WARNING--GLARING SELF-REFERENCE VIOLATION:

Election Forecasts: A Numbers Game (Charlie Cook, August 29, 2004)

The vast majority of political scientists and economists who forecast elections based on computer models will be presenting their papers at the annual meeting of the American Political Science Association this week in Chicago, and they are projecting a Bush victory over Sen. John Kerry -- in a landslide, some say. Other analysts, myself included, think Bush faces an uphill
struggle.

The boldest prediction is from Yale University economist Ray Fair, the dean of the election-forecasting academicians whose model projects that Bush will get a whopping 57.48 percent of the major-party (combined Democratic and Republican, no independent) vote. Fair's model is based entirely on economics -- the real gross domestic product growth rate and inflation -- and it carries, he says, a standard error rate of 2.4 percent in either direction. In his July 31 "Note to the Media" on his Web site, Fair cautions that a change in economic data could affect his forecast but that "no realistic economic values can bring the predicted
vote share to even about 53 percent."

Another bullish Bush prognosticator is political scientist Helmut Norpoth of the State University of New York (Stony Brook), who gives 20-1 odds that Bush will be re-elected. His model shows a Republican two-party-vote victory of 54.7 percent to 45.3 percent. Norpoth's model focuses on how well the nominees performed in their respective party's primaries, on long-term partisan trends, and on how long the incumbent's party has held the presidency.

Slightly more conservative is a model developed by Oxford University's Christopher Wlezien and Columbia University's Robert S. Erikson that projects a 52.8 percent Bush share of the two-party vote. They say that Bush has slightly better than a two-out-of-three chance of re-election.

The Wlezien/Erikson model relies upon the index of leading economic indicators, as well as on Gallup job approval ratings and trial heat data.

A model developed by the University of Iowa's Michael Lewis-Beck and Hunter College's Charles Tien shows a two-party-vote edge for Bush of just 51 percent to 49 percent. They say that "the narrow difference makes the race too close to call." The "Jobs Model," as the authors call it, focuses on job creation and presidential job approval.

At Emory University, Alan Abramowitz's "Time for a Change" model uses three variables: the incumbent's job approval in June, the change in real GDP during the first two quarters of the election year, and the amount of time -- whether one term or longer -- the incumbent party has held the White House. While Abramowitz's model projects a Bush win of 53.7 percent of the two-major-party vote, he is doubtful about whether it will work this time; he concludes that "it seems unlikely that [Bush] will receive anything close to 53.7 percent."

I had a psychology professor in college who had served on a bunch of committees with my grandfather, who'd been on the Board of Trustees for twenty years. Taking attendance on the first day of class he stopped at my name and mentioned what a great man the elder Orrin had been. When we went to his office at the end of the semester, several classmates stood dumbfounded as he announced that his first gradescale had given me a C+ but he "knew what I expected" so I redid the scale to give me a B-. Now, to begin with, a C+ would have raised my GPA and been perfectly satisfactory, even undeserved, but in the second place, Charlie Cook seems to be forecasting this race by using the same "what I expected" theory. It seems less likely to work in a presidential election than in college.

Posted by Orrin Judd at August 30, 2004 9:37 AM