contribute to “the ability of a society to resolve 1 Of course the lion’s share has gone to programs in Iraq and Afghanistan. These figures come from examining United States official development assistance data for congressional years 2007-2010, as posted to http://www.usaid.gov/results-and-data/progress-data/data. Fragile states are identified as those that are shown in the UCDP-PRIO Armed Conflicts Database (v.4) as experiencing political violence since 1990. 2 From http://www.oecd.org/statistics/. Promoting Democracy in Fragile States conflicts without violence” and therefore that “policies and programmes that foster political pluralism and competition [may] sustain stability and democracy in the long run” (Annan et al., 2012, p. 9). Violence and institutional dysfunction, combined with widespread poverty and under- development, give reason to question the wisdom or practicality of democracy promotion in fragile states (Paris, 2004). Given decades of cynical assistance to dictatorial regimes, democracy assistance from Western powers is often met with skepticism in recipient coun- tries (Hennemeyer, 2011, 54-55). Even if one thinks that effective democracy promotion is possible and worthwhile, it is not clear that strategies currently pursued are in fact effec- tive (Carothers, 1999, 15-17). Can we really expect democracy promotion interventions to make a difference in fragile state contexts? This paper addresses this question in two ways. First, we develop a theoretical frame- work for understanding strategic dilemmas that voters face in fragile state contexts. This helps us to hypothesize what kinds of interventions could be effective and why. The frame- work builds on the observation that fragile states are marked by political fragmentation and local patron-client systems (North, Wallis and Weingast, 2009). Patrons possess advan- tages in their ability to control political information and issue costs or benefits on the basis of voters’ choices. Voters face a coordination problem such that even if a majority of voters prefers some outside candidate to the patron’s preferred candidate, they may nonetheless feel pressure to vote for the patron’s preferred candidate. As a result, elections may be of poor quality, failing as mechanisms for conveying of policy preferences of voters. Third party actors may have a role to play in removing two key barriers to political expression: (i) insecurity that voters face locally and (ii) lack of citizen access to coordinating information due to a poor media environment and patrons’ incentives to hoard information. Methodologically, we use a large scale field experiment and triangulated measurement techniques to estimate effects of strategies of democracy promotion that aim to improve 2 Promoting Democracy in Fragile States access to information, voter coordination, and voter security. The experiment randomly assigned combinations of a monthly civic education and town hall program and a security committee program over the course of nine months prior to the October 2011 elections. Programs were assigned to 101 villages, with 41 villages randomly assigned to control. The civic education and town hall program used a curriculum developed by the National Elections Commission and facilitated town hall discussions in communities. The security committee program linked communities to the United Nations (UN) peacekeeping mission. Outcomes were measured using polling place data on votes, a voter survey, and a set of novel behavioral measures. The results show pronounced effects of the civic education and town hall program but negligible effects from the security committee program. The civic education and town hall program substantially increased enthusiasm for civic participation and preferences for na- tional versus parochial candidates. This is consistent with the type of “coordinated shift” that our theory proposes can occur when third parties introduce new information and co- ordination opportunities in otherwise remote communities. The civic education program also increased sensitivity to voter intimidation. The security monitoring program only had a weak effect on increasing preferences for national versus parochial candidates. We begin with our theoretical framework. This is followed by a discussion of the inter- ventions and the context. Following that is a discussion of our experimental design, how we operationalize outcomes, and our estimation methods. We then present our main results, followed by an exploration of reasons for some of the unexpected findings. A conclusion draws out implications. 3 Promoting Democracy in Fragile States 2 Theoretical Framework A fragile state is one in which coercive authority is not consolidated under either the rule of law or an encompassing system of clientelism (North, Wallis and Weingast, 2009).3 Rather, coercive authority is dispersed among elites who lead localized clientelistic systems. Such elites use their coercive power to guard their exclusive control over productive resources and expropriate from other networks. Fragility characterizes many societies recovering from civil war (Rotberg, 2003). Fragility results from violence through the death, dis- placement, or submission of those needed for collective institutions for managing conflicts peacefully. Fragility may lead to new violence due to the lack of such collective institu- tions. We would expect party systems to reflect this fragmented and localized distribution of power.4 Such conditions are reinforced when local factions are institutionally uncon- strained in their ability to coerce citizens, whether through overt violence or subtler forms 3 The current literature on democracy promotion has given extensive consideration to elections under authoritarian rule. While both authoritarian and fragile states contexts are marked by weak democratic institutions (Acemoglu, Robinson and Verdier, 2004; Pande, 2011; Simpser, 2013), the two contexts differ in how power is distributed societally, with fragile states exhibiting dispersed rather than concentrated power. Other literature has fo- cused on causes of outright violence surrounding elections in fragile states (Arriola and Johnson, 2012; Bracati and Snyder, 2013; Cederman, Gleditsch and Hug, 2012; Collier and Vincente, 2014; Flores and Nooruddin, 2012; Hafner-Burton, Hyde and Jablonski, 2013; Hoglund, 2009; Straus and Taylor, 2009; Weidmann and Callen, 2012; Wilkinson, 2004) Institutional weakness and fragmented power can undermine the quality of elections in more subtle ways too. 4 The logic we present here is similar to what Ziblatt (2009) characterizes as local co- optation of electoral institutions via loyalty networks. But, in a manner similar to Weid- 4 Promoting Democracy in Fragile States of intimidation (Weidmann and Callen, 2012). By this account, elections in fragile states take place in a landscape of localized patron- client systems. Here, we develop an analytical framework that incorporates ideas from the literature on clientelism, in particular the idea that local patrons have advantages that allow them to bring voters into line despite their policy preferences (Fernando-Medina, 2007, Ch. 5; Stokes et al., 2013; Stokes, 2005). Suppose in a community, we have N (odd) voters indexed by i, and the election determined by simple majority.5 At the ballot box, voter i may choose between a patron candidate, P , and an outsider candidate, O. If elected to office, each candidate produces public goods that the voter values as ViP or ViO under the patron or outsider candidate, respectively. Voters differ in their preferences over the public goods that each candidate has to offer. Those who prefer the patron candidate’s public goods policies have ViP > ViO , and vice versa for those who prefer the outsider. This captures the idea that candidates are associated with policies that voters may each evaluate differently. In addition, each voter receives an expressive benefit, ei , for voting for the candidate that they favor on policy grounds. We also assume that the patron candidate has an advantage over the outsider candidate: if the patron candidate wins, he can impose a cost, ki , on anyone who voted against him. This assumes that the patron can observe votes, but the logic carries through if we merely assume that voters believe that the patron candidate has some ability to reward or punish on the basis of individuals’ vote choices.6 mann and Callen (2012)’s analysis of Afghanistan, in the situation we consider the institu- tions themselves play a secondary role to direct intimidation via loyalty networks. 5 The logic carries through to situations of multiple candidates in a plurality contests if we can assume that support for the two front-running candidates is substantially higher than any other candidates. 6 Patrons may even cultivate in-group favoritism and other psychological costs to de- 5 Promoting Democracy in Fragile States Suppose that these costs are greater than the expressive benefit of voting, ki > ei , in which case voters do in fact take them seriously. A voter’s payoff from voting for the patron P instead of outsider O is ⇡i (P ) = I (P wins)ViP + (1 I (P wins))ViO + I (ViP > ViO )ei , while the payoff from voting for the outsider is ⇡i (O) = I (P wins)(ViP ki ) + (1 I (P wins))ViO + I (ViP < ViO )e, where I (.) refers to the indicator function. We consider pure strategy Nash equilibria. Under these assumptions, the community level election is a coordination problem in which the patron candidate enjoys an advantage. Voters can be distinguished in terms of whether they believe that the patron candidate is positioned to win in the absence of their vote, whether they are pivotal, and whether they prefer the patron or outsider candidate.7 Because of the patron candidate’s advantage, there is always an equilibrium where everyone votes for the patron, regardless of the preference profile. Such equilibria rest on each voter believing that the patron is poised to win and that the voter is not pivotal. In such cases, voters do not believe they can change the policy outcome and, because ki > ei , find it in their interest to vote for the patron candidate. This is how the “patron’s advantage” could allow him to induce conformity even if there is a majority that opposes his policies.8 When the patron’s policies are preferred by a majority, unanimous voting for the patron is the fecting from the patron’s network (Padro-I-Miquel, 2007). 7 Because these are relatively small elections and because voters often act in blocs, piv- otally is assumed to matter. 8 Similarly, Ellman and Wantchekon (2000) and Wantchekon (1999a,b) show that vul- 6 Promoting Democracy in Fragile States unique equilibrium. Voters who prefer the patron candidate and are pivotal will vote for the patron candidate so as to secure the policy and expressive benefits. Non-pivotal voters who prefer the patron candidate will vote for the patron so as to secure the expressive benefit (and avoid and costs of betraying the patron candidate). The outsider candidate only wins in equilibrium if there is a majority preferring the outsider and such voters can coordinate. Even in such cases, the outsider does not command unanimity: even if those who prefer the patron candidate are not pivotal, they can still vote for the patron candidate to obtain an expressive benefit, knowing that the outsider candidate is not positioned to impose costs for doing so. This analysis has implications for the types of interventions that could change election behavior. First, at the most basic level, votes against the patron candidate requires that voters view outsider candidates as preferable (note that the same is not true for votes for the patron candidate). Preferences will be a function of the information to which voters have access. Voters may have access only to biased and highly limited information about the choices available to them at the ballot box (Snyder, 2000). Patrons would presumably prevent the inflow of information that may increase accountability (Besley and Burgess, 2002; Bratton et al., 1999; Djankov et al., 2010). In fragile states, poor access to informa- tion via the media is due not only to poverty, but also such strategic incentives.9 Biased and restricted electoral information limits voters’ ability to reflect on their preferences in light of diverse points of view and credible alternatives to a patron’s parochial appeals (Fujiwara nerability to coercion causes electoral outcomes to depart from the policy preferences of the electoral majority. 9 For example, BBC Media Action (2012) finds that citizen-oriented media innovations in fragile states are typically initiated by externally supported NGOs or international orga- nizations, despite their wide popularity. 7 Promoting Democracy in Fragile States and Wantchekon, 2013). As such, voters may be more likely to vote in a “parochial” (of- ten, ethnically parochial) manner than would be the case were information on competing alternatives more readily available (Casey, 2012). Voters may also be more vulnerable to manipulation and unknowingly surrendering rights. Faction leaders may have considerable leeway in convincing non-supporters that costs of registration and voting are higher than they actually are or that political operatives have a right to intrude on balloting procedures.10 Second, voting against the patron candidate requires that voters can either coordinate against the patron or are shielded from potential consequences that the patron can impose. This means that new information, on its own, will probably have no effect on vote choices, since the patron can induce votes in his favor despite policy preferences. What would be needed are safe space for public deliberation.11 Threats of patron reprisals could be removed with the introduction of third party security providers. Their effectiveness would depend, however, on the extent to which the patron’s advantage comes in his ability to inflict harm as opposed to offering rewards (Baldwin, 2013). If voters feel they have no option but to surrender their policy-preferred vote to the patron, this would dampen citizens’ enthusiasm for electoral participation (Birch, 2010). Voter turnout may still be high if voters feel that they must turn out to avoid punishment by local factions (Acemoglu, Reed and Robinson, 2013), and so assessing whether inter- ventions allow for more free voter expression would require examining subtler aspects of participation behavior. This includes studying broader forms of participation and whether 10 Daniels (2010) provides a comprehensive discussion of these and other voter deception tactics as they have been applied in the United States. 11 Collier and Vincente (2014) examine an intervention motivated by a similar logic and helped to coordinate communities against violent party activities in Nigeria’s 2007 elec- tions. 8 Promoting Democracy in Fragile States vote choices reflect patrons’ parochial preferences. This analysis suggests that interventions that substantially increase voter information about electoral choices, opportunities for voters to coordinate, or secure voters against be- ing punished for their vote choices could change electoral participation in important ways, including increasing enthusiasm for voters to participate and reducing parochial voting. The field experiment that we describe next was motivated on these grounds. 3 Interventions and Hypotheses Our study is based on a cluster-randomized field experiment with two democracy promo- tion programs—one that introduced civic education and town halls, and another that pro- moted security by connecting communities to United Nations peacekeepers. The programs were applied in villages over the nine months preceding the 2011 general elections. The content of the programs emerged from discussions between Liberian civil society organi- zations, the civil-military cooperation unit of the United Nations peacekeeping mission in Liberia (UNMIL, by it’s official acronym), the National Elections Commission public out- reach office, and the authors.12 The programs occurred in rural communities. In addition to being underserved by infrastructure and social programs, sources of political instability have been rooted in rural areas since the onset of the first civil war in 1989 (Ellis, 2006). We collaborated with a coalition of civil society organizations and the UNMIL civil-military relations unit to implement the programs. We briefly describe each program below. 12 United Nations Security Council resolution S/Res/1938 (September 2010) authorized UNMIL to assist with the 2011 general election, with a particular emphasis on support for remote areas. 9 Promoting Democracy in Fragile States 3.1 The civic education program For the civic education and town hall intervention, ten members of local civil society or- ganization chapters were hired as program facilitators for both programs. The facilitators included seven men and three women who were civil society organization staff or profes- sional teachers from the counties or adjacent counties of the villages in which they worked. For villages assigned to receive a program (randomization is discussed below), facilitators traveled via motorbike to the villages and met with the town chief to discuss the intended program and seek their consent to work in the village. Of all the villages approached, con- sent was denied in only one instance (in Lofa county) and it was in association with the security program. Program facilitators administered a civics and elections curriculum combined with ei- ther a gender rights or peace education curriculum in program villages. Then at least an hour, and often more, was devoted to town hall style discussion of electoral procedures, candidates, and rights issues. All communities received the same civic education curricu- lum, while the complementary gender rights or peace education curriculum was randomly assigned. The curriculum was administered in nine monthly open meetings taking place inside participating villages on weekends. The civic education and elections components of the curriculum were based on Liberian National Elections Commission civic education outreach materials and provided instruction on the structure of Liberian government, in- cluding distribution of powers and office-holder duties; the candidates, parties, and offices for which votes were being cast; and rules for legal campaigning and election-day ballot- ing. (Examples of outreach materials are displayed in the supplemental appendix.) During the meetings, the facilitators would begin a lesson plan based on the outreach materials. When discussing candidates and parties, the facilitators were strictly instructed to avoid trying to persuade villagers to favor one or another candidate. At the same time, the facil- 10 Promoting Democracy in Fragile States itators were able to provide clarifying information or correct facts known to be incorrect about candidates or parties. We expected that the intervention would help to clarify candidate options available and clear misconceptions that may have arisen from biased information that favored parochial (that is, co-ethnic or otherwise local) candidates. We also expected that opportunities to discuss candidates in town hall settings would allow voters to coordinate and possibly shift away from parochial candidates. Both of these factors were expected to increase voters’ enthusiasm for participation. Based on the theoretical discussion in the preceding section and the content of the civic education we hypothesize that: H1 (civic education and coordination hypothesis): the civic education and town hall intervention should increase the quality of electoral participation in program communities, where electoral quality is measured in terms of higher enthusiasm for voter participation, less receptivity to parochial appeals or voting that reflects local security monopolies, and reduced voter manipulation or surrendering of electoral rights. 3.2 The security program The security-oriented program was a pilot of a community-based electoral “early warning” program. It sought to enhance UNMIL’s visibility in rural areas during the election sea- son as a third party security guarantor (Doyle and Sambanis, 2006; Fortna, 2008; Walter, 2001). The program institutionalized exchange of insecurity-related information between rural Liberian communities and UNMIL. The program was implemented in each village through nine monthly events in the pe- riod leading up to the elections. The first event was a village-wide meeting during which the facilitator worked from a script to explain UNMIL’s assistance role with respect to the 11 Promoting Democracy in Fragile States elections and how a coalition of civil society organizations, UNMIL’s civil-military rela- tions unit, and the researchers were testing a program to bring communities into closer contact with UNMIL to help ensure peaceful elections. Then, the facilitator worked with those attending the meeting to assemble a village “security committee” of 10 people. We randomly assigned half of the villages to have the committee selected through nomination by the chief, while in the other half, the facilitators organized a ballot process in which those attending the meetings listed up to 10 people that they thought should be on the com- mittee, with the top 10 vote earners being appointed. Members of the security committee were tasked with keeping a monthly log of insecurity related events, including violent dis- putes or aggressive actions by political actors. Once a month for each of the subsequent eight months, our facilitators arranged for two out of the 10 members of the security com- mittee to be brought to the nearest UNMIL deployment base for an afternoon discussion of security related issues. The two that were chosen to participate were to rotate through all 10 members over the months of the program. The discussion sessions would include secu- rity committee members from between five to eight of the program villages. The sessions would lead off with a presentation by the UNMIL troops on recent patrolling activities, followed by presentations by each of the village security committee delegations, and then discussion. Finally, security committee members were to report back to their villages on the results of the discussion session at regular community meetings. The goal of the security committee intervention was two-fold. On the one hand, it served as an information system for UNMIL as it sought to implement its electoral assis- tance and security mandate in the countryside. Indeed, our interviews with members of the UNMIL civil-military relations unit indicated that the program was quite useful in that re- gard.13 On the other hand, and more relevant for the present discussion, our expectation was 13 Author interviews, UNMIL headquarters CIMIC unit, Monrovia, November 2011. 12 Promoting Democracy in Fragile States that the program would also increase the sense of security among villagers in the program communities. With more accurate information about the nature of threats that villagers faced and with channels of communication established, our expectation was that villagers would find UNMIL more credible in protecting them against victimization by local party patrons. This sense of security was expected to create the feeling among participants that they had freedom to vote for whomever they wanted. Based on the theoretical discussion in the preceding section and the content of the security intervention, we posit the following hypothesis: H2 (security hypothesis): the security committee intervention should increase the quality of electoral participation in program communities, where electoral quality is measured as described above. 3.3 Synergies between security and civic education programs We also consider possible synergy across the two programs. The theoretical discussion above suggests that it may be the combination of security assistance, town hall discussion, and civic education that generates improvements in the quality of electoral participation. As such, we also included a treatment condition that combined the two programs, alter- nating from month to month as to whether the activity was a security committee meeting or a community meeting, where for the latter a shortened version of the curriculum was administered.14 For this combined program we hypothesize the following: 14 Because of the alternation between programs (which was a practical necessity), the communities assigned to the combined condition received a “lighter” version of each of the constituent programs. This will be important for understanding some of the results below. 13 Promoting Democracy in Fragile States H3 (combination hypothesis): the combination of security committee and civic education programs should increase the quality of electoral participation in program communities above and beyond any of the individual programs on their own. Our study goes beyond what existing research has shown about electoral assistance in fragile states and transitional regimess. By looking at various aspects of voter behav- ior, our analysis goes beyond research suggesting that civic education programs increase pro-democratic attitudes (Bratton et al., 1999; Finkel, Horowitz and Rojo-Mendoza, 2012; Finkel and Smith, 2011), which may be confounded by self-presentation biases among pro- gram participants. While similar in spirit to research by Fujiwara and Wantchekon (2013), our study examines the effects of third-party interventions on voter parochialism, rather than campaign restrictions, which may be difficult to enforce. By examining potential syn- ergies between civic education, town hall discussion, and security enhancements, our study also goes beyond Collier and Vincente (2014) work on how anti-violence campaigns help to increase perceived security among voters. 4 Context The study region includes rural villages in five Liberian counties: Bong, Grand Bassa, Grand Gedeh, Lofa, and Nimba, with a combined population of about 1.4 million out of the country’s total of 4 million (ca. 2010). These are areas that were hit hard during the two bouts of brutal civil war the engulfed the country between 1989 and 2003.15 Liberia 15 Detailed accounts of the conflict, which claimed some 250,000 lives and displaced over one million people, are available in Adebajo (2002), Berkeley (2001), and Ellis (2006). Narrative accounts of the post-conflict period and elections since 2003 are given in Harris 14 Promoting Democracy in Fragile States constantly ranks at the very bottom of the United Nations Human Development Index, and the study areas were especially poor and underserved by infrastructure. The 2011 election season arrived with much reason to worry about the quality of elec- toral participation in the countryside. The country still relied on UNMIL as the primary security and infrastructure provider. The Liberian national police, for example, provided very patchy coverage and were frequently accused of corruption by watchdog organiza- tions and ordinary Liberians (International Crisis Group, 2012). UNMIL was limited in the extent to which its mandate permitted its engagement in the countryside, however, leaving much of rural Liberia completely uncovered by state level institutions. In a 2011 survey, Sandefur and Siddiqi (2013) found that only four percent of crimes and disputes in ru- ral Liberia were taken to the formal justice system. In a 2008-2010 panel study in rural areas Blair, Blattman and Hartman (2012) found that while overt collective violence was rare, property disputes were pervasive, indicators of intra-communal cohesion were deteri- orating, and inter-tribal biases and perceptions of inter-ethnic inequity remained endemic. Blair et al. found that key correlates of violence included factors related to ethnic diver- sity, a legacy of the ethnically charged war (Boas and Utas, 2014). Much of the country remained accessible only via unpaved pathways, and nearly the entirety of the country out- side the capital of Monrovia lacked access to a reliable electricity grid or piped water. A 2007 survey conducted by the authors found that in rural areas, two-thirds of respondents lacked regular access to radio, the primary form of media access in the country. Word- of-mouth was the primary channel of political information in rural areas. These logistical and communication challenges greatly hampered voter registration efforts (Look, 2011). Liberians and international onlookers worried about regional crises in neighboring Cote d’Ivoire and Guinea spilling over into Liberia (International Crisis Group, 2011). (2011), Lyons (1999), and Moran (2006). 15 Promoting Democracy in Fragile States Institutionally, the 2011 general elections featured ballots for three offices: the national president, elected directly through two-round run-off voting; senators, with two elected in each of 15 counties via block voting; and members of the house of representatives, with one elected via plurality in each of 73 constituencies. Liberia’s run-off system encouraged multitudes of candidates to enter the race by the logic of “king making” (Wright and Riker, 1989), resulting in a presidential ballot with sixteen candidates. House and Senate races reflected the political fragmentation anticipated by the theoretical discussion above: in the house races, candidates from 27 parties and 131 independents received votes, with most parties and candidates having geographically-concentrated (and typically ethnically defined) support bases; for the Senate, candidates from twenty parties and 15 independents secured votes. Given low rates of adult literacy (43% in 2012, according to UNESCO, and certainly lower in rural areas) most voters relied on candidate photographs and party pictograms to identify candidates, making the issue of ballot complexity especially salient (Fujiwara, 2011; Gingerich, 2012).16 Politically, the 2011 elections were informed by the previous elections in 2005. The 2005 elections were successful in many respects, being free of any significant violence. Nonetheless, the run-off election of the Unity Party’s Ellen Johnson-Sirleaf to the presi- dency involved overcoming a ten percentage point deficit in the first round against Congress for Democratic Change (CDC) candidate, George Weah. This dramatic turnaround, which involved a coordinated switch in allegiance by ethnic blocs (Harris, 2006), left a linger- ing bad taste among CDC supporters. In 2010, Sirleaf announced that she would run for re-election, breaking a promise to serve only one term. This was much to the chagrin of her opposition and especially the CDC. The CDC still included Weah among its leaders, but it nominated Winston Tubman, former Liberian justice minister, diplomat, and nephew 16 See the supplemental appendix for an example presidential ballot. 16 Promoting Democracy in Fragile States of former president Winston V.S. Tubman (served 1944-1971), to lead the 2011 presiden- tial ticket. Aside from the UP and CDC, the Liberty Party (LP), whose candidate Charles Brumskine had placed third in the 2005 presidential election, stood to repeat as a likely third or fourth place finisher. While the UP, CDC, and LP were the primary national contenders, they had to seek the support of regional and ethnic leaders. Most prominent among these was warlord-turned-senator Prince Johnson of Nimba county, who ran for president with the National Union for Democratic Progress (NUDP) party, banking on playing kingmaker in a run-off. A few significant events surrounded the October-November elections. First was an August referendum on amendments to the constitution, which changed the legislative and municipal election system to one of simple plurality rather than the prevailing run-off sys- tem (run-off was retained for the presidency). While the UP supported the amendments, the CDC was opposed, and some tensions ensued as CDC party members agitated against the referendum. Four days prior to the October first round, Sirleaf was awarded the Nobel Peace Prize. Liberian opposition reacted to this as an intrusion and tried to rally opinion around the idea that Sirleaf was more interested in impressing international contacts than ordinary Liberians (Corey-Boulet, 2011; Reuters, 2011). A few isolated attacks targted UP and CDC party affiliates and press facilities. Then, voting proceeded peacefully in the first round, with Sirleaf winning 44% of the presidential vote, Tubman receiving 33%, Prince Johnson 12%, and Brumskine 6%, thereby necessitating a run-off between Sirleaf and Tubman. The CDC protested the first round result on the basis of fraud and elections commission malfeasance. CDC protestors clashed violently with Liberian police days prior to the run-off. While Sirleaf secured Johnson’s endorsement, Tubman boycotted the run- off. Sirleaf thus won 91% of the run-off vote, but the number of voters in the second round 17 Promoting Democracy in Fragile States was only 52% of the number from the first round.17 5 Methods In this section we describe the design and analysis of the experiment, including treatment assignment, covariate balance, outcome measures, and estimation methods. 5.1 Treatment Assignment The civic education and town hall, security committee, and combined interventions were randomly assigned according to a blocked-cluster design (Gerber and Green, 2012, Ch. 4). Random assignment was at the level of the “town,” Liberia’s lowest tier geographical administrative unit. Despite the suggestion of the word “town,” in fact towns can be ei- ther rural or urban areas. Towns consist of either a village or small group of hamlets in rural areas or a cluster of quarters in urban areas. We stratified our towns by the “clans” within which they were embedded. Clans are the next higher level of geographical ad- ministration in Liberia, and they are made up of a large cluster of towns, some of which may be very distant from each other. The blocking strategy first grouped clans together on the basis of geography, forming a set of twelve geographic blocks. Because ethnicity, political tendencies, war exposure, economic conditions, and other important factors clus- ter geographically, this geographic blocking allows us to control variation in these factors across communities in different treatment conditions. More details on the randomization procedure are given in Section C of the supporting information. Table 1 shows the distribution of towns included in the study based on whether they re- 17 The Liberia National Elections Commission makes all final vote counts publicly avail- able at http://www.necliberia.org/results2011/. 18 Promoting Democracy in Fragile States ceived any security committee and civic education programming (the 26 toward the bottom right are thus in the “combined” condition). Figure 1 shows their geographic distribution. As it turns out, two towns (one civic education town and one control town) were dropped from the end line due to access problems, in which case the final tally was 142 rather than 144 towns. Such a low rate of attrition, if ignored, should have negligible consequences for bias. The imbalanced distribution across the cells is attributable to the fact that we were able to operate more security committee programs and the variation in the density of the security committee programs across clans (see Section C of the supporting information). 5.2 Covariate Balance Section D of the supporting information displays results from an analysis of pre-treatment covariates using data from the 2008 Liberian national census. We summarize the findings. Inhabitants are predominately agricultural (around three quarters) and quite poor, as evi- denced by a very low proportion of households with domiciles built out of durable walls (about five percent) or that own mobile phones or radios (two percent and five percent, respectively). Formal educational attainment is extremely low, with about two-thirds of adults not having completed primary education. There is a slight imbalance in these ed- ucation and literacy indicators as well, with civic education and security committee com- munities exhibiting slightly lower levels of literacy on average than the control communi- ties. The communities (that is, the “towns”) have population sizes that vary considerably, ranging from as few as 60 members to as many as about 8000, with an average of about 700-800; these population sizes are in good balance across treatment conditions. Just over half of these communities’ members are under the age of 20, nearly half of households were displaced at some point during the war, and they are predominately (over ninety per- cent) Christian. The main ethnic groups represented in the sample include (in order of 19 Promoting Democracy in Fragile States Table 1: Treatment assignment distribution No civic education Civic education Total No security committee 41 17 58 Security committee 58 26 84 Total 99 43 142 Figure 1: Field locations in Liberia ● ●● ● ● ●● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● Monrovia ● ● ● ●● ● Control Civ. Ed. ● Sec. Com. Combined National boundary County boundary 0 25km 50km 100km their degree of representation) Kpelle, Bassa, Gio, Mano, and Kissi. The ethnic composi- tion of the combined-treatment communities differs somewhat, with a higher share coming from ethnic groups other than these five. Our analysis checks robustness of our findings to controlling for any unbalanced covariates. 5.3 Outcome variables and data Our aim in this study is to investigate systematically the effects of two democracy-promotion strategies on the quality of electoral participation. We conceptualize this outcome in terms 20 Promoting Democracy in Fragile States of (formal) political participation broadly defined as “those activities by private citizens that are more or less directly aimed at influencing the selection of governmental personnel and/or the action they take” (Verba and Nie, 1987, p. 2). This definition encompasses a wide range of political acts and behaviors such as turning out to vote, participating in polit- ical campaigns, in which ordinary citizens engage in order express their policy preference and influence politicians’ behaviors. The quality of participation is also based on whether votes convey policy preferences. Here our outcome measurement strategy is based on the implications of our theory, which suggest that obstacles to expression will tend to bias elec- toral outcomes in favor of parochial candidates—that is, local candidates who operate on an ethnic or region-specific appeals as opposed to national-level appeals. Finally, voters’ ability to express their preferences are a function of their sense that they will face little harm in doing so. To test whether such a mechanism is operating, we also consider outcomes related to voters’ exposure to intimidation or undue influence on their vote choices. The outcome data that we have available for testing our hypotheses include official polling-place level election outcome data, survey data, a set of behavioral measures, and qualitative reports by our field facilitators. Table 2 shows how our data are used to test the hypotheses and Section F of the supporting information provide details on how the data were gathered. The polling place data include official party vote shares for presi- dent, house, and senate races, as well as invalid vote counts. Each town included in the study was matched to its appropriate polling place. The survey data were collected during the weeks just after the October 11, 2011 first-round polls and prior to the second-round polls on November 8, 2011. The survey included modules on contact by politicians and parties, various forms of civic participation, security considerations, and vote choice. For vote choice, we asked both for actual vote choice and, to get around the endogeneity of candidate appeals being targeted to voters, we also asked respondents to comment on who was most appealing in a voting vignette that provided a slate of fictional candidates with 21 Promoting Democracy in Fragile States Table 2: Outcomes from Hypotheses and Associated Measures Outcome Measure Source Enthusiasm for Attended rallies Survey participation Registered to vote Survey Turned out to vote Survey Able to associate slogans to parties Survey Purchased news clipping Behavioral activities Signed petition Behavioral activities Submitted issues post card Behavioral activities Parochial voting Diversity of vote choice (Herfindahl index) Polling place results Actual vote for national versus local candidates Polling place results Vignette vote for national versus parochial candidates Survey Manipulation and Perception that process was unfair Survey surrendering of Exposure to intimidation Survey rights Fear of being in trouble if vote "the wrong way" Survey Exposure to threats Survey Witnessed calls to violence Survey Exposed to offers of gifts Survey Exposed to offers of ballot "help" Survey Passed ballot test Behavioral activities Valid vote proportions Polling place results appeals that emphasized either regional, ethnic, national, materialist, or religious concerns. The behavioral activities include a suite of activities meant to gauge enthusiasm for par- ticipation as well as empowerment against intimidation. Respondents had the opportunity to purchase news clippings, sign petitions for clean elections, and submit postcards (to be submitted after enumerators left the village), providing a set of costly signals of interest in politics. Respondents were also issued a test on filling out a ballot, to examine effects on vulnerability to manipulation through lack of voter knowledge. Section G in the supporting information shows sample summary statistics for all of the outcomes. 5.4 Estimating Effects To estimate the effects of interventions, we fit the following regression model, the specifi- cation of which is derived from our study design: Yit = 0 + 1 Civ. Ed.t + 2 Sec. Com.t + 3 Combinedt + Xt0 + ✏it , 22 Promoting Democracy in Fragile States where Yit is the outcome for person i in town t (with the polling place data, we simply aggregate up to town t), Civ. Ed.t is the indicator for whether the town received civic ed- ucation, Sec. Com.t is the indicator for whether the town received a security committee, Combinedt is the indicator for whether the town received the combined treatment, Xt ac- counts for both blocking and the low- and high-density security conditions by including centered block and high density indicators and their interaction with the treatment vari- ables, and ✏it is an error term assumed to be clustered at the clan level.18 All models are fit using weighted least squares, where the weighting accounts for each town’s probabil- ity of being included in its treatment condition given varying numbers of towns per clan. Thus, our weighted least squares estimate of 0 is consistent for the expected value in the study areas under control, while the estimates for 1, 2, and 3 are consistent for the av- erage treatment effects (ATEs) of civic education, security committees, and combined pro- grams, respectively. Section H in the supporting information presents results that control for pre-treatment education levels, literacy rates, and ethnic group composition, as these were found to be out of balance in our analysis of covariate data. As Table 2 shows, we have multiple measures for each of the outcomes. Following Anderson (2008), whenever possible we boost the power of our tests while also ensuring that we are not led astray by multiple comparisons problems by constructing indices in the 18 Centered block fixed effects with interactions estimates the stratified sample average treatment effect, appropriately accounting for the blocking without introducing any func- tional form assumptions (Gerber and Green, 2012; Lin, 2013, pp. 71-77). For the purposes of this paper, we do not interpret the effects of the differing density conditions. As it turns out, there were no significant effects, and so we simply average over the density conditions to simplify the presentation. 23 Promoting Democracy in Fragile States form of standardized inverse-covariance-weighted averages of the outcomes.19 Inverse- covariance weighted averaging provides an optimal way to construct a linear index of different indicators that all measure a common latent factor.20 The scale of the resulting indices is in pooled outcome standard deviations (Cohen’s d). Compositional outcomes, such as party vote proportions, are restricted in that they must sum to one. To estimate effects on compositional outcomes, we use a weighted least squares method that amounts to the first stage of a seemingly-unrelated regression.21 The estima- tor is consistent, provides estimates that are readily interpretable as differences-in-mean proportions, and makes fewer modeling assumptions (e.g., independence of irrelevant al- ternatives) than parametric methods such as multinomial logistic regression, although at 19 A similar approach is used by Casey, Glennerster and Miguel (2012) and Clinging- smith, Khwaja and Kremer (2009). 20 For now we ignore any complications due to item-level missing outcome data. In most cases rates of missingness were quite low, although for one suite of outcomes—those associated with violence and intimidation—missingness rates were as high as 13%. To construct the inverse-covariance-weighted averages when there was missingness in some outcome items, we simply imputed the expected value based on the regression specification. If missingness is random with respect to outcomes conditional on treatment and blocking strata, then this introduces no bias. Even if this condition does not hold, with missingness rates as low as they were, the consequences for either bias (or for standard errors) should be negligible. 21 Some minor efficiency gains are possible by running a second stage with feasible gen- eralized least squares, however this requires that one assumes restrictions on the error terms. Our approach is equivalent to estimates that make no such restrictions beyond the clustering that is implied by the experimental design. 24 Promoting Democracy in Fragile States some cost in efficiency (Tomz, Tucker and Wittenberg, 2002). 6 Empirical Results Our theoretical discussion proposed that interventions that increase voters’ information about electoral choices, opportunities for discussion and coordination, or security vis-a- vis political patrons should result in higher “quality” electoral participation. We define quality in terms of the ability to express policy preferences in a manner that is free from pressures of local factions. As such, we expect such interventions to produce increased enthusiasm for participation, less parochial voting, and less manipulation and surrendering of electoral rights. Tables 3 through 8 display estimates of the average effects for the three different program varieties in association with these classes of outcomes. Each of the tables shows the estimated control group mean toward the top. For tables 3, 4, 7, and 8, the first column of estimates (marked “Index”) shows results for the inverse-covariance weighted indices described in the previous section. These index results provide an omnibus test of the effects of the different programs on the entire set of outcomes items that appear in the other columns. Section H in the supporting information displays effect estimates after controlling for pre-treatment education levels, literacy rates, and ethnic group composition—covariates that were found to be somewhat out of balance. None of the conclusions of the analysis change when we control for these covariates, and so our interpretations will focus on the per-design analyses presented in Tables 3 through 8. Table 3 displays results for effects on enthusiasm for participation, which should in- crease if voters indeed feel freer to express themselves in the elections. As discussed in the theory section, measuring enthusiasm for participation requires going beyond turnout, given that turnout can easily be manipulated to produce a semblance of enthusiasm. For this reason we use seven participation-related variables that tap into various types of partic- 25 Promoting Democracy in Fragile States ipation. The results show that the civic education/town halls and combined programs sig- nificantly increased enthusiasm for participation—leading to about a quarter of a standard deviation increase based on the index that combines information from the seven indicators (first column of estimates, Table 3). The difference between the average treatments effects (ATEs) for the civic education and combined programs is not statistically significant. The security committee program registers no such aggregate effect, leading one to infer that it is probably the civic education elements of the combined program that generated the positive effect in the combined program. The indicators that contribute most to the observed effect include rally participation, the purchasing of news clippings, and petition signatures. Rally attendance and news purchases increased by about 10 percentage points and petition signa- ture rates by about 15 percentage points where civic education programs took place. The security committee communities did show gains in news purchases and petition signatures, but the effects are much smaller. By contrast, the voter registration and turnout estimates suggest a ceiling effect: these are both self-reported measures, and what we see is that in the control group, 97% and 94% of respondents indicate that they were registered and then turned out, respectively, leaving little room for any increases. It is quite possible that these self-reported turnout rates are biased by respondents indicating that they registered and turned out when they in fact did not. Such reporting bias is another reason that it is important to use alternative participation measures. Tables 4, 5, and 6 display results for effects on parochial voting, which should be lower if voters can coordinate or are shielded from the pressures of local faction patrons. We use three types of outcomes to assess this. In Table 4 the outcomes are measures of diversity of vote choice based on Herfindahl index measures for presidential, senate, and house vote tallies the program communities’ polling places. The Herfindahl index ranges from 0 to 1, with higher values indicating more homogeneity in vote shares or monopolization of votes by a single actor in each location. If a program increases diversity, this should result in a 26 Promoting Democracy in Fragile States Table 3: Effects of programs on enthusiasm for participation Index‡ Rally Register Voted Slogan Bought news Petition Issues card (Control mean) 0.05 0.24 0.97 0.94 0.42 0.45 0.58 0.64 (0.04) (0.01) (0.01) (0.01) (0.02) (0.02) (0.02) (0.01) Civ. Ed. ATE 0.22⇤⇤ 0.11⇤⇤⇤ 0.00 0.03 0.01 0.09⇤⇤ 0.17⇤⇤⇤ 0.01 (0.07) (0.02) (0.01) (0.02) (0.05) (0.03) (0.03) (0.02) Sec. Com. ATE 0.03 0.02 0.01 0.00 0.08⇤ 0.05† 0.09⇤⇤ 0.02 (0.05) (0.02) (0.01) (0.01) (0.03) (0.03) (0.03) (0.02) Combined ATE 0.28⇤⇤⇤ 0.09⇤⇤⇤ 0.01 0.02 0.01 0.07⇤ 0.15⇤⇤⇤ 0.09⇤⇤ (0.06) (0.02) (0.01) (0.01) (0.03) (0.03) (0.04) (0.03) N 1426 1384 1419 1406 1426 1368 1426 1426 R2 0.10 0.14 0.05 0.06 0.11 0.15 0.14 0.08 WLS estimates account for blocking using centered block interactions (coefficients not reported). ‡ First column outcome is a standardized, inverse-covariance-weighted average of the other columns’ outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 negative effect on the Herfindahl index. In Table 5, the outcomes are vote shares for the three national-level party candidates (the LP, UP, and CDC) as well as for parochial can- didates (“Other”) from the polling place data. Less parochial voting would be reflected by decreases in voting for the “Other” candidates in favor of a national party candidate. In Table 6 the outcomes are voteshares for the different types of candidates presented as part of the voting vignette exercise in the survey. Again, less parochial voting should be asso- ciated with increased voteshares for candidates that make national appeals over regional, ethnic, or religious appeals. The results agree in part with our hypothesized effects, but not completely. With respect to diversity of vote choice, we see that the control group levels of voter homogeneity are quite high: Herfindahl indices of circa 0.4 to 0.5 are consistent with votes going predominately to two or three candidates. There is noisy indication that the civic education program caused such homogeneity to increase, contrary to the expected effect. When we look at the changes in party voteshares in Table 5, this concentration ap- pears to have been due to a coordinated shift (of about 10 percentage points) in votes away from the “parochial” candidates toward the UP. For the security committee and combined interventions, there is some suggestion of increases in voter diversity (that is, decreases in 27 Promoting Democracy in Fragile States Table 4: Effects of programs on diversity of vote choice (Herfindahl index) Index‡ Herf. Pres. Herf. Sen. Herf. Hou. (Control mean) 0.16 0.53 0.37 0.41 (0.08) (0.01) (0.01) (0.01) Civ. Ed. ATE 0.26† 0.01 0.05⇤ 0.02 (0.15) (0.02) (0.02) (0.02) Sec. Com. ATE 0.20 0.04† 0.02 0.03 (0.17) (0.02) (0.03) (0.02) Combined ATE 0.11 0.02 0.04⇤ 0.02 (0.10) (0.02) (0.02) (0.03) N 142 142 142 142 R2 0.71 0.86 0.65 0.80 WLS estimates accounting for blocking using centered block interactions (coefficients omitted). ‡ First column outcome is a standardized, inverse-covariance-weighted average of the other columns outcomes. “Herf.” stands for Herfindahl index. “Pres.”, “Sen.”, and “Hou.” stand for president, senate, and house votes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 the Herfindahl index measures), with votes from parochial candidates being reduced again, but then being more evenly spread across the national party candidates. The increased ap- peal of national-level candidates above parochial candidates (and in particular, regional and ethnic candidates) is also apparent in the results for Table 6. Thus, while the effects on vote diversity are mixed and quite noisy, we see rather compelling evidence that all types of programs induced a shift toward national level candidates away from parochial candidates. Tables 7 and 8 display results related to manipulation and surrendering of rights, which we expect to be lower to the extent that voters are better able to coordinate and are shielded from local factions. Table 7 includes nine indicators based on the survey data, capturing various forms of manipulation, while Table 8 shows effects on invalid vote counts based on the polling place data. At first glance, the effects for the civic education program appear puzzling: based on the index measure, we find a 0.11 standard deviation increase in the index of manipulation and surrendering of rights. No such aggregate effects are apparent for the other two programs. The effect for the civic education program is driven primarily 28 Promoting Democracy in Fragile States Table 5: Effects of programs on actual vote for national versus parochial candidates LP voteshare UP voteshare CDC voteshare Other voteshare (Control mean) 0.10 0.32 0.26 0.33 (0.01) (0.01) (0.01) (0.01) Civ. Ed. ATE 0.01 0.11⇤⇤⇤ 0.02 0.09⇤⇤⇤ (0.01) (0.03) (0.02) (0.02) Sec. Com. ATE 0.02 0.02 0.01 0.04⇤ (0.01) (0.02) (0.01) (0.02) Combined ATE 0.02 0.03 0.04⇤⇤ 0.05⇤ (0.01) (0.03) (0.01) (0.02) N 142 142 142 142 R2 0.90 0.88 0.91 0.95 WLS estimates account for blocking using centered block interactions (coefficients omitted). Party share outcomes are compositional and sum to one. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 Table 6: Effects of programs on vignette vote for national versus parochial candidates None Regional Ethnic National Materialist Religious No resp. (Control mean) 0.01 0.08 0.23 0.59 0.02 0.01 0.01 (0.00) (0.01) (0.02) (0.02) (0.00) (0.01) (0.01) Civ. Ed. ATE 0.00 0.04⇤⇤⇤ 0.06 0.08⇤ 0.00 0.02† 0.02 (0.00) (0.01) (0.04) (0.04) (0.01) (0.01) (0.02) Sec. Com. ATE 0.01 0.03† 0.07⇤ 0.06† 0.01 0.00 0.02 (0.01) (0.02) (0.03) (0.04) (0.01) (0.01) (0.01) Combined ATE 0.01 0.06⇤⇤⇤ 0.06⇤ 0.10⇤ 0.00 0.01 0.00 (0.00) (0.01) (0.03) (0.04) (0.01) (0.01) (0.01) N 1426 1426 1426 1426 1426 1426 1426 R2 0.44 0.44 0.44 0.44 0.44 0.44 0.44 WLS estimates accounting for blocking using centered block interactions (coefficients not reported). Outcomes are compositional and sums to one. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 by respondents reporting that they witnessed voter intimidation and threats and that they received help in filling election day ballots. At the same time, we see that the positive aggregate effect masks the fact that for some items, such as whether respondents thought that they would get in trouble for voting the “wrong” way or whether or not respondents reported receiving gifts, we estimate substantial negative effects. The civic education pro- gram and the combined program are also associated with substantial reductions in invalid votes recorded in the polling place results; no such pronounced effect was evident for the 29 Promoting Democracy in Fragile States Table 7: Effects of programs on manipulation and surrendering of rights Index‡ Unfair Intim. Trouble Threats Call to Viol. Big Gifts Small Gifts Bal. Help Bal. Test (Control mean) 0.01 0.09 0.33 0.19 0.20 0.12 0.06 0.48 0.05 0.93 (0.03) (0.01) (0.02) (0.01) (0.02) (0.01) (0.01) (0.02) (0.01) (0.01) Civ. Ed. 0.11† 0.03 0.14⇤⇤⇤ 0.05⇤ 0.06† 0.01 0.02 0.06⇤⇤⇤ 0.05⇤⇤⇤ 0.02 (0.06) (0.02) (0.03) (0.03) (0.03) (0.02) (0.02) (0.02) (0.01) (0.02) Sec. Com. 0.02 0.01 0.02 0.03 0.00 0.01 0.02 0.02 0.02⇤ 0.01 (0.06) (0.01) (0.03) (0.02) (0.03) (0.02) (0.01) (0.03) (0.01) (0.01) Combined 0.02 0.02 0.02 0.06⇤⇤ 0.02 0.03⇤ 0.01 0.01 0.03⇤ 0.03⇤ (0.06) (0.02) (0.03) (0.02) (0.03) (0.02) (0.02) (0.02) (0.01) (0.01) N 1426 1245 1363 1262 1266 1278 1405 1401 1300 1359 R2 0.07 0.07 0.06 0.08 0.08 0.09 0.08 0.08 0.05 0.04 WLS estimates account for blocking using centered block interactions (coefficients not reported). ‡ First column outcome is a standardized, inverse-covariance-weighted average of the other columns’ outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 Table 8: Effects of programs on invalid vote proportions Index‡ Inval. Pres. Inval. Sen. Inval. Hou. (Control mean) 0.39 0.10 0.09 0.09 (0.17) (0.01) (0.01) (0.01) Civ. Ed. ATE 1.04⇤⇤ 0.05† 0.05⇤ 0.04⇤⇤ (0.39) (0.03) (0.02) (0.02) Sec. Com. ATE 0.31 0.02† 0.02 0.01 (0.24) (0.01) (0.01) (0.01) Combined ATE 0.78† 0.04⇤ 0.04† 0.03⇤ (0.43) (0.02) (0.02) (0.02) N 142 142 142 142 R2 0.51 0.57 0.57 0.52 WLS estimates accounting for blocking using centered block interactions (coefficients omitted). “Inval.” stands for proportion of votes that were invalid. “Pres.”, “Sen.”, and “Hou.” stand for president, senate, and house votes. ‡ First column outcome is a standardized, inverse-covariance-weighted average of the other columns outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 security committee program. Thus, the results for these outcomes are not quite in line with our expectations. In the discussion below we present results of some auxiliary analyses that help to make sense of these findings. 30 Promoting Democracy in Fragile States 7 Discussion The results suggest that the civic education and town hall discussions had pronounced ef- fects on participation, vote choices, and the experiences of voters as they pertain to ques- tions of voting rights. The combined program appears to have had similar but somewhat less pronounced effects as compared to the civic education program. The security commit- tee program, which had communities regularly liaise with UN peacekeepers, appears only to have reduced the extent of support for parochial candidates, but in a way that was much less pronounced than for with the civic education/town hall and combined programs. In this section we take up two questions that arise from these findings. First, why was the security committee intervention so ineffective, particularly in comparison to the civic education and town halls intervention? Second, how should we interpret the unexpected effects of the civic education and town halls program on diversity of vote choice and manipulation and surrendering of rights? That the security committee program was so ineffective warrants some reflection. Our expectation was that the civic education and security committee programs would be com- plementary, an expectation that we crystalized in our combination hypothesis (H3). The study was designed to test for a synergy between them. We found no evidence of such synergy. Two possibilities may help to explain the non-results for the security commit- tees. The first explanation derives from our theory. The fragmentation of the party system, the extreme number of parochial candidates, and the relatively high vote shares for such candidates are indicative of the extent to which local patron-client systems define political choices in rural Liberia. In our theory, the “patron’s advantage” comes from the patron’s ability to impose costs conditional on voter behavior. Such costs could come in the form of threats of physical punishment. But they could also come in the form of denying benefits. There are various ways through which local patron-client systems are enforced, and it is 31 Promoting Democracy in Fragile States not clear that the relevant mechanisms are those that rely on threats of physical endanger- ment. The non-effectiveness of security committee intervention suggests such may be the case. Moreover, the very fact that community leaders overwhelmingly accepted to host the security committee intervention may speak to this point. A second explanation is not informed by our theory but rather by our observations in the field about the way the programs ran. At the most basic level, the monthly civic education meetings involved broad participation from all segments of the community, whereas direct participation in the security committees was, by design, very limited to those few individ- uals selected for the committee and to attend the monthly meeting (cf. the photographs in section B of the supporting information). The two programs also differed markedly in the “messengers” that conveyed the content of the programs. In the civic education/town halls program, the curriculum was taught by a local civil society organization facilitator. In the security committee program, the security committee members were exchanging informa- tion with foreign peacekeepers. A basic result from the literature on diffusion of ideas is that the credibility and therefore the impact of information introduced by a change agent decreases in social distance between the change agent and the subjects (Rogers, 1995, pp. 18-19). The first unanticipated effect of the civic education and town halls program was the fact that it reduced heterogeneity of vote choice and rather induced a “coordinated shift” from a mix of parochial and national candidates to somewhat more homogenous preference for national level candidates. The fact that the shift was almost entirely in favor of incumbent UP is noteworthy. Along the lines of our theoretical discussion, this could represent citi- zens updating their party preferences: by learning more about how governance institutions work via the civic education program, citizens’ confidence in the state may have increased. Then, the most natural way for citizens to express is new confidence and to “vote for the state” as opposed to voting parochially would be to switch from parochial candidates to 32 Promoting Democracy in Fragile States the incumbent UP. Our theory suggests that such a coordinated shift away from parochial candidates would have been facilitated by the deliberation and reflection on political alter- natives that the town halls allowed. Voters in a community may have sought to coordinate their votes to maximize strategic impact (Cox, 1997; Dryzek, 2009). The second unanticipated effect was in how the civic education and town halls program increased reports of exposure to some forms of intimidation. We consider a few possible explanations. First, we wanted to be sure that we were not being misled by differential non-response. The non-response rates for the items that appear in Table 7 are consider- ably higher than for the other outcomes (see the row labeled N ). If it were the case that non-response rates were higher in the control areas, and that non-response tends to be asso- ciated with experience of manipulation or surrendering of rights, then this would result in a biased ATE estimate for the civic education program. However an analysis of non-response patterns does not show that they can explain the counterintuitive findings (see section I in the supporting information). Second possibility, the fact that the outcome used in Table 7 are based on survey self- reports mean that they combine two types of effects: (i) effects on actual levels of ma- nipulation and surrendering of rights and (ii) willingness to report such manipulation and surrendering of rights. With respect to actual levels of threats, at a theoretical level, it is when citizens’ actions disturb an equilibrium favorable to incumbent elites that you would expect to see an attempts at intimidation. Such intimidation may be wielded in attempts to prevent the change. No meaningful political change comes without some degree of conflict. But it is also possible that the civic education and town halls program increased people’s willingness to report threats, even if the level of threats did not change appreciably. We investigated these possibilities as best we could. As part of our project implementation, we had project monitors (who, for the program communities were the same as the field facili- tators, and for the control communities were different) collect community-level indicators 33 Promoting Democracy in Fragile States at baseline and then again at endline. Among the endline indicators collected were reports on incidents of voter intimidation, politically-motivated attacks on community members, disputes between the chief and the national elections commission (NEC), disputes between residents and the NEC, and any inter-group violence within the community. The incident reports suggest that such instances of overt intimidation or election-related conflict were extremely rare: for these five indicators over the 126 communities in which the indicators were collected,22 we received word of only 6 incidents. In the civic education communi- ties, there were no incidents recorded at all (see section J in the supporting information). The survey responses about perceptions of unfairness or whether respondents thoughts they would be in trouble for voting the “wrong” way are not confounded in the same way as the reports on witnessing intimidation and threats, and for both of these we see beneficial ef- fects; these are also consistent with the results for the more objective invalid vote counts and even the ballot test. The most compelling interpretation of these results is that the civic education/town halls program may have introduced some degree of local political conflict and increased people’s willingness to report on threats, but avoiding creating generating any instances of violence. 8 Conclusion We used a large scale field experiment to study the extent to which the quality of elections in war-affected rural Liberia are undermined by lack of access to information about insti- tutions and politics, few opportunities to coordinate, or insecurity vis-a-vis local political factions, and whether such barriers can be overcome through civic education and town hall 22 We were unable to collect incident reports in eight control towns, one civic education town, and seven security committee towns. 34 Promoting Democracy in Fragile States or third-party security programs. We used a combination of surveys, behavioral activi- ties, and electoral returns to measure the effects of these programs. We found that the civic education and town halls program increased enthusiasm for electoral participation and gen- erated a coordinated shift from parochial to national candidates, while possibly increasing levels of contestation within communities and heightening citizens’ willingness to report on intimidation. The security committee program had only a weak effect on preferences for national versus parochial candidates. We found no evidence of synergy between the programs. Relating these results to our theoretical discussion, the experiment provides insights on barriers to citizens’ ability to express policy preferences with their votes, and therefore the quality of elections, in fragile states. The substantial effects of the civic education and town hall program is evidence in itself of barriers to information and opportunities to co- ordinate for citizens in rural Liberia. This raises questions as to why such information and coordination opportunities are lacking. One reason for the lack of information is of course material: census data show that only about five percent of households possess a radio, in- dicative of the extent to which individuals rely on word-of-mouth, and therefore locally mediated, transmission of political information. But this lack of media access likely inter- acts with strategic factors to leave voters with access only to biased and limited information. Parochial patrons have incentive to monopolize what constituents hear to secure their po- sitions. Similarly, parochial patrons have every incentive to limit opportunities for citizens to discuss and coordinate politically. In the absence of third parties that might intervene, these strategies allow patron’s to sustain their parochial advantage, potentially preventing citizens from expressing their policy preferences through the elections. By intervening in such settings, third party actors such as civil society organizations or even foreign donors could help to remove such barriers to political expression. 35 Promoting Democracy in Fragile States References Acemoglu, Daron, James A. Robinson and Thierry Verdier. 2004. “Kleptocracy and Divide-and-Rule: A Model of Personal Rule.” Journal of the European Economic As- sociation 2(2):162–192. Acemoglu, Daron, Tristan Reed and James A. Robinson. 2013. “Chiefs: Elite Control of Civil Society and Economic Development in Sierra Leone.” NBER Working Paper No. 18691. Adebajo, Adekeye. 2002. Liberia: A Warlord’s Peace. In Ending Civil Wars: The Im- plementation of Peace Agreements, ed. Stephen John Stedman, Donald Rothchild and Elizabeth M. Cousens. Boulder, CO: Lynne Rienner Publishers pp. 599–630. Anderson, Michael L. 2008. “Multiple inference and gender differences in the effects of early intervetion: A reevaluation of the Abecedarian, Perry Preschool and Early Training projects.” Journal of the American Statistical Association 103(484):1481–1495. Annan, Kofi A., Ernesto Zedillo, Martti Ahtisaari, Madeleine H. Albright, Louise Arbour, Vidar Helgesen, Rima K. Hunaidi, Festus Mogae, Amartya Sen, JAvier Solana, Vaira Vike-Freiberga and Hassan Wirajuda. 2012. Deepening Democracy: A Strategy for Improving the Integrity of Elections Worldwide. Stockholm: Global Commission on Elections, Democracy, and Security. Arriola, Leonardo R. and Chelsea B. Johnson. 2012. “Insurgents, Institutions and Post- conflict Elections.” Manuscript, University of California-Berkeley. Baldwin, Kate. 2013. “Why Vote with the Chief: Political Connections and Public Goods Provision in Zambia.” American Journal of Political Science 57(4):794–809. BBC Media Action. 2012. “Suport to Media Where Media Freedoms and Rights are Con- strained: What Works and Why?” Posted to bbcmediaction.org. Berkeley, Bill. 2001. Their Graves Are Not Yet Full: Race, Tribe and Power in the Heart of Africa. New York, NY: Basic Books. Besley, Timothy and Robin Burgess. 2002. “The Political Economy of Government Responsiveness: Theory and Evidence from India.” Quarterly Journal of Economics 117:1415–1451. Birch, Sarah. 2010. “Perceptions of Electoral Fairness and Voter Turnout.” Comparative Political Studies 43(12):1601–1622. Blair, Robert, Christopher Blattman and Alexandra Hartman. 2012. Patterns of Conflict and Cooperation in Liberia (Part 2): Prospects for Conflict Forecasting and Early Warning. New Haven: Innovations for Poverty Action. 36 Promoting Democracy in Fragile States Boas, Morten and MAts Utas. 2014. “The Political Landscape of Postwar Liberia: Reflec- tions on National Reconciliation and Elections.” Africa Today 60(4):47–65. Bracati, Dawn and Jack L. Snyder. 2013. “Time to Kill: The Impact of Election Timing on Post-conflict Elections.” Journal of Conflict Resolution 57(5):822–850. Bratton, Michael, Philip Alderfer, Georgia Bowser and Joseph Temba. 1999. “The Effects of Civic Education on Political Culture: Evidence from Zambia.” World Development 27(5):807–824. Carothers, Thomas. 1999. Aiding Democracy Abroad: The Learning Curve. Washington, DC: Carnegie Endowment for International Peace. Casey, Katherine. 2012. “Crossing Party Lines: The Effects of Information on Redistribu- tive Politics.” Manuscript, Stanford University. Casey, Katherine, Rachel Glennerster and Edward Miguel. 2012. “Reshaping institutions: Evidence on aid implacts using a pre-analysis plan.” Quarterly Journal of Economics 127(4):1755–1812. Cederman, Lars-Erik, Kristian Skrede Gleditsch and Simon Hug. 2012. “Elections and Ethnic Civil War.” Comparative Political Studies 46(3):387–417. Clingingsmith, David, Asim Ijaz Khwaja and Michael Kremer. 2009. “Estimating the Impact of The Hajj: Religion and Tolerance in Islam’s Global Gathering.” Quarterly Journal of Economics 124(3):1133–1170. Collier, Paul and Pedro C. Vincente. 2014. “Votes and Violence: Evidence from a Field Experiment in Nigeria.” The Economic Journal 124(574):F327–F355. Corey-Boulet, Robbie. 2011. “Not All Liberians are Celebrating President’s Nobel Win.” Toronto Star . Filed in Monrovia on October 7, 2011. Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World’s Electoral Systems. Cambridge: Cambridge University Press. Daniels, Gilda. 2010. “Voter Deception.” Indiana Law Review 43:343–387. Djankov, Simeon, Rafael La Porta, Florencio Lopez de Silanes and Andrei Shleifer. 2010. “Disclosure by Politicians.” American Economic Journal: Applied Economics 2:179– 209. Doyle, Michael and Nicholas Sambanis. 2006. Making War and Building Peace: United Nations Peace Operations. Princeton, NJ: Princeton University Press. Dryzek, John S. 2009. “Democratization and Deliberative Capacity Building.” Compara- tive Political Studies 42(11):1379–1402. 37 Promoting Democracy in Fragile States Ellis, Stephen. 2006. The Mask of Anarchy, Second Revised Edition: The Destruction of Liberia and the Religious Dimension of an African Civil War. New York, NY: New York University Press. Ellman, Matthew and Leonard Wantchekon. 2000. “Electoral Competition Under the Threat of Political Unrest.” Quarterly Journal of Economics 115(2):499–531. Fernando-Medina, Luis. 2007. A Unified Theory of Collective Action and Social Change. Ann Arbor: University of Michigan Press. Finkel, Steven E. and Amy E. Smith. 2011. “Civic Education, Political Discussion, and the Social Transmission of Democratic Knowledge and Values in a New Democracy: Kenya 2002.” American Journal of Political Science 55(2):417–435. Finkel, Steven E., Jeremy Horowitz and Reynaldo T. Rojo-Mendoza. 2012. “Civic Educa- tion and Democratic Backsliding in the Wake of Kenya’s Post-2007 Election Violence.” The Journal of Politics 74(1):52–65. Flores, Thomas Edward and Irfan Nooruddin. 2012. “The Effect of Elections on Postcon- flict Peace and Reconstruction.” Journal of Politics 74(2):558–570. Fortna, Virginia Page. 2008. Does Peacekeeping Work? Shaping Belligerents’ Choices After Civil War. Princeton, NJ: Princeton University Press. Fujiwara, Thomas. 2011. “Voting Technology, Political Responsiveness, and Infant Health: Evidence from Brazil.” Manuscript, Princeton University. Fujiwara, Thomas and Leonard Wantchekon. 2013. “Can Informed Public Deliberation Overcome Clientelism? Experimental Evidence from Benin.” American Economic Jour- nal: Applied Economics 5(4):241–255. Gerber, Alan S. and Donald P. Green. 2012. Field Experiments: Design and Analysis. New York, NY: Norton. Gingerich, Daniel W. 2012. “Can Institutions Cure Clientelism? Assessing the Impact of the Australian Ballot in Brazil.” Manuscript, University of Virginia. Hafner-Burton, Emilie M., Susan D. Hyde and Ryan S. Jablonski. 2013. “When Do Gov- ernments Resort to Election Violence?” British Journal of Political Science 44:149–179. Harris, David. 2006. “Liberia 2005: An Unusual African Post-Conflict Election.” Journal of Modern African Studies 44(3):375–395. Harris, David. 2011. Civil War and Democracy in West Africa: Conflict Resolution, Elec- tions and Justice in Sierra Leone and Liberia. London: I.B. Taurus. 38 Promoting Democracy in Fragile States Hennemeyer, Christian R. 2011. Democracy in Africa: Rumours of Its Demise Have Been Greatly Exaggerated. In Elections in Dangerous Places: Democracy and the Paradoxes of Peacebuilding, ed. David Gillies. Montreal: McGill-Queen’s University Press pp. 53– 70. Hoglund, Kristine. 2009. “Electoral Violence in Conflict-Ridden Societies: Concepts, Causes, and Consequences.” Terrorism and Political Violence 21:412–427. International Crisis Group. 2011. “Liberia: How Sustainable is the Recovery.” Africa Re- port No. 177, Dakar/Brussels. International Crisis Group. 2012. “Liberia: Time for Much-Delayed Reconciliation and Reform.” Africa Briefing No. 88, Dakar/Brussels. Lin, Winston. 2013. “Agnostic Notes on Regression Adjustments to Experimental Data: Reexamining Freedman’s Crtique.” Annals of Applied Statistics 7(1):295–318. Look, Anne. 2011. “Liberia Voter Registrations Short of Expectations.” Voice of America February 23, 2011. Lyons, Terrence. 1999. Voting for Peace: Postconflict Elections in Liberia. Washington, DC: Brookings Institution Press. Moran, Mary H. 2006. Liberia: The Violence of Democracy. Philadelphia, PA: University of Pennsylvania Press. Norris, Pippa. 2004. Electoral Engineering: Voting Rules and Political Behavior. Cam- bridge: Cambridge University Press. North, Douglass C., John Joseph Wallis and Barry R. Weingast. 2009. Violence and Social Orders: A Conceptual Framework for Interpreting Recorded Human History. Cam- bridge: Cambridge University Press. OECD. 2011. International Engagement in Fragile States: Can’t We Do Better? Paris: Organization for Economic Cooperation and Development. Padro-I-Miquel, Gerard. 2007. “The Control of Politicians in Divided Societies: The Poli- tics of Fear.” Review of Economic Studies 74(4):1259–1274. Pande, Rohini. 2011. “Can Informed Voters Enforce Better Governance? Experiments in Low-Income Democracies.” Annual Review of Economics 3:215–237. Paris, Roland. 2004. At War’s End: Building Peace After Civil Conflict. Cambridge: Cambridge University Press. Powell, G. Bingham. 2007. Aggregating and Representing Political Preferences. In The Oxford Handbook of Comparative Politics, ed. Carles Boix and Susan C. Stokes. Oxford: Oxford University Press pp. 653–677. 39 Promoting Democracy in Fragile States Reuters. 2011. “Sirleaf Does Not Deserve Nobel Prize, Say Weah, Tubman.” Filed in Monrovia on October 7, 2011. Rogers, Everett M. 1995. Diffusion of Innovations, Fourth Edition. New York: Free Press. Rotberg, Robert I. 2003. The Failure and Collapse of Nation-States. In When States Fail: Causes and Consequences. Washington, DC: Brookings Institution Press pp. 1–50. Sandefur, Justin and Bilal Siddiqi. 2013. “Delivering Justice to the Poor: Theory and Experimental Evidence from Liberia.” Manuscript, Center for Global Development and Oxford University. Simpser, Alberto. 2013. Why Governments and Parties Manipulate Elections. Cambridge: Cambridge University Press. Snyder, Jack L. 2000. From Voting to Violence: Democratization and Nationalist Conflict. New York, NY: Norton. Stokes, Susan. 2005. “Perverse Accountability: A Formal Model with Evidence from Argentina.” American Political Science Review 99(3):315–325. Stokes, Susan, Thad Dunning, Marcelo Nazareno and Valeria Brusco. 2013. Brokers, Vot- ers, and Clientelism: The Puzzle of Distributive Politics. Cambridge: Cambridge Uni- versity Press. Straus, Scott and Charlie Taylor. 2009. “Democratization and Electoral Violence in Sub- Saharan Africa, 1990-2007.” Manuscript, University of Wisconsin-Madison. Tomz, Michael, Joshua A. Tucker and Jason Wittenberg. 2002. “An Easy and Accurate Regression Model for Multiparty Electoral Data.” Political Analysis 10(1):66–83. Verba, Sidney and Norman H. Nie. 1987. Participation in America: Political Democracy and Social Equality. Chicago, IL: University of Chicago Press. Walter, Barbara F. 2001. Committing to Peace: The Successful Settlement of Civil Wars. Princeton, NJ: Princeton University Press. Wantchekon, Leonard. 1999a. “On the Nature of First Democratic Elections.” Journal of Conflict Resolution 43(2):245–258. Wantchekon, Leonard. 1999b. “Strategic Voting in Conditions of Political Instability: The 1994 Elections in El Salvador.” Comparative Political Studies 32(7):810–834. Weidmann, Nils B. and Michael Callen. 2012. “Violence and Elections Fraud: Evidence from Afghanistan.” British Journal of Political Science 43:53–75. Wilkinson, Steven I. 2004. Votes and Violence: Electoral Competition and Ethnic Riots in India. Cambridge: Cambridge University Press. 40 Promoting Democracy in Fragile States Wright, Stephen G. and William H. Riker. 1989. “Plurality and Runoff Systems and Num- bers of Candidates.” Public Choice 60:155–175. Ziblatt, Daniel. 2009. “Shaping Democratic Practice and the Causes of Electoral Fraud: The Case of Nineteenth Century Germany.” American Political Science Review 103(1):1–21. 41 Promoting Democracy in Fragile States Promoting Democracy in Fragile States: Insights from a Field Experiment in Liberia Supporting Information (Not for Publication) 1 Promoting Democracy in Fragile States A Outreach materials used for civic education Figure 2: Ballot instruction, I Source: authors. 2 Promoting Democracy in Fragile States Figure 3: Ballot instruction, II Source: Lindsay Forslund, Search for Common Ground, Liberia. 3 Promoting Democracy in Fragile States Figure 4: Voting process instruction Source: UNMIL and ACE Electoral Knowledge Network (http://aceproject.org/). 4 Promoting Democracy in Fragile States B Indicative photographs of the two programs Figure 5: A security committee meeting Figure 6: A civic education meeting 5 Promoting Democracy in Fragile States C Random assignment procedure Random assignment involved three steps. First, while we were able to conduct security committee programming in all twelve blocks, resources limited us to conducting the civic education and town halls in only nine of the blocks. So, we first randomly selected which nine blocks would host civic education programming. Second, we randomly assigned clans within each of the blocks to a “high-density” or “low-density” security committee con- dition. This was in anticipation of the potential for spill-over effects from the security treatment, and this aspect of the design provided exogenous variation in each town’s sus- ceptibility to spill-over (Gerber and Green, 2012, Ch. 8). Third, a simple random sample of towns was taken from each clan, and the selected towns were randomly assigned to either control, security committee, civic education, or the combined program. In “high density se- curity committee” clans, the number of towns assigned to receive security committees was higher than in “low density” clans (either four or six towns in the high density condition, depending on the size of the clan, assigned to security committees versus two in the low density condition). The number of civic education towns within each clan was always only two per clan and so there is little scope for spill-over for the civic education treatment. The number of towns selected as control was between one and four, depending on the number of towns that remained in the clan after the other treatments were assigned. 6 Promoting Democracy in Fragile States D Covariate analysis using 2008 census data D.1 Summary statistics Table 9: Summary statistics for 2008 census variables Raw mean Raw s.d. Weighted meanc Weighted s.d.c Economic indicators: HoH primary occ. is agriculturea,b 0.73 0.29 0.77 0.27 HoH primary occ. is skilleda,b 0.66 0.39 0.67 0.38 Durable wallsa 0.06 0.11 0.05 0.10 School w/in 1 hour walka 0.75 0.37 0.69 0.40 HH owns mobile phonea 0.02 0.02 0.02 0.02 HH owns radioa 0.05 0.04 0.05 0.04 Education indicators: Adults without completed primary educationa 0.66 0.16 0.67 0.15 Male literacya (age 10 and over) 0.52 0.21 0.51 0.20 Female literacya (age 10 and over) 0.30 0.18 0.27 0.16 Demographic indicators: Total population of community 693.30 1088.65 571.98 818.01 Under 20 years of agea 0.52 0.06 0.53 0.06 Between 20-49 yearsa 0.37 0.06 0.36 0.06 HH displaced during the wara 0.44 0.39 0.46 0.40 Death rate in past year 0.01 0.02 0.01 0.03 Religion: Christiana 0.92 0.15 0.91 0.15 Muslima 0.02 0.08 0.04 0.11 Ethnicity: Bassaa 0.16 0.34 0.12 0.30 Gioa 0.15 0.34 0.07 0.24 Golaa 0.00 0.00 0.00 0.00 Kissia 0.09 0.28 0.18 0.37 Kpellea 0.29 0.42 0.38 0.46 Manoa 0.11 0.30 0.10 0.29 Other ethnicitya 0.19 0.37 0.16 0.34 a expressed as a proportion of households in the community. b “HoH” refers to head of household. c Using weights to account for the unequal assignment probabilities by randomization block. D.2 Balance analysis The tables below show the estimated covariate means for the control group (first row in each table) and then estimated differences in means between the control group and the civic education, security committee, and combined treatment groups, respectively. The 7 Promoting Democracy in Fragile States estimates are produced using the same specification as described in the Methods section of the main text for the estimation of treatment effects. This specification accounts for blocking and unequal probabilities of assignment. Table 10: Economic indicators (expressed in proportions) HoH agr. HoH skilled occ. Durable walls School w/in 1 hour Mobile phone Radio (Control mean) 0.74 0.69 0.06 0.76 0.02 0.05 (0.03) (0.07) (0.01) (0.05) (0.00) (0.01) Civ. Ed. 0.02 0.02 0.02 0.06 0.01⇤ 0.00 (0.05) (0.21) (0.01) (0.14) (0.00) (0.01) Sec. Com. 0.02 0.03 0.01 0.03 0.00 0.01 (0.04) (0.08) (0.01) (0.07) (0.00) (0.01) Combined 0.11 0.03 0.06⇤ 0.18 0.00 0.02 (0.06) (0.14) (0.02) (0.10) (0.00) (0.02) “HoH” means head of household. The first two columns report the primary occupation of the head of household. WLS estimates accounting for blocking using centered block interactions. Standard errors account for clan-level clustering. ⇤ indicates t-statistic> 2 Table 11: Education indicators (expressed in proportions) Less than primary ed. Male literacy (age 10 and over) Female literacy (age 10 and over) (Control mean) 0.64 0.56 0.33 (0.01) (0.02) (0.02) Civ. Ed. 0.07⇤ 0.11⇤ 0.09⇤ (0.03) (0.04) (0.04) Sec. Com. 0.03 0.05 0.05⇤ (0.02) (0.03) (0.03) Combined 0.03 0.06 0.02 (0.03) (0.05) (0.02) WLS estimates accounting for blocking using centered block interactions. Standard errors account for clan-level clustering. ⇤ indicates t-statistic> 2 8 Promoting Democracy in Fragile States Table 12: Demographic indicators Total Population Prop. under 20 Prop. age 20-49 Proportion displaced Death rate past yr. (Control mean) 688.38 0.54 0.36 0.38 0.00 (101.59) (0.00) (0.00) (0.04) (0.00) Civ. Ed. 146.03 0.04⇤ 0.01 0.01 0.01⇤ (134.50) (0.01) (0.01) (0.06) (0.00) Sec. Com. 11.77 0.03⇤ 0.00 0.08 0.01⇤ (211.54) (0.01) (0.01) (0.06) (0.00) Combined 98.30 0.03⇤ 0.03⇤ 0.01 0.02⇤ (133.56) (0.01) (0.01) (0.07) (0.01) WLS estimates accounting for blocking using centered block interactions. Standard errors account for clan-level clustering. ⇤ indicates t-statistic> 2 Table 13: Religion of community members (expressed in proportions) Christian Muslim (Control mean) 0.95 0.03 (0.02) (0.01) Civ. Ed. 0.05 0.04 (0.07) (0.04) Sec. Com. 0.03 0.00 (0.03) (0.02) Combined 0.01 0.01 (0.05) (0.04) WLS estimates accounting for blocking using centered block interactions. Standard errors account for clan-level clustering. ⇤ indicates t-statistic> 2 Table 14: Ethnicity of community members (expressed in proportions) Bassa Gio Gola Kissi Kpelle Mano Other (Control mean) 0.16 0.11 0.00 0.10 0.28 0.16 0.18 (0.02) (0.01) (0.00) (0.02) (0.03) (0.02) (0.02) Civ. Ed. 0.04 0.01 0.00⇤ 0.02 0.06 0.03 0.01 (0.03) (0.01) (0.00) (0.05) (0.03) (0.02) (0.05) Sec. Com. 0.01 0.04 0.00⇤ 0.02 0.01 0.05 0.02 (0.03) (0.03) (0.00) (0.02) (0.04) (0.04) (0.02) Combined 0.02 0.02⇤ 0.00⇤ 0.14 0.02 0.04⇤ 0.17⇤ (0.03) (0.01) (0.00) (0.09) (0.04) (0.02) (0.08) WLS estimates accounting for blocking using centered block interactions. Standard errors account for clan-level clustering. ⇤ indicates t-statistic> 2 9 Promoting Democracy in Fragile States E Survey non-response patterns Table 15: Effects of programs on non-response rates Non-response (Control mean) 0.65 (0.04) Civ. Ed. ATE 0.06 (0.06) Sec. Com. ATE 0.06 (0.05) Combined ATE 0.07 (0.05) N 1414‡ R2 0.36 WLS estimates account for blocking using centered block interactions (coefficients not reported). Outcome for first column is standardized, covariance-weighted average of other columns’ outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 ‡ For 12 respondents, no information was recorded as to whether the respondent was from the original list or not. 10 Promoting Democracy in Fragile States F Details on outcome data sources The polling-place level election outcome data were obtained from the official results posted by the National Elections Commission to its website.23 Each town included in the study was matched to its appropriate polling place. Polling places typically service more than one town, and so polling place results would include votes of individuals from both study and non-study towns. In no cases were there any polling places that included both treated and control communities. Thus, the fact polling places combine results for study and non-study towns introduces some noise into the outcome measures and implies some loss of power in trying to estimate treatment effects. The survey sample was drawn through simple random sampling from full listings of adults constructed for each town. We sampled 10 individuals per town, stratifying the town level sample so as to ensure a 50-50 gender balance.24 The response rate with respect to the names drawn from the town lists was 71% overall, with no evidence of significant differential response by treatment condition (see supporting information section E). All non-responding individuals were replaced by available adults from the same household. The behavioral activities were conducted after the survey modules on elections and political perceptions were complete. First, the enumerator presented to the respondent an anti-violence petition organized by the non-governmental organizations (NGOs) with which we were working. The enumerator asked whether the respondent wanted to have their name added to the petition, which the NGOs intended to circulate among political leaders. The enumerator made clear that this was totally voluntary. Next, the enumerator provided each respondent with an issues card that was addressed to the local House rep- resentative. The respondent had the option to fill out the issues post card, which provided options for them to list what policy issues should be prioritized, and then drop it off in a secure box positioned in the center of town. The boxes were then collected by our field facilitators and the cards were delivered to the respective representatives. The cards did not require that respondents fill out their names, but serial codes on the cards allowed us to link them back to the surveys. Both the petition and post card were designed to test for moti- vation and enthusiasm for civic participation. The third activity was a ballot test, in which the enumerator provided a mock version of the official ballot to the respondent, and the respondent was asked to demonstrate what was the appropriate way to fill it out. Finally, in the fourth activity, the enumerators presented the opportunity for the respondent to buy a set of news clippings from recent news wires. Such clippings are commonly sold in towns in lieu of proper newspapers, which are not available outside Monrovia. The ballot test and news clipping offer were meant to test for capability and motivation for informed, effective 23 The website is http://www.necliberia.org/results2011/. 24 Due to a very minor error in enumerator assignments, we ended up with six extra respondents than we had intended for a total of 1426 rather than 1420. 11 Promoting Democracy in Fragile States civic participation. 12 Promoting Democracy in Fragile States G Outcome data summary statistics All outcome data summary statistics below are the raw sample summary statistics. (The regression tables in the main text show estimated control group means applying the assign- ment probability weights.) Table 16: Enthusiasm for participation variables summary statistics Mean S.D. Min Max N Inverse covariance weighted mean index 0.00 1.00 -3.78 2.01 1426 Attended rally 0.26 0.44 0 1 1384 Registered to vote 0.97 0.17 0 1 1419 Voted 0.93 0.26 0 1 1406 Know slogan 0.38 0.48 0 1 1426 Bought news 0.50 0.50 0 1 1368 Signed Petition 0.64 0.48 0 1 1426 Submitted Issues Card 0.63 0.48 0 1 1426 With the exception of the index, all variables are coded such that 1=yes and 0=no. Table 17: Diversity of vote choice summary statistics Mean S.D. Min Max N Inverse covariance weighted mean index 0.00 1.00 -1.38 3.23 142 Presidential voteshares Herfindahl index 0.49 0.20 0.24 0.94 142 Senate voteshares Herfindahl index 0.36 0.16 0.18 0.86 142 House voteshares Herfindahl index 0.38 0.19 0.14 0.93 142 Table 18: Actual vote for national versus parochial candidates summary statistics Mean S.D. Min Max N Liberal Party (LP) voteshare 0.11 0.16 0.00 0.63 142 Unity Party (UP) voteshare 0.35 0.21 0.01 0.89 142 Congress for Democratic Change (CDC) voteshare 0.25 0.27 0.00 0.94 142 Other parties voteshare 0.29 0.30 0.00 0.98 142 13 Promoting Democracy in Fragile States Table 19: Vignette vote for national versus parochial candidates summary statistics None Regional Ethnic National Materialist Religious No resp. Total Frequency 18 89 284 928 51 20 36 1426 Percent 1 6 20 65 4 1 3 100 Table 20: Manipulation and surrendering of rights variables summary statistics Mean S.D. Min Max N Inverse covariance weighted mean index 0.00 1.00 -2.84 4.74 1426 Elections were unfair 0.07 0.26 0 1 1245 Witnessed intimidation 0.34 0.47 0 1 1363 Feared trouble if voted wrong way 0.16 0.37 0 1 1262 Witnessed threats† (0=never, 1=sometimes, 2=often) 0.20 0.43 0 2 1266 Witnessed calls to violence (0=never, 1=sometimes, 2=often) † 0.13 0.34 0 2 1278 Promised big gifts 0.04 0.20 0 1 1405 Promised small gifts 0.47 0.50 0 1 1401 Received ballot help 0.07 0.26 0 1 1300 Passed ballot test 0.94 0.24 0 1 1359 With the exception of the index and the two variables marked with a † , all variables are coded such that 1=yes and 0=no. Table 21: Invalid vote proportions summary statistics Mean S.D. Min Max N Inverse covariance weighted mean index 0.00 1.00 -1.70 4.21 142 Presidential invalid voteshares 0.08 0.06 0.00 0.39 142 Senate invalid voteshares 0.07 0.05 0.01 0.25 142 House invalid voteshares 0.07 0.04 0.00 0.24 142 14 Promoting Democracy in Fragile States H Results controlling for imbalanced covariates Table 22: Effects of programs on enthusiasm for participation, controlling for imbalanced covariates Index Rally Register Voted Slogan Bought news Petition Issues Card Civ. Ed. ATE 0.20⇤⇤ 0.10⇤⇤⇤ 0.00 0.02 0.01 0.11⇤⇤⇤ 0.16⇤⇤⇤ 0.03 (0.08) (0.03) (0.01) (0.02) (0.06) (0.03) (0.03) (0.03) Sec. Com. ATE 0.03 0.02 0.01 0.00 0.08⇤⇤ 0.06⇤ 0.09⇤⇤ 0.03 (0.05) (0.02) (0.01) (0.01) (0.03) (0.03) (0.03) (0.03) Combined ATE 0.30⇤⇤⇤ 0.08⇤⇤ 0.01 0.03⇤ 0.02 0.11⇤⇤ 0.17⇤⇤⇤ 0.07⇤ (0.09) (0.03) (0.01) (0.01) (0.04) (0.04) (0.04) (0.03) N 1426 1384 1419 1406 1426 1368 1426 1426 R2 0.10 0.14 0.05 0.07 0.11 0.15 0.14 0.08 WLS estimates account for blocking using centered block interactions (coefficients not reported). These estimates control for proportion without primary ed., male and female literacy rates, and proportion other ethnicity. First column outcome is a standardized, inverse-covariance-weighted average of the other columns’ outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 Table 23: Effects of programs on diversity of vote choice, controlling for imbalanced co- variates Index Herf. Pres. Herf. Sen. Herf. Hou. Civ. Ed. ATE 0.19 0.01 0.04 0.02 (0.16) (0.02) (0.02) (0.03) Sec. Com. ATE 0.25 0.04† 0.04 0.03 (0.16) (0.02) (0.02) (0.02) Combined ATE 0.27 0.05 0.09⇤⇤ 0.04 (0.17) (0.03) (0.03) (0.03) N 142 142 142 142 R2 0.73 0.87 0.71 0.81 WLS estimates accounting for blocking using centered block interactions (coefficients omitted). First column outcome is a standardized, inverse-covariance-weighted average of the other columns outcomes. “Herf.” stands for Herfindahl index. “Pres.”, “Sen.”, and “Hou.” stand for president, senate, and house votes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 15 Promoting Democracy in Fragile States Table 24: Effects of programs actual vote for national versus parochial candidates, control- ling for imbalanced covariates LP voteshare UP voteshare CDC voteshare Other voteshare Civ. Ed. ATE 0.02 0.12⇤⇤⇤ 0.02 0.09⇤⇤⇤ (0.01) (0.02) (0.02) (0.02) Sec. Com. ATE 0.02 0.02 0.01 0.04⇤ (0.01) (0.02) (0.01) (0.02) Combined ATE 0.01 0.01 0.03⇤ 0.03 (0.02) (0.03) (0.02) (0.03) N 142 142 142 142 R2 0.91 0.89 0.92 0.96 WLS estimates account for blocking using centered block interactions (coefficients omitted). Party share outcomes are compositional and sum to one. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 Table 25: Effects of programs on vignette voting for national versus parochial candidates, controlling for imbalanced covariates None Regional Ethnic National Materialist Religious No resp. Civ. Ed. ATE 0.00 0.01 0.03 0.27⇤⇤⇤ 0.39⇤⇤ 0.10⇤⇤⇤ 0.04 (0.00) (0.01) (0.03) (0.07) (0.14) (0.02) (0.05) Sec. Com. ATE 0.01 0.01 0.11⇤⇤⇤ 0.16 0.34⇤⇤ 0.03 0.01 (0.01) (0.00) (0.03) (0.12) (0.12) (0.03) (0.03) Combined ATE 0.01 0.01 0.04 0.10⇤⇤ 0.12† 0.06⇤ 0.04 (0.00) (0.04) (0.03) (0.04) (0.06) (0.02) (0.03) N 1426 1426 1426 1426 1426 1426 1426 R2 0.44 0.44 0.44 0.44 0.44 0.44 0.44 WLS estimates accounting for blocking using centered block interactions (coefficients not reported). Outcome is compositional and sums to one. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 Table 26: Effects of programs on manipulation and surrendering of rights, controlling for imbalanced covariates Index Unfair Intim. Trouble Threats Call to Viol. Big Gifts Small Gifts Bal. Help Bal. Test Civ. Ed. ATE 0.16⇤ 0.01 0.14⇤⇤⇤ 0.04† 0.07⇤ 0.00 0.01 0.06⇤ 0.05⇤⇤⇤ 0.03 (0.07) (0.02) (0.02) (0.03) (0.03) (0.02) (0.02) (0.02) (0.01) (0.02) Sec. Com. ATE 0.01 0.00 0.03 0.03 0.01 0.02 0.02 0.02 0.02⇤ 0.01 (0.06) (0.02) (0.03) (0.02) (0.03) (0.02) (0.01) (0.02) (0.01) (0.01) Combined ATE 0.02 0.00 0.03 0.07⇤⇤⇤ 0.02 0.00 0.00 0.00 0.04⇤ 0.01 (0.06) (0.02) (0.03) (0.02) (0.03) (0.02) (0.02) (0.04) (0.02) (0.02) N 1426 1245 1363 1262 1266 1278 1405 1401 1300 1359 R2 0.07 0.09 0.07 0.09 0.09 0.09 0.08 0.09 0.05 0.04 WLS estimates account for blocking using centered block interactions (coefficients not reported). The outcome for the first column is a standardized, covariance-weighted average of the other columns’ outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 16 Promoting Democracy in Fragile States Table 27: Effects of programs on invalid vote proportions, controlling for imbalanced co- variates Index Inval. Pres. Inval. Sen. Inval. Hou. Civ. Ed. ATE 1.11⇤⇤⇤ 0.06⇤ 0.05⇤⇤ 0.05⇤⇤⇤ (0.32) (0.02) (0.02) (0.01) Sec. Com. ATE 0.32 0.03⇤ 0.02† 0.01 (0.23) (0.01) (0.01) (0.01) Combined ATE 0.58 0.03⇤ 0.03 0.02 (0.39) (0.01) (0.02) (0.01) N 142 142 142 142 R2 0.54 0.62 0.59 0.57 WLS estimates accounting for blocking using centered block interactions (coefficients omitted). “Inval.” stands for proportion of votes that were invalid. “Pres.”, “Sen.”, and “Hou.” stand for president, senate, and house votes. First column outcome is a standardized, inverse-covariance-weighted average of the other columns outcomes. Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 17 Promoting Democracy in Fragile States I Missing data patterns for manipulation and surrender- ing of political rights outcomes Table 28: Effects of programs on missingness rates for manipulation and surrendering of rights Unfair Intim. Trouble Threats Call to Viol. Big Gifts Small Gifts Bal. Help Bal. Test (Control mean) 0.12 0.04 0.12 0.10 0.10 0.01 0.01 0.08 0.05 (0.01) (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.01) (0.01) Civ. Ed. ATE 0.01 0.00 0.03 0.03 0.03 0.01† 0.01 0.00 0.03 (0.03) (0.03) (0.03) (0.03) (0.02) (0.00) (0.00) (0.02) (0.02) Sec. Com. ATE 0.00 0.00 0.02 0.01 0.00 0.02⇤⇤ 0.01⇤⇤ 0.01 0.02⇤⇤ (0.02) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Combined ATE 0.02 0.02 0.02 0.00 0.01 0.00 0.01⇤ 0.02 0.02† (0.02) (0.01) (0.02) (0.02) (0.02) (0.01) (0.01) (0.01) (0.01) N 1426 1426 1426 1426 1426 1426 1426 1426 1426 R2 0.09 0.14 0.06 0.07 0.08 0.04 0.06 0.06 0.05 WLS estimates account for blocking using centered block interactions (coefficients not reported). Standard errors account for clan-level clustering. † significant at p < .10; ⇤ p < .05; ⇤⇤ p < .01; ⇤⇤⇤ p < .001 18 Promoting Democracy in Fragile States J Incident report results for manipulation and surrender- ing of rights Control Civ. Ed. Sec. Com. Combined Total Voter intimidation 0 0 0 0 0 Political attack 0 0 0 0 0 Dispute between chief and NEC 0 0 1 0 1 Dispute between resident and NEC 1 0 1 0 2 Inter-group violence 0 0 1 2 3 N‡ 33 16 51 26 126 Incident reports were not received for 16 towns, including 8 control towns, 1 civ. ed. town, and 7 ‡ sec. com. towns. 19