Essays on parties and elections in federal systems

This dissertation contains three essays on political parties and elections in federal systems. The first essay develops a formal model for the conditions under which political parties contesting elections at the national and subnational levels choose to affiliate under a common party label. Vertically integrated parties, which contest elections at both levels, gain organizational efficiencies at the expense of competitive flexibility. State and federal parties choose to affiliate unless two conditions are met: (1) the distribution of voter preferences within a state differs significantly from the distribution of preferences at the federal level, and (2) the scope for organizational efficiencies is small. Consistent with the model, political parties in Australia and the United States are vertically integrated. Parties in Canada are less likely to integrate vertically, particularly in provinces with distinctive political preferences.;The second essay develops two signaling models that demonstrate how voters can use subnational elections to send messages to parties in government at the federal level. In the first model, voters reveal private evaluations of government competence. In the second model, voters reveal private evaluations of government policy. Signaling in subnational elections is sustainable under two conditions: (1) subnational elections are sufficiently important to render the signal credible, and (2) the cost of signaling by voting for less-preferred subnational parties does not exceed the benefit of revealing information to parties in federal government. The implications of these signaling models are contrasted to balancing theories of voting in subnational elections.;The final essay develops an empirical model to distinguish between voting behavior predicted by balancing and signaling theories. Existing empirical approaches are subject to bias and fail to allow for signaling behavior. The Bayesian approach used in this essay is less vulnerable to bias and enables estimation of the probability that a signal was sent in any particular election. When applied to election results from Germany and Australia, it shows that the pattern of election results is more consistent with a signaling explanation for losses by federal governing parties. Evidence from Canada is more ambiguous, with some results consistent with signaling and others with balancing.