Heightening comparativists' concern for model choice: Voting behavior in great britain and the netherlands

Theory: As the research methodology more closely approximates the causal process being analyzed, the inferences and predictions derived from that methodology will better represent actual behavior. Statistical models were specified on the basis of accepted theories of voting behavior and political cleavages in the Netherlands and Great Britain. Hypotheses: We hypothesize that multinomial models provide a more accurate characterization of voting behavior in countries with more than two political parties competing for votes. Method: Binomial logit, multinomial logit, and nested multinomial logit models of voting behavior are estimated on Dutch and British National Election Study data. Results: Compared to binomial methods, we find that multinomial models of voting behavior produce results that are more congruent with accepted theories of Dutch and British politics.