Existing models of multicandidate spatial competition with probabilistic vo
ting typically predict a high degree of policy convergence, yet in actual e
lections candidates advocate quite divergent sets of policies. What account
s for this disparity between theory and empirical observation? I introduce
two variations on the basic probabilistic vote model which may account for
candidate policy divergence: 1) a model which incorporates candidate-specif
ic variables, so that candidates may enjoy nonpolicy-related electoral adva
ntages (or disadvantages); 2) a model which allows nonzero correlations bet
ween the random terms associated with voters' candidate utilities, thereby
capturing situations where voters view two or more candidates as similar on
nonpolicy grounds. I report candidate equilibrium analyses for each model,
which show far greater policy divergence than exists under the standard pr
obabilistic vote model. I then analyze the strategic logic which underlies
these results.