This is the ninety-seventh year in which I’ve messed around with some sort of college football rating schema – lots of changes, different viewpoints, etc – and I think we’re finally onto something. Last year I used drive success rates, which was helpful – but frankly a lot of work, and at the end of the day somewhat questionable in its utility. I am not sure we got to a better answer, although the top of last year’s charts were absolutely justifiable. Of course, what system is not justifiable? But this year, I’m doing something simpler – rigorous, but focusing on final scores. At the end of the day we have two dimensions to look at – who did you beat, and how did you play. Ultimately, what matters is who you beat (and lose to), so I was looking for a way to not lose that data, while still focusing on the wins. What we do here instead is to use the margin of victory – the best indicator of a team’s actual quality – to determine schedule strength. (instead of opponents winning percentage, focus on opponents margin of victory) Then use that schedule strength to normalize wins and losses. How does it work?
- Schedule strength
- We look at the margin of victory in every game, using a diminishing return scale. Every point up to 28 points counts (so a 28 point win is worth 14 points more than a 14 point win). After 28 points, each additional point counts a half. (so a 35 point win is worth 28 + .5*7 = 31.5 points). After 42 points, each additional point counts one tenth. So Boston College’s 72-0 win over Howard is only worth (28 + 14*0.5 + 30*0.1) 38 points.
- In the case of true road games, a road added is given to the road team. This adder is equal to the aggregate margin of victory between true home and away teams in FBS games. Last year it ended up at 3.3 points. Right now, it’s at 7.1 points (it will shrink).
- Calculating margin of victory and opponents margin of victory (in games not involving your team) is pretty straightforward from here. However, we know that in reality your opponents’ opponents scoring margin should impact your opponents strength too. Of course you can go on and on like this (a mathematical wormhole). Last year after 10 iterations, the standard deviation across teams got down to basically zero. (a point where adding more iterations did not give any more information) Of course right now, the 10th iteration is not too helpful, so we go to 20 iterations. The standard deviation will shrink over the season of course.
- We take the collected result (your margin relative to opponents margin relative to their opponents margin relative to … and so on) and scale it from 0 to 1. This scale number is the strength of the team.
- Relating it to wins
- Now that the opponents have “value”, how much is beating a team worth. If a team has a rating of t, a win is worth t and a loss is worth minus (1-t). So the top scoring margin right now is Alabama. So a win over Bama is worth 1 full point and a loss has no deduction. A win over a team with a .95 rating is worth 0.95 points, while a loss deducts 0.05 points and so on.
- In theory we could do a W/L percentage here, but the problem is that it treats all unbeatens equally, which tells us nothing about relative quality of resume. So we take the “win points” subtract the “loss points” and scale it to a per game basis. We also give extra credit and blame for true road wins and home losses (based on collective win/loss in true home-road games between FBS teams).
- All non-FBS teams are treated as “Other” in order to fairly capture their impact on schedule. Other measures like RPI disregard those games, but they are too big a portion of many college schedules to just set aside like you might for basketball.
Now the results below are preliminary – so little is known about true schedule quality. Also, not everybody has played 2 games, which would allow strength of schedule metrics to be mathematically correct. In particular pay attention to the “scale”. A lot of teams are bunched up between #16 and #42, so the relative ranking there is pretty meaningless. (as if the others are that helpful)
|84||San Diego State||1||1||-0.0486||0.420|
|92||San Jose State||1||1||-0.0716||0.407|
|123||New Mexico State||0||2||-0.6395||0.092|