Its a long read, but there are some interesting conclusions about "false superstars:" http://www.nytimes.com/2006/06/10/opinion/10berri.html The N.B.A.'s Secret Superstars By DAVID J. BERRI Published: June 10, 2006 Bakersfield, Calif. THE N.B.A. finals tipped off Thursday night, and while most reports of the Dallas Mavericks' victory over the Miami Heat focused on the stars — Dwyane Wade and Shaquille O'Neal of the Heat and Dirk Nowitzki and Jason Terry of the Mavericks — few mentioned two players who were arguably just as important to the result: Antoine Walker and Erick Dampier. To understand why these two gentlemen mattered, one has to look beyond the standard focus on points scored and consider a new way of judging athletic performance: whether a player's actions — both scoring and non-scoring — help his team win or not. For several years now, baseball professionals (and rotisserie-league wannabes) have used so-called sabermetrics to judge a player's objective value based on his individual statistics. Of course, because one bats or pitches in baseball without the help of teammates, it is easy to separate a player from his team. Plus, the numbers in baseball are easy to understand: a home run clearly creates more wins than a single. Basketball, however, is a true team sport, and thus the numbers are not as obvious. Which is more valuable, a basket scored or the assist that made it possible? Is a rebound more valuable than a blocked shot? This is not such a mystery if we are judging the individual impact of, say, Michael Jordan, a player who excelled at every facet of the game. Most N.B.A. players, though, are not "Like Mike." Rather, they contribute in only one or two aspects of basketball, and ascertaining the relative value of such diverse talents is difficult. To make it possible to compare the apples and oranges, it helps to look at each statistic in terms of how it contributes to wins. So, along with two other economists, Martin Schmidt of the College of William and Mary and Stacey Brook of the University of Sioux Falls, I developed an algorithm that measures a player's "wins produced" for his team. The algorithm begins with the factors tracked for individual players: points, field goal attempts, free throw attempts, turnovers and so on. Then, through a fairly standard statistical analysis, it links these factors to team victories. What we find is that each point, rebound and steal has relatively the same positive impact on wins. Each field goal attempt and turnover has an equal negative effect. The other statistics — like free throw attempts, blocked shots, assists and personal fouls — do matter, but to a lesser degree. After an entire 82-game season, we take all these numbers and decipher how many wins every player in the league produced for his team. How accurate is this system? Well, when we add up the wins of each team's players, the results are shockingly close to the official standings. By our measure, Dallas should have won 57.4 games this regular season, and the Heat 51.2 games. In actuality, they won 60 and 52, respectively. For the past 10 seasons, the average difference between our projection and the actual number of games won by each of the league's teams is only 2.3 wins. So what does this tell us about individual players? Well, one result is obvious. The players who excel at many aspects of the game produce a lot of wins. In the six years Michael Jordan led the Chicago Bulls to the N.B.A. title, he averaged 23.4 wins per season. Versatile players like LeBron James, who produced 20.4 wins for Cleveland this year, and Kevin Garnett, who produced 26 for Minnesota, also dominate. But there are players who score a lot but also have a significant deficiency or two, like poor shooting efficiency or lots of turnovers. These players often have a level of win productivity far below expectations. This group includes supposed superstars like Allen Iverson of the Philadelphia 76ers and Carmelo Anthony of the Denver Nuggets, who this season combined to produce only 13 wins for their teams. Among players in the finals, the Heat's Walker, a three-time All-Star, produced only 1.7 wins; and Jerry Stackhouse of Dallas, who has averaged 20 points a game in his career, actually cost his team a third of a win overall this year. Our system also shows that it is possible for a non-scorer to produce high numbers of wins. For example, extreme rebounders like Ben Wallace and Marcus Camby have produced much higher win totals than fans might expect. Wallace led the Detroit Pistons with 20.1 wins produced this season while Camby, despite missing a third of the games because of injury, produced 13.7 wins to lead Denver. So, who are the "best" — or most productive players — on the teams vying for the title? Each team has at least one player who does many things well. On Dallas it's Nowitzki, who produced 18 wins this season; the multitalented Wade led the Heat by creating 18.2 victories. O'Neal played in just 59 games this year, so his 8.5 wins produced was low by his standards; but in his career he has averaged 21.3 wins produced per 82 games played. Certainly, saying those three players are great is hardly a shock. What might be surprising, however, is the importance of some non-stars. After Nowitzki, Dallas is led in wins produced by Dampier, a low-profile center, and starting forward Josh Howard, who each accounted for 7.8 wins. For the Heat, Wade and O'Neal are primarily assisted by the workmanlike forward Udonis Haslem (7.0 wins produced) and O'Neal's backup, Alonzo Mourning (5.7 wins produced). Of these role players, only Howard averages more than 15 points per game. So why do I think Dampier and Walker were vital to Thursday's outcome? Dampier scored only eight points, but he was efficient: making three of four shots and adding seven rebounds. Walker, on the other hand, scored an impressive-seeming 17 points, but he took 19 shots and turned the ball over an astounding six times in the loss. What should you watch for in the N.B.A. finals? When looking at the scorers, think about efficiency. And keep an eye on role players like Dampier and Haslem. What the non-scorers do matters, and in fact, it is those players who might ultimately decide who gets to be crowned N.B.A. champions. David J. Berri, an economist at California State University at Bakersfield, is a co-author of "The Wages of Wins: Taking Measure of the Many Myths in Modern Sports."
Yeah that's pretty interesting. Kinda like the +/- stats on 82games.com But, Carmelo and Iverson - costing their team wins? That's a surprise.
I'm not so sure if AI "cost" his team that many wins considering the fact that he has no one around him and he carrying the load of the team night in and night out. The problem with these stats is that they try to make absolute claims such as these when really its alot more situational. Ben Wallace's high rebounding totals are not always indicative of anything. Alot of his rebounds could just be easily clearing the ball on the defensive end when all other 9 guys have already started running to the other end of the floor.
AI's number is from his abysmal shooting, though (at least in past years - I didn't pay attnetion this year). Sure, he scores 30 pts. But if he takes 30 shots to do it, how helpful is it? If Webber and others took those 30 shots, would they have combined to score 30 more points? The analysis says no, but that it would be close - thus making AI not as great as his 30 ppg may indicate. (if i understand correctly, if it would be better that Iverson didn't exist, that would be reflected by a negative win share).
It's sort of interesting. But before I can pass judgement, I'd like to see the methods used in more detail and also a list of how all the players in the league rate in terms of producing wins for their respective teams (does it pass the "laugh test"?). And if a player produced X wins, does than mean if you replaced that player with an average player the team would win X less games that season? I think that's the correct interpretation, but I'm not totally sure. It appears that Berri's rating seems to give a lot more credit to rebounding than is conventoinally given. Below are the results, using different methods, to get at the question of how many "net wins" a player produces for his team. It's based on the methods described in Dean Oliver's Basketball on Paper. I believe Berri "weights" different stats by considering how they correlate to wins/losses throughout the season. On the other hand, Oliver's approach is to use the available stats to estimate points produced/poss on offense and points allowed/poss on defense, and then derive individual wins/losses from that. Oliver's results are very much team dependent. For instance, players that a big role on bad teams will typically have negative Net Wins. Does that mean if you replace them with an "average player", the team would be better? Not necessarily. Here are the top 100 in net wins this season. Here are the bottom 100 in net wins this season.
I think his formula is looking for the "extra" production from that players normal output, when linked to wins. The same for the losses, the formula looks for below par production. I suppose.
that's why it's so difficult to do for basketball, because it's so hard to figure what those 30 shots would be. their normal shots are a result of having iverson on the court. based on his 8+ apg, he presumably creates quite a few easy shots for them. and even past that, if the rest suddenly had to create 30 extra shots split among themselves, who knows what happens to their efficiencies. maybe iverson is inefficient only b/c if he doesn't take those shots, everyone else would take even less efficient shots. or maybe iverson really is a ballhog who is getting his 30 no matter how many shots it takes and really it would be better if he didn't carry everything. i wonder how they figured the impact of each stat. did they just correlate each stat to wins to obtain a regression like: wins=41+5*(steals-8) where, hypothetically, the average number of steals for teams is 8 (and obviously the average number of wins is 41) and each extra steal for a team generally accounts for 5 more wins, or did they do something more complex. while i would think it would be more complex, this would give results pretty close to actual results b/c it would be based on what already happened. if you choose enough stats that have to be centered around the mean of 41 games, adding them up for each team would give nice results b/c their would be enough stats to even out any anomalies. i've done this before with point differential, and in two different seasons, i got wins=41+2.7*PD. it was kinda cool that i got the same 2.7 number for two different seasons. so basically, a point each game is worth 2.7 wins over the course of a season, or each point is worth about .033 wins (2.7/82). and this usually gives results pretty close to what actually happened. i wonder if it was 2.7 this season. edit: 2.746
Here's the website for Berri's book: http://www.wagesofwins.com/ Here's a post about their Win Scores method, from the book's blog:
in the most simple terms: high FG%, rebounds, steals = good. TO's, low FG% = bad the other stats matter but to a lesser degree. --------------------------------------------- so here are the Rockets who are tops on the team in the bad catagories: low FG% - bowen (30%), brunson (35%), alston (38%), bogans (39%), wesley (40%), tmac (40%), head (40%). TO's - yao (2.6), tmac 2.5), rafer (2.5). everyone else is under 2 TOPG. and now for the good: high FG% - hayes (56%), deke (53%), yao (52%), stro (49%), jho (45%). frahm is ok at 42% because they are mostly long range shots. rebounds (per 48 minutes to even the playing field)- hayes (16), deke (15), yao (14), stro (10), jho (10), tmac (8), bogans (7), head (5.5), rafer (5) steals (P48)- hayes (2.3), rafer (1.9), head (1.8), tmac (1.6), bowen (1.5), bogans (1.4), stro (1.3), wesley (1.1). the rest are under 1SPG. ---------------------------------------------------------------------- as you can see, we are a very good rebounding team. our bigs have nice FG%. however our perimeter players have poor FG% and don't cause enough TO's by stealing the ball. sure tmac's stats are messed up because he wasn't himself. rafer's stats may be worse because he was taking on too big of a role (low FG% and TO's because he was forced to do too much). the guys that we need to get rid of are: wesley - low FG%, no steals. brunson - low FG%, nothing positive. bowen - he's a wash. very low FG%, but average in steals. we could find someone more productive. Jho - decent FG% and rebounds. his problem is that stro and hayes have better #'s. questionable: bogans - low FG%, but makes up for it with rebounding and steals. rafer - low FG%, high TO's, nice rebounder for a PG and good steals. playing with a healthy team will probably increase FG% and lower TO's. worth another chance. guys that need more PT: hayes - leading the team in FG%, rebounds and steals. all 3 things on the "good" side. stro - similar #'s to jho, but higher FG% and steals.
Maybe we got a lot of rebounds because we missed a lot of shots. I know that Golden State typically gets a lot of rebounds, and they certainly miss a lot of shots.
Code: Factor Off (Rk) Def (Rk) +------------+---------+----------+ Shooting .471 (27) .472 ( 6) Turnovers .160 (23) .146 (25) Rebounding .255 (24) .746 ( 7) Free Throws .240 (21) .247 (16) +------------+---------+----------+ Rockets were 7th in defensive rebounding%, and only 24th in offensive rebounding%. Number of shots missed shouldn't impact those numbers.
It'd be funny if Avery and some pale faced engineer gave Dampier that "2nd best center in the league" line.
this is nothing new that a good coach would tell you. these stat geeks coming out of the wood works is hilarious. high fg%, rebounding, assist, defensive player gets you wins, no freakin kidding.
im not the biggest believer in stats analysis, But i believe the antwan walker comparison is spot on. He sucks.
For this to come close to being taken seriously, you'd have to quantify all this stuff for each time an individual was guarding another and break it down by individual possession. For instance, if Bowen is guarding Dirk for 30 Mavs possessions during the game, shouldn't we look at Dirk's numbers while Bowen's on him and compare those to Dirk's overall numbers and give Dirk's negatives to Bowen's positives and vice-versa? In other words, If Shaq gets 4 rebounds less than his average when Dampier is guarding him, that should be quantified in a way that is a positive for Dampier, whether he gets the extra rebounds or not... sometimes when you're playing certain folks, just preventing them from doing their usual stuff is sometimes enough to give your team a chance to win.
In science, you use models. This is a model. It makes simplifying assumptions. There's nothing wrong with that, and it's no reason to assume it has no value.
Berri's system seems horribly skewed, and to me it seems to be doing the APBR community a disservice. Anyone who watched Raptor, Bulls or Knicks games over the last few years saw Jerome Williams could do to a team. Too slow to guard small forwards, too short to guard big ones, he was the worst "great" defender I've ever seen -- constantly improvising on defense, picking up plenty of steals, blocks, rebounds, but doing very little for the overall team D. Guys, if you're going to make the jump into understanding new NBA numbers, go with Dean Oliver's book, or Hollinger's yearly run-throughs, and save your money on this crap. (and, as was noted earlier in this thread, any moron could have looked at Walker's G1 boxscore and noted that he hurt his team. 19 shots to get 17 points? Six turnovers? No FT attempts? That's easy stuff, right there)
From the book's excerpt, it appears that they guage the accuracy of their model by summing up all the player's Win Scores on a team and comparing it to the number of wins the team actually got. Apparently, they find that it's pretty close. Still, that doesn't really get at the question of whether the "win shares" are really being properly alloted amongst the players. And by BoP and Hollinger's methods, Jerome Williams looks pretty good as well, since they're still based on the conventional box score stats with some team stats mixed in for context. Incidentally, according to 82games, Jerome has had a positive impact on team defense his last 3 years (2003, 2004, and 2005): Code: [COLOR=Navy] Team On Off Net[/COLOR] 2003* Tor 94.7 99.2 -4.5 2004 Tor 98.5 103.2 -4.7 2004 Chi 105.8 106.0 -0.2 2005 Nyk 105.9 112.2 -6.3 * points allowed [i]per 48 min[/i] Now, I haven't followed Jerome Williams closely the last few years. But it's interesting that this +/- data indicate he's an impact defensive player. That appears to jive with Berri's claim. But without knowing the precise formula Berri uses, I can't say if this is just a fluke or not.
