The American Political Science Association, meeting in Chicago, assembled six of the world’s top scholars in the field of electoral studies to forecast the US presidential election. Their verdict? Bush, probably.
The Panelists:
Helmut Norpoth (SUNY Stony Brook)
Brad Lockerbie (Univ. of Georgia)
Thomas Holbrook (UW Milwaukee)
Christopher Wlezien (Oxford) with Robert Erikson (Columbia)
Alan Abramowitz (Emory)
Michael Lewis-Beck (Univ. of Iowa) with Charles Tien (CUNY Hunter)
Moderator: James Campbell (SUNY Buffalo)
“This will not be a ‘girly man’ forecast,” says the moderator. And it isn’t. Helmut Norpoth towers over the podium, grips the edges with meaty hands, the microphone reaching only just above his navel. “Bush wins with 54.7 percent of the popular vote,” he thunders, in a distinct German accent.
This being academia, bold forecasts, and even accurate forecasts, aren’t all that matters. Norpoth is certainly bold. He made his forecast back in January and hasn’t changed it.
Bold, but is it justified? All the panelists make their predictions based on statistical models of voting behavior, back-tested on data from previous elections. But not just any data. If blue-eyed candidates always beat candidates from Texas, that won’t cut it. There must be a sound theoretical basis for each model’s explanation of why Americans vote the way they do.
“The best way to predict an election is with an election,” Norpoth says, defending his approach. Most of the models include some measure of the economy – reflecting, loosely, the idea that people vote their pocketbooks. (“It’s the economy, stupid.”) Norpoth’s model is based mostly on a sophisticated reading of the New Hampshire primary results. Which is why he calculated, back in January, a 95 percent chance of a Bush win. (Bush was nearly unchallenged in New Hampshire; Kerry had a bit of a fight.)
As the Republic Convention finishes, though, Norpoth is nervous. The polls aren’t where he’d think they’d be. Likewise for Brad Lockerbie. His model, based on surveys of consumer finances, has Bush winning with a landslide 57.6 percent.
Say these scholars are far wrong. What would that mean? First of all, it would be improbable. Perhaps surprisingly – given the unpredictable nature of US politics – most of the models produce a median forecasting error of only two percent of the popular vote in past elections. That’s pretty good. (Norpoth’s model accurately picks the winner in every presidential election since 1912 except one, 1960.)
Historically, pre-convention opinion polls have been about five percent off the final vote, on average. Post-convention polls are four percent off. And the final Gallup poll, on the eve of the election, is, on average, still 2.1 percent off the actual results. Many of the panelists’ models exceed this level of accuracy, using data released months before the vote.
Impressive, but that doesn’t make them error-free. The 2000 election was so close it wasn’t decided by the popular vote. (That job was, infamously, left to the Electoral College, Florida officials, and the Supreme Court.)
So not everything is predictable. But the assembled scholars believe most of the variance in the popular vote is explainable, based on sound theory.
Alan Abramowitz, for his part, plays up the theory. He’s annoyed by a recent New York Times Magazine article on economist Ray Fair, who has a model of his own. Abramowitz isn’t buying it. “It’s not an economic model. He tweaked it. He put a ‘time for a change’ variable in there, he doesn’t tell you about that. That’s my variable. I thought of it.”
‘Time for a change’ reflects a penalty that grows the longer a party holds the presidency. Bush the elder was a victim in 1992, because the Republicans at that point had been in office for 12 years. For Bush the younger, it’s no threat, so based on the economy and his approval ratings, Bush should win handily, says Abramowitz.
And there’s reason to believe this. Most of the models are new this year, but Abramowitz has been running his model since 1988, and he’s picked the correct popular vote winner every time.
Indeed, what makes this year’s election so interesting is that almost every model predicts an easy Bush victory. Thomas Holbrook’s approach based on approval ratings; Brad Lockerbie’s equation based on prospective personal finances; and Norpoth’s numbers from the primaries – all have the odds of a Bush win at more than 90 percent. Christopher Wlezien and Robert Erikson are only slightly less certain. A 75 percent chance, they say.
In other words, this shouldn’t be a repeat of 2000 – where models that were right about the vote were wrong about the outcome. If decades of history are any guide, Bush should be far ahead. That he isn’t is puzzling. A professor in the audience stands up and shouts: “if it wasn’t for this unilateral war Bush would be winning by a landslide!”
Or maybe there’s another explanation. Michael Lewis-Beck is a respected scholar in the field, but his 2000 model was far off. So he scrapped it and, with Charles Tien, created a new version for 2004.
This new model is a little odd. He’s using a standard economy measure, but then adjusting it by one-half if there’s a ruling party candidate who’s not the incumbent. And then another adjustment in cases where the incumbent “doesn’t like” the candidate from his party. By this point, Abramowitz, the stickler for theory, is clearly fidgeting.
By the time Lewis-Beck adds in his fifth variable, Abramowitz is rolling his eyes. But Lewis-Beck’s model is the most accurate, on past data, of any of them – an R-squared of .96, for statistics buffs.
And for this year’s election, because of dismal jobs growth, the Lewis-Beck model predicts a close race. Essentially a tie in the popular vote. Which means it could go either way. Indeed, he pulls out a chart which, he claims, shows that the race will go the Electoral College and then resolve in Kerry’s favor. By this point the other panelists, who are overwhelmingly Democrats and no doubt want to believe him, are nonetheless voicing loud theoretical objections.
It seems that devotion to theory trumps party loyalty. The moderator, the lone Republican on the panel, chuckles into the open mike. He says: “face it, it’s time the Democrats took one for science!”
This article was originally published on Countryrisk.com before I sold the site to Roubini Global Economics.