Oh yeah, the suspense is killing us

In the Wall Street Journal, Steve Liesman authors an article that bears the following headline:

No Matter the Nominee, One Model Says the Democrats Just Can’t Win

Paragraph 5 contains the following:

I’m going to use columnist’s prerogative for a few paragraphs here and leave you hanging as to Mr. Fair’s precise prediction[.]

Oh, all this hanging, it’s just too much for us. It looks like Mr. Copy Editor used his prerogative to give your punch line away Mr. Liesman. A real shame that. Not that the article is especially interesting, mind you. Liesman writes:

[…] and how you can use the formula to make your own forecast, using your own assumptions.

Seriously, [sic] it’s worth it to spend a minute discussing how the formula is produced and what Mr. Fair is trying to achieve.

Right, seriously. Because we were all laughing pretty hard at that great laugh line that was “using your own assumptions.” But seriously[!], how much thinking do you need to do to figure to what Mr. Fair is trying to do? Especially when Liesman told us right in the opening paragraph that Fair tries to predict the respective voting share of the presidential candidates. This isn’t exactly The Bible Code.

We can (mostly) leave aside the fact that some models predicted Al Gore’s victory in 2000 (and right they were, though that’s a different issue,) but it’s worth pointing out that when he looks at specific sets of circumstances, Fair’s sample size becomes too small to be useful:

The formula draws on similar situations — since 1916 — when an elected Republican president ran for a second term after he took the White House away from the Democrats. Those analogs include Ronald Reagan’s manhandling of Walter Mondale, Mr. Nixon’s mauling of George McGovern and Dwight Eisenhower’s savaging of Adlai Stevenson.

Hmmm, three “similar” elections? Not that much to go on, is it? Liesman continues writing something that gets liberals in trouble because, we’re always told, it just isn’t true:

Namely, we’ve had strong GDP growth with lousy job creation. […] Mr. Fair might learn, along with an unhappy Mr. Bush, that the two aren’t interchangeable and growth that doesn’t produce jobs isn’t good economics or politics. […] Whatever the results of today’s jobs report, there’s no changing the weak jobs numbers under the Bush presidency.

You’re preaching to the choir Mr. Liesman, PTTC! Besides,

While Liesman writes that the formula has an impressive margin of error of 2.4%, we’re only left with this indication of the formula’s success at predicting the winner:

Finally, the formula has been badly wrong only once: that was in 1992, when it predicted the incumbent would win with 51.7% of the vote. It was off by 5.1 percentage points. That, of course, was the last time a man named Bush ran for re-election.

We would have liked to know how many times the formula has been wrong, not just badly wrong.

Finishing second in a race, whether it’s by 1 second or 1 hour, means finishing second. It matters little how badly you lose. And until the US drops the electoral college, state-level forecasts seem like a better (if much more complicated) tool.


Comments: 8


Sounds like Mr. Fair’s model is the most impressive scientific theory this side of scientific creationism. Note, btw, that the latest AP-Ipsos poll shows that “Bush’s 47 percent approval rating is the same as his father’s at this stage in his presidency 12 years ago before he lost to Bill Clinton.” Deja vu all over again.


“The Savaging of adelai Stephenson” sounds like the title for a movie of the week to me


I hope there aren’t any pot-smoking, fellatio-enjoying, Osama-loving Democrat Party liberals involved! @:->}}}}


I stopped reading the WSJ piece when I realized that model who was predicting stuff wasn’t Kate Moss or some other super model who might be expected to have superhuman precognitive powers.


And sometimes … coming in first, with 1 more vote or 500,000 more votes is coming in second.


If memory serves, GHWB didn’t receive 46.7% of the 1992 poll, but rather 37%, a difference of 14+% from the model.


Professor Fair’s equation is an example of regression analysis. If you go here:
then (a) you might learn something and (b) you might note that he hopes Bush loses.
He specifically cites the error rate of the prediction and notes that he is trying to predict the percentage of the two party vote (Nader and Perot don’t count).
This site:
is trying to do a state by state projection, but there is no evidence that there is any regression analysis used to verify the forumla used.


then (a) you might learn something and (b) you might note that he hopes Bush loses

Regression analysis on a sample that size is problematic, and while we have no real issues with Fair’s work, the author of the article is clearly out of his depth when it comes to writing about statistical analysis. What Fair hopes outcome wise is of no relevance here, and in a close election country-wide support for the two major parties isn’t necessarily going to be that useful.


(comments are closed)