3) We’ll try to be delicate here but why is his system inferior in your opinion to your own?
These two questions are related — and I will try and be diplomatic.
When I started looking doing economic research with NBA statistics I needed a measure of player performance that connected what a player did on the court to team wins. Although measures like PER — and NBA Efficiency — are popular, these measures are not highly correlated with team wins (as I have shown in published research).
So, I set out build a model connecting a player’s box score statistics to team wins. The result of this effort is Wins Produced, a measure that translates all the numbers we see in the box score into each player’s production of wins.
The big difference between Wins Produced and PER is that the former is based on a model connecting each factor in the box score to team wins. In other words, the value of factors like offensive rebounds, steals, or turnovers is based on a statistical model connecting each of these variables to how many wins we observe.
When Hollinger describes how PER was constructed he doesn’t really argue that the weights he assigned reflects how that particular factor impacts outcomes (and he certainly doesn’t provide evidence that his weights reflect actual outcomes). What he does is choose weights that seem to make sense to him. And what seems to make sense to him leads to a measure that confirms what people tend to think about NBA players.
As a result, we see two clear problems with PER.
• A player who shoots better than 30% from two point range (and even less from three point range) can increase his PER by simply taking more shots. So PER rewards inefficient shooting.
• Consequently – because inefficient shooting does not translate into wins — measures like PER do a very poor job of explaining team wins.
As a result, it doesn’t appear that PER is a very good measure of a player’s actual contribution to outcomes.
PER is highly correlated with Game Score (Hollinger’s simple box score metric) and NBA Efficiency. Each of these measures overvalues inefficient scoring and does a poor job of explaining wins. But all three do a very good job of explaining a free agent’s salary (something also noted in published research).
This indicates that PER is a statistical measure that captures perceptions of value in the NBA. For example, PER tells us that players like Allen Iverson and Carmelo Anthony are really very good. And therefore both should get paid large sums of money (and Hollinger once thought that these two combined on the Nuggets would lead to a title contender). That's why John Hollinger is a cunt.
Bookmarks