Tuesday, October 2, 2007

2007 Week 5 Predictions

The predictions are based on yards per play on rushing and passing plays (sacks and sack yards included), sack rates, third down conversion rates, penalty yards per game, and fumble and interception rates. The models are trained on the 1996-2006 seasons. Linear regression predicts the final score margin (home team points - away team points). Logistic regression estimates the probability that the home team will win.

Quick recap of last week's predictions: The logistic regression based on adjusted stats was 8-6.

Based on adjusted stats

GamePredicted Final Score MarginP(Home Team Wins)
DAL @ BUF-4.701231.1754
CLE @ NE9.321776.5465
SEA @ PIT1.625354.3521
MIA @ HOU4.436763.7474
TB @ IND5.029967.0868
ATL @ TEN3.705765.2925
SD @ DEN9.191678.9935
JAX @ KC-0.643347.1863
NYJ @ NYG10.550083.2431
DET @ WAS5.512466.2263
CHI @ GB6.267870.6839
CAR @ NO-2.248843.9337
ARI @ STL-6.184228.9171
BAL @ SF-3.576441.8151

Based on unadjusted stats

GamePredicted Final Score MarginP(Home Team Wins)
DAL @ BUF-11.955414.7143
CLE @ NE14.834787.3854
SEA @ PIT5.016666.0379
MIA @ HOU6.108868.8604
TB @ IND2.690859.2368
ATL @ TEN5.559372.6257
SD @ DEN12.960985.6946
JAX @ KC2.977559.3624
NYJ @ NYG7.271776.1502
DET @ WAS3.852959.9085
CHI @ GB7.583574.1125
CAR @ NO-6.679028.5680
ARI @ STL-7.049926.4957
BAL @ SF-5.407236.0024


Brian Burke said...

Derek-Are the adjusted numbers on top or bottom?

Brian Burke said...

Your system beat mine this week!

Do you use any sort of method for accounting for early-season statistical extremes? Such as de-tuning extremely high or low performance stats?

I'm finding my model needs some sort of compensation for say, Denver's early season boom, or San Deigo's early struggles. It's more likely than not that each team will regress to their true mean throughout the year, and their to-date stats don't accurately reflect true team capabilities.

Derek said...

No, but looking at the power rankings after this week's action so far, SD shot way up and DEN fell way down. They're both in the middle of the pack now.

It's something I've been thinking about, though. Take the Giants' 12 sacks against the Eagles for example. Yes, their defensive line performed really well, but that extremely high sack rate is going to unfairly skew the averages. Maybe some sort of sigmoid function to define upper and lower limits for some of the stats would help.