I read a cool series today. In between working on a consulting gig at home, I read this piece from deadspin. In it they take it upon themselves to rank the 100 worst nfl players of all time. I was inspired to rank the 100 worst statistical seasons of all time (or at least since 1978) as well as the worst players.
Now, they didn’t get all statistical and set up a method for doing this. They just followed their gut. I’m going to be a little more scientific
Some quick ground rules:
- Go read the Basics if this is your first rodeo
- I’m looking for cost to the team, so each season will be ranked by the total differential from the players win production to the average players win production with similar minutes. So minutes count.
- My blog, my definition of value
Image courtesy of xkcd.com
I looked at every player and every season in the the database and this is what I came up with:
And that’s pretty cool, but I really wanted to take it further so I decided to average the standard deviations for every players season’s and use them to rank the players themselves. I also took out anyone who didn’t (or hasn’t) played three years in the league. That looks like this:
Sorry Devin, Bargs only rates number 3.
Now since I’m a little swamped, I’m letting all pertinent discussion in the care of my regulars.
Tomorrow we go around the WOW.
nerdnumbers
11/21/2010
Yes! Excellent post Arturo. I have to guess that when accounting for salary above standard salary and wins below standard wins that Bargs may jump to 1
some dude
11/21/2010
any worst players list without Kwame Brown is one that disappoints me. How is he not on this list!?
BTW, another thing to add on would be to see how many of these seasons were employed by teams. Who employed the most worst individual seasons? Toronto? The Knicks? I must know!
EvanZ
11/21/2010
And folks wonder how OKC is winning without Jeff Green…
EvanZ
11/21/2010
unfortunately for them, Krstc is still healthy.
Guy
11/21/2010
This is an excellent, if inadvertent, illustration of why you need to fix WP’s rebound problem. Let’s look at Clifford Robinson, who really stands out here for being allowed to play 18 seasons while, allegedly, subtracting wins from his teams. From age 23 to 37 Robinson played for a single team each season, usually as a regular. Over these 15 seasons, playing for 4 different teams, WP says that Robinson cost his teams 4.7 reb per 48 minutes (treating him as 50% C and 50% PF). So WP says that Robinson cost his teams a staggering 3,726 boards, or about 250 rebounds per season and a loss of 7.5 wins per season.
Now let’s look at his teams. How many rebounds below average were his teams? If WP is correct about diminishing returns, it should be somewhere in the ballpark of -3700 (he had lots of different teams and teammates, so overall it’s likely they were about average). The answer is +1,411 — his teams had a net rebound advantage of 1411 boards, or about 94 per season. So we expected to see -3700, but instead see +1400. Where did these extra 5,000 rebounds come from? Does anyone seriously think that Robinson just got “lucky” and had teammates who were 5,000 rebounds above average?!?
And if you do think that, then try to explain why the same story holds for every other big man on the list:
Morrison 2007: -320 rebounds. Team: -113. WP Fail.
Mashburn 1994: -334 rebounds. Team: -92. WP Fail.
Amaechi 2001: -176 rebounds, team -76. WP Fail.
Collins 2007: -159 rebounds. Team: -33. WP Fail.
Every single “bad” rebounder is largely offset by above-average teammates. Every one, it never fails. Or rather, WP always fails. Just as all “great” rebounders have lousy teammates, all bad rebounders have very good teammates.
Cliff Robinson was basically a league average NBA player for many seasons. The idea that he was literally subtracting wins from his teams is just totally implausible. You WOW network guys aren’t doing Prof. Berri any favors by ignoring the problem. You should work it through, find a solution, and help him fix WP. Until you do, WP just won’t be taken seriously outside a very small circle. If you don’t, you’ll end up claiming funny things like Kevin Love is the 2nd most productive player in the NBA. (Oops….. :>))
EvanZ
11/21/2010
By what measure could Clifford Robinson possibly be considered anything more than a bad rebounder? Even without using WP, he averaged 7.6 reb/36 in his *best* season. His career average is much, much worse (5.3 reb/36). Ok, so if he didn’t bring the rebounding, he must been a lights-out (efficient) shooter, right? Nope. Career 51% TS. These numbers are so far below the average F/C in the NBA, it’s hard to explain how he lasted so long.
I usually think of this as the “Vladimir Radmanovic” effect, in which his coaches are under some kind of magic spell that convinces them to give him unwarranted minutes. Maybe I should rename this the “Clifford Robinson” effect. I never realized he was so bad. Thing is, you don’t need to know anything about WP to see that in the statistics.
ilikeflowers
11/21/2010
Are all of Guy’s posts about the overvalued rebounds myth? The rebounds critique has been dealt with over and over again – from looking at what happens to individual stats when players switch teams to looking at what happens to wp48 when rebounds are devalued (almost nothing) to providing mirror arguments to all of the silly anecdotes that apply equally to scoring.
Guy
11/21/2010
I’m interested in lots of issues, flowers. Want to discuss the usage/efficiency tradeoff? But you are badly misinformed on this issue. When players change teams it often has a significant effect on other players (Rodman reduced his teammates rebounds dramatically wherever he went, for example). And I’m not offering “anecdotes,” I’m saying EVERY “bad rebounder” has only a small negative effect on team rebounds, and every great rebounder has only a small positive effect — much less than WP claims. If you disagree, find me some counterexamples.
And seriously, where do you think Robinson’s team found 5,000 extra rebounds?
EvanZ
11/21/2010
“I’m saying EVERY “bad rebounder” has only a small negative effect on team rebounds, and every great rebounder has only a small positive effect”
That’s not necessarily an “effect” of a bad rebounder or a great rebounder. It may be that teams generally try to reach some “sufficient” level of rebounding, and not much more.
Guy
11/21/2010
Evan, the WOW story is that rebounds are not valued by NBA decision-makers. They play no role in drafting players, or setting salaries. That’s the only way you could explain a player as bad as Robinson allegedly is playing for 18 seasons. So you’re saying all of that is true, BUT teams simultaneously are so concerned about rebounds that they are extremely careful to make sure every poor rebounder is matched by good-rebounding teammates? Indeed, every single team manages to do this every year, while also not realizing how important rebounds actually are? I’m sorry, but that story makes no sense at all.
Why do people keep trying to invent theories to dismiss this evidence, theories which are always totally inconsistent with the WOW analysis anyway? Wouldn’t it be better (and more interesting) to try to figure out how to fix the metric?
EvanZ
11/21/2010
Well, Guy, I’ll say this. You’ve piqued my curiosity about it. That’s a start.
ilikeflowers
11/21/2010
The pay for scoring math speaks for itself and is rock solid, your anecdotes on this point are worthless.
EvanZ
11/21/2010
” I’m saying EVERY “bad rebounder” has only a small negative effect on team rebounds, and every great rebounder has only a small positive effect — much less than WP claims. If you disagree, find me some counterexamples.”
Just thought I would mention David Lee as a counter example. He’s made a huge impact. The Warriors went from the worst rebounding rate in the league to somewhere near league average. They’re starting to drift below that since he’s been out. To say his effect has been small is simply untrue.
arturogalletti
11/21/2010
Guy,
The math is not that simple.
If these players played with all average rebounders then your statement would be factually correct. It isn’t . A more valid test would be comparing the result of replacing one player with another. The David Lee example that Evan brings up is a better and more telling example.
Guy
11/22/2010
Arturo: of course any one example could be explained by a player having very good (or very bad) teammates. But my point is that it’s ALWAYS the case that so-called “bad” rebounders have good rebounding teammates, while good rebounders always have bad rebounding teammates. In the aggregate, that obviously spells “correlation,” which I know you understand. And the correlation is huge. Here’s something you could easily do with your data: compare a team’s rebounds above/below average at center with the rebounds from the other 4 positions. Do the same at PF. My prediction is you will find a very large negative correlation (in ballpark of -.7). That wouldn’t be true if there were only minimal diminishing returns, as WP assumes.
I don’t know what the Lee example shows us. Of the top 6 GSW players in MP this year, only two of them played significant time for GSW last year. A lot of moving parts there. But I’m certainly not arguing that players have zero effect on rebounds. I’m sure Lee will add some rebounds this year. Rodman added rebounds (but not the 400-500 that WP credits him with.) The question is how many?
EvanZ
11/22/2010
Guy, having now read the diminishing returns articles by Eli and Nichols, which I’m sure you are familiar with, (I assume you are the “Guy” that Eli mentions several times), I found their data/regressions convincing, to say the least. To my mind, anyway, it does seem like a significant problem that should be addressed.
Couldn’t WP be changed simply by using the 0.7/0.3 weights (like AWS does)?
Guy
11/22/2010
Evan: .7/.3 (OReb/DReb) seems like a reasonable starting point. There are some reasons to think Eli Wittus’ approach understates the extent of diminishing returns, as he acknowledges. Using a different methodology, he came up with much lower coefficients. My guess is the right coefficients are something closer to .5/.2, but this is a good area for further research (and the answer may vary by position). I’m not familiar with work by Nichols — can you post a link or reference? Thanks.
EvanZ
11/22/2010
“The Diminishing Returns of Rebounds and Other Stats” (Jon Nichols, Dec. 2009)
http://basketball-statistics.com/blog1/2009/12/06/the-diminishing-returns-of-rebounds-and-other-stats/
Guy, I assume you frequent the APBR forums? What do you think about DSMok1′s advanced SPM stats measure? I found it interesting, and am thinking about developing something similar, except using Monte Carlo simulation to play around with some of these effects.
EvanZ
11/22/2010
Also, Guy, can you give me some links to your work on the subject (assuming it’s public)? TIA.
Guy
11/22/2010
Thanks for the link. Very interesting work by Nichols. The assist finding is intriguing. The one limitation for both Nichols and Witus’s models is that they are, to some extent, predicting themselves since they use same year reb% to predict what each lineup will do (the season rate already includes the lineup being “predicted”). So that will tend to understate diminishing returns, maybe by a fair amount. If you used prior-year reb%, and looked at teams with significant personnel changes, you would probably get a more accurate estimate.
I haven’t done any rigorous studies on this. I did some comparisons of rebound variance at the player level (large) and team level (small), which shows that there must be massive diminishing returns. I think I posted those somewhere at APBR. But that approach would also pick up efforts by team to balance good and bad rebounders, as you speculated above, so it has limitations too.
Another way to see the diminishing returns is to regress player WP48 on rebounds vs. position average, which shows rebounds is hugely important in predicting player WP48. But if you do the same thing for team WP48, rebounds play a much smaller role in predicting team wins. Obviously, both relationships can’t be the true measure of rebounds’ contribution to winning. (The team regression is more correct, IMO.)
ilikeflowers
11/21/2010
No, you are badly misinformed on this issue. Your opinions on me and on this issue are irrelevant. The evidence is on wow. You’re on a wow network site. YOU bring the counter evidence. I believe you posted something about rebounds from that Ivey League student who dberri already ripped to shreds once before – is there anything else you have?
Guy
11/22/2010
Sorry, flowers, I have no idea what you’re talking about here.
Raspu10
11/21/2010
So many “old friends” on the list… hi Cliffy, hi Duck, heya Jellybean…. Evanz, I’d call it the Bryant Effect, since Kobe’s Dad was equally good at it.
Devin
11/21/2010
As I was looking at the first list, the first thing that came to mind was “where the hell is Bargnani?!?!? He only shows up once at #45!”
Then I see the second list, and all is right with the world.
Really…Millsap had the highest PAWS of any college player in the 2006 draft. Instead of picking him (or Roy, or Rondo, or…anyone), Colangelo decided to go with “the mystery box” and select Bargnani.
BARGNANI WAS A #1 OVERALL PICK!!!!!!!!!!!!!! AAAAAAAARGGGGGG!
You can bet a link to this is going on NBeh?.
Rad E. Cool
11/22/2010
You gotta feel sorry for John Wall, or at least Flip Saunders!
The Wiz trot out the worst player of all time and 3 of the top 50! Luckily they cut Adam Morrison or else it’d be 4.
some dude
11/22/2010
David Lee is not a valid example of swapping players this year.
Golden State often ran out a 4 guard lineup last season and Biedrens missed a ton of time. Golden State, iirc, was the worst rebounding team in NBA history. Lee isn’t going to a somewhat below average rebounding team, he went to a team that played midgets at the forwards and center spots (NBA speaking) and was the worst ever.
You can’t look at Lee and scream “See! Look at their rebounding now!” Had they simply played slightly below average rebounders for their PF/C positions this year, they would have improved drastically.
Shawn Ryan
11/24/2010
“1. Go read the Basics if this is your first rodeo”
Hehe, how often is the advice “Go read this” given to someone because they are about to partake in their first Rodeo? Sometimes taking idioms literally can yield some funny interpretations.
Daniel
06/10/2011
Nice use of the past participle form “costed”. I don’t see those particularly often…