Positional PER

[Alfred]

PER at each position is radically different. It's much higher at the PG and PF positions, and much lower at the wing positions.

Because positional crossover isn't common, shouldn't we be comparing players to other players at their positions? For example, which has higher value, a 17 PER player at the PG position, or a 15 PER player at the 3? The comparative production of the SF might outstrip that of the PG.

Is there a way of quantifying this discrepancy?

[turk3d]

I think you make a great point and this is something I've been thinking about. Wouldn't it be better to statistically compare teams (as well as players) by position? You could even do a +/- by position which would tell you where your teams weaknesses (and strengths) are, as well as look at individual players and how they stack up against the opposition. Maybe this exists already but I think it would be quite useful. It actually could simplify things a little.

[branny]

I think you make a great point and this is something I've been thinking about. Wouldn't it be better to statistically compare teams (as well as players) by position? You could even do a +/- by position which would tell you where your teams weaknesses (and strengths) are, as well as look at individual players and how they stack up against the opposition. Maybe this exists already but I think it would be quite useful. It actually could simplify things a little.

Would be quite useful/interesting to look at!

[Doctor MJ]

PER at each position is radically different. It's much higher at the PG and PF positions, and much lower at the wing positions.

Because positional crossover isn't common, shouldn't we be comparing players to other players at their positions? For example, which has higher value, a 17 PER player at the PG position, or a 15 PER player at the 3? The comparative production of the SF might outstrip that of the PG.

Is there a way of quantifying this discrepancy?

Wait, it's much HIGHER at the PG? Please elaborate on this. Traditionally top point guards are the most underrated players by this metric.

To your general idea, I'd put this in terms of an "uncanny valley" of sorts. You're quite write that different types of players get over & underrated by the raw PER metric, so further adjustments make sense. However, just splitting it up by position doesn't help that much because "position" according to the NBA is one of only 5 roles despite the fact that there are vastly more roles than that, and PER manages to over & underrate within the same position.

[turk3d]

There is a difficulty due to the way the NBA is run these days (positions seem to float all over the place). However, to simplify things there are always only 5 players on the floor (hence 5 positions for want of a better term, feel free to change it to another name. What we are accustomed to is 1-PG, 2-SG, 3-SF, 4-PF and 5-C. Regardless of how it stacks up, Each of the 5 players on each team has to match up (play defense) on 1 of the 5 players on the opposing team.

If you measured the +/- impact on each of these matchups, you'd get a pretty clear picture of who's doing their job. I feel that most of these metrics (even the +/- 5 player units) is a bit unfair because they basically don't tell you what's really important imo how a guy stacks up against the player (or players) he's actually matched up against, he's subject to what the rest of his teammates do.

Now I realize that the justification for doing it this way is that is that it's about how a guy makes his teammates "better" or "worse". But my answer to that is what happens if the same player is afforded the same opportunity to play with the same unit (for the same number of minutes to get an equal sample size)? Just because the +/- may be + for certain unit doesn't mean it would be even better (more +) if another player were afforded the same opportunity.

Typically what happens if it's positive it may bet stuck with, however an individual player on that unit may be getting beat out on whoever it is that he's individually matched up with and might be carried by the rest of his teammates giving what appears to be a positive result. This does not mean that the team wouldn't perform even better with another player in that same position.

To me the only way to compare players fairly is to look at them individually (before looking collectively). If your 5 players outscore the oppositions 5 players (which you would see if you looked at each position individually 1-5) I don't care by how many points, you win every time.

[Alfred]

PER at each position is radically different. It's much higher at the PG and PF positions, and much lower at the wing positions.

Because positional crossover isn't common, shouldn't we be comparing players to other players at their positions? For example, which has higher value, a 17 PER player at the PG position, or a 15 PER player at the 3? The comparative production of the SF might outstrip that of the PG.

Is there a way of quantifying this discrepancy?

Wait, it's much HIGHER at the PG? Please elaborate on this. Traditionally top point guards are the most underrated players by this metric.

I used Hoopdata to determine what the average PER is for players playing over 15 minutes a game (why? Thought it would filter out some of the "noise" generated by players who might not have a defined position playing so few minutes. This might be totally incorrect.)

http://www.hoopdata.com/advancedstats.aspx?team=%&type=pg&posi=PG&yr=2013&gp=0&mins=15
http://www.hoopdata.com/advancedstats.aspx?team=%&type=pg&posi=SG&yr=2013&gp=0&mins=15
http://www.hoopdata.com/advancedstats.aspx?team=%&type=pg&posi=SF&yr=2013&gp=0&mins=15
http://www.hoopdata.com/advancedstats.aspx?team=%&type=pg&posi=PF&yr=2013&gp=0&mins=15
http://www.hoopdata.com/advancedstats.aspx?team=%&type=pg&posi=C&yr=2013&gp=0&mins=15

If you check out last year as well, it's similar, with the PERs being higher at PF, PG and C, and much lower at the swing positions.

[Alfred]

You'll have to copy and paste those into your browser if they aren't fully loaded.

[Doctor MJ]

]
If you check out last year as well, it's similar, with the PERs being higher at PF, PG and C, and much lower at the swing positions.

I like that you're trying to analyze it like this but you said "radically differs".

The point guard number there is well under 1 point different than the small forward number, and if you move the threshold to 25 minutes, the gap becomes less than 0.15. This is not a radical difference. This is the same ballpark, influenced by the players involved in the league at the moment.

In other words, shooting guards come across sucking here because shooting guards in the league happen to suck right now.

Historically for example, Magic never had a season with a PER north of 24.0 until he became the team's first scoring option, which means that PER view the true pass-first version of Magic being roughly equal to Gilbert Arenas. Now, I suppose you could say that Arenas is technically a point guard, but that's really my point. Arenas is, whether he's a shooting guard or a point guard, was certainly not getting underrated by the stat, but the true point guard get horribly jobbed.

Getting back to the bigs, I think you've got a stronger case there. The gaps I see on Hoopdata are bigger there. I don't know if that's necessarily wrong though. Even ignoring the whole sample size of the moment issue, wingmen have had no problem putting up GOAT level numbers in the stat. Do you look at PER and think that the greats at these positions are terribly off the mark? If they seem okay, then is this truly a bias?

[Alfred]

In other words, shooting guards come across sucking here because shooting guards in the league happen to suck right now.

You're missing the point. I don't care about the historical. I care about the present. How good is a shooting guard right now compared to other shooting guards in the league? If a SG has a PER of 15, and every other SG in the league has a PER of 10, then that player is going to be worth more than a PF with a PER of 15 in a league where an average PF has a PER of 20.

Also, a league-wide PER difference of ~4 between bigs and swings IS massive. The sample size is also pretty big too (this season and last).

[mysticbb]

If a SG has a PER of 15, and every other SG in the league has a PER of 10, then that player is going to be worth more than a PF with a PER of 15 in a league where an average PF has a PER of 20.

Why should that be the case, if the overall game concept consists of having 5 players on one side against 5 players on the other side? Basketball is not a game of 5 individual 1on1 games, but one 5on5 game, in which each team tries to outscore the opponents in order to win the game. How much a player contributes to the win is NOT determined by how much his PER is higher than the PER of the player who is supposed to play the "same" position on the other team.

If PER is a useful representation of the player's impact on the outcome of the game, no positional adjustment is needed at all. If it is not, no positional adjustment whatsoever will change that.

Also, a league-wide PER difference of ~4 between bigs and swings IS massive. The sample size is also pretty big too (this season and last).

The difference is much smaller due to the fact that hoopdata incorrectly is giving out the simple average instead of the minute weighted average. The difference between bigs (PF+C ~ 15.85 PER) and wings (SG+SF ~ 14.09 PER) is 1.76 or 12.5%. That is hardly massive at all. One of the reasons: not so skilled bigs are getting replaced by more skilled wings players, reducing the amount of minutes low PER bigs are getting in order to give more minutes to skilled wing players (players like James, Durant, Anthony, etc. getting minutes at PF for example). Right now the wings have 10000 minutes more playing time than the bigs. If that would be shifted back to the lower PER bigs as it was before, the bigs minute weighted PER would drop again. At the same time lower PER wing players are getting more minutes, which is decreasing the average for those wings. But the point of the game, as I mentioned before, is to win as a team anyway.

Historically, PER has no specific bias towards any position, just a bias towards scoring and against defense, if a player does not create boxscore entries.

[turk3d]

Why should that be the case, if the overall game concept consists of having 5 players on one side against 5 players on the other side? Basketball is not a game of 5 individual 1on1 games, but one 5on5 game, in which each team tries to outscore the opponents in order to win the game.

True, but the fact remains that if you were to take total points scored at each position for each time and added them up, you'd get the same result. Hence, this becomes a simplified way to find out where the pluses are coming from, as well as the minuses and allow you to attempt to make the appropriate adjustments, by making substitutions in order to improve that outcome (it's just a slightly different way to quantify things).

And also, I agree who cares about the past? What's important is the present and what's going to improve your present situation. Past data sure has it's uses, but during the season you're more concerned about improving the present. One of the problems is that the past doesn't take into account (especially the more years a years a player has behind him) is if/when a guy might be having a breakout year (or for that matter a downer year). I think there could be value in doing it both ways.

[Alfred]

If a SG has a PER of 15, and every other SG in the league has a PER of 10, then that player is going to be worth more than a PF with a PER of 15 in a league where an average PF has a PER of 20.

Why should that be the case, if the overall game concept consists of having 5 players on one side against 5 players on the other side?

Because generally, one PG, one SG, one SF, one PF and one C are on the court at the same time. That isn't always the case, but positions do matter.

Who you are replacing a player with is just as important as how good that player is. If their competition is terrible, then they become more valuable. It's like in baseball -- a catcher hitting a so-so average is decidedly more important than a first baseman hitting a so-so average.

Also, a league-wide PER difference of ~4 between bigs and swings IS massive. The sample size is also pretty big too (this season and last).

The difference is much smaller due to the fact that hoopdata incorrectly is giving out the simple average instead of the minute weighted average. The difference between bigs (PF+C ~ 15.85 PER) and wings (SG+SF ~ 14.09 PER) is 1.76 or 12.5%. That is hardly massive at all. One of the reasons: not so skilled bigs are getting replaced by more skilled wings players, reducing the amount of minutes low PER bigs are getting in order to give more minutes to skilled wing players (players like James, Durant, Anthony, etc. getting minutes at PF for example). Right now the wings have 10000 minutes more playing time than the bigs. If that would be shifted back to the lower PER bigs as it was before, the bigs minute weighted PER would drop again. At the same time lower PER wing players are getting more minutes, which is decreasing the average for those wings. But the point of the game, as I mentioned before, is to win as a team anyway.

I don't really see why we should care about the weighted minutes? The value of the proposed stat is that it allows you to gauge a player based on the strength of his peers at his position. If he is producing at a level that outstrips his competition at that position, his value is higher than someone who doesn't.

[Doctor MJ]

In other words, shooting guards come across sucking here because shooting guards in the league happen to suck right now.

You're missing the point. I don't care about the historical. I care about the present. How good is a shooting guard right now compared to other shooting guards in the league? If a SG has a PER of 15, and every other SG in the league has a PER of 10, then that player is going to be worth more than a PF with a PER of 15 in a league where an average PF has a PER of 20.

Granting that if the variance is big enough that could be an issue, to me the big thing I'm seeing from what you're saying that leaves me saying "Whoa hold on!" is the idea that the average shooting guard is by definition as valuable as the average big, and by extension that the average shooting guard in any one year must be by definition as valuable as the average shooting guard in any other year.

mystic's saying good things relating to this but in my own words:

If the quality of 2's in the league goes down, more responsibility gets shifted to other positions. Yes to an extent the fact that there is an average size associated with 2's means that there is a replaceability aspect of the 2 that cannot be taken up by another position which limits how far the value of the average 2 can fall, but it certainly does not mean that it cannot fall at all.

Of course there is a question of how big that fall is relative to the typical average value of the position. If you believe that fall can't possibly be major, then you'd certainly want a stat that judges the positions separately.

This seems rather absurd to me. Let's consider the career of Derek Fisher. The dude gets way too much hate because he plays a role that most teams don't really have: secondary guard, smallest defender. People judge him compared to point guards because that's his official position, but that's not his job. His job is to fill in the gaps, just like pretty much any other role player. It is something to be values...

this does not make him as valuable as Kobe Bryant though. And if every team in the league decided to run offenses that let their 2 be the lead guard, then this would change the 1's in secondary guards who were not as valuable.

Now, teams haven't decided they no longer want people of 2 size to be their lead guard. That isn't way 2's have dropped. It isn't quite THAT dramatic. However, the fact remains that talent can ebb and flow, and that to the extent PER is measuring everything we want to measure (it isn't of course but that's another thread) there SHOULD be some some drop when the talent drops.

[Doctor MJ]

I guess more succinctly, we're running into a bit of a recent pet peeve of mind. I see threads saying, "Is OJ Mayo a top 5 shooting guard?", and it makes think "Why would anyone care?". Even if he is Top 5 at his position, if he's a volume scorer who can't volume score super-effectively, I'm not going to pay him like he's an elite player.

[mysticbb]

True, but the fact remains that if you were to take total points scored at each position for each time and added them up, you'd get the same result. Hence, this becomes a simplified way to find out where the pluses are coming from, as well as the minuses and allow you to attempt to make the appropriate adjustments, by making substitutions in order to improve that outcome (it's just a slightly different way to quantify things).

The conclusion you are drawing about players is wrong. It really doesn't matter at all, whether a player is outscoring a supposed to be direct opponent player in order to make a positive impact. In fact, a player, who has his basic positive impact due to his team and help defense, can even get "outscored" by 10 or 15 points while still being the main reason for the win. Ben Wallace got constantly outscored and yet he had a great positive impact.
The issue is that the players are having specific roles, which are not just determined by their "positions", but also by their skillset and skill level.

And also, I agree who cares about the past? What's important is the present and what's going to improve your present situation. Past data sure has it's uses, but during the season you're more concerned about improving the present. One of the problems is that the past doesn't take into account (especially the more years a years a player has behind him) is if/when a guy might be having a breakout year (or for that matter a downer year). I think there could be value in doing it both ways.

Reducing the overall sample might very well lead to the wrong conclusion, let alone that PER has not a very good ability to predict the outcome of games anyway.

Because generally, one PG, one SG, one SF, one PF and one C are on the court at the same time. That isn't always the case, but positions do matter.

That doesn't explain that at all. Those "position" do not present a specific role for each player, because that is also determined by the skillset and skill level.

Who you are replacing a player with is just as important as how good that player is. If their competition is terrible, then they become more valuable.

No, that is a wrong conclusion. You need to understand the concept of the game before drawing conclusions like that. As DocMJ pointed out, a player does not become a better piece to build around only because the average PER of players who are supposed to have the same position is lower than for other positions. It is a flawed concept, which is not representative for the real basketball game.

I don't really see why we should care about the weighted minutes?

Because you want to compare the player to the average value on his position, and that is determined by the minute weighted average not by the simple average.

The value of the proposed stat is that it allows you to gauge a player based on the strength of his peers at his position.

You misunderstand the purpose of the stat here, if you think that the stat is meant to be "gauge a player on the strength of his peers at his position". The idea is that PER represents a value, which is supposed to show the overall impact on the outcome of the game regardless of any position.

If he is producing at a level that outstrips his competition at that position, his value is higher than someone who doesn't.

That is nonsense within the framework of the game.