PER and Usage Rate

[BlazerCaravan]

Being right on this board seems to be more important than actually looking at things through a different prism.

I have a machete, but I can't cut through the irony... it's just too thick!

 

[PapaG]

I have a machete, but I can't cut through the irony... it's just too thick!

Hey, at least I'm trying though. Stating that a 20% variance isn't significant and is just "noise" is false in terms of any scientific/mathematical model.

Again, thanks for your input in this thread. It was the one idea that actually expanded my own thoughts and improved on them, IMO.

Feel free to bash away/ignore.

 

[BlazerCaravan]

Feel free to bash away/ignore.

Oh, quit pretending to be so delicate. You know you go after that "I Was Right On The Internet" trophy as much as anyone here (myself included). I'm glad to have brought some substance into the thread, even though my statistics seem to be surrounded by noise. Like most threads, this one will end poorly.

 

[PapaG]

Oh, quit pretending to be so delicate. You know you go after that "I Was Right On The Internet" trophy as much as anyone here (myself included). I'm glad to have brought some substance into the thread, even though my statistics seem to be surrounded by noise. Like most threads, this one will end poorly.

I don't claim to be right on the internet. Your own chart shows me that I've been wrong about Andre Miller, as I admitted in this thread.

Whatevs...

 

[andalusian]

In terms of statistical significance/variance, the "noise" that you mention most definitely matters in terms of assessing the statistics offered in this thread. 1.0 is very significant, and even rare. Even a 0.1 variance is also significant, at least in terms of raw data. If you have a counterpoint other than what you've offered so far in ths thread, please address the stats, because mathematically, you really haven't shown anything to counter the stats.

No it's not when it is based on a piece of information that is already in the original formula with better constants in front of it being massaged by a proper analyst...

Not sure why you're arguing an obvious statistical point, but whatever. Being right on this board seems to be more important than actually looking at things through a different prism.

Because I actually work with statistics day in and day out for over 20 years now - and while I am not an analyst - I spent enough time with them to get a clear idea of what makes sense and what does not - and what I see from the formula and your discussion does not make sense. But, if it makes you happy - go ahead, enjoy.

 

[PapaG]

No it's not when it is based on a piece of information that is already in the original formula with better constants in front of it being massaged by a proper analyst...



Because I actually work with statistics day in and day out for over 20 years now - and while I am not an analyst - I spent enough time with them to get a clear idea of what makes sense and what does not - and what I see from the formula and your discussion does not make sense. But, if it makes you happy - go ahead, enjoy.

You claimed that a Usg/PER ratio of 1.0 is relatively normal. I've provided links that show it is a rare exception, and not even close to the rule.

You have yet to address this, and if this basic statistical formula doesn't make sense to you, then it is hardly my fault that it's over your head. You could choose to actually do some research to back whatever your point is; instead, you decided to put forth a demonstrably false idea, and instead of wondering how you missed on it, you instead try to patronize me.

LOL

This place is hilarious.

 

[andalusian]

You claimed that a Usg/PER ration of 1.0 is relatively normal. I've provided links that show it is the exception, and not the rule.

I claimed nothing of the sort. I said that what Minstrel showed you with very different players getting something of the same order shows you that usage is already ingrained in PER - only much more refined and with less noise.

You have yet to address this, and if this basic statistic doesn't make sense to you, then it is hardly my fault that it's over your head. You could choose to actually do some research to back whatever your point is; instead, you decided to put forth a demonstrably false idea, and instead of wondering how hyou missed, you instead try to patronize me.

You are right. It is over my head how replicating less refined data that is already in PER with a lot more detail and a lot of thought put into it is an advantage - but if you feel happy about using it - go ahead.

 

[espn_hall_of_famer]

I am sorry - but if you look at the formula - you will see that this is exactly right - the value is efficiency per minute, not per action - that's why uPER is calculated with a (1 / MP) * .... - thus, what I have said is correct - the efficiency formula captures production (per minute) by default - if you have someone that is on the team and is an amazing shooter but only takes 1 shot per 10 minutes - his PER will be low. The usage/PER correlation is implicitly built into the formula - because we are talking about efficiency per minute. If someone is fantastically efficient per action - but his usage% is low - by definition - his PER will be low.


This basically tells you that generally speaking - coaches in this league have a pretty good understanding of how to use players - as the PER/USAGE correlation is pretty uniform as Minstrel implied.

In other words, doing some kind of a usage multiplication with PER is a rather useless operation imho - the usage is implied within the formula.. - if you look at the formula - it is "all the usage statistics that can be measured specifically in gain or loss directly" divided by minutes played and normalized to the team/league - adding the usage parameter which is basically "percentage of possessions when on the court that the player is involved in" basically adds nothing more than noise into the formula - because it does not distinguish between good and bad touches (assists vs. turn-overs for example) - and it duplicates the more refined data that is already there.

Just because the number is there - does not mean that it makes sense to combine it with another number to try to look for meaning. It just does not make too much sense, mathematically.

Well said Grasshopper.

 

[PapaG]

I claimed nothing of the sort. I said that what Minstrel showed you with very different players getting something of the same order shows you that usage is already ingrained in PER - only much more refined and with less noise.

Dude, just stop lying. This is what you posted.

as the PER/USAGE correlation is pretty uniform as Minstrel implied.

It's not at all close to uniform. You asserted it; back it up.

 

[PapaG]

Well said Grasshopper.

The Usg/PER ratio is not at all uniform.

Will you stop pretending to be John Hollinger now? I've had to correct you before on "your" own stats. Anybody can read a statistical forumula; Looking at ways to interpret that data is how the use of statistics advances over time.

It's my thought that a high usg/PER ratio should also be considered when judging players solely on "PER", which I see daily on this board to justify the ability or value of a given player.

 

[espn_hall_of_famer]

The Usg/PER ratio is not at all uniform.

Will you stop pretending to be John Hollinger now? I've had to correct you before on "your" own stats. Anybody can read a statistical forumula; Looking at ways to interpret that data is how the use of statistics advances over time.

It's my thought that a high usg/PER ratio should also be considered when judging players solely on "PER", which I see daily on this board to justify the ability or value of a given player.

Part of statistics is that there are many interpretations and not all are right and wrong ways of looking at it. And I would hope you could correct me on my stats as seeing another person's view on a statistic is how one improves in their own future intpretation and adjusts the variables to accomodate. And I don't try and "pretend" to be anyone. You can call a zebra a horse, but it's still a zebra. I've never tried to pretend to be a horse.

 

[PapaG]

Part of statistics is that there are many interpretations and not all are right and wrong ways of looking at it. And I would hope you could correct me on my stats as seeing another person's view on a statistic is how one improves in their own future intpretation and adjusts the variables to accomodate. And I don't try and "pretend" to be anyone. You can call a zebra a horse, but it's still a zebra. I've never tried to pretend to be a horse.

Well, that is a refreshing post! Thank you for acknowledging that there may be more than one way to assess the impact of a statistic, and how other variables may impact that statistic.

Who would be the "best" player under the PER system, in your opinion?

A player with a usage of 25 and PER of 25
A player with a usage of 25 and a PER of 15
A player with a usage of 15 and a PER of 15
A player with a usage of 15 and a PER of 25

These questions interest me, and coaching can also be called into question when assessing this statistical ratio.

 

[andalusian]

It's not at all close to uniform. You asserted it; back it up.

That's a mighty high zebra you are perched on. I told you exactly why I find it meaningless. Forget the uniform part - what I should have said was that Minstrel showed you vastly different players with pretty close ratio - not letting you learn much from it.

I am pretty much out of this. You want to take Hollinger's careful work on PER and make it more meaningful by multiplying it with a noisier and less precise bit of information that's already in there - go for it and enjoy it. I think I gave you a pretty good idea why I think this is not relevant - but if you do not like it - so be it.

 

[PapaG]

That's a mighty high zebra you are perched on. I told you exactly why I find it meaningless. Forget the uniform part - what I should have said was that Minstrel showed you vastly different players with pretty close ratio - not letting you learn much from it.

I am pretty much out of this. You want to take Hollinger's careful work on PER and make it more meaningful by multiplying it with a noisier and less precise bit of information that's already in there - go for it and enjoy it. I think I gave you a pretty good idea why I think this is not relevant - but if you do not like it - so be it.

In other words, you were completely incorrect on your 1.0 gaffe. Thanks for somewhat admitting it! No wonder you're "pretty much out of this". Hollinger himself admits that an inefficient shooter who puts up a high volume will have an inflated PER. You also glossed over my posts on a PER multiplier being worth looking at, but that doesn't surprise me, either.

 

[BrianFromWA]

Well, that is a refreshing post! Thank you for acknowledging that there may be more than one way to assess the impact of a statistic, and how other variables may impact that statistic.

Who would be the "best" player under the PER system, in your opinion?

A player with a usage of 25 and PER of 25
A player with a usage of 25 and a PER of 15
A player with a usage of 15 and a PER of 15
A player with a usage of 15 and a PER of 25

These questions interest me, and coaching can also be called into question when assessing this statistical ratio.

On a pure "best" scenario, I'd say the 25/25 guy. He's performing at a high level, and doing it with a) a large sample size and b) presumably more attention paid from the defense.

I think the 15/25 guy would be an excellent 3rd/4th option. Maybe like Pippen in 91...where he eventually proved (in 94) that he COULD be the man, but he hadn't shown it yet.

I think there are a lot of reasons someone might be a 25/15 and it's "ok", but he's not the best. If his 15PER is the highest, then he should be getting more of the touches. But that just means that your team isn't good.

 

[BrianFromWA]

Though I've been going with the premise that PER generally goes down as usage goes up for a mediocre player, and PER goes up with usage if you're an elite player. I'm not sure that assumption is correct. In Haberstroh's article about Horford today he makes it out to be relatively remarkable that Horford's PER, shooting% and turnover% have improved as his usage went from "4th-option" 17.6% last year to 20.4% this year.

 

[andalusian]

In other words, you were completely incorrect on your 1.0 gaffe. Thanks for somewhat admitting it! No wonder you're "pretty much out of this". Hollinger himself admits that an inefficient shooter who puts up a high volume will have an inflated PEr.

Uniform does not mean 1. It means that it is pretty consistent, and for the most part, it is. I actually told you that it is 1.0 +/- 0.2 which would be my guess. Just to look at it - I randomly chose 125 players that play at least 20 MPG and have at least 20 games (just to get a proper sample size) from this year, and only 9 of those did not fall within the +/- 0.2 I told you. Let's say I got really lucky - and there are 3 times as many that do not fall within this range - you are still looking at the vast majority falling within this range and even those that are out of it are very close. It is pretty uniform, as I have said.

But again, you are missing the big pictures. Who cares about that, what is important is that you are adding a number that is already implicitly in the formula, only doing it with a much less precise version of it - which makes it worse than PER by itself - and somehow think this adds value. Large data sets have lots of noise in them. The value of statistics is to try and minimize that noise and learn something from it. You are doing the exact opposite and somehow get all offended when we tell you that.

 

[BrianFromWA]

I also wonder about the usage rate straight-up as a factor. Jordan's highest usages were in the 38% range, Nic's at 17% now (highest of his career). Roy was between 23 and 28. Blake's hovered around 15% the last few years. Travis was low-20s. I just don't know how to work with it, I just think that there may not be a linear effect

 

[PapaG]

I also wonder about the usage rate straight-up as a factor. Jordan's highest usages were in the 38% range, Nic's at 17% now (highest of his career). Roy was between 23 and 28. Blake's hovered around 15% the last few years. Travis was low-20s. I just don't know how to work with it, I just think that there may not be a linear effect

I'm not saying it's a linear effect; hence some sort of multiplier that tries to explain high PER/ high usage players.

 

[PapaG]

The value of statistics is to try and minimize that noise and learn something from it. You are doing the exact opposite and somehow get all offended when we tell you that.

Actually, a value of statistics is to try and understand how a given variable correlates (or may even cause?) to the statistic. I'm not at all trying to minimize "noise". I'm trying to understand some "noise", and how it impacts the final result of the equation. I see a correlation that jibes with my own observation. If you're not interested in actually addressing it, please feel free to start your own thread.

 

[andalusian]

Though I've been going with the premise that PER generally goes down as usage goes up for a mediocre player, and PER goes up with usage if you're an elite player. I'm not sure that assumption is correct. In Haberstroh's article about Horford today he makes it out to be relatively remarkable that Horford's PER, shooting% and turnover% have improved as his usage went from "4th-option" 17.6% last year to 20.4% this year.

Usage is a dirty statistic. It has a lot of stuff that is either good or bad. Every time a player touches the ball within a possession - it is added to his usage. Unfortunately, you do not know if it was good (score, assist, rebound, block) or bad (turn-over, miss). PER tries to use the stuff we know for sure - and put the correct sign (+ or -) and some kind of a constant according to it's "value". Since we do not know what usage contains (the good or the bad parts) - it is not always direct correlation between more usage and better PER or the other way around. Also - possessions where the player touches the ball and are not accounted for by the normal direct statistics (scoring, missing, assisting, turnovers, blocks etc...) - can still be good (the player has so much attention on him that he started swinging the ball until a free player is found for an uncontested shot) or bad (dominated the ball, pounding it and allowing the defense to get set and protect the basket better).

In other words - it really does not give you too much information all by itself...

 

[THE HCP]

Well said Grasshopper.

Hey man, saw you in the media room last night.

 

[Minstrel]

I'm not saying it's a linear effect; hence some sort of multiplier that tries to explain high PER/ high usage players.

A multiplier would still be linear. A non-linear function is one that changes slope at different times during the output...simply adding a multiplier doesn't create that effect. If an extra percentage of Usage doesn't always mean the same thing in terms of how it impacts the team, then it's a non-linear factor and dividing it by PER will still yield a non-linear function...no matter what multiplier you put on it. In which case, just looking at Usage/PER is meaningless unless you know where the value of Usage changes and how.

Since Usage is on a completely unrelated scale from PER and, as andalusian said, incorporated into PER, it isn't at all clear that simply dividing Usage by PER is ever meaningful. Any more than dividing steals per game by winning percentage is particularly meaningful.

 

[PapaG]

A multiplier would still be linear. A non-linear function is one that changes slope at different times during the output...simply adding a multiplier doesn't create that effect. If an extra percentage of Usage doesn't always mean the same thing in terms of how it impacts the team, then it's a non-linear factor and dividing it by PER will still yield a non-linear function...no matter what multiplier you put on it. In which case, just looking at Usage/PER is meaningless unless you know where the value of Usage changes and how.

Since Usage is on a completely unrelated scale from PER and, as andalusian said, incorporated into PER, it isn't at all clear that simply dividing Usage by PER is ever meaningful. Any more than dividing steals per game by winning percentage is particularly meaningful.

It would be linear in the sense that the multiplier would add a linear effect, so long as you stay consistent in your equation. I should hope that any advancement of statistics is linear in its application. Also, I'm noticing a correlating trend, not some sort of mathematical fact, and I'm asking for it to be explained. Thus far, all I have is a lie about how a 1.0 Usg/PER ratio is close to a mean. Clearly, it isn't.

Any statistic is on an unrelated scale from PER. That seems obvious, doesn't it? Do you work with numbers for a living?

You were wrong on your 1.0 Usg/PER ratio assumption, but that's not a big deal to me. That another poster seems stuck on your false assumption and has ended up hijacking this thread is unfortunate, but I do feel that there is value in looking at the components of an equation, and how they may impact the answer of that equation.

I'm not saying that usage is or is not a valid statistic; I am saying that it may seem to play a role in PER inflation, that it is a part of the PER equation, that it may be able to separate players of a similar PER in terms of their on-court value, or even (gasp!) their "usage" by their own coach.

That's all I'm asking in my brainstorm. The non-linear answers that are all over the board, yet don't attempt to refute the Usg/PER statistic, don't surprise me.

 

[PapaG]

Hey man, saw you in the media room last night.

espn_hall_of_famer isn't John Hollinger.

 

[PapaG]

Since Usage is on a completely unrelated scale from PER and, as andalusian said, incorporated into PER, it isn't at all clear that simply dividing Usage by PER is ever meaningful. Any more than dividing steals per game by winning percentage is particularly meaningful.

I never said it was clear. I offered up a simple equation. One that you falsely refuted by guessing about a ratio that you thought was ordinary, but was actually quite extraordinary.

This place is crazy. A bunch of people who think that they are much smarter than they are in reality. I'm asking for help, which BlazerCaravan actually tried to give and did well, IMO, and the rest of the "experts" lie about a statistic that doesn't even exist and continue to dwell on that false argument.

 

[andalusian]

Fuck this place. Someone asks for help, you explain to him what something does not make sense mathematically - and he goes all upset at you for trying to answer him.

 

[PapaG]

So ... usage is a bad stat? Great

That wasn't at all the point of the thread. I noticed a correlation that may exist between PEr and usage. That's the only point.

 

[PapaG]

Fuck this place. Someone asks for help, you explain to him what something does not make sense mathematically - and he goes all upset at you for trying to answer him.

How can unrelated mathematical components make "sense" mathematically unless you explore the supposed connection?

You lied/were ignorant about Usg/PER and the 1.0 ratio being consistent. Admit that and we can proceed. As it is, I can only think that you're full of shit, which surprises me, because I've always enjoyed your posts in the past.

 

[PapaG]

How does trying to see if a correlation exists between two different data sets not make sense?

Do you really work with numbers, andlusian? Sabermetrics was born out of taking different sets of data and then trying to combine them into a new answer. You know this, right? You take as much data that is available to you as possible, and you try to make sense of them. At times, you try to combine them.


That you were so far wrong on your Usg/PER assumption tells me that you're not capable of advanced mathematics.