SOTD - Different NBA positions and PER differences

[Double Helix]

I'm likely to make some bad typos as I'm posting this from a train on a phone. With that out of the way, like many people, I find PER useful as a quick, one-glance indicator of dominance league-wide. Obviously no single formula or index can account for everything but one problem I've often had with PER is that few people tend to view PER from a positional context.

If you look at all players who've played at least 20 games and who average at least 20 minutes of playing time in any one NBA season it quickly becomes apparent that it's unreasonable to be comparing the PER of bigs with wings. If I had more time and wasn't posting from a phone I'd provide more historical data but for the 2013 season so far, here are the league-wide PER averages for every position when looking at all impact players who play solid minutes.

PGs - 16.28
SGs - 14.15
SFs - 14.94
PFs - 15.78
Cs - 17.19

It makes sense when you stop and think about it but all posters should be taking these differences more into consideration when comparing bigs with wings and vice versa. So, is a 16.0 PER wing above average for his position and a 16.0 PER PF/C merely average? Perhaps.

EDIT - Dhackett has taken this concept further by creating different restrictions based around the concept that there are often different minutes distributions going on, depending on the positions played. What follows is probably the best look at the subject in general. Thanks, Dhackett for your work here. For what it's worth, I feel the last one is most reflective of the league as a whole.

Total minutes played (not per game) weighted, with no game restriction:

PG: 15.27
SG: 13.79
SF: 14.46
PF: 15.35
C: 16.45

Total minutes played (not per game) weighted, with 10+ game restriction:

PG: 15.35
SG: 13.82
SF: 14.48
PF: 15.35
C: 16.46

Minutes per game weighted, with 10+ game restriction:

PG: 14.83
SG: 13.48
SF: 13.76
PF: 15.04
C: 15.98

[J-Roc]

That's very interesting, thanks. PER definitely could use context like this.

[pbj]

Genius.

However, based on this, would you argue that players like Lowry and Ed Davis have an "inflated" PER relative to Demar Derozan?

Perhaps a more accurate calculation of PER would be something like Player PER/League Position AVG..

"Relative PER"
Lowry = 20.4/16.28 = 1.253
Derozan = 14.6/14.15 = 1.032
Gay = 21.0/14.94 = 1.406
Amir = 17.8/15.78 = 1.128
Gray = 9.6/17.19 = 0.558

Bargnani = 12.4/15.78 = 0.786

[Indeed]

Interesting, but I wonder how we calculate this when teams are going small.

[dhackett1565]

Another consideration - what if the wings are just populated by worse players than the other positions? Perhaps this just means that wing players in general should be of less concern to a team. Or perhaps it means that efficiency is less important at the wings than at other positions. Hard to draw any concrete conclusions from this.

Also, what happens if you remove the minutes restriction? It seems to me there are a lot of bigs that sub for 15 minutes a night or so, depending on matchups and fouls, more so than with wings and guards.

[CreaM]

I think this thread needs to be stickied. Its something I wanted to bring up because it seems some here over-value this stat.

Per tends to reward guards who dominate the ball. Point guards have a chance to pad assists when they draw the defence and are able to find team-mates for the catch and shoot. It also rewards players who take a lot of threes. Point guards are perfect for this as they are the head of the snake and have the ball in their hands more than any other player on the court, so they are able to do more with it. A shooting guard doesn't have this luxury unless they are a play-making type like Tyreke Evans (although he cant shoot threes) or Steph curry.

Big man rarely have to create their own shots as they are mostly set up on pick and rolls/pop and are relegated to put backs on offence, making their efficiency a lot higher as they don't have to create their own shot. Pair that with rebounding availability and blocked shots and this makes for a good PER.

Shooting guards and small forwards don't really have unique designated stat like rebounds and assists, so if they fail to shoot threes at a high rate, it will most likely affect their PER output. This is why I don't see PER as a perfect stat as some do here. It rewards different positions differently.

[Double Helix]

Another consideration - what if the wings are just populated by worse players than the other positions? Perhaps this just means that wing players in general should be of less concern to a team. Or perhaps it means that efficiency is less important at the wings than at other positions. Hard to draw any concrete conclusions from this.

Also, what happens if you remove the minutes restriction? It seems to me there are a lot of bigs that sub for 15 minutes a night or so, depending on matchups and fouls, more so than with wings and guards.

The average minutes per game of all NBA players is apparently 21 so 20 seems like a reasonable cut off to me.

The average PG plays 21 minutes
The average SG plays 21 minutes
The average SF plays 20 minutes
The average PF plays 18 minutes
The average C plays 20 minutes

[Double Helix]

Genius.

However, based on this, would you argue that players like Lowry and Ed Davis have an "inflated" PER relative to Demar Derozan?

Perhaps a more accurate calculation of PER would be something like Player PER/League Position AVG..

"Relative PER"
Lowry = 20.4/16.28 = 1.253
Derozan = 14.6/14.15 = 1.032
Gay = 21.0/14.94 = 1.406
Amir = 17.8/15.78 = 1.128
Gray = 9.6/17.19 = 0.558

Bargnani = 12.4/15.78 = 0.786

I had considered doing something similar. let's be honest. Those numbers above (or below) the positional averages do seem to make sense.

And it also makes sense to me that the Jeff Teagues, Will Bynums and Jameer Nelsons of the league are essentially producing at an average clip for the position they play due to the talent level available at the PG position whereas the JR Smiths, OJ Mayos and Kevin Martins of the league this year, despite having similar PER scores to the PGs, are actually producing at an above average clip for the position they play due to the scarcity of talent at that position the nature of the position itself.

[Indeed]

I think this thread needs to be stickied. Its something I wanted to bring up because it seems some here over-value this stat.

Per tends to reward guards who dominate the ball. Point guards have a chance to pad assists when they draw the defence and are able to find team-mates for the catch and shoot. It also rewards players who take a lot of threes. Point guards are perfect for this as they are the head of the snake and have the ball in their hands more than any other player on the court, so they are able to do more with it. A shooting guard doesn't have this luxury unless they are a play-making type like Tyreke Evans (although he cant shoot threes) or Steph curry.

Big man rarely have to create their own shots as they are mostly set up on pick and rolls/pop and are relegated to put backs on offence, making their efficiency a lot higher as they don't have to create their own shot. Pair that with rebounding availability and blocked shots and this makes for a good PER.

Shooting guards and small forwards don't really have unique designated stat like rebounds and assists, so if they fail to shoot threes at a high rate, it will most likely affect their PER output. This is why I don't see PER as a perfect stat as some do here. It rewards different positions differently.

As you mentioned, rebounding is part of the PER, therefore, I think it is fairly distribute (usually bigs have higher PER). I think the biggest issue of PER is, it does not represent your role in the team. Many 1st option may not have high PER.

[Blast Tyrant]

Great thread. I've always wondered about this. I guess this makes Demar and Rudy have slightly more impressive numbers. What's scary is how bad the SF position must be beyond the PER gods such as Lebrons and KD.

[pbj]

Also.. I think we can "equalize" PER based on positional averages but my logic is probably flawed.

Average of those 5 positional PERs is 15.67
15.67 / Positional PER =
PG = 0.9624
SG = 1.1073
SF = 1.0487
PF = 0.9929
C = 0.9115

And the re-apply that to our guys' PER..

Lowry = 19.63
Derozan = 16.17
Gay = 22.02
Amir = 17.67
Gray = 8.75
Bargnani = 12.31

Compared to untouched PER courtesy of bball-ref
Lowry = 20.4
Derozan = 14.6
Gay = 21.0
Amir = 17.8
Gray = 9.6
Bargnani = 12.4

[BC4CB4]

Very interesting stats - thanks DH!

[Double Helix]

Great thread. I've always wondered about this. I guess this makes Demar and Rudy have slightly more impressive numbers. What's scary is how bad the SF position must be beyond the PER gods such as Lebrons and KD.

Both guys are certainly boosting the SF average significantly due to their individual dominance. They're true outliers. They're the most dominant players in the game for numerous reasons but also because the talent gap between themselves and an average replacement at the position is ridiculous. A PER 20.0 SG or SF is just flat out more rare and further from the average player at the position than a PER 20.0 C, PF or PG.

[dhackett1565]

Another consideration - what if the wings are just populated by worse players than the other positions? Perhaps this just means that wing players in general should be of less concern to a team. Or perhaps it means that efficiency is less important at the wings than at other positions. Hard to draw any concrete conclusions from this.

Also, what happens if you remove the minutes restriction? It seems to me there are a lot of bigs that sub for 15 minutes a night or so, depending on matchups and fouls, more so than with wings and guards.

The average minutes per game of all NBA players is apparently 21 so 20 seems like a reasonable cut off to me.

The average PG plays 21 minutes
The average SG plays 21 minutes
The average SF plays 20 minutes
The average PF plays 18 minutes
The average C plays 20 minutes

Sure, but looking at the fraction of players in the groups that exceed 20 MPG:
PG: 54/100 (54%)
SG: 48/83 (57%)
SF: 46/96 (48%)
PF: 44/99 (44%)
C: 36/71 (51%)

So PF's are being undersampled by including only players playing well above median minutes. Where SG's are being oversampled, including a number of below-median minutes played players.

[Double Helix]

Another consideration - what if the wings are just populated by worse players than the other positions? Perhaps this just means that wing players in general should be of less concern to a team. Or perhaps it means that efficiency is less important at the wings than at other positions. Hard to draw any concrete conclusions from this.

Also, what happens if you remove the minutes restriction? It seems to me there are a lot of bigs that sub for 15 minutes a night or so, depending on matchups and fouls, more so than with wings and guards.

The average minutes per game of all NBA players is apparently 21 so 20 seems like a reasonable cut off to me.

The average PG plays 21 minutes
The average SG plays 21 minutes
The average SF plays 20 minutes
The average PF plays 18 minutes
The average C plays 20 minutes

Sure, but looking at the fraction of players in the groups that exceed 20 MPG:
PG: 54/100 (54%)
SG: 48/83 (57%)
SF: 46/96 (48%)
PF: 44/99 (44%)
C: 36/71 (51%)

So PF's are being undersampled by including only players playing well above median minutes. Where SG's are being oversampled, including a number of below-median minutes played players.

Well, if you remove the minutes restriction and instead look at all players in the league the averages simply drop even lower across all of the positions, as would be expected. The biggest difference there is that PFs rise up to second above PGs but Cs still stay remain their top status and SGs and SFs remain the positions with the least talent out there in the league.

PG - 14.02
SG - 12.97
SF - 13.17
PF - 14.59
C - 15.05

[ItsDanger]

Coming from someone who uses stats in their job, I find the recent stat development in sports to be severely flawed, especially baseball. Too many qualitative factors at play.

[Too Late Crew]

There certainly is validity to a theory of scarcity of talent at a position raising or lowering g the bar.

Unfortunately using position as separator has less impact than say 10 years ago,

Should Lebron be lumped in with Sf or PF.

Most would say he's a SF a position that seems to be lacking in talent. yet 82games.com claims he plays over 60% of his minutes as a PF

JJ Hickson is some 4 inches shorter than Aldridge yeat he apparently is a C 90% of the time

Positions are so blurred now that there are many guys who could go from below to above average simply by swapping the letter next to their name on the box score.

[dhackett1565]

Looking at the total data set (hoopdata.com - I believe they use unforced PER, which doesn't force the average to 15), the averages break down as follow:

PG: 12.2
SG: 12.22
SF: 12.11
PF: 13.69
C: 14.38

And with a minutes restriction of 10 MPG:

PG: 13.83
SG: 12.8
SF: 12.91
PF: 14.69
C: 15.2

And with a games played restriction of 10+, with no minutes restriction:

PG: 13.43
SG: 12.5
SF: 12.53
PF: 13.94
C: 14.36

And using players above the median minutes played for each position:

PG (21.5): 16.0
SG (23): 14.4
SF (19.1): 15.4
PF (18.5): 15.9
C (20.7): 17.1

All of these sets show a less extreme variation than the 20 MPG restriction. That last seems the most reliable to me - it does support the inflation of C's PER, and similar patterns on the wings, with a less extreme distribution. So, similar conclusions, but with a little less impact (18% versus 22% maximum difference, 4.5% vs 5.8% average deviation).

[Double Helix]

Looking at the total data set (hoopdata.com - I believe they use unforced PER, which doesn't force the average to 15), the averages break down as follow:

PG: 12.2
SG: 12.22
SF: 12.11
PF: 13.69
C: 14.38

And with a minutes restriction of 10 MPG:

PG: 13.83
SG: 12.8
SF: 12.91
PF: 14.69
C: 15.2

And with a games played restriction of 10+, with no minutes restriction:

PG: 13.43
SG: 12.5
SF: 12.53
PF: 13.94
C: 14.36

And using players above the median minutes played for each position:

PG (21.5): 16.0
SG (23): 14.4
SF (19.1): 15.4
PF (18.5): 15.9
C (20.7): 17.1

All of these sets show a less extreme variation than the 20 MPG restriction. That last seems the most reliable to me - it does support the inflation of C's PER, and similar patterns on the wings, with a less extreme distribution. So, similar conclusions, but with a little less impact (18% versus 22% maximum difference, 4.5% vs 5.8% average deviation).

Not utilizing a games played restriction isn't wise, in my opinion. You now have Daniel Orton's 62.8 PER and Viacheslav Kravtsov 34.37 PER affecting things, as well as some negative and 0.0 data.

A games played restriction of some kind is a must, in my opinion. I appreciate your work on the last offering though. That's helpful. Is that WITH a games restriction of 10 games as well or no?

[StopitLeo]

Coming from someone who uses stats in their job, I find the recent stat development in sports to be severely flawed, especially baseball. Too many qualitative factors at play.

True story, but that won't stop people from trying to examine things quantitatively. My issue is trying to create individual player metrics like PER in a sport like basketball where: 1) discrete traditional positions are less clear, and 2) accounting for the impact of the other 4 players on the court is extremely difficult.

Nevertheless, I appreciate the OPs thread as I find dissecting PER interesting given its suspect fundamentals in statistics.