Testing Formal Models of Political Competition with Candidate-Choice Experiments

Abstract

Formal models of political competition have faced persistent skepticism about their assumptions and predictions. This study addresses these criticisms using data from candidate-choice conjoint experiments, which uniquely allow candidates to adopt a full range of policy positions, including match-ups rarely observed in real-world elections. Using over 20,000 candidate-choice responses, we estimate the ``empirical best response function'' of political candidates in response to how respondents vote across the full range of positions on social and economic policy dimensions. Our results reveal conditional support for the convergence prediction of the median voter theorem. We also find that voters often, though not always, have single-peaked preferences and vote according to those preferences. We also empirically test key features of formal models, including valence advantages, multidimensional politics, and turnout effects, shedding new light on their validity. By demonstrating how conjoint experiments can overcome the limitations of observational data, this study underscores the value of experimental approaches in advancing formal theory and bridging the gap between theoretical predictions and empirical realities.