APBRmetrics

Dean Oliver write about a new ranking tool for college team that ESPN is using:

http://espn.go.com/mens-college-basketb ... -explained

In contrast to other established ranking methodologies, it takes into account injuries which is a significant advance in my opinion. It also gives decreasing weight to larger victory (or loss) margins (so, a 30-point win doesn't count twice as much as a 15-point win), which also makes sense to me. Perhaps more controversially, every win is better than every loss.

Has anyone tried incorporating these factors into an NBA power ranking system yet (publicly, I suppose some teams may be doing so)? It looks like that might just be around the corner.

A point for discussion I'd think:

"In BPI, a close win at home is better than a close loss on the road against the same opponent."

How can that be true unless home court advantage is irrelevant?

This isn't how the legal betting markets deal with results like that...and those still represent the standard in all the studies I've seen. Wonder if participants from the markets were consulted in the creation process. Wonder what the conversations sounded like to convince the analytical team that a one-point home win over an opponent is more impressive than a one-point road loss against that same opponent when home court is worth something in the range of 3-4 points in general...

Maybe it's not supposed to be a prediction tool, as for seeding the tournament.
Maybe it's just a "fair" reward metric for what everyone mostly cares about, which is Wins.

Mike G wrote:Maybe it's not supposed to be a prediction tool, as for seeding the tournament.
Maybe it's just a "fair" reward metric for what everyone mostly cares about, which is Wins.

Good point. DeanO does say that it seemingly is as good as or better than other tools for predicting, but that could be due to the other factors it takes into account.

I'd like to see retrodiction results for prior seasons. How did they validate this?

"In BPI, a close win at home is better than a close loss on the road against the same opponent."

So, a close win when you had a head start is more impressive than a close loss when the other guy had a head start?

I think they made a mistake when deciding that they don't want to optimize for prediction. Why make another team rating system anyway when that's not your goal? Weirdly, after not caring (at design stage) that they want good prediction they go on to say that it's better at predicting tournament matchups than Sagarin.. while using a whopping sample size of 4 seasons. Why not look at the last 10 or 20 seasons? If they had actually cared about prediction from the start they would have probably kicked the "Diminishing returns for blowouts" and "All wins are better than losses (before Opp Adj)" parts

I'd like to see retrodiction results for prior seasons

I think they did that, but just for '07 to now.

"we do think it is the best power ranking available."
I'd like to see how it does versus SPM, as it can factor in injuries with the most precision...
I'd bet money that SPM models or the LRMC are better, if only by some.


Note to self: convince Daniel to include efficiency * minutes% to improve SPM's R^2 against rAPM to create world's best power rating system...

Jerry, it can be reasoned that effects that are marginalized in the NBA can be seem more widespread in college ball, with the greater scale of competition.

There is also evidence in college ball that home vs. road performance is not accurately captured by point margin (i.e. the diff in win% is greater than the diff in point margin predicts), so that might be some of their reasoning for the "wins are always better" thing? Not sure I agree at any rate.

And adjusting for blowouts has some merit depending on how you look at it (Daniel has measured how teams play up or down to opponents, which is one version of this).

Has anyone seen any discussions anywhere in the college hoops world about that decision to weight close home wins as better than close road losses if the same teams were involved? Was hoping Dean might pop in here to explain why they went that route.

Saint Louis looks to be in interesting big picture benchmark team. I'm skeptical that the best team in the A10 is a top 10 team in general. And, this team in particular has played few top notch opponents. Both BPI and Pomeroy currently rank them at #10. Sagarin has them at #16. The team rankings site has them at what seems like a more logical #24 (logical because I'm market biased). Boooeee's most recent market reflection showed them at #23...but that's a week old now. Maybe he'll be posting the new numbers later today.

Anyone following college hoops who has other teams that has attracted their attention because of varied rankings?

The team rankings site is:
http://www.teamrankings.com/ncaa-basket ... ng-by-team

Hopefully falling to #243 Rhode Island will knock Saint Louis out of the top dozen...quite the mystery team given the differences of opinion on them. Ugly result there...