Updated and Revised SportVU aided "TrueUsage"
Over the last month-plus, Kevin Durant has quickly changed the conversation from "maybe LeBron James isn't the MVP for this season" to "maybe LeBron isn't the best player anymore?" While I'm not ready to countenance the latter question, the former is, barring injury, plague or LeBron tiring of this talk and showing us mortals what's what, settled. KD's last month or so has sealed it. He has done so by not only using up basically all the possessions formerly the purview of Russell Westbrook, but using them at an incredible rate of efficiency.
Just over a month ago, I took a first look at a metric designed to describe an individual players overall involvement in an offense that included not just shooting, but also passing and turnovers. I was unsatisfied with the standard "Usage/TS%/TOV%" trinity of stats to describe players' respective loads, efficiency and ability to hang onto the ball. Using SportVU data from NBA.com, I was able to calculate the approximate percentage of of possessions involving that player while he was on the floor. I say "involving" rather than "using" because I think that the concept of crediting only one player with each offensive possession is unnecessarily reductive and binary. Further, it doesn't really align with what we see on the court. Guys who couldn't "create" their own shot in an empty gym get all the credit for using possessions on which they shoot, while the guy who draws the triple team in the post leading to the shot gets none.
So why not relax the assumption that only one player can "use" each possession, and instead try to estimate the number of possessions in which a player was integrally involved? That's essentially what True Usage attempts to do.
The additional benefit is to be able to evaluate players propensity for turning the ball over not just in comparison to their shot attempts, but also with respect to their involvement in setting teammates up for shots (make or miss - assist % is obviously very teammate dependent, whereas assist chances as measured by SportVU are less so). As I mentioned in the initial run through, the fact that point guards in general have significantly lower TO rates than big men by this accounting is an accurate reflection of reality; they are point guards because they handle the ball well to begin with.
So why not relax the assumption that only one player can "use" each possession, and instead try to estimate the number of possessions in which a player was integrally involved? That's essentially what True Usage attempts to do.
The additional benefit is to be able to evaluate players propensity for turning the ball over not just in comparison to their shot attempts, but also with respect to their involvement in setting teammates up for shots (make or miss - assist % is obviously very teammate dependent, whereas assist chances as measured by SportVU are less so). As I mentioned in the initial run through, the fact that point guards in general have significantly lower TO rates than big men by this accounting is an accurate reflection of reality; they are point guards because they handle the ball well to begin with.
So to return to Kevin Durant and his amazing run. The first time I looked at True Usage was shortly after Westbrook's injury was announced. I wondered how the Oklahoma City offense would adjust considering Westbrook had the highest (actually second highest to Chris Paul when I cleaned up the data a little bit) True Usage of any player in the league. Could Durant compensate for this huge loss? Would Reggie Jackson step up?
Well, first let's look at OKC's rotation players who I captured in the first iteration
Well, first let's look at OKC's rotation players who I captured in the first iteration
| Player | TS% | USG% | TrueUsage | TrueTOV% |
|---|---|---|---|---|
| Russell Westbrook | 51.80% | 32.20% | 58.49% | 9.88% |
| Kevin Durant | 62.94% | 29.40% | 46.25% | 8.56% |
| Reggie Jackson | 53.93% | 22.77% | 43.68% | 8.78% |
| Jeremy Lamb | 57.13% | 20.25% | 30.20% | 5.36% |
| Serge Ibaka | 55.65% | 18.94% | 25.07% | 8.95% |
| Thabo Sefolosha | 51.75% | 11.63% | 20.29% | 9.27% |
Unfortunately due to the parameters, I didn't get numbers for Derek Fisher or Steven Adams and since SportVU data doesn't have a date filter (grumble grumble), I can't go back and look, but here are the current stats through Thursday's games.
| Player | TS% | USG% | TrueUsage | TrueTOV% |
|---|---|---|---|---|
| Russell Westbrook | 51.80% | 32.20% | 58.49% | 9.88% |
| Kevin Durant | 64.50% | 31.70% | 49.01% | 8.72% |
| Reggie Jackson | 52.60% | 24.00% | 44.26% | 9.11% |
| Jeremy Lamb | 54.20% | 19.90% | 30.73% | 5.38% |
| Derek Fisher | 52.10% | 13.80% | 26.99% | 7.33% |
| Serge Ibaka | 56.40% | 20.10% | 25.93% | 9.37% |
| Nick Collison | 59.90% | 11.60% | 22.06% | 9.68% |
| Thabo Sefolosha | 51.30% | 12.20% | 20.44% | 8.84% |
| Steven Adams | 53.60% | 13.00% | 19.67% | 16.18% |
| Kendrick Perkins | 47.30% | 11.70% | 18.70% | 20.25% |
| Perry Jones | 57.70% | 12.80% | 17.80% | 7.50% |
So of the 5 non-Westbrook players (Durant, Jackson, Jeremy Lamb, Serge Ibaka and Thabo Sefolosha) over Durant's 30 point game streak (and actually including the game before he started), here are the True Usage stats:
| Player | MPG | TrueUsage | TrueTOV% |
|---|---|---|---|
| Lamb | 25.2 | 31.91% | 5.42% |
| Ibaka | 32.2 | 28.45% | 10.48% |
| Sefolosha | 28.2 | 20.79% | 7.92% |
| Jackson | 33 | 45.56% | 9.76% |
| Durant | 39.6 | 56.62% | 9.08% |
So while Jackson, Lamb and Ibaka each took on a little bit more of the offensive load, Durant went from 29th in the league and 6th among non point guards to the highest in the league over that span. Higher than LeBron with Wade missing time. Higher than every point guard (including John Wall, who is 3rd overall for the season, but been involved with about 55% of Washington's offense over the same time period).
The most amazing thing about the stretch of games is that Durant has carried this offensive load with only a slight uptick in turnovers and an overall offensive efficiency through the roof resulting in a somewhat absurd 71% TS rate, a figure reserved for either catch and dunkers like Tyson Chandler or the very best three point bombers, not players who create over half the shots taken by a team's offense.
The most amazing thing about the stretch of games is that Durant has carried this offensive load with only a slight uptick in turnovers and an overall offensive efficiency through the roof resulting in a somewhat absurd 71% TS rate, a figure reserved for either catch and dunkers like Tyson Chandler or the very best three point bombers, not players who create over half the shots taken by a team's offense.
So, moving on from Durant. First of all, I've posted a doc with TrueUsage and TrueTOV% stuff for the whole league (well, players who have played 20+ games at 10+MPG, all 325 of them) here. The one thing I'm not terrible pleased with is the ability to track "playmaking efficiency" in a way that will allow a TS% or EFG% style number to compare to TrueUsage. The points per assist chance stuff seems extremely noisy with no rhyme or reason why certain players do well and others don't.
Perhaps there will be a way to track assists and assist chances versus Hickory High's "Expected Points Per Shot," but at this point I'm not sure I have the raw data to work with to say anything nearly as definite about playmaking efficiency as TS does about scoring efficiency.
As a final note, I want to make exceedingly clear that True Usage is intended as a descriptive rather than normative stat. I don't think it's "better" or "worse" to have a high involvement on this metric, though I do think on a team level more assist chances (and thus more aggregate True Usage) probably correlates to some degree with scoring at higher clips just based on the types of shots which tend to generate assist chances (Cuts, Spot ups, Transition, PnR roll man) as compared to those that don't (Iso, PnR ball-handler, post ups)
Perhaps there will be a way to track assists and assist chances versus Hickory High's "Expected Points Per Shot," but at this point I'm not sure I have the raw data to work with to say anything nearly as definite about playmaking efficiency as TS does about scoring efficiency.
As a final note, I want to make exceedingly clear that True Usage is intended as a descriptive rather than normative stat. I don't think it's "better" or "worse" to have a high involvement on this metric, though I do think on a team level more assist chances (and thus more aggregate True Usage) probably correlates to some degree with scoring at higher clips just based on the types of shots which tend to generate assist chances (Cuts, Spot ups, Transition, PnR roll man) as compared to those that don't (Iso, PnR ball-handler, post ups)
RSS Feed
