Here are a few semi-random weekend thoughts:
More on Shaq and Marion
A few days ago I posted a comment on the Shaq trade. In this column was the following argument:
Let’s imagine that Shaq and Steve Kerr are right. The issue was a lack of motivation in Miami and in Phoenix Shaq will return to the player we saw in 2004-05. That year Shaq posted a 0.306 WP48. If that’s the O’Neal that arrives in Phoenix, will the Suns contend for a title in 2008?
Unfortunately, the answer is probably no. Marion was already offering a 0.314 WP48. So even if the old Shaq can become the Shaq of old, it’s not clear this team is better than the Lakers (with a healthy Bynum). They are certainly not better than the Boston Celtics.
Matthew Yglesias responded with this comment: Any Way You Look
The Yglesias post begins as follows:
It occurred to me that maybe Dave Berri has some counterintuitive argument as to why the Shaq-Matrix trade makes sense for Phoenix. The answer is no. Instead, he has a counterintuitive argument that even if Shaq were to return to his 2004-2005 season level of production the trade still wouldn’t help Phoenix. And, of course, that’s not going to happen.
When I read this I thought it would be worthwhile to elaborate briefly on the career performances of Shaq and Marion. Entering this season, Marion had a career WP48 [Wins Produced per 48 minutes] of 0.303. His peak was 2000-01, when his mark hit 0.414. Despite this mark, in five seasons Marion’s WP48 ranged between 0.200 and 0.300.
In contrast, Shaq’s average mark from 1992-93 to 2004-05 was 0.344. In only one year did Shaq have a mark below 0.300 (1995-96) and his peak of 0.428 (1999-00) eclipsed the best mark of Marion.
In sum, if we compared Shaq’s career to Marion’s career, we would conclude that Shaq has been the more productive player. I would emphasize, though, that the difference is not very large. If Shaq posts a 0.344 WP48 for the Suns in the second half of this season, Phoenix could expect to win about one more game (relative to what it would have received from Marion).
Of course, few people expect Shaq to return to what he was for much of his career. So this argument is probably not one we will be able to empirically check.
Value of Midpoint Analysis
We can, though, look at the evidence behind a related point. Can we look at a player’s first half performance and know the remainder of the season?
Just like I did this year, in 2006-07 I downloaded each team’s data after 41 games. Today I looked at the relationship between a player’s per-minute Win Score over the first half of the 2006-07 season and
1. the player’s per-minute Win Score in the second half of 2o06-07.
2. the player’s per-minute Win Score at the end of the 2006-07 season.
Looking at players who played at least 500 minutes in each half, for the first relationship – first and second half per-minute Win Score – I found a 0.84 correlation. This is basically what you see when you look at performance from season-to-season, so again we have evidence that performance in the NBA is fairly consistent across time.
When we look at the second relationship – first half and final per-minute Win Score – we find a 0.96 correlation. This is not too surprising, since half of the final mark is comprised of the first half numbers. What it does tell us, though, is that what we have seen for the first half of the season from each player is roughly what we are going to see when the season is over. So Dwight Howard can expect to win this year’s M2P [Most Productive Player] award (not that he gets anything for this).
Playoff Picture
Quick observation on the playoffs… right now the New Jersey Nets hold down the eight spot in the Eastern Conference with a mark of 21-29. The Pacers, 76ers, Bulls, Bucks, and Bobcats are each posting marks that are even worse, yet each team can also claim it is within three games of the playoffs. In other words, the Bucks – who I noted were expected to be a bad team this year – might still make the playoffs because so many bad teams play in the East.
In contrast, the Sacramento Kings are only three games below 0.500. Such a mark would give the Kings the 7th seed in the East. But in the West, Sacramento is 6.5 games out of the playoffs and already dreaming of lottery balls.
Roger Clemens Column
One final note… on Saturday Justin Wolfers gave me a heads up on his latest column. Writing with three other Wharton School professors at the New York Times, Wolfers and company consider the statistical evidence that Roger Clemens did something he might regret. Here is a link to their Keeping Score column:
Report Backing Clemens Chooses Its Facts Carefully
In addition to recommending this column, I would also recommend the work of JC Bradbury on this subject.
- DJ
Our research on the NBA was summarized HERE.
The Technical Notes at wagesofwins.com provides substantially more information on the published research behind Wins Produced and Win Score
Wins Produced, Win Score, and PAWSmin are also discussed in the following posts:
Simple Models of Player Performance
What Wins Produced Says and What It Does Not Say
Introducing PAWSmin — and a Defense of Box Score Statistics
Finally, A Guide to Evaluating Models contains useful hints on how to interpret and evaluate statistical models.
30 responses so far ↓
William // February 10, 2008 at 7:11 am
Perhaps the first-half second-half analysis needs to be broadened somewhat? Why don’t you look at the productivity of players who get traded (and receive statistically significant minutes with each team)? Test whether it’s true that some guys just need a change of scenery.
You’ll probably have to include a larger number of seasons in the sample, but I can think of three high-profile players who got traded recently: AI, Ray Allen, and KG. Also, you’d have to control for ‘aging-SG’ for Ray Allen. Plus, this time next year you could add Kyle Korver, Pau Gasol and Shaq/Marion to the sample.
I say this because Shaq/Marion won’t be in the same situation they were in the first half. If pace and usage don’t change productivity, then they won’t cloud the ‘being-traded’ effect.
Kent // February 10, 2008 at 9:25 am
Wolfers copied Ray Fair! — http://dberri.wordpress.com/2007/06/04/rocket-science-clemens-and-%E2%80%98roids/
Stan // February 10, 2008 at 2:28 pm
Hmm … interesting analysis. My gut reaction still says Shaq will do better with better teammates around him, but then again my gut tells me the Dallas Mavericks will be champions every year.
Quick question, how were you able to get mid-year stats? Thanks for your great site. I read it every day!!!
Jason // February 10, 2008 at 4:50 pm
i’m pretty sure the Suns won’t be as good in the regular season – the real question is what happens in the post season. it’ll be interesting to see how the bigs in the west will fare against the Suns IF Shaq is playing reasonably well.
dberri // February 10, 2008 at 6:12 pm
Stan,
I just downloaded these stats last year as each team hit the 41 game mark. I did the same this year, so I have two years of data now.
Mark // February 10, 2008 at 8:35 pm
It will be interesting to see if shaq and marion both maintain their existing rebounding rates, and so do their teammates (a key WoW contention).
dberri, what are your expectations for how each team (and the traded players in particular) in terms of how they will rebound?
Dan // February 10, 2008 at 8:46 pm
Shaq is another guy whose value is higher than his WP, because his distorting effect on opposing defenses isn’t reflected in the box score. This guy has been double and even triple teamed for years to the detriment of his numbers and the benefit of his teammates’. Furthermore his physical impact on the opposing center certainly softens the lane as a game progresses, making things easier for teammates.
Dan // February 10, 2008 at 8:49 pm
Thought of it right when I posted….the combined physical AND mental effects he has on opposing centers due to the focus he commands….makes things easier for everyone else. Erick Dampier always played Shaq scared, and that effect reverberated.
And of course, Shaq will have to play better than he has recently to reap these intangible benefits.
RG73 // February 11, 2008 at 9:03 am
If Shaq has some influence on the opposing team it will necessarily be reflected in the box score if it is a real effect, as opposed to some hand-waving just so story. For instance, you’d expect to see it in offensive efficiency ratings (for both Shaq and his team). Given he is coming from the worst team in the league (which was the worst when he was playing for them), with the 29th worst offensive rating in the league, there isn’t the slightest evidence to indicate Shaq is having a “distorting” effect on any defenses.
Shaq does, btw, routinely demand double or triple teams anymore. He does not intimidate (everyone remembers that he was contained by 6′7″ Ben Wallace and 6′8″ Nocioni in the playoffs last season). The concern is for his coaches who worry about his talent for fouling now (1 every 7 minutes or so) and his inability to make a free throw when he is inevitably fouled.
Nope, there’s no intangible, magical, invisible Shaq effect. You either have impact or you don’t. Every other player in the league who supposedly has these intangibles, well, gee, they tend to have high WP48 scores–so it shows up in the boxscore. Shaq is not the exception. One is especially hard pressed to explain how Shaq had impact that showed up his entire career in WP48, but now, despite his declining numbers, the numbers that measured his impact don’t, in fact measure his impact anymore. Does that make any sense?
Dan // February 11, 2008 at 10:10 am
I should have noted that my remarks referred to Shaq’s historical numbers – when he did always draw doubles – and the premise that even at his best he wasn’t significantly better than Marion.
“For instance, you’d expect to see it in offensive efficiency ratings (for both Shaq and his team).”
My point exactly, Shaq’s dominance and the attention it drew made things easier on his teammates, but the increased attention would NOT have made his statistics or WP better, that is irrational.
“Nope, there’s no intangible, magical, invisible Shaq effect. You either have impact or you don’t.”
Who’s saying he doesn’t have impact? All I’m saying is that WP doesn’t tell the whole story, and anyone who watches basketball should understand that the dynamics of team play aren’t all reflected by statistics.
jwstringer // February 11, 2008 at 1:49 pm
Just hoping to verify my comment will be displayed.
Thanks!
jwstringer // February 11, 2008 at 1:55 pm
Is there a place I can view the guidelines for commenting on blog postings here on the site?
Thank you.
Dan // February 11, 2008 at 6:59 pm
I read this again and understand I sound like a critical jerk. Just to be sure, I wouldn’t read the site if I didn’t like the system, its impressive. And Dave I obviously am not trying to rip your intelligence nor basketball understanding.
I think my only problem is when guys are misidentified as poor players when they are doing more……whether its Iverson drawing guys away from the hoop or Shaq pounding centers into submission, these things play a significant role.
The sticking point is always that WP is astonishingly accurate at predicting win totals, but I think its because everything nets out between teammates – i.e. Iverson’s speedy slashing creates easier offensive rebounds. My guess.
SHC // February 11, 2008 at 8:07 pm
Wow, talk about trusting his simplistic model to death. If Shaq becomes his dominant self again (I, just like everyone else, am highly doubtful that this will happen), the overall effect is not going to be a simple substitution of Marion’s numbers with Shaq’s numbers.
Let’s imagine another scenario – let’s assume that Marion was traded for Duncan (both of them have similar WP48 this season). Does Berri honestly believe that with such a trade, Suns would be no better than what they are right now?
The biggest problem with Berri’s analyses is that he assumes that basketball is a collection of individual talents – each of whose production is completely independent. Terrible, terrible assumption.
Mike H // February 11, 2008 at 8:48 pm
“Terrible, terrible assumption.”
Strong words. dberri has laid out his argument and his evidence in the book and on this site. He also frequently concedes the obvious that the model is not perfect. Do you have any evidence for your statement, or should we just trust your basketball expertise?
SHC // February 11, 2008 at 9:55 pm
Even with a flawed system like WP48, there are already plenty of evidences that certain players benefit from playing in certain teams. Look at Boston Celtics’ for instance, guys like Eddie House, Kendrick Perkins – their WP48’s are way better this year than last year’s (at least up to this point).
How about Mike Dunleavy? He’s playing so much better after changing his team.
Then there’re players whose productions increase because their roles expand – Scottie Pippen had his best years in 93-94-95 when Jordan was playing baseball and he was the focal point of the team.
That basketball player production is not an independent event should be pretty self-evident. Dramatic changes are probably rare, but if you get .3-.4 WP48 changes on several players due to a team style change, coaching change, a trade, or whatever, those translate into a lot wins.
Also, with a prime-time Shaq, you become a strong half-court team, with a prime-time Marion, you’d struggle mightily to score in half-court sets. That would have huge implications in playoffs for teams like Suns.
Lastly, WP48 is not a unique solution. using the exactly the same methodology but using different equations for possessions, you should be able to come up with different weight sets that would require some different ‘team’ adjustments, but would have exactly the same predictive power as WP48.
Mike H // February 11, 2008 at 11:13 pm
paragraph 1 & 2 – maybe it’s evidence, maybe it’s a temporary blip. Either way it’s anecdotal – that’s 3 out of whatever number of players we have data on that have changed teams.
paragraph 3 – good players suppress the w48 of their teammates
paragraph 4 – “dramatic changes are probably rare”, can be taken as evidence against the terribleness of the assumption (that you claim dberri has made).
paragraph 5 – nothing useful
paragraph 6 – nothing useful
So, is there any more evidence as to the terribleness of the assumption that would justify such a strong statement? Even more importantly, is there any evidence as to dberri making the assumption that basketball is a game of five on five?
Mike H // February 11, 2008 at 11:16 pm
LOL, that should be “Even more importantly, is there any evidence as to dberri making the assumption that basketball is essentially a game of five one on one contests, without any team effects?”
antonio // February 11, 2008 at 11:35 pm
read his book. in his book he states his evidence. it is probably also somewhere on this site, but i am not really sure where. I can tell you his conclusion is something like player stats dont really change, even if players switch teams and play with new teammates, so how much of an effect can teammates have? And if teammates do have an effect, than all teammates just about have the same effect considering that almost every team plays a tradition pg, sg, sf, pf, c. (the warriors really don’t) and players play the same regardless of where they are
antonio // February 11, 2008 at 11:36 pm
and his test was done on the whole nba, so yea, you could point out a few players who changed, i am sure, but his data and conclusions showed that overall there was not a significant change
SHC // February 12, 2008 at 12:20 am
Mike-H, well you just said yourself that good players suppress teammates’ WP48. Aren’t you already saying that a player’s production is not an independent event? Way to contradict yourself.
Also, I believe that dberri made statements that players’ production only changes perhaps within 10% even if they change teams. But let’s say, that a great player made 8 of his teammates produce 10% more on the average. This could add up to .100 or more WP48. This would translate into quite a few more wins.
And you’re saying that there’s nothing useful in saying that WP48 is not a unique solution?
Look at http://www.wagesofwins.com/CalculatingWinsProduced.html
In his possession computation, he doesn’t use opponent’s rebounds AT ALL. This is a completely arbitrary choice. But there’s absolute no reason why there should not be opponents rebounds here. The only thing this indicates is that berri wanted to make computations more convenient (minimizing ‘team’ adjustments).
Mike H // February 12, 2008 at 1:00 am
Paragraph 1 – No, that’s what dberri’s research suggests. You are the one saying that dberri’s makes the assumption that a player’s production is an independent event.
Paragraph 2 – anecdote
Paragraph 3 – no, I’m not
Paragraph 4 – I guess I’ll have to take your word on it that there’s ‘absolute no reason why’ and on what it indicates that dberri did.
Mike H // February 12, 2008 at 1:01 am
Basically, this is the scenario we are in:
[The theatre darkens and a booming voice speaks]
They wrote a peer-reviewed book containing a model for measuring individual basketball performance…
He said the model contained a “terrible, terrible assumption”…
He said “O Rly?”…
Coming soon to an intertube near you…
SHC fights against all odds as, ‘Some Internet Dude’
Mike H gives the performance of a nighttime as, ‘Some Other Internet Dude’
And the enigmatic dberri reprises his role as, ‘The Professor From Cal State, Bakersfield’
In…
‘Who Needs to Provide More Evidence in Order for their Statements to be Taken Seriously?’
The Franchise // February 12, 2008 at 9:00 am
This is the classic argument here, though. The rubric is something like this:
“The model overvalues rebounds/undervalues scoring.”
“How so?”
“It’s just obvious. Don’t you watch the games?” (This may be accompanied by one example or vignette.)
“Is there evidence to support your assertion that the data there can be generalizable to the league as a whole?”
“…”
Here’s the thing: this is the best system to use only box score data to explain and predict wins. That simplicity makes it a very useful tool. Additionally, it seems to be a better system than certain techniques that involve more work to collect data, like +/-. Can a more accurate system be developed? Absolutely. However, it would require a much more complicated model to even garner slight improvements.
This model was not built with certain results in mind. It was built correctly, starting from the results–wins–and comparing the correlations of the measured inputs made by respective laborers. The WP model is the result. This is the right way to build a model. The wrong way would be to decide various weights arbitrarily, or based on raw intuition, like PER does. (Or like the posters that “know KOBE is the greatest EVAR!” because they watch him sink a game-winner on tv.)
Jason // February 12, 2008 at 10:59 am
“In his possession computation, he doesn’t use opponent’s rebounds AT ALL. This is a completely arbitrary choice. But there’s absolute no reason why there should not be opponents rebounds here. The only thing this indicates is that berri wanted to make computations more convenient (minimizing ‘team’ adjustments).”
Arbitrary? Here are the things that are counted: FG made and FT made (or alternatively, points scored, though this results in different ‘costs’ for attempts) FGA and FTA, steals, rebounds, turnovers, blocks and assists.
Missed shots minus offensive rebounds *does* count the opponent’s defensive rebounds since a missed shot, unless there’s an offensive rebound, results in a defensive rebound for the opposition. Counting it twice would be just that: counting a single event twice. This is not arbitrary. It’s a product of possession logic.
Alternatively, to calculate possessions, one could look at opponents total defensive rebounds and ignore your team’s missed shots, but this leads to the problem of ‘team rebounds’ (usually missed shots that go out of bounds resulting in a change of possession). Counting missed shots *instead* of defensive rebounds by the opposition solves this problem on one half of the court (possessions utilized) since it doesn’t matter who got the rebound to determine the number of possessions, only that the possession ended.
SHC // February 12, 2008 at 2:08 pm
Well, let’s look at dberri’s statement in THIS article : ‘If Shaq posts a 0.344 WP48 for the Suns in the second half of this season, Phoenix could expect to win about one more game (relative to what it would have received from Marion).’
This sounds as if he’s assuming a player’s production is independent of other players’ productions. If he thinks that a player can affect other teammates, he certainly doesn’t say it.
Also, you think that a published book is an academically accepted research work? Publishing a book doesn’t require approvals from other experts. Only papers for respectable journals or conferences go through rigorous peer-reviews. Additionally, papers for less-than-respectable journals/conferences are often filled with errors and loopholes.
Jason, just to show you an example of ‘infinitely many’ ways of computing possessions, you can simply swap how he computes PE and PA : PE = FGM + 0.47*FTM + OppREBD + TO + OppREBTM. PA = OppFGA + 0.47*OppFTA + OppTO – OppREBO. All of these compute the correct number of possessions.
Others have already shown that you can put different weights on opponents’ rebounds, field goals missed, own rebounds and still compute the correct number of possessions.
http://sonicscentral.com/apbrmetrics/viewtopic.php?t=1595&postdays=0&postorder=asc&start=15
Mike H // February 12, 2008 at 4:27 pm
Paragraphs 1 & 2 – dberri says that W48 is pretty consistent for players even across teams. If some internet dude wants to claim otherwise, then fine. If some internet dude wants to pick a post where the exact (seemingly small) team effects are not explicitely acknowledged (and aren’t particularly salient to the point of the post), then fine.
Paragraph 3 – No, I don’t. However, dberri’s book was peer reviewed (aren’t all Stanford University Press books?). If some internet dude thinks that his anecdotes justify his strong unequivocal statements as to the incorrectness of a model from a peer-review book because sometimes papers that pass peer-review turn out to have flaws, then fine.
Paragraph 5 – that’s great for the advancement of the debate as a whole, but what does that have to do with you making such concrete statements against a particular model (peer-reviewed, by an expert, yadda yadda..) yet providing only a few anecdotes as evidence?
SHC // February 13, 2008 at 3:48 pm
dberri has only said that a player’s productions are marginally different throughout different teams on the average. This is NOT the same thing as saying that a player doesn’t affect the collective performance of his teammates. If a player helps his teammates be 10% more productive, that would obviously not show up as dramatic improvement for individual players. But he is in fact responsible for adding 4-6 wins for the whole team.
Also, I’m pretty sure that when you are publishing a book (through a university press or not), you are not putting your work through rigorous technical analyses of other domain experts. Sure, you might get some reviews from editors who probably have no domain knowledge, and you might get it proof-read by some of your friends. But nobody is going to stop you from publishing it.
As for your last point, I was trying to show that the model used by dberri derives from possession equations, and yet those possession equations are NOT unique. It doesn’t matter whether I came up with the counter-examples or somebody else has already come up with them. You keep mentioning that I’m only coming up with a few anecdotes to prove my point but there’s nothing anecdotal about proving mathematical facts.
SHC // February 13, 2008 at 3:57 pm
dberri said that a player’s productions seem to be marginally different from team to team. However, this is NOT the same thing as saying that a player affects the collective performance of a team. If a player helps his team achieve ~10% more, he is personally responsible for 4~6 more wins.
I’m pretty sure that publishing a book (through a university press or not) is not subject to expert approvals. He might get it reviewed by editors (with probably very little domain knowledge), and get it proof-read by his ‘friends’. But nobody’s going to stop him from publishing it. After all, the book doesn’t explain a lot about mathematical details, so who’s going to verify all the steps?
As for your last point, I was trying to show that dberri’s possession equations (that form the basis of his WP48) are NOT unique. I proved that by simply showing a few counter-examples. It doesn’t matter whether those were my original ideas or not. Mathematical facts are what they are, and they are not for ‘debate.’
SHC2 // February 13, 2008 at 3:59 pm
Did I just get banned? My responses are not showing up. Lol