Page 1 of 1

Possession cost scheme (Charles, 2007)

Posted: Fri Apr 22, 2011 2:25 pm
by Crow
Charles



Joined: 16 May 2005
Posts: 107


PostPosted: Sun Dec 09, 2007 9:10 am Post subject: Can some one explain the “possession cost” scheme? Reply with quote
I don’t understand this.


If I take two players and give each an "average" record based on playing 10% of their team's minutes, except one with all his points from two-pointers and the other with all his points from three-pointers the player shooting three-pointers gets 55 less PER credits.

Each player took the same number of shots. Each scored the same number of points. Yet, the two point shooter gets more PER credits (based on less FGx)

Am I missing something, or does a pound of stone actually weigh more than a pound of feathers?

Last edited by Charles on Sun Dec 09, 2007 10:41 am; edited 1 time in total
Back to top
View user's profile Send private message
deepak_e



Joined: 26 Apr 2006
Posts: 373


PostPosted: Sun Dec 09, 2007 10:05 am Post subject: Reply with quote
I'm looking at the PER formula right now, and here are the relevant terms:

3P + (2 - factor*(TmAst%))*FG - VOP*DRB%*(FGA_miss)

So, I think you're assuming that TmAst% is 0, DRB% is 1, and VOP is 1. Then, it reduces to:

3P + 2*FG - FGA_miss

Per possession, we expect 1 points to be scored (VOP = 1). If you have a player who scores 60% of the time on 2-pointers, then we expect:

.6*2 - .4 = 0.8 credits earned per shot attempt

And if you have a player who scores 40% of the time on 3-pointers, we expect:

.4*3 - .6 - 0.6 credits earned per shot attempt

So, according to PER, if all misses are rebounded by the defensive team, then it's better to score your X points at a higher percentage inside the arc then at a lower percentage outside the arc. This can easily be rationalized. I imagine that teams tend to be more efficient in their possessions when it follows a miss by the other team, versus a make since defense has less time to get set. So, a miss doesn't only represent lost opportunity on the offensive end, it also impacts the ensuing defensive possession. Not sure if it was intended, by Hollinger's scheme does seem to take this into account in a clever way.
Back to top
View user's profile Send private message
Charles



Joined: 16 May 2005
Posts: 107


PostPosted: Sun Dec 09, 2007 10:44 am Post subject: Reply with quote
You are correct deepak, I grossly over-simplified, since you can't determine the exact values without league averages. Could you comment on the modified first post.
Back to top
View user's profile Send private message
Charles



Joined: 16 May 2005
Posts: 107


PostPosted: Sun Dec 09, 2007 10:50 am Post subject: Reply with quote
deepak_e wrote:
So, according to PER, if all misses are rebounded by the defensive team, then it's better to score your X points at a higher percentage inside the arc then at a lower percentage outside the arc. This can easily be rationalized. I imagine that teams tend to be more efficient in their possessions when it follows a miss by the other team, versus a make since defense has less time to get set. So, a miss doesn't only represent lost opportunity on the offensive end, it also impacts the ensuing defensive possession.


Fair enough, although it is rather speculative. However, do you think this possibility makes up for about 8% less PER credits -- despite allowing 50% more offensive rebounding opportunities? This seems way out of balance to me.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Sun Dec 09, 2007 11:41 am Post subject: Reply with quote
Here is Berri on PER...

"Hollinger argues that each two point field goal made is worth about 1.65 points. A three point field goal made is worth 2.65 points. A missed field goal, though, costs a team 0.72 points.

Given these values, with a bit of math we can show that a player will break even on his two point field goal attempts if he hits on 30.4% of these shots. On three pointers the break-even point is 21.4%. If a player exceeds these thresholds, and virtually every NBA played does so with respect to two-point shots, the more he shoots the higher his value in PERs. So a player can be an inefficient scorer and simply inflate his value by taking a large number of shots.

But again, our model of wins suggests that inefficient shooting does not help a team win more games. Hence the conflict between PERs and Wins Produced. Hollinger has set his weights so that inefficient scorers still look pretty good. We argue that inefficient scoring reduces a team’s ability to win games, and therefore these players are not nearly as effective as people might believe."

Hollinger also has GameScore, a simplified version of PER. Although PER is extremely complicated, it has a 99% correlation with Game Score, or at least so I have read over at the WOWJ. The formula is....

(Points x 1.0) + (FGM x 0.4) + (FGA x -0.7) + ((FTA-FTM) x -0.4) + (OREB x 0.7) + (DREB x 0.3) + (STL x 1.0) + (AST x 0.7) + (BLK x 0.7) + (PF x -0.4) + (TO x -1.0)
Back to top
View user's profile Send private message
Charles



Joined: 16 May 2005
Posts: 107


PostPosted: Sun Dec 09, 2007 11:51 am Post subject: Reply with quote
Do you agree with Berri on this point, Flint?

I do. I see the PER approach as a compensation for the minimal weight given to shot creation. I realize shot creation is difficult to determine using nothing but official stats, but that's the challenge, because shot creation is essential to understanding the reality of basketball.

When you combine PER's implementation of "lost opportunities" with the error caused by assuming everyone has the same percentage of their shots assisted (Amare actually has about 65% of his baskets assisted compared to 20% for Nash) and a very loose rationalization for the weighting of assists versus points scored) it becomes easy to understand why Nash’s plus/minus has averaged about 7.3 points per 48 minutes higher than Amare’s since they began playing together. Plus minus is noisy, but a seven point difference over 6,000 minutes played is a lot.

Incidentally, I would be reluctant to criticize these weights if they were produced through regression or some other verifiable analysis rather than logic.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Sun Dec 09, 2007 12:26 pm Post subject: Reply with quote
Yes. I definitely agree with Berri. And I don't think there is any objective basis for Hollinger's weights.

You were discussing Joe Johnson in the other thread. Here is a comp of 03-04 and last season.

http://www.basketball-reference.com/fc/ ... 02&y2=2007

To me, it seems very odd that went from a 15 PER to a 19.5 PER. What changed? His ts% was the same, 55.6% and 55.8%.

In y2, 1.4 less rebounds, .8 more assists, 1.6 more tumovers, and of course, 5.4 shots more per 48 and 8.3 points more per 48. To me, that is valuing shot creation a great deal more than necessary.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 422


PostPosted: Sun Dec 09, 2007 12:27 pm Post subject: Reply with quote
Flint, the WOW logic about the team possession lost after a FGMade could be acceptable, but his approach is wrong and he created a big mess. He penalizes scorers (FGMade) by attempts and rewards the inbounds (oppFGMade) in the opp. team, by minutes. He should have done this also by attempts (if he considers FGMade a matter of individual attempts rather than team), and then scorers just recuperate the most of this team possession lost/gain-with the ball and the clock out of play (just a rule). There's no any straight relationship between FGA and player minutes for him to attempt to cancel one thing with the other, and he's just rewarding low FGA-high minutes players.

PER could be overrating a bit, but WP is overpunishing there. They did the opposite with Rebounds.

Last edited by Harold Almonte on Sun Dec 09, 2007 2:38 pm; edited 2 times in total
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 53


PostPosted: Sun Dec 09, 2007 12:35 pm Post subject: Reply with quote
Quote:
Do you agree with Berri on this point, Flint?


LOL.

Quote:
But again, our model of wins suggests that inefficient shooting does not help a team win more games. Hence the conflict between PERs and Wins Produced.


If the distinguishing feature of WP is it's treatment of shooting efficiency, I wonder, "Flint," if you could explain to us why WP has a lower correlation with TS% than does PER? And lower than NBA Efficiency? The truth is that because WP penalizes all shots, not just missed shots, it rewards efficient shooting less than PER.

What really distinguishes WP is that it gives zero credit to shot-taking, not its treatment of efficiency. This means that high-usage/low-efficiency players will indeed be rated very low compared to other metrics, but it also means that high-usage/high-efficiency players will be rated lower than in other metrics.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Sun Dec 09, 2007 12:51 pm Post subject: Reply with quote
Guy - What do you think of PER?
Back to top
View user's profile Send private message
Charles



Joined: 16 May 2005
Posts: 107


PostPosted: Sun Dec 09, 2007 1:08 pm Post subject: Reply with quote
Guy wrote:

What really distinguishes WP is that it gives zero credit to shot-taking, not its treatment of efficiency. This means that high-usage/low-efficiency players will indeed be rated very low compared to other metrics, but it also means that high-usage/high-efficiency players will be rated lower than in other metrics.

Guy, have you seen evidence that low percentage shot taking deserves credit? I am not talking about shot creation - which is the "unassisted" portion of an unassisted field goal - but, simply shot taking itself?

The analysis I have done says an assisted two-point field goal attempt has no value beyond the actual points scored (in fact it has a small negative value.) Assisted three point attempts do have value, even when they miss, but they are easily separated out.
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 53


PostPosted: Sun Dec 09, 2007 6:57 pm Post subject: Reply with quote
Flintberri: My point is not to argue for (or against) PER. My point is simply that WP does not value efficiency more highly, as claimed, as demonstrated by its lower correlation with shooting efficiency. The metric is so dominated by rebounds that it can't give proper credit even to many of the efficient shooters that Berri celebrates.

Charles: I wasn't trying to draw a distinction btwn shot-taking and usage, and wouldn't claim to know the right way to value usage. (My intuition would be to try to establish a reasonable replacement level, which certainly would be lower than league average.) But I feel pretty confident that these equations from the WP perspective are wrong:
* 0-0 = 2-4 = 12-24
* 2-4 > 11-23
* Net value of all shooting by NBA players = zero.

WP says that all below-average shooters hurt their teams, so logically I assume that means these players should stop taking shots. But here's the problem: in this new WOW version of the game, half of those still permitted to shoot will be below the new, higher ts%. So WP will then tell us again that they too should stop hurting their team by shooting, having committed the sin of being below average. Eventually, I suppose each team will have one player who takes all of its shots. Personally, I don't find this terribly plausible. Perhaps Flintberri can make sense of it?
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 237
Location: Pleasanton, CA

PostPosted: Sun Dec 09, 2007 9:33 pm Post subject: Reply with quote
Charles wrote:
Fair enough, although it is rather speculative. However, do you think this possibility makes up for about 8% less PER credits -- despite allowing 50% more offensive rebounding opportunities? This seems way out of balance to me.


I think you've made a good point. The damage of a missed shot may go beyond the opportunity cost (and the chance of an offensive rebound) because the defense has less time to setup, but this should be established and measured first. Until then it seems like just an excuse.

And it seems to me the only way to avoid this discrepancy between 2p and 3p shooting is to subtract the same opportunity cost from all attempts -- makes and misses -- like you were suggesting in the other thread.

(I'm pretty sure the opportunity cost should not be '1' though.)

Quote:
Incidentally, I would be reluctant to criticize these weights if they were produced through regression or some other verifiable analysis rather than logic.


After seeing what a disaster WoW is, I don't think you should be reluctant to criticize weights just because they were produced through regression. These things don't run themselves. You always need to bring in some logic.
Back to top
View user's profile Send private message
gabefarkas



Joined: 31 Dec 2004
Posts: 980
Location: Durham, NC

PostPosted: Sun Dec 09, 2007 9:37 pm Post subject: Reply with quote
Flint wrote:
But again, our model of wins suggests that inefficient shooting does not help a team win more games.


Oh really? Who is the "our" in your sentence referring to? You and Mr Berri? Because I wholeheartedly disagree with this contention. Take a look at Iverson's 6ers teams, for one.
Back to top
View user's profile Send private message Send e-mail AIM Address
94by50



Joined: 01 Jan 2006
Posts: 458
Location: Phoenix

PostPosted: Sun Dec 09, 2007 10:06 pm Post subject: Reply with quote
Is it just me, or are the following possible?

* The break-even marks for 2pt and 3pt shooting in PER are too low.
* The break-even marks for 2pt and 3pt shooting in WP are too high.
* The ideal mark is somewhere in the middle.

Just eyeballing it, for example, if I take the 30% break even mark for 2pt shooting in PER, and split the difference with league average 2pt shooting (the break even mark in WP), which is, what, about 49%... the average of the two is in the high 30s. I suspect the ideal break even mark is somewhere in that neighborhood. Just a hunch.

Also, regarding the idea that "inefficient shooting does not help a team win more games"... well, of course it doesn't. The question is, how poorly does a player have to shoot before they shouldn't be shooting at all? The Sixers were more than willing to live with Allen Iverson's 40% field goal percentage for quite a while. Why is that? (And I don't think the correct answer is, "Because Iverson is a scorer and teams overvalue scorers.")

Mike G



Joined: 14 Jan 2005
Posts: 3621
Location: Hendersonville, NC

PostPosted: Mon Dec 10, 2007 7:27 am Post subject: Reply with quote
Doesn't the usefulness of the shot depend on a few things besides the % of time you make the shot? Like:

FT likelihood and %
the shot clock
one's teammates

In another thread we went into depth about how good/bad Willie Green is. His TS% is .470, so he should probably shoot less. How much less?

Updated, here's a chart like one I posted earlier. eTS% is estimated TS%, in which I've added (TS%-eFG%) to 82games.com's eFG% for each time interval. These shooting% are thus adjusted upward for each player; moreso for those who make more FT/FG. Assuming FT accompany FG equally at all stages of the shot clock.
Philly players, ranked by total shot attempt frequency:
Code:
eTS%% at seconds:
FG/40 FT/40 Sixers 0-10 11-15 16-20 21+
14.8 5.6 Iguodala .62 .53 .47 .39
15.0 5.4 Williams .64 .55 .56 .36
15.7 3.6 Miller .55 .54 .49 .38
16.2 2.2 Green .53 .46 .48 .43
15.0 2.2 Korver .51 .53 .49 .35
9.8 5.5 Dalembert .64 .61 .53 .52
10.9 3.9 Smith .60 .42 .48 .48
6.4 4.0 Evans .59 .24 .66 .19

Sixers .59 .50 .51 .39

The bottom line shows the team's eTS% in each time interval; estimated, again, by assuming FTA/FGA are evenly distributed in each interval.
Willie Green was criticized for taking low-% shots in the 11-15 second interval (when he had hit only 39%), thus depriving better shooters. Passing on a 46% shot to get a 50% shot is less a big deal.
But taking a 49% shot at 18 sec. is a lot better than a 39% shot at 22 sec. When you pass the ball, a few seconds have ticked away. And shooting specialists Williams and Korver seem to be worst at buzzer-beaters. Green is apparently the best (among those who actually get shots).

'Who your teammates are' should also include offensive rebounding. If 30% of missed FG are offensively rebounded, then even on a 30% FGA the odds of 'something good' happening are .30 + (.30*.70) = .51
When FT add .044 to a player's efficiency (league TS%-eFG%), then a 30% shot is 55% 'good'.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Mon Dec 10, 2007 3:09 pm Post subject: Reply with quote
If for a fairly scoring rating we need to rate: the assisted factor, the dificulty factor (how was guarded the scoring attempt, and by whom-s), and this kind of decission making factor. Why everybody say that defense is more dificult to rate than offense?
Back to top
View user's profile Send private message
Jacob



Joined: 29 Nov 2007
Posts: 19


PostPosted: Mon Dec 10, 2007 3:22 pm Post subject: Reply with quote
People worrying about the break-even marks in PER completely miss the point, and this includes our friend Berri. The talent level in the NBA dictates the break-even mark, not John Hollinger. A player with a 31% 2P% will never play significant minutes, let alone attempt significant shots. This is why it drives me crazy when people claim that "PER unfairly rewards 35% shooters who take 30 shots a game". Like these players actually exist in the NBA, the environment PER was designed for.

I guess there's also a continued misunderstanding what PER measures. It doesn't quantify a player's contribution to team wins, but it pragmatically identifies player quality. It recognizes the burden some players are made to carry by their coaches; it never mistook Allen Iverson for a bad offensive player, even in 03-04, and doesn't love him all of a sudden now that he shares the floor with Carmelo Anthony.
Back to top
View user's profile Send private message
John Hollinger



Joined: 14 Feb 2005
Posts: 175


PostPosted: Mon Dec 10, 2007 3:34 pm Post subject: Reply with quote
Part of the problem with setting a higher break-even mark for FG% and 3P% is that we only get half the story in the stats. What I mean by that is there is a stat for assists -- good passes that lead to a made basket -- but not for anti-assists -- bad passes or passes not made that lead to a missed shot. The consequence is that the shooter splits the credit for a made basket, but takes all the blame for a miss. If we can work out that quandary, we can reset the break-even threshold. Otherwise, if you set it any higher, you unfairly punish the shooter for misses that may have been the fault of others (such as the dreaded "flaming bag" pass, when a guy get thrown the ball when he's covered 20 feet from the hoop with two on the shot-clock.)

As an aside, somebody with league-average stats in every other category and the "break-even"FG% will end up with a PER around 8 or 9 -- you'd have to take a LOT of shots at that low efficiency mark to end up anywhere close to normal, and that would presumably only happen on on a horrendously bad team.
Back to top
View user's profile Send private message AIM Address
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Mon Dec 10, 2007 8:07 pm Post subject: Reply with quote
Mike G - They are discussing your post over at the Wages of Wins in the Kevin Garnett thread.

http://dberri.wordpress.com/2007/12/10/ ... -nba-team/
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Mon Dec 10, 2007 8:51 pm Post subject: Reply with quote
If it were prohibited to take Off. Rebs., then every FGMissed would be the same than a turnover, and will be straight canceled by DRebs.

No matter if you think or not an OReb. starts a new possession for a teammate (and he is not loosing any credit), a simple TEAM regression (used by WP to weight stats) would weight the cost of a FGMissed as a 70% of a turnover, because that is how much the opponent can steal (rebound) it. Then, I don't know why some stats are supposed to be weighted by team win regressions, and other ones by personal logic, and how some possessions are responsibility of the usager, and other ones (opp.FGMade-inbounds, opp. not stolen TOs, etc.) are for the team.

I don't understand how you can be stolen by a teammate. Or is he stealing from the opponent? Did the opponent have a possession with the ball in the air? It's not a pass, but It's not a double steal situation. It's not transfering credits from the off. rebounder to the shooter, is transfering the lost credits from the shooter to the off. rebounder of course, but not a full possession lost. The possession never was lost at all.
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Mon Dec 10, 2007 11:10 pm Post subject: Reply with quote
Flint: Conceptually, there's nothing inherently wrong with basing a valuation system on average efficiency, as Berri does, rather than replacement level. However, you then have to find a separate way to value usage, so that you don't end up saying that 0-for-0 shooting is as valuable as 15-for-30 shooting, which clearly isn't true. Setting a lower break-even level provides a mechanism for valuing usage and efficiency together. You may want to quibble with the precise break-even points in PER, but until/unless Berri amends WP to penalize players for not shooting as well as for inefficient shooting, I don't see how WP is an improvement.

In the thread you link to, Berri objects to any linear weight model (like PER) that sets the "break-even" rate below 50% on the basis "that it allows an inefficient shooter to increase his value just by taking more shots." He has made this point dozens of times, and seems to feel it's a debate-ending observation -- that any metric with this feature must be wrong. But the idea that below-average players must lose value the more scoring opportunities they consume is hardly self-evident. For example, the best value metrics in baseball, based on replacement-level production, would all say that a 10%-below-average hitter is worth more when consuming 600 plate appearances than 300 plate appearances.

This determination to penalize usage among below-average shooters seems rooted in WP's radical claim that the average level of efficiency has zero value. If 1-for-2, 5-for-10, and 15-for-30 shooting all have equal value, then clearly that value is zero. I'm not aware that Berri has ever offered an explicit argument for why 50% shooting has no value -- if he has, can you explain it? It seems to me that case requires demonstrating that the skill required to maintain league-average efficiency over 48 minutues is ubiquitous -- that any team should always be able to get the ball in the hands of an average shooter. (Of course, one has to overcome a number of uncomfortable facts, such as that about half the teams in the NBA fail to achieve this standard every season, and that teams with "zero-value" shooting tend to win about 41 of 82 games.)

And, I'm still interested in your thoughts as to why Wins Produced has a weaker correlation to shooting efficiency than PER, if Berri is right that a greater emphasis on efficiency -- rather than overvaluing rebounds vis-a-vis scoring -- is what makes WP different.
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 245
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 12:10 am Post subject: Reply with quote
And if average shooters are worthless, shouldn't the same thing go for average rebounders? But a player who never shoots and just rebounds at an average rate will do well with WoW. I think that was part of the point NickS was making in this post. Whatever criteria you enforce on shooting, you have to also enforce on rebounding.

And this got me thinking about the original point Charles made about PER here. I think he's right that PER is flawed in not counting an opportunity cost for a made basket, but at the same time it didn't seem like PER would be improved if you made only this change. I began to think that PER ended up working out kind of right, but for the wrong reasons. PER was inflating the value of made baskets, but only as a response to the inflated value of rebounds.

I did some research, and I again rediscovered another excellent post by NickS where I think he gets the right answer later in the thread:

NickS wrote:
If you deducted the full value of a possession from made baskets in PER all of a sudden you would find that the people it would rate very highly would be rebounders. Finding some balance between scoring and rebounding is necessary for the reason I laid out in the "two meanings of zero" post. Again, PER accomplishes that by setting a value for Rebounding, and then deflating it by adding extra value for made baskets.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 12:16 am Post subject: Reply with quote
Guy - How much you score matters in WP. The way I think about it is that there is a points scored benchmark and an efficiency benchmark, via fga, fta, ftm's, etc. A player is penalized if he doesn't match the benchmark for his position. So, if you have two players who both shoot a ts% of 55%, and one scores twice as many points as the other, he will produce more wins. What happens when you have two below average efficiency scorers is unclear to me, but I think you get penalized the more opportunities you use.

I think a lot of people get confused by WS, and the fact that you can shoot 20-40 on 2 pt fg's and not accumulate any WS. But people always forget to add in the effect of 3pters and fts. When you do that there is much more variation. I watch games calculating WS, and most player don't shoot over 50% from the field. But when you throw in fts and three pointers, that changes. And when you throw in position adjustments it changes again.

To me, usage is baked into WP, or at least as much as it needs to be. If you are a player who doesn't score points, its going to be harder to meet your overall target. If you don't notch assists ditto. You have to make it up in other parts of your boxscore. Yes, Steve Nash commits a lot of turnovers. But that's because he is accumulating assists at an even higher rate.

People maintain that WP has a problem evaluating high usage players. I don't really see that. Most of the top players in WP are scorers and high usage players. To me, it has always been much more clear that PER has a huge problem with evaluating low usage players. The fact that Dennis Rodman had a below average career PER and Eddy Curry is above average has always been my touchstone in that regard for me.

I don't know if PER correlates more closely with shooting efficiency. If you have done a study which shows that, post it and I will comment.

I can't say I really understand your replacement level player argument. I am not a sabermetrician, or a statistician for that matter. But I don't think its true that average efficiency has zero value, as you say above. A team of average players will win 41 games, which is a damn sight better than 0, and much better than what the Knicks will put up this year.
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3621
Location: Hendersonville, NC

PostPosted: Tue Dec 11, 2007 9:56 am Post subject: Reply with quote
The benchmark for shooting% that has >0 value cannot be static. In an 84-80 (avg score) playoff series, it may be that a 43% FGA is a 'good' shot.

In the '98 Finals, the Bulls forced the Jazz to shoot just 45% FG (down from 49% in the season). More aptly, Utah's TS% was off, from .567 to .506 .
Despite running 3 fewer possessions per game, Utah was forced into 1 extra turnover from their season rate.
Together, these resulted in an offensive efficiency drop of 17%, from their norm of 1.13, to .93 Pts/possession. Any shot better than about .390 eFG% was a 'better than average' possession.
That's assuming their offensive rebounding (26%) led to just their average (.506) scoring rate, with no TO and no 2nd OReb.

There were some playoff series ('80s-90s) between Phx and Den, in which a 50% FG try was almost a waste.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Tue Dec 11, 2007 10:54 am Post subject: Reply with quote
Flint wrote:
Quote:
To me, usage is baked into WP, or at least as much as it needs to be. If you are a player who doesn't score points, its going to be harder to meet your overall target.

A usage value is indeed baked into WP, and that value is zero. Usage is rewarded only for above-avg-efficiency shooters, and has negative value for below-avg-efficiency. Overall, usage has no net value, and WP has virtually no correlation with usage at the player level. And for half of the players your second sentence is demonstrably false: if you are a below-avg-efficiency shooter, the fewer points you score the easier it is to reach the overall target. A 48% TS% player is more valuable taking zero shots per game than taking 20.

Quote:
I don't think its true that average efficiency has zero value

Then you need to study WP more closely. An average efficiency player receives identical WP when taking 2, 10, or 20 shots per game. If 2*X = 10*X = 20*X, and I solve for "X," I get X=zero; don't you? This shouldn't surprise us: Berri says "efficient shooting" (by which he means above-average) creates wins and has positive value, while "inefficient shooting" causes losses and has negative value. So it follows logically that average shooting will have zero value.

Quote:
I don't know if PER correlates more closely with shooting efficiency. If you have done a study which shows that, post it and I will comment.


Dan Rosenbaum has studied this and found a WP-TS% r of .556, compared to .613 for PER-TS%. I believe his paper will be published fairly soon, so you (and Berri) will have the chance to study and rebut it. But just as a matter of algebra, it's easy to see that PER punishes inefficient shooting more than WP if you hold shot attempts constant. WP will value inefficient/high-usage players lower than PER, but will also value efficient/high-usage players lower than PER. So my guess is Rosenbaum's calculations are correct.

* *

To some extent, the two main criticisms of WP -- undervaluing scoring, overvaluing rebounds -- are just two sides of the same coin. If either claim is true, the other has to be at least somewhat true. But it is useful to consider them separately. While I believe Berri is wrong that usage has no value and there is no usage/efficiency tradeoff (basically the same claims), I don't think critics can claim to have proven that point definitively. And a weak version of Berri's argument -- that teams could improve by giving somewhat more shots to high-efficiency shooters -- is not inherently implausible.

On rebounds, however, there is no doubt that Berri's core assumption -- that all player rebounds equal a net additional rebound for his team -- is false. A player's REB total is highly influenced by that of his teammates. For example, if you run a correlation between the REB total for a team's top rebounder and the REB of his teammates, it's about -0.80. In fact, the negative correlation is so strong that the standard deviation for rebounding is actually much higher at the player level than at the team level. The difference between good and bad rebounding teams is fairly small, and that simply couldn't happen if the real differences in players' rebounding ability were as large as raw REB stats imply.

Berri's post yesterday on Garnett and the Timberwolves is a good example of how this gets Berri in trouble. WP estimates that KG was about 23 wins above average per year in his peak years (02-03 thru 05-06). About 45% of this comes from rebounding, as he averaged 330 rebounds/season more than an average F. But, while WP assumes this meant 330 extra rebounds for the T-Wolves, in fact the team was only 33 rebounds above average in these years. While KG was +330, his teammates were -297 -- what bad luck!

Interestingly, the now terrible KG-less T-Wolves are averaging more rebounds per game (42) than in Garnett's last two seasons (40). Isn't it surprising that Minn could improve its rebounding while losing KG's 330 boards? That would mean that the rest of the team has suddenly improved by 6 REB per game, an astonishing improvement. And if rebounding is so important, shouldn't we be shocked that a team can get much worse while improving its rebounding? Yet Berri doesn't even mention it.

You find the same pattern with Rodman. At his peak, 91-92 to 97-98, Rodman was +543 REB per year (vs. avg F). Yet his teams were only +162 overall. Rodman's teammates were a horrific -381 REB on average, and were well below average in every one of these seven seasons (covering 3 different franchises). Another great rebounder saddled with lame teammates. Bad luck yet again! (I haven't done the work, but I'm pretty sure you will find that virtually every great NBA rebounder has had similar misfortune.)

The final test of what matters in winning is whether an attribute is actually correlated with winning. What led Bill James to discover that stolen bases aren't very important for winning baseball games was observing that winning teams didn't have many more SBs than losing teams, and high-SB teams didn't tend to win many more games than low-SB teams. The same is true for rebounds relative to scoring in the NBA. The correlation between point differential and shooting (PPS) is about 3 times higher than the correlation between point differential and rebounding. However, WP is more highly correlated with rebounding than with scoring (and much more correlated with rebounding than PER). Berri's whole case is built on the notion of explaining team wins. But while rebounds don't explain very much of the variance in team wins in the NBA, they explain a lot of the variance in Wins Produced.


Charles: sorry to hijack your thread.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 4:46 pm Post subject: Reply with quote
Guy - I don't think I can give you satisfaction here. If everything that has already been written by Berri hasn't been enough to convince you, including his direct responses to you, I don't know what more I can offer. Reading your post, what I can say it that I don't think it is an accurate depiction of Wins Produced.

The way you are thinking about rebounding and the way it is treated by Wins Produced, for instance, strikes me as flawed. Specifically, your characterization of....

"Berri's core assumption -- that all player rebounds equal a net additional rebound for his team"

"Jason" had what I hope is a germane comment at 1042 today in the Garnett post. From that:

"WP" doesn’t say that a 10 rebound/48 player replacing a 15 rebound/48 player will result in a net increase in 5 rebounds 48. It says that the increase or decrease in wins will be roughly equivalent to the win probability contributed by the player-assigned stats. The model does not say that any isolated components must stay constant. It is important not to confuse the two."

I think actually reading the rest of that post by him, and the others he has written today might be a better response to what you posted above than what I can offer you.

My interest in basketball stats is a byproduct of reading the WOW and of work done in public policy, investing, and also a very serious poker hobby. I am not a basketball statistician or professional economist. While I understand most of the arguments made, I don't think I can really engage you at the appropriate level.

Re PER

I guess unfortunately I operate by sense of smell, like some on this board, and I think like most people who watch basketball. I just have a different sense of smell, for reasons which I have articulated in the past (and been ridiculed for, which was sort of hilarious). I am also a Knicks fan, which means I have been watching Eddy Curry for the past 2+ years.

Anyway, bottom line, PER tells us that Curry has been a better player than Samuel Dalembert and Tyson Chandler in his career. It tells us also that Curry has been better than Dalembert this year and as good as Chandler.

And that's kind of what it boils down to for me. Is Eddy Curry really an above average NBA center? Is he really as good as Tyson Chandler, currently anchoring the fourth best defense in the league. Is he really only slight worse (.5 PER) than Marcus Camby, currently anchoring the second best defense in the league? Is he really dramatically better than Ben Wallace, the center on the fifth best defense, as PER suggests?

PER would have us believe, it seems to me, that if you were to swap Curry for Camby, Wallace, Chandler, Dalembert, Oberto, Ilgauskas, or any other number of big men he scores more then, the Knicks wouldn't improve. That the only serious upgrade we could make at the center position would be Ming, Bynum, or Howard. And that's just, well, incredibly bizarre to me.

Wins Produced does not ask me to believe that. It tells me in fact that getting rid of Curry and adding any of those players would help the Knicks a great deal. That makes sense to me. Argument from anecdote I know, sorry not be able to give you anything else...
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 245
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 5:32 pm Post subject: Reply with quote
Flint wrote:
Anyway, bottom line, PER tells us that Curry has been a better player than Samuel Dalembert and Tyson Chandler in his career.


Flint. If you're going to criticize PER, you should at least get the basics right. PER doesn't make statements about who's an overall better player because it only measures what happens in the box score -- which ignores most defensive contributions. PER summarizes box score contributions only and estimates how many points those contributions would make for a typical team.

So you should say that PER tells us that Curry has been a better box score producer than Dalembert or Chandler -- which is a very different argument. Maybe you think that is incorrect as well, but at least you are characterizing PER correctly.

In the same way, PER does not say that Rodman was an average player -- another of your favorite examples. It says that he was an average box score producer. His defense would certainly have him ranked much higher. Again, maybe you think that box score analysis is incorrect, but you're not helping anyone by beating up on a strawman.

Quote:
PER would have us believe, it seems to me, that if you were to swap Curry for Camby, Wallace, Chandler, Dalembert, Oberto, Ilgauskas, or any other number of big men he scores more then, the Knicks wouldn't improve.


Again, PER makes no such claim. (Hollinger is frequently critical of the Knicks' interior defensive problems.) You've seriously misunderstood the scope of these stats. You need to re-approach all of these players and numbers from the mindset of measuring box score contributions only. Ben Wallace, who does so much off the box score, shouldn't be rated to his full actual value. That would actually be a warning flag that something is wrong.

So, of course, WoW also measures only box score contributions -- and not much of defense. If you find WoW is measuring players accurately in an overall sense, then this is actually a problem. That would imply that WoW is getting the right numbers for you, but for very wrong reasons -- the method is flawed.

Last edited by asimpkins on Tue Dec 11, 2007 5:56 pm; edited 1 time in total
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 5:56 pm Post subject: Reply with quote
Asimpkins - I have heard this argument before and I find it baffling.

Would it be accurate to say:

PER is designed to provide a summary statistic measuring a players production of box score stats. It is not designed to capture a player's actual contribution to winning basketball games for his team.
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 245
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 5:58 pm Post subject: Reply with quote
It would be correct to say:

It is not designed to capture a player's full contribution to winning basketball games for his team.

Re: Possession cost scheme

Posted: Fri Apr 22, 2011 2:25 pm
by Crow
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 6:00 pm Post subject: Reply with quote
Ok. What percentage of a players contribution does it ostensibly capture?
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 244
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 6:12 pm Post subject: Reply with quote
A good question, and a complicated one.

I couldn't give a definitive answer, but for the sake of discussion I would estimate that the box score doesn't represent somewhere between 25%-50% of the important things that happen in the game.

And I know you're more comfortable attacking PER than considering flaws in WoW, but PER is based on the exact same limited information as WoW is. Whatever shortcomings the box score represents, they apply to WoW just as much.

Neither is a measure of a full contribution to winning a game. I don't think even Berri would argue with that.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Tue Dec 11, 2007 6:24 pm Post subject: Reply with quote
And the defense rating he created to make up for the lack of information in boxscore, is so fairly distributed that doesn't give more valid information than the unadjusted +/- that some ratings use to compound with PER, or than PER counterpart.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 7:30 pm Post subject: Reply with quote
Mike G wrote in the player losses thread that his intention was to create:

"a win-based player metric that stood up to translation from one team environment to another."

Isn't that the goal here? Isn't that what an advanced player metric should be?

And given what you have said about PER, can it really even be considered an advanced player metric in that sense? It seems like a no to me.

Maybe that shouldn't be the goal, maybe it's an impossible goal, I don't know. But I do have trouble with the fact that PER is popularly perceived to be an accurate guide to a player's full value, when in fact, as you say, it isn't even actually intended to be that.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Tue Dec 11, 2007 7:48 pm Post subject: Reply with quote
I think nobody thinks is a full accurate guide, just that the logic is better and more acceptable, and doesn't try to overdo statistical methods like to do a team win regression to a stat whose action is shared in the game with another one that is not boxscored, that's is to overrate the stat.
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 244
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 7:53 pm Post subject: Reply with quote
Flint. I've stated it twice already, and you don't seem to have absorbed it. Maybe you are psychologically blocking out words that offer any criticism to WoW? But I'll try one last time.

WoW is based on box scores, just as PER is. Thus, WoW offers an incomplete picture of player contributions towards wins, just as PER does. In this sense, PER and WoW are the same. WoW is as much an advanced metric as PER is. Any criticism you level at PER on this matter applies equally to WoW. It is a feature of any box score only metric.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Tue Dec 11, 2007 8:02 pm Post subject: Reply with quote
Do you think that a proffessional statician who knows and applies the odds for a block to be defensive rebounded (applies to block only of course), can do the blind to the same odds between an OReb and a DReb. after a FGMissed?
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 8:10 pm Post subject: Reply with quote
Ok. Clearly this is a discussion that has happened before. I got it. What you are saying is that no method based on tracked box score stats can capture a player's full value. Only plus/minus data can do so, and only when proprietary techniques are applied.

That's fair enough, duly noted. I disagree of course.
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 244
Location: Pleasanton, CA

PostPosted: Tue Dec 11, 2007 8:35 pm Post subject: Reply with quote
I haven't said anything about plus/minus or proprietary techniques??

I can only take this as another sign that you aren't actually having a conversation with me... you're just here to preach the Wages of Win?
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Tue Dec 11, 2007 8:56 pm Post subject: Reply with quote
You believe you cannot paint a complete picture of a player's value using box score stats tracked for teams and individuals, correct?
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Tue Dec 11, 2007 9:57 pm Post subject: Reply with quote
Quote:
"Jason" had what I hope is a germane comment at 1042 today in the Garnett post. From that:
"WP" doesn’t say that a 10 rebound/48 player replacing a 15 rebound/48 player will result in a net increase in 5 rebounds 48. It says that the increase or decrease in wins will be roughly equivalent to the win probability contributed by the player-assigned stats. The model does not say that any isolated components must stay constant. It is important not to confuse the two."


Yes, Jason does make this point, but it's a very strange argument indeed. The idea, I think, is that if we replaced an average F on a team with KG, the team might not gain 330 additional rebounds -- or perhaps any at all -- but would still gain the 23 wins projected by WP. So how does that work? Are the rebounds a proxy for some other important contribution not captured in the boxscore stats? Hard to see what that could possibly be -- we're talking about 4 REB a game in phantom value here. And shouldn't Berri be able to tell us what that invisible contribution is, by tracking what happens to teams that add high-WP players? If rebounds don't go up, then team defense or something else we can measure will. Surely it's not just team spirit these rebounds are indirectly measuring?

Of course, this is a very convenient argument, in that it allows Jason to dismiss any assumption of WP which is shown to be implausible (like player rebound = net team rebound): "OK, maybe Rodman didn't really add 500 new rebounds for his teams, but trust me, he still created a ton of wins." But that doesn't mean he's wrong. The only way to test this is to see how well WP predicts future wins, especially when players change teams or MP. As far as I know, only Dan Rosenbaum and Dave Lewin have done that. Their analysis shows that WP does quite a poor job of predicting future wins, somewhat worse than PER and much worse than a revised WP that uses very different coefficients for REB and FGA. So, unless someone else finds that WP does in fact predict future wins more accurately, that seems to definitively establish that WP is a less "advanced" metric than PER (which isn't to say a still better mousetrap may not be out there).

Flint, I hope you will invite Jason to come over to this board and engage the APBR community at some point. He seems like a sharp fellow, and I think he would find it more challenging -- and interesting -- than lecturing the regular flow of newbies over at Berri's site on the revealed wisdom of WOW.
Back to top
View user's profile Send private message
asimpkins



Joined: 30 Apr 2006
Posts: 244
Location: Pleasanton, CA

PostPosted: Wed Dec 12, 2007 1:03 am Post subject: Reply with quote
Flint wrote:
You believe you cannot paint a complete picture of a player's value using box score stats tracked for teams and individuals, correct?


Yes.
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Wed Dec 12, 2007 12:29 pm Post subject: Reply with quote
Flint: Jason flies solo over at WOW today, with a post on the consequences of Houston losing Olajuwon for a chunk of the 1990-91 season. I think it may help clarify the discussion/debate over what Berri calls "diminishing returns," the idea that rebounds is largely -- though not entirely -- a zero-sum game. Jason believes the Olajuwon story refutes the diminishing returns idea, because the rebound rates for the 4 other starting players don't change much in his absence. But to me it's a perfect illustration of the zero-sum idea, because the team rebound rate was essentially unchanged without Olajuwon.

To be true, the zero-sum argument doesn't require that the other four starters obtain the rebounds when a star rebounder leaves the court. It simply says that the team -- including the star rebounder's replacement -- will get nearly as many rebounds no matter who is on the floor (within reason). And that's exactly what Jason's story illustrates. The fact that a 6'-8" bench player near the end of his career can replace a 7-foot all-star center, with no net loss of rebounds, is hardly a vindication of Berri's analysis -- it's more a refutation of it. According to WP, Olajuwon was creating over 4 rebounds per game above an average center, worth about 10 wins per season to his team. Thanks to Jason, we can see that wasn't really true.

Perhaps you can communicate this important clarification to Berri and Jason, so they will better understand the zero-sum argument. Even if they aren't persuaded, at least they can then engage the actual criticism being made.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Wed Dec 12, 2007 2:54 pm Post subject: Reply with quote
Guy - The point of the post is that the Rockets were able to replace Olajuwon's rebounding. Larry Smith was a better rebounder than Olajuwon in his career, and in that season, even at the end of his career, he was as good in terms of rebounds per 40 and rebound rate. But Olajuwon was a much more prolific scorer than Smith. That is the major difference between them (although blocks and steals look very significant also). Ergo, how hard is it replace scoring? If you substitute a low usage player for a high usage offensive player, how hard is it to make up the scoring slack. The answer, in this case, and you would argue the point of course, is not very hard. How hard is it in general to replace scoring, as opposed to rebounding? It's an interesting debate I think, and I know where I stand at the moment.

This is argument from anecdote, which I of course love, and you could probably employ it to your advantage by finding a case that makes your argument. This one doesn't though. It's not a test of the case that if you put in an average rebounder in Olajuwon's place, you wouldn't notice the difference in the teams rebounding margin, because they didn't replace him with an average rebounder.

A better example for you might be:

Eddy Curry is a bad rebounder, perhaps the the worst rebounding starting center in NBA history. He increased his minutes dramatically last year, his rebounding numbers got significantly worse from the year before, and the Knicks still improved their defensive rebounding mark. How does this work? That seems to be the gist of what you are saying. Other players simply picked up the slack, as Q, Lee, Balkman, etc did. What does this say about Berri's view of rebounding etc etc?

I confess to being a bit puzzled by that. But what does determine a team's rebounding margin if not the proven abilities of the players on the court? It seems logical to me that the more players you have on the court with a proven ability to rebound, the better your rebounding margin will be, and the more likely you are to win. And that is the logic employed by the WOW. That may be wrong, I would be happy to see the case against that statement. It may be fairly random, or depend on coaching, or the mysterious way the players work together on the court, i don't know. But I will need a bit of persuasion. While there is significant variation in rebounding margin in the NBA, player rebounding totals seem to be quite consistent across time and teams, when adjusted for age and pace. There are differences of course, diminishing returns as Berri says, but not huge differences. Jason Kidd has always been a great rebounding point guard, probably the best ever as a pure point. Last year, he averaged 8.9 per 40, beating his career high at age 33. Clearly this has something to do with his teammates, the absence of Jefferson and Kristic. And when you consider the fact that his WP was the best in the league last year, and significantly higher than what he had posted previously in his career, you take that into account somewhat. But that doesn't change the fact that Jason Kidd is a great rebounder and had a great season last year. Is it really a coincidence that the Nets are the third best defensive rebounding team in the NBA after being the ninth best last year? What exactly does this have to do with Kidd? And how much does it affect their win total?

And why is it that Mikki Moore's rebounding average didn't skyrocket last year stepping into the same void Kidd was operating in? His rebounds were actually lower, per 40 than in the two previous seasons.
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Wed Dec 12, 2007 4:02 pm Post subject: Reply with quote
Flint wrote:
Larry Smith was a better rebounder than Olajuwon in his career, and in that season, even at the end of his career, he was as good in terms of rebounds per 40 and rebound rate.....It's not a test of the case that if you put in an average rebounder in Olajuwon's place

Fair enough -- it's not a perfect test. And one anecdote of course can't prove anything. But to use this as an example of how diminishing returns is mistaken is just silly. You replace Olajuwon with an old, part-time player who's 4 inches shorter, see no loss of rebounds, and conclude "see, no diminishing returns"??? And the example does illustrate nicely the relative scarcity of rebounding talent vs. scoring talent. A guy who got zero starts the year before, 7 the next season, and never drew a million dollar salary can grab rebounds as well as an All-Star center. How scarce can this rebounding "skill" be?

Quote:
Ergo, how hard is it replace scoring?....How hard is it in general to replace scoring, as opposed to rebounding? It's an interesting debate I think, and I know where I stand at the moment.

This is indeed the central question, and Jason expressed the same view today: "I suspect that replacing someone’s scoring load is generally easier than replacing someone’s rebounding load. Again, an empirical question to be addressed." I don't understand why you, Jason, and Berri treat this as some great mystery. We can answer it easily by looking at team variance. The SD for Reb% is about .014, or 1.25 Reb/Game. The SD for PPS is around .035, or approximately 2.8 points per game. (People here with large datasets should weigh in and correct me if I'm off base.) So we know that teams vary much more in their ability to score than their ability to rebound. If scoring were easy to find -- i.e. less scarce -- than teams would use those players and scoring variance would decline. Conversely, if it was hard to find good rebounders, the teams with the "best" rebounders would dominate and rebounding variance would grow. The SDs tell us exactly how scarce the talents are. I'm surprised Berri doesn't know this. The much greater variance in scoring also means that differences in scoring ability explain much more of the variance in wins -- ostensibly Berri's bottom line -- than does rebounding ability.

Quote:
But what does determine a team's rebounding margin if not the proven abilities of the players on the court? It seems logical to me that the more players you have on the court with a proven ability to rebound, the better your rebounding margin will be, and the more likely you are to win. And that is the logic employed by the WOW. That may be wrong, I would be happy to see the case against that statement.

I'm sure there is some truth to this. But what you're missing is that teams have the ability to assign the rebounding role to certain players, giving them vastly more opportunities (which are finite) than their teammates. They may be asked to do this because of their rebounding skill, because of their deficiencies in other areas, or both. And not all teams do this the same way. The result is that some players can grab a lot of rebounds without really increasing the team total. And the fact that they play this role year after year, even when switching teams, doesn't tell us anything about how valuable that role is. (I don't understand why Berri keeps pointing this consistency out. Odd.) Similarly, putting a lot of good rebounders on the floor will likely increase rebounds a little, but not much.

The question is how strong is that relationship between player and team totals? For each extra rebound recorded by a given player, how much did that add to the team total? WP assumes it's one to one. I've shown you that KG's rebounds were almost perfectly offset by the "shortcomings" of his teammates, and that Rodman's teams actually only gained a small fraction of what he appeared to deliver. But those are just anecdotes, right? So I also pointed out the large negative correlation between the top rebounder's total and total team rebounds, and that if player rebounds were truly independent, the SD for team rebounds would be about 6 times as large as it actually is. Still you say you would be "happy to see the case", so I have to ask whether you really have an open mind on this issue.

Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Wed Dec 12, 2007 6:36 pm Post subject: Reply with quote
"And that's exactly what Jason's story illustrates. The fact that a 6'-8" bench player near the end of his career can replace a 7-foot all-star center, with no net loss of rebounds, is hardly a vindication of Berri's analysis."

He isn't making an argument about rebounding. His argument is about usage and scoring. The fact that Larry Smith, who was a terrific rebounder, replaced Olajuwon's rebounding, is entirely unsurprising. Just as it was sort of unsurprising in retrospect that a healthy Mutombo was able to replace Ming last year for a long stretch. Replace one rebounder with an equally good rebounder, nothing changes on that front, so what... I don't think that part of the anecdote proves anything either way. It certainly proves nothing about the scarcity of rebounding talent.

I think you ought to be looking for cases where a great rebounder went down with injury and was replaced by an average rebounder, and the rebounding margin stayed the same. That would be the way to support your theory, anecdotally.

That may have happened very frequently, I don't know. But it seems to me that usage is very easy to replace. It's pretty easy to add usage, its not easy to add inches of height and vertical leap. Arenas going down in Washington I think may provide an interesting anecdote. How much will they miss him, his 29.8 usage, and his 24 per from last year?

Can you look at Team Usage? Is there such a concept? And see how Usage changes when a high usage player leaves?

In regard to standard deviation, I don't know if your numbers are correct. I don't really think about PPS, unless your PPS include foul shots, which usually they do not. There is a lot of variation in efg%, but there is much less in ts%, at least that is my impression. The league average was 53.5% or so last year I think, and the vast majority of players fell within 10% of that on either side. i.e. between 48%-59% or so.

I don't have a great deal of statistical dexterity, but it seems unlikely, both from examining stats and thinking about it, that scoring ability is more scarce than rebounding. Unusually good rebounding generally requires both athleticism and height. Scoring does not. Anyone can score, but by and large you have to be big and highly athletic to rebound well. Basically I buy the "short supply of tall people" argument. (somewhat tautological argument there maybe) But also, how do you define great scoring. Is Iverson a great scorer? Is Nash, with a ts% of 65% last year? Was David Lee last year with a 65ts%?

re "The question is how strong is that relationship between player and team totals? For each extra rebound recorded by a given player, how much did that add to the team total? "

Again, I just don't think this is how to think about it. And I dont know how to respond. Its not about adding to the teams rebound total necessarily. I think I am going to still defer to Jason's comment about the probability of winning. I know it seems to you like a ghost in the machine, basically just a statement that given a player who is able to capture more rebounds, you are more likely to win, rather than having a clear connection. Can't really help you.

What i do know for sure is that Eddy Curry is not better than any starting center in this league. WP tells me that, PER doesn't.

I have to go try and decipher Rosenbaum's paper, and hopefully that will provide more clarity in this debate, and who knows, perhaps even some closure. (lol, don't think that's happening...)

Oh and re Rodman, what about his offensive rebounding? The guy averaged 6+ offensive rebounds per 40 even on a Bulls team with some, ahem, efficient scoring. I wrote a long post a while back pointing out that you can make a huge offensive contribution simply by being that good a rebounder and not turning the ball over very much. I think I may even have gotten some backup there from BChaikin if I remember correctly, (thanks for that btw if you read this...)
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Wed Dec 12, 2007 6:47 pm Post subject: Reply with quote
A question for a statician who can help me. If I do a team win regression on DRebs (I don't want just to forecast team wins from Rebs, but to weight the stat according to its real level of game action), is the defense on the shot considered a residual? and is not a residual when is a block only? In a block, why is not the DReb. the residual? Why are not points allowed credited to rebounders only?

If the defense on the shot can produce variations of about a 50% on the shooter's FG% depending on its contested level, isn't this still enough to be considered as a percent of block rather than a residual?

Can I sacrifice the player rating for me to be forcasty?
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 127


PostPosted: Thu Dec 13, 2007 12:26 am Post subject: Reply with quote
Flint wrote:
"He isn't making an argument about rebounding. His argument is about usage and scoring.

Now, Flint, I have to wonder if you actually read Jason's post. He wrote "so too some have argued that a significant portion of many a player’s rebounds are ‘taken’ from teammates. This suggests that if you remove the superior rebounder from the game, his teammates will start to grab some of those freed rebounds, negating some not-insubstantial part of the loss. Perhaps such returns have been witnessed elsewhere, but no such fortune befell Olajuwon’s regular front-court mates." I take your point that Smith-for-Olajuwon may not be a good test case either way, but the notion that it refutes the idea of diminishing returns, well, just can't be taken seriously. In focusing only on the four teammates, Jason reveals that he doesn't fully understand the argument he's trying to rebut. I think that's what often makes debates about WOW methods so unsatisfying -- Berri doesn't understand a criticism (because it doesn't fit in his framework), his reply therefore appears non-responsive, and his critics then feel he's being evasive or just knocking down a strawman. And round and round we go.....

Quote:
In regard to standard deviation, I don't know if your numbers are correct. I don't really think about PPS..... There is a lot of variation in efg%, but there is much less in ts%,

OK, let's use TS%. The team SD over past 2 years is .014, or about 2.6 points per game. This is almost exactly twice the SD for reb%. So can we agree that, in the actually existing NBA, the disparity in scoring talent is a) much larger than the disparity in rebounding talent, and b) explains a much larger proportion of the variance in team wins? (Why do I think your answer will still be 'no'?)

Quote:
I don't have a great deal of statistical dexterity,

Fair enough. But Jason and Berri should have enough training to understand these points. Send them over here for a good discussion.

Quote:
I have to go try and decipher Rosenbaum's paper, and hopefully that will provide more clarity in this debate, and who knows, perhaps even some closure. (lol, don't think that's happening...)

Give it a go. But without "statistical dexterity" I think you may find Rosenbaum's serious and complex analysis pretty tough to crack. Comments by Berri and Jason today seem to indicate they don't yet really understand it. Hopefully they'll eventually take the time to work through it and provide a thoughtful response.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Thu Dec 13, 2007 9:06 pm Post subject: Reply with quote
Guy - I agree the post does not refute the idea of diminishing returns in rebounds. I don't think it was intended to, which is what I was trying to say. I am sorry that it wasn't the response you hoped it might be. I think Jason mentioned it simply as an aside. This simply isn't a case that bears on that question.

I don't know about Jason, but I really don't know why DBerri would post here. This is after all a place where people say things like "Berri hasn't proven he's worthy to carry Dan or Dean or whomever's jock (so to speak), much less sit at the table with them as equals." (I think I actually joined the site in response to that comment.)

What about Arenas? He had a PER of 24 last year. Antonio Daniels had a PER of 15. Arenas had the fourth best adjusted +- last year, Daniels was at .46. Yet the Wizards are playing pretty well without him so far. WP indicates that Arenas was only slightly more productive than Daniels last year per minute. Therefore a WOW person could offer a cautious prediction that they won't miss him too much, that his absence will only cost them a couple games at most.

That is the kind of thing that makes me think the WOW really is an excellent metric, not just in results, but also in design. Thoughts?
Back to top
View user's profile Send private message
Guy



Joined: 02 May 2007
Posts: 127


PostPosted: Thu Dec 13, 2007 10:17 pm Post subject: Reply with quote
Flint wrote:
Guy - I agree the post does not refute the idea of diminishing returns in rebounds. I don't think it was intended to, which is what I was trying to say.... This simply isn't a case that bears on that question.

I think we mainly agree. Actually, it's also a rather poor test of the usage/efficiency question: if you replace a 17 FGA/51 ts% player on a 49 ts% team, why would you expect to see measurable change over 25 games (assuming a usage/eff tradeoff exists)? But Jason probably had a point in there somewhere.

Quote:
I don't know about Jason, but I really don't know why DBerri would post here. This is after all a place where people say things like "Berri hasn't proven he's worthy to carry Dan or Dean or whomever's jock (so to speak), much less sit at the table with them as equals."

Agreed -- it is too much to ask Berri to post here at this point. But he should provide a serious response to Rosenbaum's paper somewhere. His current position -- "Using residuals is wrong. End of discussion." -- is just not serious. He either knows he is wrong on this point, or hasn't even read enough of the paper to realize what Dan and Dave are actually doing. He diminishes himself with this posture.

And I would hope that you, Jason, and/or Berri will at some point try to back up the claim you've all made that rebounding talent is more scarce than scoring talent. I've provided a wealth of evidence to the contrary here, most importantly the simple observation that variation in scoring ability at the team level in fact explains team wins far more than rebounds. You haven't responded to any of it yet. Hopefully one of you will.

And regarding disrespectful treatment on this site, aren't you perhaps projecting a little? This summer, you said "Rosenbaum himself said he couldn't carry Berri's jock academically." That's an unfair translation of Dan's rather gracious concession that Berri's academic credentials as an economist are stronger than his. However, credentials doesn't equal ability. More importantly, as analysts of basketball statistics -- which economists really have no particular expertise in -- from what I've seen I'd say Dan is easily his equal (being charitable to Berri).
Back to top
View user's profile Send private message
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Thu Dec 13, 2007 11:11 pm Post subject: Reply with quote
Flint wrote:
What about Arenas? He had a PER of 24 last year. Antonio Daniels had a PER of 15. Arenas had the fourth best adjusted +- last year, Daniels was at .46. Yet the Wizards are playing pretty well without him so far. WP indicates that Arenas was only slightly more productive than Daniels last year per minute. Therefore a WOW person could offer a cautious prediction that they won't miss him too much, that his absence will only cost them a couple games at most.

That is the kind of thing that makes me think the WOW really is an excellent metric, not just in results, but also in design. Thoughts?

This is probably the sad part of what is being lost in this whole discussion. I would bet that many in this community agree that scorers are overrated by PER and by many NBA decision-makers. The irony is that most of Berri's biggest detractors are folks who would like to be allies with him. But Berri goes so overboard that his views end up being worse than conventional wisdom, which in a nutshell is what my paper with Dave Lewin shows.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
Charles



Joined: 16 May 2005
Posts: 134


PostPosted: Fri Dec 14, 2007 2:13 am Post subject: Reply with quote
Jacob wrote:
People worrying about the break-even marks in PER completely miss the point, and this includes our friend Berri. The talent level in the NBA dictates the break-even mark, not John Hollinger. A player with a 31% 2P% will never play significant minutes, let alone attempt significant shots. This is why it drives me crazy when people claim that "PER unfairly rewards 35% shooters who take 30 shots a game". Like these players actually exist in the NBA, the environment PER was designed for.


I understand your argument and, if the weights were the product of a study examining NBA data then I might agree with it. For instance, I have no problem with the logical incongruities Dan Rosenbaum’s statistical plus/minus displays when taken out of context. However, a system that is based purely on basketball logic should apply to any basketball environment. When all you have to go on is logic, then the logic should be air tight.

Any way, let’s plug in some data from this season and see how these elements of PER fare in the environment they were designed for. For the 2007-08 NBA season to date I have these values: Factor = .593, tmAST/tmFG (Phoenix) = .655, VOP = 1.034, DRBP = .735. Therefore, according to the PER formula, I get,

Two pointers are worth 1.61
Three pointers are worth 2.61
Missed field goals are worth -.76

Now, from 82games.com we can see that Amare Stoudemire has made 61 of 87 “inside” shots this season, while Steve Nash has made 38 of 79 three point attempts. Both plays produce better than 1.40 points per attempt which, in this league, is a devastating inside/outside combination. Here is a summary of the data (for the sake of clarity, I’m going to add eight more attempts and four more makes to Nash’s totals),

Code:
Att-Made Points Points/Attempt
Nash three-pointers 87-42 126 1.45
Stoudemire “inside” 87-61 122 1.40


Nash’s three pointers: 42 * 2.61 + 45 * -.76 = 76.9 credits
Stoudemire’s two pointers: 61 * 1.61 + 26 * -.76 = 78.4 credits

The result is that Nash has taken the same number of shots. Scored four more points. Provided nineteen extra offensive rebounding opportunities. And received less credit from PER. Is that acceptable?

It's not about theory or about players scoring 50 points while shooting 31%, it's about the way the numbers represent actual player data.
Back to top
View user's profile Send private message
Mike G



Joined: 14 Jan 2005
Posts: 3484
Location: Hendersonville, NC

PostPosted: Fri Dec 14, 2007 5:01 am Post subject: Reply with quote
Charles wrote:
... Nash has taken the same number of shots. Scored four more points. Provided nineteen extra offensive rebounding opportunities. And received less credit from PER. Is that acceptable?
...

Consider too that Amare has gotten 133 FTA, representing some (133*.44) 58.5 attempts. From these (104 FT), he's averaged 1.78 Pts/Att.

He's been fouled (2 FTA) just about once for each 5 FGA; probably a higher ratio on 'inside' attempts. Plus those and-1's.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Fri Dec 14, 2007 8:59 am Post subject: Reply with quote
Quote:
The result is that Nash has taken the same number of shots. Scored four more points. Provided nineteen extra offensive rebounding opportunities. And received less credit from PER. Is that acceptable?


Mathematically probably no, but the reality is the inside shot produce more fouls-FTs, less DR opportunities (changing the point o view)=less FGMissed. But, as deepak said that implicit reallity is probably a coincidency than an clear intend. The 3p shooter is obtaining an extra point (+1) when he scores, but when fails is (-0.7), the extra -0.7s are more than the extra +1s to balance with the 2p shooter. The 3p shooting is a good skill as a strategy, but not to be overrated too much. But if you do FGMissed=-1, you can change the reallity against the 3p shooter.

Last edited by Harold Almonte on Fri Dec 14, 2007 9:43 am; edited 1 time in total
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Fri Dec 14, 2007 9:15 am Post subject: Reply with quote
Quote:
Flint wrote:
What about Arenas? He had a PER of 24 last year. Antonio Daniels had a PER of 15. Arenas had the fourth best adjusted +- last year, Daniels was at .46. Yet the Wizards are playing pretty well without him so far. WP indicates that Arenas was only slightly more productive than Daniels last year per minute. Therefore a WOW person could offer a cautious prediction that they won't miss him too much, that his absence will only cost them a couple games at most.

That is the kind of thing that makes me think the WOW really is an excellent metric, not just in results, but also in design. Thoughts?

This is probably the sad part of what is being lost in this whole discussion. I would bet that many in this community agree that scorers are overrated by PER and by many NBA decision-makers. The irony is that most of Berri's biggest detractors are folks who would like to be allies with him. But Berri goes so overboard that his views end up being worse than conventional wisdom, which in a nutshell is what my paper with Dave Lewin shows.


Berri has a good approach about diminishing return, but when substituting some scorers only (but you can extract this from PER too, but with logic not with numbers), is not the same approach when substituting higher rebounders usagers, because he hides that the relationship DR/OR is the same at the team level like the individual level, and a matter of floor positioning advantage, even more than height (and this thing explains a 20% of team wins only if you don't overrate). The relationship difficult shots-scorer/easy shots scorers it does exist, but usage (quantity of FGAs) is not the way to search it. That correlation is circunstantial to some scorers and some teams, not everybody, and probably Arenas is not making teammates as better, he can't be made up.

Last edited by Harold Almonte on Fri Dec 14, 2007 10:40 am; edited 2 times in total
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Fri Dec 14, 2007 10:13 am Post subject: Reply with quote
A high rebounder usager (and every other shared and not shared stat high usager-the only shared stat in boxscore is AST) can make things easy for teammates doing the most of the rebounding job, but teammates are improving his stats letting him to do that thing they probably could some way make up. It's a relationship that linear metrics don't show very well. Is it as high this relationship as an 80% team / 20% player? probabbly no. Is it as high as the +/- distribution of credits? probably not either. But, who knows?

Last edited by Harold Almonte on Fri Dec 14, 2007 10:50 am; edited 2 times in total
Back to top
View user's profile Send private message
basketballvalue



Joined: 07 Mar 2006
Posts: 204


PostPosted: Fri Dec 14, 2007 10:26 am Post subject: Reply with quote
Flint wrote:

What about Arenas? He had a PER of 24 last year. Antonio Daniels had a PER of 15. Arenas had the fourth best adjusted +- last year, Daniels was at .46. Yet the Wizards are playing pretty well without him so far. WP indicates that Arenas was only slightly more productive than Daniels last year per minute. Therefore a WOW person could offer a cautious prediction that they won't miss him too much, that his absence will only cost them a couple games at most.


Flint,

This is an interesting point. As someone who's not delved into the details of WOW, I have a question: If we had said in September that Arenas would have been injured when he did (or out for the season if that simplifies things), and the minutes would fall as they have, what exactly would have been the prediction for the Wizards record? What if he hadn't been injured?

Thanks,
Aaron
_________________
www.basketballvalue.com
Follow on Twitter
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 860
Location: Washington, DC

PostPosted: Fri Dec 14, 2007 10:52 am Post subject: Reply with quote
The problem with the Arenas "analysis" being proposed here is that Daniels for Arenas isn't the only thing that's changed from last season. Haywood is playing better -- and was, before Arenas was re-injured this season. Blatche is playing far better than he did last season. Nick Young and Roger Mason have given the team someone who can hit a shot instead of Jarvis Hayes (edit to clarify -- Young, Mason, Blatche and Songaila are getting minutes that went primarily to Jarvis Hayes and Etan Thomas last season).

Well, what about Caron Butler, right? He's shooting more and better -- PROVING that increased usage doesn't mean diminished shooting efficiency, right? Except, there are other factors at work. Butler came into the season in the best shape of his life -- lighter, stronger, quicker. He improved his diet so his energy is better (cutting out high fructose corn syrup, for example -- the guy used to drink a two-liter Mountain Dew at every game, half before the game, half at half time). He also did something else this offseason -- he spent a lot of time working with new Wizards assistant coach and shooting instructor Dave Hopla.

So yeah, the Wiz may end up with a record this season that's similar to the record they had last season. But it's a whole lot more complex than swapping out Arenas for Daniels and saying -- see, the team doesn't miss Arenas much.

Last edited by kjb on Fri Dec 14, 2007 1:41 pm; edited 1 time in total
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
HoopStudies



Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA

PostPosted: Fri Dec 14, 2007 12:10 pm Post subject: Reply with quote
kjb wrote:
The problem with the Arenas "analysis" being proposed here is that Daniels for Arenas isn't the only thing that's changed from last season...


This is something that came up in conversation with Mike G -- it's easy to find anecdotes where one methodology looks good or bad. Maybe they look surficially good, but look worse upon inspection (as kjb suggests here), or vice versa.

Anecdotes should be collected as examples to be explained. Collect them all in one place. Arenas goes down this year. Joe Johnson switches from Phoenix to Atlanta. Steve Nash goes to Phoenix from Dallas. Ron Harper joins the Bulls from the Clippers. The Lakers' superstar edition of 2004.

Single anecdotes have driven me crazy as cherry-picking or, as I somewhat regretfully called it, "smearing." WoW does some of this cherry-picking, too, looking at simple personnel reasons for a team to underperform or overperform. The good thing is that it does look at most every team. It gets old, but it's actually useful to read the different stories. Today's blog mentions the Indiana Pacers and how the key was Jermaine O'Neal before the season, but he has been awful, according to WP, but the team is still ok -- an admission that the basic method isn't working in that case. Other articles are more proud suggestions that it is working. Basically a ton of case studies around one metric.

What makes sense is to identify good case studies, add them to a long list, and test more than one metric. Dan and Dave's Pot is something along this line, but broader. I read part of it yesterday and understand better what it was doing. Gotta find time to read the whole thing. The test comparing to actual team wins seems like a fair one; because of concerns I have about adj +/-, I would rely less on that. But, as broad as their study is, I wonder if there is value in just looking at what seem to be the good anecdotal case studies. I have generally found this to be better than looking at an overall population for quickly drawing inferences.

Just interjecting some broader thinking...
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
Charles



Joined: 16 May 2005
Posts: 134


PostPosted: Fri Dec 14, 2007 1:36 pm Post subject: Reply with quote
I don’t want to interrupt the Arena’s discussion because it is quite interesting.

However, going back to the original topic, I would still like to see an advocate of the FGA-FGM method explain how there can be no possession cost for a made field goal.

The NBA is not like MLB where a successful plate appearance has no “outs” cost. It’s not like three on three “make it and take it” playground ball. In fact, PERs accounting would be appropriate under those rules. But, in the NBA you have to give up possession after you make a shot. There is simply no reason to subtract FGM from FGA when assigning possession cost.

Re: Possession cost scheme

Posted: Fri Apr 22, 2011 2:27 pm
by Crow
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Fri Dec 14, 2007 1:36 pm Post subject: Reply with quote
DeanO's point about adjusted +/- is ironic, because if we send this off to a high profile economics journal, we will probably need to strip out all of the team wins stuff. The identification is much more complicated with team wins and that group would find the adjusted +/- much easier to follow (even though they won't know anything about +/- prior to reading the paper). But I do understand why the team wins stuff would be more convincing for this group (and for lots of sports economists, for that matter).
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
Charles



Joined: 16 May 2005
Posts: 134


PostPosted: Fri Dec 14, 2007 1:56 pm Post subject: Reply with quote
I agree with the above post. What is currently lacking is good measures against which to test data. Adjusted plus/minus is a major step in that direction. I think people are much too concerned with the noise at the player/season level. As Dan points out in the paper, the fact that the correlations vary as much as they do should alleviate that concern.

Now if someone who calculates adjusted plus/minus would make it publicly available...
Back to top
View user's profile Send private message
Charles



Joined: 16 May 2005
Posts: 134


PostPosted: Fri Dec 14, 2007 2:13 pm Post subject: Reply with quote
One more attempt. Rather than defending PER’s methodology, perhaps someone could criticize, what seems to me like the obvious procedure: award the points scored and account for the value of lost possessions (attempts – missed shots * offensive rebounding percentage ) * value of possession. From the above example.

Code:
Value of lost
Points possessions Net
Nash three-pointers 126 76.4 49.6
Stoudemire “inside” 122 82.4 39.6


That seems like the correct result. If you take the same number of shots, but produce four more points and 19 extra offensive rebounding opportunities, that should be worth just about ten points. Of course, Stoudemire should get credit for the extra free throws he draws, but those are already counted elsewhere.

And, if you want to make adjustments for assists or other shot creation factors that’s great, but why would they affect calculations which they are based on the same number of shots attempts?
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Fri Dec 14, 2007 3:01 pm Post subject: Reply with quote
Guy - If I have been disrespectful to anyone on this site, I regret that.

re "that rebounding talent is more scarce than scoring talent." I just don't know how to engage with you on this point. To me, what I am saying is that you have to look past a player's scoring profile to see how good he is. It's more than rebounding, a player's value is determined by much more than his ts% and volume (even if his salary is not.) And that is true even if the variance in ts% is the most decisive aspect influencing team wins, which is what I gather you are saying. To me, you have to look at everything a player does, his ability to gain and maintain possession, avoid fouls, help his teammates, block shots, steal the ball, and participate in team defense.

I don't feel like PER does that, a feeling based on looking at ratings, and not from deconstructing the metric itself as Charles is trying to do in admirable fashion. Look at Kobe and Michael at age 28, the comparison suggested at Basketball Reference for the player comparison machine. They have exactly the same scoring profile: same volume, same ts%. And their PERs are very similar, Jordan's is 6% higher. However his wp48 is .392 v Kobe's .248. And that's because his ancillary stats are much better. The evaluation that Jordan was much better than Kobe makes sense to me. The PER rating doesn't. And I don't think the difference in their abilities as defenders is what is missing.

Basketball Value -

DB has posted on the Wizards twice recently:

The preview piece before the year started was at:

http://dberri.wordpress.com/2007/10/01/the-anti-nets/

And he recently did a piece on the Wizards "survival" addressing how it is that the Wizards are over .500 this year:

http://dberri.wordpress.com/2007/11/23/ ... s-survive/

Hopefully he answers your questions.

KJB - In response to your question about Butler, I would agree he is playing better. In that post there is chart showing how well each Wizard had performed up until the point it was written, and he was at .204, which is a small improvement on th .178 he posted last year, so about 12% better. His PER this year though is at 23, compared to 18.3 last year. You are correct that other things have changed. Berri notes a big difference in Haywood's rebounding, and he also notes Blatche's play. Anyway, read on...

Charles - Sorry for getting in the way of your thread, carry on...
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Fri Dec 14, 2007 3:12 pm Post subject: Reply with quote
Quote:
deconstructing the metric itself as Charles is trying to do in admirable fashion.


Charles hasn't said, how is he going to distribute the team possessions gained after a FGMade when the ball is inbounded, which cancels this punishment. How would you prefer Charles, by minutes or by attempts? What do you think is the fairest way, and why? Would you prefer this thing stays without any credit like rebounds stay without any punishment?

Last edited by Harold Almonte on Fri Dec 14, 2007 3:50 pm; edited 1 time in total
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 864
Location: Washington, DC

PostPosted: Fri Dec 14, 2007 3:26 pm Post subject: Reply with quote
Flint wrote:


KJB - In response to your question about Butler, I would agree he is playing better. In that post there is chart showing how well each Wizard had performed up until the point it was written, and he was at .204, which is a small improvement on th .178 he posted last year, so about 12% better. His PER this year though is at 23, compared to 18.3 last year. You are correct that other things have changed. Berri notes a big difference in Haywood's rebounding, and he also notes Blatche's play. Anyway, read on...


I read Berri's blog daily. My point was really in response to the notion that the Wizards don't miss Arenas (much). I don't think the evidence we've seen so far proves that point. There are other factors at work that can be quantified. I think that a healthy Arenas in place of Daniels would likely mean more wins for the Wiz so far this season. The Wiz haven't had a healthy Arenas, however.
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Fri Dec 14, 2007 3:58 pm Post subject: Reply with quote
"My point was really in response to the notion that the Wizards don't miss Arenas (much)."

I agree. The general point I was making is not that they won't miss him, but that they will miss him less than one might think, given his PER last year, his adjusted +/- last year, and those numbers for the players who are taking a lot of his minutes.

You can't lose your best player and be unaffected. But you won't see the kind of collapse like that caused by Lebron's absence. And that is because Arenas is less productive than commonly believed and because his replacement is more so. That is not the case in Cleveland.

Varejao returned last night, had a very solid outing. The Cavs won, which is extremely strong evidence that Varejao is more important than James.

just kidding....

It is interesting though that Varejao was 22nd in adjusted +/- last year. If he is as good a defender as indicated on 82games, that would give him a significant boost in his WP. It's nice to see two metrics agreeing on a player...
Back to top
View user's profile Send private message
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Fri Dec 14, 2007 10:58 pm Post subject: Reply with quote
HoopStudies wrote:
kjb wrote:
The problem with the Arenas "analysis" being proposed here is that Daniels for Arenas isn't the only thing that's changed from last season...


This is something that came up in conversation with Mike G -- it's easy to find anecdotes where one methodology looks good or bad. Maybe they look surficially good, but look worse upon inspection (as kjb suggests here), or vice versa.

Anecdotes should be collected as examples to be explained. Collect them all in one place. Arenas goes down this year. Joe Johnson switches from Phoenix to Atlanta. Steve Nash goes to Phoenix from Dallas. Ron Harper joins the Bulls from the Clippers. The Lakers' superstar edition of 2004.

Single anecdotes have driven me crazy as cherry-picking or, as I somewhat regretfully called it, "smearing." WoW does some of this cherry-picking, too, looking at simple personnel reasons for a team to underperform or overperform. The good thing is that it does look at most every team. It gets old, but it's actually useful to read the different stories. Today's blog mentions the Indiana Pacers and how the key was Jermaine O'Neal before the season, but he has been awful, according to WP, but the team is still ok -- an admission that the basic method isn't working in that case. Other articles are more proud suggestions that it is working. Basically a ton of case studies around one metric.

What makes sense is to identify good case studies, add them to a long list, and test more than one metric. Dan and Dave's Pot is something along this line, but broader. I read part of it yesterday and understand better what it was doing. Gotta find time to read the whole thing. The test comparing to actual team wins seems like a fair one; because of concerns I have about adj +/-, I would rely less on that. But, as broad as their study is, I wonder if there is value in just looking at what seem to be the good anecdotal case studies. I have generally found this to be better than looking at an overall population for quickly drawing inferences.

Just interjecting some broader thinking...

I find the whole argument in this post interesting. It appears that DeanO is arguing that rather than doing systematic analysis using as many observations as possible, it might be better to focus on a few case studies. This might be reasonable once we figure out how to put standard errors around what we learn from these case studies. But until then I suspect that the confidence intervals in most cases are really huge - so huge as to not be able to rule out any reasonable hypothesis. Most of the issues we debate about are subtle enough that there won't be enough power in case studies to learn much. The case studies are very effective in selling a particular theory or finding, but I am skeptical of their value in evaluating them.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
Guy



Joined: 02 May 2007
Posts: 128


PostPosted: Fri Dec 14, 2007 11:18 pm Post subject: Reply with quote
HoopStudies wrote:
Anecdotes should be collected as examples to be explained. Collect them all in one place.

And if we collect enough, we can call our collection "data." Smile

Seriously, the confidence level issue Dan raises is legitimate. On the other hand, talent differences in the NBA are so large relative to other sports, and effective sample sizes so large (90+ scoring opportunities per game), that it wouldn't shock me if you could learn things from some fairly small samples. But at least as big a problem is how you select your anecdotes. For this to work at all, they have to be selected randomly from whatever team/player category you want to study (big rebounder moves to high-rebounding team; high usage player injured; whatever). But what's likely to happen instead is that anecdotes get selected because they appear to support a belief of the person submitting the anecdote (or they're unusually memorable, perhaps for some circumstances that make it atypical of your category). So you have to give some thought to avoiding selection bias, even if you can overcome sample size issues.
Back to top
View user's profile Send private message
HoopStudies



Joined: 30 Dec 2004
Posts: 705
Location: Near Philadelphia, PA

PostPosted: Fri Dec 14, 2007 11:56 pm Post subject: Reply with quote
Guy wrote:
HoopStudies wrote:
Anecdotes should be collected as examples to be explained. Collect them all in one place.

And if we collect enough, we can call our collection "data." Smile

Seriously, the confidence level issue Dan raises is legitimate. On the other hand, talent differences in the NBA are so large relative to other sports, and effective sample sizes so large (90+ scoring opportunities per game), that it wouldn't shock me if you could learn things from some fairly small samples. But at least as big a problem is how you select your anecdotes. For this to work at all, they have to be selected randomly from whatever team/player category you want to study (big rebounder moves to high-rebounding team; high usage player injured; whatever). But what's likely to happen instead is that anecdotes get selected because they appear to support a belief of the person submitting the anecdote (or they're unusually memorable, perhaps for some circumstances that make it atypical of your category). So you have to give some thought to avoiding selection bias, even if you can overcome sample size issues.


The concepts were addressed in Blink. Basically, there are tons of things that we CAN look at, but the key things are the ones you should look at. I think the Blink example was from predicting what couples would survive. You can build a statistical model using all the data they collected in their interviews, but only a few pieces of info end up significant. The interviewers started off looking at EVERYTHING and couldn't predict what couples would survive. But after many interviews, they'd figured out what the characteristics were, even if they couldn't say it explicitly, and they could predict much much better what couples would survive.

What I'm saying is that some of our anecdotes are getting at the significant things. Focusing on those can potentially help us get at answers more quickly.

(This is also shown in a more philosophical way in some poli-sci references I've seen. There are "experts" who know how to predict things and averaging those expert predictions does very well predicting the future, but the problem is that there are lots of people who aren't experts but are strongly correlated to the experts. If you use all the people (experts and non-experts) and average them together, you can end up with biased estimates because some experts have more followers than others.)

The problem is getting the good anecdotes or getting the right experts. It's so much work that people don't typically do this...

Heading way too far down Philosophical Avenue...
_________________
Dean Oliver
Author, Basketball on Paper
The postings are my own & don't necess represent positions, strategies or opinions of employers.
Back to top
View user's profile Send private message Visit poster's website
Dan Rosenbaum



Joined: 03 Jan 2005
Posts: 541
Location: Greensboro, North Carolina

PostPosted: Sat Dec 15, 2007 8:07 am Post subject: Reply with quote
HoopStudies wrote:
The concepts were addressed in Blink. Basically, there are tons of things that we CAN look at, but the key things are the ones you should look at. I think the Blink example was from predicting what couples would survive. You can build a statistical model using all the data they collected in their interviews, but only a few pieces of info end up significant. The interviewers started off looking at EVERYTHING and couldn't predict what couples would survive. But after many interviews, they'd figured out what the characteristics were, even if they couldn't say it explicitly, and they could predict much much better what couples would survive.

What I'm saying is that some of our anecdotes are getting at the significant things. Focusing on those can potentially help us get at answers more quickly.

What this points out is that anecdotes/case studies can be very helpful in making the right simplifying assumptions for a statistical model and in some cases they may point out when it will be difficult for a statistical model to do better than "intuition and human judgment." In a nutshell that is the thesis of my paper with Dave Lewin. But it sounds like the ultimate evaluation of predictive ability was still done using statistical analysis. Maybe another way of saying the same thing is that anecdotes/case studies can be very helpful in shedding light on how a statistical model is doing poorly, but they are not particularly good at evaluating whether it is doing poorly.
Back to top
View user's profile Send private message Send e-mail Visit poster's website Yahoo Messenger
schtevie



Joined: 18 Apr 2005
Posts: 408


PostPosted: Sat Dec 15, 2007 11:25 am Post subject: Reply with quote
It seems to me that testing theories on how well they explain particular sets of "anecdotes" would be a productive exercise in persuasion. These particular sets would be defined by roster changes involving statistical outliers. And there are rigorous ways to define such sets (i.e. cases where there are net X standard deviations in variable Y).

And I think we could appropriate as a working title for such a procedure a familiar term: the laugh test. Because if a theory can't give a plausible result for extreme cases, well.....hee hee.
Back to top
View user's profile Send private message
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Sat Dec 15, 2007 3:18 pm Post subject: Reply with quote
Although those examples of players replacement have some random, I think we can, specially from superstars and high usage players replacements, understand about the replacement performance and the diminishing return. How much from a player's skills (in quantity and in quality) can be replaced by another player, or by a set of players inside the team. How much room a player has to improve his ordinary stats when needs to replace a supposed superior player (Diaw-Amare). Weather two players with identical stats or ratings could be replaceable or not each other (the five Rodmans team example), etc. But that just carry us to a dilemma: What kind of rating do we need, a season performance (inside his context) player rating, or a skill (all contexts) player rating. Because, like stop% and Drtg., they seems to be similar, but sometimes they aren't.

I think it has some link with the magic word: predictibility.
Back to top
View user's profile Send private message
Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Sat Dec 15, 2007 4:25 pm Post subject: Reply with quote
Schtevie - Nice shout out from Yglesias re your Celtics projection. What kind of projection would you have made for the Wizards without Arenas?

And speaking of which, the Arenas experiment has taken a new twist, with Daniels out for 2-4 weeks. The drop-off was slight between Arenas and Daniels, but the difference between Daniels and the other guards on the team seems very large, although there is really only sketchy data on them. Will be interesting to see how it works out.
Back to top
View user's profile Send private message
basketballvalue



Joined: 07 Mar 2006
Posts: 208


PostPosted: Sat Dec 15, 2007 10:20 pm Post subject: Reply with quote
Charles wrote:
Now if someone who calculates adjusted plus/minus would make it publicly available...


Charles,

I am happy to say this is something I am in the process of working on adding to BasketballValue.com. It's slow going with my day job, but I'm hoping it won't be too far in the future....

Flint,

Thanks for the pointers. I'll take a look.

Aaron
_________________
www.basketballvalue.com
Follow on Twitter

Flint



Joined: 25 Mar 2007
Posts: 112


PostPosted: Thu Dec 20, 2007 9:13 am Post subject: Reply with quote
Berri reviews the loss of Agent Zero, with some thoughts on usage mixed in....

http://dberri.wordpress.com/2007/12/20/ ... gent-zero/
Back to top
View user's profile Send private message
kjb



Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC

PostPosted: Thu Dec 20, 2007 10:00 am Post subject: Reply with quote
Here's what I posted over at Berri's blog:

The usage/efficiency question is an interesting one. I’m not sure this is going to be a great test case (although I’m sure folks will try to use it). The key issue in the usage/efficiency question is one of ability — can a player increase his usage while maintaining efficiency? However, the fact that Butler is doing it now does not necessarily mean he could have done it last season. There’s plenty to suggest that Butler has actually improved — that his skills have not remained static.

Specifically, Butler improved his workout regimen and came into the season lighter, stronger and more flexible — in the best shape of his life. He improved his shooting by working with the team’s new assistant coach and shooting instructor Dave Hopla. And, he improved his diet by cutting out stuff like high fructose corn syrup. Last season (and throughout his playing career) he drank a 2 liter Mountain Dew during each game (half before the game, half at halftime). This season he’s cut that out and his stamina is improved. In other words, the fact that his efficiency has improved with more usage THIS season does not prove that he could have done it at any time previously. The evidence suggests that there has been actual improvement in his physical capabilities and basketball skills.

The comparison of shots per game doesn’t really address thinking about shot creation. Teams are going to take shots because of the shot clock. All shots are not created equal, however. Some are more difficult, some are easier. I think one of the perceived benefits of high-usage “stars” is that they have the ability to attempt shots while well-defensed that have some chance of going in where teammates with lesser ability would have less of a chance of making a similar shot.

The thinking here is that when the shot clock is running down on a vigorously defended possession, SOME shot (even a shot that has a 1-in-10 chance of going in) is better than NO shot. Someone has to take the shot, and the ball usually ends up in the star’s hands in that situation. The issue is not (and never was) that removing a guy who puts up 20 FGA per game will reduce his team’s FGA by 20 per game (or some other number). It’s that removing his shot attempts then pushes the possessions (even the more difficult ones) onto other players and a drop in efficiency results.

I use the term “efficiency” intentionally, because the drop may not necessarily come in shooting percentages. Someone at APBRmetrics posted at some point that the effect may show up in increased turnovers. Dean Oliver posted his finding (though he did not describe his method in any detail) that for every 1% increase in possession usage, there’s a 0.6 drop in efficiency (using Dean’s offensive rating measure).
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Harold Almonte



Joined: 04 Aug 2006
Posts: 616


PostPosted: Thu Dec 20, 2007 10:21 am Post subject: Reply with quote
Good comments. That post doesn't say that Arenas is a "PG" with less than 6 assists/g, and unlike A.I., or Nash, etc, he could be just creating shots exclusively for him, and not making anybody better at scoring. Haywood is not becoming better with or without Arenas, because he is a center scoring dependant (wouldn't he "better" if were assisted more?), but Buttler had skill-room to improve his scoring with or without Arenas.