Additive and Multiplicative Models of the Voting Universe: The Case of Pennsylvania: 1960–1968

Of all the fields of political science where quantitative methods have been developed over the past generation, probably the one where scholarly understanding has been most enriched has been that of mass voting behavior. But while we know vastly more about this behavior on the individual and aggregate level than we did a quarter-century ago, there are still large territories on the map which are blank, or in which exploration has only very recently begun. There remain a number of doubtful areas in which issues of methodology and of substantive interpretation are still very much open to systematic inquiry.One such area is that associated with the interrelation of socio-economic correlates of the vote. That is, there is a real question as to whether such independent or predisposing variables should be conceptualized as making mutually independent or, alternatively, interdependent contributions to the prediction of voting patterns. The normal practice in research involving multiple correlation of aggregate voting behavior with a set of independent variables has been to assume implicitly that the relationship of these variables is additive (i.e., non-interactive) and that the appropriate theoretical representation is of the general form y = b + m1x1 + m2x2 … + mnxn. Such an assumption seems plausible so far as individual voting for American major parties and their candidates is concerned. Thus, for example, the authors of the MIT 1960 simulation study found strong evidence that predispositional factors summate, i.e., are indeed additive in character.