We consider a political contest where data is common knowledge and candidates compete by announcing ideologies (statistical descriptions of the data-generating process). The election winner will make policy decisions according to the announced ideology. Voters form a prior on the data-generating process on the basis of the announced ideologies. They choose the candidate who maximizes their expected payoff. We prove that the ideology selected by candidates in equilibrium maximizes the product of three factors that can be interpreted as: fit (likelihood of the observed data within the ideology), simplicity (a penalty on the ability of the ideology to explain a lot of data realizations), and fear (the ability of the ideology to evoke bad outcomes of competing ideologies).