Election Forecasting for Turbulent Times

Abstract

Election forecasters face increasing turbulence in their relevant environments, making predictions more uncertain, or at least apparently so. For US presidential contests, economic performance and candidate profiles are central variables in most statistical models. These variables have exhibited large swings recently. Before the 2008 US presidential election, the economy fell into a Great Recession, and the candidate of one of the two major parties was, for the first time, a black man. These unprecedented conditions were trumpeted in the media, with heightened frenzy over the “horse race” question of who was going to win the White House. In the press, many forecasts appeared, taking different forms—polls, models, markets, pundits, to name some—offering a broader range of possible outcomes than ever before. Just looking at the predictions of the statistical modelers alone, we find that for 2008 many teams offered estimates of the incumbent (Republican) vote, ranging over an 11 percentage point spread. At one extreme, Lockerbie (2008) forecast 41.8% while at the other extreme Campbell (2008) forecast 52.7%. Of course, other methodologists offered their own, different, forecasts. The media, in its various forms, added to the hyperbole, aggressively reporting different forecasts on an almost daily basis.