Deep n' Insightful stats (a bit long)

By Craig Burley

Some recent reading (and subsequent work) I have done on basketball statistics is beginning to confirm to me that, like with baseball, basketball is remarkably quantifiable. As a former basketball coach and player and a longtime devotee of the sport, I found this to be intensely surprising, but also intensely pleasurable. I have often in the past derided basketball statistics as hiding more than they reveal, but like baseball stats, when properly accounted for and subjected to rigorous analysis, they are capable of producing insight that is startling for its originality and power.

A huge factor in my recent conversion into a basketball stathead has been the work of John Hollinger, author of Pro Basketball Prospectus, which is a book you should RUN to the store to buy. John sets out his methods in just fourteen pages of a 300-page bible to the past season, but they are the second most important thing I have ever read on basketball. (The most important thing? Easy... a short article from 1986 in which a young Providence College coach named Rick Pitino outlined the fundamentals of his 1-2-1-1 pressure defense).

Anyway, without spoiling too much of the fun, Hollinger's approach takes every event recorded in a statsheet and assigns it a value, denominated in points, based on its impact on the game. The point value of a steal? Well, it's the value of one possession, so a steal garners a player 1.001 points, which is the average value of a possession in 2003/04. The point value of an offensive rebound? That's the value of one possession, times the chance the defense would have gotten the ball otherwise (or the "league defensive rebounding percentage").

He does more than that, though. The most important insight in the whole book is that all statistics need to be read within the context of a team's pace... the number of possessions both teams use in a game. On teams that play to a slow pace, each possession (and each point) matters just that bit more. Also, did you know that the "Pythagorean Theorem" that we can use to predict team records in baseball from runs scored and runs allowed, also works in basketball? The exponents are different, is all.

Anyway, you can add every element of a player's record -- his 3-pointers, assists, rebounds, steals, blocks, fouls, made and missed field goals and free throws, and turnovers -- and get a nearly complete picture of that player's contribution to his team. What it misses, is the ability of a player to spring other players open, and on-ball defense (including help defense). These are very important, but that's all that's missing from this superstat, which is called Player Efficiency Rating or PER.

PER isn't perfect. Some events in the game are double-counted; others are single-counted. For example, a steal is counted as a turnover for the defense and a steal for the offense. A made field goal, on the other hand, counts as a positive for the offense (part credited for the assist, part for the make) but nothing is counted against the defense. However, PER is more than good enough for government work... it's a terrific shorthand at how much a player is contributing to his team's offense and defense.

PER itself is a per-minute stat, but it can also be calculated in its "raw" format, which lends itself to MVP-type analysis. It also can be given a Linear Weights type analysis, and is particularly amenable to a "replacement-player" analysis. PER can also be broken down into its component parts. Hollinger has a measure he calls Offensive Percentage, which uses the offensive components of PER and compares it to the number of possessions a player uses to generate that offense.

I have two metrics of my own design, called Individual Defensive Value and Adjusted Defensive Value, which use the defensive components of PER. Adjusted Defensive Value uses a PER-type analysis to measure a team's field-goal defense, for which value is then distributed amongst the players on that team and added to their Individual Defensive Value.

The creme de la creme of the metrics, however, is my Wins Above Replacement metric. More than even PER, Wins Above Replacement (which is in the very early stages of development) is a basketball superstat, similar to VORP or Win Shares for baseball. Essentially, the metric measures the total points a player is credited for in the PER system, subtracts the points a "replacement level" player (usually with a PER of about 11.00 on the normal scale where 15 is average, but it can be moved up or down), adjusts for the "double counting" problem, and then plugs that into an average team's results using the Basketball Pythagorean Theorem.

Wins Above Replacement, as I have indicated, is in the early stages. I find that it does a good job of capturing value, and curiously the sum of a team's WinsAR is very close to a team's actual wins no matter whether the teams are good or bad teams. This would indicate that a team of "replacement level" players wouldn't win very many, if any, ballgames; a theory that does appear to be borne out in practice. Again, on-ball defense is not counted in this metric, and I am working on including it, as well as putting players in team context instead of league context. For now, though, the WinsAR metric will help us look at the midseason candidates for NBA MVP.

The MVP Candidates

First, let's get the list out of the way. The top 20 NBA players in 2003/04 to date (actually to February 6, the day for which I currently have data) are:

Player Team WinsAR
Kevin Garnett MIN 12.3
Tim Duncan SAN 10.3
Tracy McGrady ORL 9.8
Sam Cassell MIN 8.3
Peja Stojakovic SAC 8.2
Andrei Kirilenko UTA 7.8
Shareef Abdur-Rahim ATL 7.6
Elton Brand LAC 7.5
Stephon Marbury NYK 7.1
Jermaine O'Neal IND 7.1
Jason Kidd NJN 7.0
Zach Randolph POR 6.8
Yao Ming HOU 6.7
Shawn Marion PHO 6.7
Paul Pierce BOS 6.6
Dirk Nowitzki DAL 6.5
Baron Davis NOR 6.4
Andre Miller DEN 6.4
Donyell Marshall TOR 6.3
Pau Gasol MEM 6.2

Something should leap out at you immediately from that table. One is that the Timberwolves have a 35-14 record (through Feb. 6) based mostly on the superlative play of two players (no other team placed two players in the top 20!). Second is that there are three players in the NBA this year who have been head and shoulders above everyone else... Kevin Garnett, Tim Duncan, and Tracy McGrady.

You may notice something else about this list. The top scorers are not necessarily on it. Allen Iverson, Kobe Bryant, Vince Carter, LeBron James... these guys are high on the scorer's list, but not as high in PER and similar metrics. Iverson leads the NBA in scoring, yet is 26th in the league in WinsAR at 5.5, tied with Carlos Boozer and Rashard Lewis. Why is this?

The analogy I use is that points scored is an awful lot like RBI in baseball. Both are primarily a product of opportunity. The top scorers, like the top RBI men, are all good players. They are all good at converting opportunities into baskets or runs. But that does not make them the best players; the players who are setting up those opportunities have equal claim to offensive prowess. The players who are more efficient (who take fewer shots and waste fewer of a team's opportunities) have a better claim to offensive prowess. Possession in basketball is like outs in baseball... it is your commodity.

Also, in basketball, there is another wrinkle. Like in baseball (except for DH baseball) all basketball players are two-way players. But in basketball, that defensive role is equal for all five players (whereas in baseball as much as two-thirds of the defensive responsibility falls on the pitcher). In baseball, non-pitchers' defense may have considerable value, as it does in basketball.

Let's get back to the candidates.

Kevin Garnett, Minnesota

Garnett is having a monster season, full stop. Not only is he fourth in the NBA in scoring and first in rebounding, he has played more minutes than any other big man and his passing, shotblocking, and other skills have been terrific. He leads the NBA in PER with a 29.06 figure, well ahead of Tim Duncan at 26.56 and in the usual range for an MVP (which is around 27.5 to 30).

Garnett is over 12 wins above replacement; which is about the distance Minnesota is over .500 (taking away 12 wins would put them at 23-26, which is probably where they would be if you replaced Garnett with Horace Grant or Donnell Harvey). His teammate Sam Cassell is 8 wins above replacement. With KG but without Cassell (replacing him with Dan Dickau or somebody) the Wolves would be about 27-22, shockingly similar to Wolves teams in the recent past.

[Editor's note: A 27-22 record works out to 46 wins over the course of a full season. The Wolves' win-totals over the past four years: 51, 50, 47, 50.]

Garnett is clearly the real MVP right now; he has never approached these levels of performance before. He is grabbing far more rebounds (he's grabbing over 20% of available rebounds for the first time in his career; 20% is the hallmark of the game's truly great rebounders). He is making far fewer turnovers. And most importantly, he is creating far more offensive opportunities for himself than ever before. Put it together, and Garnett is carrying one of the league's best teams on his shoulders. It has been a Shaq-like performance.

Tim Duncan, San Antonio

Ho hum, another Tim Duncan season. Duncan's statistical profile is different in type, but equally valuable, as last season, when all he did was lead the Spurs to the NBA's best record and the championship, and won the MVP to boot. A player can do no more.

Duncan's doing it differently, as I mentioned. He has been creating far more offensive opportunities for himself, much as Garnett has - but like Garnett he is actually a less effective shooter than in the past (is this a product of the evolution of NBA zones? If so, yuck). Also like Garnett, his assists are down marginally, but so are his turnovers.

Duncan's 10 wins above replacement are impressive; the next best players on the Spurs are Manu Ginobili at +4.1 and Tony Parker at +2.8, so if San Antonio's 34-18 record is disappointing, it's not down to Duncan. But there is one other stat that goes a long way in explaining San Antonio's "disappointing" season.

Remember how earlier I was explaining that the Pythagoeran Theorem works in basketball too? Well, every team in the NBA is within three games of their Pythagorean record. Except for the Spurs and the Portland Trail Blazers. The Blazers are four games worse than their record indicates; but the Spurs are *under*performing their Pythagorean Expectation by a whopping SIX games.

The Spurs' "true" record is 40-12; better than Sacramento and indicating that if you can find a nice price on the Spurs you may do well to take it. They have been unlucky so far this year; I would not bank on that continuing. The Spurs are trying to set all kinds of records for defensive excellence; they allow a ridiculously low 91.3 points per 100 possessions by their opponents, easily the best in the NBA in some time. And we all know that in the playoffs, it all starts with defense. Duncan's contributions to that outstanding defense are a good argument and what must underlie any argument for him as MVP.

Tracy McGrady, Orlando

A great lesson in how a player can get lost when he's on a lousy team.

Aaron and I have discussed McGrady a couple of times now and Aaron rightly points out that McGrady has "Vince Carter disease" and doesn't slash hard to the basket to create offensive opportunities, instead hanging out at the 3-point arc and expecting others to make shots for him. It's a symptom of living without hope; McGrady's not likely to apply himself when the only thing that will accomplish is fewer balls in the lottery for the Magic. McGrady was the league's real MVP last year; he won't get there this year, but he might have if the Magic could play defense.

The Magic have been a tiny bit better than their 13-39 record would indicate, but nearly all of their problems this year have been defensive (opponents score 107.8 points per 100 possessions against the Magic, easily the worst in the NBA). On offense, they rank 12th, in large part thanks to McGrady, who creates shots for himself with ridiculous ease and does it without turning the ball over.

On a per-minute basis, McGrady is not in fact the third-best player in the NBA this year. Elton Brand has a PER of 25.56 to McGrady's 24.94, but he hasn't had the minutes because of a broken foot. T-Mac isn't a serious MVP candidate right now, not on a godawful Magic team. But it goes to show how a guy can be written off despite a good year, merely due to team context.

This is originally posted @
http://www.aarongleeman.com/2004_02_08_baseballblog_archive.html

 

francis4prez,

Interesting analysis.

I don't think winsAR stat is kept anywhere or is even calculated daily. I certainly hope it gets more popular and would shine more light on those players who really does make a difference to their teams fortune.

BTW, you may still want to click on that link and may be there is some extra info I didn't cut 'n paste here.