Ray C. Fair is a professor of economics at Yale University. In this Wall Street Journal ($) article, , Professor Fair’s new book — Predicting Presidential Elections and Other Things — is reviewed and it sounds like a winner:
How can you guess who might be having an extramarital affair? This is an important question, and it deserves to be treated with scientific rigor.
Start with a theory. As a first approximation, it seems reasonable to suppose that the likelihood of having an affair depends on income, age, number of years married, marital satisfaction and religiousness. Next, find some data — say, a sex survey published in a magazine like Psychology Today or Redbook. Now fit the data to the theory (which means having your computer run a line through a cloud of points — a technique called linear regression) and do a statistical test to see whether the theory is any good. And what do you know? It is!
Now comes the fun part: prediction. Using the results, you can guess which of your friends and neighbors might be straying from the matrimonial paddock. Likely candidates for an affair are those who (1) have a high wage rate, (2) have been married a long time, (3) are relatively young given the length of their marriage, (4) aren’t very happily married and (5) aren’t particularly religious. Want something more quantitative? Well, all else being equal, an extra 10 years of marriage increases the predicted number of adulterous encounters per year by about six. (Warning: Blackmail based on these findings is strongly discouraged.)
Predicting adultery is only one of the interesting subjects that Professor Fair addressed. However, during this political season, the most interesting subject is his model for predicting Presidential elections:
By trial and error, Mr. Fair comes up with a list of eight: the growth rate of the economy, inflation, the number of economic “good news” quarters leading up to the election, whether an incumbent is running, how long the incumbent party has held the White House, whether there is a war on and, finally, a “party variable” in case the electorate has an innate preference for one party over the other. As data, he uses election results from 1996 (when President Clinton beat Bob Dole) back to 1916 (when President Wilson beat Charles Hughes).
After fitting the data to the theory, Mr. Fair finds that all eight variables affect voting at greater than chance levels.
And applying Professor Fair’s model to the Presidential elections from 1916 through 1996 reflects that it is pretty darn accurate:
From 1916 to 1996, Mr. Fair’s theory only calls two elections incorrectly. In 1960 Nixon received 49.9% of the vote, but according to the theory he should have received a 51.1% — a relatively small discrepancy. More embarrassing to the author’s analysis is the 1992 election, in which President Bush’s predicted share of the major-party vote was a winning 50.9%, whereas his actual share was 46.5% — a whopping 4.4 percentage-point error.
Moving to the 2000 election, which lies outside the data set used to construct the theory and is therefore a good test of its validity, Al Gore should have received (a losing) 49% share of the vote that went to the two major parties, but he actually got (a losing) 50.3% share. Not bad.
So, how does the Professor size up the 2004 election?:
Mr. Fair’s analysis will be cheering to President Bush, who, as a Republican president running for re-election when the Republicans have been in power only one term, enjoys the best possible incumbency situation. The only way he can lose, the theory suggests, is if the economy suddenly tanks.
Looks like another book to add to my reading stack.