ESPN NBA

In today's bullets, I wrote:

ESPN's Bill Simmons predicted that one day a stat geek would declare that Kevin Garnett was as good as Tim Duncan. That day is today.

In the second link, David Berri, lead author of "Wages of Wins," makes the case for Garnett. After a lot of statistical stuff that you should read, Berri concludes:

The moral of this story is that players do not win games. Teams win games. If you surround an exceptional talent with other productive players, your team might contend for a title. But if you surround an exceptional talent with unproductive talent, your team has a good chance of missing the playoffs entirely. And there is nothing one exceptional talent can do to change that basic fact. 

I just got a very interesting email, though, from a basketball fan who is also a marketing professor who does applied econometric research at the University of Southern California. Kenneth Wilbur writes:

Duncan tends to stand closer to the basket while his teammates have the ball. Garnett usually stands farther away.

Thus, when a teammate drives for a lay-up, Garnett's defender is better able to help defend against the driving offensive player. This makes Garnett's teammates more likely to be going 1-on-2, while Duncan's teammates are going 1-on-1. Duncan's defender is less likely to help, since Duncan is just a short pass away from a high-percentage dunk attempt.

Win Score, PER, and other systems do not observe where a player stands when other players attempt to shoot. Thus they may misattribute performance among individual players.

There are many other unobserved aspects of play that could have similar effects. For example: willingness to set a pick, deliver a hard foul, move without the ball, and many aspects of team defense. People who know basketball would probably agree that Duncan is better than Garnett in many of these areas.

Some suggestive evidence is available from changes in teammate performance when they join or leave Garnett's or Duncan's teams. Players' roles and abilities change over time, so take this with a grain of salt. But you can see some evidence for Simmons' argument.

Marko Jaric
.414 FG% w/ Clippers in 04-05
.399 FG% w/ T-wolves in 05-06

Mike James
.469 FG% w/ Raptors in 05-06
.432 FG% w/ T-wolves in 06-07

Ricky Davis
.464 FG% w/ Celtics in 05-06
.429 FG% w/ T-Wolves in 05-06

All three are "slasher" types who are potentially affected by the unobserved positioning variable discussed above.

The larger point is that Win Score, PER, and the rest of current-generation basketball statistics should be seen as a complement to informed basketball observation, but they ARE NOT a substitute for experts' basketball observations. (Though they are may be better than many casual fans' basketball observations.) It is possible to correct the deficiencies in these statistical measures, but doing so will require much better data and a more sophisticated class of models.