Calculation of player efficiency rating (PER)

[molepharmer]

I never visit this forum so apologize in advance if this is a common topic.....

I'm trying to calculate PER given a player's stats and am not coming up with the same value as shown at basketball-reference. I have an Excel spreadsheet with all the necessary numbers (e.g. league stats, pace, etc.) and am using the equation at basketball-reference and in Wikipedia (which are the same).

Calculation of 'factor', VOP and DRBP seem straightforward - 'tho the equation for 'factor' seems a little odd (0.5 in numerator and 2 in denominator). Determining the lg_aPER isn't straightforward either but that doesn't seem to be where my error is occurring because not all of my calculated PER numbers are off the same amount from the reported PER.

Anybody ever try these PER calculations with success ??? Or something called WARP ?

[Don Draper]

I think PER is a load of crap.I don't believe in anything that uses weights which are arbitrary and have no statistical foundation.

[Malinhion]

I think PER is a load of crap.I don't believe in anything that uses weights which are arbitrary and have no statistical foundation.

It does have a statistical foundation. It tries to reduce everything to point production. Not a flawless method, but nothing really is. It's actually not that bad, but it should be pared down and reduced to an offensive measure, only.

[tsherkin]

PER is a summary of box score contribution, with a fairly specific description of what a player has done and yes, the weights do have statistical foundation (most, anyhow).

You have to understand what PER is saying to really get how to use it properly.

[Don Draper]

Why are fouls even considered?

Why are asts scaled down to 2/3?

I understand the .44 for the adjusting of free throws that end in possessions and the .5 for 3 pointers but the rest is less clear.

[tsherkin]

Assists are scaled down because a large chunk of the work done in the assist-basket play is the conversion of the basket, which has nothing to do with the passer.

Fouls are presumably considered (there's a legit reason, I just don't remember) because they are a negative impact and, if offensive, can also be turnovers.

[Ryoga Hibiki]

Fouls are considered because they lead to FTs for opposing teams.

Btw, imo PER is useless because it doesn't understand the limits of boxscore stats.

[tsherkin]

PER is not useless, just limited.

[Bobbcats]

rofl at a stat needing to "understand the limits of box score stats"

The real problem is understanding the limits of PER. For some reason the people who hate PER want to see it as the complete encapsulation of a player into a number.

[Don Draper]

My questions his how did he scale an assists down to 2/3. I am very skeptical of any stat analysis that arbitrarily weights stats without any concrete statistical significance over a long period of time.

For example it was found out that from the 2002-2006 season 43.8% of free throws ended in a possession. Hence the .44 scale for free throws. I would like to see some date for his assists and free throw weights.

I personally use usage% and offensive rating.

Offensive rating tells you how efficient a player is offensively, and added with usage% you can roughly determine how involved the player in the offense when he is on the floor (of course minutes per game also factor into this). Furthermore any weight used in offensive rating has been proven.

But PER tries to capture individual defensive performances which I believe are almost impossible to capture with the official stats given. Even Dean Oliver's defensive rating formula has seriously flaw in that since defensive is more a team concept bad defenders can get good ratings on good defensive teams and vice versa.

[erudite23]

I hated PER at first, but I've come around a bit. If you use it as simply a per-minute, comprehensive measure of statistical accumulation its pretty good. There are serious flaws, and its amazing to me the way that it continually underrates the great PGs in history--likely due to the devaluing of the assist--but its still a useful tool.

For the most part, a list of highest PER lines up roughly with what one might venture is the greatest players in the NBA, with a few notable exceptions. Often times these exceptions illuminate a particular trend with the player who seems out of place at the spot he is ranked--be he much higher or much lower than one might think he should be from mere casual observation--and therein lies its value. A good example this season is TMac. He didn't crack 20 this season, and was waaaay down the list, obviously. Well, a closer looks reveals a player who was not scoring efficiently at all, and whose other numbers (aside from his passing numbers) have taken a serious down turn, yet most would name him a top 15-20 player in the game for this past season. So it does its job, though it should obviously never be used as a hard and fast rating system.

[Malinhion]

I think McGrady's PER totals from last year might just seem "off" because he's been one of the most consistently overhyped wings in the NBA. Really, he just hasn't been that good in that long. His efficiency is atrocious.

I do believe that his impact extends beyond his statistical contribution, but it's like nobody has even really taken a look at T-Mac's numbers in the past four years!

[Ryoga Hibiki]

rofl at a stat needing to "understand the limits of box score stats"

The real problem is understanding the limits of PER. For some reason the people who hate PER want to see it as the complete encapsulation of a player into a number.

there's nothing to laugh about, because it's pretty clear just watching its formula what should be the point of PER: a summary of all statistical contributions of a player. So? What should I know ranking players using PER? What is PER telling me that I wouldn't know just looking at basic stats?
Player X scores 28, Player Y 22: is that a remarkable difference? Does it say X has a bigger impact? A bigger impact on offense? A bigger impact not considering individual defence (the only flaw I've seen JH admitting)? What?

Statistical analysis should be all about using different metrics to measure different aspects of each player and than test them following the scientific approach to understand if they're really helping the team and how (the only thing that really counts).
You can see a high rebound rate but than discover it's inflated by bad teammates, easy def reb stolen, atc.
You can see a not very efficient scorer who actually creates a lot, always moves the ball, a takes shots only when the plays are broken and there are few seconds on the shotclock.
How can you do the same when PER did put everything together in a big melting pot?

PER = useless, I don't remember a single time it helped me when analyzing a player.

[Copperhead]

I think a lot of people just don't understand PER nor do they feel like taking the time out to understand it. I think the consensus is that pretty much everyone knows it's not perfect.

Player Efficiency Rating (PER) is a statistic created by John Hollinger. PER is a great way to look at and synthesize the statistical contributions of a player in the relation to the other players in their league.

The important thing to consider when talking about PER is that it is a per-minute stat. Per-minute stats are the holy grail of basketball statistical analysis. In the NBA historically, per-minute stats remain constant despite increases or decreases in a particular player playing minutes.

PER is also adjusted for pace (number of possessions). Teams that run a run-and-gun offense, like Mike D’Antoni, will have their per-game statistics skewed because they will have more possessions in a game. Likewise, conservative game styles (ala Dennis Wolff) will also have their per-game statistics skewed because they will have less possessions in a game. Pace adjusted stats allow us to compare players with statistics regardless of the tempo that they play at.

The problem with PER, like most stats, is that not everything in basketball can be measured with statistics. Certain players will have low PERs because the value that they give their teams does not translate into statistics. The best example of this is Bruce Bowen of the San Antonio Spurs. Bowen has a low PER because his outstanding perimeter defense does not result in a lot of steals or blocks. Being assigned to the opposing team’s best perimeter player also limits the number of rebounds he can get. However, his value to the Spur’s last three championships is unquestionable.

http://blog.thebuzoo.com/2008/07/06/sho ... ating-per/

A Year For the Ages: 35.0
Runaway MVP Candidate: 30.0
Strong MVP Candidate: 27.5
Weak MVP Candidate: 25.0
Bona fide All-Star: 22.5
Borderline All-Star: 20.0
Solid 2nd option: 18.0
3rd Banana: 16.5
Pretty good player: 15.0
In the rotation: 13.0
Scrounging for minutes: 11.0
Definitely renting: 9.0
On next plane to Yakima: 5.0

[Ryoga Hibiki]

I think a lot of people just don't understand PER nor do they feel like taking the time out to understand it.

Why don't you answer my question: what is PER measuring and how am I supposed to use it?

http://blog.thebuzoo.com/2008/07/06/sho ... ating-per/

A Year For the Ages: 35.0
Runaway MVP Candidate: 30.0
Strong MVP Candidate: 27.5
Weak MVP Candidate: 25.0
Bona fide All-Star: 22.5
Borderline All-Star: 20.0
Solid 2nd option: 18.0
3rd Banana: 16.5
Pretty good player: 15.0
In the rotation: 13.0
Scrounging for minutes: 11.0
Definitely renting: 9.0
On next plane to Yakima: 5.0

this implies it's a good way to rank players, isn't it?
when a difference is going to be significant?

[Copperhead]

I think a lot of people just don't understand PER nor do they feel like taking the time out to understand it.

Why don't you answer my question: what is PER measuring and how am I supposed to use it?

http://blog.thebuzoo.com/2008/07/06/sho ... ating-per/

A Year For the Ages: 35.0
Runaway MVP Candidate: 30.0
Strong MVP Candidate: 27.5
Weak MVP Candidate: 25.0
Bona fide All-Star: 22.5
Borderline All-Star: 20.0
Solid 2nd option: 18.0
3rd Banana: 16.5
Pretty good player: 15.0
In the rotation: 13.0
Scrounging for minutes: 11.0
Definitely renting: 9.0
On next plane to Yakima: 5.0

this implies it's a good way to rank players, isn't it?
when a difference is going to be significant?

I don't know actually. I've never taken the time out to delve that far into PER. Sorry. Maybe someone else here will help you though. But what I'm saying is that I can understand how some people brush off what they don't understand. IF there's even anything to understand at all. If your question isn't answered in the explanations in PER articles or if no one here can help you or no one at sonicscentral/apbrmetrics can help you, then I don't know what to tell ya. Maybe you can email Hollinger directly.

[Ryoga Hibiki]

It's not that I don't understand PER, it's just that it looks clear to me what are JH's assumptions and where are the flaws.