Sunday, August 16, 2009

NFL: Pythagorean Projection IV: Picking Games

Update 14 Feb 2010: on further review, I had a bug in the program I used to generate the numbers I had originally posted here. I was picking games using the projection for the same season (rather than using the previous season's projection). Naturally, this made the picks look better than they should. I have updated the table below to reflect picking one season's games based on the projection from the previous season's numbers.

I've also added the results for the 2009 season.

Introduction

  • Part I: how Pythagorean Projection did over the 1994-2008 seasons (this also contains an introduction of what Pythagorean Projection is).

  • Part II: a team-by-team look at how Pythagorean Projection dis for the 2007->2008 season.

  • Part III: the 2008->2009 projections

Using Pythagorean Projection to Pick Games

In this post, I will look at how the Pythagorean Projection does in picking the winner for individual regular season games for the following season. The method is very simple: the team with the highest projection is picked to win.

The following table details the tallies resulting for this method for then 1994-2008 seasons. As usual, tie games are counted as a push.

The 'correct percentage' is then computed, using the usual (w+t/2)/(w+l+t) formula.

SeasonRightWrongPush Correct %
199413391059.4%
199512583060.1%
199614595060.4%
1997129109254.2%
199814892061.7%
1999124108053.4%
2000143105057.7%
2001137110155.4%
200214099158.5%
2003136120053.1%
2004146110057.0%
2005149107058.2%
2006140116054.7%
2007142114055.5%
2008140115154.9%
200916492964.1%

Conclusion

As it turns out, Pythagorean Projection does a surprisingly good reasonable job picking games (producing results a bit better somewhat worse than last year's regular season numbers for APR.

More on this in a couple of weeks. I will post the picks for Week 1 based on the Pythagorean projections from last season (as well as the Line picks as usual).

Friday, August 7, 2009

NFL: Pythagorean Projection III: 2008->2009

In Part I, I looked at how Pythagorean Projection did over the 1994-2008 seasons. In Part II, I looked at how it did team-by-team for the 2007->2008 season.

In Part III, the projected wins for the 2009 season is presented. As you should be aware by now, you can expect an error rate of ±4 or more games for as many as 10 teams.

Team Projected 
Wins
Comment
Rams 2.5824 New HC
Lions 2.7847 New HC, (and QB?)
Bengals 3.2359 Carson Palmer should be back
Chiefs 4.3666 New HC, QB
Browns 4.3836 New HC, Anderson/Quinn back
Raiders 4.5542 New HC
Seahawks 5.3737 New HC
Jaguars 6.1842
Broncos 6.2170 New HC, QB
49ers 6.8998 Newish HC
Redskins 6.9572
Texans 7.3029
Bills 7.8322
Cowboys 7.9218
Cardinals 8.0222
Bears 8.6526 New QB
Dolphins 8.7997
Packers 8.9221 New defensive coaching staff
Buccaneers 9.0484 New HC, QB
Jets 9.2131 New HC, QB
Vikings 9.2171 New QB?
Saints 9.5347
Falcons 9.7252
Panthers 10.1263
Colts 10.1733
Chargers 10.1735
Patriots 10.5850 Tom Brady is back
Eagles 11.2536
Giants 11.3240
Steelers 11.8460
Ravens 11.9466
Titans 12.0570 No more Albert Haynesworth

Comments:

  • The Bengals, Brown, and Seahawks are almost guaranteed to do better if they can avoid the serious injury problems they had last year.

  • Would Tarvaris Jackson qualify as a new QB for the Vikings? What they really need is some consistency at the position. They haven't had a QB start all 16 games of the season since 2004—Daunte Culpepper's last full (healthy) season with the franchise.

  • Albert Haynesworth falls outside of the usual HC/QB issues of other teams. But in the 2007 season, the Titans' ability to play defense pretty much disappeared when he was out injured. And for a defensive-oriented team like the Titans, that could be a big, big deal.

  • On the other hand, Dan Snyder has an extensive history of spending money on big-name players with unremarkable results, so Redskins fans shouldn't get too excited about having him on their favorite squad...

Coming up

The Pythagorean projection can be regarded as a kind of predictive power rankings... how well does it do predicting week 1 games?