Game probabilities for week 13 are available at the New York Times. This week I look at Aaron Rodgers's value to the Packers.
...The model used here to produce the game probabilities is a “parametric” model, meaning its inputs are parameters that can be varied easily. That’s obviously a good thing, as team statistics certainly vary from week to week. It’s also handy because those parameters can be manually adjusted to account for injuries and just to play what-if...
By replacing Rodgers’s numbers with Flynn’s, and holding all other factors as is, we can get a rough idea of how the game probability should change. Unsurprisingly, the numbers would now heavily favor Detroit...
Only 58% for the Patriots in Houston? That seems odd.
Hi Brian. That's fascinating. Does the model also significantly downgrade Cleveland's 82% win probability if you substitute the passing stats accumulated under Weeden's tenure for those of Hoyer/Campbell? Also, can the regression be improved by disaggregating team passing stats into some sort of individual player statistic parameters to catch the effect of players like Calvin Johnson? I guess a small sample size problem would quickly start to appear, no? What a challenge NFL predictive models present! Few games in season from which to calibrate, constantly changing cast of players, players themselves who change capabilities significantly due to injury or simply to age. Whew.
"If Rodgers were able to play the entire season, his numbers would add four wins to make the Packers more like an 11-win team"
I dunno, if one player is really worth a good four wins over only a part of a season all by himself (compared to an average player as Flynn is stated to be) then a serious market failure is implied: QBs at this level are being seriously under-paid by the league's GMs.
I probably missed this in the posts, but when did you change the HFA for the game probabilities?