All data originated from Pro Football Reference and nflFastR.
Week 3 was a disappointing result for the model. Too many upsets and a tie caused me to only go 9 for 16 and as I suspected, I dropped another game to Vegas because my stupid model thinks Jeff Driskel is better than Tom Brady. Actually, ESPN ranks Brady 26th and Driskel 27th in QBR, so by that measure I wasn’t too far off.
Vegas is widening the gap both in wins and in point spread accuracy. I’m going to have to right this ship and since we mostly agree on games each week, time is running out.
Vegas and I agree on 14 of 16 predicted winners which includes TEN/PIT, so maybe we only agree on 14 of 15.
Every week I am amazed that my model matches Vegas so closely. I thought we would agree on maybe 12-13 games a week but so far that has averaged 14.5 games each week, with the point spreads being around 2 points apart on average.
The needle moved on expected wins. With week 3 data rolled in, the model now thinks weeks 4 - 17 look better by about 0.4 games, up to 9.7 expected wins for the season.
So far, my model expected the Colts record to be 1.997 - 1.003. Way off.