Can information that raises voter expectations improve accountability?

Can information that raises voter expectations improve accountability? A field experiment in Mali Jessica Gottlieb Stanford University, Political Science May 8, 2012

Overview Motivation: Preliminary studies showed voters uncertain about more than just previous government performance Question: Does voter uncertainty about what their governments can and should do undermine democratic accountability? Strategy: Field experiment manipulates this type of information and explores effects on political behaviors Findings: Increased performance-based voting in hypothetical simulations, increased likelihood of openly challenging leadership, and decreased transparency of leaders

Effects of information asymmetries on accountability Theory explains link between information asymmetries and democratic failure (Lipset 1959; Almond & Verba 1963; Ferejohn 1986; Besley 2006) Relative consensus among non-experimental studies that access to information reduces corruption, increases responsiveness of government (Brunetti & Weder 2003; Ahrend 2002; Besley & Burgess 2002) Experimental studies generate mixed results (Chong et al 2012; Banerjee et al 2010; de Figueiredo et al 2011)

Missing piece: low voter expectations If voters underestimate the value of government, they will care less about government performance than they should Performance information vs. relative performance information Argument Information that raises voter expectations of what government can and should do will make performance-based voting more likely, thereby improving accountability.

Context: Mali s troubled democracy Good example of state with robust democratic institutions and weak accountability - Democratic since 1992, often hailed as beacon of West African democracy (with exception of recent 3-week coup) - Still one of the poorest countries in the world; Malians twice as poor and half as literate as average sub-saharan African - Anecdotal evidence of poor governance: corruption, nepotism, poor public service provision Conducive to within-country study - In 1996, decentralization reform created 703 communes with elected councils controlling budget for local public goods Generalizability issues - Highly rural 80% of population - No systematic civic education

Linking information and expectations Evidence of low voter expectations in Mali: Afrobarometer shows a majority thinks the national rather than local government is responsible for local public goods Half of people in my survey incorrectly believed the local government couldn t afford small public goods projects Two types of information can raise voter expectations: 1 information about what governments should do, or are legally responsible for doing 2 information about what governments can do, or what resources they have at their disposal

Overview of research design Among 95 communes in Mali, randomly assigned a 2-part civics course (31 control communes) - All rural communes across 5 districts in Mali - Stratified by incumbency, poverty, and geographic location 2 treatment variations: 1 In 32 communes, provided information about local government capacity and basics of democracy/decentralization 2 In 32 communes, provided the same plus information about relative government performance Evaluation strategy: - Household surveys - Town hall meetings - Chief and leader surveys

The intervention 2-part course in which T1 receives only first component and T2 receives both: 1 Component 1: expectations of local government - Responsibilities of local government to provide public goods - Annual budget of a rural commune with examples of line items - Basics of democratic accountability 2 Component 2: relative government performance - Regularity of town meetings - Concentration of public goods financed in the commune seat - Number of projects financed by the commune in each village - Rates of tax recovery Policy motivation for separating components in this way

Motivation Research design Theory Data analysis Conclusion

Formalizing a new type of uncertainty A representative voter is uncertain about whether or not the politician has a budget for a public good in the village; the politician knows whether the funds are available. The voter can either condition re-election on gifts or on gifts and public goods. If the probability of there being a budget for public goods is low, or if the voter thinks it is low, then the politician gets sanctioned even when he is acting responsibly. This makes it more attractive for the incumbent to shirk, leaving the voter with nothing including gifts. Comparative static The more uncertain a voter is about whether the government has a budget for public goods in their village, the less likely they are to ever condition their vote on the provision of public goods.

Hypotheses H1 Increasing voter information about potential government performance will increase the cut point at which poor-performing candidates are sanctioned. H2 Increasing voter information about potential government performance will increase the likelihood of voting along the performance dimension. H3 The more public the information signal, the greater the treatment effect on voter behavior.

Econometric specification for survey data Average treatment effects account for blocked randomization: ATE ˆ j = 1 N N (y Tj i y Ci ) where i is the block assigned to each commune in randomization, N is the total number of blocks (31 or 32), y Ci is the average outcome in the control group in block i, and y Tj i is the average outcome in treatment group j in block i. i=1 When comparing T to C, y Tj i = y T 1 i +y T2 i 2.

H1: Poor-performing candidates sanctioned more often

H2: Increased likelihood of performance-based voting Mean vote share for Candidate A Control T1 T2 Baseline 0.288 0.315 0.427 Kin 0.419 0.404 0.346 Chief support 0.376 0.361 0.357 First differences Control T1 T2 Difference (Kin - Baseline) 0.132*** 0.090** -0.081* (0.039) (0.040) (0.043) Difference (Chief - Baseline) 0.088** 0.046-0.070 (0.039) (0.036) (0.044) Difference-in-differences T1 - C T2 - C T2 - T1 Kin condition -0.049-0.208*** -0.170*** (0.053) (0.055) (0.051) Chief condition -0.047-0.154** -0.116** (0.051) (0.057) -(0.053)

Impact on behavior: challenges to leadership at town halls Mean number of challenges by group Control T1 T2 Challenges 4.125 5.93 5.83 N 8 13 12 Mean differences Difference p value (2-sided) exact p T1 - C 1.80 0.125 0.148 T2 - C 1.71 0.109 0.178 T - C 1.76 0.086 0.113 Example challenge from a woman in a treated village Since decentralization the village of Gombala has paid their taxes at 100%, but they have never received any investment from the commune. What is the reason for this?

Mechanism I: Raised expectations Mean effects analysis: index of respondent expectations Variable Coefficient M1 M2 M3 T1 0.031 0.046* 0.087** (0.036) (0.028) (0.041) T2 0.015 0.047* 0.104** (0.036) (0.028) (0.043) Majority party 0.036 (0.066) Majority party x T1-0.179** (0.084) Majority party x T2-0.228*** (0.082) Intercept 0.002 0.133-0.009 (0.026) (0.111) (0.029) N 31 5,452 95 Controls N Y N

Mechanism II: Coordination Mean squared error around true vote share for Candidate A Baseline Kin Chief Control mean 0.146 0.122 0.130 T1 (ATE) -0.027-0.012-0.006 (0.020) (0.013) (0.016) T2 (ATE) -0.041** -0.007-0.004 (0.019) (0.011) (0.015) N 31 31 31

Differential effects H3 Treatment effects are stronger when a majority of villages in the commune are treated. Treatment effects are stronger the lower is baseline civic knowledge (measured by outlying villages vs. commune seat). Treatment effects are stronger in more politically competitive villages, providing support for complementary relationship between information and political competition.

What kind of information works? T1 and T2 have effects in the same direction, but effects of T2 are generally stronger and more significant: No significant difference between T1 and T2 in level of knowledge/expectations, even among participants T2 has bigger/stronger effect in voting simulations, on ability to coordinate One interpretation: an additive effect of information Information about government capacity (T1) works to raise expectations, but is not sufficient to induce changes in behavior. For that to occur, information about government performance (T2) is required.

Cautionary tale: Incumbent response Survey administered to councilmember in 95 sample communes finds suggestive evidence of a decrease in transparency of elected officials 1 A list experiment finds there is a nearly significant decrease in the proclivity of councilmembers to say they will campaign on transparency in the next election. 2 The number of public meetings held in the few months between the intervention and the survey decreases from an average of about 7 in the control group to about 4 in the treated groups.

Take-away Even with a very small-scale, brief treatment, we observe effects of information that raises voter expecations on both citizen behavior and on how voters make decisions in hypothetical situations. But there may also be an adverse effect on behavior of elected officials, at least in short run.