Maurice Brooks (a feature writer at NBA.com) and John Jacobson (an editor at NBA.com) had an on-line debate last week concerning the relative merits of Deron Williams and Chris Paul. Here is an excerpt from their conversation:
Here is an instant messenger conversation I had with NBA.com editor John Jacobson.
John: Mo, take this as constructive criticism. There is no way that the Jazz’s Deron Williams is better than the Hornets Chris Paul.
Maurice: Good afternoon to you too.
John:: Oh my bad, how’s it going?
Maurice: Anyway, I don’t know about that. I originally thought that Paul was the more talented of the two, but last season Williams passed him by.
John: How do you figure?
Maurice: Where do you work at again? With the exception of Phoenix’s Steve Nash and New Jersey’s Jason Kidd, Williams was the best point guard in the league last season. He averaged 16.2 points, 9.3 assists and 3.3 boards a game and he led the Jazz to the Northwest Division title.
John: Williams’ postseason performance and Paul’s injury woes have wrongly convinced people that Deron had a better season, but if you look beyond the most basic statistics you would see that Paul was more productive. The Jazz would be better off with Paul at the helm and any Gm would be crazy not to take CP3 over his Utah rival.
John: And by the way, if you want to simply compare basic stats, Paul’s numbers were great. He put up 17.2 points, 8.9 assists and 4.3 rebounds. Across the board, he had better stats.
Maurice: Last time I checked, 9.3 was greater than 8.9. I thought you graduated from Princeton?
There is more to this conversation, but I think you get the basic idea. I would note that the link I placed in bold takes you to the following column, which I wrote last May.
This post compared Chris Paul and Deron Williams both in college and the NBA. What I would like to do today is update this column. But before I get to that, I want to note a few comments offered on this subject by Tim Legler of ESPN.com.
First, let’s look at Paul. He has put up great numbers, but the thing I like about him the best is he’s a great decision maker. Since he first had the ball in his hands at the NBA level, he was the leader of that team. …. Williams, however, will be an All-Star this season. What he did in the playoffs last season put him on the map as probably a top four point guard in the league. … If I had to take one, I’d pick Williams, because I think he’s got a little bit more in each area. I like a point guard with more size (6-3, 208 pounds) who’s more physical, who can take contact at the rim. Still, you’re not going to go wrong with either guy.
Okay, let’s review the arguments. Both Maurice and Tim place Deron Williams among the top point guards in the league. And although each thinks Paul is very good, both prefer Williams.
John differs with this perspective. And although everyone cited numbers, only John noted analysis that looked “beyond the most basic statistics” and considered the wisdom (or something like wisdom) of Win Score and Wins Produced. And as noted back in May, when we consider The Wages of Wins metrics it’s clear that Williams ain’t no Paul. At least, that’s the story told by the following tables (re-posted from the earlier column).
Table One: Williams and Paul in College
Table Two: Williams and Paul in 2006-07
What about this season? Each player has only played eight games, which is not much of a sample. Still, we have numbers so let’s see what they say.
Table Three: Chris Paul and Deron Williams After 8 Games in 2007-08
Let’s start with Williams. So far this year (and yes, this is just eight games), Williams is having his best season as a pro. Relative to an average point guard, he is above average with respect to shooting efficiency, points scored, blocked shots (just barely), assists, and personal fouls. His WP48 [Wins Produced per 48 minutes] stands at 0.189, which is well past the average mark of 0.100. And if he maintains this performance, and plays all 82 games, he will produce 12.3 wins this season.
To put that in perspective, here are the top five point guards in Wins Produced from 06-07.
Jason Kidd: 24.8 [0.405 WP48]
Steve Nash: 19.4 [0.348 WP48]
Chauncey Billups: 13.4 [0.252 WP48]
Chris Paul: 13.2 [0.270 WP48]
Gilbert Arenas: 11.0 [0.180 WP48]
Again, this is only eight games. But it looks like Williams might be a top five point guard right now.
But is he better than Paul?
Going back to Table One we see that Paul exceeds the productivity of Williams (not just the average point guard, but Williams) in the following statistics: Points scored, free throw percentage, rebounds, steals, assists, personal fouls, Win Score, WP48, and Wins Produced. Yes, Paul beats Williams in everything except shooting efficiency, blocked shots, and turnovers.
When we turn to WP48 and Wins Produced, we see how big of a difference there is betweent these two players. Again, Williams has posted a 0.189 WP48 and is on pace to produce 12.3 wins. Thus far this season, Paul – and again, this is only after eight games – has a WP48 of 0.457. That translates into 26.5 wins across 82 games. In sum, Paul is not just a top five point guard. Relative to last year, Paul would have been the most productive player in the league at any position.
Allow me to summarize. Williams is a good point guard. Last year only 35 of the 458 NBA players finished in double figures in Wins Produced. And Williams is on pace to join this elite group.
Paul, though, is on pace for one of the most productive seasons in recent NBA history. This can be seen when we consider what each player means to his team. Both the Hornets and Jazz are off to very good starts. But New Orleans is much more dependent on its star point guard. Whereas Williams produces about 19% of his team’s wins this year, Paul is responsible for 43% of his team’s victories.
Now Legler might be right that Paul is also a great leader. And he could be the second coming of Douglas MacArthur (or whatever great leader you wish to note). Paul’s contribution, though, is indeed captured by the numbers. And these numbers tell us that this is not a contest. There is just no way, and I mean no way, that Williams is as good as Paul. It wasn’t true in college. It wasn’t true the first two years of their NBA careers. And it ain’t true this season.
- DJ
Our research on the NBA was summarized HERE.
The Technical Notes at wagesofwins.com provides 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
135 responses so far ↓
Sam Cohen // November 14, 2007 at 1:00 am
I was watching when Legler made his comments. While he did end up saying he preferred Williams in the long-run, I thought his comments about Williams making the all-star game this year were more about Williams having greater public exposure (relative to Chris Paul) based on the Jazz’s playoff run last season than on the current difference in ability between the two. Probably doesn’t change the bottom line in the analysis (WP says take CP3, Legler says give him DW), but I thought that context for his all-star remarks was somewhat important.
Paulo // November 14, 2007 at 4:36 am
Sam, I agree that DW has a great chance to be an All-Star this year, but I think Legler forgot how many quality guards the West has. Off the top of my head, Nash, Kobe and T-Mac (if he’s listed at G, which I think he will be this year) would be shoo-ins. That leaves CP3, AI, Tony Parker(?), Manu, Brandon Roy (?), DW, Baron Davis, etc. fighting for, at most, two spots (2 voted by the fans, 2 reserves by coaches and 2 any-position by the coaches — one of which would most probably be a forward; of course, injuries happen). But if WP becomes a more popular measure of performance, which I think is gaining some sort of momentum, I think CP3 has to get one of those spots.
People mention how Deron has the bigger body, but CP3 isn’t frail himself; he’s got a thick body as well. Moving forward, I’d take CP3 simply because of the consistently high level of production, and how “Since he first had the ball in his hands at the NBA level, he was the leader of that team.”
Bob H. // November 14, 2007 at 4:45 am
Deron Williams does all the little things well that don’t show up in box score statistics.
Ryan Schwan // November 14, 2007 at 8:41 am
No matter what article you read that selects Deron Williams over Chris Paul, the three arguments cited are: Deron has played in the playoffs, he’s bigger, and he’s more durable.
I won’t argue about the playoffs. Chris needs to get there, and he will. Still this has more to do with quality teammates than anything else.
The bigger argument is just lame. Marko Jaric is bigger than either of them. I don’t see him on the top five list. Steve Nash is the same size as Paul. Does Nash suck? The advantage players get for being bigger, in general, is they are better rebounders, defenders and can finish better at the rim. Chris Paul draws more free throws, gets more rebounds, and according to 82games.com improves his team’s defense when on the floor much more than Deron does.
So Deron must finish better at the rim, right? Deron shoots 31% if his shots close to the basket, hits 54% and gets 19% blocked. Paul shoots 29%, hits 53% and gets 13% blocked. Both players dunk 1% of the time, Deron hitting 77%, Paul hitting 100%. Seems to me Paul’s a slightly better finisher. Oh, and Deron needs to be assisted 31% of the time on his close shots. Paul creates his own shots, being assisted only 12% of the time.
Lastly, let’s look at durability. Paul has missed a total of 21 games over two years. Deron has missed 4 games over two years. Deron has played 5561 minutes. Paul has played 5445 minutes. Considering Paul is twice the player by most statistical measures, I’d still rather have Paul playing 98% of Deron’s minutes for my team.
Mike // November 14, 2007 at 9:06 am
Ryan – Steve Nash is definitely bigger than Paul. A big deal is made of Deron Williams’ size b/c he resembles Kidd who is playing magnificently at age 35 despite having microfracture surgery. Hollinger’s analysis has pretty conclusively demonstrated that (except for Nash), smaller PG’s do not age well. Thus, although Paul is far superior than Williams, once their career is over and done with, that may not be the case…
Rasta // November 14, 2007 at 9:53 am
Are there any Jazz fans that regret drafting Williams over Paul? Are there any Hornets fans that wish Williams had fallen to them instead of Paul?
It seems to me that both teams are happy with their selections so far. They are both already among the top PGs in the league, and look to be so for many years to come.
Ryan Schwan // November 14, 2007 at 10:10 am
I won’t argue that – though I will point out that Hollinger says a player needs to meet 2 of 3 criteria to have a long career. Paul meets one of them – passing – and has improved his shooting every year. If his shot remains effective, he’s likely to play past 30 for a while.
I think of it like this: Paul is 22, and even if he can’t go far past 30, the Hornets will get a good 10-12 years out of him by his career’s end. Paul had almost twice the WP48 of Deron last year. An almost exact match in the difference of Win scores last year is Rashard Lewis and LeBron James. Would you like to have 10-12 years of Lebron James, or 16-17 years of Rashard Lewis?
T.G. Randini // November 14, 2007 at 10:20 am
Mr. Berri,
I have an MBA in Finance from the University of Chicago with Highest Honors… so I’m not a Neanderthal jock who hates stat geeks. I enjoyed WAGES OF WINS when it first came out and have enjoyed reading sports analysis and sabermetrics for more than 25 years now. (I bought the very first Bill James Abstract.)
BUT… (major BUT here…)
The MORE that most of you statheads over-analyze the game… the FURTHER AWAY you get from the truth.
Topic A: (Example… Joe Dumars)
There was a comment yesterday in your column “ARE SCORERS OVER-HYPED” (a comment not made specifically by you) about Mr. Dumars (a shooting guard) only averaging 3 rebounds per 48 minutes.
In your ‘win calculations’… this would be detrimental to Mr. Dumars’ ‘Wins Produced’.
WHAT you are failing to see… and what many others are failing to see… is that Joe Dumars is not playing on a team of FIVE JOE DUMARS… he is playing on a team called the DETROIT PISTONS.
Basketball is a team sport. Someone coaching the team (Chuck Daly) figured out that Mr. Rodman was more efficient at grabbing rebounds than Mr. Dumars… so he had Mr. Rodman concentrate on that aspect of the game. This same coach (again, Mr. Chuck Daly) ALSO figured out that Mr. Dumars was better at hitting 20 foot jumpers than Mr. Rodman was. So he had Mr. Dumars take most of those jumpers… and not Mr. Rodman.
So… Mr. Rodman camps out near the basket and does what he is best at doing… and Mr. Dumars hangs back and does what HE is best at doing.
And the team wins a couple of championships. Mr. Rodman is happy. Mr. Daly is happy. And Mr. Dumars is happy… until he reads one of the stathead blogs and sees that he is an inefficient rebounder and a less efficient player than he could be.
WAKE UP EVERYBODY!
1. Joe Dumars’ personal ‘Wins Produced’ are lower (low rebounding) only because he has fewer opportunities for gaining rebounds because he has an exceptionally efficient rebounder on his OWN TEAM (Mr. Rodman) limiting his chances at rebounding.
2. But this is NOT BAD because Mr. Rodman is on THE SAME TEAM AS MR. DUMARS.
3. Mr. Dumars is only doing what Mr. Daly wants him to do. (Hit 20 foot jumpers and play tenacious defense.)
4. But Mr. Dumars is punished by stathead geeks in their formulas.
5. Note however… the championship rings he wears.
Basketball… unlike baseball… is a continuous-activity sport rather than a discrete-activity sport.
Where baseball players’ stats are more like the stock market (a game where everybody can win)… basketball players’ stats are more like the commodities markets (zero sum games where there is a loser for every winner.)
To wit:
If Alex Rodriguez hits a homer… it doesn’t affect HIS TEAMMATE Derek Jeter’s opportunity to hit a homer. (Baseball = discrete) (Baseball is not a zero sum game)
But if Dennis Rodman grabs a defensive rebound… it affects HIS TEAMMATE Joe Dumars’ opportunity to grab the rebound. If fact, it reduces his chance to ZERO. (Basketball = continuous)
Mr. Berri, you (and many others) are analyzing each tree with a microscope… and TOTALLY failing to see the forest.
Topic B: (Example… Bruce Bowen)
According to your stat analysis… Bruce Bowen isn’t fit to play on my playground. But Mr. Bowen has been a STARTER on several NBA championship teams.
What does Gregg Popovich know that you and your fellow statheads don’t know?
A LOT.
Although nearly everyone acknowledges Mr. Bowen as an excellent defender, Mr. Popovich knows that you can’t measure individual defensive ability in basketball through stats even though position-to-position-matchups-during-minutes-played is now available. He knows that although basketball defense is primarily man-to-man… ‘help defense’ and ‘switching’ is paramount to success (team activities).
This is because the playing of basketball is a continuous activity.
On the other hand, the playing of baseball is a discrete activity. You can measure defensive ability there because the players (when on the field/defense) are playing a ZONE defense with defined areas of responsibility awaiting a discrete activity (the hitting of the baseball by the batter).
In discrete activities… the total = sum of discrete activities.
In continuous activities… the total can be > sum of activities due to synergies between the various activities.
Mr. Popovich assigns roles to specific members of his team so that the (qualitative) TOTAL is GREATER than the (qualitative) SUM OF THE PARTS.
He knows that he is not playing FIVE TIM DUNCANS against FIVE STEVE NASHES. He is playing Duncan-Parker-Ginobli-Bowen-etc. against Nash-Marion-Stoudamire-etc.
He is not playing Mr. Bruce Bowen to win an MVP award. He is playing him as a vital cog with defined roles to create synergies with other vital cogs (Duncan, etc.) to give his TEAM the best chance to WIN.
And the proof is in the pudding.
JChan // November 14, 2007 at 10:21 am
As a Jazz fan, even seeing the numbers and knowing that Chris Paul is better, I’m still pretty happy about having Deron on the Jazz. Deron is getting better and better. If he could cut down on his turnovers, he’d be fantastic.
Also we have to look at the small sample size. If we did this comparison after 7 games, since game 8 was a blowout win over Sacramento where Deron didn’t play great and didn’t shoot much, his WP48 would have been over .200. If he can possibly stay over .200 for the year, that’s a guy that can lead a championship team.
And that’s what it boils down to. We can compare this player to that one, but as a fan, I just want to know if the players my favorite team has can win a championship. There’s no chance the Jazz can get Chris Paul on the team. But they have Deron Williams, and hopefully, that gives the Jazz a chance to compete for the title.
Bob H. // November 14, 2007 at 11:15 am
Tom Randini, you need to read the book. They look statistically at diminishing returns, etc.
JChan // November 14, 2007 at 11:19 am
Also, concerning the durability issue, I think a team would prefer to have their starting point guard on the court 35 minutes every game, rather than 40 minutes for 60 games. The quality of the backups behind both Deron and Chris Paul are not going to be winning many games when they have to start, but they can be a decent 10-12 minute stopgap in one particular game.
Also, to Mr. T.G. Randini, you make some very good and very valid points. One key point you miss, however, is that Mr. Berri and other statheads do not consider their work to be the only tool in evaluating players. He, and I, and most people that enjoy these new stats, consider them a starting place for evaluation, not to replace scouting and great coaching, but to supplement them.
Sure, Chuck Daly understood the best way to put together a basketball team. But some of these high-paid executives and coaches don’t seem to. This leads to things like the Heat not realizing how much they are going to miss James Posey. And the Knicks thinking Zach Randolph is the solution to their problems. These stats allow us to gain some insight into what the few truly great basketball minds already know. And they provide a starting point for making better decisions.
You should stick around and discuss this more. It’s not often that we see intelligent opposing views at this blog.
JChan // November 14, 2007 at 11:20 am
One other thing, Mr. Randini, if I may….
You don’t need to YELL to get your point across.
Thanks.
Ryan Schwan // November 14, 2007 at 11:33 am
Yeah, my whole minutes argument was pretty lame, wasn’t it JChan.
Anyways, it still remains that the durability “issue” isn’t an issue yet with me. Paul missed 17 games last season because of a terrible high ankle sprain. High ankle sprains are temporary injuries that lots of players get, and it’s not something caused due to being “injury prone.” If he had needed ankle surgery or had knee or back issues, then I’d be concerned.
Jason // November 14, 2007 at 11:43 am
Mr. Randini, why are you YELLING so much? It’s been my experience that the intended effect of allcaps, putting emphasis on points, does not produce the desired consequence of making the point look more important. It merely makes it look like you’re more intent on forcing your opinion on others than engaging in any discussion. It is especially distressing when the points you yell about aren’t particularly novel; they’ve been expressed here before in both more civil and more eloquent ways by others.
Bob H. // November 14, 2007 at 11:53 am
TJ RANDINI, No need to capitalize “team game” and “five joe dumars” every time you write one of those phrases.
dberri // November 14, 2007 at 12:24 pm
TJ,
Just a quick check on your story.
Rebounds per minute for Joe Dumars with Rodman on his team: 0.065.
Without Rodman: 0.059
Dumars was also slightly more productive (in terms of WP48) with Rodman on his team.
Yes, diminishing returns do exist in the NBA. But they are often overstated in importance.
Guy // November 14, 2007 at 1:15 pm
“Yes, diminishing returns do exist in the NBA. But they are often overstated in importance.”
Dave: It’s true we can’t know for sure what the diminishing returns are for Dumars, or any other specific player (w/o using something like plus/minus). But as you know, we can certainly estimate it in general. For example:
* the correlation between RB/G for a team’s top rebounder and RB/G for the other players on that team is -.76. Clearly, many of the RBs by the top rebounders are just taken from other players. (As TG notes, we don’t see this in baseball.)
* the SD for RB/48min at the player level is around 3.8, but at the team level is just 1.4. It’s actually lower at the team level! (If each player’s rebound opportunities were independent of his teammates’, then the team SD would be about 8.5, or 6 times as big.)
These results tell us there is a huge negative correlation between a players’ rebound totals and that of his teammates. It would take some work to figure out how much each player rebound increases the team total, but it appears to be somewhere between .1 and .4. It can’t possibly be anywhere close to 1, which is what WP basically assumes (beyond a simple adjustment for guards having fewer RB opportunities than non-guards). I’d suggest that figuring out the correct weight for rebounds should be an important priority for WOWII.
Pete // November 14, 2007 at 1:24 pm
“the SD for RB/48min at the player level is around 3.8, but at the team level is just 1.4. It’s actually lower at the team level! (If each player’s rebound opportunities were independent of his teammates’, then the team SD would be about 8.5, or 6 times as big.)”
Guy Washington, Very creative analysis. I like it.
Joey C // November 14, 2007 at 1:59 pm
Guy, So you’d like the wins produced value of rebounds to at least be reduced in half?
Joey C // November 14, 2007 at 2:02 pm
Guy W., your SD analysis looks off. (and Pete you were too quick to praise it.) Don’t the TEAM statistics implicitly normalize the minutes whereas the variability of the player level will be greater b/c of uneven minutes distribution?
dberri // November 14, 2007 at 2:32 pm
Guy,
Again, as far as I know, I am the only one to estimate the level of diminishing returns. Your faith in plus-minus is nice, but so far it is just faith.
By the way, why are you not as obsessed with diminishing returns with respect to scoring?
And is it just me, or are we having exactly the same conversations every few days. Someone should just record these and I will just put it in a single post.
Guy // November 14, 2007 at 2:42 pm
Joey: I was using Reb/48 minutes. But it really doesn’t matter. If the difference between a very good rebounding team and a bad rebounding team is just 2.4 rebounds/game (adjusting for opportunities), then I guarantee you that Tyson Chandler is not adding 7 net rebounds for his team each game (as Wins Produced tells us).
Dave: You say “By the way, why are you not as obsessed with diminishing returns with respect to scoring?” (as you invariably do when this issue is raised). And my answer is: I agree it’s important with respect to scoring. However, while Wins Produced corrects for diminishing returns on scoring — by subtracting FGAs — it doesn’t do this on rebounds. So the question back to you is: why are opportunities consumed important when evaluating scorers, but not rebounders?
Here’s a useful thought experiment on the question of valuing rebounds vs. scoring. Suppose that the NBA had decided years ago to have scorers credit a “missed rebound” (MRB) for each rebound by an opposing team, defined as the player judged to have had the best opportunity to get the ball on the non-rebounding team. So “Rebound opportunities” (RBO) = RBs + MRBs, and for each player we knew their RB% (RB/RBO) which averages 50% (c. 70% on defense, 30% on offense). However, this NBA decided it was not important to record FGAs.
In this parallel universe, the WOW authors would run their regressions and discover this Win Score formula: Points + Rebounds – Rebound Opportunities + Steals + .5*Assists + .5*Blocks – TO – .5* FTAs – .5*PF. (That is, RBOs replaces FGAs.) The authors got a lot of attention when they identified several “rebound hogs” who had impressive raw rebound totals, yet actually rebounded at less than a 50% rate and so were not helping their team. Top scorers, however, are rated very highly because there is no adjustment for scoring opportunities consumed.
Sound familiar? This if of course simply the WOW story in reverse. Everything Dave says about why it’s a mistake to just assume players with high point totals are great players applies exactly to how rebounds are handled in Wins Produced. It assumes that rebound opportunities are distributed randomly (except big men have more than guards), so that every rebound above average represents a net additional rebound to the team. But that’s no more plausible for rebounds than for scoring. We don’t know what the correlation is between RB and RBOs, because RBOs aren’t recorded (at the player level), but the fact that teams differ very little in their overall rebound % tells us this correlation has to be very high. Points/minute has a .83 correlation with FGAs, and the RB/RBO correlation is likely as high or higher. And that means WP should value rebounds at something like .2 or .3, rather than 1.0.
P.S. My analysis literally has nothing to do with plus-minus, so I don’t know what “faith” you are referring to. Diminishing returns can be demonstrated by comparing player and team variance, with no need for plus-minus.
T.G. Randini // November 14, 2007 at 2:52 pm
Jeez, Louise… substitute italics for CAPS. I’m writing this on the fly and the program doesn’t allow italics. Mrs. Ackerman, my third grade English teacher, would be proud of all you guys who are micro-managing punctuation instead of looking at the big picture… or shall I say, gazing at the moss surrounding the trees, instead of looking at the forest. But anyway, you all get extra credit and a brownie from the teacher.
(And so now I am whispering in parentheses… especially to Jason who thinks my points are not novel. No, they aren’t novel. In fact, some of my points are incorporated into the black-box algorithms Mr. Dean Oliver had devised. Mr. Oliver, as I’m sure all of you are aware of, wrote the finest book of basketball analysis ever… and now he doesn’t publish or blog anymore. That’s because one of the NBA teams realized he’s the finest analyst and went out and HIRED him. Whoops. I got carried away there and forgot I was whispering. I meant (hired him.) Sorta like the Red Sox did with Mr. Bill James. That sort of worked out well. Even if Mrs. Ackerman didn’t like his punctuation.
Kent // November 14, 2007 at 2:57 pm
“And is it just me, or are we having exactly the same conversations every few days.”
True. We just need Harold A to pop in with a weird metaphor about how rebounding is a team effort because of boxing out.
T.G. Randini // November 14, 2007 at 3:07 pm
Mr. dberri,
I’m thinking and writing on the fly here so forgive my punctuation… but vis a vis Econ and avg. cost/marginal cost… can’t we think in terms of avg. utility v. marginal utility for high volume shooters? Just brainstorming… but if say for example each team throws up 80 shots per game (and now over-simplifying)… half or 40 shots are always higher percentage shots and the other half are lower percentage shots by necessity… (meaning half the shots are contested much more highly by the defense than the easier shots)… aren’t the high volume shooters being punished for risking the lower percentage shots because someone has to take them? And aren’t the higher volume shooters the ‘high volume shooters’ because they are the best shooter on the team at making the lower percentage shots?
Thus, I think, there should be a weighting that understands this marginal utility and gives credit for it instead of merely depressing the average utility for high volume shooter.
There.
No caps.
T.G. Randini // November 14, 2007 at 3:09 pm
for high volume shooters (plural)
dberri // November 14, 2007 at 3:24 pm
TG,
As Marty noted in June of 2006, there isn’t much of a link between shot attempts and shooting efficiency.
But even if there were, the focus on shooting efficiency is a “why is the player productive” question. And that’s a good question.
The answer, though, starts with “how” productive is the player”. The “how” question, in turn, depends on the impact the stats have on wins. At least, that is the approach taken here.
To summarize, I think the better approach to analysis is to first figure out what the stats mean in terms of wins. This tells us how productive the player has been. Then we can look at all the factors (including usage rates if we can find conclusive evidence that these matter) that cause a player to get the numbers we see.
T.G. Randini // November 14, 2007 at 3:37 pm
To spell the marginal utility idea out further:
Let’s do some simple very rough calculations. 100 shots per team (to make the percentages easier… again, rough…) 50 high percentage shots (60% makeable on average.) 50 low percentage shots (30% makeable on average.)
Five starters play the entire game.
Let’s take just two starters and analyze a bit.
Starter A takes 10 shots. He converts 5. His efficiency is under par. Why? Because each starter has 10 high percentage shots he could take * five starters = the 50 high percentage shots for the collective team. And high percentage shots should be converted at 60%. But starter A only converted 50%.
Starter B (the shooter) takes 24 shots and converts 11. This doesn’t look so good at first glance.
But… par would be 60% of 10 plus 30% of 14 = 10.2.
And Starter B converted 11… so he is over par.
Starter B contributed more in efficiency than Starter A did in regards to scoring WHEN the distribution of difficulty is taken into account.
The key point is that not all shots are equal. Some are wide open and others are heavily contested. And Starter A always passes the ball when his shots are contested and lets Starter B do all the dirty work.
The stat blogs love Starter A. But the coach loves starter B.
And P.S. dberri… don’t mis-understand me. Your top-down, or macro oriented, approach is valuable as a first pass. I wouldn’t have finished your book or be posting on your site if this wasn’t the case. I just think it would be interesting to break down the elements on a more micro-level where the sum of the micro elements add to the team wins and losses, instead of the regressions looking at the more macro-type variables. (e.g. macro is shots… but micro takes into account the type, quantity, or difficulty of shots.)
T.G. Randini // November 14, 2007 at 3:49 pm
P.S. to studies linking amount of shot attempts to shooting efficiency:
The logic is faulty in these studies. The comparison is always between pools of players taking either high or low volumes of shots. This is meaningless. It is also meaningless to compare the same player on days when he takes a high volume of shots to days when he takes a lower of shots. There are way too many variables here… quality of opponent, quality of person guarding him, hot nite versue cold night, etc. etc. and macro regressions will not see through the smoke.
Let’s just do an Albert Einstein thought experiment here:
You are Michael Jordan or Allen Iverson for a month. Don’t you think your shooting percentage will rise if you only toss up high percentage shots? So instead of your normal 25 shots… you limit yourself to the best 15 or 16 opportunities? And you do this nite in and nite out… 15 or 16 higher quality shots. And you pass the ball off to Luc Longley or Eric Snow when you don’t have a high quality opportunity and let them deal with the low quality shot.
Don’t you think your shooting percentage will rise?
dberri // November 14, 2007 at 3:53 pm
T.G.,
Another way to think about shooting efficiency and shot attempts…
Every trip down the floor is an independent event. Your odds of hitting a shot on that trip depend on your ability. From this perspective, more shot attempts don’t impact efficiency because ability doesn’t change with more shot attempts. All that has changed is that you are taking more of your team’s shots.
In other words, a player doesn’t start the game saying I am going to only take 15 shots, so I better just take the really good ones. Or tonight I am going to take 25 shots, so I can take a few bad ones.
T.G. Randini // November 14, 2007 at 3:53 pm
But after a month goes by where Luc misses even more of them than you (you=Michael) would… (much more of them than you, Michael, would… Coach Phil Jackson calls a meeting with you in his office. And he says three words to you.
“Shoot, Michael, Shoot!”
T.G. Randini // November 14, 2007 at 4:05 pm
dberri,
“…ability doesn’t change with more shot attempts.”
Of course, I agree.
“Your odds of hitting a shot on that trip depend on your ability.”
Totally, totally, totally wrong. So wrong I can’t put enough totallys in front of it. You have completely missed the central point in my whole argument… a point that is crucial in actually attempting to understand the game of basketball.
Your odds of hitting a shot depend on your ability PLUS the incremental difficulty or ease of the particular shot.
This difficulty factor is determined by:
1. Quality of the opponent’s team defense.
2. Quality of the player matched up with you.
3. On the given play… how contested is the particular shot? Is it a transition layup? Is it a wide open look at a three? Or is it 2 seconds left on the clock and someone has to take the shot and you have to try to get it up even though you are double teamed?
Mr. dberri, you are missing HALF the equation and ALL the variables. Spend more time watching the games. Chart the difficulties and variations of shots.
As Walt Whitman might say… there are multitudes.
T.G. Randini // November 14, 2007 at 4:08 pm
In the land of the blind, the one-eyed man is king.
I depart this land, my sceptor in hand.
dberri // November 14, 2007 at 4:30 pm
TG,
Of course all that you cite matters. A player will shoot worse against the Spurs than he will against a less capable defensive team. But across a season, quality of opposition will somewhat even out (although not entirely). So I am not sure why this impacts our analysis of individual players across an entire season.
Joey C // November 14, 2007 at 4:40 pm
TG Randini is very supercilious in his style of argumentation. I find that of a lot of the Univ of Chicago MBAs I meet.
Jason // November 14, 2007 at 4:51 pm
(Well, at least he *claims* to be departing.)
T.G. Randini // November 14, 2007 at 4:53 pm
dberri,
I certainly don’t want to be a Jesuit priest that tries to ‘convert’ savages to Christianity… so I give up. I’m neither priest, nor a typical Christian.
I will leave just one more thought with you.
After watching about a thousand Jordan-led Bulls games (watching the games, not just gazing at box scores), and watching many finals games in-person… the Shot, the triple overtime, the Utah blowout, and many others… I can categorically say this:
when there was five seconds left on the shot clock and everyone was covered, whoever had the ball looked for Michael. As if the ball was a hot potato.
Michael gets it with two or three seconds left and has to create off a double or triple team.
That happened several times a game, each and every game, every week for six or eight months a year, for several years.
Give it to Michael when all hope is lost.
The point? Not all shots are created equal. Not all shots have the same difficulty level. And typically… the crunch-time shooters (like Bird, etc) have a steeper hill to climb regarding your Berrian efficiency.
It’s SO much easier to merely pass the ball to someone else, collect an assist and let HIM do the dirty work.
The same for rebounding.
It’s so much easier to catch the ball off a backboard than to throw it from a distance into a round hoop.
Hey… you don’t get penalized in the Berrian system for missing a rebound, do you? You don’t get penalized in the Berrian system for missing to see someone wide open and not getting an assist, do you?
But you get penalized for missing a shot.
If you can’t see the dichotomy here, well, you just can’t see it.
T.G. Randini // November 14, 2007 at 5:04 pm
And (seriously) thanks, Jason. The whispering parenthesis made me laugh.
And Joey, I generally can’t tolerate Chicago MBA’s either… but you’re the first to call me supercilious. Everyone else just says I’m always being facetious.
Owen // November 14, 2007 at 5:39 pm
T.G. – It sounds like you think you have invented the wheel here. But from what I can tell from your garbled transmissions, you seem to be expounding a version of the Difficulty Theory proposed, much more clearly, by Oliver in Basketball on Paper. You can find it on page 143 of that book. Berri and Co address it in pages 91-2 of the WOW.
Why not read what others have said already, then start over, with perhaps a little less pep?
Also, re “Hey… you don’t get penalized in the Berrian system for missing a rebound, do you? You don’t get penalized in the Berrian system for missing to see someone wide open and not getting an assist, do you?”
You do get penalized. A rebound you don’t capture is a point of winscore you don’t get credit for. You certainly get penalized relative to a player who does capture that rebound. An assist is .5 of a point you don’t get credit for. And if the opposing team secures the offensive rebound, then scores, that is a deduction as well to the defending team, which will also act as a penalty on the individual level.
T.G. Randini // November 14, 2007 at 6:03 pm
I gave credit to Mr. Oliver earlier on in these posts. In fact, I gave him quite a rave review.
We still discriminate against the shooter, though. If a passer gets an assist (good thing) when a shooter makes his shot (good thing), then why doesn’t the shooter get an ‘assist’ when he misses and there’s an offensive rebound?
There wouldn’t have been an offensive rebound (good thing) unless he missed his shot.
Haha.
T.G. Randini // November 14, 2007 at 6:05 pm
Regarding Oliver, this is what I had previously said:
“No, they aren’t novel. In fact, some of my points are incorporated into the black-box algorithms Mr. Dean Oliver had devised. Mr. Oliver, as I’m sure all of you are aware of, wrote the finest book of basketball analysis ever… and now he doesn’t publish or blog anymore. That’s because one of the NBA teams realized he’s the finest analyst and went out and HIRED him. “
Owen // November 14, 2007 at 6:18 pm
I don’t see difficulty theory there. But basically, you write as if Berri and Co had not responded to this approach, when in fact they have.
At this point, you are sort of spinning your wheels.
I would though, love to know what Oliver had to say about hiring Iverson.
Mike H // November 14, 2007 at 6:31 pm
And he wouldn’t have been able to shoot if he didn’t have the ball and so on…
We have fed you troll. Begone simple thing!
Joey C // November 14, 2007 at 6:56 pm
It seems there’s one of these people that pops up every day. Yesterday it was Jon Posner.
Pete // November 14, 2007 at 7:39 pm
The great TJ Randini called us “statheads.” Whoa!
Joey C // November 14, 2007 at 8:23 pm
Pete, I’ve been called worse.
Pete // November 14, 2007 at 8:45 pm
Me too. But never before by a University of Chicago MBA.
Pete // November 14, 2007 at 10:58 pm
Mike H, And he wouldn’t have made the shot if he wasn’t wearing sneakers. If DBerri ever watched a basketball game instead of just scouting the box scores, he would realize ALL the players wears sneakers.
Jason // November 14, 2007 at 11:22 pm
It’s gotta be the shoes.
Okapi // November 14, 2007 at 11:27 pm
Jason Kidd on the 7-0 Celtics:
“They are not about stats,” New Jersey’s Jason Kidd said. “They are just about wins and you can see that out there.”
http://sports.yahoo.com/nba/recap;_ylt=AnWaUQjtfHOremuL_R5JraA5nYcB?gid=2007111402&prov=ap
dberri // November 14, 2007 at 11:45 pm
Okapi,
Great quote. There is a column in there someplace.
Mike H // November 15, 2007 at 8:30 am
Well that just confirms my suspicion that the model is woefully inadequate. May I propose adding sneaker brand to the boxscore regression?
Danny N // November 15, 2007 at 8:45 am
I’m from Utah, I’m a believer in WinScore, and I even believe that Chris Paul is better than Deron Williams when Paul is playing. They are both excellent, of course. The Jazz picked Williams over Paul in part because because D-Will is bigger and allegedly more durable. Paul missed 18 games in 06-07; Williams sat out only 2. So their Wins Produced were comparable despite Paul being more productive per 48 min. Given some players’ susceptibility to injury, which Paul may have, shouldn’t that be considered when comparing players? So I love Chris Paul, but I’d still rather have Williams.
Harold Almonte // November 15, 2007 at 9:16 am
Kent-”weird metaphors”
I think that after these betters: Guy’s “metaphore” about the needs of a more fair way in measuring rebounding and other no scoring stats’s break evens, and Randini’s “metaphore” about how scoring volume-dificulty makes changes in the supposed similar break even (33%, leage AVG, or 50%) for all scoring usagers, and why it’s unfair a unique break even, and neccesarily to have different break evens for different usage levels on every basketball action;
all my former posts became obsolete. My respect to them.
Jon // November 15, 2007 at 11:51 am
dave, why did you never respond to Guy when he brings up the missed rebounds? are you and everyone else just going to ignore this or is a good explanation going to be given?
another question that i dont know the answer to. Has a study been done to see what is easier to replace, scoring, or rebounding? I would suggest doing this by looking at the rebounding rate of the pistons with wallace, then the rebounding rate of the pistons without wallace. then compare this to the offensive efficiency of the sixers with iverson compared to without. Maybe Wallaces 15 rebounds are easier to replace and thus not as valued… Maybe they are not easier to replace and are really worth their value. Anybody willing to do a study?
Jon // November 15, 2007 at 11:53 am
obviously, the replacement players (miller and mohammed) would have to be factored in. Don’t know exactly how to do it, but you guys are much smarter than me and im sure can figure it out
dustin // November 15, 2007 at 12:33 pm
Jon, Owen answered your question
“Also, re “Hey… you don’t get penalized in the Berrian system for missing a rebound, do you? You don’t get penalized in the Berrian system for missing to see someone wide open and not getting an assist, do you?”
You do get penalized. A rebound you don’t capture is a point of winscore you don’t get credit for. You certainly get penalized relative to a player who does capture that rebound. An assist is .5 of a point you don’t get credit for. And if the opposing team secures the offensive rebound, then scores, that is a deduction as well to the defending team, which will also act as a penalty on the individual level.
“
Harold Almonte // November 15, 2007 at 1:21 pm
Not getting a credit isn’t to be penalized. That’s adding from zero. It’s like suddenly we suggest FGMissed (or FGA) are not a lost of possession (whole or most probably partial and not individual) and doesn’t need to be penalized. Zero can’t be a break even, and sharing penalization among teammates can’t be a privilege for rebounding only.
It’s probably that the real stats break evens are league average efficiency at each different usage levels, and position adjust (not necessarilly position=usage level exactly) and team defense adjust (in this case penalization teamed) are not the best make up for this (why not team adjust everything?), but of course is the most accomplished way to do that actually. But still one action is penalized too much, and others even aren’t. The easy way is a traitor thing.
Joey C // November 15, 2007 at 2:34 pm
Almonte, you don’t lose points for passing up shots either. Isn’t that the same thing?
mrparker // November 15, 2007 at 2:34 pm
RE: MIssed rebound argument.
Are you guys kidding with this argument?
THe whole point of the missed shot argument can be underscored very easily.
Any one of us could walk out onto an NBA court and take a shot. However none of us could walk onto the court and grab a rebound.
Similaryly five players can simultaneously go for a rebound while 5 players cannot simultaneously shoot. Therefore there needs to be a much larger penalty for missed shots than rebounds. Being that this is an approximation and not exact science what is wrong with Mr Berri’s conclusions.
I keep waiting for all the naysayers to come up with a better approximation than Mr. Berri.
How many of you thought Philly got better after last year’s trade? How many of you thought San Antonio was the best team?
Jon // November 15, 2007 at 2:44 pm
first, regardless of whether there is a better metric out there, which im not going to argue, does not make this metric right. if you believe this metric is the better than other metrics, that is not a defense of using this metric. it would just be better than using a different metric. and with missed shots, being able to take 25 shots in a game is a skill that you would not be able to do on the court. and no, i don’t believe any of us could walk on to a court and shoot either.
Jon // November 15, 2007 at 2:45 pm
and owen did not answer the question, getting penalized means taking away, as field goals attempted takes away. not getting a rebound does not take away from win score, it just doesn’t add to it. that is a terrible answer if that is the answer you come up with
mrparker // November 15, 2007 at 3:01 pm
jon,
I didnt say you we could make a shot I said we could take. If Allen Iverson misses a shot thats something that anyone of us could do. But if any player misses a rebound that is not something that we could replace. Hence its not as damaging as a missed shot.
You still didnt address the fact that five people can go for a rebound but cannot simultaneously take a shot.
Owen // November 15, 2007 at 3:38 pm
Jon – I think of things relative to the average for the position. Someone who grabs one less rebound than the average guard is penalized for not doing so, on average.
Jon // November 15, 2007 at 4:01 pm
explain how they are penalized???? they are just not rewarded. there is a difference. if he were penalized, there, like there is with fg, he would be reducted for every rebound he misses. that is not the case.
and i know what you said, mr parker, and i dont believe you have the ability to miss as many shots as iverson. first of all you need to get the ball, then manage to get your shot off before it is stolen. something neither of you can do. and i am confused about the simultaneous thing. i dont’ get it. so if you could explain that would be helpful. Iverson shooting the ball as many times as he does is a skill. if many players attempted to shoot this much, they would be benched. it would not be allowed because they cant do it as well as he can
demcavs // November 15, 2007 at 4:13 pm
I completely agree with Jon. A guy like Iverson is talented enough to miss 60% of his shots. Tim Duncan, even at his absolute best, will only miss like 48-50% of his shots.
Jon // November 15, 2007 at 4:27 pm
i am not saying iverson is better than duncan. My point is, if many guards were in fact to attempt to do what iverson does, they would shoot much, much worse. And the fact is somebody on a team needs to shoot. A team has roughly 96 possessions a game on average. Iverson is able to shoot over 20 shots a game and still put up respectable averages. (No not good averages, but they certainly arent miserable.) By him shooting this much, he is allowing other players to stay in their comfort zone and not shoot as much. Obviously, on a team of all-stars, this is not necessary, because they can all get their shot off fine. But this is not the discussion anyways. This is way off topic.
The point is, that Guy pointed out, if a player is penalized for shooting the ball and missing, as he is in this formula, why is a player not penalized for not missing a rebound. That is the question.
Jon // November 15, 2007 at 4:28 pm
i mean for missing a rebound, my mistake in the 2nd to last sentence
The Ghost of TG // November 15, 2007 at 4:36 pm
If a guard does not grab a rebound, he is never punished with a debit. But he is punished in relation to other guards, if another guard grabs the ball.
But… if you are doing position adjustments and a guard doesn’t grab a rebound, he isn’t penalized in relation to other guards if a forward or center grabs the rebound.
Jon // November 15, 2007 at 4:43 pm
but couldnt you then say that when a player misses a shot, he is being double penzalized in this formula. One for missing the shot, which is shown in -fga, and then because about 70% of the time a player on the other team gets the rebound, driving up their score? Or is that wrong thinking?
Owen // November 15, 2007 at 4:50 pm
Jon – I think its just semantics. I don’t know. Willie Green averaged 4.1 rebounds last year. The average shooting guard averages 5.6. This means that, relative to the average shooting guard, Willie Green is deducted, penalized, whatever you want to call it, 1.5 units of WinScore.
Also, you aren’t actually penalized for a missed shot. The deduction is for all shot attempts, whether they go in or not. This is actually the primary difference between WP and NBA Efficiency and PER.
Jon // November 15, 2007 at 5:01 pm
Either I don’t know the formula for Win Score, or I am terribly confused. So Willie Green is being deducted for total shot attempts. Why is he not also deducted for total rebound attempts. And I understand how he is deducted 1.5 Win Score, but isnt his shooting deducted once by the total field goal attempts, and then a second time if he averaged less points than the average shooting guard. So unless I am mistaken about the formula, he is penalized twice in shooting, once in rebounding.
Correct me if I am wrong.
Harold Almonte // November 15, 2007 at 6:02 pm
Owen-How many offensive rebounds gives up a shooting guard attempting to rebound? and a forward?, How many boxing out and not attempting to?. How many times a player loose his man gambling a steal? How many fouls trying to block, and how many points not even trying to do that?
The fact is you need an efficiency level at every action (a rate), and a break even according to the usage level. The higher usagers (most skilled players) of any action (offense, defense, or rebounding), not only monopolize the good volume, but also the entropy (negative stats and dificulty) in favor of lowers. But once you have this properly rate (eFG%, RBr%, etc.) for every kind of players (not exactly positions), you can then worry about wether an action (rebounding) produces more wins than another (scoring).
But of course, if the method is rating against attempts, you need to know the possession value of all those non boxscored attempts, and that means boxscoring them and regressing to team wins later, or team adjusting by attempts and not by minutes. That’s not possible actually.
Harold Almonte // November 15, 2007 at 8:29 pm
The ideal thing would be a player action get rated against himself, teammates and opponents at the same time ( that is his attempts against teammates attempts-or diminishing/increasing return, and against opponent action- or miss), but even Rbr% rate against teammates by minutes.
And also these attempts possession value, or weight, should probably be obtained taking account the intrinsic probability of the team to loose the possession.
Then, we come back to the beginning: in a continuation play, what are the probabilities for the team to loose the possession? in a continuation play, are the possessions a team property, or an usager’s? Must rebounding share credits with a prior shot defense and be underrated a bit? Must be rated against the offensive rebounding 30% standard probability? Must a FGMissed + a probable OReb be the counterpart of the shot defense + a more probable DReb? Is this really a summatory of four possessions? three? or just one team possession changing or not?
Roger Goodell // November 15, 2007 at 9:36 pm
“I keep waiting for all the naysayers to come up with a better approximation than Mr. Berri.
How many of you thought Philly got better after last year’s trade? How many of you thought San Antonio was the best team?”
My god you people are dense. Philly’s defense improved when Iverson left, & their offense got worse. Replacing Iverson the chucker with Miller the efficient scorer did not cause them to improve on offense and go .500 the rest of the way, but Miller’s better D did cause them to improve (which WP doesn’t really address, unless you think steals/blocks/defreb mean good D).
Anybody who looked at point differential could predict San Antonio winning it all. Shoot, even Hollinger was touting them the whole time.
So stop bringing up examples that prove nothing except your own ignorance.
mrparker // November 15, 2007 at 9:54 pm
Harold,
You just don’t get it. The less posessions the other team has the better your defense is. Hence the more steals and defensive rebounds my team obtains the better our defense is.
I acutally have my own set of numbers for player production. But my numbers do not have the ability to predict wins only evaluate players. as compared to any average player.
It doesn’t matter how a player gets it done as long as they get it done. Who cares if philly’s offensive rating went down if their overall performance produced more wins?
Its funny that numbers that time and again point to what we see can be disputed. Its no coincidence that the players that win score has liked the most also have produced the most championships. Doesn’t that tell you to stop disputing the numbers and embrace them.
At least you proved I was ignorant
mrparker // November 15, 2007 at 9:55 pm
not harold…..roger goodell
Roger Goodell // November 15, 2007 at 10:23 pm
Ugh. Every team will basically have the same # of possessions as their opponents, it’s an alternating possession sport! So the statement “The less posessions the other team has the better your defense is” shows that you’re the one who doesn’t get it. Also, on defense, forcing missed FG (of which blocks are a very small subset) are far more important than steals, etc. WP doesn’t track forced misses (nobody does) b/c they aren’t in the boxscore.
“It doesn’t matter how a player gets it done as long as they get it done. Who cares if philly’s offensive rating went down if their overall performance produced more wins?”
WP maintains that a player like Iverson who takes many shots but doesn’t make a lot is detrimental to an offense, & a player like Miller who takes fewer shots but makes a higher % is more valuable to an offense. That’s the basis behind saying “Philly will get better when they replace Iverson w/ Miller”. He had no way of predicting their D would get better b/c WP doesn’t tell you the important things about a player’s defense. But under WP he would predict that their offense would improve w/o Iverson, b/c WP values efficiency and not creation.
But their offense didn’t get better. It got worse. Their D just happened to improve as well, which accounted for the winning. This matters because when you acknowledge these facts the AI trade can’t be used as evidence of WP. It didn’t correctly predict what it was supposed to, that Philly’s offense would get better w/o Iverson.
Pete // November 15, 2007 at 11:07 pm
Roger Goodell, this would be imperfect but why can’t defense be assigned to individual players based on how the opposing player at their position does? Or do too many teams use a zone defense such as to make this a bad approximation? Tks.
Roger Goodell // November 15, 2007 at 11:21 pm
Adding counterpart stats like 82games has would definitely improve the accuracy of any linear weights method, whether it is WP or PER or tendex-like. Switching defenders (ie, Bruce Bowen guarding a SG) happens, but something is better than nothing.
Still, even if you did that, it wouldn’t improve WP’s .94 r-squared. That just comes from the relationship between pt-differential and wins. But it would almost certainly increase the predictive accuracy going forward with players who change teams or see different roles.
dberri // November 16, 2007 at 12:00 am
Roger,
Obviously this is not the NFL commissioner. Odd choice for on-line pseudonyms. Does anyone use their real names on-line? Or is that somehow against the rules?
Okay, about what you are saying. The prediction that Philadelphia would improve without Iverson did not depend upon team defense (and yes, I wasted time tonight checking on this).
And I have to ask, how do you know combining plus-minus data with a linear weight measure would improve accuracy going forward? And if that were established, wouldn’t a better way to put it be that the box score data improves the accuracy of plus-minus? Plus-minus data is not a very good forecaster of future wins or even of itself.
One last note… try and discuss this without calling people names. We are not “dense” here and so far you have not demonstrated that you possess any special insights.
mrparker // November 16, 2007 at 8:32 am
Roger,
You’re one of those debate team guys who depend on semantics to make a point aren’t you. Don’t worry my wife does the same thing so you are in good company.
Substitute “chances to shoot” for “possessions “and my argument makes sense for those of us who make inferences and for those of us who can’t tell when someone is writing at 3am in the morning.
Win produced does not argue for offense or defense. It incorporates both. Ben Wallace in his hayday was not good for an offense but had one of the highest wins produced around. Contrarily Steve Nash who consistenly scores over .3 is more heralded for his offensive production than his defensive and certainly is not brought to any team to make the defense better.
mrparker // November 16, 2007 at 8:32 am
WP is not choosy. It likes both defensive and offensive production. If you have ever read Dean Oliver, you know that what Mr. Berri incorporates in WP is exactly what Oliver uses in his defensive rating. The formula is a little different but the results are the same. Its not surprising that the guys with the highest defensive rating are for the most part the guys who we all think are the best defensive players.
I don’t know what your specific beef with “us” over at the wages of wins is but it seems as if you are one of the guys that is of the opinion that all you need is the eye test.
mrparker // November 16, 2007 at 8:33 am
I used to think the same way. I used the eye test to rate players and when I became interested in making predictions it never worked. I was left thinking that everything must be governed by the chaos theory. Then I went to sites like footballoutsiders, Kenpom, and here and found that there is some sense of order after all. It left me able to make my own metrics that work best for what I am trying to do. However it also left me with a think tank of like minded people who found the light. I don’t worship the sites but I do believe that there is order and everything which can be visually represented. That can be numbers or charts or tables but it is what I look for when trying to make sense out of sports, economics, the stock market, etc. The eye test got me nowhere. Visual representation has allowed me to be writing from my own computer on my own time during what used to be work hours.
The proof is in the pudding.
I don’t mean any harm to you or anyone who I get into an argument with online. Its only that the numbers are pretty good at what they represent and I get a little defensive of them especially considering that I’ve had to go through two bad olympic cycles because the leaders of one of my favorite sports won’t look at them.
mrparker // November 16, 2007 at 8:38 am
ps 5am in the morning….ive got my timezones a little mixed up
Rasta // November 16, 2007 at 9:29 am
For me, the WoW Journal is like the Bible: I read it everyday; I take it’s message seriously, but I don’t take it literally.
The reason I don’t accept WP at face value is the same reason that so many others are critical: penalizing the shooter for all FG attempts, not just FGs missed.
(I take issue with WoW in a few other areas, but they are immaterial in comparison.)
In my spare time, I’ve been working on a model very similar to Wins Produced, except using a more conventional approach to FG attempts. The results are very encouraging.
We all know that Point Differential alone results in R-squared of 0.94 and estimates wins within a margin of error of 2.2 games. IMO, any model that claims to do better than that is fudging something.
My model, which was inspired by and loosely based on WoW, results in R-squared on 0.92 and a margin of error of 2.8 wins.
Again, the main difference between my model and WP gets back to FG attempts. My website and research will be up and running soon.
dustin // November 16, 2007 at 10:22 am
Made field goals are always worth at least 1 point of win score. Missed field goals are always worth -1. Seems to me the only penalty is for missed FGs.
Owen // November 16, 2007 at 12:47 pm
There are deductions for turnovers, pf’s, fg attempts, and ft attempts in Win Score.
dustin // November 16, 2007 at 1:22 pm
Owen, I know that fg attempts are subtracted from win score. The point I was trying (and aparently failing) to make is that a made field goal always results in net + to win score, and I don’t see how that can be called a “penalty.”
Jon // November 16, 2007 at 1:29 pm
can someone with good knowledge of win score answer this for me? I posted this already but did not get a response. I am probably wrong, but I am just wondering why.
“Either I don’t know the formula for Win Score, or I am terribly confused. So Willie Green is being deducted for total shot attempts. Why is he not also deducted for total rebound attempts. And I understand how he is “deducted” 1.5 Win Score for averaging 1.5 rebounds less than the average shooting guard, but isnt his shooting deducted once by the total field goal attempts, and then a second time if he averaged less points than the average shooting guard? So unless I am mistaken about the formula, he is penalized twice in shooting, once in rebounding.
Correct me if I am wrong.
Owen // November 16, 2007 at 1:47 pm
Dustin – You are correct that a made field goal will give you one or two points of winscore. I confess, I don’t exactly understand what your question is. In general I think the approach used is to look at how many points a player scores relative to his field goal and ft attempts, then compare that to the average for his position. There is no made field goal/missed field goal distinction. Other metrics do make that distinction. For instance, NBA Efficiency deducts for missed shots rather than for all shots taken, with the result that anyone who shoots over 33% will see a “return” on their shooting.
Jon – Willie Green scores more points than the average shooting guard. I am not sure if this answers your question, but my sense of the logic of WP is that if you are well below average in shooting efficiency as he is, you are probably penalized more for taking an above average amount of shots, relative to the average guard. So perhaps in a sense, he is penalized both for being high volume and low efficiency.
A high efficiency player like David Lee would perhaps be “penalized” in some sense, (don’t really know if the term is the right one) for scoring fewer points than the average power forward. However given that he is much more efficient than average, overall, his return from scoring is much higher than most PF’s.
Jon // November 16, 2007 at 1:58 pm
owen that doesnt answer my question. it doesn’t matter whether he scores more or less then the average. what if he scored less then the average? Then he would be penalized twice, (once for scoring less, once for every field goal attempt) compared to just once for rebounding (only penalized for rebounding less than average) can somebody else try (preferrably jason or berri)
Rasta // November 16, 2007 at 2:07 pm
Dustin,
Say a player makes 1 out of 2 FG attempts. Under Win Score, he’d be credited for the 2 points scored, then debited for the 2 FG attempts. The net WS of those two shots is 0.
Under virtually every other metric, the player would be credited for the 2 points scored, then debited 1 (or some fraction thereof) for the missed FG. The net of those two shots is 1.
These represent two opposite ends of the spectrum. The conventional metrics like Efficiency and PER reward poor shooters as long as they exceed the absurdly low threshold of .333 FG%. Win Score, on the other hand, penalizes any shooter who fails to hit .500 FG%. The ideal solution lies somewhere in between.
(Clearly, I’m only referring to 2-point FG attempts in the above paragraph.)
dberri // November 16, 2007 at 2:17 pm
Rasta,
A couple of questions to ponder as you think about building your own model.
Why does the “ideal” lie somewhere in between? What is your criteria for choosing the “ideal”?
Jon // November 16, 2007 at 3:04 pm
dave, can you clear win score up for me?
Harold Almonte // November 16, 2007 at 3:19 pm
I’m also a believer of the “somewhere in between” issue. I also sustain is something near to league avg. (46%) or so. There’s a big error in sustaining a FGMissed=1 point, because you are saying a team is recovering (off. rebounding) the other possible scored point in every missed (or recovering something near to 50% of misses-because some recovered will be made), and that’s not true. You can’t say you lost just the possession like a turnover, because you shot and the probability these 2 points be scored does exist (the attempt already matters, no?). What is more probably is both, the FGmissed and the OReb worth more than 1 point, and less than a whole 2.
And here I’m granting an alibi for off. rebounders.
Rasta // November 16, 2007 at 3:25 pm
Dave,
Here’s my thought process so far:
Look at Ben Gordon and Ben Wallace last year. Gordon took about 23 shots per 48 minutes; Wallace took about 7 shots per 48 minutes. Assume, for illustrative purposes, that the average player (regardless of position) took 15 shots per 48 minutes.
So, per 48 minutes, Gordon took +8 shots; Wallace took -8 shots. Obviously, this difference in shot attempts is a tactical decision by Skiles, for very sound reasons.
However, compared to the average SG, Gordon’s PPG (and therefore his conventional metric) is disproportionately high. Vice versa for Wallace verses the average C.
In Gordon’s case, we can adjust his metric downward by calculating how many points her scored (and how many shots he missed) in those +8 marginal shot attempts.
We can do the opposite for Wallace.
In theory, this adjustment is a transfer from one teammate to another, and is required to compensate for the uneven distribution of shot attempts (versus the positional average) tactically decided by the coach.
Player, Wins Produced, Modified WP
Deng 14.9 12.8
Gordon 4.5 5.1
Hinrich 9.9 12.5
Thomas 2.1 2.9
Duhon 4.9 8.3
Wallace 15.8 10.1
Nocioni 1.9 1.7
Griffin 1.7 1.9
Brown -0.7 0.3
Allen -1.5 0.5
Sefolosha 0.9 -0.0
Khryapa -0.2 -0.2
Sweetney 0.0 -1.6
Total 54.2 54.2
Westy // November 16, 2007 at 3:37 pm
mrparker said, “I keep waiting for all the naysayers to come up with a better approximation than Mr. Berri.”
I for one am pretty interested with what Rasta shows above. Is this an answer mrparker? I think Rasta’s justification for looking for something “in between” is indeed the fact that the NBA average FG% is below 50%. Rasta, do you adjust the rebound weights at all?
A few posts ago, I also noted a WP alternative that keeps the team totals the same. I noted, “DRs and ORs need to be split at the individual level.
A player’s valuation would include a factor of 0.6*DR plus 0.1*(other DRs while they’re on the floor). As well, each player would get 0.9*OR plus 0.1*(the shots they took that actually resulted in an OR). Missed shots would still be -1.0.” Those weights could be adjusted to best reflect actual team worth.
So I guess that’s another alternative.
Owen // November 16, 2007 at 3:54 pm
Westy – I think that you need to look past fg% here. Fg% doesnt account for the fact that 3 pointers are more valuable than 2 pters. And it doesnt account for the added efficiency from free throws.
I am not sure if it would help Rasta to add fts to his analysis. But it might help people who are puzzled about WS and the “50% break even.”
I do find it helpful to think about TS%. The league average ts%, which incorporates fg’s and fts, was around 53-4% I think.
Also, another comment. Win Score is a simplified version of Wins Produced. DB has posted very helpful technical notes on WP for those looking for more info…
dberri // November 16, 2007 at 4:00 pm
Rasta,
My sense is that when you are done, your rankings will have a 0.95 correlation with Wins Produced. For most players, you will be telling the same story. I don’t see how this approach improves predictive ability. It certainly makes things more complicated (which is not a positive). And the theoretical foundations behind what you do is not entirely clear to me.
Still, it is good to see people thinking about this stuff.
Jon,
I have no idea what a total rebound attempt is. So that prevents me from dealing with what you are saying.
Harold Almonte // November 16, 2007 at 4:31 pm
Rasta-Can you better explain (with real stats and eff.) how did you do that scoring diminishing/increasing return adjust?
Jon // November 16, 2007 at 4:32 pm
a total rebound attempt would be any rebound that your team does not get while on the floor. maybe more specifically, it could be done by position. For example, for jason kidd, his total rebounds would be subtracted by how many rebounds the other teams point guards get. i dont know if this is a good idea or not.
Owen // November 16, 2007 at 5:36 pm
Jon – It seems to me that a missed rebound attempt is basically the same thing as an offensive rebound, correct? It seems like you are saying that there should be a deduction for the other teams offensive rebounds.
There isn’t in Win Score. But in Wins Produced, the complete metric, there is I would say, at least from my perspective. When a team grabs an offensive rebound it has another opportunity to score. If the team that allowed that rebound then allows a bucket, the defending team is deducted for a made opponent fg. This debit, over a season is divided equally amongst players based on minutes played. They are penalized for allowing that rebound, to the extent that points are scored off them.
So, in a sense, a missed rebound attempt can end up in a deduction.
Hope that is both correct and helpful…
Harold Almonte // November 16, 2007 at 6:03 pm
The in between issue:
Making FGmissed=-1 point or poss., gives us a 2p 33% break even. making FGmised=-2 points, or penalizing every attempt is the same, gives us a 50% break even.
About a 25% of FGmissed are ofensive rebounded (the possession is not lost), and about half those offensive rebounded will be missed and defensive rebounded by the opp., then we have about a 12% of FGmissed recovered by the team (I’m talking about team, because a team regression do the same). That gives us a FGmissed and OR weight = about 1.75 or so. That gives us a 47%-48% break even (that’s very close the 2p league average FG%).
If we accept, and we are not obligued to, the relationship between rebounds 2/3-1/3, and weight them according to this dificulty and probability factor, then a DReb=0.58, not 1.
What could this mean? a probably Dreb overrating because regressing before the team adjust, and taking some of the FGmissed defense weight?
PEte // November 16, 2007 at 6:10 pm
Harold A, are offensive and defensive rebounds weighted the same in wins produced?
Rasta // November 16, 2007 at 6:14 pm
Owen,
My model does include FTs and 3-pointers. I omitted them from my earlier post in order to keep the example simple.
Harold,
I will be glad explain the adjustment with a detailed post. Probably not until Monday though.
Dave,
I would guess that correlation between Wins Produced and my modified version (which I’m referring to as “sCore”) will be lower than 95%. Probably below 90%. Some players, like Gordon, will be very close. Others, particularly heavy rebounders, will be more variable. At the team level, the correlation between models will be close to 99%.
Anyway, thanks for all your work on this site, which has become one of the best places for this type of stat discussions. Don’t let the critics get you down.
Harold Almonte // November 16, 2007 at 6:25 pm
I don’t know the WP formula and how much varies from WS, just that WP defenders sustain all R= 1 poss = 1 point. Even, any answer Berri gives to me about his regressing procedures, I wouldn’t be skilled enough to refute it, but maybe a real statician would.
But, don’t be nervous, rebounders’s worth are coming back again, but those who are good off. rebounders.
Rasta // November 16, 2007 at 6:27 pm
Westy,
I’ve treated all rebounds equally. Meaning that an offensive rebound=defensive rebound. And that the rebounder gets 100% of the credit.
Logically, it makes sense to share the credit in some manner, but I don’t see how to do so given our current boxscore limitations.
Drew // November 16, 2007 at 6:37 pm
Just one important thing to note about the All Star game this year with regard to DW or Chris Paul getting selected: the game’s in New Orleans.
Harold Almonte // November 16, 2007 at 6:45 pm
I think everybody forgot the main story. CP and DW are both point-guards. CP a better point, DW a better guard. And everybody is pleased.
Christopher // November 16, 2007 at 9:38 pm
“dave, why did you never respond to Guy when he brings up the missed rebounds?”
Because “rebounds attempted” is nonsense? Shall we add steals attempted and blocks attempted? How do you track this? What does it mean? For a FG this is all very easy.
dberri // November 16, 2007 at 10:28 pm
Christopher,
I can’t respond to everyone (although I try). In this instance, though, I thought the story pointless. We don’t have data on rebounds attempted. I am not even sure what that means. Given that, I can’t see where a discussion of this leads.
Jon // November 17, 2007 at 12:21 am
my last word and then i am not bringing it up again. it is easily tracked. it is whatever amount the other teams have. and just because it isnt “easy” is not good for a metric trying to provide accuracy. so how are you not even sure what it means
mrparker // November 17, 2007 at 7:02 am
Jon,
What exactly do you mean by attempted rebound?
Number of rebound opportunities while on the floor?
Jon // November 17, 2007 at 1:02 pm
it would simply be total rebounds. yes
Harold Almonte // November 17, 2007 at 1:25 pm
PETE-I actually could see how WP was build and his adjustments, and yes DR and OR have de same marginal value. And like I suspected, the problems begin in what is called PA (possession acquired), thare are a lot of ways team and players acquire possession, and I could see the only responsibles for acquiring from FGmisses are DRebs. Probably scorers shot all his attempts uncontested.
DRebounders are taking credits from teammates, and this must be fixed or adjusted before the regression.
mrparker // November 17, 2007 at 5:12 pm
Last thought then I will leave the subject alone.
It seems as if many posters are trying to perfect DJ’s system.
Why not just KISS it. When it comes to making predictions and you find something this good messing with it(for lack of a better term) has proven time and again to produce little extra value.
His evaluation may wrongly “dislike” some our favorite players but there is where the eye test can back up his findings.
Bruce Bowen may be a great man defender but if Tim Duncan isn’t grabbing all of his man’s misses then he becomes of less use.
Thats just an example but is a case of a great player making their teammates better. Steve Kerr was a great player for the bulls because he had to do little else besides what he does best(shoot 3 pointers). Everything other aspect of the game was covered.
WP may not like their value as producing wins is concerned but as evidenced by drating or orating these players add more to their teams than most other players could.
mrparker // November 17, 2007 at 5:13 pm
should have spell checked
Westy // November 19, 2007 at 12:24 pm
Good synopsis, mrparker. Our eyes still add value.
And Rasta, I do think there is a fairly easy way to split rebounds with only the box score. Granted, it would be an approximation, but in my opinion would give more accurate player valuations. Now Dave might argue this is making the formula more complex, but in the age of Excel, it’s really not more work once you create the formulas if you already have the ability to turn a box score into WP. Here is how it would go:
Remember, a player’s valuation would include a factor of somewhere around 0.6*DR plus 0.1*(other DRs while they’re on the floor). As well, each player could get 0.9*OR plus 0.1*(the shots they took that resulted in an OR). Missed shots would still be -1.0.
These can be approximated if you assume the pace of the game (and thus the rate at which rebounds occur) is somewhat constant. Thus, to determine the number of defensive rebounds that occurred while a player was on the floor that he didn’t get, first determine how many d. rebounds/minute the team as a whole got. Then, multiply out how many d. rebounds occurred while the player was on the floor using his minutes played. Finally, subtract the rebounds he himself acquired (which he’s getting 0.6 for) and you have the number of other rebounds that occurred while he was out there contributing defense and the like (and he’ll get 0.1 for these).
On the offensive end, you know how many shot attempts the team had. You also know how many offensive rebounds they got. Thus, you can determine the rate at which the team’s shots were offensively rebounded. Then, using that rate, distribute that 0.1 (or whatever weight it shakes out to be) to all the shot attempts. So if a team shoots ten times, and offensively rebounds it 4 times, there’s 0.4 points to distribute to shot attempts. Thus, each of the ten shot attempts is worth 0.04 positive points using these weightings.
Thus, using only the box score statistics, you can easily add the ability to split credit for rebounds and still produce the same team total win score.
Jason // November 19, 2007 at 12:56 pm
The issue isn’t just whether or not the formula is more complex, it’s whether it is more useful. Your division of a rebound and credit for a missed shot makes an assumption that this division will more accurately measure the individual contributions of players. It may be true; it will not change the team score, but if it doesn’t explain something else better, then the move seems more one made out of a sense of equity rather than one derived from necessity.
What would it need to explain to justify such a change? A change that could modify player ratings would have to better explain the individual contributions of a player such that if the player was removed from the team and placed on another team, his performance *and the performance of the new team* would be more accurately gauged based on the individual’s rating.
While it may appear more fair, it is an empirical question as to whether or not it makes the formula more *accurate* in terms of its predictive power. If you are rewarding players for rebounds they did not gather or giving less than full credit for a rebound that someone did pull down, you have to justify that making this adjustment will better explain future returns.
It still seems like the arguments against weighting a rebound as a rebound are philosophical rather than empirical.
Jon // November 19, 2007 at 1:00 pm
Jason, I agree completely with what you are saying. But one way you could look to figure this out is to see how well Detroit rebounded with and without Ben Wallace. If Ben Wallace really did contribute all those rebounds himself, there would be a large change in the teams rebound rate. If it was not just him, the rebound rate for the team would be roughly the same. Obviously the test would have to include who replaced Wallace’s minutes and how well they rebounded. Wouldn’t this give us an idea (not a definite answer because its only player) of how much credit a rebounder deserves?
Jon // November 19, 2007 at 1:04 pm
its only 1 player i mean
dberri // November 19, 2007 at 1:06 pm
I really need to stop leaving comments. But I am going to chime in here because it might save us all the trouble of going around and around the same issue.
The issue is not just rebounds, but everything your teammates do. When they take a shot, you don’t take the shot. When they get a steal, you didn’t get the steal.
Given that is everything, the way to get at this is to look at the link between your productivity (all that you do) and your teammates productivity (all that they do). And this is what we did in The Wages of Wins. The impact, as I have said before, is statistically significant. But it is rather small.
Jon // November 19, 2007 at 1:16 pm
Just for clarification, I don’t know this stuff well. Are you saying that either everything should be divided between teammates or nothing should? Like the reason the rebound is not divided is because a point is not divided or an assist is not divided?
But would that mean if it was possible to figure out how much credit each person deserves for each box score statistic (which so far no one has been able), the system would be better? I am just curious
Jason // November 19, 2007 at 2:05 pm
Jon, looking at a single instance like Wallace’s departure is anecdotal. It can suggest something to look at further, but on its own it cannot make or break the model. A statistical model is drawn from and evaluated with statistical observations. It’s not just a matter of how Detroit did but how in general teams do when new players move in and out. It is not possible to evaluate this based on a single instance and conclude that it is correct or not based on the single instance.
Cases that appear to fall out of the model can suggest other models or modifications to a model to pit models against each other, but saying that another model is better based on anecdote is particularly bad science.
Jon // November 19, 2007 at 3:13 pm
thats why i said it would give an idea (not definite because it is 1 situation). But if somebody was willing to do it, they could do this for many different players and if it was noticed one way or the other, while it would not offer a solution, it would just let you know something is wrong with the model. I am not saying this is a way to come up with another model, I am saying this is a way to let people know whether there is a flaw in that area and improvement is needed. I am not saying it will work one way or the other. I have not done the test on Wallace, nor have I done it for any other player. But if a trend occurred after this test was done a number of players, I believe that would be a good indicator of whether or not other people should get credit.
Jon // November 19, 2007 at 3:20 pm
All I am saying is recently on this website there has been an argument about whether or not a rebound should be distributed. Everybody has their own opinion, and no matter how good a point anyone makes, nobody is changing there mind. I believe if this test was done enough to transactions that teams made, it would provide an answer to that question. If, in fact, the people who claim the rebound should be distributed were right, this would not offer the solution to how it should be distributed. (Although maybe it would, like when a player leaves, a team is able to replace X amount of his rebounding). I believe the test would just say, yes, it should be distributed or no, it should not be distributed. This could help end the seemingly endless argument about rebounds on this website and than maybe people could move on to more productive conversations
Jon // November 19, 2007 at 5:28 pm
“but saying that another model is better based on anecdote is particularly bad science.”
Also, not once have I mentioned any other models. I have never said any model is better. This is solely a discussion of WoW and specifically rebounding. I am not on a side and am just offering a solution to the ongoing argument about the distribution of a rebound.
dberri // November 19, 2007 at 5:59 pm
Jon,
I noted the following above. I am not clear why this doesn’t answer the question.
The issue is not just rebounds, but everything your teammates do. When they take a shot, you don’t take the shot. When they get a steal, you didn’t get the steal.
Given that is everything, the way to get at this is to look at the link between your productivity (all that you do) and your teammates productivity (all that they do). And this is what we did in The Wages of Wins. The impact, as I have said before, is statistically significant. But it is rather small.
Jon // November 19, 2007 at 6:47 pm
I was just responding to Jason, not what you said. I asked a question about what you said before because I am a little confused by what you were saying.
“Just for clarification, I don’t know this stuff well. Are you saying that either everything should be divided between teammates or nothing should? Like the reason the rebound is not divided is because a point is not divided or an assist is not divided? As in, teammates contribue to every statistic, not just rebounds, so it is pretty fair to just assign everybody the statistics they accumulate?
But would that mean if it was possible to figure out how much credit each person deserves for each box score statistic (which so far no one has been able), the system would be better? I am just curious
dberri // November 19, 2007 at 7:14 pm
Jon,
What I looked at was the relationship between your teammates wins, which is derived from the box score, and your productivity (also derived from the box score). Again, there is a relationship. The more productive your teammates, the less productive you will be. But it is not a huge effect.
In essence, if we simply say “you get credit for any stat assigned to you” we are not going to be too far off the mark.
Jon // November 19, 2007 at 7:30 pm
ok, thanks
dberri // November 19, 2007 at 8:04 pm
Jon,
Thanks for the comments. I appreciate your interest and your comments force me to try and explain things better.
Michael Whittenberg // December 3, 2007 at 12:04 pm
I like Williams better than Paul, and really do Williams will be the better PG, especially playing for Jerry Sloan.
http://www.bleacherreport.com/articles/4332-NBA-Deron_Williams_and_Chris_Paul_The_Future_Has_Arrived-011207
Chris Paul, Deron Williams, and the Surprising Hornets? « The Wages of Wins Journal // January 30, 2008 at 1:06 am
[...] Chris Paul vs. Deron Williams, Again [...]