In real-life elections, voters do not have full information over the policy platforms proposed by political parties. Instead, they form (imprecise) beliefs. I propose a new model of partisan competition to represent the interaction of these beliefs with platform selection. Both parties gain more from appealing to the voters with more precise beliefs over their platform. Minority candidates viewed with less precision overall gain relatively more from outliers. Therefore, the Median Voter Theorem is recovered if and only if voters' policy preferences are uncorrelated with the precision of their beliefs about each candidate, and preferences are distributed symmetrically. Otherwise, even election-motivated parties diverge away from each other. As the population becomes polarized in how they form beliefs about politics, they will become polarized on political grounds as well, providing a new explanation for recent political polarization in the United States which, under reasonable assumptions, is more in line with the stylized facts than models with perfect observability.