The latest issue of PS, a political science journal put out by the American Political Science Association, features a series of articles forecasting the 2004 presidential vote using various models that have been developed. Though they all illustrate that political scientists are much better at predicting the past than the future, it's useful to take a look at how well they did.
(Note: Bear in mind that many of these models have been shaped by how well they predict past elections, a type of specification search that can be especially pathological given how few data points we have. That's why it's important to point out that the 2004 results don't necessarily demonstrate that certain models are better or worse than others. Revising your model in response to each new data point is a bad idea.)
Here's a summary of the contenders from an APSA press release (236K PDF):
In the earliest completed forecast, made in late January 2004, Helmut Norpoth (Stony Brook University) finds that the true determinant of a presidential victory lies in the performance of presidential candidates in primary elections. His model examines candidate support in presidential primaries since 1912, accurately calling the general election winner in every race except in 1960—the century's closest presidential contest. Norpoth’s forecast makes Bush a 20-1 favorite, predicting a 54.7% to 45.3% Bush victory.
In late May 2004, Brad Lockerbie (University of Georgia) used two variables that are decided well in advance of the presidential conventions to forecast the presidential election: the amount of time a party has controlled the White House, and voters’ expectations concerning their financial well-being over the course of the next year. Lockerbie predicts that President Bush will garner 57.6% of the two-party popular vote en route to victory in the 2004 presidential election.
Alan I. Abramowitz's (Emory University) model, completed at the end of July 2004, is based on the assumption that a presidential election is fundamentally a referendum on the performance of the incumbent president. Abramowitz’s model employs three variables: the incumbent president’s approval rating in the final Gallup Poll in June, the change in real GDP during the first two quarters of the election year, and the amount of time that the incumbent party has held the White House. He predicts that President Bush will collect 53.7% of the 2004 presidential election two-party popular vote.
Christopher Wlezien (Oxford University) and Robert S. Erikson's (Columbia University) forecast model analyzes a combination of leading economic indicators, income growth, presidential approval ratings, and trial-heat polls to predict the election’s final outcome. Completed in late August just in advance of the Republican convention, their model predicts that Bush will win between 51.7% and 52.9% of the two-party presidential vote.
Michael Lewis-Beck (University of Iowa) and Charles Tien (Hunter College, CUNY) weigh the president’s low popularity and the net loss in jobs during his term with the powers of incumbency and economic growth to break from the ranks of the other forecasts and predict a razor-thin margin of victory for Senator Kerry. Completed in late August 2004, the Lewis-Beck/Tien model predicts that Bush will gain 49.9% of the two-party vote and 241 Electoral College votes.
Thomas M. Holbrook's (University of Wisconsin-Milwaukee) forecast model integrates economic news and trends, voter level of satisfaction, and personal finances, weighed with the power of incumbency and presidential popularity. Holbrook’s model, completed in late August 2004, predicts President Bush will receive 54.5% of the two-party presidential vote.
Symposium editor James E. Campbell's (University at Buffalo, SUNY) forecast model integrates the incumbent’s trial-heat Labor Day Gallup numbers and the real growth in the GDP in the second quarter of the election year. Completed in early September 2004, the Campbell model predicts that President Bush will win 53.8% of the vote.
Their median prediction was that Bush would receive 53.8% of the two-party vote, and according to the totals on CNN.com this morning he actually received 51.5%, which is pretty close. Despite the predictions from many pundits, Iraq didn't appear to drastically change the fundamentals of this election -- presidential approval and the state of the economy (the core of most models) appear to have again played a decisive role, though we'll have to wait for all the post-mortems to see for sure.