Why do voters often elect governments in which several parties share power? To answer that question, we propose a dynamic model in which in every period, a voter elects a unified government---in which a single party controls policy making---or a divided government---in which the agenda and veto powers are held by different parties. The elected government then observes the state of nature and decides which of two policy dimensions to reform, if any. On the consensual policy dimension, both parties and the voter have congruent preferences (e.g., infrastructure), whereas on the divisive dimension (e.g., taxation), they disagree with positive probability. Crucially, the government has limited legislative time in that in any period it can change only one dimension of the status quo, which is inherited from the previous government. We show that a divided government leads more often to gridlock. However, the voter elects a divided government, because she fears that a unified government will waste its legislative time on divisive reforms in line with its own ideology, which may crowd out beneficial consensual reforms. (with Wiola Dziuda)