Filed under:

Expected Wins after Week 12 and Picks for all Remaining Games

Taking a look at the model's accuracy in its first season, and how it predicts the remaining games

FBS Expected Wins after Week 12

With only two weeks left in the regular season, it's time to start looking at how the model performed.

First, though, your updated expected wins after Week 12.

I use the same model to calculate expected wins as well as to make the weekly picks.  The expected wins is a combination of total wins to date, plus the sum of the remaining game probabilities of win.  Entering Week 13, for instance, Nebraska has seven wins and two games remaining in the regular season.  The model gives Nebraska a a 0.64 chance of beating Penn State and a 0.44 chance of beating Iowa.  7+ .64 + .44 = 8.10 expected wins.

Generally, the model has been pretty successful at predicting the winner using a methodology of scoring offense versus scoring defense. Not surprisingly, though, it's much more accurate when the probability of win is skewed heavily towards one team.

This graph shows the average accuracy of a prediction based on the forecast winner's probability of win.

For matchups where the winner's probability of win is between .50 and .60 the model's accuracy is about .58. It jumps up quite a bit for matchups with a winner's probability of win between .60 and .65, and then takes a substantial drop from .65 to .75. Matchups with a winner's probability of win of greater than .75 have an accuracy of .89. The points well below the .90 line are primarily the results of the FCS upsets of FBS teams early in the season.

Overall, the model predicted 74.7% of games correctly. The challenge for my off-season is to fine-tune the way the model incorporates strength of schedule. Calculating strength of schedule isn't a difficult thing to do. Figuring out how it should weigh into the algorithm is difficult.