Modeling in Circles

Strange phenomena in NBA player-performance modeling as the Wages of Wins Journal discusses Rashard Lewis:

Lewis is a comb0-forward, which means he has logged minutes at power forward and small forward in his career. WP48 is calculated by comparing a player relative to the average player at his position. Because power forwards tend to rebound at a higher rate than small forwards, power forwards tend to offer higher levels of productivity. So when Lewis is compared to players at the four spot, he tends not to look so good. Relative to small forwards, though, he can be very good.

To illustrate, consider last season. When Lewis played power forward his WP48 was only 0.096, which is close to average. At the three spot his WP48 was 0.209, which is above the “perfect” mark.

This is more than a little perverse. Good power forwards are hard to come by. That Rashard Lewis is capable of performing competently in that role is an asset he has as a player. But thanks to the WoW position-adjustment method, it registers as a problem for his game. If he was much, much worse at playing the 4, he'd never be asked to do it, and his WoW rating would look much better. But in the real world, he'd be a less valuable basketball player.