We develop a dynamic repeated election model in which citizen candidates are distinguished by both their ideology and valence. Voters observe an incumbent's valence and policy choices but only know the challenger's party. Our model provides a rich set of novel results. In contrast to existing predictions from static models, we prove that dynamic considerations make higher-valence incumbents more likely to compromise and win re-election, even though they compromise to more extreme policies. Consequently, we find that the correlation between valence and extremist policies rises with office-holder seniority. This result may help explain previous empirical findings. Despite this result, we establish that the whole electorate gains from improvements in the distribution of valences. In contrast, fixing average valence, the greater dispersion in valence associated with a high-valence political elite always benefits the median voter but can harm a majority of voters when voters are sufficiently risk averse. We then consider interest groups (IGs) or activists who search for candidates with better skills. We derive a complete theoretical explanation for the intuitive conjectures that policies are more extreme when IGs and activists have more extreme ideologies, and that such extremism reduces the welfare of all voters. Copyright 2011, Oxford University Press.
