State-Level Public Opinion Polls as Predictors of Presidential Election Results

Abstract

Forecasting presidential elections results with polls prior to the election has become an important activity and test of voting theory. Such efforts are hampered by the few number of observations, however. In 1996, state-level polls for most of the states have become available. A model to account for state-level presidential election results is built using these polls and other commonly used variables. The model performs quite well, suggesting that these polls, which are mostly commercial, can profitably be employed in future modeling and perhaps forecasting efforts. However, the questions used in these polls, like trial heats and positive-negative name recognition, are not well understood and thus should be examined more closely with an eye on theory building and conceptualization.