Post

RodmanSpecific

Broader Analytical

Part 1(a): Rodman vs. Jordan

 Dennis Rodman has dominated Rebounding Percentage more than anyone has dominated any major stat.
 Before Rodman, we should have expected a rebounder of that quality to appear about once every 400 years.

 Use standard deviations to measure relative greatness
 Outliers can skew their own data against them

Part 1(b): Defying the Laws of Nature

 Rodman had an almost unnatural ability to dominate rebounding on both ends of the court simultaneously
 Rodman showed no tradeoff between offensive and defensive rebounding rates

 The tradeoff between offensive and defensive rebounding exists, and is generally a completely separate phenomenon from rebounding ability.

Part 1(c): Rodman vs. Ancient History

 Contrary to popular opinion, Rodman was a much better rebounder than Wilt Chamberlain or Bill Russell, and it’s not close

 Total Rebound Percentages for players prior to 1970 can be estimated with extreme accuracy

Part 2(a)(i): Player Valuation and Conventional Wisdom

 On induction, Rodman will be the worst scorer and the best rebounder in the Hall of Fame
 Rodman scored even less than we would expect the lowest scoring Hall of Famer to have
 Rodman was an even better rebounder than we would expect the best rebounder in the Hall of Fame to be
 The Hall of Fame likes pointscorers better

 Everyone uses statistics, yet no one listens to statisticians—in part because statisticians build overreaching models, then believe and defend them
 Rebounding percentage correlates more strongly with winning than points per game.
 Added: Individual rebounding percentage has a more causative effect on team rebounding percentage than individual PPG does on team PPG.

Part 2(a)(ii): Player Valuation and Unconventional Wisdom

 Player Efficiency Rating ranks Dennis Rodman as the 7^{th} best player on the 199596 Bulls championship team
 Player Efficiency Rating is terrible
 Extreme rebounding ability like Rodman’s may have exponential value

 Player Efficiency Rating fails completely as a predictor of true player value
 PER rewards Usage rate (shooting), despite no correlation between Usage and shot efficiency
 PER’s many layers of complications and adjustments are demonstrably counterproductive

Part 2(b): With or Without Worm

 Rodman has the highest Margin of Victory differential of any player since 1986 with a remotely similar sample size
 Rodman’s value comes mostly from extra possessions from extra rebounds
 Despite claims that he was exclusively a defensive player, Rodman’s teams played significantly better on offense with him in the lineup, even after accounting for his offense rebounding

 Introduce my brand of gamebygame “With or Without You” stats
 Two main areas of player impact: Reciprocal Opportunities, and Reciprocal Efficiency
 Impact on one aspect of a game can easily be reflected in other areas statistically
 Completely independent confirmation of the results of previous analysis is powerful evidence

Part 3(a): Just Win Baby (in Histograms)

 Rodman’s Win Percentage differential is even better than his Margin of Victory differential
 Specifically, his Win% differential is #1 of the 470 players who qualified for the study—by a wide margin
 Adjusting for the quality of teams Rodman played for makes his differential even better

 Introduce win percentage differential, which is incredibly useful for research and hypothesistesting in many contexts
 It is harder to have a big impact on better teams
 In a game of small margins, exceptional performance in limited areas can be more valuable

Part 3(b): Rodman’s XFactor

 Not only is Rodman’s Win% differential greater than his (already great) MOV differential, it is greater by one of the largest margins of any player
 After adjusting for sample size, Rodman’s XFactor is by far the largest of any qualifying player
 The most plausible explanations for this disparity suggest that, in Rodman’s case, his Win% differential may be the more trustworthy metric

 “XFactor” is the difference between MOVpredicted and actual Win% differentials
 Higher MOV’s and larger sample sizes should correspond to smaller XFactors
 3D plots are a visually appealing way of identifying lessobvious outliers
 There are several plausible causes for real team and/or player Xfactors

Part 3(c): Beyond Margin of Victory

 Using a standardized model for combining MOV and Win% differentials (which weights MOV more heavily), Rodman still places #1 in the set of 462 qualifying players (many of whom have MUCH smaller samples)
 Depending on which metric you favor, Rodman’s differentials place between the 98^{th} percentile and the 99.98^{th} percentile among fulltime players (approximately 5% make the Hall of Fame)

 The statistical community overvalues Margin of Victory and undervalues raw winning percentages
 Winning is a provably existent skill, separate from scoring and allowing points.
 Predicting regular season win expectations is best done with a combination of Margin of Victory AND Win percentage
 The larger the sample size, the more heavily Win % should be weighted

Part 3(d): Endgame: Statistical Significance

 Rodman’s win differentials alone are statistically significant well beyond the 99% level of confidence.
 Looking at the overall statistical significance for player win differentials over the broadest possible pool of 1539 players, Rodman ranks between 2st and 8^{th}, depending on your preferred metric.
 Rodman’s average ranking across metrics is second only to Shaquille O’Neal (who’s sample includes over twice as many qualifying games).

 Introduce “Black Box” as a term for when variance gets eaten up by events that have binary outcomes.
 Standard deviations for win differentials can be found by sampling
 These can be adjusted to different sample sizes mathematically
 Using this process, we can measure the statistical significance of individual win differentials for many players who didn’t have sufficient samples to qualify for earlier comparisons

Part 4(a): AllHall?

 There are indirect reasons to believe Rodman’s Win Differential is more reliable than his Margin of Victory Differential

 Occam’s Razor and Bayes Theorem reasoning can make unlikely or only mildly supported independent hypotheses much more likely

Part 4(b): Rodman v. Jordan 2

 Rodman’s unusual consistency and otherworldly impact per possession used suggest that he may have had even more value than his Win Differentials indicate

 Introduce “Invisible Value” and “IFactor”

Awesome, I had forgotten how much information was in the earlier posts, great series all around. What can we be expecting next?
I’m glad to see that our conversation from the firm holiday party made it into not one, but several blog posts. I wonder how Kevin Love’s stats this season — closing in on the consecutive doubledouble mark, leading the league in rebounding (per game and per 40) — compare to those of the Worm. Though Love’s accomplishments are pretty impressive, maybe he is just stealing rebounds from his teammates. And he sure can’t be contributing much in the win column.
Hi Eric,
I actually had a long conversation about Love at the SSAC yesterday. He definitely seems to be having a statistically rare season. The biggest problem with him is that, in the NBA, great players pretty much don’t lose ever. He’s going on his third straight season with a winrate of <30%. The only marquee players to do this in any seasons over the past 30 years are Dwyane Wade (who did it once) and Pau Gasol (twice).
[…] Rodman owns by a large margin the highest career rebounding rate, a metric he dominated more thoroughly than Jordan did scoring. Though I’m admittedly out of my element here. Smarter people than me or Pesca have hashed out the Rodman debate, and if you fashion yourself one of these, I’d direct you to this labyrythian minefield of numbers. […]
Hi guys, new to the site. How would you respond to the assertation that RB% does not work for “role players” (can be defined as players who don’t play the most on their team) since they can expend more energy on rebounds/get more rest, etc. Rodman didn’t necessarily play the number of minutes you’d expect from an elite player; that said, he still compiled otherwordly ratestats (RPG, RP40, etc). Do you feel RB% has this limitation?
Yes, I would say RB% is definitely affected by role, as is virtually every other stat.
[…] truly wonderful statistical analysis making a case for Dennis Rodman. One of my favorite lines: Before Rodman, we should have expected a rebounder of that quality to […]
[…] is a case for Rodman as the greatest player in NBA history. It’s worth reading it all. It has a dozen different charts, measuring different aspects of […]
[…] is being misrepresented (and have a few days to spend being entertained) go read Ben Morris’ Rodman Opus and don’t miss the bit about rebounding being […]
[…] related. His value as a baller is a great topic, and to learn more about that I direct you to a series of articles far better than anything I have ever written. This article, though, will mainly focus on […]
I’ve looked at this (but not read every single word) but how do you go from his point margin difference with and without, which is barely ahead of others, and come back with the incredible high winning %?
Basically I agree through 3A in the chart above, but then I come to a leap that I don’t get (yet)
thanks
I’m with you through 3A. How do you go from a point margin barely better than number 2 to a winning % far better than anyone else?
That’s my problem with this so far, and the leap I want to understand
thanks
The Win% differential is calculated separately using the method described. It isn’t projected based on points, it’s the actual difference in his teams’ winning percentages with and without him.
Rodman’ benefits greatly from a .381 differential from the 95 pistons, which had the 20 game minimum you used.
Had you used 21 games for your cutoff (1/4 of the season, rounded up) you would have different results.
Fair enough, but I don’t think the point is particularly probative: His differential was high with every team he played for, and can drop a data point or two and still be one of or the best ever.
I don’t have the data in front of me, but to dampen outlier seasons for whatever reason (such as age, or, incidentally, a better example is Ron Artest, whose suspension after the “Malice in the Palace” coincided with many teammates and inflates his career differential substantially), I tried dropping each player’s “best” and “worst” impact years, and it had little effect that would be relevant to this analysis (aside from dramatically shrinking the pool of qualifying players).
You also get MJ who has only one qualifying season – 2002 Washington after 2 retirements where is differential is still .182.
His return season 1995 had 17 games and he had a differential of .242 – if I weight those 2 you get .208 – and neither season was his peak.
This is such a cherry picking exercise that it doesn’t really mean a lot.
Meh, you have to draw a line somewhere. I was acutely aware of the fact that adjusting the filters a small amount one way or another would include or exclude certain players and/or be more or less favorable to them. So I tried to set lines that were general, rationalizable on their own terms, and as effective and probative as possible with my particular goal in mind: evaluating Dennis Rodman. There’s a constant tradeoff between breadth and accuracy, and I went with a balance that I thought best for testing my hypothesis about Rodman having HOFlevel value.
If you read the last post in this series, you know that I agree MJ’s Win % diff is probably higher than Rodman’s (though, given sample sizes, it is not more statistically significant). I have also noted in several different places that I think Rodman has certain advantages in these comparisons: For example, he missed a relatively high percentage of games due to suspension rather than injury, and he didn’t stay around in the league several years past his prime as many other great players did. But the fact that there are large margins of error is offset by Rodman’s statistically extreme performance. The whole point of showing that he comes out on top in this reasonably broad pool of players isn’t to show that he is actually the best, but to show how unlikely it is that he wasn’t very very good.
well put statement. Good job
I’m not impressed.
In the “Outlier in motion” graph, Rodman seems to make a huge statistical leap in rebounding prowess between the 1990 and 1991 season. What’s happening there? Did his MOV or win% diff also change noticeably (assuming we have a large enough sample size)?
If we’re looking for the bizarre loophole, maybe looking at other examples of large statistical leaps from year to year+1 would yield worthwhile insights.
I also wonder if Rodman can recall any conscious change in his approach or strategy for games during those years. Worth asking him before he gets locked away in North Korea? :)
[…] women’s hardcourt players over 30 so small sample sizes and all that, but it’s still a Rodmanesque level of […]