how useful is PER?

[Jimmy76]

In your view how effective is PER at measuring a players contribution to winning

[supersub15]

PER should be stripped of its so-called defensive metrics because it tells us absolutely nothing about defense.

For instance, you can steal the ball, but it doesn't mean that you're playing team defense. Just ask Dikembe how many times he had to bail out Allen Iverson when he gambled on steals. There are things that the NBA should start measuring, like charges taken, ball deflections, jumpballs won, etc. Then, you can add those to PER, and it could become a more complete metric.

So, if we go by oPER (offensive PER), then we can at least measure a player's offensive output, without worrying about the other side of the ball.

My 2 cents.

[tsherkin]

Is PER as useful as offensive win shares or ORTG, though? Win shares seem overly dependent on team success (as they are limited by the number of wins a team had in a given season), and ORTG is a little ambiguous.

PER isn't that valuable; it's entirely too sensitive to per-minute production, which can be very misleading. It's also over-sensitive to volume scorers.

I'll give it this: if you adjust for actual minutes played, it stratifies reasonably well; you don't find a lot of guys it misses. But it's only a peripheral tool for casual assessment; it doesn't tell you anything other than your guy loads the box score, but this is also a system that annually makes ridiculous projections and frequently features puzzlers in the rankings because it's taking a 20 mpg player and ranking him up with the production of honest starters because it overvalues per-minute production.

PER is a candy stat, like most all-in-one stats. I don't think those types of stats are especially illuminating.

[turk3d]

I find that per is a useful tool for measuring ones potential. It identifies players who may not be getting as much minutes as perhaps they should. Of course there may be other factors (playing against opponent backups instead of starters, playing less meaningful minutes rather than during clutch times and perhaps a few others). It is, however, a good way to measures some of the younger, non-starters in determining whether or not they merit more playing time. Other than that, I do think people may go overboard on its true value.

[tsherkin]

If you're looking for potential, ask The_Secret_Weapon about his Diamond Rating.

[TrueLAfan]

^^ Totally agree about The_Secret_Weapon.

How useful is PER? It depends on what you mean by "useful." Can it identify which players are truly great? Yes?

But...stop and think. How hard is it to spot a great player? Does anyone look at Chris Paul or Dwyane Wade or Dwight Howard and say, "Not that good"? As far as I can tell, the primary argument for PER is that it can "tell" you when a player is good. But that isn't hard. As I like to say, it isn't pulling a rabbit out of a hat...it's pulling a rabbit out of a box of rabbits. I came up with a system a couple of years ago that pretty accurately identified top players. I came up with it in about two minutes in the shower, and modified it for about five minutes afterward. It isn't hard to do.

But identifying the obvious and being able to differentiate between level of goodness, sorta goodness, and mediocrity...that's a totally different thing. The logic behind claiming PER is extremely effective goes something like this: "Well, if you can spot what's big and obvious, you can see what's smaller and less obvious." Sadly, that's not true. It's why we go to the optometrist and get glasses.

And I'm not necessarily sold on PER's ability to identify and clearly compare/delineate players that good or above average. Zach Randolph was better than Steve Nash last year (or any year)? Andris Biedrins was better than Carmelo? Nate Robinson was better than Chauncey? Erick Dampier was better than Artest? Hakim Warrick as better than Lamar Odom or Rashard Lewis? Really? Those aren't isolated instances. There were only 64 players that played 2000 minutes and had a PER of over 16 last year. They don't seem to be compared all that well. Derrick Rose, John Salmons, Ron Artest, Al Harrington...they don't even make that PER cut. Those guys were below average starters last year. (The average PER for starters is almost always over 16.)

Then there's the inability to compare between eras/years/ PER is zeroed out at 15 each year. A player who had a PER of 15 (or 25 one year) may be better or worse than a player who had a PER of 15 (or 25) in another year based on whether the system thinks overall play has improved or declined between the two years.

I figure PER is possibly, marginally useful in noting/comparing elite players and becomes less so as the list gets longer. And doesn't work between years or eras. So it's better than flipping a coin, but isn't a real comparative tool. If you're not a particularly good dart thrower, I guess I'd say it's somewhere between throwing darts and a Magic 8-Ball.

[panth181]

I think you are wrong on the year to year part - this is one of the things that Hollinger set PER up to be able to do.

15 is the league average every year, and the system makes no effort to differentiate between levels of play in various years. So while you may have an opinion on one year being stronger than another and therefore a 15 in one year being better than a 15 in another, PER does not.

[tsherkin]

That would be correct BUT...

aPER = (pace adjustment) * uPER

PER = aPER * (15 / lg_aPER)

PER is normalized against 15 and league-average PER. League-average changes year to year.

[Don Draper]

Still to this day I don't understand why fans love PER. If you use PER you might as well you NBA Efficiency or Game Score, because they will basically give you the same results. Also if you wanted to use oPER you are better off using Offensive Rating. Lastly, combining offensive and defensive ratings doesn't make any sense. PER makes the assumption that for every player, individual offense and defense hold the same weight, which is nothing more than an assumption. When people use PER as a source, I generally take what the have to say with a grain of salt.

[charrua]

I find it pretty useful, in fact. It depends on what do you want to do. Do you want to see which guys drafted in 2006 are doing well right now? PER can do it. Do you want to see how players's production declines with age? PER can do it. Do you want to see which guy may deserve more minutes? PER can do it.
It's not a perfect measure of player productivity, but it's reasonable enough.
It has weaknesses, of course. As all boxscore based systems, defense is poorly measured. It assumes a certain value to shot creation that may be debatable. It assumes that player productivity remains the same regardless of minutes played. Etc. etc.
I personally use it because: a) its formula is well known, so you can dig into the data and see the reasons for the numbers (while APM, for example, is a black box). b) it accounts for efficiency AND usage (while ORTG, for example, only accounts for efficiency) c) stability. Its results are reasonably stable over time. Look at Gasol Winshares, for example, and they are all over the place. 7 wins, 12 wins, 8 wins, 13 wins, etc. And Gasol is a guy with an unusually stable production. d) ease of use. This is laziness on my part, but being boxscore based and widely available IS an advantage.
As I said, it depends on what do you want it for. I would never consider it the "final word" on a player, but if you want a quick and reasonably accurate look at who is doing well, it works.

[panth181]

That would be correct BUT...

aPER = (pace adjustment) * uPER

PER = aPER * (15 / lg_aPER)

PER is normalized against 15 and league-average PER. League-average changes year to year.

Yes, league average would change every year. But this accounts for league wide changes in offensive efficiency. IMO, it would be better to compare a player from the mid-90's and a player from the mid-80's to their respective league averages than to each other unadjusted. Different times, different games. I'm not saying there are changes in quality of play over time, but it seems to me that adjusting to the league average is the best way to compare across eras...

[5DOM]

There are things that the NBA should start measuring, like charges taken, ball deflections, jumpballs won, etc. Then, you can add those to PER, and it could become a more complete metric.

That would be nice but weighting them would be really hard.

[tsherkin]

Yes, league average would change every year. But this accounts for league wide changes in offensive efficiency. IMO, it would be better to compare a player from the mid-90's and a player from the mid-80's to their respective league averages than to each other unadjusted. Different times, different games. I'm not saying there are changes in quality of play over time, but it seems to me that adjusting to the league average is the best way to compare across eras...

I'm only saying that for the purposes of comparing across eras, it makes for trouble. Comparing two guys in the same season, sure, but that's it. I think if the difference is significant, which it usually is (especially given the way stats were recorded before 74), then that causes problems.

[panth181]

Are you saying that comparing to the league average would be less valid than using unadjusted statistics to compare players from different era's?

[mopper8]

PER should be stripped of its so-called defensive metrics because it tells us absolutely nothing about defense.

For instance, you can steal the ball, but it doesn't mean that you're playing team defense. Just ask Dikembe how many times he had to bail out Allen Iverson when he gambled on steals. There are things that the NBA should start measuring, like charges taken, ball deflections, jumpballs won, etc. Then, you can add those to PER, and it could become a more complete metric.

So, if we go by oPER (offensive PER), then we can at least measure a player's offensive output, without worrying about the other side of the ball.

My 2 cents.

I agree with this assessment entirely (as well as TrueLA's). PER is logically incoherent. It accounts for "measurables" that aren't necessarily valuable barometers of player value while ignoring other 'measurables" in the same vein that might well be of value (i.e. deflections, charges drawn, etc).

PER basically takes all the statistics we're given in a standard NBA boxscore, standardizes them across minutes and pace, compiles them into one metric that's then set as a true number based on comparisons to other players. As such it's only really valuable insofar as the information in a box score is valuable. Of course, I think we're all aware that box scores are entirely inadequate at gauging player value...as such, it follows the PER is too.

I'd also add that it's very limited in accounting for the value of being able to play multiple positions. Even ignoring all the above about PERs limitations, it still has problems putting any value on versatility. For example, say Dwyane Wade has a PER of 29.5 at SG and 25 at PG. Say, too, that a PER of 25 would at PG would make him the 2nd best PG in the league (based solely on PER) when he's playing at that position. Say, as well, that Kobe Bryant has a PER of 29 at SG and plays exclusively at SG...no other position. Say also that Chris Paul has a PER of 28 at PG and plays exclusively at PG. Finally, let's assume that D-Wade plays enough PG in this thought experiment that his overall PER turns out to be 27.5.

Now, in this scenario, even granting that PER is an accurate barometer of a player's performance in comparison to other players, we'd say that in the above situation, PER gets it wrong. In the above world, D-Wade is a better shooting guard than Kobe Bryant based solely on PER when at SG, and yet he is being punished in his overall PER for his versatility....for his ability to swing to the point and be the 2nd best PG in the league. Meanwhile, Chris Paul also has a higher PER than D-Wade, even though Wade, based on PER, is a near-comparable PG and a superior player overall when at SG, and has the ability to play both spots, unlike Paul.

I think it's important to note that I made those numbers up and the rankings/names are near-arbitrary, and a discussion of how those 3 stack up against one another belongs in another thread, as I'm not trying to make any actual claims on their comparative values...just demonstrate yet another failing of PER. You could do the same thing with Lebron at the 3 and 4 vs a strict 3-man and a strict 4-man. Or whatever.

[tsherkin]

Are you saying that comparing to the league average would be less valid than using unadjusted statistics to compare players from different era's?

I'm saying that if you've got a stat that compares a player against the league average in his own season and you have two players from vastly different years compared against one another, you can draw relative measure against league-average competition in-year, but have little clarity for the direct comparison.

Like this:

Player A's PER is 18, and league average is normalized to 15, but it's a different league average than Player B, so even if his PER is also 18, is that a really meaningful comparison?

Unadjusted stats don't help much either, but I'd rather use a host of little stats than PER, because PER is a weak stat even if the era normalization DOES work.

[musiqsoulchild]

New to this forum. Long time poster on the Bulls board and occassionally on the Knicks board. Hi to all!

My two cents on this are:

PER is like salary. So, if A makes 50,000 USD and B makes 100,000 USD it may feel like B makes bank compared to A.

However, this ignores all context:

1) How long was B an employee as compared to A?
2) What are the median salaries in the 2 industries that A and B come from?
3) Who pays more taxes?
4) Does B spend more than A to an extent that effectively the person with the lesser salary saves more money after bills and expenses?

All important questions.

But, like PER, the guy who has the 100K salary is probably pulling in more girls in the weekend than the guy pulling in 50K.

Meaning, PER is a glittery stat. It will do alrite for a one night stand. You want a more meaningful relationship? Screw PER. Metaphorically, that is.

[fancynapkin]

The biggest problem with it is that it tries and fails to incorporate defense. It would be much better if it were just offense and didn't pretend to be all encompassing.

[blabla]

The biggest problem with it is that it tries and fails to incorporate defense. It would be much better if it were just offense and didn't pretend to be all encompassing.

That's a good point. Though, at the time of its' invention there weren't many (even the simplest) +/- statistics in existence, and thus, it was even harder to measure defensive contribution in a decent manner.
I don't think you can blame Hollinger for trying to make use of BoxScore stats that were easily accessible (steals, blocks, dreb). If he had omitted those I'm sure lots of fans would have preferred another stat over PER "because it also measures defense" (even though it would not have been actually good at measuring defense)

[CousinOfDeath]

New to this forum. Long time poster on the Bulls board and occassionally on the Knicks board. Hi to all!

My two cents on this are:

PER is like salary. So, if A makes 50,000 USD and B makes 100,000 USD it may feel like B makes bank compared to A.

However, this ignores all context:

1) How long was B an employee as compared to A?
2) What are the median salaries in the 2 industries that A and B come from?
3) Who pays more taxes?
4) Does B spend more than A to an extent that effectively the person with the lesser salary saves more money after bills and expenses?

All important questions.

But, like PER, the guy who has the 100K salary is probably pulling in more girls in the weekend than the guy pulling in 50K.

Meaning, PER is a glittery stat. It will do alrite for a one night stand. You want a more meaningful relationship? Screw PER. Metaphorically, that is.

I don't really like this comparison. There's basically no situation where a person with a 50k salary makes more money than someone making 100k.