Abstract

This paper examines two models used in survey research to explain voting behavior. Although the models rely on the same data they make radically different predictions about the political future. Nevertheless, both models may be more or less correct. The models represent interacting systems and it may be impossible to get a super model of the interactions between their elements. In the natural sciences causal relationships between the elements of interacting models can often be ignored. Because voting behavior models describe phenomena that are roughly the same size, the reciprocal relationships between elements of different models severely restrict the predictive power of voting behavior models. Certain analogies, and disanalogies, between the use of models in natural and social science explain why the social sciences cannot predict many of the events they are able to explain.