Modeling the Future: Lessons from the Gore Forecast

Extract

Our 2000 presidential election fore cast was considerably off the mark. We predicted Gore would receive 56.9% of the two-party popular vote. Gore did get the majority of this popular vote (about 50.2%), as we predicted, but his total fell many points short of our forecast. Does such a loarge error (of 6.7 points) invalidate our model? We are hesitant to say so, since we forecast Clinton's victory in 1996 almost exactly using the same model, and more closely than other political scientists. While the Gore result has some outlier characteristics, incorporating it into the model does not really budge the structural coefficients. Put another way, introducting the 2000 data does not alter the magnitude of the regression estimates for the three independent variables: GNP change, popularity, peace and prosperity. As shown in the second column of Table 1, the coefficients from the 1952–2000 estimates are, respectively, 1.81, .14, and .14, while the coefficients from the 1952–1996 estimates are, respectively, 1.83, .16, .14.