SYMPOSIUM - THE DIRECTIONAL THEORY OF ISSUE VOTING .2. DIRECTIONAL AND PROXIMITY MODELS OF VOTER UTILITY AND CHOICE - A NEW SYNTHESIS AND AN ILLUSTRATIVE TEST OF COMPETING MODELS
S. Merrill et B. Grofman, SYMPOSIUM - THE DIRECTIONAL THEORY OF ISSUE VOTING .2. DIRECTIONAL AND PROXIMITY MODELS OF VOTER UTILITY AND CHOICE - A NEW SYNTHESIS AND AN ILLUSTRATIVE TEST OF COMPETING MODELS, Journal of theoretical politics, 9(1), 1997, pp. 25-48
Under the assumption that a voter's utility is maximized when s/he cho
oses the candidate/party that is closest to the voter's own most prefe
rred set of policies, the Downsian proximity model of voter choice has
become the standard method for modeling the linkage between the polic
y preferences of voters and the policy positions of candidates. Altern
ative spatial models of voter utility and voter choice - based on dire
ctional criteria - have been proposed by Matthews and by Rabinowitz an
d Macdonald. The relative fit of models can best be addressed by nesti
ng seemingly disparate models in a unified statistical framework which
embodies proximity, directional and intensity components and which ha
s each of the 'pure' models as a special case. Theory suggests the nee
d to distinguish the ability to predict distinct shapes of voter utili
ty functions from the ability to predict voter choice. Using data on t
he voter utility functions for major candidates for the US presidency
during the period 1980-92, we show that the best fit incorporates all
three components with intensity significantly more prominent for chall
engers while the Matthews directional model - which de-emphasizes inte
nsity - is preferred to the Rabinowitz/Macdonald version for incumbent
s. Differing utility functions for incumbent and challenger imply that
the former should seek the center while the latter espouses strong st
ands. We show that the marked preference for the model obtained by Rab
inowitz and Macdonald using mean issue placements of the candidates is
greatly reduced with voter-specific candidate placements, even after
adjusting for projection. However, when we shift from utility function
s to voter choice, we find no significant difference in predictive pow
er between the proximity and directional models.