Wins Produced?!

[mystic]

Sensing the danger, Berri himself has jumped in to try to derail the discussion.

Wow, Berri's comment shows how less he is interested in actually discussing the issues. WP does not predict well on any kind of out of sample tests, no matter how large or small. Saying in hindsight how many wins a team should have is no prediction.

Also, Arturo uses the year-to-year correlation as evidence that a metrics works, while not knowing that a simple metric which uses points per 100 poss scored by a player adjusted via a team defensive adjustment shows a very similar year-to-year correlation. That tests is basically useless here, because we are not talking about an unbiased sample. The players are preselected by coaches and will most likely be used in a similar fashion by any coach in the league. A scorer in one season will most likely be again a scorer in the next season. We have a selection bias which is forcing the year-to-year correlation, that is not a result of the metric itself. All 3 things, the WOW crew is using to justify the metric (regression, correlation to wins and year-to-year correlation) are neither unique to WP nor are they sufficient to say anything about the ability of the metric to evaluate individual players.

[DSMok1]

An exchange from: http://wagesofwins.com/2012/09/24/the-p ... a-analysis

“Unfortunately not all analysis is equal (a point made in the post.) Can the stats attributed to players explain wins? Yes! Now if the question is if credit is properly distributed, that’s fine. That does not take away from the explanatory power though.”

There are many ways to regress box score stats in such a way that they match up to point differential and thus “explain” wins. Explanatory power as such is basically a necessary but not sufficient criteria of a good metric.

This statement….

There are many ways to regress box score stats in such a way that they match up to point differential and thus “explain” wins. Explanatory power as such is basically a necessary but not sufficient criteria of a good metric.

is simply not true. At least, not unless you just add the error term back in to your model (a common method in some circles). Then, of course, there are an infinite number of ways to run the regression.

How many can you come up with, following Dave Berri's rules of not adding back in a "team adjustment"? You could still add in team things like "Opponents Field Goals Made".

[DSMok1]

Here was my first crack:

Here is one basic box-score regression that has an adjusted R^2 of 0.935 on wins at the team level:

Player Points: 0.03305
Opponents Points Scored: -0.033116
Intercept: 41.4971

(For a player-level stat, you would use 0.03305*Points Scored, -0.033116*player minutes/team total minutes, and 41.4971*player minutes/team total minutes)

That would, at the team level, have an R^2 of 0.935 on wins, and would have good YtY correlation as well, fulfilling both criteria used to evaluate Wins Produced.

[Guy]

Daniel: I was going to bravely predict that your excessively cogent comments were soon going to result in your being banned there, but having now clicked over to WOW I see that I am too late. But in the spirit of Wins Produced, I will nevertheless "predict" your banning, and I feel confident my R^2 will exceed even Berri's famous .94.

You are to be commended for your civility and (excessive) patience. It's too bad that it wasn't reciprocated. Truly, Berri doesn't seem to be trying even to maintain a facade of intellectual integrity anymore. Faced with criticisms, he just kicks up a lot of dust to confuse the issue, signals Dre and his other lieutenants it is time to attack the infidel (you, in this case), and hopes that most of his readers will conclude "I guess the guy with the PhD and articles in refereed journals must be right." Surely he knows better, but at this point he's "all in" on defending WP to the death. It's sad to watch.....

[Crow]

a small follow-up to this comment from Mark Cuban at wages of wins:

Mark Cuban on September 4, 2012 at 1:02 pm said:

"Dre makes my point. How many wide open threes do you think dsteve, jet or jkidd get without diek on the floor.

Jet was the only one who played any meaningfully minutes w out dirk. To say dirk was our fifth best player doesn’t pass the most basic of smell tests.
Who should have taken dirks minutes? "

nba.com shows that Terry actually took more 3 attempts without Dirk but he hit far less of them.

http://www.nba.com/advancedstats/player ... 2;season=r

So Mark's argument is partially supported. But maybe Terry shouldn't take so many 3s without the apparent advantage that comes with playing with Dirk. Does any Boston player offer the same advantage for him? Probably not.

Terry on the bench, Dirk had better individual stats but the team did worse. Will that stay the case? The new guys will have their direct and indirect impact too.

[mystic]

So Mark's argument is partially supported.

Sorry, but his argument is fully supported. Without Nowitzki on the court the Mavericks became worse, substantially worse, and that for now 12 years in the row!

But maybe Terry shouldn't take so many 3s without the apparent advantage that comes with playing with Dirk.

That is a naive idea and basically flies on the same level as the stuff WOW is saying. It is far more likely that Terry took the best available shots, just that those shots were less easy to convert than when he had more space with Nowitzki on the court.

Does any Boston player offer the same advantage for him? Probably not.

High pick&roll with Garnett or Brandon Bass? Maybe even Jeff Green? Why shouldn't that work?

Terry on the bench, Dirk had better individual stats but the team did worse. Will that stay the case? The new guys will have their direct and indirect impact too.

Terry was replaced with a worse player, for sure the team gets worse overall. The question is can Mayo fill in? Maybe not in the same fashion offensively, but he can be an upgrade defensively over Terry which might compensate the offensive drop. Also, the Mavericks major improvement seems to be rather the frontcourt with Brand and Kaman instead of Haywood and nobody formerly known as Lamar Odom. In order to compensate for Kidd's ability to make an entry pass to Nowitzki, we have to see how that works with either Mayo or Carter, because the smaller guys like Collison and Beaubois might have similar problems with that as Barea, Harris or Terry had.


@DSMok1

Run the points scored per 100 possession with a defensive adjustment (simple DRtg minus league average DRtg). I got R^2=0.941

Code: Select all

    Model Summary(b)
                         Change Statistics
    Model   R   R Square   Adjusted R Square   Std. Error of the Estimate
    1     ,970a      ,941      ,941           ,0377361   
    a. Predictors: (Constant), DEF_A, PTS_P
    b. Dependent Variable: Win%


  Win% = -2.805+0.155*PTS_P+0.031*DEF_A

Now I can make up any kind of formula to determine PTS_P which sums up to the ORtg of the team (given the boxscore, we can basically get that to 100%) while the whole high correlation to win% is controlled by DEF_A anyway.

Analysis of the effect of the DEF_A on the overall player ranking:

Code: Select all

Correlations
                         Rank_Pts     Rank_adj
Rank_Pts  Pearson Correl          1       ,994**
          Sig. (2-tailed)                    .000
          Sum of Squares8271032.000   8217375.000
          Covariance      17902.667     17786.526
          N                     463           463
Rank_adj  Pearson Correl    ,994**              1
          Sig. (2-tailed       .000
          Sum of Squares8217375.000   8271032.000
          Covariance      17786.526     17902.667
          N                     463           463

As we can see, there is a 0.994 correlation between the ranking of the players by points scored per 100 possessions and the same with the DEF_A. So, the DEF_A is a small thing which essentially doesn't change the player ranking.


Then the proof that the DEF_A controlls the high correlation despite having a minor influence on the player ranking:

Code: Select all

Correlations
                               PTS_P           Win%
PTS_P       Pearson Correl               1       ,680**
            Sig. (2-tailed)                         .000
            Sum of Squares         469.918        66.054
            Covariance                .564          .079
            N                          834           834
Win%        Pearson Correl         ,680**              1
            Sig. (2-tailed            .000
            Sum of Squares          66.054        20.096
            Covariance                .079          .024
            N                          834           834

Only 0.68 instead of 0.97 correlation.

Last but not least, the year-to-year correlation for the Top 292 minute getters in 2010 and 2011:

Code: Select all

Correlations
                               FM48_2010         FM48_2011
FM48_2010    Pearson Corre                   1        ,852**
             Sig. (2-tailed)                             .000
             N                             292            292
FM48_2011    Pearson Corre             ,852**               1
             Sig. (2-taile                .000
             N                             292            292

Well, typical high correlation is seen.

Oh well, Berri's criteria are completely met:

1.) It's based on a good theory, Scoring per 100 possessions + Defense = Wins.
2.) It correlates well with Wins (0.97 correlation coefficient)
3.) It is predictable (a players FM48 from a previous year explains 80+% of future years.

[Crow]

I guess I missed the word "open" 3s. Thanks for that correction. My response was based & focused on the total quantity of 3s increasing. My questioning the wisdom of that still stands a topic that needs further research to answer.

"It is far more likely that Terry took the best available shots, just that those shots were less easy to convert than when he had more space with Nowitzki on the court."

Maybe, but you are making an assumption here.



"High pick&roll with Garnett or Brandon Bass? Maybe even Jeff Green?"

How good are they and their team at this? I don't have Synergy. I'd be somewhat surprised if any of these three are notably above league average on these plays.

[mystic]

Maybe, but you are making an assumption here.

Yeah, the assumption that the coaches and players on the Mavericks are actually not idiots and really trying to win games. Don't you think that at some point someone would have said something and corrected things, especially on a team like the Mavericks?

"High pick&roll with Garnett or Brandon Bass? Maybe even Jeff Green?"

How good are they and their team at this? I don't have Synergy. I'd be somewhat surprised if any of these three are notably above average on these plays.

Synergy had Garnett last season with 1.04 ppp (points per play) as 49th in the league. Bass was at 0.95 and 72nd in the league. Both are noticable above average pick&roll players. And sorry to be blunt, every basketball fan who is a bit more interested in the game and especially analysing players should have known the strength of rotation players like Bass or Garnett. Both are automatic from midrange when they are open, thus the defender has to stay with them. Rondo isn't able to convert up on this, but Terry will take advantage of those situations.

[Crow]

Maybe, but you are making an assumption here.

Yeah, the assumption that the coaches and players on the Mavericks are actually not idiots and really trying to win games. Don't you think that at some point someone would have said something and corrected things, especially on a team like the Mavericks?

"High pick&roll with Garnett or Brandon Bass? Maybe even Jeff Green?"

How good are they and their team at this? I don't have Synergy. I'd be somewhat surprised if any of these three are notably above average on these plays.

Synergy had Garnett last season with 1.04 ppp (points per play) as 49th in the league. Bass was at 0.95 and 72nd in the league. Both are noticeable above average pick&roll players. And sorry to be blunt, every basketball fan who is a bit more interested in the game and especially analyzing players should have known the strength of rotation players like Bass or Garnett. Both are automatic from midrange when they are open, thus the defender has to stay with them. Rondo isn't able to convert up on this, but Terry will take advantage of those situations.

"Both are automatic from midrange when they are open"

Sorry to be blunt but 1.04 and 0.95 makes those shots (overall, does Synergy give you open shots to make your claim about them?) below league for all shots and not really that impressive to me and certainly not "automatic". But I guess most other players really suck on those plays. I didn't immediately know for sure that they were better and don't trust my eyes without question to "automatically" know that. I prefer to know the figures, so I asked. Thanks for sharing those figures.

What was league average anyways? You didn't say. How far above league average was Bass? If the right people take more of these shots, the mean is probably higher than the median players efficiency. And you left Green out. Did he not support your argument against my cautious, need the facts perspective? If Bass is not notably above the mean that would make 2 of 3 not notably above as I stated, assuming Green is not, since you didn't mention the facts with regard to him. So looking forward to receiving that information to resolve whether it was so obvious.

[Crow]

Maybe, but you are making an assumption here.

Yeah, the assumption that the coaches and players on the Mavericks are actually not idiots and really trying to win games. Don't you think that at some point someone would have said something and corrected things, especially on a team like the Mavericks?

Maybe or even probably so, but it is still an unsupported assumption. And overall they optimized their team eFG% to ... 14th best in the league last season. So you trust them to have done right on this detail, when the facts say they were not notably efficient at shot selection overall. As you wish. Something was pulling them down. Can't really say whether Terry shooting 32% from 3 without Dirk was helping much without more facts. Did Synergy give you data on 3pt efficiency of players purely off of pick n action of other players or even specific players or are you assuming even more? I assume you would have to watch all those plays and count them using this criteria yourself to know, using the public version, but correct me if I am wrong or if you have done that.



I would have assumed that the Celtics would recognize that they were extremely high on frequency of low-efficiency mid-range shots and correct things... but they didn't, all season.

Terry may or may not help, depending on how many shots he takes compared to the mid-range happy shooters left and whether he gets many open 3 pt looks or not off of high pick n roll or not, given that the ball is in Rondo's hands a ton. Will they play Terry at back-up PG? That is another thing I wonder about. Will Terry even play much more than half his minutes with KG, Bass and Green? It might be close to that level.

I understand the logic of getting him though, if you have alienated Allen and vice versa and / or you decide you can't continue to do (with their high pick n roll and everything else) what produced the 27th best offensive efficiency... any longer.

[mystic]

"Both are automatic from midrange when they are open"

Please Crow, that conversation makes not much sense, when you constantly missing the point. Open, is the key word here. The Synergy data contains ALL situations, even those when they don't even shoot due to a turnover.

In average the league shots about 38% from 10 to 23 ft, Bass and Garnett are at about 48%.

below league for all shots and not really that impressive to me and certainly not "automatic".

The point is not whether you are impressed by that or not, the point is that both players can play the high pick&roll with Terry like Nowitzki done it. Obviously, neither has the range of Nowitzki, which will probably lead to a lower efficiency increase than with Nowitzki for Terry, but nonetheless it should look better for Terry than as he looked without Nowitzki on the court. They can run the same staggered picks like the Mavericks, they can the pick&pop, they can do that too. This is the point.

But I guess most other players really suck on those plays.

There is an VERY important detail you are missing, shot selection is not just determined by the players own will, but more importantly by the shot the defense is willing to give up. And usually the defense is trying to give the opponent a low percentage shot. Some players don't have the talent in order to get better shots, but the shot clock is forcing the players to take a shot, because no matter how low the probability is to convert that shot, it is still more likely to score than by accepting a shot clock violation.

And you left Green out. Did he not support your argument against my cautious, need the facts perspective?

Green didn't play last season, there is no data available for him. It would be good, if you educate yourself first before making baseless assumptions. And Jeff Green had about 48% from 10 to 23 ft on the Celtics as well.

The point for Terry isn't even how good those roll man are at converting, but that Terry is an improvement over Rondo in terms of converting as pick&roll ball handler. Rondo scored on 42 eFG% in those situations while Terry on 49 eFG%.

Maybe or even probably so, but it is still an unsupported assumption. And overall they optimized their team eFG% to ... 14th best in the league last season. So you trust them to have done right on this detail, when the facts say they were not notably efficient at shot selection overall.

It makes not much sense to engage in a conversation when someone like you are completely missing the point or just blantantly makes up strawman.

It is not a question of the results here, it is a question of the opportunities. My point is that it is likely that they just didn't have the opportunities to take better shots rather than thinking they were just morons who couldn't see the obvious. Heck, the Mavericks had Roland Beech on the bench, drawing up charts on the fly, you know, the guy who runs 82games.com, and you really think that they were just unable to see the thing which seems so obvious to you?
Yeah, it is easy to say that they should take different shots, but it not that easy to get that better shot against an opponent who wants you to not get that better shot. A part of the equation you seem to ignore.

I would have assumed that the Celtics would recognize that they were extremely high on frequency of low-efficiency mid-range shots and correct things... but they didn't, all season.

If the Celtics could have gotten easier shots, they would have taken them. That is the reality. The shot selection is determined by the own talent and the opponents defense, not by pure will. Rondo can't create better opportunities, thus the Celtics accomodated that by using their better midrange shooters.

I understand the logic of getting him though, if you have alienated Allen and vice versa and / or you decide you can't continue to do (with their high pick n roll and everything else) what produced the 27th best offensive efficiency... any longer.

The Celtics were above average for isolations or for the rollman in the p&r. Do you want to know where they were way below average? The pick&roll ballhandler! Rondo running the pick&roll was a problem here. Terry will improve that aspect. And Doc Rivers already said that they want to use Terry as backup point guard, maybe even reducing Rondo's minutes. I wouldn't be suprised, if they let Terry get the minutes at PG at the end of close games, while Rondo is sitting on the bench.


Short summary:

1. Shot selection is not determined by will, but rather by talent and opponents defense.
2. The Celtics have a couple of potential pick&roll partners for Terry.
3. Terry is better as the pick&roll ball handler than Rondo.

[DSMok1]

Daniel: I was going to bravely predict that your excessively cogent comments were soon going to result in your being banned there, but having now clicked over to WOW I see that I am too late. But in the spirit of Wins Produced, I will nevertheless "predict" your banning, and I feel confident my R^2 will exceed even Berri's famous .94.

You are to be commended for your civility and (excessive) patience. It's too bad that it wasn't reciprocated. Truly, Berri doesn't seem to be trying even to maintain a facade of intellectual integrity anymore. Faced with criticisms, he just kicks up a lot of dust to confuse the issue, signals Dre and his other lieutenants it is time to attack the infidel (you, in this case), and hopes that most of his readers will conclude "I guess the guy with the PhD and articles in refereed journals must be right." Surely he knows better, but at this point he's "all in" on defending WP to the death. It's sad to watch.....

I wasn't banned over there, just warned. But it was unnecessary and unprofessional on my part to even start the debate. All of the issues are well known, so it was rather pointless.

[Guy]

I wasn't banned over there, just warned. But it was unnecessary and unprofessional on my part to even start the debate. All of the issues are well known, so it was rather pointless.

Dre effectively said any comments from you that criticized the WP model were banned. So while you personally were not banned, many of your ideas were. To me that is not materially different, since it's the ideas that are important.

You said over at WP that "there are better and certainly more productive ways to discuss these matters." What are some of those better ways? If the point is to engage proponents of WP in discussion, I don't think there is any other place you can do that. So I'm not sure what alternative you have in mind. Of course, you may now feel that no discussion can be productive, or simply choose not to engage in such a discussion for whatever reason. But I don't see anything "unprofessional" about your attempt to discuss the merits of WP with its proponents.

[Bobbofitos]

I wasn't banned over there, just warned. But it was unnecessary and unprofessional on my part to even start the debate. All of the issues are well known, so it was rather pointless.

Dre effectively said any comments from you that criticized the WP model were banned. So while you personally were not banned, many of your ideas were. To me that is not materially different, since it's the ideas that are important.

You said over at WP that "there are better and certainly more productive ways to discuss these matters." What are some of those better ways? If the point is to engage proponents of WP in discussion, I don't think there is any other place you can do that. So I'm not sure what alternative you have in mind. Of course, you may now feel that no discussion can be productive, or simply choose not to engage in such a discussion for whatever reason. But I don't see anything "unprofessional" about your attempt to discuss the merits of WP with its proponents.

I think DSmoki was just being polite. No need to really break it down any further; one party had class and dignity, the other did not. Nothing new to report on that front.

[DSMok1]

Class is often lacking in discussion on the internet; I don't need to add to the problem! Picking an argument where someone has to be defensive is never right; I put them on the spot--I could easily have just written a blog post, or emailed Dave privately.