John Hollinger's PER: What a Crock!

by Wayne Vore (ATS) on Nov 6, 2008 2:27 PM CST

Ok, crock is a strong word for it, he seems like a cool guy and all. BUT, and this is a big BUT, his stats have absolutely nothing to do with playing winning basketball or assessing whether or not a player is contributing to winning a basketball game.

How do I know this? Because I am a Spurs fan. Right now, the Spurs are playing terrible basketball. They are 1-3. They got blown out once. They were very fortunate to win the one game they have one. Despite that, here are John Hollinger's PER rankings for our Spurs:

Tony Parker - PER ranking #1 - he clocks in a at a stunning 39.41. Next closest is LeBron way back in the 32's.

Tim Duncan - PER ranking #9 - 28.68. Just to get perspective on Tony's numbers. There are six players in the 3 point range between Tim and LeBron.

Roger Mason - PER ranking #25 - 23.21. And 15 between Tim and Roger's 5.6 point difference.

Amazingly, we have a fourth Spur in the top 25:

George Hill - PER ranking #22 - 23.51 (just ahead of Steve Nash)

The Spurs have 4 of the top performing players in the league. Hah!

I'll tell you right now, Tony, Tim, Roger, and George could play like they have the first 4 games for the entire year, but if the Spurs play the same defense then we'll be under .500. So, can we just throw all the PER talk away? It doesn't really mean anything other than you are effective putting up stats that don't necessarily contribute to winning. Me, I'll take Bruce Bowen and his 5.74 (44th among small forwards) and defensive effort.

Tweet Comment 17 comments | 0 recs |

Do you like this story?

Comments

Metrics seem to work better for baseball

So far anyways….it seems the past few years, people have been trying to find new, esoteric ways to quantify basketball players.

Maybe PER is more of an offensive stat?

Okay, just so I understand it... in your wildest fantasy, you are in hell. And you are co-running a bed and breakfast with the devil.

by bren on Nov 6, 2008 2:54 PM CST reply actions

I would agree that metrics work better for baseball. I think it is because baseball is dominated by the pitcher-batter relationship. The ability of the defensive players, baserunning skills (not to be confused with stolen bases), etc. is relatively inconsequential compared to the pitcher-batter. Especially over a 162-game schedule.

I think the baseball equivalent to PER would be if they only had a radar gun reading as a stat. It’s impressive. Higher is better, in general, but hardly the whole story.

Basketball doesn’t really have any good individual defensive statistics. Not ones that matter anyway. There are good team stats like: defensive rebounding % (I think the Spurs lead the league last year) and defensive FG% (Spurs are always top 3).

I’m no statistician, but I would like to know what the “fit” for Hollinger’s PER is to wins. The basic idea is that you would plot PER vs wins and see how well they correlate.

Just to give you an example — and an example of how geeky I am — me and a buddy analyzed our softball team’s statistics. I kept our individual and team OBA for 3 years. A total of just over 100 games. My buddy and I got in an argument about how important taking the extra base is. I stated that going 4-4 with 4 singles is better than 3-4 with a double, triple, and homer. So, we crunched the numbers and we got a 90% fit for OBA vs Runs. You interpret that to mean that 90% of all run production is directly attributable to OBA. This is an off the charts correlation which means you don’t even look any further for other factors.

by Wayne Vore (ATS) on Nov 6, 2008 4:11 PM CST up reply actions

Well, one could argue that if there’s any game in which the sum of the PER ratings of the players from the losing team is higher than the sum of the PERs of the players in the winning team, then it’s a flawed stat. But it’s all a matter of intent. Is PER supposed to reflect the team’s performance to that degree? I know there’re several ways in which the performance of the team affects the stat, but it’s impossible to find a complete formula for that pesky PER.

That’s what bothers me about it, or at least one of the things. I don’t have Hollinger’s books, and although I understand he wants to sell them, it’s difficult to judge its worth without the formula. And more than the formula itself, it’d be interesting to analyze the parameters he chose, and their origin. Weighted stats can offer wildly different results because of small changes in those parameters, and honestly, in my mind I can’t imagine how he performed the necessary statistical analysis to obtain them. What’s more important for a team? An assist? A rebound?

And as you said, what’s probably its biggest flaw is how it fails to take defense into account. That’s half the friggin’ game, isn’t it?

I honestly like Hollinger, really. He writes a lot, has some sound articles and actually sounds impartial all the time. I respect that. My problem with him, as I said in the reply to bones’ recap, is that he tends to lose himself in his own numbers, and follows them blindly even though some good ol’ basketball savvy could cast some doubts on his conclusions. Like last season, when he put Manu in the All Decline Team. But what really, really irks me is that he uses his own (suspect) stats to justify his opinions. “I think this. PER agrees. So it’s true.” That’s pretty much how I read most of his opinions, and there’s not a better example I think than yesterday’s Daily Dime (linked also in my reply to bones’ post), where he used a brand new stat to “prove” that Amare’s night was better than Tony’s. Eh? Really? Can’t help but wonder what the individual PER for the night was for Tony and Amare… Wouldn’t it be ironic if PER disagreed with that new stat?

Anyway. PER is okay. Just another fun stat to throw around… but it’s not the end-all of basketball analysis. Hollinger disagrees.

by LatinD on Nov 6, 2008 11:56 PM CST up reply actions

The “fit” term I was looking for is called r-squared. It’s a regression analysis type thingy.

by Wayne Vore (ATS) on Nov 7, 2008 9:13 AM CST up reply actions

i think all stats in basketball are based more on offense than defense. they track rebounding, steals and blocks, but that doesn’t really tell the whole story, hence marcus camby’s defensive player of the year award.

but hollinger has stats on offensive and defensive efficiencies that seem to be a pretty good standard. i think hollinger even said it himself in a PER diem column, tony and tim are playing at an elite offensive level, but still the spurs can’t seem to win (this was before last night). he’s been knocking our defense, too, which is dead last in his defensive efficiency ranks (as of a few days ago).

you ain't a beauty but hey you're alright.

by kalone on Nov 6, 2008 4:11 PM CST reply actions

Did you know that 56% of all statistics are made up? How do I know? I read it on a wet-nap at Hooters. If Hooters says it, it must be true.

"Well if I am truly crazy, don't you know I like my life that way"

by SinCitySpur on Nov 6, 2008 4:24 PM CST reply actions

I respectfully disagree

PER is a pretty good tool for evaluating how well and individual is playing offensively. It’s far better than, say, points, or even shooting percentage. Of course it doesn’t tell the whole story, but put it another way: who in the league is playing better offensively (through 4 games) than Tony? TD is tearing it up, as is Mason; the problem is that they are not stopping anyone, and they’re not getting any production at all from anyone else on the team.

As for George Hill’s PER being better than Steve Nash: that is like a minor-league call up going 3-4 with a homer in his 4 pinch-hit at bats. His stats will say he’s better than A-Rod, but it’s highly unlikely he’ll keep up that pace for the whole season.

IMO, Hollinger, Charley Rosen, and to some extent David Thorpe, are the only basketball columnists worth reading (maybe the Wages of Wins guys too). They seem to have a far better handle on what actually wins games than dumbasses like Chris Sheridan, Mark Stein, JA Adande, or Screamin’ A. Despite the fact that those guys watch tons of basketball, they seem incapable of anything but the most superficial analysis.

Catch the Spurs Spirit! It's a Fast-breakin' Fiesta!

by tomasito on Nov 6, 2008 9:18 PM CST reply actions

Evaluating any stat based on a sample size of 4 games is dumb.

PER is used to assess overall individual accomplishment. There’s no better single stat that I know of.

And Bruce Bowen sucks.

by sungo on Nov 6, 2008 10:58 PM CST reply actions

I meant to say “There’s no single better stat that I know of that does a better job of doing what it does.”

(Like that clears anything up.)

by sungo on Nov 6, 2008 10:59 PM CST up reply actions

This would be much clearer:

“Of all the stats I know, there is no stat of which I know that does a better job of doing what it does in knowing what I know.”

chaos... panic... pandemonium... my work here is done.

by rick2g on Nov 6, 2008 11:02 PM CST up reply actions

You left out a “does” and you need to close the ) or the of statement won’t work right.

by sungo on Nov 6, 2008 11:09 PM CST up reply actions

I see them in my sleep...

… man… you… you just don’t tease a lisp programmer about parentheses, man… you just don’t do it. that ain’t cool, man.

chaos... panic... pandemonium... my work here is done.

by rick2g on Nov 6, 2008 11:36 PM CST up reply actions

There are no lisp programmers left. They died with the computer dinosaur known as Symbolics. IF, in fact, you are still programming lisp for work I bet I have a 50/50 chance of naming the company you work for.

by Wayne Vore (ATS) on Nov 7, 2008 9:06 AM CST up reply actions

Ah… but there are still many derivatives of lisp in wide use, and countless University CS divisions out there are replete with aging and bitter professors still trying to push lisp on a younger generation.

Pure lisp, tho… nope. They’re dead. They’re all dead, and shalt recurse no more.

chaos... panic... pandemonium... my work here is done.

by rick2g on Nov 7, 2008 9:47 AM CST up reply actions

I’d wait at least 10 games into the season before anybody starts pulling out stats. The only important numbers that come out of four games are 1 and 3

by r21x on Nov 6, 2008 11:13 PM CST reply actions