Pythagorean Wins Revisited

I was browsing Football Outsiders today and noticed that they posted a link to a Pro Football Reference article regarding Pythagorean Wins.

At bottom, the guys at Pro Football Reference wanted to know if capping blow out wins in the NFL would make the Pythagorean projection more accurate.

And what did they find?

They found that capping blowout wins in the NFL does not make the formula any more accurate in terms of determining how good teams truly are. However, they did note that capping blowout wins in college football would make the that formula more accurate in terms of determining how good teams truly are.

If you recall from making posts on Pythagorean Wins, I have been doing the same thing for months in regard to college football. I've been making the argument that games against the Sisters of the Poor (Western Carolina, etc.) should not be used in Pythagorean projections because the talent disparities between the teams are so great, and one team is effectively guaranteed a win. The bigger opponent can basically name the score on the Sister of the Poor in question, and that massive blowout win inflates that team's Pythagorean projection to make it look like they should have more games than they did.

At bottom, when you do it for the entire SEC, it essentially means that basically every SEC team underachieves in a year, i.e. they didn't win as many games as the formula says they should have.

The same analysis, as the guys at Pro Football Reference noted, does not work for the NFL because the teams are so similar. Even when the best NFL teams play the worst, for example the Raiders, the teams are still relatively closely matched. You never run into anywhere near the disparities that you do in college football when, for example, when USC plays Idado (as they do in the 2007 season opener). To get that same disparity in the NFL, you would have to have a team like the New England Patriots play someone, for example, from NFL Europa, and of course that never happens.

So what does it mean for Pythagorean wins for college teams?

It means, as I've been saying for months, you should not factor in the cupcake games, and focus much more on conference games.

Perhaps I should tweak my Pythagorean formula slightly more, say including all "big" games (conference opponents and legitimate out-of-conference opponents, including possibly bowl games), or perhaps I should count the games against the Sisters of the Poor, but somehow weight down that game to where it is not very important. Doing so, of course, has pitfalls of its own (namely, you are comparing apples and oranges because then teams are playing very different schedules, unlike when you compare only conference play when schedules are very similar, so strength of schedule would became a major factor that would have to be evaluated), but perhaps it would make the formula more accurate.

Maybe that should be a research topic for another day.