Joined: 14 Jan 2005 Posts: 1523 Location: Delphi, Indiana
Posted: Sun Sep 23, 2007 11:50 pm Post subject:
Each player's stats generate a T rate (see link, above).
eW = (T - TR)*Min/X
where TR is the T rate of a 'replacement player' -- a player who adds no wins, but only fills minutes.
X is a floating number that causes league eWins to add up equal to league games played.
Last year, TR was 11.72; x was 5525. _________________ 40% of all statistics are wrong.
Dan and I are going to focus primarily on simple and well known metrics. We will do this both for simplicity sake, and because it would not be fair to our employer to supply the competition with too much information.
Joined: 03 Jan 2005 Posts: 671 Location: Washington, DC
Posted: Mon Sep 24, 2007 9:58 am Post subject:
To underscore Dan's point about ANY metric describing wins well once a team adjustment is added... I ran a simple Tendex-style formula on each team for last season. In raw form (no adjustment for pace or team defense), that formula had a .69 correlation with efficiency differential (ortg - drtg). Berri's Win Score had a .81 correlation. Then I added a defensive adjustement based on the team defense's distance from average. I applied the adjustment identically to both my rating system and Berri's. With that adjustment, my simple Tendex system had a correlation of .97 with efficiency differential -- Berri's was at .965.
Just to see how much work my defensive adjustment could do, I applied it strictly to team points scored. The correlation with efficiency differential went from .33 to .82.
Note -- My defensive adjustment is NOT how Berri does his adjustment. _________________ My blog
Joined: 30 Dec 2004 Posts: 534 Location: Near Philadelphia, PA
Posted: Mon Sep 24, 2007 11:27 am Post subject:
Mike G wrote:
Each player's stats generate a T rate (see link, above).
eW = (T - TR)*Min/X
where TR is the T rate of a 'replacement player' -- a player who adds no wins, but only fills minutes.
X is a floating number that causes league eWins to add up equal to league games played.
Last year, TR was 11.72; x was 5525.
My recollection in looking into evaluating this method was that I could not find formulas for "Sco", "ast", "reb", etc. - the things that go into T - and I couldn't reproduce numbers posted. That could be a reason eWins wasn't evaluated. _________________ Dean Oliver
Author, Basketball on Paper
http://www.basketballonpaper.com
It would surely be desirable to have all of the fomula but you only a satisfactory dataset on EWin scores to evaluate the performance of the metric. You need the formula to try to analyze why it performed the way it did. Every last detail of Wins Produced is not out there yet though Dave Berri indicates it will be in 2008. A large share of overall adjusted +/- has been shared but maybe not every last detail.
Based on DLew's statement and others I assume that some version of adjusted +/- scores well in explaining team wins. I can understand protecting the exact score or any innovations added beyond what is previously described. Still the fascination of knowing it and the speculation that other worthy contenders could be out there now or developed and wanting to know the complete finish order of the race.
As far as I know all released information about "overall +/-" was for a 2 year period of 2002-3 to 2003-4. It is harder to score that from the outside, but maybe a reasonable estimate could be derived, backed into by knowing the statistical +/- weights and perhaps someone else computing the pure +/- for that timeperiod and trying to replicate the blending approach for each year. Or, as noted earlier, doing the same for a more recent timeperiod where pure +/- is available already.
Last edited by Mountain on Mon Sep 24, 2007 12:00 pm; edited 2 times in total
Joined: 14 Jan 2005 Posts: 1523 Location: Delphi, Indiana
Posted: Mon Sep 24, 2007 11:54 am Post subject:
HoopStudies wrote:
...I could not find formulas for "Sco", "ast", "reb", etc. - the things that go into T ...
That's right, these quantities are proprietary. I can't even get them, as my spreadsheet has taken complete control. Suffice it to say they are per-36-minute rates, standardized for teams' opponents averaging 100 Pts and 44 Reb; also adjusted to MPG, % of games started, and further et-ceteras. 'Sco' is scaled to (TS%/.53)^.6, as of last year. But that, too, changes, for example within a given playoff series.
At some point I figured no one really follows the formulas, but some will scan the results and decide if it is/isn't believable. Then you may proceed on faith or request a copy of the spreadsheet. I will send it if you will sign a medical disclaimer against future damages. _________________ 40% of all statistics are wrong.
Joined: 31 Dec 2004 Posts: 881 Location: Durham, NC
Posted: Tue Sep 25, 2007 8:06 am Post subject:
Mike G wrote:
That's right, these quantities are proprietary. I can't even get them, as my spreadsheet has taken complete control. Suffice it to say they are per-36-minute rates, standardized for teams' opponents averaging 100 Pts and 44 Reb; also adjusted to MPG, % of games started, and further et-ceteras. 'Sco' is scaled to (TS%/.53)^.6, as of last year. But that, too, changes, for example within a given playoff series.
At some point I figured no one really follows the formulas, but some will scan the results and decide if it is/isn't believable. Then you may proceed on faith or request a copy of the spreadsheet. I will send it if you will sign a medical disclaimer against future damages.
I've been waiting, pen in hand, for some time now ready to sign the disclaimer and see the spreadsheet, but you've never gotten back to me. Does this mean I didn't pass the pre-screening process?
Joined: 14 Jan 2005 Posts: 1523 Location: Delphi, Indiana
Posted: Tue Sep 25, 2007 8:39 am Post subject:
OK, you asked for it. Am currently mailing myself a copy and will forward it to Gabe and Mtn (or anyone else). You might get explanations for this or that, privately or on forum. It looks like I changed some weights and maybe deleted a vital macro. Still, you might read formulas, look for errors, make it better, etc.
Edit: Please don't direct questions to this thread; better in the adjoining one.
-- In the sheet, ignore the 'team' line above the players' lines. It isn't to the same scale, and some parameters don't make sense; they just got in the way of some formulas.
-- Positions (and raw stats) are all from DougStats. (Don't blame me.) _________________ 40% of all statistics are wrong.
Joined: 13 Oct 2005 Posts: 374 Location: Atlanta, GA
Posted: Tue Sep 25, 2007 10:48 am Post subject:
Mike G wrote:
OK, you asked for it. Am currently mailing myself a copy and will forward it to Gabe and Mtn (or anyone else). You might get explanations for this or that, privately or on forum. It looks like I changed some weights and maybe deleted a vital macro. Still, you might read formulas, look for errors, make it better, etc.
Edit: Please don't direct questions to this thread; better in the adjoining one.
-- In the sheet, ignore the 'team' line above the players' lines. It isn't to the same scale, and some parameters don't make sense; they just got in the way of some formulas.
-- Positions (and raw stats) are all from DougStats. (Don't blame me.)
Hey Mike, I want in on this, too! I've wondered what makes eWins tick for a long time...
Joined: 14 Jan 2005 Posts: 1523 Location: Delphi, Indiana
Posted: Thu Sep 27, 2007 7:19 am Post subject:
davis21wylie2121 wrote:
... at the team level, DRebs are a proxy for the entire process of forcing a zero-point possession by the opponent. So unless the individual actually forced the miss himself (and also performed all of the defensive actions that led up to forcing that miss), he's being overvalued for his individual DReb. ..
This is a belated response, but DW's (fine) critique was applauded by a couple others, who had no quibble with this passage. Yet I don't think this is one of the valid criticisms of B-Wins.
The player who scores isn't the only guy involved in setting up the score. The guy who is credited for the assist isn't necessarily the one who made the best pass on the play. Defender A may tip a pass that player B steals, or force his man into the area of a teammate, who blocks the shot.
Almost all stats are the result of team play, and only one player gets credit for any one (FG, Stl, Reb, etc). But in the course of a game, no one player gets all the rebounds or anything else. They tend to be distributed among the players that most 'deserve' them.
In the course of a whole season, the distribution of DReb among players may be a reasonable proxy for how well they placed themselves on defense. Better defenders get more playing time; and if the team plays better D, there are more DReb available.
In conclusion: I don't think the Berrites overvalue rebounds; I thing they undervalue other things. Scoring is the biggest part (and I haven't looked closely, so can't comment on the rest.) This of course effectively becomes 'overvaluing rebounds', relative to everything else. _________________ 40% of all statistics are wrong.
1. I agree that most stats involve giving one player the credit for something the team accomplished, so that on the surface it doesn't seem like anything new to do this with defensive rebounds. But it seems to me that defensive rebounds are typically more extreme in this funneling than most others cases. Multiply that with the frequency of defensive rebounds and you're going to get some absurd results. Perhaps it's the degree that's important here?
2. You mention that defensive rebounds may be a proxy for further defensive ability. I'm not sure that's close enough to a universal rule to be useful, but it's hard to say. It seems to be a better proxy of being big and near the basket. Are there any studies on this?
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
