A few days ago I posted a comment on Carmelo Anthony receiving a maximum contract. In that post I noted that Anthony is a leading scorer in the NBA, but not a leader in the production of wins. One criticism people have directed towards our valuation of NBA players is that it appears to under-value scorers. I think it might be a good idea to address this issue. Specifically, can a player be above average in terms of both scoring and wins production?
Actually we explicitly comment on the valuation of scorers in the book in our discussion of Ben Gordon’s rookie campaign. Although Gordon was a unanimous selection to the All-Rookie team and also won 6th man of the year, his rookie campaign was hardly a success in terms of Wins Production. At least, relative to other players at his position, he did not appear very productive.
As we note in the book, maybe scorers need to be compared to scorers. After all, someone needs to score if a team is going to win. With this in mind, we stated the following on p. 153:
Still, we agree that someone must score at some point if a team is going to succeed. In our analysis we have made it a point to compare a player relative to his position. Perhaps the same thinking applies to evaluating a scorer like Gordon. Comparing a scorer to a non-scorer might be unfair, since each plays a different role on the team. Perhaps we would be better off comparing Gordon to other scorers.
Of course, to make this comparison we need to define what we mean by the term “scorer.” In the book we defined a scorer as any player who played at least 2,000 minutes and averaged one point scored every two minutes played. In 2004-05 33 players fit this definition. This was the same number of players I find when I look at 2005-06.
Each of these 33 players were analyzed in terms of Wins Production and Wins Production per 48 minutes (WP48) during the 05-06 regular season.
Now I tried adding a table with this analysis as a picture, but it didn’t work too well. One day I will figure out how to get tables into these posts. For now, please go HERE for the analysis of scorers in 2005-06.
The results reveal that most scorers look pretty good in terms of our performance metrics. The average team wins 0.500 games per 48 minutes, so the average player will produce 0.100 wins per 48 minutes played. Of the 33 scorers examined, 26 or 79%, posted an above average WP48 in 05-06. Interestingly this was the same percentage we observed in 2004-05. Given these results, if you assumed that a scorer was an above average player, you would be right quite often.
But you won’t be right all the time. Seven of these players were not above average. These include the aforementioned Carmelo Anthony and Ben Gordon. Additionally, Mike Bibby, Chris Webber, Richard Hamilton, Al Harrington, and Zach Randolph offered below average levels of wins production in 05-06. So relative to their scoring peers, these players don’t appear to be very productive.
The results with respect to Al Harrington are especially significant. If we look over Harrington’s career we see that his performance in 2005-06 was in some sense unusual. Prior to this last season Harrington had never met our definition of a “scorer.” So his scoring totals in Atlanta were somewhat out of character. His lack of wins production, though, was not. Harrington has never been an above average producer of wins in the NBA. Given this history, it will be interesting to see what kind of contract he will sign this summer. Will people over-value Harrington’s one year of scoring? Or will his lack of wins production dampen enthusiasm for his services?
And more importantly, if Harrington doesn’t cash in, will he think this is my fault?
— DJ
Wages of Wins Journal 
KC
July 12, 2006
Oh s#!^, I knew the minute I heard that Mullins is guning to sign Harrington that something was wrong in the univers–or in the case of the Warriors, perhaps all is proceeding according to the same g*&&@m plan we’ve been suffering though since Cohan stole the team.
Has there ever been a wages of wins analysis of owners? I’d love to see Cohan exposed for the scourge that he is…
dberri
July 12, 2006
KC,
The latest rumour I saw floating around mentioned the possibility of a sign and trade where the Warriors would send Troy Murphy to the Hawks for Harrington. The Warriors had four above average performers last year: Murphy, Baron Davis, Jason Richardson, and Biedrins. Given the productivity levels we see for Murphy and Harrington, swapping these two players does not look like a trade that moves Golden State closer to a winning season. So you have some reason to be concerned.
Jason
July 12, 2006
Concern? It’s more or less a forgone conclusion that the Warriors will make a lousy move or two or three and it looks like this could be one of them.
dberri
July 12, 2006
Jason,
One rumour I saw said the Warriors could send Murphy or Pietrus. If it is Harrington for Pietrus, then the Warriors are not necessarily worse off. Not better off, but it doesn’t look like it hurts them too much. Of course, at another place I saw that to make the salaries numbers work it had to be Murphy and Pietrus for Harrington. And that really doesn’t help the Warriors.
It does move the Warriors closer to being the New York Knicks of the Western Conference. With a team of scorers that doesn’t win they can spend an entire season arguing about team chemistry and how a new coach would solve all their problems. So that might be fun.
KC
July 12, 2006
The NYKnicks? We of the long suffering Warrior tribe would love to be the Knicks. The Cohan-owned W’s have no parallel in the NBA. No team comes close to their tradition of utter nothingness.
Is there some site that publishes these stats? I would love to see the season ending stats by player by team.
dberri
July 12, 2006
KC,
Be careful what you wish for. The Knick fans can’t be a happy bunch right now.
As for the stats… as some point I will try and get more of this info on-line. Not sure how long that will take me, though.
Tom Mandel
July 13, 2006
Lets assume for a moment that you were a GM making use of your results as an economist – and that you were running the Washington Wizards.
What deals would you try to make to improve the Wizards?
Tom Mandel
July 14, 2006
Well, I don’t seem to be eliciting a response — which may be perfectly reasonable.
Except that 1) unless these tools of analysis can be applied to actually improve a team, one might wonder what their value is (“wonder” I said, not doubt that they have any value), and 2) I really need an answer; Grunfield is being very quiet this off-season in an Eastern Conference roiling with change, and I’m getting kind of nervous….
Any thoughts – from you, DJ, or from other readers of the WoWJournal?
dberri
July 14, 2006
Tom,
I’m thinking……
There is a problem with the Wizards. You have a collection of players who are all just a bit above average. So it is not real clear where you can improve easily. For most teams there is one place on the roster where you can say “hey, if that position or those group of players were much better, then your team would really improve.” The Wizards don’t have that one glaring weakness.
Of course, that previous response is the product of me thinking about this for about 30 seconds. So perhaps I just need to think more.
p-dawg
July 14, 2006
I really wish you’d hop over to Celticsblog.com and explain to all those knuckleheads why the Celtics stink even though they’re won of the highest scoring teams in the league and why Iverson is not the answer!
Tom Mandel
July 15, 2006
DJ – thanks for weighing in on the Wiz. You make an interesting point, and it’s possible that you are giving a thumbs-up to Grunfield’s strategy, which seems to be ‘improve from within.’ There seems to be some interest in acquiring Magloire, but it’s unclear what they can offer in a trade. It seems as well that there’s some interest in acquiring Deshawn Stevenson, who does nothing for me.
So… maybe we have to count on the development of Andrey Blatche, the increasing chemistry the team showed in the 2d 1/2 of last season (reflected in an improving record), and… Oleksiy Pecherov?
How does Brendan Haywood rank using your WP48 rating method (and how could I calculate thaat myself w/o building the model on my own in Excel)?
Jason
July 15, 2006
Indeed, if there’s somewhere I could get a better handle on calculating wins produced, I’d greatly appreciate it. I’ve scoured the book (and the 1999 journal article), but my attempts to build the model in excel seem to be off when I compare them to the numbers I’ve seen posted here. Any pointers?
dberri
July 15, 2006
Jason and Tom,
Chapters six and seven in the book were easily the hardest to write. The objective was to explain how to calculate Wins Produced but not use any math. My hope was that the explanation would be clear enough that anyone could easily follow how this is done.
After a few inquiries, though, I see that this is not true. People have trouble following the strict verbal description.
This next month I am going to be revising the working paper that describes the Wins Produced methodology. This paper has all the equations. When that is done I will either post the paper (which I may or may not be able to do depending upon who decides to publish this work) or an abbreviated version. This paper will definitely be published, so at some point everyone will have access to non-abbreviated (is that a word?) paper.
In the meantime, I can only ask for patience.
Tom Mandel
July 17, 2006
for “non-abbreviated” you might substitute “complete”.
I was able to follow the conceptual outlines of your book’s WP metric. It is your simpler metric combined w/ some math that somehow factors in the influence on a player’s score of the number of assists others on the team produce.
Is that correct?
Tom Mandel
July 17, 2006
In re: do you undervalue scorers — if so, and given that the wp metric is 96% associated w/ # of a team’s wins, then you would have to be *undervalueing* something else.
Have you divided the league into teams that win 20-29 games, those that win 30-39 games, those w/ 40-49 wins, those victorious between 50-59 times, etc. and determined whether wp of all team players tracks each category equally (or close)? Have you tracked it against any other order of team success (least points scored against, etc.)? Are there outliers anywhere at all?
Tom Mandel
July 17, 2006
Sorry, I meant to say, obviously, that if you are undervalueing scorers, you must be *overvalueing* something else.
dberri
July 17, 2006
Tom,
I really don’t think we undervalue scorers. The point of the post was that most scorers also rank quite high in Wins Produced.
Jason
July 18, 2006
I don’t think it’s so much “overvaluing” or “undervaluing” particular things as it is seeing if a particular player’s contribution is reflected in their statistics
For the *team* the regressions seem to hold very well. However, partitioning the individual contributions isn’t necessarily as straightforward, and this isn’t just a diminishing returns of having a team full of Ben Wallaces. I’m thinking particularly about players who have reasonable “win scores” or “wins produced” but do not seem to have an actual impact on the team’s wins.
For example, Troy Murphy does reasonably well by a measure of ‘win score.’ Even adjusting for positions (which is where I seem to get off when I try to replicate your numbers) he seems to be above average, yet the Warriors were significantly more ‘negative’ in their +/- when he was in the game. Their team rebounding % fell during the time he played and their FG% yielded to opponents rose. Both important measures of success for the team and indicate that despite his *individual* accomplishments, when he’s been in the game, the Warriors haven’t done as well as when he’s been out. If win-score works for the team, I’d then have to conclude that Murphy is a detriment to teammates’ wins produced (e.g. he interfers with a teammate’s rebounding, he allows the opposition to shoot the ball better).
These aren’t necessarily ‘intangibles’ but are an indication that the partitioning of the variables may not completely reflect the values to whom they ultimately get creditied and perhaps that the partitioning of some stats is as important as the value itself.
Tom Mandel
July 19, 2006
Hmmm, Jason, that’s kind of interesting. I’d be interested to hear DJ’s take on this.
dberri
July 19, 2006
Tom,
I really don’t want to get into the difficulties with plus-minus at the moment. We do talk about it briefly in the book so you can get a sense of my thoughts in Chapter Six.
Perhaps I will blog on this in the near future.
Qball
July 20, 2006
This may or may not be the most suitable place for this comment but since we’re saying that scores might be undervalued by Win Produced (the exact metric for which I have still yet to see…I definitely don’t mind seeing the math behind this one…but I’ll assume to be quite similar to Win Score), I would argue that playmaker are even more undervalued than scores.
In Win Score, assists are only counted half as much as rebounds, steals, etc. However, the conventional wisdom says that playmaker (mainly point guards) are essential to the success of a team. The PG are the one who gets the assists and probably also has the highest turnover amounts on the team (and turnover are a full point deduction). Of course, guards generally gets their share of steals to raise their win score contribution. However, just from looking at the win score metric (and not really crunching any data yet), it would appear to me that playmakers are going to be significantly undervalued.
dberri
July 20, 2006
Qball,
Okay, I think people missed the point of this post. I am saying exactly the opposite. Scorers are not undervalued by Wins Produced. Most scorers are above average in Wins Produced.
Dean Oliver suggested that I post the analysis of Reggie Miller. Miller is basically a pure scorer. He was also consistently well above average in Wins Produced during his career. So I do not think the evidence suggests that our methods undervalue scorers.
As for assists, I will repeat what I said in response to your other post. If you believe our approach is flawed then you need to provide evidence. Specifically, if we value assists as you suggest, can we increase explanatory power? Can we increase predictive power? If we can, then we have a better model and that is great.
In the end, we need an accurate measure of a player’s marginal productivity to do research. Anyone who can help us by providing an estimate of marginal product that improves upon our model’s explanatory and predictive power is welcomed.
Jason
July 20, 2006
Coming up with models that can explain what happened (measured by how close the extrapolation from recorded data matches what happened) is the easier part.
We know that the average margin of victory (or defeat) is a good indicator of the team’s record, but deciding who on the team was responsible takes another level of analysis. A player’s scoring average doesn’t capture much of this since it captures none of what the other team did. HOwever, player plus minus as presented as scores per 100 possessions for the team and opponents, converted into a pythagorean win expectancy and then weighted by playing time comes *very* close to an actual team’s record. That’s because it’s really just a product of the team’s points scored for and against which usually does a very good job of indicating the team’s records. By separating it out as an individual’s +/- we’re able to get an idea of who was more or less responsible for the wins. It’s not perfect, but it certainly explains most of what we’ve seen (emphasis on seen as in past tense).
But the problem with +/- comes in a couple forms. A player paired with lousy players will tend to have a poor +/- and a player paired with good players will have a better one and this is likely true without much regard for what that one guy does. In this regard, this is probably true of a wins produced model as well. Being on the court with poor defenders mean the opposition makes more shots, provides fewer opportunities for rebounds (and a better chance that the other team can grab those rebounds).
This influence of teammates may reflect negatively on wins produced. However, the word “may” is important here as it’s not clear to me how substantial this is to calculated wins produced. It’s certainly less clear than +/- and in this regard, wins produced is a superior model because it provides some ability, even if imperfect, to separate out an individual component of players that isn’t wholly reflective of the results of the team as a unit.
The other side of comparing models is prediction and separated +/- doesn’t seem to hold all that well as a predictor of future +/- and consequently of future team results. If the explanation of the past doesn’t provide a tool to predict the future, there’s no reason to head off to Vegas with model in hand and it’s a good indicator that the method of explanation wasn’t capturing something.
It may be that there’s no way to completely isolate out individual contributions to the degree that there is in baseball where hitting is largely an isolated contest between pitcher and hitter and the influence of situation and thus influence of teammates is minimized, but a model that captures more of *future* performance than random noise does indicate to me that the factors measured that go into the model are both important to the outcome and are not influenced by the behavior of other players to such a degree that their measures are meaningless. It may be that Iverson would shoot better if he had better teammates regularly or Harrington would have a higher win score if the rest of the team didn’t suck, but I suspect that these influences are not the majority of the influence else players’ wins produced should fluctuate greatly and there shouldn’t be as much covariance in summed wins produced and a team’s actual record.
dberri
July 20, 2006
Jason,
Brilliant comment. I think you summarized the issues quite well. By the way, do you know how well +/- predicts future +/-? I have never seen research on this but I hear it is not a good predictor of the future.
I would add that we don’t want a model that predicts the future perfectly. Not sure anyone wants to watch a sport where we know the outcomes with certainty before the games start to be played.
Jason
July 21, 2006
I’ve just done an ad hoc analysis of some +/- and noticed that they move all over the place. Players can go from good to bad in the course of a season.
Interestingly enough, Al Harrington hasn’t been good by this measure either, though he’s fluctuated from marginally negative to very negative relative to his teams when he wasn’t on the court. However, over the last 4 seasons, both the Hawks and Pacers have had better results when he wasn’t in the game. The Hawks were terrible and a bit worse with him. The Pacers were pretty good, but not as good with him.
disappointmentzone
July 24, 2006
Harrington has cashed in for six years and $57 million according ESPN:
http://sports.espn.go.com/nba/columns/story?id=2528093
dberri
July 24, 2006
Thanks Disappointmentzone (or Joel Witmer) for the update on Al Harrington. For those interested in a neat site on Cleveland sports, check out http://disappointmentzone.wordpress.com/