Print and Go Back TrueHoop [Print without images]

Tuesday, October 9, 2007
John Hollinger's Explains PER, and Wants You to Join his Cult

ESPN's statistical wizard John Hollinger just released a massive trove of player projections for the upcoming season. He was nice enough to answer some questions about his work.

People get crazy about your statistic: PER. Some are hypnotized by it, and treat it like religion. Others can't believe anyone would pay it any mind. There has been a lot of internet discussion about it lately. Can you give me the layman's explanation of what PER is, and how it should best be used?
PER is absolutely a religion and should be treated as such, with me as the all-knowing maharaja whose authority is unquestioned. OK, not really. But that would be so cool if it happened.

Anyway, PER is a per-minute rating of a player's statistical productivity. It's great for measuring a player's "tangibles"; the one area it struggles is with defense because so much of that remains a black box for the analytical community.

I use it so much because it makes it very easy to make comparisons between players who play differing minutes, or in different systems or what-not -- comparisons which, using conventional stats, are almost impossible.

We're not saying it should be used in a vacuum, right? It's not like you're saying teams should fire all their scouts.
Firing all the scouts and the aforementioned maharaja idea would both be wonderful for me, actually, but for the teams I wouldn't recommend it. PER is a great way to evaluate a player's *statistical* production, but that still leaves hundreds of elements that you can't deduce without watching players on the court and, perhaps more importantly, off it.

For instance, here's one crucial one that comes to mind: How likely is this guy to keep working and improving that PER in coming years?

What was wrong with the old numbers?
You can't make comparisons with them. Points per game is so heavily context dependent that it's really not very useful, and the other per-game averages have similar shortcomings.

You can use them to tell that Kobe Bryant is a better player than Scott Padgett (as if you'd need stats to make that brilliant deduction), but if you try to compare Richard Jefferson to Lamar Odom you're not going to get very far.

If PER is a better measure of a player's value than points and rebounds and all that, wouldn't it be cool to use PER as a basis for a new fantasy basketball game?
I've found that fantasy rules value things so much differently than PER does that using PER for fantasy is essentially useless. Well, unless you made PER the only stat for the league and based everything off it. If I become the leader of a PER cult in Guyana I'm totally doing that.

I have always wondered: what's the physical process like when you do these rankings? I assume you have all these numbers in a computer? How do they get updated? And is the whole thing laborious? I mean, if I were to give you box scores from my rec. league, and ask you to determine the PER of all the players, would it take long? (By the way, you could probably make a lot of money selling that kind of stuff in rec. leagues, come to think of it.)
Well, ESPN has automated the process now so folks can see it every morning without me lifting a finger. But yes, I just download the stats, drop them into an Excel sheet, and voila. It can be very fast if everything is formatted correctly.

Alas, it's probably not useful for a rec league unless somebody is tracking things like opponent turnovers and the league average pace factor.

Now, today we are seeing your PER projections for the upcoming season -- as in, here's a very refined statistical analysis for something that hasn't even happened yet! People are going to think you're just sitting in your basement saying "lookie here, I'm feeling like Yao's going to be about a 27 this season." But that's not how it works, right? Can you explain the science?
Of course that's not how it works -- I don't have a basement. Come to think of it there are one or two other differences.

Here's how it really goes down: I take a player like Yao Ming, and compare him to all the players who were his 2006-07 age from the past 20 years, and find the ones who were most similar statistically over the previous three-year period (and yes, height is one of the "stats").

Then I see how those players did the following season, weighting the results of the most similar players the most heavily, and apply those gains or losses to Yao's stats from the past season. And in Yao's case, what I find is that the players who were similar to him improved sharply at his age, and that's why his projection is so positive for this year.

Are there any players out there that you never really noticed or cared about, but then saw they had surprisingly good PER numbers? Or have you pretty much gotten to the point where you can see good PER in real time?
For the most part it's the latter, but there are exceptions. Like Antonio Daniels. For some reason every summer when I sit down to write I'm like, "Wow, his PER was that good?"