Bank of Maimonides
And so the Great Mainstream Stat Wars of Summer 2007 continue. Ziller fires back with his critique of TOIH, and Silverbird defends our honor thus in the comments section:
A couple of points. First, it seems like you are using season-level data here (please correct me if I'm wrong about this). This presents several problems. One that has been mentioned already is that players improve, so there will be covariance between quality and minutes increased (this is something you try to account for, I recognize). Another reason is that season-to-season increases in minutes often result when players are traded/signed to weaker teams, and thus are sharing the ball with weaker teamates. Likewise, players who join better teams and play fewer minutes may see an increase in efficiency, e.g. Matt Bonner (one of the players on our list) who, when he joined the Spurs, saw his playing time fall sharply but his PER jump 20%. Both of these cases are consistent with our original argument, which is that PER inflation reflects imbalances between teams. Career level data just doesn't work here. Better would be intra-season data on large mpg increases, but here you'd have to control for things like quality of opponent (since bench players generally see more time against crappy teams). Maybe I'll try to do this at some point, when i have the time.
Second- I'm not sure why you restrict your data set only to players whose mpg increase from the previous season. What about the players whose mpg decrease? (i.e. Matt Bonner). Maybe your results will be the same. Regardless, it seems like an odd restriction. After all, the relationship you're testing for is between minutes played and efficiency, not minutes increased and efficiency.
Third- We never claim that "if you increase a player's minutes, his efficiency will suffer". I'm sorry if you interpreted it this way. What we said is that very large increases in mpg - the kind that change a bench player into a starter - will decrease efficiency, all else equal (the "else" here being age, the team played for). Again, the only way to really evaluate this claim is with within-season data.
Finally, since our entire argument was about players with above-average PERs (the group we looked at was the top 150 - that is, players who should be starters, according to Hollinger), it seems a little odd to present your overall coefficients as somehow refuting us us. Your subgroup of players with 15+ PER is really the group we're talking about. And although I certainly wouldn't claim the negative coefficient (-.22) as some vindication - it is, as you say, still very small - it certainly seems important.
This is the part where I admit that I'm beyond my depth. On a common sense level, the original Silverbird/Shoals hypothesis still makes sense to me. On the other hand, it is entirely within the teachings of FreeDarko to think that all benches are stocked with hell-bugs awaiting their moment. I would only be really bummed if NOTHING happened when these players got their added PT.
Another note: Tom and I were talking earlier today about how deadest summer is the perfect time to have this discussion. I'm aware that some people spend all season up in this vein, but here you have minds addressing these issues from a range of perspectives. Super-stats can't wholly supplant subjectivity or intuitions, and most of us involved in this debate want to talk about ways that all these facets of NBA knowledge can co-exist. These forces should feed and learn from each other; when there's a clash, it's time to figure out the interaction between the two realms, not break out the scalp knives.
posted by Bethlehem Shoals @ 9/12/2007 01:09:00 AM
16 comments
16 Comments:
This is like the science versus religion debate, or Jack versus Locke.
Just want to say I love how Silverbird took some of our intuitive qualms to a completely new level of academic bad-assness. I liked TZ's post and thought it raises some interesting points, but you're spot on in that there's a difference between the question of, "What happens when Millsap gets another year under his belt and is rewarded more PT?" and "What happens if you suddenly make Ronny Turiaf a starter?" All else equal, I'm still a believer in the TOIH, and haven't seen anything yet that would indicate it doesn't hold.
(I'll follow the trend and leave my response to Silverbird here.)
As Jason notes, I think what we can get out of this (as well as Kubatko and Hollinger's previous studies) is that we cannot make generalized assumptions about how per-minute production will be effected by changes in minutes, regardless of other changes. If anything, we can generalize vaguely that per-minute production seems to have a positive correlation with an increase in minutes played... across nearly all subsets. (And that -0.22 from +15 PER guys wasn't a correlation coefficient -- it was the average change in PER among the 53 players. The correlation coefficient is actually +0.23.) And as I wrote in the post: "over-15 PERs who then got at least 10 extra minutes per game the following season, the average change in PER was +0.18." You say this is whom your theorem is after -- good players who see a huge increase in playing time.
So that finding is just as valid as the -0.22 finding, which is to say not very valid at all. If I posted the list of players in each case, we could both sit here and justify every single one based on age, team, situation. But that's not what this is about. It's about trying to address widely-held assumptions and finding out if they are right or wrong. In this case, I think we've shown pretty conclusively (especially combined with the previous studies) that increased playing time does not lead to decreased per-minute production universally. (I know you're after a smaller subset -- one we don't have in large enough fashion here -- but the claim has still resonated.)
I look forward to any work you end up doing with in-season data. But until then, there's no way I can reasonably assume TOIH is correct -- the evidence thus far just wholly disagrees.
This is correlation and not causation, though, right? It seems to me more likely that players would get more minutes because they got better, not vice versa. Especially if you're talking one season to the next.
In fact, intuition might tell me that this could be an attrition effect, in which young players that don't get better from years one to two, or two to three, will probably exit the league rather than stay in and push down the data. If this is true, then you'd expect the numbers you're getting. There should be a clever way of measuring this, although it would probably take a scientist to figure it out.
salt: I'd be more apt to understand your intuition if the data didn't say the opposite time and again.
I mean, even older high-PER players who saw their minutes increase saw their average per-minute production basically stay the same.
Tz: This may partially disprove a rookie attrition effect, but it still doesn't show causation one way or the other. It's still perfectly reasonable to think that players who (for whatever reason and whatever point in their careers) get better will experience a concurrent jump in minutes played (importantly: this increase will be at the expense of some other player who has worsened). Until you can produce data that better measures causation, it may be that your numbers simply illustrate that coaches are handing out playing time in a rational manner. Note that I said "may". It could also be that your theory is true, and I'm wrong. What I'm saying is that the numbers don't necessarily give that part of the answer.
One last thing: I'm not necessarily on one bandwagon or the other, but you will get people who don't like PER, and will ignore your data simply because they believe PER isn't a good measure.
In fairness to TZ and Jason, they do use a subgroup of 5+ and 8+ year vets, and find the same positive relationship. And this also holds when they include only above average players with 10+ mpg increases, if i remember correctly.
The problem remains that many remaining cases of season-to-season mpg increase probably correspond to changes in roster or team played for. It's just hard to imagine there are that many cases where bench players suddenly become starters without some dramatic change in the situation. And such changes aren't controlled for in their analysis. One way to do this would be to look at cases where a starter gets injured mid-season and see what happens to the PER of the player who replaces him. A kind of natural experiment for TOIH.
also, thanks to Carter for comments here and elsewhere. A lot of my own comments basically just repeat what he's already said.
When there's a clash I recommend silly self-fulfilling prophecies not be played out at the expense of people who like to keep it objective, and unPlaschke.
you, sir, are no tom ziller.
After Marbury finished testifying, he rode down 23 floors in a crowded elevator, saying to no one in particular: "Money makes you do crazy things, man."
Then he left the courthouse, smiling and singing a song out loud. When pressed to divulge more about his relationship with the intern, he instead commented cheerfully on a reporter's shoes.
wv: atsvloo
all toronto saw vince leap over olowokandi
raise thine hands, ye who are bored to effing smithereens about this! who hate charts! and espn anything! because both are BORING!
seriously, y'all brought us the awesome 70's mascot t-shirts and then, you bring us this. i have victim-blames-themself syndrome from it.
I saw with shock that Jackie Christie has a mini-column at Hoopshype.com. But I immediately felt better after reading the short summary under the link:
"Jackie Christie thinks his husband Doug and many other veterans trying to make a comeback still have a lot to offer to an NBA team."
All I'm asking for is some common sense. There doesn't have to be some middle ground hacked out through debating the merits of PER. That debate has already been done more than a few times over. The information is already out there about every advanced metric 'cept WoW.
Short of Rainman, no one sees numbers when they watch basketball. That's why they're so useful. PER, game flows, +/- rating, etc. allow someone to develop an opinion, or state a fact that they would have never realized from watching a game. It beats raw intuition, and it's a rational approach to analyzing the game.
Stats are "a rational approach to analyzing the game?" WHO CARES? Would you put stats on sex if you could? Basketball should be a passion, an equal escape for the over and under-educated alike. PER, efficiency, it's all the same: a way for people who can't play to ruin the act of watching basketball for millions of people.
"Sincerity we measure in Spelvins on a scale of zero to ten," Dashwood went on, totally absorbed in his subject. "Hedonism in Lovelaces-we've been lucky there; subjects are able to distinguish sixteen graduations. Finally, there's the dimension of Tenderness-we find zero to seven covers that, so that the perfect Steinem Job, if I may use the vernacular, would consist of ten Spelvins of Sincerity, sixteen Lovelaces of Hedonism, and seven Havens of Tenderness."
ODEN IS OUT FOR THE YEAR
Post a Comment
<< Home