Sunday, November 18, 2007

Election 2008: Second Forecast

My eighth forecast is here.

REVISED, 5:05 PM, 11/18/07

The Presidency - Method 1

Intrade posts odds on which party's nominee will win in each State and, therefore, take each State's electoral votes. I assign all of a State's electoral votes to the party that is expected to win that State. Where the odds are 50-50, I split the State's electoral votes between the two parties.

As of today, the odds point to this result:

Democrat, 302 electoral votes

Republican, 236 electoral votes

(No change since the first forecast, 11/16/07.)

The Presidency - Method 2

I have devised a "secret formula" for estimating the share of electoral votes cast for the winner of the presidential election. (No, it's not "method 3," described here.) The formula is based on the 35 presidential elections from 1868 through 2004. The standard error of the estimate is 3.6 percentage points, as against the winner's average share for the 35 elections, which is 71.7 percent. Here's how formula-based estimates for the 35 elections compare with the actual results of those elections:

The only "wrong" pick is the one for the election of 1876, which was decided even more crookedly than the 1960 election (see below).

The largest errors (greater than 5 percentage points) occur in these seven instances:
  • McKinley's re-election in 1900 -- EV share underestimated by 8.2 percentage points, that is, by 37 EVs.
  • T. Roosevelt's election in 1904, after having become president following McKinley's assassination in 1900 -- Share overestimated by 5.3 percentage points, 25 EVs.
  • FDR's re-election in 1940 -- Share underestimated by 5.1 percentage points, 27 EVs.
  • Truman's election in 1948, after having become president following FDR's death in 1945 -- Share overestimated by 5.7 percentage points, 30 EVs.
  • Kennedy's election in 1960 -- Share underestimated by 5.6 percentage points, 30 EVs. (Illinois, thanks to Chicago's mayor, Richard Daley, delivered its 27 EVs to Kennedy. Otherwise, the legitimate outcome, 276 EVs for JFK, would have been uncannily close to my estimate of 273 EVs for JFK . I assume, perhaps wrongly, that JFK's narrow win in Texas, with its 24 EVs was owed to LBJ's ability to pull in votes as JFK's running mate, not to LBJ's ability to rig elections.)
  • Reagan's initial election in 1980 -- Share underestimated by 8.1 percentage points, 44 EVs.
  • G.W. Bush's re-election in 2004 -- Share overestimated by 6.1 percentage points, 33 EVs. (By the way, I chose to use a model that's wide of the mark in 2004, for the sake of getting a better fit across the board. It bothered me, at first, to show a bad estimate at the right-hand edge of the graph, where it looms so obviously. But a statistical model should be chosen for how well it fits all of the observations on which it's based, not just the outliers -- statistical or graphical.)
My formula currently yields these estimates of the outcome of next year's presidential election (CORRECTED, 12/13/07):
Democrat nominee -- 241 to 280 EVs

Republican nominee -- 258 to 297 EVs
In sum, the prospect of a Democrat victory isn't as clearcut as method 1 suggests.

Will method 1 or method 2 prove to be the more accurate one? The answer is less than a year away. Stay tuned.

U.S. House and Senate


* * *

How did I do in 2004? See this and this.