Predicting remaining W's on ND's schedule

For those that have noticed, I've been playing with a variety of stats this year doing some analysis for fun. I've been a bit facinated with yppd - yards per play differential (e.g. if we average 6.0 yds/play and our opponents average 5.0 yds/play, we are at +1 yppd).

Well, I've decided to expand my model and I've included a number of things, including total yards/game, home field, and winning % - to come up with a predictor model. Essentially, by combining these variables, I've create a formula that correctly 'predicts' the winner to a game approximately 80% of the time (based on data from 2008-2010). For 2011 so far, this formula would predict the winner 85% of the time (likely due to the tougher part of the schedule not yet being played).

The predictor works by having a % chance of winning - 0 to 100%. If it's above 50%, I say it predicts a win, below 50%, a loss.

Some more detail on its accuracy -



[% chance of winning] [% of times right] [# of Games in sample]
0-19 94% 950
20-39 77% 709
40-49 54% 538
50-59 56% 405
60-79 76% 756
80-100 94% 940



So, in the middle of its predictions, it's right about 55% of the time. But is right 3/4 to more than 9/10 once it gets outside the 60% predictions.

Anyway - the point of everything I wrote before this is... the numbers in this model are complex, but are a decent predictor of who wild win.

So - let's apply this to ND's schedule this year. We get:


[Opponent] [Prediction] [% chance]
South Fla. win 64%
Michigan loss 19%
Michigan St. loss 40%
Pittsburgh win 78%
Purdue win 82%
Air Force win 81%
Southern California win 55%
Navy win 88%
Wake Forest win 51%
Maryland win 81%
Boston College win 103%
Stanford loss 19%



BC is above 100% because there's a home field advantage factor that sometimes pushes % above 100% and below 0%.

Now, the model gets better as the year goes on, so the numbers will change as we and everyone else plays more games. But, based on what's been played so far, Michigan and Stanford will be our toughest games. USC and (gasp) Wake are our hardest games left outside of Stanford.

For fun, here is what the model would predict for past years:

2008 (predicted 7-6)

San Diego St. win 94%
Michigan win 92%
Michigan St. loss 32%
Purdue win 77%
Stanford win 75%
North Carolina loss 35%
Washington win 94%
Pittsburgh loss 46%
Boston College loss 33%
Navy loss 42%
Syracuse win 96%
Southern California loss 3%
Hawaii win 65%


2009 (predicted 3-9)

Nevada loss 46%
Michigan win 56%
Michigan St. loss 49%
Purdue loss 42%
Washington win 70%
Southern California loss 33%
Boston College loss 50%
Washington St. win 98%
Navy loss 33%
Pittsburgh loss 11%
Connecticut loss 50%
Stanford loss 15%


2010 (predicted 8-5)

Purdue win 87%
Michigan win 66%
Michigan St. loss 28%
Stanford loss 21%
Boston College win 52%
Pittsburgh win 56%
Western Mich. win 83%
Navy loss 27%
Tulsa win 50%
Utah loss 46%
Army win 64%
Southern California loss 44%
Miami (FL) win 58%