Matt Runnels: I had a stat question that wasn't in the book but came up on another website regarding survival football pick'em. Are upsets/close games more likely in interdivision games? (Some people want to take New England over Buffalo as a sure pick, while others are saying "NO, THEY'RE BOTH AFC EAST!")
So if New England played two teams that were roughly equal, but one was in the AFC East, would the AFC East team have a better shot due to playing each other twice a year/knowing each other better/hating each other more/etc...?
Sure seems like it, right? We all can remember significant upsets where a bad team beat a much better division rival. Just a few examples: It sure seems like the Broncos and Raiders always play close games, right? Remember when the Dolphins were awful and they beat the Patriots on Monday Night Football right before the Pats won their third Super Bowl? The Colts struggle more with Houston and Tennessee than with any other opponents, right?
I could research this using all sorts of complex variables, but I wanted to just run a quick regression to get a general idea of whether this axiom is true. I took every game going back to the 1995 expansion and ran a regression with the dependent variable as binary (win or loss) and the following independent variables:
Some readers may have noticed that this XP was up a couple hours ago, then disappeared. That's because I mistakenly used every game for every team, and it came out that "division game" was completely, totally meaningless. Of course it did. I was looking at every game twice. Duh.
So I re-did the regression, only from the point of view of the team with more Pythagorean wins.
Like the original, faulty analysis I did, the variable for "division game" was completely, totally meaningless. Regressions produce this thing called "P-value" which is used to tell if the variable is significant or not. In general, any variable with a P-value below 0.1 is considered significant. Here are the P-values for the four variables:
That's less significant than an album of Stone Temple Pilots covers by the Olsen Twins. However, the results were slightly different when I used margin of victory as the variable, not just win/loss. The P-value was .11, not significant but close. The coefficient was -.68. In other words, in a division game, if you feel daring, you could give the underdog an extra seven-tenths of a point.
I also tried the regression looking at games where one team was clearly better than the other. In games where one team was two or more Pythagorean wins better, the result is virtually the same -- the coefficient is -.70 and the P-value is .20. In games where one team was four or more Pythagorean wins better -- the really surprising upsets, the Dolphins over Patriots on Monday Night Football stuff -- the coefficient is +.13 and the P-value is .86. In other words, when there is a huge difference between the teams, the fact that they play in the same division does not mean anything.
The verdict: If our readers in Nevada or England feel like buying an extra half-point when betting on a division game, sure, feel free. But it doesn't mean anything for your survivor pool.
One more note: The regression gives the home team an extra 2.7 points, pretty close to the three-point standard for home-field advantage used in Vegas.