Voting theory generally concludes that, in first-past-the-post elections, all votes should go to effective candidates (Duverger's Law), and parties should adopt a similar platform (median voter theorem). However, such predictions are not always met in practice. We show why divergence and vote dispersion is a natural outcome when (a) parties are opportunistic, (b) there is uncertainty on the position of the median voter and (c) elections are repeated. “Voting for losers” increases the informational content of elections, and may induce mainstream parties to relocate towards extremists. As a result, to maximize their probability of being elected, they do not adopt median platforms, but instead diverge to a certain extent. (JEL: D72, D81)