Comments on Advanced NFL Stats: Win Probability Model Accuracy

"There isn't a handy goodness-of-fit stat...

2009-08-08T07:03:49.799-04:00

"There isn't a handy goodness-of-fit statistic to cite."

Not true! Whatever statistic is of interest (accuracy, quadratic loss, AUC, etc.) could be estimated on hold-out data.

Mark-Welcome back. No idea. I'm guessing we us...

2009-07-29T18:02:34.685-04:00

Mark-Welcome back. No idea. I'm guessing we used similar data sets, so if there were quirks in the data they'd probably show up in any model.

Hey - just came back to the site for the first tim...

2009-07-29T01:56:30.196-04:00

Hey - just came back to the site for the first time in the offseason. In our WP probability, we ran the exact same graphs as above with identical results. My one question - do you have any idea why the greatest diversion points are at ~20% and 80% ? We observed a similar phenomena....

-Mark Kamal (Protrade - mkamal@protrade.com )

Nice blog . i kept it as a bookmark. thanks a lots...

2009-07-19T05:47:59.396-04:00

Nice blog . i kept it as a bookmark. thanks a lots.

Brian,have you tried to running a WP calibration u...

2009-07-09T14:03:51.634-04:00

Brian,have you tried to running a WP calibration using an atypical group of games,for example a group where there is a very strong home favourite.

Your observed tight symmetry may be a reflection of the make up of the non training set very closely resembling the make up of the training set.

Systematic errors in these kind of data driven models usually become more evident when you apply them to extreme match ups.

Ian-True. If I add in a HFA adjustment, even one d...

2009-07-07T17:26:08.527-04:00

Ian-True. If I add in a HFA adjustment, even one done badly, the confidence levels will improve.

81-Probably not. The model's achilles heal currently is 4th down decisions late in games. It has a hard time telling when teams will actually go for it. The WPs can be drastically different if a team kicks or goes for it. In that case, DEN went for it and won.

So in the Broncos 39-38 win over the chargers last...

2009-07-07T17:13:31.897-04:00

So in the Broncos 39-38 win over the chargers last year, your model has the chargers winning that game 99% of the time at one point. Did the Broncos really only have a 1% chance there?

With regard to improving general accuracy, would y...

2009-07-07T11:21:42.673-04:00

With regard to improving general accuracy, would you look to add in some form of adjustment to take account of home-field advantage?

I suppose with that one major issue is how you fade out home field advantage and bring in game situation (i.e. at the start of the game, the home team wins more often than away teams, but if the scores are tied at the start of the fourth quarter we'd expect somewhere closer to 50-50, although home teams probably slightly ahead)