Ken Pomeroy discusses updates to ratings methodology

[Marcus92]

http://kenpom.com/blog/ratings-methodology-update/#more-1155

A lot of this is way beyond my understanding. But there are some interesting ideas here on offensive/defensive efficiency as predictive measures, strength of schedule and how things like recency are weighted.

One thing that caught my eye is in the last paragraph: he plans to make the home-court advantage values specific to the court. The assumption is that Duke's advantage playing at Cameron is different than DePaul playing at Allstate Arena.

[Henry Sugar]

thanks for sharing. I missed this when it was originally posted.

Here are my notes and commentary:

The switch to Adjusted Efficiency Margin as the ranking is interesting and makes life a little more user friendly. The pythagorean exponent was originally borrowed from baseball, but the exponent was way different, and Pomeroy had changed it a few times.

The additive approach vs the multiplicative approach is also interesting and also makes life a little more user friendly. The previous view was (Team A offense * Team B defense) / overall average offense. In this view, let's say MU scores 1.07 ppp but they play Villanova, who allow 0.90 ppp. MU would be predicted to score 0.97 ppp against Villanova, adjusted for home/away. I'm going to have to fix my kalman filter model to account for the new setup.

I have no comment about his change to recency or game importance, other than I have also looked at predictions of teams with wildly different ability (ie - cupcake games), and those predictions are typically bonkers. If he has adjusted to account for the spread, that will be good.

I'll be interested to see the results of the SOS and conference ratings this season.

Last, for the site specific home court advantage, I believe he's referring to a team like Villanova (again) that plays in two different arenas.