Paul Beard writes to ask what the economists and social scientists at Crooked Timber think of this interview with economist Ray C. Fair. His model that is predicting that Bush will get 57.5% of the 2-party votes.
Well, the economists and social scientists at Crooked Timber are in bed, so I’ll try. Here’s the model and relevant webpage. Fair is a Yale professor who has been tweaking his model, and successfully publishing papers on it, since 1978. I am a lowly uncredentialed blogger who has spent less than an hour looking at said papers. If I were a betting man, I’d have to bet that Fair is right and I’m wrong when I raise a criticism. If I’m lucky, maybe he’ll respond to my email and humiliate me in comments.
Having said that, a few comments:
1. The model is stunningly simple. There are only two inputs: inflation and two measures of real per capita GDP growth. And the output is not the expected vote percentage of the incumbent, but of the Republican candidate. (There’s also an incumbency variable implicit in the equation he’s posted for the 2004 election.)
2. If I wanted to put this model into words, it would sound something like: “Republican presidential candidates will get more of the vote when real GDP growth is strong and inflation is relatively low.” This has a few counterintuitive implications:
Good GDP growth and low inflation hurts Democratic candidates. (Even when they’re the incumbent? I’m not sure which effect is more powerful.)
Unemployment, polls, fundraising, immigration, wars, etc., don’t make much difference in how people vote for President.
If you share the reasonable belief that the President doesn’t have much control over GDP or inflation, the model implies that the candidates, even sitting Presidents, can’t do much about the percentage of the vote that they obtain.
Now, “counterintuitive” certainly doesn’t mean “wrong”, but I’d need some convincing.
3. The input GOODNEWS is rather delicately specified; it’s defined as “number of quarters in the first 15 quarters of the Bush administration in which the growth rate of real per capita GDP is greater than 3.2 percent at an annual rate”. I can’t imagine that there’s a theoretical justification for it, as it changes in different versions of the model. It seems that it’s simply a number that most accurately retroactively predicts the old elections.
Is it an overspecified model, good at describing past history but not at making predictions? It seems awfully unlikely that I’d be the one to discover this. We’ll know more in November.
Professor Fair updates his model after every election to better accomodate the new data point. I’d be interested in seeing how well each iteration predicted the elections since 1978 before they were retroactively tweaked. It did pretty well in 2000.
UPDATE: Evan Roberts at Coffee Grounds points out that I’ve missed something important. The party of the incumbent is a variable; benefits of a good economy don’t automatically benefit Republicans.
{ 19 comments }
Mark Byron 08.18.04 at 12:56 pm
The logic that I would follow out of things is that when things are going good, people are willing to let the economy roll and not have government solve things, leaning to wards the laissez-faire GOP approach.
When the economy isn’t growing and inflation’s nasty, government intervention starts to look good.
Matt 08.18.04 at 1:32 pm
I think the relevant saying is “It’s hard to make predictions, especially about the future.” Predicting the past with a simple model is not really that much of a trick.
Ken Houghton 08.18.04 at 2:27 pm
People are reporting that it failed miserably in 1992. Which also implies that it should have been dubious in 1980–the two times in recent history when improvements were on the horizon but not yet reflected in the marketplace (replacing Miller with Volcker in Carter’s case, though any sentient life form would have worked).
Now, of course, the “improvements” are there, but–again–the marketplace doesn’t reflect them.
Fair apparently forgets that people are not (yet) automatically disenfracnhised when they are downsized.
Ken Houghton 08.18.04 at 2:34 pm
Update: Fair confirms that there was a “large error” in 1992 in making his 1996 update.
Chris in Boston 08.18.04 at 2:44 pm
Admittedly econometric modeling is over my head. I still can’t help but feel that this is positivism pure and simple, an attempt to fetishize empirical observation through number crunching without being able to identify the true causes for people’s voting behavior. In this instance, it’s taking a fairly agreed-upon premise (that the nation’s economic health influences voter behavior, even when they’re unaware of that as a cause for a given vote), and ignores any other causes concscious or unconscious that may underlie voting.
I’m curious to hear why Fair’s model is better than the Super Bowl model of predicting elections:
http://www.slate.com/id/83375/sidebar/83381/
Erik 08.18.04 at 2:45 pm
The model works quite well historically. As a predictive vehicle, it can probably not outperform simply looking at subjective appraisals of how well the president is doing. Moreover, the model cannot incorporate the extent to which the war plays a role in the current election. I have a working paper with Paul Brewer that has some evidence that it is mostly evaluations of the war that have been driving presidential approval, although there is some evidence that the economy is becoming more important as the election nears (not all the latter evidence is in the paper yet).
John Isbell 08.18.04 at 3:17 pm
SO for 2000, it predicted the SCOTUS intervention? Now that’s a cool model.
Matt Weiner 08.18.04 at 4:49 pm
The embarrassing thing is that the model Chris linked made an absolutely dead-on prediction for the 2000 vote.
In this it beats my favorite model: The winner of the election will be the one who gets the most officially tabulated votes. Worked for every election in the 20th century, but is 0 for 1 in the 21st. It also has the problem that it’s difficult to gather the data necessary to make the prediction.
(Preemptive strikes against nitpicks: The Super Bowl model predicts share of the two-party vote, so 50.1 for Gore is just about right; and there’s a fresh can of whoop-ass sitting on my desk waiting for the first person to claim that 2000 was in the 20th century.)
Dan 08.18.04 at 6:21 pm
From the Slate article featuring the Super Bowl model:
“Yale economist Ray Fair picked the wrong winner in 1996 and 1992, even though he’s been refining his model since at least 1976 (he got that one wrong, too). Reading about forecasters’ track records is like reading Money magazine’s stock and mutual fund picks. They remind you of their successes but seldom mention their failures.”
Andrew Edwards 08.18.04 at 7:02 pm
“Yale economist Ray Fair picked the wrong winner in 1996 and 1992, even though he’s been refining his model since at least 1976 (he got that one wrong, too).”
Hmmm. I have a prediction model for every election from 1976 onward. It gets it right in every election except ’76, ’92, and ’96. Ready?
“Republicans will always win”.
That seems to be Fair’s model. Either that, or he’s also missed intervening elections, and the above model is better than his.
Andrew Edwards 08.18.04 at 7:03 pm
“Yale economist Ray Fair picked the wrong winner in 1996 and 1992, even though he’s been refining his model since at least 1976 (he got that one wrong, too).”
Hmmm. I have a prediction model for every election from 1976 onward. It gets it right in every election except ’76, ’92, and ’96. Ready?
“Republicans will always win”.
That seems to be Fair’s model. Either that, or he’s also missed intervening elections, and the above model is better than his.
Andrew Edwards 08.18.04 at 7:05 pm
Oops. Sorry.
Matthew 08.18.04 at 7:10 pm
Would employment growth, or the unemployoment rate be a better measure than GDP growth?
If (as I keep reading, though I don’t know how true it is) then this has been a recovery in which fast GDP growth has outpaced employment (i.e a jobless recovery in the parlance) then perhaps that could explain the difference betwwen the forecast result and what the polls tell us.
Matt Weiner 08.18.04 at 8:35 pm
Andrew, not quite–he predicts the popular vote; so in 2000 his model predicted (correctly) that the Democrat would win. It predicted Gore with 50.8% of the two-party vote; Gore got 50.3%.
Still, he lost to the Super Bowl model, which predicted 50.1%.
son volt 08.19.04 at 2:24 am
I grudgingly suspect Fair is onto something. For one thing, that presidential elections track fairly well with macro-economic performance needn’t suggest that people don’t have other concerns. Rather, the model would suggest that those other concerns are over-determined by manipulative, hot-button advertising paid for by the sort of actors who do pay close attention to macro-economic performance.
Paul Rosenberg 08.19.04 at 2:50 am
Remember in 2000? Virtually all the modellers predicted Gore would win. Which, of course, he did. But not by as much as they predicted.
Why? By empirical observation (see the Daily Howler archives), because the press lied its ass off about Gore. So what does this say about this time–not just this model, but about these models in general? Perhaps that with so damn few data points, it’s a wonder they predict anything at all, ever. 2000 was just a stunning reminder.
You’d think that they’d learn. You’d think that they’d want to avoid the public humiliation. But making those models is just so much fun (you remember being 8 years old, right?)
If you want to climb down from the realms of powerful model building, however, you can see amazing things. Such as the fact that Bush’s re-elect number only rose 3 points during the Iraq war invasion bounce:
“In mid-March, just before the outbreak of war, 45 percent in a Gallup poll said they would like to see Bush re-elected, rather than have a Democrat win in 2004. That’s gone up only to 48 percent in the current Pew survey.”
— Ruy Teixeira, “Public Opinion Watch: April 10 to 18”
http://www.tompaine.com/feature.cfm/ID/7633/view/print
I read Teixeira religiously, so way back then, I expected to see Bush scrambling right…about…NOW!
dsquared 08.19.04 at 11:18 am
I think it was me that coined the phrase “cheap suit econometrics” for models which “fit where they touch”.
Andrew Edwards 08.19.04 at 12:17 pm
Andrew, not quite—he predicts the popular vote; so in 2000 his model predicted (correctly) that the Democrat would win.
Fair enough, although a model that predicts, as I understand it, only one Democratic win from 1976 to 2000 still might have a partisan tilt to it. Whether intentionally or not.
Great metaphor, D^2.
Matt Weiner 08.20.04 at 4:35 pm
Andrew, I agree (on both points). Anyway, the Super Bowl model predicts Kerry with 59.33%, so we’ve got a powerful test this year.
Comments on this entry are closed.