Is anyone super into "PER"?

[kozlodoev]

Just curious... is there any stat junkies out there who are huge fans of "PER"?  I think most of us don't take it seriously, but do any of you stand by it? 

It's interesting, because Rondo historically has a pretty horrible "PER". 

Last year he was 129th with a PER of 15.34
This year he is 150th with a PER of 15.18

Meanwhile, Brandan Wright is a poster-boy for PER.

Last year he was 13th with a PER of 23.60
This year he is 6th with a PER of 26.18

If ever there was a trade that put John Hollinger's reputation on the line... it was the trade we just saw.  Boston moved arguably one of the most overrated players in the league for one the most underrated.  I can't help but be curious to see what happens here.  This is either going to put the nail in the coffin for "PER" or prove it as relevant.

PER is essentially a one-stop shop for per-minute stats, the way I understand it. So you take per-36 numbers, give them some coefficients, calculate a single number and normalize it so that the mean value for the league is 15. It's not particularly "advanced", other than it tries to reduce the hassle of interpreting a whole box score to looking at a single number.

It is useful for what it is (a quick and dirty representation of player production, but suffers the same flaws your usual per-minute stats do. So I'd say that if this "kills" the stat for you, you probably don't understand how it worked in the first place.

[hwangjini_1]

Just curious... is there any stat junkies out there who are huge fans of "PER"?  I think most of us don't take it seriously, but do any of you stand by it? 

It's interesting, because Rondo historically has a pretty horrible "PER". 

Last year he was 129th with a PER of 15.34
This year he is 150th with a PER of 15.18

Meanwhile, Brandan Wright is a poster-boy for PER.

Last year he was 13th with a PER of 23.60
This year he is 6th with a PER of 26.18

If ever there was a trade that put John Hollinger's reputation on the line... it was the trade we just saw.  Boston moved arguably one of the most overrated players in the league for one the most underrated.  I can't help but be curious to see what happens here.  This is either going to put the nail in the coffin for "PER" or prove it as relevant.

PER is essentially a one-stop shop for per-minute stats, the way I understand it. So you take per-36 numbers, give them some coefficients, calculate a single number and normalize it so that the mean value for the league is 15. It's not particularly "advanced", other than it tries to reduce the hassle of interpreting a whole box score to looking at a single number.

It is useful for what it is (a quick and dirty representation of player production, but suffers the same flaws your usual per-minute stats do. So I'd say that if this "kills" the stat for you, you probably don't understand how it worked in the first place.

thanks to both koz and dos for helping relieve my headache.

so, what i take away, and this is only for me, is that PER as currently imagined sheds less light on performances than we would hope for, while simultaneously not really resolving the issue of how to compare players in a meaningful way.

add in the distinct possiblity of author bias in its creation and its propensity for encouraging lazy thinking...and we have a stat that i will try to steer clear from.

[mgent]

Put it this way, would anybody think a stat for cars/trucks that ranked them by mushing all the categories into one is meaningful?  Of course not, each is built for different things. 

I've heard the built around Jordan thing before too, and I think that's equally dumb.  To use the same metaphor, that's just as bad as Bugatti building a car around 0-60 numbers.  What would be meaningful is if you just made the best stat and you could, and then Jordan just happens to top it.

[Boris Badenov]

The PER model is like any other statistical model - it's been "fitted" to describe the data. It describes some slices of the data better than others.

Now with PER there's a further complication which is that it's user-defined. Hollinger picked the parameters himself, and necessarily so because there is no objective summary measure of player quality (hence demand for something like PER).

Given that it's user-defined, Hollinger surely "fit the model" so that it would pass a sanity test. My guess is that he understood that for most people the sanity test would involve comparing the PER-based top 20 list to their own list. It might also involve comparing the PER-based list of "unappreciated gems" to their own internal lists (hello, Manu Ginobili). So, Hollinger almost certainly tweaked his parameters to dovetail with those external pressures.

What this means is that PER was built to give a pretty good ranking of "typical" star and starter-type players, who all play a lot of minutes. It's also somewhat useful for identifying valuable bench players (though no more so than per-36 stats).

Now, like any model it's not going to be very good at describing or ranking "outliers," and both Wright and Rondo are just that. Point guards who rebound like power forwards, or who shoot 60% from the line just are not going to be well-captured by something like PER, because the model wasn't built to describe/compare players like that. It was built to compare Nowitzki to Wade, and 2008 Lebron to 1988 Jordan.

Similarly, guys who shoot 74% from the floor and have terrific per-minute block numbers, but who also can't get more than 18mpg, are outliers.

No summary model is going to accurately capture guys like Rondo and Wright, because they just lack "comps" which is what any informative statistical comparison requires.

So in sum: PER is flawed, but not much more so than any other summary measure of things that are complicated. And I think if you recognize what Hollinger probably intended it to do - which is provide "sensible" top-50 lists and comparisons of great seasons across players - its anomalies become more understandable.

And to answer the OP, comparing Rondo and Wright based on PER or any summary measure just misses the point because they're so unique. It's like using one of those summary "city quality of life" indexes to compare New York and Boulder.

[Rondohara]

PER, TS% are junk stats.
Heck even old EFF is better in some ways.

But now we have per possession stats, point per shot, full shotchart, defense dashboard, all the other player tracking data, and a multitude of available stats that have meaning, anyone that still talk about those in 2014 has problems.

[fairweatherfan]

I don't know why PER is getting so much criticism over its evaluation of Wright, when a commonly touted alternative - Win Shares - is even more favorable toward Wright, ranking him as the 3rd (!) best player in the league on a per-minute basis, behind only Davis and Curry.

...that's not to say that PER is better than Win Shares - it's almost certainly not - just that PER is far from the only advanced stat to have Wright as a huge outlier.

[Chris22]

PER works pretty well for players who get a lot of minutes.

http://insider.espn.go.com/nba/hollinger/statistics

[D.o.s.]

This is either going to put the nail in the coffin for "PER" or prove it as relevant.

This, by the way, is wrong. ESPN's not going to give up one of it's proprietary stats simply because of this trade.

Just curious... is there any stat junkies out there who are huge fans of "PER"?  I think most of us don't take it seriously, but do any of you stand by it? 

It's interesting, because Rondo historically has a pretty horrible "PER". 

Last year he was 129th with a PER of 15.34
This year he is 150th with a PER of 15.18

Meanwhile, Brandan Wright is a poster-boy for PER.

Last year he was 13th with a PER of 23.60
This year he is 6th with a PER of 26.18

If ever there was a trade that put John Hollinger's reputation on the line... it was the trade we just saw.  Boston moved arguably one of the most overrated players in the league for one the most underrated.  I can't help but be curious to see what happens here.  This is either going to put the nail in the coffin for "PER" or prove it as relevant.

PER is essentially a one-stop shop for per-minute stats, the way I understand it. So you take per-36 numbers, give them some coefficients, calculate a single number and normalize it so that the mean value for the league is 15. It's not particularly "advanced", other than it tries to reduce the hassle of interpreting a whole box score to looking at a single number.

It is useful for what it is (a quick and dirty representation of player production, but suffers the same flaws your usual per-minute stats do. So I'd say that if this "kills" the stat for you, you probably don't understand how it worked in the first place.

thanks to both koz and dos for helping relieve my headache.

so, what i take away, and this is only for me, is that PER as currently imagined sheds less light on performances than we would hope for, while simultaneously not really resolving the issue of how to compare players in a meaningful way.

add in the distinct possiblity of author bias in its creation and its propensity for encouraging lazy thinking...and we have a stat that i will try to steer clear from.

Well it definitely did one thing -- it got Hollinger a job at ESPN.

[The Oracle]

  The NBA's tracking data is game changing!  It is exposing once thought good defenders and the like as they truly are!  It is also exposing the lazy and will give praise to the effort guys!  PER was never a stat that could measure any of this and completely misses the boat on role players!  There will be many studies done from this data that show who the real contributers are and those that are not!
   Whats much worse than PER is when people use opponent PER and try to measure whether they are outplaying their opponent!  This data is so intermixed you can't get anything from it!  For example Rondo mostly guarded shooting guards and Bradley mostly guarded PG's but the sites will mark them the exact opposite and assume PG's guard PG's and such!  This makes Bradley look awful and Rondo much better than he should!  PER should just be disreguarded as it is inaccurate and obsolete!

[Beat LA]

While I think PER has some merit, I just don't think advanced stats work for basketball like they do for baseball.  There are way too many variables in basketball to quantify players with any kind of equation.

Well said, TP.  I'm a numbers guy, myself, but sometimes it just gets to be too much, and all these stats seem to do is to cause paralysis by over-analysis.