I study a dynamic, multi-dimensional model of electoral competition in which two parties learn about voters' policy preferences by observing vote-shares from an election. What parties learn depends on the platforms voters are asked to evaluate. In equilibrium, politics is essentially two-dimensional, with one dimension capturing the conflict between the two parties, and another populist dimension describing the conflict between party elites and voters. If parties are farsighted, they prefer not to learn voters' preferences, and they moderate their platforms to slow the rate of learning. The reason why is that more precise information intensifies competition on the populist dimension of politics, which party leaders would rather cooperate to avoid. Even if there is only one policy issue, parties only prefer to learn if there is a large shock to voter preferences between elections. In this case parties choose more extreme platforms to speed the rate of learning.