Candidate Competition and Voter Learning in Sequential Primary Elections: Theory and Evidence

We develop a model of sequential presidential primaries in which several horizontally and vertically differentiated candidates compete against each other. Voters are incompletely informed about candidate valence and learn over time from election results in previous districts. We analyze the effects of learning about candidate quality, and the effects of candidate withdrawal on the vote shares, using data from the 2000-2012 Democratic and Republican presidential primaries. Consistent with the predictions of the model, the withdrawal of a candidate has a bigger effect on the vote shares of candidates in the same political position, vote variability declines over time in a pattern consistent with learning, and a tilt of the electorate towards a particular political position disproportionately increases the vote shares of the weak candidates espousing that position (relative to the strong candidates in that position). JEL Classification Numbers: D72, D60.