I was considering e-mailing Tom to point out that two totally different questions were being conflated in this debate:
1. Is PER a valid way to sum up other statistics?
2. Are per-minute statistics reliable for low-minute players?
Obviously, he needed no such suggestion, tackling the second issue in the post linked above. I think many of us are at least somewhat wary of PER and other linear-weights statistics, so I didn't really have an issue with point 1; point 2 is a different story.
Tom has another worthy look at per-minute numbers, but these studies are hurt by the factors mentioned in the comments, namely that causality in the minutes-performance relationship should go both ways.
I think the only study which has ever looked at in-season playing time changes caused by outside factors has been Hollinger's first Pro Basketball Prospectus, which concluded:
"Every player in the above list experienced major performance drops in games where they played fifteen minutes or less, and their fluctuations in playing time had almost nothing to do with their in-game performance."
Since it would be better not to start another PER thread, I'd like to add to the discussion.
Ever since I read BoP, I've wedded myself to the idea of matching everything to actual points scored and points allowed. A team's points scored and allowed should be the foundation for judging a team, and by extension for assigning credit to its players.
But PER simply credits players with points for their box-score stats (or debits them in the case of turnovers, missed shots, and fouls). It doesn't fit within the points scored/allowed framework, and because it doesn't, I'm left wondering what the points that are awarded in PER are actually measuring. The points that are deducted from a player for turning the ball over - where do they actually show up in the team ledger? It's hard to say exactly, because teams don't lose points for turnovers.
So how do I reconcile the "free-form" addition and subtraction of points in PER with my idea that everything should be strictly related to the team's points scored and allowed? Is it possible to reconcile the two?
Joined: 13 Oct 2005 Posts: 373 Location: Atlanta, GA
Posted: Thu Sep 13, 2007 12:01 am Post subject:
94by50 wrote:
Since it would be better not to start another PER thread, I'd like to add to the discussion.
Ever since I read BoP, I've wedded myself to the idea of matching everything to actual points scored and points allowed. A team's points scored and allowed should be the foundation for judging a team, and by extension for assigning credit to its players.
But PER simply credits players with points for their box-score stats (or debits them in the case of turnovers, missed shots, and fouls). It doesn't fit within the points scored/allowed framework, and because it doesn't, I'm left wondering what the points that are awarded in PER are actually measuring. The points that are deducted from a player for turning the ball over - where do they actually show up in the team ledger? It's hard to say exactly, because teams don't lose points for turnovers.
Great, great post. Thank you. I've been saying this for about a year now. I'm glad someone else actually gets it, too.
94by50 wrote:
So how do I reconcile the "free-form" addition and subtraction of points in PER with my idea that everything should be strictly related to the team's points scored and allowed? Is it possible to reconcile the two?
You don't, and it's not. PER may be a quasi-useful measuring stick for player performance in the absence of everything else, but it shouldn't be considered anything more. And I say this as someone who joined this board and this movement specifically because of John Hollinger's writing.
Joined: 14 Jan 2005 Posts: 1506 Location: Delphi, Indiana
Posted: Thu Sep 13, 2007 8:41 am Post subject:
The bloggers who 'started' this debate would seem to be second-generation stat-users who 'grew up with' PER. But these same debates have made the rounds at APBRmetrics (and before, at APBR-analysis) every year, for 7 or 8 years.
Indeed, we are never more than a week or two removed from a similar debate here, are we?
Similar studies (by several among us) have done the minutes/effectiveness thing, and always the results are inconclusive. There are always players whose optimal minutes are 14, others are 44, most are in between. Since we generally don't know why a guy is getting more minutes one night, his minutes may or may not depend on his performance.
Ideally, all improvement should result in more minutes for a player. So the correlation should be closer to 100% than to the 67-70% found by TZ (and others). Perhaps only a couple-% of these were given more minutes before showing any sign of improvement, and subsequently improved.
Since the correlation is 'only' 67-70%, I have to believe there's a good bit of the supposed 'myth' being played out: A sizable number of players do worse when given more minutes.
From one season to the next, players often markedly improve (or regress) . Their minutes should absolutely change accordingly. From one game to the next, a player isn't the same effectiveness; a good coach sees this and plays him more or less.
The 'exception to the rule' is when a player is dramatically better or worse than before -- last year or last week -- and his minutes do not change. _________________ 40% of all statistics are wrong.
But PER simply credits players with points for their box-score stats (or debits them in the case of turnovers, missed shots, and fouls). It doesn't fit within the points scored/allowed framework, and because it doesn't, I'm left wondering what the points that are awarded in PER are actually measuring. The points that are deducted from a player for turning the ball over - where do they actually show up in the team ledger? It's hard to say exactly, because teams don't lose points for turnovers.
This is missing the point, though. Let's say Player A turns the ball over under his own basket and the opposing team kicks out and nails a three-pointer. Player B, on the other hand, commits a very similar turnover under his own basket but the kick-out flies out of bounds (meaning that Player B's team recovers). By your logic, Player A should be penalized 3 points, and Player B not penalized at all - yet both committed the same error.
Anyway, I know this wasn't the real point of your post. What does a PER of "18.6" mean? Well, nothing, unless taken in context. But this is true of many statistics - what does a slugging percentage of .530 MEAN? This player is likely to produce 0.53 total bases for each at-bat - which is gibberish in the context of any individual at-bat, of course, but meaningful over the course of a season.
I guess I just think that many of the criticisms of PER come from trying to ascribe too much meaning to it. Hollinger has loudly argued in the past with any attempt to use PER as a "ranking system" of players, and his player comments in his books have always acknowledged the context behind each player's PER. Hollinger would never have suggested that PER be used to strictly assign playing time. It has shortcomings, but I think it is a valuable short-hand tool, and the only valuable statistic I can calculate on my own. _________________ No books - no articles - no website.
Just opinions.
Hollinger has loudly argued in the past with any attempt to use PER as a "ranking system" of players, and his player comments in his books have always acknowledged the context behind each player's PER.
Well, this may be the case, BUT he still ranks his top 50 on his site every season. If it shoudln't be used to "rank" players - why does he do it? _________________ www.goodstats.net
I'm not sure S.K. was correct to say that PER has nothing to do with ranking players. I think it obviously does. The important point is that it hasn't been argued as a definitive and final word on ranking players -- there are some things it measures well, some things it measures less well, and some things it doesn't measure at all.
So expecting Hollinger to never provide a list of players ranked by PER isn't really reasonable. PER is not infallible, but what is? Every ranking could be shot down by that argument.
I'd much rather read through his list ranked by PER scores than alphabetically. And I know that any given player on the list isn't necessarily better than the man below him.
Hollinger has loudly argued in the past with any attempt to use PER as a "ranking system" of players, and his player comments in his books have always acknowledged the context behind each player's PER.
Well, this may be the case, BUT he still ranks his top 50 on his site every season. If it shoudln't be used to "rank" players - why does he do it?
Well, it's still a stat - you can rank players by PER, but it doesn't suggest that the top five PERs are the top five players in the league, and so on. Hollinger is always careful to say "his best season *by PER*", and I remember him being very angry last pre-season when ESPN billed his season preview as "John Hollinger ranks every player in the NBA!"
edited to add: Of course asimpkins beat me to the punch. _________________ No books - no articles - no website.
Just opinions.
To me PER is like the big bang theory before the discovery of microwave background radiation. It sounds nice and all, but where's the evidence that it's correct? Someone needs to device a method whereby PER can be used to make predictions which turn out to be correct. That's how science is supposed to work.
Well, this may be the case, BUT he still ranks his top 50 on his site every season. If it shoudln't be used to "rank" players - why does he do it?
Doesn't that seem similar to any other stat. Players are ranked by points per game, assist ratio, and true shooting percentage. This only displays that Player A is better than Player B by that particular stat. _________________ No fancy website
No fancy book
Since the correlation is 'only' 67-70%, I have to believe there's a good bit of the supposed 'myth' being played out: A sizable number of players do worse when given more minutes.
The basic idea is that per-minute productivity will decline with large jumps in mpg because of a) the increased quality of teammates with whom production is shared, and b) the increased quality of defenders. I use play-by-play data to measure the aveage quality of both teammates and opponents for each player-game observation in the 2006-2007 season. I find that while the effect of teammate-quality is neglibile, the effect of opponent quality is both negative and significant across all specifications. That is, increasing the average PER of the opposing unit one shares the court with leads to lower individual production. Since mpg has a positive effect on opponent quality, I conclude that increased minutes should lead to decreased productivity, all things equal. And that this is especially true for bench players, for whom large increases in mpg will correspond to large increases in opponent quality. For these players, per-minute measures of productivity (like PER) will be inflated.
Anyway, if your interested, check out the article for more details.
Joined: 14 Jan 2005 Posts: 1506 Location: Delphi, Indiana
Posted: Sun Sep 16, 2007 1:17 pm Post subject:
Very interesting. I studied players who started and subbed at least 10 games (each) last year, and found that playing with subs gives you a lower TS% and fewer assists, but more of everything else.
A starter who plays more than about 32 minutes is actually going to face more subs, the more minutes he plays. A starter going all 48 minutes will typically have played against about 2/3 starters and 1/3 subs. That is, lots of garbage time, in the course of a season.
Bench players with low minutes face the fewest starters.
Starters with low minutes face the highest % of starters.
Your conclusions are quite believable. In fact, I still find it hard to believe anyone thinks otherwise. _________________ 40% of all statistics are wrong.
Everyone believes that players' effectiveness will go down with more minutes, because they'll face better players. Logically, this makes total sense.
However, any effort to quantify this effect will lead you to the exact opposite conclusion. Players tend to fare better when given more minutes (within reason), especially when they go from very short snippets of playing time to longer ones. Study any group of players who see sudden increases or decreases in playing time due to injuries or trades, and you'll see this effect.
Value above replacement or value above average might be worth trying to bring more prominently into the media and to more fans. To address, in a way, the issue of minutes of demonstrated performance. Especially if PER were split into offensive and defensive components. To me, offensive and defensive VAR or VAA would help describe and rank players better than roll-up PER, which of course can still continue to be used for what it is good for, a quick composite assessment, though not as complete on defense.
And O and D VAR or VAA splits taken to position specific replacement and average levels would be another option that might appeal to some fans interested in further refined analysis and insiders.
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
