Filed under:

# Week 12 college football win probability charts

Taking a look at some of the biggest and most exciting games of the week

This offseason I developed a win-probability model for college football. The model is built off of play-by-play data from 2009-12 and considers the Down, Distance, Spot, Lead, and Time Remaining of each play, as well as pre-game factors such as who is home and FBS vs. FCS teams. You can read more about the development and accuracy here.

Feel free to request a game I haven't included in the comments. Also the model does not include overtime for now.

### Mississippi at Alabama

Their team colors are so similar I put Mississippi State as the black dots.

### Northwestern at Notre Dame

I answered some questions about Brian Kelly's decision to go for two at the end of the game for a Notre Dame fan

I have to give the model a down and distance so the win probabilities are from the perspective of having kicked off and given Northwestern the ball at the 25, so basically an average kickoff.

ND's predicted WP up 11: 90.8
ND's predicted WP up 12: 92.8
ND's predicted WP up 13: 94.4

The historic data takes a little more work because you have to decide how specific you want to get. Here are the results for how many times the home team wins when the away team has the ball, down by 11/12/13, with 10-11 minutes left in the 4th, and has 1st and 10 at the 20-30:

Historic WP up 11: 8 wins out of 9 chances
Historic WP up 12: 4 wins out of 4
Historic WP up 13: 1 win out of 2

I can relax the time constraint to only be 9 or more minutes in the 4th quarter, but I just want to caution that when I say chances I really mean plays that have occurred in the area we want. So the same team may be in the group more than once from the same game if they run multiple plays that fit the conditions. Its just a limitation I didn't think about when I made the data.

Historic WP up 11: 54 wins out of 58 chances (95.1%)
Historic WP up 12: 29 wins out of 31 chances (95.5%)
Historic WP up 13: 26 wins out of 30 chances (86.7%)