RealGM

Obviously, neither are perfect measures (or even adequate in some cases), but which correlates better with actual player ability?

win shares per 48 is the best measure

The way Wins Produced gives such huge weight to rebounds, I'm inclined to side with Win Shares.

Win Shares is an earnest attempt to apportion player credit for team wins.

WP is a psychology experiment to see how many people can be persuaded to believe something, no matter how absurd it is.

Wow, that's incisve, jicama. Do you get paid big bucks to write such well-argued opinion copy as that? Facts, reasoned argument, detailed evidetiary analysis, that post had it all!

With respect to the good Doctor, I actually quite like Wins Produced despite (or perhaps because of) its counter-intuitive results.

Rebounds are a tricky beast to quantify the value of (like assists). A missed shot will generally mean a rebound for SOMEONE, and positioning on the floor means that the likely rebounder will be a defensive player. To what extent do you value the individual ability to rebound and/or prevent the opposing team from rebounding?

Personally, I feel that dictating the terms of possession is quite important, which is what rebounding is all about. More rebounds equals more possessions (and/or more/few shooting opportunities per possession, depending on how you conceptualise it). More opportunities plus better efficiency will generally equal a win, all things being equal.

The other criticism of Wins Produced is that it overly privileges low-use, high efficiency offensive players as compared to high-volume, somewhat less efficient shooters like Kobe and AI.

Unlike PER, it doens't value shot creation per-se, which it's possibly a weakness (although I don't think gunners necessarily SHOULD get a free pass).

Don't quite know how well winshares track to actual wins, but as far as I'm aware wins produced generally follow quite closely.

EDIT: Berri's own views of its weakness are:

Dave Berri wrote:Dr. Berri's own suggestions for possible weaknesses
1. Wins Produced does not give credit to players for creating shots. This is generally perceived as the biggest weakness. This perception is based on two beliefs. a) it is difficult to get a shot off in the NBA and b) the more you shoot the lower your efficiency. I am not sure there is much empirical evidence behind either proposition. Plus, if you give credit for taking shots then inefficient shooters will look better. Nevertheless if you believe shooting is difficult, Wins Produced will disappoint. And connected to this point…if you believe scoring is the most difficult thing a player does in the NBA, and therefore scorers are the most valuable, Wins Produced will disappoint also.

2. How vs. Why: Wins Produced tells you how productive a player is, but it does not tell you why. Performance is impacted by age, injury, roster turnover, coaching (in a few cases), and the productivity of teammates. The latter issue is the idea of diminishing returns. The more productive your teammates, the less productive you will be. This is a real effect in the NBA, although in the aggregate it is rather small. Still, these effects are not part of Wins Produced and these issues can impact what we see in the future from a particular player.

Those really looking to understand the cons of Wins Produced, should check out this conversation:

http://sonicscentral.com/apbrmetrics/vi ... sc&start=0

There's a lot to it, it's been a while since I've read it, and looks to me like more conversation was added.

To give the gist:

1) The heavy hitters in the community don't understand why WP is calculated like it is, and when they ask the questions they used to asking in the statistics field, Berri responds with a brush off. This isn't something where Berri's a genius and the other guys are morons. The guys he's talking with are statisticians with as much education as he has, and considerably more credibility in basketball, and they feel like they're asking basic questions and not getting reasonable answers.

2) Berri indicates that WP was designed to correlate with actual wins, but others point on that you can do that with basically any stat. They are alarmed that that when you use WP to actually predict future wins, it actually does a worse job than if you go simply take a basic stat like points scored, which itself is much worse (as you'd expect) than pretty much any of the major formulas.

With all that said, like I said, I haven't followed this closely in a while. If people have an update, I'd like to here it. I have to say, I'm honestly quite surprised that Berri's work has achieved such prominence given the way rest of the statistical community - which is very supportive, and typically pretty tolerant of new ideas - reacted with such a decisive thumb's down.

rrravenred wrote:Rebounds are a tricky beast to quantify the value of (like assists). A missed shot will generally mean a rebound for SOMEONE, and positioning on the floor means that the likely rebounder will be a defensive player. To what extent do you value the individual ability to rebound and/or prevent the opposing team from rebounding?

Personally, I feel that dictating the terms of possession is quite important, which is what rebounding is all about. More rebounds equals more possessions (and/or more/few shooting opportunities per possession, depending on how you conceptualise it). More opportunities plus better efficiency will generally equal a win, all things being equal.

The other criticism of Wins Produced is that it overly privileges low-use, high efficiency offensive players as compared to high-volume, somewhat less efficient shooters like Kobe and AI.

Unlike PER, it doens't value shot creation per-se, which it's possibly a weakness (although I don't think gunners necessarily SHOULD get a free pass).

Don't quite know how well winshares track to actual wins, but as far as I'm aware wins produced generally follow quite closely.

Let's take a look at the case of Marcus Camby.

Here's how WP shows the Nuggets in Camby's last 3 years:

'06 - Camby leads the team in WP despite playing only 1800 minutes, whereas other players play more than 2900.

'07 - Camby leads the team 17.6 WP. 2nd best on the team is 5.6.

'08 - Camby leads the team 21 WP, 2nd best is 9.5.

So he's way, way, way better than everyone else right? What happens when he leaves the next year? Not only does the team get better, but it starts outrebounding opponents which it was not doing before.

I understand thinking that maybe Berri's on to something with his emphasis on rebounding, but I honestly don't know how anyone can look at that and think Berri is valuing individual rebounding accurately. In the stat community there's something called "the laugh test" - if you get a result that's so obviously wrong any objective observer will laugh, then you need take gain some humility and either correct the algorithm, or explain the limitations of the algorithm. This result is frankly as embarassing as anything I've ever seen from a major stats name, and Berri is the least willing of any of the major stat guys (except Winston if you include him) to embrace humility. Yes, he shows some as in what you quoted, but on the whole he's very cocky, and even in his humility, he doesn't address the obvious problem explicitly.

Re: PER favoring chuckers, that idea needs to be tempered. Here's the issue that Berri's refering to: PER (and similar stats), penalize a guy blowing a possession. A missed shot, doesn't always blow a possession, so PER penalizes it based on an estimate for how often a missed shot blows a possession. Berri points out correctly that a player's PER will keep increasing with volume even at pretty low FG% which seems like a weird way to do a stat because you could theoretically inflate your PER while doing stuff that hurts your team. Practically though this isn't much of an issue. Fans of players who score at high volume but not great efficiency are constantly trashing PER because such players don't come off as well in the stat as the fans think. The typical basketball viewer thinks more highly of Kobe than PER does for example, and up until recently, thought WAY higher of Iverson than PER did.

Getting back to the estimate for how often a missed shot leads to a blown possession, I want to make sure it's clear what that means, because other than the problems with defense, the use of estimates are what I consider to be the biggest weakness in PER. If the box score actually listed what shots led blown possession, we'd have a really great idea how much a player's chucking was hurting his team. Undoubtedly, if a player started chucking worse, and defenses saw it coming, the offense would be less able to get offensive rebounds, and it would all get factored in to PER. PER doesn't use such fine grained information, it estimates. This is a weakness, but it's not something caused by weak logic. It's not something that in practice rewards chuckers, it's something that simply adds margin of error to algorithm.

One last note: It's funny that Berri says it may be a flaw of his formula that it doesn't reward shot creation - because this is actually a problem with PER. PER seeks to give some of the credit for scoring to the assister - but it doesn't know whether any particular bucket was assisted, so it docks all scoring equally. This really hurts the players that are basically always creating for themselves and others (distributors - who also are the guys who are most likely to be acting as coach on the floor which is another skill PER can't evaluate). As a result, volume scorers are actually inflated by PER relative to distributors, but again, it's simply due to lack of precision. There's one clear way to solve it, and WP isn't doing it either.

rrravenred wrote:EDIT: Berri's own views of its weakness are:

Dave Berri wrote:Dr. Berri's own suggestions for possible weaknesses
1. Wins Produced does not give credit to players for creating shots. This is generally perceived as the biggest weakness. This perception is based on two beliefs. a) it is difficult to get a shot off in the NBA and b) the more you shoot the lower your efficiency. I am not sure there is much empirical evidence behind either proposition. Plus, if you give credit for taking shots then inefficient shooters will look better. Nevertheless if you believe shooting is difficult, Wins Produced will disappoint. And connected to this point…if you believe scoring is the most difficult thing a player does in the NBA, and therefore scorers are the most valuable, Wins Produced will disappoint also.

2. How vs. Why: Wins Produced tells you how productive a player is, but it does not tell you why. Performance is impacted by age, injury, roster turnover, coaching (in a few cases), and the productivity of teammates. The latter issue is the idea of diminishing returns. The more productive your teammates, the less productive you will be. This is a real effect in the NBA, although in the aggregate it is rather small. Still, these effects are not part of Wins Produced and these issues can impact what we see in the future from a particular player.

And that is the reason I think Berri is the one with the least amount of basketball knowledge. The guy just has no clue. It is not about creating a shot, every shot attempt even a bad shot gives a team a better chance for a basket (and winning) than a turnover. What is the point? Not shooting means a shotclock violation, the result will be a turnover for a team. Even bad shots have a higher probability to went in than every turnover. Not accounting for that is either complete ignorant or just blantant stupid. Berri is both in my eyes.

On another note: The variable in the boxscore are not competetly independent, which makes a regression analysis a tough job. That the amount on rebounds depends on the amount of missed shots is a very important thing. Taking into account what I wrote before, it becomes rather clear why his marginal values overrate rebounding.
It also happens that the boxscore is biased towards offense. Any boxscore tells you more about individual offense than defense. He runs into the same problem as Hollinger. Well, Hollinger's PER shows a better correlation towards winning than WP, if you take out the postional adjustment and the team factor for WP. What is the point? Berri is cheating to get a better correllation to winning. That might work on a team basis, but if you go down on the player basis your results become flawed.
There is no way to adjust correctly for positions in basketball, because nearly every player (except of some point guards and centers) don't play an explicit position anymore.

@Doctor MJ

It is not that easy to give credits for assists correctly. I choosed a way to do it on a player-by-player basis on each team. That means every assists gets 1/2 of the value a team usually scores with a field goal. On the other side I subtract those points from the scorer via team ast%.

To show the effect:
PER has Amare Stoudemire with 22.6 ranked 12th right now and Steve Nash with 21.7 is 16th. My PRA (just simple player rating, nothing fancy at all) has Steve Nash at 7th with 16.7 (league average is 10) and Stoudemire with 14.8 at 17th. Comparing those things with the +/- numbers my boxscore rating seems to make a better job. Of course the rating has the same flaws, underrate individual defenders who are not producing blocked shots, steals or rebounds.

Btw: Good example with Marcus Camby. When a boxscore rating is so way off in comparison to +/- numbers, it should definitely be reviewed.

http://www.wagesofwins.com/WOWCh8.htm

Enjoy.

mysticbb wrote:@Doctor MJ

It is not that easy to give credits for assists correctly. I choosed a way to do it on a player-by-player basis on each team. That means every assists gets 1/2 of the value a team usually scores with a field goal. On the other side I subtract those points from the scorer via team ast%.

To show the effect:
PER has Amare Stoudemire with 22.6 ranked 12th right now and Steve Nash with 21.7 is 16th. My PRA (just simple player rating, nothing fancy at all) has Steve Nash at 7th with 16.7 (league average is 10) and Stoudemire with 14.8 at 17th. Comparing those things with the +/- numbers my boxscore rating seems to make a better job. Of course the rating has the same flaws, underrate individual defenders who are not producing blocked shots, steals or rebounds.

Btw: Good example with Marcus Camby. When a boxscore rating is so way off in comparison to +/- numbers, it should definitely be reviewed.

It's definitely true that it's hard to value assists, and it's even harder to value the things a floor general does that aren't assists. I'm talking about PER undervaluing a distributor's scoring. I think we can all agree that a player who scores without a teammate getting an assist on average should get more credit that the guy who scores with a teammate assist - even PER agrees with that, but it can't do anything about it with the box score statistics available.

In terms of how to value assists, I don't really advocate anything. It's pretty clear to me that different offenses generate assists in different ways and frequencies, and they don't correlate that well with actual team effectiveness - so the answer is not to inflate assist weighting until you get something that looks good.

Doctor MJ wrote:Those really looking to understand the cons of Wins Produced, should check out this conversation:

http://sonicscentral.com/apbrmetrics/vi ... sc&start=0


To give the gist:

1) The heavy hitters in the community don't understand why WP is calculated like it is, and when they ask the questions they used to asking in the statistics field, Berri responds with a brush off. This isn't something where Berri's a genius and the other guys are morons. The guys he's talking with are statisticians with as much education as he has, and considerably more credibility in basketball, and they feel like they're asking basic questions and not getting reasonable answers.

I will admit, I only got a few pages in before my lack of formal mathematics training did for me.

However, a lot of it appears driven by Dan Rosenbaum, to which Berri offered this somewhat condescending response.

It's not pretty, but no academic fight ever is. What's the old saying about in academia, the fights are so vicious because the stakes are so small?

One note is that Rosenbaum (and Winston, whom his work is substantially based on) sold their methodology to teams (Cavs and Mavs, respectively) so have a vested interest in devaluing competing systems. Berri, OTOH, seems quite interested in selling his books, so he's no cleanskin in this regard.

Doctor MJ wrote:2) Berri indicates that WP was designed to correlate with actual wins, but others point on that you can do that with basically any stat. They are alarmed that that when you use WP to actually predict future wins, it actually does a worse job than if you go simply take a basic stat like points scored, which itself is much worse (as you'd expect) than pretty much any of the major formulas.

Well... the arguments I've seen indicate that they're disturbed by team adjustments, and note that even jersey numbers have a correlation to wins when a strong enough adjustment is used. It's a fair point, and is inherently a fudge. But box-score stats especially are inherently weak explanations of what actually goes on on a basketball court (with the "official" exceptions of points scored and fouls incurred), and you DO need to factor in context, although doing so is inherently contentious.

Doctor MJ wrote:Let's take a look at the case of Marcus Camby.

Here's how WP shows the Nuggets in Camby's last 3 years:

'06 - Camby leads the team in WP despite playing only 1800 minutes, whereas other players play more than 2900.

'07 - Camby leads the team 17.6 WP. 2nd best on the team is 5.6.

'08 - Camby leads the team 21 WP, 2nd best is 9.5.

So he's way, way, way better than everyone else right? What happens when he leaves the next year? Not only does the team get better, but it starts outrebounding opponents which it was not doing before.

And the additions are: Chris Andersen (post-suspension?) 6.2rpg, .242 WP48 (7.3 wins), Nene (post testicular cancer?), 7.8 rpg, .146 WP48 (7.6 wins), Chauncey Billups .171 (9.7 wins).

JR Smith that year also increased from .097 (1.9 wins) in 2007 to .165 (7.7 wins). Now there's a reasonable amount of improvement in that from the previous year independent of Marcus Camby, I would think.

Would those additions conform with the increased success and increased rebounding shown after Camby left?

Doctor MJ wrote:I understand thinking that maybe Berri's on to something with his emphasis on rebounding, but I honestly don't know how anyone can look at that and think Berri is valuing individual rebounding accurately. In the stat community there's something called "the laugh test" - if you get a result that's so obviously wrong any objective observer will laugh, then you need take gain some humility and either correct the algorithm, or explain the limitations of the algorithm. This result is frankly as embarassing as anything I've ever seen from a major stats name, and Berri is the least willing of any of the major stat guys (except Winston if you include him) to embrace humility. Yes, he shows some as in what you quoted, but on the whole he's very cocky, and even in his humility, he doesn't address the obvious problem explicitly.

Well, he's made attempts. Here he claims that even if you significantly devalue rebounds in Winscore (WP's little brother) the metric doesn't change markedly.

I somewhat disagree with your characterisation of the laugh test. What people laugh at is socially constructed, and the decisions we make are strongly influenced by non-rational factors ((i.e. the behvioural economics school that Berri adheres to). Are the results wrong or counter-intuitive? If they're wrong, point at some evidence that says they're wrong. Otherwise, stop laughing.

If someone were to say that an MVP 30ppg scorer was not an especially valuable player in a their MVP year (yes, we both know the player), then would people initially laugh? I don't think people laugh at that statement now (although many people disagree, passionately).

Am I saying that Berri should sail ahead, ignoring all critique? Certainly not! If he's been arrogant in asserting the predictive power of his model then he should be knocked back on his ass quick-smart.

Doctor MJ wrote:Re: PER favoring chuckers, that idea needs to be tempered. Here's the issue that Berri's refering to: PER (and similar stats), penalize a guy blowing a possession. A missed shot, doesn't always blow a possession, so PER penalizes it based on an estimate for how often a missed shot blows a possession. Berri points out correctly that a player's PER will keep increasing with volume even at pretty low FG% which seems like a weird way to do a stat because you could theoretically inflate your PER while doing stuff that hurts your team. Practically though this isn't much of an issue. Fans of players who score at high volume but not great efficiency are constantly trashing PER because such players don't come off as well in the stat as the fans think. The typical basketball viewer thinks more highly of Kobe than PER does for example, and up until recently, thought WAY higher of Iverson than PER did.

I'd note that WP thinks Kobe is an excellent player despite the offensive inefficiency. And any fan will bash a metric that doesn't confirm their own biases about players or teams.

Doctor MJ wrote:Getting back to the estimate for how often a missed shot leads to a blown possession, I want to make sure it's clear what that means, because other than the problems with defense, the use of estimates are what I consider to be the biggest weakness in PER. If the box score actually listed what shots led blown possession, we'd have a really great idea how much a player's chucking was hurting his team. Undoubtedly, if a player started chucking worse, and defenses saw it coming, the offense would be less able to get offensive rebounds, and it would all get factored in to PER. PER doesn't use such fine grained information, it estimates. This is a weakness, but it's not something caused by weak logic. It's not something that in practice rewards chuckers, it's something that simply adds margin of error to algorithm.

I understand that's one of the major criticisms of PER, that its values are all estimated by Hollinger. Now individually none of those estimations fail the laugh test, but if the cumulative effect raises a giggle in terms of compounding the MoE, then there's a problem.

Doctor MJ wrote:One last note: It's funny that Berri says it may be a flaw of his formula that it doesn't reward shot creation - because this is actually a problem with PER. PER seeks to give some of the credit for scoring to the assister - but it doesn't know whether any particular bucket was assisted, so it docks all scoring equally. This really hurts the players that are basically always creating for themselves and others (distributors - who also are the guys who are most likely to be acting as coach on the floor which is another skill PER can't evaluate). As a result, volume scorers are actually inflated by PER relative to distributors, but again, it's simply due to lack of precision. There's one clear way to solve it, and WP isn't doing it either.

Assists suck as a measure, although they're not horrible as an indicator. I'd ditch them in any stat assessing team offensive efficiency.

Doctor MJ wrote:In terms of how to value assists, I don't really advocate anything. It's pretty clear to me that different offenses generate assists in different ways and frequencies, and they don't correlate that well with actual team effectiveness - so the answer is not to inflate assist weighting until you get something that looks good.

True, that is the reason why applying a static value to assists is not working at all. It is also the case that assists seems to be a not given out in the same fashion by each scorekeeper in the league. For me assists are a somewhat subjective stat.

@rrravenred

I also read some of the stuff on APBRmetrics.com and I'm in nearly complete agreement with Dan Rosenbaum. It was the first thing I noticed when I first read Berri's paper back in 2007 that Berri is using a team adjustment. I trusted Berri that this adjustment is a minor one, not really questioned it and I used Wins Produced as the main boxscore stat in addition to my own a lot until around mid 2008. In summer 2008 I wanted to improve my own rating a bit more, and thought about some additions. While I was at it, I recreated Dave Berri's Wins Produced and realised that without the team adjustment and the positional adjustment the rating is actually working worse than NBA Efficiency. Ever since I followed the discussions about that and Berri never admitted that his adjustments weren't in fact not minor important, but very important especially for his claim that 95% of the wins would be explained by his metric.

Using a team adjustment is rather simple. Knowing that the boxscore is biased towards offense means a team with better offense, but also bad defense will most likely score better ratings than a team with worse offense, but better defense. That is in fact the case for nearly any boxscore based rating. Even the pythagorean expectation is doing that. In this season for example the Bulls have a better than average defense, but a much worse than average offense. The Pyth. Exp. would give them 35 wins, but they have actually 39 wins. On the other side a team like the Warriors is better on offense than average, but much worse than average on defense. They are expected to have 30 wins by now, but in reality they only have 25. Knowing these things you can easily improve your rating by using a "team factor" just like Berri is doing it. I can actually apply such an factor to my rating and get a nearly 100% correlation towards winning. But using such a tool for evaluating players on a boxscore basis will give you nothing at all.

At the end of the day I don't check Berri's blog anymore, I also don't use Wins Produced anymore, because I don't get any real valuable informations out of it. If I want to have a metric which will give credit to players for wins based on boxscore I use Win Shares and my own rating to judge players. I can get a different view by using +/- numbers (adjusted or Net or unadjusted). There is really no use for me for Wins Produced anymore, especially since b-r.com provides WS/48 values on their website (I don't need to calculate that by myself anymore).

mysticbb wrote:.. assists seems to be a not given out in the same fashion by each scorekeeper in the league. For me assists are a somewhat subjective stat.
.

You can eliminate much of the subjectivity of assists by noticing the great differences in various teams' home/away Ast/FG ratios. This is readily downloadable here:
http://www.dougstats.com/09-10.html

On the theory that 29 scorekeepers can't be highly prejudiced for/against any given team's playmakers, a team's away Ast/FG ratio is almost certainly more indicative of their real Ast/FG rate.
Furthermore the ratio of Away/Home Ast/FG tell you what % of their Home assists are real.

From these ratios, and knowing # of games at home and away, you can guess what % of a player's 'official' assists are real by league-average standards. This year, you could apply these ratios to players' assist rates:


The Hawks Ast/FG is .491 on the road, and .625 at home. That's a 27% markup by Atlanta scorekeepers.
At the other end, the Kings' scorekeepers are relatively stingy at home. Their players have gotten 8.5% more Assists at other venues.

rrravenred wrote:And the additions are: Chris Andersen (post-suspension?) 6.2rpg, .242 WP48 (7.3 wins), Nene (post testicular cancer?), 7.8 rpg, .146 WP48 (7.6 wins), Chauncey Billups .171 (9.7 wins).

JR Smith that year also increased from .097 (1.9 wins) in 2007 to .165 (7.7 wins). Now there's a reasonable amount of improvement in that from the previous year independent of Marcus Camby, I would think.

Would those additions conform with the increased success and increased rebounding shown after Camby left?

But surely you see how bizarre the story has to become to stay coherent? Camby is over twice as good of a player as anyone else on the team, he leaves, the team becomes better in the areas of Camby's specialties (defense and rebounding), and it's explained because the Nuggets just happen to get a bunch of great players back? I recognize that Nene and Anderson did come back - that's relevant and a good point - but look at Anderson relative to Billups. Anderson's playing half the minutes of Billups and he produces almost as many wins? I mean literally, WP per minute says that Chris Anderson is the best player on the team quite easily and Karl thinks the team is better off with Anderson on the bench most of the time - I guess he's a moron.

I guess in the end to me this is about recognizing value over replacement. If you can replace Camby with a journeyman and achieve similar results, then he's not actually giving value over replacement - and hence not that valuable.

rrravenred wrote:
Well, he's made attempts. Here he claims that even if you significantly devalue rebounds in Winscore (WP's little brother) the metric doesn't change markedly.

I somewhat disagree with your characterisation of the laugh test. What people laugh at is socially constructed, and the decisions we make are strongly influenced by non-rational factors ((i.e. the behvioural economics school that Berri adheres to). Are the results wrong or counter-intuitive? If they're wrong, point at some evidence that says they're wrong. Otherwise, stop laughing.

If someone were to say that an MVP 30ppg scorer was not an especially valuable player in a their MVP year (yes, we both know the player), then would people initially laugh? I don't think people laugh at that statement now (although many people disagree, passionately).

Am I saying that Berri should sail ahead, ignoring all critique? Certainly not! If he's been arrogant in asserting the predictive power of his model then he should be knocked back on his ass quick-smart.

Yeah, that link is crazy man. He says this:

"So what’s the difference? One can note that Win Score has different values for assists and blocked shots. One can also note that we incorporate personal fouls. But these are not the truly important difference. The important difference is how we value shooting efficiency."

Okay, so is that how Camby ends up with a sky high WP? No, because Camby's efficiency is terrible. In '08 his TS% was under 50 - so it's pretty clear that shooting efficiency is NOT what's causing Camby to look like a megastar, and so it is NOT what the important difference is.

Point taken on the "laugh test" in principle, but hits just keep coming against WP. The combination of 1) his results being far outside the norm, 2) his arrogance being well above almost everyone in the field, 3) answers like this one where rather than deal with the issue at hand he brings up something else that doesn't resolve the issue just kill the dude's credibility.

rrravenred wrote:I'd note that WP thinks Kobe is an excellent player despite the offensive inefficiency. And any fan will bash a metric that doesn't confirm their own biases about players or teams.

The point is that PER in and of itself values volume scoring less than typical basketball viewers do. Now, you can always say "yes, but it STILL overrates them, just not as bad as basketball viewers", but that's not what he's saying. He's choosing to pick a battle against PER instead of viewers, and one would think he's doing that on purpose.

rrravenred wrote:I understand that's one of the major criticisms of PER, that its values are all estimated by Hollinger. Now individually none of those estimations fail the laugh test, but if the cumulative effect raises a giggle in terms of compounding the MoE, then there's a problem.

It's an estimate man. No it's not perfect, of course not. Have less than perfect precision on something, magnify an unrealistic scenario, you'll always be able to find something that doesn't look good. The real point is that the only way to make it clearly better is with more data - there's not some magic algorithm that can make it work without that.

What one guy (now working for the Cavs, I think) said about WP:

... because Berri sets the threshold efficiency for scoring at league average, the sum of Wins Produced by scoring in the league is zero. Thats right, Berri's metric says that throughout the entire league scoring did not contribute to any wins, that all 1230 wins in the league were created acquiring possession of the ball.

http://sonicscentral.com/apbrmetrics/vi ... hp?p=14738

So, unless players are shooting better than the league average -- fewer than half of all players do this -- their scoring effort actually counts negatively in their WP totals.

Rebounds are all good, however. There's nothing subtracted for the boards you didn't secure. You might think everything below 1/10 of available rebounds -- average by definition -- would be subtracted. But no. Because everything must revolve around the mantra, Points are Overvalued, rebounding by default rises to the top of the heap.

In this scenario, guards have almost no value. They only get boosted by their 'positional adjustment'. If you are a G-F who doesn't rebound much, you better hope he calls you a G.

Bret at Hoopinion destroys Berri's book and logic:

http://www.hoopinionblog.com/2010/04/bo ... html#links

This thread makes me feel good. I had generally not looked at wins produced, because I've had concerns about its credibility, and it appears I was right.

Doctor MJ wrote:The way Wins Produced gives such huge weight to rebounds, I'm inclined to side with Win Shares.

Berri ran regression and results showed how valuable rebounds are. No place to discussion here unless someone will provide other result done on bigger sample of data.

Doctor MJ wrote:Let's take a look at the case of Marcus Camby.

Here's how WP shows the Nuggets in Camby's last 3 years:

'06 - Camby leads the team in WP despite playing only 1800 minutes, whereas other players play more than 2900.

'07 - Camby leads the team 17.6 WP. 2nd best on the team is 5.6.

'08 - Camby leads the team 21 WP, 2nd best is 9.5.

So he's way, way, way better than everyone else right? What happens when he leaves the next year? Not only does the team get better, but it starts outrebounding opponents which it was not doing before.

Lets look
2008
Camby 21.0
AI 9.5
Melo 9.1
Carter 5.3
JR 2.9
Kenyon 2.3
Kleiza 1.7
Najera 1.3
Nene 0.2

2009
Billups 9.7
JR 7.7
Nene 7.6
Andersen 7.3
Melo 5.9
Balkman 4.9
Carter 4.4
Kenyon 2.9
Jones 1.7
Kleiza 0.7

So healthy Nene + Andersen and Balkman (together 19.8 WP) contributed in place of Camby. WP perfectly explains what happened in Denver between 2008 and 2009.

mysticbb wrote: I can actually apply such an factor to my rating and get a nearly 100% correlation towards winning.

I saw statements like this before, but nobody do this. So feel free, do adjustments the way Berri done and show us and him (Berri) the results.

WP doesn't explain anything then, if the #2 player in WP can get so easily replaced.