View Full Version : Calculating Wins Produced And What It Implies

01-07-2014, 04:13 PM

For the most part, player performance in the NBA does not depend on the player’s teammates. This can be easily seen when we consider how consistent a player’s P48 is across time. From 1973-74 to 2010-11, about 80% of a player’s P48 in the current season is explained by the player’s P48 in the past season (from a sample of nearly 5,000 player observations; players needed to play in at least 20 games and have at least 12 minutes per game in season to qualify for sample). And as noted in Stumbling on Wins, players in the NBA are much more consistent than what we generally see in the NFL or Major League Baseball.

Although players are consistent, it is the case that we do see some interaction effects in the NBA (as noted in The Wages of Wins and academic articles that were written well before The Wages of Wins appeared). To ascertain these effects, the following model was estimated:

player per minute statistical production (i.e. defensive rebounds per minute) = f(per minute performance in stat the previous season, age, age squared, percentage of games played last two seasons, dummy variable for position played, dummy variable for new coach, dummy variable for new team, dummy variable for year, stability of roster, and the teammates’ per-minute production of statistic).

This model allows us to see how a teammate’s production of a given statistic impacts a player’s per minute performance (after controlling for other factors that explain performance).