I have been a longtime reader of Gregg Easterbook's weekly
Tuesday Morning Quarterback column. ONe of Easterbrook's recurring features is a summary of the often dubious success rate of the many people who publish predictions of each week's NFL games. This led him to propose this brain-free formula: "Home team wins." As he notes, this requires no knowledge of any kind except which teams are playing and where the game is being held. Given a modest home-team-advantage edge (put at 3 points by
Jeff Sagarin), this formula should be right a little better than half the time over the season, and in fact it was in 2006 (53%) and 2007 (57%). During the 2006 season, it occurred to me that this formula ignored the fact that, despite the home field advantage, it was more likely that the better team would win. How do we know the better team? Well, on average the better team will win more games. So I started tracking this simple two-part formula: The team with the better record wins; if the records are the same, home team wins.
Easterbrook first published this prediction in his
Bad Predictions Roundup column in Feb. 2007, noting a success rate of 56%.
Then in
a 2007 pre-season column, Eastbrook wrote: "This year, I will endorse the generic forecast advocated in the offseason by about 20 readers, pulling from a hat the name Catey Tarbell of Kirkland, Wash. Catey's Law: Best Record Wins -- Unless Records Equal, Then Home Team Wins. TMQ will track this off-price house-label generic forecast against the predictions of full-time football experts who possess incredible insider information." As any good academic does, I recognized immediately the plagiarism of my own work, fired off a note to Easterbrook, who kindly corrected the record in the next day's
reader-response column:
"Tuesday I wrote, 'This year I will endorse the generic forecast advocated in the offseason by about 20 readers, pulling from a hat the name Catey Tarbell of Kirkland, Wash. Catey's Law: Best Record Wins -- Unless Records Equal, Then Home Team Wins.' Professor Eric Isaacson of Indiana University notes that last year I attributed this idea to him. Hey Eric -- you don't think I can actually remember everything in Tuesday Morning Quarterback, do you?! At any rate, from now on I will call the generic prediction in question the Isaacson-Tarbell Postulate."
Easterbrook has reported interim and final results several times since then. Before I link to these, I want to note that, unfortunately, Easterbook consistently over-reports the success of the formula. I've not verified the cause but believe it is because he is looking at the teams' records
after the game has been played, not before. This has the effect of changing the formula's prediction in favor of the winning team after the game is over! After each of the links below I'll provide the correct results.
October 10, 2007 (TMQ reports: 52-24, 68%. Actual: 48-28, 63%)
November 7, 2007 (TMQ reports: 94-36, 72%. Actual: 87-43, 67%)
February 8, 2008 -- Quoting from the season wrap-up:
"The Isaacson-Tarbell Postulate ruled the landscape of NFL predictions, finishing the season 183-84, or 69 percent correct. Of the dozens of NFL predictors TMQ tracked, John Czarnecki of Fox Sports had the best finish, at 183-84; ESPN's best predictor was Keyshawn Johnson at 176-91. Since its first mention in TMQ, the Isaacson-Tarbell Postuate has been referenced throughout the blogosphere."
The correct figure is: 175-92, 66%
November 18, 2008 (TMQ reports 107-51-1, 69%; Actual: 96-64, 60%)
February 10, 2009 -- Quoting from the season wrap-up:
"The Isaacson-Tarbell Postulate, proposed by TMQ readers Eric Isaacson of Indiana University and Catey Tarbell of Kirkland, Wash., holds: Best Record Wins Unless Records Equal, Then Home Team Wins. If you'd picked 2008 NFL games using this simple algorithm, you would have picked 175 of 267 contests correctly (.665), counting London as a "home" game for New Orleans, Toronto as a "home" game for Buffalo and the Super Bowl as a "home" game for Arizona, which was the home team of record. The Isaacson-Tarbell Postulate is a huge time-saver because you don't have to think about picks -- you don't even need to know who's playing!"
The correct results are 159-108 (60%).