Well, I have a couple challenges. They continue to insist usage and efficiency aren't correlated, so I want them to directly answer how good this lineup would be:
PG:
Rondo (pretend he's healthy), 0.202 WP/48
SG:
Landry Fields, 0.155 WP/48
SF:
Thabo Sefolosha, 0.218 WP/48
PF:
Reggie Evans, 0.203 WP/48
C:
Larry Sanders, 0.218 WP/48
Bench:
Earl Watson, 0.082 WP/48
Jimmy Butler, 0.249 WP/48
Andris Biedrins, 0.164 WP/48
Pretend the rest of the roster is near replacement level. How good would this team be? I used the minutes distribution of last year's Hawks for their top seven guys, assuming things like Rondo playing 2835 minutes in a season, sanders 2657, etc. Over 26% of the minutes were given to guys not in the top 7 of the rotation, so this wasn't a top heavy minutes team; it's a fairly conservative estimate for how this team would play. Even so, based on Wins Produced, this would be projected to win 65 games.
Now my question for Wages of Wins is, does that seem right to you? What objections do you have? (And ignore anything quibbling with the numbers or something unrelated to the spirit of the question.) Can these group of guys who doesn't shoot often win that many games?
The other day I linked to a study that looked at usage and efficiency in the Wages of Wins comment section. Well, at least I tried to: the comment had to be approved by a moderator, meaning if you want to post evidence of the Falsehoods spouted by the website you could be censored. For years, Berri and company have stated that usage does not influence efficiency, that having a "shot-creator" is useless. Well, I found a study that looks into this question and finds that when a bunch of low usage guys are on the court, the efficiency goes down: the Diminishing Returns for Scoring - Usage vs. Efficiency. However, they did not approve my comment, for fear it would upset their preconceived notions.
Here are some of the key portions of the study:
"For each lineup, I used the ORtg and Poss of all five players to project the lineup’s points per possession. First I converted each player’s Poss into %TmPoss, the percent of their team’s possessions that the player typically uses when on the court. Then I took a weighted average of the five ORtg’s, with the weights being each player’s %TmPoss."
"For lineups composed of low usage players, where the sum of the five players’ %TmPoss is less than one, the projection assumes that the players in the lineup will maintain their efficiencies (ORtg) even though they will be forced to ramp up their usages above their typical level. And on the other side, if the lineup is a collection of high usage players with a summed %TmPoss of more than one, the projection assumes that even though the players will have to decrease their usage, their efficiencies will not increase."
"If a 1% increase in a lineup’s summed usage results in a drop of 0.25 points per 100 possessions, what does this translate to in terms of individual usage and individual efficiency? We can translate this pretty simply by multiplying 0.25 by five, which suggests that for each 1% a player increases his usage, his efficiency drops by 1.25 points per 100 possessions."
I would love for Wages of Wins to acknowledge this study because they insist there's no evidence of usage and efficiency correlated. And no quick two sentence dismissal either.
In the vein of academic rigor, I double dog dare you.
Justin,
ReplyDeleteI recently did a very similar study, and I found a quite similar result. However, I would say the exact numbers (each 1% increase leads to a drop of 1.25 points) may not be accurate. My study confirms that the relationship exists and is statistically significant at the 99% level. However, the 95% confidence interval is quite wide, so I think he's getting a bit ahead of himself when he gives those exact numbers.
Regardless, there is certainly evidence that the relationship exists, and anybody who denies this fact is either statistically illiterate or intentionally being stupid.
I would love to see your study. I agree about the interval part. The statistical (old) standard is the alpha level of 0.05, but that can lead to problems. If a p-value is 0.03, that doesn't mean it's an open and shut case.
DeleteI wouldn't accept that study as evidence. Regardless, I came around specifically looking for criticisms of their method, and I am not finding what I am looking for. Mostly, your disapproval of the model just seems to be a personal issue with it's proponents. They may be the arrogant jerks you (and many on the ABPR boards) portray them as. I don't know. I'm only interested in measuring productivity. So far, if I keep in mind the scale of the criticism (noisiness in +/-, box score limitations in WP), I am going to be using their model as a baseline. As I find +/- in it's various forms to be so noisy as to be worthless, the only basis of comparison that is convincing is actual wins. The Sports Skeptic has an interesting take on it, by the way.
ReplyDeletePersonality-wise, they seem to be losing. But I will say this: I am finding their insular little community to have a more scientific approach to the data. The ABPR community seems to have serious difficulty discarding their assumptions, (what the box score misses matters, a model that doesn't agree with my eyes is wrong, usage and eficiency interact on a scale that is large enough to matter against the noise, the "experts" know what they are doing, etc) and has difficulty in applying the same rigor to ideas they like and don't like. I have only read a small sample of your postings, but this seems to prevail here as well.
I hope that was polite. Take care.
Scott, you assume that the box score doesn't miss anything important? Why is that more plausible than suggesting the box score does miss something? What evidence supports the claim that the box score captures everything important?
DeleteIt's been shown in multiple studies that Wins Produced does a horrible job (relative to other advanced metrics, even PER) of predicting future outcomes, particularly when players are put in new contexts w/ different teammates. So use their metric as your baseline at your own peril. The proof is in the predictive pudding, and they've failed on that front repeatedly.
Delete