We develop a spatial model which embeds regional- and national-level elections in a single theoretical framework, and links the two via the assumptions that: a.) voters punish parties for intraparty platform disunity, and b.) regional candidates value having their party in national office. Although candidates are purely office-seeking, polarization occurs when platforms are captured by regional extremists. Interestingly, it is in `decentralized' systems that national politicians can escape this centrifugal force, and adopt the national median as their platform. Conversely, party capture only occurs in sufficiently `centralized' systems, where pressures for party unity generate incentives for regional pandering. In centralized systems, platform polarization depends in surprising and non-monotonic ways on geographic polarization in the electorate: decreasing geographic polarization at times leads to increased polarization of political parties. In sufficiently centralized systems, a form of asymmetric party system polarization may emerge despite the presence of a moderate electorate.