Player Efficiency Rating is one of the oldest advanced NBA stats, devised by John Hollinger and popularized by ESPN back in the early days of the league’s rise out of the Dark Ages in the mid-aughts.
In theory, it takes the totality of a player’s contributions on the basketball court, distills it into one number, sets the league average at 15, and is sufficiently position-agnostic to make it possible to compare guards to centers the way you compare catchers to shortstops by Wins Above Replacement in baseball.
In practice, it’s been all but completely superseded by Win Shares (and its derivative WS/48 minutes) and Value Over Replacement Player.
Which is a good thing. PER has never effectively translated to team wins the way those other two stats have—just in the past week, I’ve demonstrated on this here site that WS and VORP correlate amazingly well with team success to the point where those two advanced stats are nearly as reliable as the actual score of each game in determining how a team did over 82 games.
To wit, let’s do a little experiment, shall we?
Let’s take the 236 NBA players who played at least 1,230 minutes in 2018-19 (15 minutes x 82 games), put their PER and WS/48 through a simple calculator of correlation coefficient, and see what we get.
Unlike those VORP and WS team pieces, I’m not going to list the raw data—I’ll just link to the Basketball Reference page I got it from and let you look at it yourself if you want to see.
Anyway, here’s what the calculator shot back. An r-squared of 0.677.
Now granted, that’s not insignificant. It does demonstrate that as PER goes up, so does WS/48. And since we’ve learned that WS/48 and VORP are correlated strongly, it then stands to reason that you will get more wins with high-PER players than low-PER players.
Further proof of this is easily found in noting that the top 5 guys for PER are Giannis Antetokounmpo, James Harden, Anthony Davis, Karl-Anthony Towns, and Nikola Jokic; the stat isn’t so useless that it produces “top guys” who aren’t, y’know, top-tier NBA players.
But the goal of any advanced stat should be a super-strong almost 1:1 correlation with wins on the actual basketball court. If it’s not that, it doesn’t have predictive value and therefore doesn’t belong in a serious analysis of the game—the entire point of the analytics revolution.
We know—because we tested and proved it—that minute-weighted Win Shares are all but synonymous with actual team wins. The r-squared on that was 0.963.
Consider it this way…as a winning percentage, .963 is a 79-3 record. .677 is 55-27. Good, but it wouldn’t even get you home court in the second round in the East last season.
And because of that, it’s time to retire PER from the discussion. If you want to say Giannis or Harden or the Brow are good, you can just as easily point at their WS/48, so you should do that instead.