Learning and Coordination in the Presidential Primary System

To analyze the advantages and disadvantages of the U.S. presidential primary system, we develop a model in which candidates with di erent policy positions and qualities compete for the nomination, and voters are uncertain about the candidates’ valence. This setup generates two problems that are a ected by the temporal organization of primaries: First, voters in late-voting states can use the results in early elections to update on candidates’ valences. Second, candidates who o er the same policy split the voters with a policy preference for their position, so that a candidate with a di erent position may win even if he is not the Condorcet winner; this problem is particularly prevalent in simultaneous elections. The advantage of sequential voting is to minimize vote-splitting in late districts; however, coordination may occur on the wrong candidate. We structurally estimate the model using the 2008 Democratic presidential primaries, and find that allowing for voter learning and imperfect substitutability of candidates across di erent political positions is quantitatively important. We use the parameter estimates to conduct policy experiments such as replacing the current system with a simultaneous system or other proposed systems. Our results indicate that the current system is preferable to simultaneous voting. Even better is a sequential system that avoids bunching of elections (“Super-Tuesday”) and the possibility of too early dropout of unsuccessful candidates.