The second part of my Q&A with Michael Schwartz (of Valley of the Suns) has now been posted. In part two, we discussed
- why the Suns were wise to allow Amare Stoudemire to depart.
- how the Suns might be able to overcome the loss of Stoudemire.
At the very end of the discussion, I note that Suns will probably make the playoffs in 2011 and Amare will probably be sitting at home. Upon further review, I still think the Suns are a playoff contender. But I am not sure about the Knicks. At least, I think it is possible the Knicks can contend for a final playoff spot in the East.
Although our discussion focused on the Phoenix Suns – and one presumes that the readers at Valley of the Suns are interested in that topic – the comments seemed to focus on a different topic. In discussing the value of Amare Stoudemire, I noted the Top 30 players according to Wins Produced in 2009-10. This list – judging by the comments – made some people unhappy. Specifically (and this was surprising since this was a website devoted to the Suns), people were unhappy that Kobe Bryant – who produced 9.9 wins and ranked 32nd in the league last year – was not on the list of Top 30 players.
Here is one comment that captures the general theme: “I don’t know what’s worse: saying Lamar Odom is better than Kobe or saying David Lee is the 11th best player in the league. C’mon Schwartz, if you’re going to do advanced stats, you might as well use the ones that make sense.”
Although I prefer Wins Produced (for reasons I have stated many times in the past), one should note that Win Shares (at Basketball-Reference) lists David Lee as the 12th most productive player in the league (ahead of Kobe). Adjusted plus-minus (a model that I think has some significant problems) does place Kobe ahead of Lee and Odom. But it also says Matt Bonner is the 12th most productive player in the game (ahead of Tim Duncan).
As I noted back in 2007, all of the “advanced” statistical models (well, maybe not the Player Efficiency Rating) create rankings of players that defy conventional wisdom. That’s because conventional wisdom is driven by points scored. And the “advanced models” are not driven by scoring (well, except for the Player Efficiency Rating). So if you are an adherent to the conventional wisdom of the NBA, you are probably never going to like any of the advanced models.
And that means fans of the NBA have a choice. They can simply follow the conventional wisdom. That means you look at a player’s scoring and believe that Allen Iverson, Carmelo Anthony, Rudy Gay, etc… are above average (if not absolutely great) players. Of course, when your team gets these players you may not see as many wins as you like. But then you can turn to the subject of chemistry (or fairy dust, astrology, etc…) and still remain pretty happy.
If that approach doesn’t work for you, there are a few advanced statistical models you can examine. Of course, there is a problem with this approach. More than one model has been offered. How do you decide which model is the “best”?
Well, the first step is to spend some time trying to figure out how each model is constructed. That means you should spend some time reading Mathletics (by Wayne Winston, who explains adjusted plus-minus) and Basketball on Paper (by Dean Oliver, who provided the analysis that serves as the basis of Win Shares). And for Wins Produced, you should spend some time reading Stumbling on Wins (and reviewing material at stumblingonwins.com and in this forum).
Once you understand how the models are constructed then you have to think about how models should be evaluated. Some time ago I offered A Guide to Evaluating Models. There is also some discussion of model evaluation in Stumbling on Wins. And I have a chapter in a forthcoming sports economics collection that comments on this subject. Obviously I think each of these works will help. But whether you agree or not, it is clear that you will have to do some thinking if you choose to look at “advanced” models.
So the advanced models can provide a better way to look at basketball (for whatever that’s worth). But they do come with a cost (i.e. you have to do some thinking to appreciate how these models work). And this is why I think we are unlikely to see a day when most NBA fans abandon the “conventional” wisdom.
– DJ
P.S. Let me offer a pre-emptive comment. There are those who claim: “I don’t adhere to any model. I look at everything because all models are useful.” I find this approach to be less than convincing. At the end of the day, you still need to have some criteria in evaluating any model. If your criteria tells you that all models are useful, then I think you need to re-think your process. Although I am not sure there is one “best” model (for example, Ty Willihnganz offers a very interesting variation on Wins Produced at Courtside Analyst), I do think there are models that are better than others. And if you are looking at the models that are not as good, I am not sure you are making a good use of your time (again, for whatever that is worth).
Wages of Wins Journal 
robbieomalley
July 27, 2010
Yeah, I was going to add that Ty’s work is also very interesting. Ty’s posts are always very good, makes me try to step my game up.
Additionally, has anyone examined Hollinger’s Estimated Wins Added (EWA)? I’ve been looking at this for roughly the past week. It attributes a number of wins that each player generates above what a “replacement” player would be expected to produce.
I’d really like to do a post on the accuracy of this model, but first I must figure it out what a replacement level player produces. So far it seems like it would be easier to reconcile quantum physics and general relativity.
robbieomalley
July 27, 2010
I would note that the Nets EWA adds up to 22.7, that is above what a team of replacement players would achieve. So Hollinger is saying the Nets should have won at least 22 games (they won 12).
That doesn’t seem very close.
That is a small sample but the story on the other teams I’ve looked at isn’t much different.
Chicago Tim
July 28, 2010
There are some of us who are not qualified to evaluate statistical models. I’m afraid I evaluate models based on their predictive results, and not very scientifically, at that.
Therefore I try to keep an open mind about all of the models (even PER), and I enjoy reading predictions based on those models and attempting to keep track of the results. It’s not very scientific, but it’s much more fun than insulting statisticians who challenge conventional wisdom when I’m not qualified to get into a statistical argument.
Chicago Tim
July 28, 2010
I remember your argument that Michael Jordan might have outperformed Dennis Rodman even in a year when Rodman apparently produced more wins. You noted that Jordan was far more productive than the average shooting guard, while Rodman was not as far above the average power forward. Here’s the link:
https://dberri.wordpress.com/2006/12/04/on-jordan-and-rodman-again/
Would your rankings of the players change if we took into account standard deviations above the average for each position? Although Kobe is overrated by those who just look at scoring totals, would he be higher on the list if we took into account that he is a shooting guard, and that it is harder to find above-average shooting guards than it is to find above-average players at other positions? Should teams take that into account when they evaluate and pay for players?
Patrick
July 28, 2010
How about this: the best players may well be those who manage to score the most b/c scoring the most leads to the greatest rewards.
I.e., AI and Amare are arguably better players than Rondo and Kevin Love because they more successfully apply their athletic talents towards the particular basketball endeavor associated with the greatest rewards – scoring a lot of points, by any means necessary.
Of course they are not as productive – everyone here agrees on that. But that doesn’t necessarily mean they’re not better players; or, more accurately, that they have (or had) the potential to be better.
One has to assume that everyone wants a max contract – it’s only rational – along with all the media, peer, and fan acclaim that goes along with having a super-high scoring average. But Rondo and Love probably won’t ever have high scoring averages. Is it really because they somehow recognize that playing the game the way they do now makes them more productive than many of their higher-scoring (and more highly compensated) brethren? Or have they chosen to play as they do because they realize they’re not talented enough to compete effectively on scoring, the most valued commodity in the market?
Conversely, if the incentive structure in the NBA were magically rationalized (if, say, contracts and media, fan, and peer acclaim correlated with WP rather than scoring average), might not the Amares and AIs of the world still come out on top (or at least become much more productive) by adjusting their play to maximize their evaluation under the new (more appropriate) metric?
Just a thought. I don’t know if this is actually true – Rondo might just be flat out much, much better than AI, just as WP would lead one to think. But I think the possibility that all these volume shooters could actually be much, much more productive than they are (but have little reason to be) has been under-considered. I also think this hypothesis helps bridge the gap between CW player evaluation and the more enlightened stats-based evaluation. I.e., AI was not necessarily unproductive because he inherently ‘sucks’ or something, but because he was not properly incentivized.
Nick
July 28, 2010
@Patrick:
Noone is claiming Wins Produced makes a player, “better”, or “more talented”. It just means they contributed the most towards a teams victory statistically.
One you lose the perception that “more productive” means “better” or “more talented” these things make more sense.
It’s more valuable in a lot of scenarios to have a player who is one of the top rebounders in the league, which leads to more shots for your team, than it is to have a shooting guard, who scores at a high volume, but only moderately above average percentage.
Also, I do not believe AI, or Amare is so selfish that they would never “change their game” just because of the money. I really do think, these guys feel that this is the best way to win. And against a lesser defence, when they can keep their efficiency up (think NCAA & high school) it works.
Amare is a good player. WP seems to imply he’s above average. It’s just that he’s not dominant, like his salary will be paying him to be.
jbrett
July 28, 2010
Chicago Tim,
I think that’s a good point. I remember a fantasy football theory on player selection (back before I ripped out all my hair and started wearing this sackcloth and ashes) that argued you should target the players whose production most exceeds the others at the position; Manning’s 40 TD a year, when the next guy has 31, is more useful than LT’s 21 TD if there’s a bunch of RB in the 17-19 range. And I seem to remember the same point being made in The Wages Of Wins, comparing MJ’s ’96 with David Robinson; the Admiral had 28 WP to Jordan’s 25, but was only +2.6 SD at his position to +3.2 for MJ.
I conclude that–man, I hate saying this–there is a valid reason to say there CAN often be a difference in the Most Productive Player and (cringe) the Most Valuable Player. Last year LeBron is probably both; each position has a handful of players above .250 WP48 with major minutes, and he is further above the others at SF. The previous year, Chris Paul led the league in Wins Produced. He was leaps and bounds ahead of Rondo at PG, but LeBron was even more of an outlier at SF. I don’t have standard deviation data, but simple addition says LBJ and Rondo (#2 PG) had 45.2 Wins Produced; Paul and Gerald Wallace (#2 SF) had 44.01–maybe too close to be instructive, but illustrative of the concept.
Arturo,
Is this an idea worth pursuing with the database? Value over replacement might be interesting, but I’d be more curious as to how often the MPP and MVP are different–and for that matter, how exactly do we choose to define the MVP.
Tom Mandel
July 28, 2010
Nick makes a great point — the goal of WP48 as a measure is to understand why teams win by assigning player contribution to those wins accurately.
That’s very different from discussions of “better” and “best” — which seem to devolve quickly into highlight reels in people’s minds.
Remembering that basketball is not just competition — it’s also *entertainment* — can remind us how fun it can be to watch guys who aren’t necessarily extremely productive. Iverson would be a great example. When he was young, he was a lot of fun to watch! In a legitimate sense of the word, he was “great.”
some dude
July 28, 2010
“Noone is claiming Wins Produced makes a player, “better”, or “more talented”. It just means they contributed the most towards a teams victory statistically.”
Only when you look at it through an individual prism.
Player A can impact player B’s statistics without there being a direct relationship between the two.
If Paul Pierce gets doubled up top and swings the ball to Rondo who finds Perkins open for a dunk, Rondo and Perkins will receive all the statistical attention for making plays any player can make, but it was Paul Pierce’s game that created that play and he will receive no credit for it statistically.
So no, it doesn’t mean they contributed the most, statistically. It simply means they contributed the most through an individual box score stat assessment.
Cool_Hand_Luke
July 28, 2010
Standard deviations away from a position is interesting, but I think people are over-selling it in this discussion. The mean at all positions is .100, so let’s say (made up) SGs have a standard deviation of .050 while Cs has a deviation of .100. We’ve already got average players at both positions and we’re looking at buying a .300 C or a .250 SG for the same cost. Even though the SG is better in terms of SDs (+3 rather than +2) we should choose the C, as he will produce more wins (assuming same minutes and all that jazz).
The real question is how much you can improve, from average values, obviously the .300 C is a bigger improvement. If we already have +1 SD players at both positions then the two propositions are generally equal (.300+.150 = .250 + .200) though due to injury the latter looks like lower risk/reward. And maybe due to playoff minute distribution it’s simply better to have the best star (which in this is the .300 C). I’m not sure. The above example also assumes we lost the previous player at each position, if we kept them (a .200 C and a .150 SG) then the getting the .300 C is obviously better than the .250 SG as his backup produces more too.
Whatever the SDs are, the way to check what deal is preferable, what improvement is preferable, or who produced more is always the raw wins produced value rather than SDs away from the mean (even if you’re evaluating vs what you have, doing it with improved WP is better than improved SDs of WP). While deviations might show one season to be more impressive, it doesn’t show who was more valuable to their team.
I suspect jbrett’s point is moot though in the case of cp3 and lbj, because usually the higher average height a position has is the same as which has larger SDs, thus I’d guess CP3 was both better in absolute terms and relative to SDs at his position.
I’d keep talking but I think I’ve gone on incoherently for long enough.
Mike
July 28, 2010
It seems some people are missing Patrick’s point. I think Patrick is trying to argue that, due to inherent athleticism, if a player like AI or Amar’e were given the proper incentives (if they were getting paid by WP numbers rather than scoring averages), they would likely produce better WP numbers.
But I guess that is an argument not based on evidence, as we have no way of knowing if that is true other than our perceptions about the apparent athleticism of each player. Sorry Patrick, I tried.
BV
July 28, 2010
With their recent trade with the Warriors, the Knicks are definitely making the playoffs.
Chicago Tim
July 28, 2010
Cool_Hand_Luke
Isn’t your best available alternative important when deciding whether to enter a contract with a player? In your example, the .300 C will produce more wins than the .250 SG, but what if there are lots of great Cs available and only a few great SGs available? Might it be worthwhile to contract with a .250 SG and a .200 C rather than a .300 C and a .100 SG, if that’s your choice?
Cool_Hand_Luke
July 28, 2010
You’re totally right Tim, best alternatives are undoubtedly what you’re calculating based on.
But your example isn’t quite fair. It assumes a very good C is easily gotten, but only an average SG and that’s what you’re improving on. Surely it would be fairer to compare improving on a good SG and good C? Or average and average? Let’s use examples where you have equally good players (judged by SDs away from the mean) at both positions rather than giving an average player at one (SG in your example) and good at the other C. Thus if you say the SG is .100 you should also call the C a .100. If your C is a .200 then the SG should be however far away from the mean the C was, which using my numbers would be .150.
You have to use appropriate numbers, but even then, if the conclusion we draw is judged on totals (.300 + .100 = .400 < .250 + .200 = .450 in your version), then why use SDs in determining value?
If we change the example to two players of the same WP at different positions, and this might be what you're getting at, but one is more SDs away from the mean I would take that player, because it will be easier to obtain someone decent at the other position, but it's more of a tiebreaker in that case than anything else. So yes, if MJ and Rodman produced similar amounts then MJ should be preferred. When we have two very similar producers then the calculations definitely prefer to go for whoever is scarcer.
If we go back to my first examples and assume we're improving on players +1.5 SDs above the norm, then suddenly the SG becomes a better choice. But in this case it's because our current players are already really, really good (.175 SG and .250 C). Generally this question is (if we say SG and PG are comparable): would you prefer Deron Williams or Gasol? Generally I'd take Gasol unless our center was already producing far more than the PG.
LakerPhil
July 28, 2010
I’ve just recently started following this blog, so please excuse me if I’m touching on something that has been discussed at length in the past. Let me also preface my post with the admission that I am no statistician, so please educate me wherever my understanding seems to be insufficient.
Ok, to my questions…
I’ve come to accept WP as the most accurate predictor of team wins, and therefore, have accepted that, of all the advanced statistics out there, it is best at evaluating which players contribute most to a team’s wins.
With that being said, I have a couple problems with it that I’m hoping some of the sages on here can clarify.
1) I’m not sure I understand how one can simply compare the WP of two players and accept that one player is better than the other. I think I’m going to touch a bit on what Chicago Tim has brought up and that is positional adjustments. It would seem to me that the more accurate determination of a player’s worth is his standard deviation from the average WP for his position. I suppose this gets to positional scarcity a bit, but it would seem to make sense to make this adjustment when comparing players and determining how much money they should be making. After all, you can’t have a whole team of C. You do need someone at the SG spot.
2) There seem to be major components of the game that cannot be captured by box score statistics, and therefore, cannot be captured by WP. This would seem to especially undervalue positional defenders (like Bruce Bowen and Kwame Brown) and overvalue gamblers (like Chris Paul and Marcus Camby). A man who plays perfect straight-up defense and causes a bad shot or a non-steal induced turnover (a travel, dribbling out of bounds) would get zero credit for what is obviously a valuable skill, whereas a guy who gambles for steals will not get dinged for his guy scoring on a backdoor cut, but will only get credit for his risky attempts at steals.
Again, I’m sure someone has brought up these concerns in the past, but if someone would be so kind as to address them for me again, it’d be much appreciated.
jbrett
July 28, 2010
USA Basketball cut Gerald Wallace, kept Jeff Green. How funny will that explanation be?
dberri
July 28, 2010
LakerPhil,
Let me touch upon question two…
WP — like Dean Oliver’s work — treats defense as a team activity. So the opponent’s stats in the model are allocated across players according to minutes played. This does make some sense, since defense is often a team activity in the NBA.
But this is not the only approach possible. About three years ago I looked at allocating defense according to data at 82games.com (hopefully someone can find the post). I don’t recall that it made much difference.
Ty W. at the Courtside Analyst takes a different approach. He uses Win Score and the Win Score of a player’s opponent to evaluate players. The difficulty with this approach is that the opponent’s data isn’t very consistent across time. That being said, I link to Ty’s work all the time since I think it is a reasonable approach.
At the moment, those are the menu of choice available with respect to defense. No matter your preference, defense is part of the evaluation of player performance.
Hope that helps.
dberri
July 28, 2010
By the way…. should I just write a post on the whole defense issue? Let me know if there is any interest in me addressing this (again).
robbieomalley
July 28, 2010
I’d be interested if you want to write something.
robbieomalley
July 28, 2010
Also, http://hoopdata.com/defrebstats.aspx, provides data for drawing charges. Wouldn’t that count as effectively a steal plus a foul drawn. So would a charge be worth 1.5? And if we included this in the winscore formula, how would that change things? Would it be more accurate?
dberri
July 28, 2010
Robbiemalley,
I think that a charge currently counts as turnover. So charges are there. Using this data would change the evaluation of players somewhat. But I don’t think the impact would be that huge. This is something we should do at some point, though.
arturogalletti
July 28, 2010
jbrett,
*Nods Heads*. Let me see what I can come up with.
Westy
July 28, 2010
Count me in the camp of those who would love to see these additional data such as charges taken or the other defensive info at 82games.com added.
Yeah, for now each incremental addition may not have a huge impact. But as more defensive stats are taken, each bit of additional info will improve the player valuations. And even slight improvements are worthwhile in my estimation.
nerdnumbers
July 28, 2010
“Each bit of additional info will improve the player valuations. And even slight improvements are worthwhile in my estimation.”
I don’t entirely disagree with it. I also don’t agree with it entirely. Some data can be very noisy, and I argue adding noise to a model is note tremendously useful. Also consider something like the QB rating system. This is a terribly complicated model that is hard to interpret. If your model gets too complicated and filled with noisy data it become less useful.
Your last point is true, slight improvements are worthwhile especially considering the stage. A win can have huge ramifications for a team, so things that help that reliably are certainly worth investigating.
brgulker
July 29, 2010
I’m also interested in the analysis on defense.
Someone please correct me if I’m wrong on this. As things stand currently, the box score “debits” the offensive player with committing a TO when that player commits a “charge” (or other offensive foul).
However, the defender who draws the charge is not credited with creating an additional possession for his team.
I realize that not that many charges are drawn in the NBA, and that would seem to indicate that the impact of including them in Win Score wouldn’t be huge. But, wouldn’t it be worthwhile simply from a credibility standpoint? If the data exists, and it’s reliable, shouldn’t it simply be included just because it’s data we have?
It would also make sense to my mind given how important possession creation is in the NBA. Certainly not every offensive foul committed in the NBA is caused by the defender, but I’d have to think that many of them are.
(okay, that’s a bit oversimplistic, but I think it gets the point across)
Hayden
July 29, 2010
I think the point of all our moaning on Valley of the Suns was that putting players into maths and squeezing the entirety of basketball into a statistical model distorts the reality of the game.
Stuff like chemistry isn’t fairy dust or astrology just because it can’t be measured. Anyone who has played for a team that really gels knows understanding between players actually makes a difference. And a player like Kobe may not do well in a statistical model, but there is a benefit to the team from having someone who is considered transcendent playing next to them. That’s not something you can measure either.
I’m a musician, and I know music can be measured entirely in mathematical models. But if you try to run songs through a statistical formula to find out what makes them great, I’m sure someone’s eventually going to come up with a system that says Take On Me is the best song ever made.
And people say beauty is all about ratios. That’s valid. But I think most people would argue there’s more to the attractiveness of their partner than well aligned cheekbones.
No matter how well thought out they are, there are intangibles that just won’t show up in statistical formulas. They are a useful bit of the text, but they don’t tell the whole story.