I've had this question for a while and am finally getting around to asking about it. It concerns the Player Efficiency Rating (PER) model and whether it's appropriate to use it for collegiate basketball players (NCAA Division I in particular).
I know that the PER was originally developed for evaluation of NBA players, although in recent years the model has been used for calculating PER numbers for collegiate players as well, with the results published by various media sources.
The problem as I see it is that from what I understand, the coefficients used in the model were based on historical NBA information. Beyond that, the numbers are adjusted to account for NBA league averages and normalized to an assumed PER value of 15 for an 'average' NBA player.
My question is when looking at collegiate players, would this model still be valid (given its basis on NBA, not NCAA data) and if not, is there a modified version that someone has developed which is more appropriate ?
FWIW, the NCAA does publish historical statistical averages for Division I as a whole, so that potentially could be a source of "league-average" and historical data to arrive at a more realistic collegiate modification of the PER model, if it hasn't been done already.
Question about PER Model for Collegiate Players
-
AcrossTheCourt
- Posts: 237
- Joined: Sat Feb 16, 2013 11:56 am
Re: Question about PER Model for Collegiate Players
Well, first of all, the numbers aren't calculated for an average NBA player; they're for an average player of whatever league you're calculating. You'd use the average for college players.*
There are a number of league rates calculated, not assumed from historical data, like factor = (2 / 3) - (0.5 * (lg_AST / lg_FG)) / (2 * (lg_FG / lg_FT)).
There are some assumptions about the value of certain stats, however. But the problem is I don't think they're all based on historical data. They're just ... assumptions. For example, Hollinger uses a 2/3rds factor for assists, and I remember him saying this is because the player assisting does two steps of the assisted play: finding someone open and passing the ball. The last third is the player making the shot.
I'm not too familiar with the collegiate PER model though. There probably are some tweaks.
*Here's the distinct problem of doing this for the NCAA: strength of schedule, which is normally a small factor in the NBA. And how much of the NCAA do you base this average PER on?
There are a number of league rates calculated, not assumed from historical data, like factor = (2 / 3) - (0.5 * (lg_AST / lg_FG)) / (2 * (lg_FG / lg_FT)).
There are some assumptions about the value of certain stats, however. But the problem is I don't think they're all based on historical data. They're just ... assumptions. For example, Hollinger uses a 2/3rds factor for assists, and I remember him saying this is because the player assisting does two steps of the assisted play: finding someone open and passing the ball. The last third is the player making the shot.
I'm not too familiar with the collegiate PER model though. There probably are some tweaks.
*Here's the distinct problem of doing this for the NCAA: strength of schedule, which is normally a small factor in the NBA. And how much of the NCAA do you base this average PER on?
Re: Question about PER Model for Collegiate Players
To echo on the above, I think you would definitely need to factor in Strength of Schedule as well as RPI to an extent if you were going to compare players across the spectrum of both the league and the NCAA as a whole.
I've been toying with formulas to judge my own teams individual contributions by distributing weights of the statistics by position and so forth.
If you come up with something, I'd like to see what you do.
I've been toying with formulas to judge my own teams individual contributions by distributing weights of the statistics by position and so forth.
If you come up with something, I'd like to see what you do.
Re: Question about PER Model for Collegiate Players
Why not look at my rankings (HnR) - which are linear based, take into account SoS and pace, and also adjust for player playing time in relation to quality of teammates (to help adjust for player factors positive or negative not found in the boxscore stats). They are converted to a 100 "average" - like looking at baseball's ERA+ or OPS+, 150 is 50% better (theoretically) than the average player.CoachO wrote:To echo on the above, I think you would definitely need to factor in Strength of Schedule as well as RPI to an extent if you were going to compare players across the spectrum of both the league and the NCAA as a whole.
I've been toying with formulas to judge my own teams individual contributions by distributing weights of the statistics by position and so forth.
If you come up with something, I'd like to see what you do.
http://www.hoopsnerd.com/2012-13CollegePDFs.html
I even have an impact rating (HnI) that accounts for plyers that missed chunks of games due to injury or suspension - for example, Nerlens Noel was ranked "only" 75th nationally in HnR partly due to missed games because of injury - but was ranked 26th in HnI when ignoring missed games: http://www.hoopsnerd.com/uploads/2012-2 ... sFINAL.pdf . Looking at the 2011 rankings - Kyrie Irving ranked 54th nationally in HnR - but 2nd nationally in HnI (ignoring missed games): http://www.hoopsnerd.com/uploads/2010-2 ... sFINAL.pdf . HnI obviously is better to look at in projecting future (pro?) performance than HnR - HnR is better at looking at who had the better season ALL things considered (including missing games).
BTW - adjusting weights based on subjective positions is the wrong way to go imo. I spent a large amount of time figuring out what appeared to be appropriate weights for college that seemed to translate effectively (pass the laugh test) across all positions and SoS's.