I have never said this before, but I must: Nate Silver is wrong.
First, some background. (Non-sports fans, you are warned)
Carmelo Anthony is either an elite player in the NBA, or he isn’t. This year, Carmelo Anthony is paid $16 million — that’s elite player pay.
Advanced NBA stats, such as David Berri’s Wins Produced metric, generally rate Anthony much lower, however. These wins, when added together, correlate with team wins. In the Wins Produced system, an elite player generates >0.300 Wins Produced per 48 minutes. (Lebron James, for instance, has 10.9 WP with 0.345 WP48.) Carmelo Anthony has 3.5 WP and a 0.151 WP48. Thus, Carmelo is not worth his contract. (See Berri’s post for more)
Nate Silver, a god among men, disagrees. He notes that other players, when playing with Carmelo, improve their shooting efficiency. That is to say, Carmelo makes his teammates better.
The problem, of course, is that if Carmelo is inefficient enough, the improved efficiency of his teammates is not enough to make up for the possessions-shot-attempts hole of Carmelo.
The reason I’m bringing all this up at all, however, is the link with the increasing use of “value-added” teaching statistics, as well as the increasingly quantitative approach to occupational statistics. Organizations are depending on numbers to help clear things up for them, to help identify the “good” teachers and the “bad” teachers.
Worse, unlike sports, teaching (or lawyering, for that matter) is not easily quantified with winners, losers, and points. And even in a sport like basketball, with its copious data-taking and constant reassessment on every sports radio station across the country, cannot come to a consensus on the productivity of its best players when each win is worth about $1.6 million. And let’s be clear, just because a statistic is advanced — that is to say, it amalgamates a bunch of other individually observed statistics — doesn’t mean it’s any good for predicting next year’s performance.
This is not to say that we should not use advanced stats. We should use them and they are important. But we should not be so blind as to believe that statistics will tell us everything about the employee productivity. Generally, stats may get the story right, but overvaluation and undervaluation will come out of any statistical system.
Carmelo is still not that good, though.