Home » Blogs » Pete Davis's blog

Why Republicans Are In Trouble In 2008

May

18

Pete Davis

Ray Fair has one of the best presidential election prediction records of anyone around, and he's an economics professor at Yale University.  His only big miss since he started predicting in 1978, was the 1992 election, when he missed Bill Clinton's win by 5.1%.  He predicted the 2000 election would be razor-thin, and he overestimated President Bush's winning margin in 2004 by 3.4%.  All of his other predictions have been much more accurate, with a standard error of 2.54%.  His latest prediction for 2008 is that the Republican will get only 47.8% of the popular vote, implying the Democratic candidate will win.  Of course, the Electoral College vote, not the popular vote, is what will actually determine who wins.

His equation is a simple function of real per capita GDP, inflation, and dummy variables for which party controls the White House, and whether the incumbent president is running for reelection. His web site documents his predictions and how little his estimating methodology has changed over time.

This is economics at its best: simple, straightforward, based upon well documented data and methodology, and with useful predictions that have stood the test of time.

On 21:49pm, May 18th, 2008 Yet Another Budget Wonk (not verified) said:

Did you read the same web

Did you read the same web site I did?

In October 2004 the model predicted a Bush victory by 15 points.
http://fairmodel.econ.yale.edu/vote2004/vot1004.htm

In 1996 his equation was predicting a dead heat, while the result was decisive Clinton victory. http://fairmodel.econ.yale.edu/vote2008/vote.htm

His predictions only look good when adjustments to the model are applied retroactively.

On 19:02pm, May 19th, 2008 Pete Davis said:

Ray Fair's Presidential Election Predictions

I didn't look at every year. In 2004, he predicted President Bush would win 57.8% of the popular vote, and Bush actually won 51.2%. In 1996, he called it a dead heat. His model predicted Clinton would win with 51.2%, but that was not large enough compared to the 2.54% standard error to make a prediction. Clinton actually won 54.6%. Those are large errors, but the point remains that Fair's equation has called the winner every time since 1978 except in 1992. All econometricians are handicapped by data revisions. This is particularly true in this case. Almost any equation will look much better after revisions improve the accuracy of the data. I wouldn't blame that on the estimator or on the quality of the equation. I know these errors seem large, but I'm amazed they're not larger given the ebbs and flows of electoral politics and whether eligible voters show up and vote at all. My experience estimating the revenue effects of tax policy changes using micro models with actual tax return data has demonstrated to me that large errors crop up all too often, even when you're very confident of the data and the methodologies you have. Those without the humbling experience of having to frequently estimate based upon flawed data sometimes reject all estimating, but that just leads to more subjective and erroneous estimates. I still remain impressed by Ray Fair's estimates despite their occasionally large errors. Just getting the winner right that often is pretty good estimating in my book.
On 20:00pm, May 19th, 2008 Yet Another Budget Wonk (not verified) said:

The Fair Doctrine

He taught his daughter well as well: Fix the model when it doesn't work.

On 15:55pm, May 20th, 2008 Barry (not verified) said:

"All of his other

"All of his other predictions have been much more accurate, with a standard error of 2.54%. "

Um, this is called 'trimming the data'. If my worse mistakes aren't counted, I'm pretty good, as well.