APBRmetrics

Mountain · Joined: 13 Mar 2007 Posts: 221

There are ways to translate PER into an estimate of average player win contribution. It would only be rough, not precise but PER Wins would answer the question what does PER measure.

It could be done by correlation of PER and actual wins or some exisitng player win estimate like winshares or another. Perhaps Dan's regression work comparing explanatory power of major methods already has the answer. You could perhaps concede it fails to explain somewhere near 25% of wins that are accounted for by shot defense and plug in an esimated value for that from a different source. You could then compare expected PER Wins for a team for a completed season to actual win and measure the error and the lost wins from expected for cause that you then could try to track down. Lost expected PER wins could suggest issues with coaching or player performance volatility in general or issues with clutchtime. (Clutch PER has been calculated before and could be again.)

The also appears to be a rough relationship between minutes weighted own cumulative team PER compared to league and offensive efficiency compared to league average with the ratio of team PER/75 (PER 15 * 5 positions) about twice as far from 1 as the team offensive efficiency / league efficiency ratio suggesting doublecounting of nonscoring events that improve scoring opportunities and the scoring itself. (This is also evident when looking at team point differential and tendex differential at dougsstats.com.)

ziller · Joined: 30 Jun 2005 Posts: 118 Location: Sac Metro

To those who have been following along, we have another study up at Ballhype -- this time looking at players who received more minutes due to injury. The usual results follow.
_________________
SactownRoyalty.com
tziller@gmail.com

Eli W · Joined: 01 Feb 2005 Posts: 287

Great work by Tom and Kevin and all the bloggers and forumites involved in this debate. I think this is a sign that statistical analysis of basketball has really made progress from the fringes to the mainstream. It hasn't permeated as far as baseball stat analysis yet, but it's on its way.
_________________
Eli W. (formerly John Quincy)
CountTheBasket.com

Mountain · Joined: 13 Mar 2007 Posts: 221

Yes the Ziller / Pelton study makes a solid contribution using the best cases available.

As for trying to help statistical analysis of basketball make progress from the fringes to the mainstream, people here and similarly inclined around the blogosphere could discuss what one or several contributions from the world of stat analysis could be the targetted push elements for the season or the half decade...

John Hollinger · Joined: 14 Feb 2005 Posts: 77

BTW, we're working on a VAR (or VORP, depending on your preferred acronym) type method to complement PER this year. More details to come ...

lowpost · Joined: 12 Oct 2005 Posts: 4

Mountain · Joined: 13 Mar 2007 Posts: 221

Sounds good John. Will be interested in any twists in the approach over previous ones out there.

Nikos · Joined: 16 Jan 2005 Posts: 339