|
|
Same day, another article about the BCS computers and how they can't use score margin. This one comes in response to Mike Huguenin's article about the first BCS rankings this week.
In an overall decent look at how the initial BCS rankings shape up, he writes:
USC is 10th in the computer rankings and has no remaining games against teams currently in the BCS top 25. The Trojans beat Ohio State by 32, yet are ranked five spots behind the Buckeyes in the computers.
That statement is unfair to the poor computers. (Not that their feelings can get hurt). I agree it's worthwhile to point out that Ohio State is ahead of USC by the computers' estimation, since that is unusual and very unintuitive. But to point out that USC beat Ohio State by 32 is not relevant to evaluating the computers' performance, because the computers aren't allowed to "know" the score of that game, only that USC won it.
Put into perspective: what if USC had won by a single point? They subsequently lost to Oregon State, giving them one loss as well. And for all the computers know, that loss was by 50 points. But just assume we know two things: USC beat Ohio State in Pasedena, then later lost to Oregon State.
Is it so ridiculous to put Ohio State ahead of them? Maybe it still is. But some of the computers will evaluate Ohio State and conclude that they've won seven games, and lost one game, on the road, to a very good team. USC, on the other hand, beat five teams, one of them a very good team, but lost to a mediocre team. That's how Ohio State ends up ahead in four of the six computers, and tied in another. Only Richard Billingsly's algorithm has USC ahead, by a single place.
Add in the score margin and almost any computer algorithm will put USC in front. They're #2 in my Strength Power Rating, which uses score margin, while Ohio State is #13. In my Success Power Rating, which like the BCS computers does not use score margin, USC is #6 and Ohio State #8. The Trojans are still ahead, but as you can see, remove score margin from the equation and they are a lot closer.
BCS computers can't use score margin because the BCS committee decided to make it that way. On one hand, it makes the power ratings less accurate, especially at mid-season, and allows examples like this one to make the computers look stupid. On the other hand, it gets rid of incentive for coaches to run up the score. However, coaches have enough incentive to run up the score anyway: to impress the human pollsters, though this can sometimes backfire.
Another statement Huguenin makes is also a bit misleading:
While strength of schedule isn’t a BCS component, all six computers have a strength-of-schedule factor in their rankings.
|
|
It's true the BCS used to add in a ridiculous "strength of schedule" component, back when the formula was overcomplex and they tinkered with it every year to try to somehow make it match the AP and Coaches polls' #1 and #2 teams. They got rid of the strength factor, as well as the "quality wins" bonus, leaving it 2/3rd human polls, 1/3 computers.
The computers include a strength-of-schedule factor by their very nature. There is nothing "added in" as far as I know, to any of the formulas; it's just part and parcel of how they work. Imagine a computer ranking system that didn't include strength of schedule in any way. In other words, the computer couldn't know which teams are playing each other.
That's right: by letting the computer know which teams are playing each other, strength of schedule gets factored in no matter what. Without it—and also banning score margin—the computer might as well rank teams by winning percentage and let that be it. Or add in a home/road factor of some sort. There's nothing else to go on. It's the way the computer uses the head-to-head comparisons between teams that makes a power rating. Very few power ratings, if any, do what the BCS used to do, which is figure out a strength-of-schedule score and add it in later, as its own component.
My problem with some computer ratings, as I've mentioned before, is when strength of schedule is average rather than aggregate. When it's averaged, you can have a situation where an 11-0 team would be ranked higher if they didn't play Washington State in their last game. In an aggregate system, they'd play the Cougars, gain few if any points, and have basically the same rating they started with. If their strength of schedule is averaged, they lose points for winning a game.
That's all for now. I'll be back the next time someone writes something erroneous about the BCS computers, even if it's just to take a shortcut because they understandable don't want to have to write ten paragraphs about it. I'll do that part.
Comments