Policy Deliberation and Electoral Returns: Experimental Evidence from Benin and the Philippines Leonard Wantchekon IGC Growth Week LSE Fall, 2014 Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 1 / 56
Acknowledgments 1 Fujiwara, Thomas, and Leonard Wantchekon. 2013. Can Informed Public Deliberation Overcome Clientelism? Experimental Evidence from Benin. American Economic Journal: Applied Economics, 5(4): 241-55. 2 Wantchekon, Leonard, 2013. How Does Policy Deliberation Affect Voting Behavior: A Field Experiment in Benin. Working Paper. Princeton University. 3 Wantchekon, Leonard, Gabriel Lopez-Moctezuma, Thomas Fujiwara, Cecila Lero and Daniel Rubenson. 2014. Policy Deliberation and Electoral Returns: Evidence from a Campaign Experiment from the Philippines. Working Paper, Princeton University. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 2 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 3 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 3 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 3 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 3 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 4 / 56
Motivation Clientelism profoundly shapes the conduct of democratic elections and government policies (Easterly and Levine [1997]; van de Walle [2003, 2007]): 1 State resources used for short-term electoral gains. 2 Voters make decisions based on immediate material gains (e.g., vote-buying, patronage, particularistic spending) rather than long-term policy. Previous literature has focused on uncovering the structural causes of clientelism and its effects (Brusco et al. [2013]). Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 5 / 56
Motivation This research addresses institutional reforms that would facilitate the emergence of efficient redistribution (Dal Bo et. al [2008]; Olken [2008]), even under slow growth and weak state capacity. We focus on deliberative electoral campaigns (i.e., public town hall meetings where voters debate about programmatic policies) as a solution to reduce the prevalence of clientelism (Fujiwara and Wantchekon [2013]). We present experimental evidence on party-endorsed town hall meetings in Benin (Wantchekon [2013]) and in the Philippines (Fujiwara et al. [2014]). Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 6 / 56
Motivation Clientelistic platforms perform better than programmatic policies, as they generate a reciprocity between candidate and voters through the discretionary distribution of transfers (Wantchekon [2003], Finan and Schechter [2012]). However, programmatic policies under deliberation can generate this connection through a two-way communication campaign: Horizontal Communication among voters. Vertical Communication from voters to candidate. Platform transparency and deliberation may make voters more receptive to programmatic policies. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 7 / 56
Motivation Treatment Effect Deliberation could be a tool for both mobilization and support for programmatic policies. 1 Direct exposure on attendees: Voter coordination. Learn about each other s preferences and beliefs. Platform transparency. Better understand the candidate s platform. Platform customization. Actively influence policy by debating with the candidate. 2 Indirect exposure on non-participants: Information sharing. Learn about the candidate s platform from attendees in your social network (Contagious voting as in Nickerson [2006, 2008]). Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 8 / 56
Preview of the Results 1 Town hall meetings have a positive effect on turnout and on electoral support for treated candidates/parties. 2 Presence of direct effects on attendees and of indirect effects on non-participants. 3 Homogenous effects of town hall meetings across all segments of the population consistent with same programmatic platform of candidates in Benin. 4 Heterogenous effects by education, income and gender consistent with the platforms of parties in the Philippines. 5 The effects are driven by audience effects and information sharing (in Benin) from meeting attendees. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 9 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 10 / 56
Deliberative Campaign Experiments Context - Benin Among top ten most democratic countries in Africa. 31st in human development. 18th in economic governance. Nonetheless, lower levels of FDI than Cote d Ivoire and Burkina Faso. Analysts have blamed poor economic performance on electoral corruption and patronage politics. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 11 / 56
Deliberative Campaign Experiments Benin Experiment March 2011 Presidential Campaign. The top three candidates collaborated with the experiment: Yayi Boni-incumbent (53.16%) Adrien Houngbedji (35.66%), Abdoulaye Bio Tchane (6.29%). Randomized Block design. 1 Use RNG again, to select 5 villages in each district and assign two to treatment and 3 to control. We have 30 districts, 60 treatment villages and 90 control villages. 2 In collaboration with the campaign management teams, districts were assigned to treatment candidates. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 12 / 56
Deliberative Campaign Experiments Benin Experiment Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 13 / 56
Deliberative Campaign Experiments Context - Philippines Traditional political parties as shifting coalitions of elite families. In fact, 50% elected politicians are dynastic (Querubin [2011, 2013]). 20% of the House of Representatives are elected through an alternative PR election of closed party-lists. Party-lists are supposed to give representation to minority groups in Filipino society (e.g. peasants, urban poor, indigenous communities). Each party that receives 2% of the total gets one seat and an additional seat for every 2% thereafter. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 14 / 56
Deliberative Campaign Experiments Philippines Experiment May 13, 2013 Party-List Legislative Election. Two party-lists collaborated with the experiment: Akbayan (2 seats) and Umalab-Ka (no seat). Randomized Block design. Use RNG to select 13 cities/municipalities. 7 cities belong to the National Capital Region (NCR) and 6 belong Calabarzon. Use RNG to select 3 villages in each district and assign 1 village to treatment and 2 villages to control. We have 13 treatment villages and 26 control villages. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 15 / 56
Deliberative Campaign Experiments Philippines Experiment Akbayan is one of the most prominent party-lists in the country, consistently winning a seat since its foundation in 1998. Founded as a left pluralist national party, it is a multi-sectoral organization that runs mainly on a feminist and environmentalist platform. It is comprised of labor, peasants, fisherfolk, urban poor, women organizations, and LGBT formations. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 16 / 56
Deliberative Campaign Experiments Philippines Experiment Umalab-Ka was founded in 2003, but it was until 2013 that it participated in the electoral process. It is an organization aimed at the urban poor. Their legislative priority is to protect informal workers through social security programs. It is composed mainly of drivers, street vendors, and house servants. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 17 / 56
Deliberative Campaign Experiments Philippines Experiment Figure: Philippines Regions NCR and Calabarzon. Other Calabarzon NCR (Metro Manila) Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 18 / 56
Deliberative Campaign Experiments Philippines Experiment Figure: Selected Cities for the Experiment. Other Akbayan! Umalab Ka! Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 19 / 56
Deliberative Campaign Experiments Philippines Experiment Figure: City of Baras (Party Tratment: Umalab - Ka ) Other Control Treated Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 20 / 56
TREATMENT Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 21 / 56
Deliberative Campaign Experiments Town Hall Meetings One staff and one candidate representative implemented between 2 and 3 town-hall meetings (around 40 participants each). Town hall meetings lasted 90-120 minutes distributed in three stages: Introduction(10-15 minutes). Introduction to programmatic platform from the candidate. Deliberation (70-95 minutes). Rounds of questions/comments and debate. Participants were encouraged to propose amendments to platform. Resolution (10 minutes). Summary of meeting proceedings and commitment to transmit information to party leaders. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 22 / 56
Deliberative Campaign Experiments Town Hall Meetings: Benin Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 23 / 56
Deliberative Campaign Experiments Town Hall Meetings: Benin Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 24 / 56
Deliberative Campaign Experiments Town Hall Meetings: The Philippines Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 25 / 56
Deliberative Campaign Experiments Town Hall Meetings: The Philippines Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 26 / 56
CONTROL Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 27 / 56
Deliberative Campaign Experiments Control Villages Business-as-usual campaign. No instructions to candidates on what campaign strategy to follow. Local brokers organized between two and three political rallies: Festive atmosphere with music, dance, and sometimes gift distribution. Speech(10-20 minutes). Candidate (or representative) gave a speech outlining the policy agenda. One-way communication without debate and voters participation. Mobile propaganda teams using a sound system roving within villages. Posters in visible public spaces with parties name and logo. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 28 / 56
Deliberative Campaign Experiments Control Villages: Benin Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 29 / 56
Deliberative Campaign Experiments Control Villages: the Philippines Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 30 / 56
Deliberative Campaign Experiments Control Villages: the Philippines Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 31 / 56
Deliberative Campaign Experiments Benin vs. Philippines In the Benin experiments: Under deliberative campaigns (treatment), candidates offered universalistic policies (education, health, employment, corruption), as designed in advance and communicated to candidates. Under business-as-usual (control) candidates offer a mix of clientelistic and universalistic policies policies. Politician s platform is a function of treatment status The treatment effect in the Benin experiment comes from a combination of both programmatic platform and deliberation. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 32 / 56
Deliberative Campaign Experiments Benin vs. Philippines In the Philippines experiment: We try to isolate the effect of deliberation on electoral returns. We focus on legislative parties that offered specific programmatic platforms targeted to minority groups in both treatment and control villages. We study small parties that do not have the means to make credible clientelistic appeals (i.e., PL do not exert discretionary control over the distribution of goods to particular villages). Under a PR (closed list) system, party-lists have less benefits from modifying their platform in order to appeal to broader interests. Small evidence of vote-buying behavior in party-list election in the past (around 4% according to our post-electoral survey). Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 33 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results Treatment Effects Conditional Effects Attendance Effects Causal Mechanisms 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 34 / 56
Treatment Effects Benin Experiments Table: Treatment Effect on Turnout (Official Results) Dependent variable: Overall Oposition Yayi (1) (2) (3) Treatment 3.309 2.654 5.110 (1.737) (1.591) (4.872) Constant 85.48 87.59 79.66 (1.541) (1.463) (3.714) Observations 150 110 40 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the Commune Level. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 35 / 56
Figure: Treatment Effect on Turnout (Official Results) Treatment Effect on Turnout 5 0 5 10 15 95 % Confidence Intervals Overall Opposition Yayi Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 36 / 56
Treatment Effects Benin Experiments Table: Treatment Effect on Vote Shares (Individual Results) Dependent variable: Overall Oposition Yayi (1) (2) (3) Treatment 5.988 8.641-1.009 (1.177) (1.561) (1.151) Constant 67.82 57.92 94.91 (4.137) (4.216) (3.070) Observations 4529 3285 1244 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the District Level. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 37 / 56
Figure: Treatment Effect on Vote Shares (Individual Results) Treatment Effect on Votes 10 5 0 5 10 15 20 95 % Confidence Intervals Overall Opposition Yayi Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 38 / 56
Treatment Effects Philippines Experiments Table: ATE on Party-list Vote Shares at the Village Level. Dependent variable: Overall Akbayan Umalab-Ka (1) (2) (3) Treatment 2.157 2.683 0.575 (1.251) (1.342) (0.341) Constant 1.859 4.548 0.226 (0.609) (0.568) (0.085) Observations 37 37 37 Adjusted R 2 0.070 0.087 0.133 Note: p<0.1; p<0.05; p<0.01 Note: Robust Standard Errors in parentheses. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 39 / 56
Treatment Effects Philippines Experiments Figure: Heterogeneous Effects of Town Hall Meetings on Vote Shares. Estimated Treatment Effect 5 0 5 10 Estimated Treatment Effect 0.5 0.0 0.5 1.0 1.5 2.0 Luisiana Malate Marikina City Quezon City Sta Maria Taguig Baras Imus Los Banos Paranaque City Pasay Pateros Valenzuela Akbayan Cities Umalab-Ka Cities Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 40 / 56
Figure: Conditional Effect on Turnout (Benin). 95 % Confidence Intervals Marginal Effect of Treatment on Turnout 15 10 5 0 5 10 0 5 10 15 20 25 Poverty Conditional Treatment Effect on Turnout (Education) Conditional Treatment Effect on Turnout (Gender) 10 5 0 5 10 15 95 % Confidence Intervals Low High Low High Low High Overall Opposition Yayi 5 0 5 10 15 95 % Confidence Intervals Overall Opposition Yayi Male Female Male Female Male Female Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 41 / 56
Figure: Conditional Effect on Vote (Benin). Marginal Effect of Treatment on Vote 50 40 30 20 10 0 10 20 95 % Confidence Intervals 0 5 10 15 20 25 Poverty Conditional Treatment Effect on Vote (Education) Conditional Treatment Effect on Vote (Gender) 10 5 0 5 10 15 20 95 % Confidence Intervals Low High Low High Low High Overall Opposition Yayi 10 5 0 5 10 15 20 95 % Confidence Intervals Overall Opposition Yayi Male Female Male Female Male Female Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 42 / 56
Figure: Conditional Effects by gender (Philippines). Conditional Treatment Effect on Turnout (Gender) Conditional Treatment Effect 0.0 0.1 0.2 0.3 0.4 0.5 0.6 Male 95 % Confidence Intervals Female 0.1 0.0 0.1 0.2 0.3 Male 90 % Confidence Intervals 95 % Confidence Intervals Female Conditional Treatment Effect Conditional Treatment Effect 0.1 0.0 0.1 0.2 0.3 Male 90 % Confidence Intervals 90 % Confidence Intervals Female 0.1 0.0 0.1 0.2 0.3 Male 90 % Confidence Intervals 90 % Confidence Intervals Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 43 / 56 Female
Figure: Conditional Effects by income (Philippines). Marginal Effect of Treatment on Turnout 0.0 0.5 1.0 1.5 95 % Confidence Intervals Marginal Effect of Treatment on Vote 0.4 0.2 0.0 0.2 0.4 95 % Confidence Intervals 1.0 1.5 2.0 2.5 3.0 3.5 4.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Income Income 95 % Confidence Intervals 95 % Confidence Intervals Marginal Effect of Treatment on Vote for Akbayan 0.1 0.0 0.1 0.2 0.3 0.4 0.5 Marginal Effect of Treatment on Vote for Umalab Ka 0.4 0.2 0.0 0.2 1.0 1.5 2.0 2.5 3.0 3.5 4.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Income Income Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 44 / 56
Figure: Conditional Effects by education (Philippines). Marginal Effect of Treatment on Turnout 0.0 0.2 0.4 0.6 0.8 95 % Confidence Intervals Marginal Effect of Treatment on Vote 0.2 0.0 0.2 0.4 0.6 95 % Confidence Intervals 1 2 3 4 5 1 2 3 4 5 Education Education Marginal Effect of Treatment on Akbayan Vote 0.00 0.05 0.10 0.15 0.20 95 % Confidence Intervals Marginal Effect of Treatment on Umalab Ka vote 0.1 0.0 0.1 0.2 0.3 0.4 95 % Confidence Intervals 1 2 3 4 5 1 2 3 4 5 Education Education Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 45 / 56
Attendance Effects Benin Experiments Table: Effect of Attendance on Turnout (IV Results) Dependent variable: Overall Oposition Yayi (1) (2) (3) Individual Attendance 6.699 5.446 9.573 (3.753) (3.663) (9.020 ) Constant 84.81 86.39 78.69 (1.762) (1.767) (4.533) Observations 4727 3472 1255 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the Commune Level. Note: 2SLS. Instrument: Treat i. Instrumented: Individual Attendance. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 46 / 56
Attendance Effects Benin Experiments Table: Effect of Attendance on Votes (IV Results) Dependent variable: Overall Oposition Yayi (1) (2) (3) Individual Attendance 11.45 16.62-1.729 (8.175) (9.204) (5.311) Constant 66.36 59.87 91.44 (3.725) (4.101) (2.856) Observations 4238 3073 1165 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the Commune Level. Note: 2SLS. Instrument: Treat i. Instrumented: Individual Attendance. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 47 / 56
Figure: Effect of Attendance (Philippines). Probability of Turnout 0.6 0.7 0.8 0.9 1.0 1.1 95 % Confidence Intervals Probability of Voting for the Treated Parties 0.02 0.00 0.02 0.04 0.06 0.08 95 % Confidence Intervals 0.00 0.05 0.10 0.15 0.20 0.00 0.02 0.04 0.06 0.08 0.10 Probability of Attendance Probability of Attendance Turnout Vote (Overall) Probability 0.00 0.05 0.10 0.15 0.20 0.25 95 % Confidence Intervals Probability 0.05 0.00 0.05 0.10 0.15 95 % Confidence Intervals 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.1 0.0 0.1 0.2 0.3 Probability of Attendance Probability of Attendance Vote (Akbayan) Vote (Umalab-Ka) Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 48 / 56
Causal Mechanisms Town Hall Meetings may be effective because: they generate transparent policy platforms for attendees. they facilitate voter coordination among attendees. active information sharing by those who attended the meetings. The results lend support to platform transparency (OLS) and activism (causal mediation analysis - Imai et al [2011]) mechanisms. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 49 / 56
Table: Treatment Effect on Mediator Variables (Benin). Dependent variable: Audience Overall Oposition Yayi (1) (2) (3) Treatment 0.802 0.755 1.129 (0.172) (0.190) (0.175) Constant 1.287 1.326 0.973 (0.195) (0.242) (0.211) Observations 733 533 200 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the Commune Level. Audience: (1) The meeting helped you know what other villagers think. Audience: (2) You get to know the candidate better after the meeting. Audience: (3) You felt listened after the meeting. Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 50 / 56
Table: Treatment Effect on Mediator Variables (Benin). Dependent variable: Information Sharing Overall Oposition Yayi (1) (2) (3) Treatment 0.430 0.382 0.455 (0.066) (0.059) (0.079) Constant 0.177 0.196 0.254 (0.081) (0.073) (0.092) Observations 1248 930 318 Note: p<0.1; p<0.05; p<0.01 Note: Clustered Standard Errors at the Commune Level. Information Sharing: Did you share information about the meeting with other people? Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 51 / 56
Figure: Causal Mediation Analysis (Benin). ACME ACME ADE ADE Total Effect Total Effect 0.0 0.1 0.2 0.3 0.4 0.0 0.1 0.2 0.3 Treatment Effect Treatment Effect Mediation Effect of Audience Mediation Effect of Information Sharing Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 52 / 56
Figure: Treatment Effect on Audience (Philippines). 30 25 20 20 15 Density Status Control Treatment Density 10 10 5 0 0 0.00 0.05 0.10 Audience Effects 0.05 0.10 Audience Effects Audience Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 53 / 56
Figure: Causal Mediation Analysis (Philippines). ACME ADE Total Effect 0.05 0.10 0.15 Mediation Effect of Audience Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 54 / 56
Outline 1 Motivation 2 Deliberative Campaign Experiments 3 Main Results 4 Conclusions Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 55 / 56
Conclusions We present deliberative campaigns as a complementary institution that limits the electoral appeal of clientelism. In addition, we show that voters seem to reward this campaigning strategy at the polls. Further research: In-depth analysis of the intrinsic institutional effects of town hall meetings from its policy effects by looking at meeting proceedings. Is the effect on attendees driven by horizontal communication or vertical communication, or both? Follow the process of voting contagion from attendees to other voters. Through which channels attendees share the information of the meetings with other voters? Leonard Wantchekon (LSE) Policy Deliberation and Electoral Returns 2014 56 / 56