Wages of Wins has gotten around to its Washington Wizards summary and reaches an unorthodox conclusion. The 'Zards, in Dave Berry's view, had a strong supporting cast ("Only Boston, Dallas, Indiana, Miami, and San Antonio managed to have this many above average performers") but Agent Zero himself isn't especially good ("Of last year’s playoff teams, only the Sacramento Kings and Milwaukee Bucks were led in Wins Produced by a less productive player"). This sort of seems like one of these conclusions one could reach if and only if one had never actually . . . watched the team play or seen the Wizards offense go to shit without the professor of Gilbertology on the floor.
« Strategic Restraint | Main | Iraq as Humanitarianism »
The Unbearable Averageness of Gilbertology?
02 Oct 2006 10:22 am
Comments (21)
This is the kind of statistical analysis that drives me nuts. Here is a talented sport economist, a guy with a reputation as "stats genius" who writes an article that fails to note that, while the Wiz were average, they were average by being a very good offensive team (7 best offense in the league) and a very poor defensive team (23 best defense in the league). Then he is mislead by his ratings system (which overrates rebounders, for example) into a conclusion that clearly defies conventional wisdom and doesn't even tries to explain it. He doesn't break down his numbers to try to show you why everybody is mistaken in his evaluation of Arenas, he doesn't make a case or tries to understand what's going on. His method is the absolute truth and that's it.
I don't think the conclusion is in any way at odds with the offense "going to shit" without him. He's still their best player, but he's just the third worst best player from one of last year's playoff teams. (Also, if you put Ron Artest on the Kings for the full season, Arenas would probably move up to second worst best player.) This doesn't seem hard to believe at all.
Without a bit of statistical analysis, let me take a guess at his competition for worst best player on a playoff team: Duncan, Bibby, Nowitzki, Gasol, Anthony, Brand, Marion, Bryant, Wallace, Redd, James, Jefferson (or Carter), the Pacers' O'Neal, Wade, [insert some Bull here].
Looking at that list, I observe (by actually watching the games, per your standard) Arenas to be better than only Bibby, Redd, and whoever that Bull is.
So perhaps he was judged one spot too harshly. The fact that he was the best player the Wizards had is no indication that he is a good best player to have. A team full of slightly better than average players isn't going to be very good without its best player on the floor, not when other teams have one or two guys who are dramatically better than average and other players who are slightly above and slightly below. Against playoff-caliber competition, they're going to keep struggling.
Oh, and by the way Arenas TS% is the highest on the Wizards (58.1) and nobody else comes close.
While not convinced that WoW's WP48 score is that great, I'm not a fan of Matthew's debunking either. The on-court/off-court numbers can be awfully misleading too. After all, we don't know who else was on-the-court and off-the-court at the same time Gilbert was. The on-court/off-court numbers suffer from an unstated ceteris paribus assumption, and there is no reason to make that assumption.
Now, it's true that Arenas isn't in "superstar" (Duncan, Wade, Nowitzki, Lebron) level yet, but how can you write an article about the Wizards and not note that improving their defense to average would do wonders for them?
jhupp - the problem with your analysis is that a team filled with slightly above average players should have much less of a falling off when a player goes out (even the best player) than a team with a few great players and the rest below-average players. The reason is simply that, in the former case, the drop-off between the player leaving and the player entering should not be that great, whereas in the latter case the dropoff is likely to be very great.
Now, of course, Matthew hasn't given us the on-court/off-court numbers for other teams's best players. For all I know, the differentials could be even greater than the differential Matthew points out for Gilbert. (I knid of doubt it, though.)
BTW - anyone else notice that, according to WoW, Jared Jefferies was the 'Zards' third-best player last year? Ahead of, say, Jamison? Hmmm....
jhupp:
He doesn't just say that, of the best players on playoff teams, Gilbert was the third least valuable ; he says (or at least strongly suggests) that Gilbert is worse than at least 36 other players in the league. Don't buy that, and don't think many GMs would, either.
the problem with your analysis is that a team filled with slightly above average players should have much less of a falling off when a player goes out (even the best player) than a team with a few great players and the rest below-average players.
I'm not sure that's right. The WoW guys are specifically concerned with wins produced. You would, one assumes, do best to look at records over injured-player periods rather than offensive or defensive numbers. But I can't quite articulate why that seems right.
Oh, and by the way Arenas TS% is the highest on the Wizards (58.1) and nobody else comes close.
Right -- I think this is the relevant thing. It's not as if Gilbert's scoring all these points purely by being some kind of crazy chucker who really ought to pass to his teammates more. He's not the most efficient scorer in the world, but he's taking the lion's share of his team's shots and doing so efficiently relative to his teammates. In other words, he's a very good player whose skills are covering for the weaknesses of many of his teammates, not a pretty good player played with a strong supporting cast.
He's not the most efficient scorer in the world, but he's taking the lion's share of his team's shots and doing so efficiently relative to his teammates. In other words, he's a very good player whose skills are covering for the weaknesses of many of his teammates, not a pretty good player played with a strong supporting cast.
Isn't the point of WoW's analysis that skills other than scoring are undervalued? (Or, alternatively, that scoring is overvalued?) What does Gilbert do other than score fairly efficiently? And how does his skill set, other than scoring, compare to the skill sets of other players that are also good scorers?
Now, I could see scoring (and scoring efficiency) being more important for a #1 player on a team that doesn't have any other players who can score well. But, at least theoretically, the 'Zards have another couple of guys that should be able to pick up some scoring slack. The strange thing is that the scoring is SO down when Gilbert is off the court - why don't Antawn and Butler step up???
I'm not sure that's right. The WoW guys are specifically concerned with wins produced. You would, one assumes, do best to look at records over injured-player periods rather than offensive or defensive numbers. But I can't quite articulate why that seems right.
Can you explain? I don't think I understand your point.
This is the same guy that Malcom Gladwell profiled, right?
His attempt to use brute regressions to create a "win shares" stat a la Bill James is horribly misguided. Baseball is inherently an individual sport. The pitcher and hitter have great control over the outcome of each event (pitch), which is then repeated ~300 times per game, 162 games per year. Sure, a really great defense might help a pitcher, but we can actually isolate this out with a stat like DIPS (defense independent pitching stats) ERA and see how "lucky" a pitcher was for a given season. Same thing goes for comparing Roger Clemens to Cy Young. We have a lot of really good data for how tough the National League was in 2006 vs. the American League in 1903. Even park effects (think Fenway's Green Monster) are slowly being controlled for as baseball stats get better and better.
By contrast, basketball is a pure team sport. Every single play involves the interaction of 5 moving parts reacting to the opposing 5 defenders. Even when an Allen Iverson or Michael Jordan dominate the ball and "take over", there are many important functions that their teammates play. There is also a huge tendency to make unselfish players look better than they are. Basketball teams need to take 80+ shots to win a game and some players (Kobe Bryant and Gil Arenas are two) are simply better equipped to take 25 shots a game. And basketball is unique in that the best offensive players already are handicapped by being the focus of every opposing defense. Mark Blount is a very efficient scorer, but if you made him the center of an offense surrounded by average talent it is likely that all of his rate stats (FG%, points per shot, etc.) would decline.
82games.com is a great start at looking for how these interactions actually work, but they have a huge sample size problem. Arenas played 85% of the available minutes last season and was in 18 of the 20 most common lineups used by the Wizards. Antonio Daniels got a few minutes at backup, but most of them went to Chucky Atkins. His basic +/- is then skewed by the fact that Chucky is pretty much a terrible player. Over the long term +/- is a great stat, but there is still a lot of work to be done for making basketball as friendly to analytics as MLB.
Tim - To be clear, the following exercise is purely for the sake of argument. I don't necessarily think they're better than Arenas, but I think they're all at least comparable. I'd rather have Gil than at least eight of these guys. The point, though, is that it's not hard to find 36 who could be considered better; here are 43 that I came up with without thinking twice.
Pierce, James, Nowitzki, Terry, Anthony, B. Wallace, R. Wallace, Billups, Davis, J. Richardson, Yao, McGrady, J. O'Neal, Stojakovic, Brand, Cassell, Bryant, Odom, Gasol, S. O'Neal, Wade, Garnett, Kidd, Jefferson, Carter, Marbury, Francis, Paul, Howard, Iverson, Webber, Nash, Marion, Stoudemire, Randolph, Artest, Duncan, Ginobili, Allen, Lewis, Bosh, Kirilenko, Okur
Again, I'm just exploring here. But I don't think it's so absurd.
www.82games.com has 2 intersting reviews on the book. If you follow the formula, it massively overrates rebounding and massively underrates the ability to create lots of shots. John Hollinger's PER, another complex stat, has Arenas as the 10th best player in the league last year, though unilke Wow it doesn't include individual defense.
I was gonna write a long post on this but Dan K pretty much nailed what I was gonna say.
Can you explain? I don't think I understand your point
I'm not sure I do either, in part because I don't know how the numbers at WoW or 82games are derived. Nonetheless, I'll try to explain what I was getting at.
WoW is looking at how many games Arenas wins for the Wizards, and 82g is looking at the net points he gives the Wizards. Is there any reason to believe those two metrics need not move in tandem? According to 82g, Arenas makes the Wizards about 2.5 pts./game better than its opponents. Without him on the floor, they're about a half point worse than their opponents. I'm not sure that such small differences--both between the team with him and without him, and between the Wizards and their opponents--offer much ability to predict the number of wins. You could imagine the Gilberted-Wizards winning by six one night and losing by 1 the next, over and over, throughout the season. That would make the Wizards a .500 ballclub, and Gilbert would still be worth about 2.5 pts. per game. You could do the same for the club without him: they serially win a game by one and lose the next by two, and still end up with a .500 ballclub. So how do we look at those numbers and know that the Wizards are a better team--one predicted to win more games--with him than without him? I don't know. I would think a better predictor, and one more similar to the WoW number, would be the Wizards winning percentage in games with him and the percentage in games without him.
(NB: This analysis is all less-than-back-of-the-envelope, so it's entirely possible I'm missing something obvious here.)
@Daniel Warren:
Its my understanding that WoW doesn't include individual D, it includes some team adjustment that they say doesn't affect the relative rating of players much (http://sonicscentral.com/apbrmetrics/viewtopic.php?t=877&start=0 first post in this thread). I haven't read the book. If it ever shows up in my library I might, but as I've mentioned before I've read reviews by experts in the APBRmetrics community who I trust and they think it has many flaws.
Crack
While I'm a big Wizards fan, and a big Gilbert fan too, I don't think it's as easy to critique WoW as some of you do.
Note that the big claim they make for their stats is that if you add up the WP for each player on a team, the total closely tracks their season wins - and does so for all teams.
That is a strong test. Any other stats have to be compared on this basis. Unless, of course, you think 'great player' is a category that exists independent of the actual games played.
Wizards' brass think that they now have a team capable of winning 50 games in the upcoming season. Lets hope they're right and the WoW analysis is wrong.
Dan K pretty well nails it. In a fluid, team sport like basketball the idea that a formula can deliver a number is laughable. Stick with Washington and go back before their name change. At one point they started Charles Jones at center for the better part of a season. Jones was a great defensive center - he regularly faced O'Neal, Ewing, Olajuwon, etc and held them below their average PPG. That's the good news. The bad news is he was incapable of scoring - averaging something like 1.5 PPG. Say O'Neal comes in averaging 25 PPG and CJ holds him to 18. That’s a damned fined performance. Unfortunately if he only scores 3 points the Bullets have spotted Orlando 15 points at that one position and have no one on the roster to make up the difference.
Put CJ on a team like the Bulls with Phil Jackson platooning him – only putting him in when he needs to slow down an opponents center – with no need to worry about his inability to score due to Jordan and Pippin – and people would have been singing his praises.
Same player, same skill set, yet vastly different results. Basketball players can’t be reduced to a simple statistic, there are far too many variables.
Comments closed October 16, 2006.
| 
Heads in the Sand
I think this comes down to Arenas' modest field goal percentage, modest rebounding, and so-so defense. Gilbert's rate states looks an awful lot like Allen Iverson's, and there are plenty of people (yours truly included) who think Iverson is the most overrated player in the NBA due to his poor shooting percentage.
Posted by Nicholas Beaudrot | October 2, 2006 11:17 AM