The politics of policy reform: Experimental evidence from Liberia

Public service reform often entails broad benefits for society and concentrated costs for interest groups. These groups’ political responses determine whether electoral incentives exist to improve public services. This paper examines the electoral effects of a randomized Liberian school reform which increased student learning but antagonized teachers. On average, this policy reduced the incumbent party’s presidential vote share by 3 percentage points (10%). It had no significant impact on legislative races, consistent with correct attribution by voters; information experiments with candidates and voters further suggest a well-informed electorate. The policy also reduced teachers’ job satisfaction by 0.18s and lowered their participation in political activity by 0.22s. I use the policy’s pairwise randomization to study how its electoral effects varied across the (orthogonal) distributions of treatment effects on student learning and teacher political activity. The policy increased vote share more where it caused greater student learning, and reduced vote share more where it caused greater political disengagement of teachers. (Treatment effects on student learning and teacher political involvement were uncorrelated.) Back-of-the-envelope calculations suggest that the policy could have won votes on net if the floor on learning effects had been the 27th percentile, and the floor on teacher political involvement effects had been the 30th percentile. This paper shows empirically that electoral rewards correlate with the size of public service improvements, but that politically feasible reforms must balance voter rewards with the costs of alienating interest groups.