Political Stake and Policy Experimentation

Abstract

Policy experimentation is an increasingly common practice for institutional reforms. Meanwhile, policy makers may be politically motivated to bias the information revealed by experiments. This paper develops a model of Bayesian persuasion to study how political stake shapes policy experimentation. The model shows that the optimal experiment almost never elicits information perfectly when the policy involves some political stake. The optimal experiment is conducive to type-I error (over-reform) when the stake is large and type-II error (under-reform) when the stakes are small. Experimentation is most likely to enhance the probability of reform when the political stake is distributed within an intermediate range. The paper examines this argument empirically against case studies in Uganda, Kenya, the Soviet Union, and Vietnam, as well as through an investigation of the reform of the Household Responsibility System in Chinese counties in the 1980s.