Henry Abbott titled his Q&A series with statistical analyst Wayne Winston "Mark Cuban's Stats Expert Isn't Bashful". As it turned out, Winston saved his boldest declaration for last Friday, the same day Abbott reported that the Mavericks had decided not to renew the Indiana University professor's contract, ending a relationship that dated back to 2001. Kevin Durant's poor adjusted plus-minus rating, Winston was asked whether he would have advised the Mavericks to acquire the budding Oklahoma City star if he was freely available.
"I'd say probably not," Winston answered. "I would not sign the guy. It's simply not inevitable that he'll make mid-career strides. Some guys do. But many don't, and he'd have to improve a lot to help a team."
While the TrueHoop post contrasted Winston's opinion with those of NBA scouts, who remain unanimous in their belief in Durant's skills, this certainly isn't a Moneyball-style stats-vs.-scouts debate. As Abbott noted, stats derived from the box score showed Durant to be a valuable contributor last season. He posted 8.6 WARP, tops on the Thunder and good for 35th in the league. If anything, the discrepancy is an example of the schism in the APBRmetrics community over the value of adjusted plus-minus I discussed in March.
How do we explain the discrepancy between the different statistics? And what does that tell us about Durant specifically and adjusted plus-minus in general?
Adjusted plus-minus often fails to shed light on the reasons behind the results. Analysts have been able to break it down into offensive and defensive components, but beyond that the "why" is often missing. In cases like this, where adjusted plus-minus defies conventional wisdom, even of the statistical variety, that's important to have.
While offense/defense splits of adjusted plus-minus are not available for last season, Eli Witus did offer them for 2007-08, Durant's rookie season. He ranked among the league's 10 worst players in defensive adjusted plus-minus, while his offensive adjusted plus-minus was slightly below average. Unadjusted net plus-minus numbers from last year show something similar; per BasketballValue.com, Oklahoma City allowed 7.9 more points per 100 possessions with Durant on the court, while the Thunder's Offensive Rating dipped by 0.8 points per 100 possessions when he was in the lineup.
That Durant has such a negative impact on defense is hardly a stunner. The counterpart statistics we tracked for Durant last season were not nearly so poor, though. I am a believer in Durant's defensive potential, but it is reasonable to assert that he is still struggling to convert those gifts into production at the defensive end.
Durant's non-impact on offense is much more difficult to explain. As a rookie, Durant was an inefficient scorer as he honed his shot selection and learned how to play against NBA-caliber defenders. That work paid off last season, as Durant posted an impressive .577 True Shooting Percentage, solidly above league average and second on his team behind role player Nick Collison. Durant used 28.4 percent of Oklahoma City's possessions while on the floor, putting him 13th in the league. That means in his absence, less efficient Thunder teammates had to create far more of their own offense. In that context, it's surprising that the Oklahoma City offense did not fall apart, let alone play better.
To better understand this phenomenon, I used the stats on lineup duos tracked by 82games.com to explore how Durant's main Thunder teammates played with and without him on the floor. From those stats, I estimated each player's usage rate with and without Durant as well as a simple Offensive Rating--points scored divided by possessions used by the player multiplied by 100.
-- W/DURANT -- -- W/O DURANT -- -- CHANGE --
Player Usg ORtg Usg ORtg Usg ORtg
Watson .150 63.1 .195 76.9 +.045 +13.8
Westbrook .247 80.8 .298 81.8 +.051 + 1.0
Sefolosha .119 81.5 .161 92.6 +.042 +11.1
Weaver .112 81.7 .151 93.5 +.039 +11.8
Green .206 92.3 .243 96.3 +.037 + 4.0
Collison .135 101.5 .139 113.8 +.005 +12.3
In Durant's absence, despite taking on larger roles--only Collison did not significantly increase his usage--his teammates scored at a more efficient clip. We really have two different stories here. There are three players--Thabo Sefolosha, Earl Watson and Kyle Weaver--who picked up the majority of the minutes when Durant was on the bench. All of them played significantly better in this role. Meanwhile, Jeff Green and Russell Westbrook saw little change to their minutes when Durant sat but played go-to roles on offense and did so without sacrificing any efficiency.
What is unclear is that either of these stories reflects poorly on Durant. Let's take the role players first. Sefolosha is an easy point of comparison because he spent the first half of the year in Chicago before being traded to Oklahoma City. Shortly after his arrival, he was a key figure in the Thunder's 5-2 run without Durant in early March. However, it is Sefolosha's play as Durant's wingman that is more consistent with what he accomplished as a Bull. Sefolosha's simple Offensive Rating in Chicago was 84.3, and that was the best mark of his three-year career. In that context, it's hard to argue Durant was holding Sefolosha back or anything of that nature; he simply played some of the best basketball of his career in Durant's absence.
Along those lines, there is a simple and reasonable explanation for why Watson and Weaver played better with Durant sidelined. This meant an opportunity to play heavier minutes and log extended stretches on the floor. Studies on playing time have shown that reserves tend to play better when temporarily thrust into larger roles because of injury. (While Collison was not playing more, his uptick without Durant also seems out of place in his career. He shot 60.0 percent from the field with Durant off the floor; his overall 56.8 percent shooting in 2008-09 was a career high.)
The way Green and Westbrook responded to Durant's absence is also interesting. Looking more deeply into the statistics, both players shot a lower percentage from the field when asked to shoulder a heavier offensive load (Westbrook significantly less, his FG% dipping from .406 to .380). However, their True Shooting Percentages improved. Why? Free throws. Westbrook attempted .445 free throws for each field-goal attempt when Durant was sidelined, up from .369 alongside Oklahoma City's go-to player. Green took an even bigger leap, from .255 free throw attempts per field-goal attempt with Durant to .404 without him.
While I haven't watched these games, I'll say that it seems unlikely that Sefolosha and Weaver were setting up Green and Westbrook. Instead, with Durant on the bench, Thunder coach Scott Brooks was calling their number and asking both players to operate one-on-one or play pick-and-roll basketball. As it turns out, both responded well to this role change, getting to the basket and drawing fouls. The suggestion is apparently that, at this stage in their careers, Green and Westbrook are better suited as go-to players on offense rather than playing away from the ball. This is not surprising for Westbrook, a poor shooter who got most of his points as a rookie off the drive. Since Green was a high-percentage three-point shooter and is viewed as a complementary option on offense, this theory does contradict conventional wisdom.
The way Durant's absence influenced how Green and Westbrook were used on offense helps illustrate my biggest quibble with Winston's comment denigrating Durant. It ignores the fact that adjusted plus-minus captures a player's value only in the context of a certain role with a certain team. On the Mavericks or any other team, Durant might thrive in a different role. This is a weakness of any statistical rating system, not something germane solely to adjusted plus-minus, but that's why analysts must be careful to supplement the bottom-line numbers with skill-based statistics and observation.
Adjusted plus-minus is unique in its unreliability. As bad as Durant's adjusted plus-minus was last season, there is still a small chance--about one in 40--that he was actually an above-average player, based on the standard error calculated by BasketballValue.com. Assuming a normal distribution, here's a graph that shows the distribution of Durant's "true" adjusted plus-minus in 2008-09. Highlighted in orange is the above-average portion of the distribution.
Winston has argued that he has proprietary methods that reduce the size of the standard errors associated with adjusted plus-minus, but without an explanation or even examples, it is impossible to assess this claim. BasketballValue and Stephen Ilardi have relied on using data from multiple seasons to create more accurate estimates of adjusted plus-minus, but Durant is an excellent example of the danger of this approach. By all other accounts, Durant improved dramatically from 2007-08 to 2008-09, but the Ilardi method treats him as having the same value for both seasons.
None of this is to completely absolve Durant of any responsibility for improving his adjusted plus-minus. While a poor rating in a single season can easily be a fluke, consistently bad adjusted plus-minus numbers are a sign something is wrong. If Durant is one of the NBA's worst players by adjusted-plus minus again in 2009-10, it will be hard to argue in his favor. For now, I'd be willing to take that chance--especially if Durant were offered for free.
Kevin Pelton is an author of Basketball Prospectus.
You can contact Kevin by clicking here or click here to see Kevin's other articles.