APBRmetrics

Harold Almonte · Joined: 04 Aug 2006 Posts: 616

OK Forget it. I catched it again.
WOW did:
Scoring Eff=P - FGA
=(+P - FGMade) - FGmissed
=+FGMade weight - FGMissed
Def. EFF=oppFGA - oppP
=+oppFGMissed + (oppFGMade(tm.adjusted) - oppP)
=+DR - oppFGMade(tm. adusted) weight

My proposals still the previous:
Scoring eff.=+P - FGMade (tm. adjusted, or *0.2) - FGMissed*DBRP / % contested and heavy contested defense
Deff. eff.=+DR*0.7*DRBP + oppFGMissed(tm. adjusted, or *0.3) - oppFGMade(tm. adjusted)

This is almost PER with the points allowed punishment!

Dan Rosenbaum · Posted: Fri Nov 23, 2007 12:40 pm Post subject:

I think the nomenclature of "possessions employed" and "possessions acquired" makes this more confusing than it needs to be. What we have are possessions produced when the "own team" has possession and possessions produced when the "opponents" have possession.

The problem with Wins Produced is that it assumes that possessions are produced in two completely different ways depending on whether the "own team" or "opponents" have possession. When the own team has possession, offensive rebounds are equal to a full possession, but they are worth nothing when the opponents have possession. When the opponents have possession, defensive rebounds are worth a full possession, but they are worth nothing when the own team has possession.

This just does not make sense and this theoretical flaw generates most of the bizarre results that people have been complaining about for more than a year. Assume that both the "own team" and "opponents" generate possessions using the same possession production function and most of the problems with Wins Produced go away.

My co-authors and I touch on this point in our JQAS article, although it is not the focus of the paper.

NickS · Joined: 30 Dec 2004 Posts: 384

Longer response later:

Dan Rosenbaum · Posted: Fri Nov 23, 2007 3:10 pm Post subject:

Conan the Librarian · Joined: 03 Sep 2007 Posts: 35

I still don't understand why it makes any sense to penalize a player for taking, and making, a field goal.

mateo82 · Joined: 06 Aug 2005 Posts: 211

Conan the Librarian · Joined: 03 Sep 2007 Posts: 35

How is taking a field goal a lost opportunity cost? That, or free throws, is the only positive outcome of any possession. What else could you possibly do with that possession, other than turn it over, which would be a lost opportunity cost in that you've lost the ability to take a field goal attempt.

I understand the desire to "punish" highly inefficient shooters, but you have to shoot. And since it is impossible to determine through boxscores whether any given shot is a "good" shot, I can't understand how every field goal attempt is a "punishable" action. Rather, it is the desired end result of every possession, to get a shot off, preferably a good one that goes in.

Harold Almonte · Joined: 04 Aug 2006 Posts: 616

**Eli W** · Joined: 01 Feb 2005 Posts: 402

I just posted a comparison of Wins Produced and PER on my blog. It looks at their consistency from year to year. I'm not sure if it really tells us anything (well, I'm sure it tells us something, but I'm not sure what), but I figured it was worth looking at the numbers.

http://www.countthebasket.com/blog/2007/12/04/evaluating-player-ratings-year-to-year-correlations/
_________________
Eli W. (formerly John Quincy)
CountTheBasket.com

mtamada · Joined: 28 Jan 2005 Posts: 377

Harold Almonte · Joined: 04 Aug 2006 Posts: 616

That looks tricky, and probably has nothing to do with any metric. Rebounding roles are well stablished each position (depending of your skill of course). Wallace was changed to a team with more available rebounds and probably better rebounder teammates, and he kept his reb% the same. Meanwhile Iverson lost FGAs (available shots) when he was changed. Do it changing positions and scoring options!

I wonder how look the standard deviation curves of points, FGA, and rebounds for starter players, by position.

Guy · Joined: 02 May 2007 Posts: 128

Eli: the strong YTY correlation for REB is not only a measure of context impact, or even sample size (per Tango). It is also hugely impacted by the variance among players: the larger the variance, the higher the correlation (other factors equal). And of course there is a huge REB variance, in part because of positional differences. For that reason I think your failure to make a position adjustment does matter: a lot of the strong correlation you're finding just reflects the fact that centers/forwards have far more rebounds than guards.

The larger problem, as you say, is that knowing that players are consistent in their rebound totals doesn't tell us anything about how it impacts team wins. The YTY r for player height is almost 1.0, but so what? WP assumes that each additional rebound is a net addition of 1 rebound for the team. But that's clearly not true: the correlation between the number of rebounds grabbed by a team's top rebounder and team rebounds is only about .2. Or let's look at correlation to team wins. At the player level, WP has a slightly higher correlation with REB than with TS%. But at the team level, the correlation between REB and wins -- once you control for REB opportunities -- is pretty weak, much weaker than for TS%. So we know empirically that rebounding ability is much less important for explaining team wins than shooting efficiency, yet WP values it more highly. I don't see how Berri can explain that away. (But maybe "Flint" will join us and take a stab at it.)

HoopStudies · Posted: Wed Dec 05, 2007 12:41 pm Post subject:

Guy · Joined: 02 May 2007 Posts: 128

**Eli W** · Joined: 01 Feb 2005 Posts: 402