How far in advance can one predict the next presidential election? Forecasting as late as election eve is trivially easy, at least since the advent of polling. Forecasting as early as two or four years in advance is virtually impossible (see, e.g., Norpoth 1995). It is at some time between the previous midterm election and election day that electoral expectations come into focus.

The evolving expectations of the 1996 election cycle offer a familiar example. A year ago, Clinton was widely considered unelectable. As of this writing (November 1995), Clinton is perceived as a mild favorite for 1996. By the time you read this essay in spring 1996, a new consensus may emerge. Expectations will harden by election day, when the verdict is not likely to surprise (Gelman and King 1993).

“Trial heat” polls are one way to predict elections (see, e.g., Campbell and Wink 1990), but they are notoriously unreliable when conducted more than a year in advance. A year in advance of the two Reagan elections, for instance, Reagan trailed Carter in 1979 and Mondale in 1983 in the polls. We can, however, examine other early indicators of election “fundamentals.” These indicators include the results of another poll question that is asked regularly throughout the electoral cycle: the president's approval rating in the Gallup poll. Various earlier models of the presidential vote have incorporated a measure of approval (Sigelman 1979; Brody and Sigelman 1983; Abramowitz 1988; Lewis-Beck and Rice 1992).