After refining pass efficiency and turnover stats, I redid the linear regression model for season wins as the dependent variable. (As a refresher, the regression model estimates season win totals based on relevant team stats.) R-squared for the model is 0.74, not any better than the previous model, but it is better in several other respects.
First, the true pass efficiency stats are used. The effect of passing game efficiency is captured more accurately. True pass efficiency counts sacks as pass attempts.
Second, turnovers are no longer lumped into one big 'net turnover' variable. Instead, each type of giveaway and takeaway is given its own variable which are per snap. For example, I use interceptions per pass attempt, defensive forced fumbles per rush/sack/completion, and offensive fumbles per rush/sack/completion. Although not as powerful as net givaway-takeaways, these 4 variables quantify the effect of each type of turnover on both sides of the ball.
Third, I've added a new variable to the model I haven't used before--penalties. Using either penalties or penality yards has roughly equal effect. Both offensive and defensive penalties are significant at the p=0.05 level. It doesn't add much to the goodness-of-fit of the model, but it does help rectify the problem of defensive run efficiency insignificance.
Defensive run efficiency becomes more significant in the model when penalties are added, and further as densive forced fumble rate is added. In this version of the model, defensive run efficiency (DRUNAVG) is now significant at p=0.05. I'll need help figuring out why.
*** denote levels of significance
- Home
- Unlabelled
- Revised Season Win Model
Revised Season Win Model
By
Brian Burke
published on 4/08/2007
Subscribe to:
Post Comments (Atom)
Sorry I'm a little late on this one. Just found your site and am really enjoying it.
With regards to the significance of defensive rush efficiency...I would imagine that adding defensive penalties has accounted for some of the variance in yards gained by the offense that indirectly acted as noise for the dependent variable (win probability).
a defense that has a lot of penalties gives up more yardage which would hurt their chance of winning, but would not be explained by any of the other defensive variables.