We return to our discussion of advanced basketball stats with a look at Win Shares, both offensive and defensive. Basketball Win Shares were derived based on the baseball Win Shares model established by Bill James, modified to fit the NBA game. The idea is to try to calculate how a player's performance translates to games won, without being unduly penalized for having a poor team. If you go through the roster of a team that wins 50 games, for instance, the total number of Win Shares earned by the players on that team should come out to about 50. For specific details on exactly how basketball Win Shares are calculated check out the Basketball-Reference site, but here I will give my interpretations on the stat's usage and usefulness.
What it is: A Win Share is an estimation of how much a particular player contributes towards a win when compared to an average player. The calculation relies heavily on Dean Oliver's Offensive and Defensive Ratings, which we talked about two weeks ago. For both offense and defense, a Win Share is roughly the player's marginal Rating "normalized" to the league average. In other words, a Win Share is kind of like baseball's VORP stat…it relies on the assumption that efficiency is the key to wins, looks at a player's offensive efficiency compared to the league average, a player's defensive efficiency compared to the league average, corrects for minutes played, then adds up the corresponding Offensive and Defensive Win Shares to get one number that estimates how many wins that player was worth.
Strengths: Win Shares is one of the stats that we will look at that estimates exactly how many wins an individual player is responsible for. Unlike PER, which outputs numbers that may appear arbitrary, the output Win Share is very straight forward: if player A contributed 10 wins vs. player B's 7 wins vs. player C's 5 wins, the user can easily gauge exactly how effective each player was with respect to the other. Whereas if player A had a PER of 18, player B a PER of 16, and player C a PER of 14 there is no easy way to say just how much better one is than the others in any kind of practical terms.
Another major strength is that this measure attempts to correct for teammate caliber. Offensive/Defensive Ratings are heavily influenced by how good the team is, which makes it difficult to separate an excellent player from poor teammates or vice versa (see the Kevin Garnett in 2007 vs. 2008 comparison in the Ratings blog). Looking only at Offensive/Defensive Rating it would appear that Garnett's play in '08 dwarfed what he did the previous two years ('08: ORTG = 118, DRTG = 94 vs. '06 – '07: average ORTG = 113.5, DRTG = 99.5). With Win Shares, though, we see that Garnett's 12.6 Win Shares in '08 is almost exactly the same as the 12.6 Win Shares that he averaged between the '06 and '07 seasons. The fact that his '08 team won more games than his '06 and '07 teams combined (66 wins vs. 65 total wins) was weeded out in the Win Shares calculation, whereas the teammate quality showed up in the Ratings.
Finally, Win Shares are a function of both efficiency and production. While PER and the Offense/Defense Ratings are normalized by possession and thus can yield unintuitive results such as role players like Carl Landry being in the top-5 in Offensive Rating, Win Shares takes production and minutes played into account (after measuring efficiency) and thus the leaderboard tends to pass the sniff test.
Weaknesses: The actual math and calculations that go into Win Shares is kind of dense. It can be navigated, but in order to understand exactly why each number is used would require a lot of back-research. So while the measure is user friendly to use, the calculation itself is a mysterious kind of a black box for most people.
The complex math ties into the major question mark: how do we know that this is really the best way to apportion credit for wins? The numbers produced are somewhat intuitive, and total Win Shares totals for teams tends to match fairly well (around +/- 3.5 wins for the average team), which is a good first level sanity check. But the actual justification for the system is also complicated and relies on the user to believe in the model with it's underlying assumptions and exceptions. Consequently, it is hard to use this type of mathematical model as evidence in a discussion with a skeptic who simply refuses to believe that a number can capture the essence of winning basketball (with all of the nuances implied). When debating or building a case, simple is usually good while complicated is easily ignored.
Usage: Win Shares is a stat that purports to measure a player's total impact on games won and takes both offense and defense into account. Therefore, one could make the statement that "according to Win Shares, player X is a more productive player than player Y." But like always, you have to understand that any mathematical model has weaknesses so this can't stand as it's own argument. It must be a supplement to a case built on personal observation, personal analysis and other stats/opinion.
One can also break Win Shares into Offensive Win Shares and Defensive Win shares to try to estimate who the best players are on each end of the court. For instance, one of the talking points in the Kobe vs. LeBron debate is that Kobe still is a better defender than LeBron. According to Defensive Win Shares Kobe was slightly more productive than LeBron on defense last season, but this year LeBron has blown past him. Taking all stats with a grain of salt, this at least makes one think critically about whether the blanked statement that "Kobe is the better defender" is actually true or not.
Win Shares Leaders this millennium:
'09 LeBron James (11.60, through 48 games)
'08 Chris Paul (17.31)
'07 Dirk Nowitzki (15.91)
'06 Dirk Nowitzki (17.18)
'05 Dirk Nowitzki (15.90)
'04 Kevin Garnett (18.10)
'03 Tracy McGrady (16.48)
'02 Tim Duncan (17.68)
'01 Shaquille O'Neal (15.02)
Highest 1-season Win Shares (only available since 1974): Michael Jordan (20.35, 1988)
Quick Links for series:
Intro
PER
+/-
Offense/Defense Rating
Kobe vs. LeBron by the numbers