PDA

View Full Version : Division I Computer Rankings for 11/7/04


LakersFan
11-10-2004, 10:31 PM
The updated computer computer rankings, including RPI, are avaible on my website http://vortex.bd.psu.edu/~rutter/WomensHockey/0405WomensD1Rankings.html. Here are this weeks top ten:

1 Minnesota
2 Dartmouth
3 Minnesota-Duluth
4 Harvard
5 New Hampshire
6 Wisconsin
7 St. Lawrence
8 Princeton
9 Brown
10 Ohio State

Mercyhurst drops out of the top 10, while Brown slides in at number 9. Harvard took a little tumble, but their first two games were against weak opponents. UMD jumped up from 7th, after two wins at UW.

Check out the web page for other rankings as well, including RPI. Things are very early, but it is interesting to note how the ranking systems compare this early in the season.

Last week, Patman made the comment that he wasn't quite sure if he liked the Bayesian approach to the current ranking system, as it adds bias to the results. As a Bayesian, I don't share those same concerns, since I feel that it is important to use all of the infromation that is available to me when doing a statistically analysis. That information includes last year's perfomances. The model is set-up in such a way that as more results from this year enter the model, the data will overwhelm the prior, and the bias will be little, if any. We will be able to see the effects of this bias as the season goes on. At my website, the KRACH rankings are not Bayesian, and KRACH ranks New Hampshire number 1 for comparision. The bias of Bayesian estimates will be an argument that will continue to rage among statisiticans unitll all the non-Bayesians are converted :). It is interesting to note that Mease-Rutter closely mirrors the USCHO.com poll, an indication how the human pollsters are also using previous results to rank the teams.

Patman
11-12-2004, 11:15 PM
I've only started my graduate work, but Bayesian approaches appears a little sketchy to me, it works on what is presumed to be knowledge and is applied to minimize variance while introducing bias... a possibly unholy combination. Again this is all from first glance. I don't, on the whole, like making guesses and it appears at times that Bayesian statistics is some guess work. I'd rather see things operate from pairing down the impossible and the likely improbable instead of a fudging calcuation. The profs here at UConn have 3 1/2 years to convince me otherwise, in the meantime for sports data I would VERY much avoid any system statistical bias, because like real bias, it contains a lot of possible unfounded fudgefactor elements.