This paper provides a political economic analysis of policy learning. We formally model the decision of an incumbent government to run (or allow) a program evaluation prior to implementing a policy reform. Focusing on randomized controlled trials (RCTs), we argue that the bene fits of program evaluation for the incumbent are greater in more politically polarized polities, where the opposition's ideal policies are further away in policy space. Moreover, the effect of polarization on the probability to host RCTs is stronger in more politically competitive environments since the incumbent is more likely to be replaced by the opposition. We empirically examine the derived hypotheses with two methods. First, we provide a detailed case study of the Liberian government's decision to commission a third-party RCT evaluation of a proposed primary school privatization reform, where we focus on the wide gap between the incumbent and opposition on their preferred education policies. Second, we have compiled a unique cross-country panel data set on RCTs in development policy since 1996, with which we demonstrate that RCTs are more likely to occur in polarized societies, and that the effect is amplified by the degree of political competition.