You know that multiple regression analysis is not ideal for dichotomous dependent variables, such as vote for president (1=Bill Clinton, 2=Young Bush) in the 2000 presidential election. The preferred alternative method is binomial logistic regression.

Given a polychotomous dependent variable, such as the vote for president (1=Clinton, 2= Perot, and 3= Old Bush) in the 1992 presidential election, multiple regression is completely inapplicable. When the dependent variable has more than two categories, then multinomial logistic regression could be used.

But there is another alternative for analyzing polychotomous dependent variables, discriminant analysis. This method requires some stronger assumptions about the data, but computer programs for discriminant analysis offer more informative data displays. These are especially helpful when your cases are not anonymous respondents but meaningful units like nations, states, or cities.

My presentation will compare discriminant analysis with multinomial logistic regression in explaining the 1992 vote for president. Then I will illustrate the technique in greater depth for the analysis of comparative political parties.