As basketball statistics get more sophisticated, and everyone from bloggers to broadcasters focus on efficiency, it’s surprising that the work of Dean Oliver doesn’t get more attention. Authors with bigger web presences, namely Hollinger and Berri, overshadow Oliver and his book, Basketball on Paper. But it talks at length about how teams are put together, whereas it feels like a lot of current statistical discussion concentrates solely on evaluating individual players’ performance*.
In particular, Oliver devotes an entire chapter to an aspect of basketball that is critical to team offensive output, but is often obscured by focusing on individual stats: the fact that NBA superstars sometimes help their teams by performing less efficiently. More and more I see analysts and commentators praise individual efficiency as the golden standard, but Oliver presents evidence that we don’t want to get carried away and make it our sole focus.
They key is usage rate, which is the percent of a team’s possessions a player uses when on the court. The stats show that, for all players, as he uses more possessions, his efficiency decreases. How much that efficiency decreases and at what usage rate we see significant declines vary from player to player. What defines a superstar, in Oliver’s statistical analysis, is that he can shoulder a larger proportion of a team’s possessions with only a relatively small drop in efficiency. Meanwhile, the opposite is also true: players perform more efficiently when they are asked to use fewer of their team’s possessions. As a result, the greater burden on the superstar means that supporting players maintain low usage rates, allowing them to operate closer to their peak efficiency. By balancing usage rates and the varying offensive ratings of the five players on the court, a team can achieve optimal offensive output for the personnel.
The concept is a little counter-intuitive, but powerful. By looking at offensive efficiency in tandem with usage rate, stat nerds like myself get a more nuanced picture of a player’s output. Other metrics, such P.E.R., Win Score, etc., use a single number to measure performance, which is good for ranking players but not as good for understanding how team’s perform. Oliver’s measure of individual offensive efficiency, which he calls offensive rating, has the added benefit of being expressed in the same terms as team efficiency: points scored per 100 possessions (for a super-quick primer on these stats, see the second footnote**). It is more informative, at least to me, to see that LeBron‘s offensive rating is 117 (very good) and his usage rate is 34% (remarkably high, leading the league this year). This is especially true when comparing LeBron to a good role player like Carl Landry, whose offensive rating is 135 (still crazy-high for a guy who doesn’t shoot 3’s and sucks at FT’s) but whose usage rate is only 19%. Those stats are more descriptive to me than saying Lebron has a PER of 30.6 while Landry’s is 23.8, even though PER incorporates usage rate. It relates the individual to the team better.
Let’s take an example from Oliver’s book: the 2002 Lakers. That year L.A.’s starting lineup of Shaq, Kobe, Derek Fisher, Robert Horry, and Rick Fox had a very solid offensive rating of 114.
- Kobe — individual offensive rating: 114 / usage rate: 30%
- Shaq — 117 / 30%
- Fisher — 115 / 16%
- Horry — 114 / 14%
- Fox — 108 / 10%
Through a measure that Oliver calls “skill curves,” the details of which are best left to another post, he demonstrates that if you change the distribution of how this lineup uses possessions, it significantly affects team performance.
- Kobe — 116 / 25%
- Shaq — 121 / 25%
- Fisher — 90 / 20%
- Horry — 93 / 17%
- Fox — 107 / 13%
With that usage distribution, the lineup’s offensive rating falls to 107 — a substantial drop. The important thing here is that Shaq and Kobe are playing more efficient basketball, but it’s actually hurting their team. Because Fisher and Horry’s offensive rating drops off so sharply with only a slight increase in usage, the entire team suffers when Shaq and Kobe don’t carry more of the burden.
Here’s the crazy thing about Oliver’s example: By looking at Rick Fox’s performance as a member of the Celtics, when he had a higher usage rate, Oliver extrapolates that Fox’s mediocre efficiency holds steady even with extra possessions. So keeping Shaq and Kobe at 25%, if the Lakers just bumped Fox to an 18% usage rate instead of burdening Fisher and Horry, their offensive rating would bounce back to a respectable 112. In this scenario, as Oliver points out, Fox acts a buffer, ensuring that Fisher and Horry can keep their usage rate down even if Shaq and Kobe are having trouble getting their shot off for some reason. The Lakers actually benefit from having a less efficient player in that lineup.
The implications of everything above are huge. For one, it gives a very clear sense of how having good teammates in supporting roles makes a superstar better and vice-versa. Second, it gives us a better way of deciding what adjustments teams should make if they’re struggling. Third, it punches a gaping hole in the seemingly provocative argument that players like Allen Iverson are overvalued — an argument that Dave Berri used to attract The New Yorker‘s interest and sell a lot of books. A.I.’s high usage rate and modest efficiency weren’t ideal, but it was better for the team than dumping more possessions on the likes of Eric Snow. The flip side, of course, is that Oliver can also use his stats to show that Pistons-era Jerry Stackhouse really did shoot too damn much.
Finally, it provides a slightly better mechanism for predicting how players will perform on new teams or in new roles. I say slightly, because the big problem with Oliver’s analysis is limited data. There simply weren’t enough games when Kobe used less than 20% of team possessions or when Derek Fisher used over 25% to say with confidence how the Lakers would have performed if those players switched roles. Oliver admits that his numbers may not have the down-to-the-decimal precision we want, but even in their imprecision they manage to provide a more accurate picture of team performance.
—
* Hollinger’s most recent article, a breakdown of the Suns since the Shaq trade, is actually an exception. I think it’s one of the more informative pieces he’s written, and it doesn’t mention P.E.R. once. Too bad it’s now doomed to Insider irrelevance.
**Leaving aside the details of the math, all efficiency stats follow vaguely the same process. (1) The statistician first acknowledges that point differential is a better predictor of a team’s ability than actual win-loss record. (2) The statistician controls for differences in pace between teams by looking at the numbers on a per-possession basis. The differential is usually expressed as points scored/allowed per 100 possessions, which is both easy to understand and fairly close to the average NBA game. (3) The statistician devises his own way to break down the stats at our disposal to assign how much each one contributes to points scored/allowed. This is where the statisticians diverge, as each one has his own unique method of assigning value to the stats at hand. Debating the merits of each method would take many blog posts to cover. (4) A lot of stat guys then devise another layer of abstraction in the hopes of simplifying their calculations. That is, they try to format their numbers in such a way that they reflect, say, a share of that team’s wins (e.g., Win Score) or the player’s performance relative to the rest of the league (e.g., PER). (5) If the statistician is worth a damn, he reminds the reader that basketball statistics fail to capture a ton of what happens on the court — everything from setting screens to rotating properly on defense to that elusive “chemistry” good teams thrive on.
Blogjammin' - An NBA Blog 