In Method 1, I assign all of a State's electoral votes to the expected winner in that State, according to TradeSports.com. A price of greater than 50 indicates a Bush win; a price of less than 50 indicates a Kerry win. (A winning bet of $50 on Bush at a price of 50 returns $100, for a $50 profit; a winning bet of $60 on Bush at a price of 60 also returns $100, for a $40 profit; a losing bet on Bush at a price of 60 pays off those who bet on Kerry; and so on.) If the price is exactly $50, I record the electoral votes as a tossup and don't allocate them to either candidate.
In Method 2, I allocate all of a State's electoral votes to Bush if the TradeSports.com price is 55 or greater, and all of a State's electoral votes to Kerry if the Tradesports.com price is 45 or less. For prices between 45 and 55, I allocate a State's electoral votes according to Method 2. Method 2 has no predictive power; it simply measures the uncertainty around the estimate yielded by method 1.
Method 3* translates the expected share of two-party popular vote into electoral votes, based on a statistical relationship for presidential elections from 1952 through 2000. I use two sources to estimate the leader's share of the two-party vote: the popular vote-share share market at Iowa Electronic Markets; the leader's share of the Bush-Kerry vote according to the Rasmussen tracking poll. I use those share estimates in the following regression equation:
Fraction of electoral vote going to the popular-vote leader =Electoral-vote percentages for the elections of 1952-2000 fell within or very close to the normal range of the estimates (mean, plus or minus standard error). However, Bush's percentage in 2004 (53.2 percent) fell markedly below the normal range (62.4 to 70.2 percent). That result is consistent with a pattern that has emerged since 1980, when Reagan's electoral-vote share was above the normal range of the estimate. Since then, the electoral-vote share of the popular-vote leader slipped steadily through the range, hitting bottom in 1996 and 2000, then dropping below the range in 2004.- 8.327 (a constant term)The r-squared of the equation is 0.95; the standard error of the estimate is 5.8 percent; and the t-stats on the coefficient and three variables are -2.908, 2.836, -2.509, and 2.507, respectively.
+ 29.249 x the leader's fraction of the 2-party popular vote
- 23.161 x the square of the leader's fraction of the 2-party popular vote
+ 0.0696 (if the leader is Republican, otherwise 0).
Based on further analysis of the elections of 1952-2004, I have concluded that the Republican electoral-vote advantage applies only when the Republican candidate is winning decisively in the two-party popular vote (54 percent, or more). Thus, in tight races, method 1 is the best way to estimate the electoral vote. At any rate, it worked well this year.
I hereby retire methods 2 and 3. It's method 1 for 2008.
* Revised slightly on 11/18/04 to correct a minor data entry error.