Please visit the new home of Majikthise at bigthink.com/blogs/focal-point.

« Tipping the scales: desert and RE | Main | President Bush "living in the future" »

August 21, 2004

Ray Fair Election Equation

VOTE = 55.57 + .691*GROWTH - .775*INFLATION + .837*GOODNEWS

Ray Fair'selection equation is getting a lot of attention. The media hook is obvious: a distinguished scholar defies conventional wisdom to predict Bush landslide. Over the last few months Fair's predictions have been cited by such influential sources as the The New York Times, The New Yorker, MSNBC, and the NRO. Eugene Volokh has even commented the Fair coverage , so we know we've facing a phenomenon.

Unfortunately, the press has glossed over the reasoning behind Fair's predictions. The celebrated election equation is a linear regression on economic and electoral data for American presidential elections (1919-2000). The line that best fits the data is expressed as an equation which assigns different weights to a handful of economic and political variables. The model assigns weights the following factors: incumbency, economic growth and inflation during the election year, and "goodnews" (how many of the last 15 quarters saw growth >3.2%). Fair also claims that his model captures a causal relationship between the economy and electoral results. Allegedly, the model predicts because people vote to maximize their expected utility. Fair makes the further claim that voters calculate their expected utility based on a handful of current economic indicators and the assumption that they are better off under an incumbent with a reasonable economic track record.

Some question Fair's assumption that the economy will get a historically average amount of attention in 2004. But even if we set aside these Humean worries, Fair's model is seriously compromised. The unreported story is that bad economic news has steadily eroded Bush's projected lead. Even so, Fair steadfastly maintains that Bush has this in the bag (unless the model is deeply flawed). To satisfy readers that Bush's projected advantage is not due to overly optimistic projections for Q3 2004, he allows readers to plug their own data into his equation. The model projects a Bush victory even for dire Q3 predictions. Why? Largely because incumbents have historically enjoyed such a big advantage over challengers.

One reason for Bush's strong standing is that he gets a substantial head start by virtue of the incumbency effect. As calculated by Fair, any member of the incumbent presidential party is presumed to derive some benefit, but none more than a Republican whose party has been in office for only a single term. Based on Fair's analysis of elections since 1916, Democrats always have a slight disadvantage, and voters tend to tire of the incumbent party the longer it has occupied the White House. [MSNBC]

Fair is predicting one of the biggest landslides in American political history, a prediction sharply at odds with polling data. The latest polls give John Kerry a slight edge in a very close race.

When asked to explain the gap between his predictions and the best measures of political opinion, Fair dismisses polls as "flaky" [NYT]. Fair can't afford to be so cavalier. His model ought to fit the preliminary data, but it doesn't. The model assumes that voters deliberate rationally and that they assign the same weight to each variable as the model does. Fair believes that people vote their economic self interest based on the assumption that an incumbent with a decent economic track record will help them.

If Fair is right, voters already have most of the information that they will use to make up their minds in 2004. They know who the incumbent is. They know about the disappointing second quarter data. They know that Bush only has two "goodnews" quarters in his entire presidency.

Today's polls must already reflect Bush's incumbent's advantage and his economic performance to date. If Fair's model is correct, Bush should already enjoy a substantial lead. The fact that he's tied or trailing is already worrisome from Fair's point of view. Moreover, if the electorate were to shift its opinions to match Fair's projections, his theory couldn't explain why. Fair claims that his model is successful because it captures the rational weighting process that voters use to choose their candidates. But if voters do stampede for Bush in the waning days of the campaign, their behavior won't conform to Fair's weighted averages. We know this because the public already knows most of what Fair says they need to know to make up their minds. So, Fair is in a double bind. If his model gives the right answer, we will know it did so for the wrong reasons.

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c61e653ef00d834214ed853ef

Listed below are links to weblogs that reference Ray Fair Election Equation :

» Predicting the election from Pharyngula
Majikthise does a simple take-down of Ray Fair's magical election equation that predicts a Bush landslide victory. I'll just add a few tiny points. Correlation is not causation. The model gets tweaked after every election to fit the data.... [Read More]

Comments

Actually, Fair's model makes a lot of sense--and I'd say Hillary Clinton believes it since she didn't bother to run for the Dem nomination. The only thing you can criticize Fair for is not having enough data points; there have been 22 elections in the period he covered. Econometricians usually want a minimum of 30.

The model assumes that voters deliberate rationally and that they assign the same weight to each variable as the model does. Fair believes that people vote their economic self interest based on the assumption that an incumbent with a decent economic track record will help them.

No, no, no, no, no!

Fair isn't conjecturing how voters think! Instead, he's saying that voters behave in a certain way, X, that can be predicted by other factors (economy, incumbency, etc.)

People can have their own reasons. As some idiot once said, the heart has its reasons that reason cannot know.

But people don't actually say well I weight incumbency at .691X

sigh

No no no no... no wait, in fact, "yes."

"Fair isn't conjecturing how voters think!"

Let's go back to the original paper, shall we?

In Fair's words, "This paper has two main purposes. The first is to present a model of voting behavior that is general enough to incorporate what appear to be most of the theories of voting behavior in the recent literature..."

And then he goes on to elaborate what those two theories are, what he labels the "standard economic theory," a second theory that we might call "Kramer's theory," and a third theory he attributes to "Downs."

All three theories of voting behavior make explicit statements about what evidence a voter evaluates, or weighs.

No one, least of all "Majikthise," thinks Fair is claiming that people sit down and think, "what exact numerical weight should I give incumbency this year?" But to say that he isn't "conjecturing how voters think" is counter to pretty much everything he's written about the model; the whole thing is built up to test different theories about how voters think.

And anyway, the main problems with Fair's model are in the small number of datapoints, the large number of variables, the selection of variables so as to best fit the data (without any discussion of the "power" or "complexity" of the set of all possible models), and the updating of the model parameters after each election.

Taken together, those concerns add up (I think) and reduce this "equation" from its press-anointed status as "miraculous mathematical predictor" to "a standard exercise in model building with fairly simple pieces." Which is pretty... boring.

Chuck is correct. Fair's model is of behavior, not of rational deliberation.

I think its one failure--in 1992--shows its strength. Because it predicts how the electorate will choose to split its votes between Republicans and Democrats, when a wealthy third party candidate (with a personal animus toward one of the major party nominees) came along and self-financed his own campaign, that threw a monkey wrench into it.

Whether its boring or not is irrelevant. What matters is its predictive value, and that's been better than any other model of which I'm aware.

The comments to this entry are closed.