Exactly this, and this again.

That said, ESPN the magazine has managed to figure out clubhouse chemistry with numbers! Well, numbers and racism:
http://assets.espn.go.com/magazine/0331TEAMAL.pdf
http://assets.espn.go.com/magazine/0331TEAMNL.pdf

It may just be some sort of elaborate joke. A joke on me, personally.
I know it's about baseball but for ****'s sake, stat nerds. Get your **** together. You're making us look bad.

Well, on a surface level, the data seems to suggest that this isn't true. If steals are generally associated with negatives such as compromised team defense, the regression should account for that. The numbers at the end also do suggest, in a way that seems convincing to me, that "team defense" actually doesn't play a major factor in steals, and that they appear to be almost solely a measure of individual ability.

-

The major issue that affects the reliability of the estimates, as far as I can tell from the article, is going to be the omitted variable bias that results from a model that just uses steals, assists, points, rebounds, and blocks to model basketball. "One steal" appears to be measuring not just the impact of one steal in a single game of basketball, but everything associated with a player averaging 2.0 steals vs. one averaging 1 steal (such as superior defense on non-steal possessions) that the model wouldn't otherwise know about or account for.

Perhaps the other way around?

The results discussed in the article focus on the relative value of box-score stats and why. Obv there is a lot more that goes into assessing value of NBA players than those things.

Isn't his dependent variable some variant of plus-minus? Therefore the bad side effects of going for steals (i.e. gambling) would be accounted for in the regression.

edit -

Variance in a player's scoring efficiency does account for variance in their impact on winning more than the variance in their steal rate (I don't think the article claims otherwise). But I imagine not by as much as people would guess.

Turnover correlation not positive as mentioned in footnotes.

You would be right that, taken in a vacuum, turnovers can have a positive predictivity (like if you did a regression on FG% and turnovers), but as long as all variables use the same scope it's not a problem.

that's fine but now you are the one fitting the narrative

I think he did a fine job letting the data speak for itself, its just a matter of if what he did is valid and/or if he is drawing the appropriate conclusions

Ben, just write better articles man. I think I speak for everyone when I say we all are pulling for 538, and very excited for the potential+. I certainly am in the support boat, as in I will read your stuff and so on. This article is better, but a lot of holes still remain

Article was more precise with language regarding this than I'd remembered in the first half. (i.e. specifying that the "value of a steal" was measuring its predictive associations rather than "a steal.") Fair, apologies.

I think a lot of the issues I had in interpreting the article could've been clearer if you'd showed the relationship from each stat to "team performance" instead of "relative to one box-score-average ppg" at the beginning. You briefly acknowledging, then moving on from, the possibility that underlying relationships (rather than irreplaceability or similar factors specific to the steal act itself) would be the major driver of the huge coefficient estimate didn't make sense at all to me my first couple readthroughs.

It made a lot more sense when I worked out the rough mental reconciliation of "each ppg worth ~.17 points based on reliability charts, then that multiplied by 9.1 = 1.52 points, which is in the ballpark of the theoretical value of .96*a steal". And then you address the innate value of a steal (.96 * a couple points) late in the article, but it doesn't explicitly tie back into the "a steal is worth nine points" element of the beginning, which -- imprecise reading or no -- a lot of people here zoomed in on.

Based on the date, I thought it was an elaborate April Fools joke, slightly premature because of set publishing dates, but apparently it isn't.



I hope Adam Silver doesn't catch wind of those jerseys. Not sure why they are dribbling the ball with a stick, though.

Hypothetical to help me grasp this a bit better: Reginald van Pickypocky, a 6'5 SG, is a league-average starting wing in every way--except for his steal rate. He averages 7.5 spg in ~30 mpg of play without compromising much at all on non-steal possessions.

He is roughly the n^th most valuable player in today's NBA. n = ___?

Sorry, not trying to bog down yall's thread with nitty-gritty reactions. Most of the non-kneejerk responses seem fine to me, if a bit focused on stuff that is consciously outside scope of article (though maybe some room for clarification on things like: duh, turnovers are bad).

But your theory about Ricky Rubio's involvement in the research/writing/editing process is, um, cray cray.

Some counterpoints to Rubio:
-The guy who has filled in (getting his minutes when out) has been JJ Barea. I don't think Barea is terrible, but he's not much better than replacement value. With any "with or without" raw study, it's appropriate to look at backups rather than gift someone into "clear +impact!"
-Rubio is pretty good. But I don't think it has much to do with his steals!!! In fact, the gambling nature of the Wolves' defense is actually pretty bad. Other conventional boxscore numbers can speak for themselves for Rubio; for example, he's a brilliant passer, and he rebounds well for a PG.
-What about guys who get steals simply for steal sake? For example, Monta Ellis. He's an oft-cited guy whose performance has completely waned and waxed inversely proportional to his steal rate. There's a reason for this.
-Lastly, TONY ALLEN GOAT.

Fine. Whatever. But that's how it appears to me. If it wasn't like that, ok.

I want to form a team w/ Reginald Van Pickypocky and Robo Rebounder. Would win all the championships

I don't think he was alleging that Ricky Rubio was involved in the writing process, so much as it seems that just like with the Rodman series you've constructed a statistical justification to support your thesis, rather than the other way around.

lol

It doesn't say that afaik. I just mentioned it as an example of a variable that is both very important to team performance and was not directly included in the model (and thus would theoretically flow through to the other variables, affecting estimates and possible interpretations).

At the end, I see why you were satisfied the model was reaching reasonable conclusions that predominantly reflected "the value of a steal" and not "the value of the bundle of attributes that is associated with averaging one more steal with points/rebs/blks/tos constant." Because, in the end, the estimate on how much a steal was worth accorded with its replaceability * its inherent value. But imo the conclusions at the end could've been more clear/cleanly tied back in with the model partway through, which seems to be the source of the majority of issues.

I still feel that "one marginal individual average ppg", though it made sense in theory, was just not an intuitive unit of measurement. It's prone to misunderstanding by people confusing it with a marginal point for the team and it lacks the immediately obvious context and interpretation of "team SRS".

Like you framed the story:
OK, for one thing, you never get around to explaining what is mysterious about the Wolves(they lose a ****load of close games?) or what is curious about Rubio(that he is good at some things but not good at other things?).

Actually, I'm just going to leave it there and focus at one thing at a time. What the **** are you even talking about?

oh thats the guy everyone dumped on for the rodman article?

didnt know he was a 2p2er

im glad someone else brought up the 'steals' article

pretty clear the author liked whatever his model was but also knew that it was deeply flawed and therefore did not even try to explain how a steal could be worth 9 points

also the replaceability premise mentioned didn't talk about fouls

so deeply flawed article


i do think some of the other articles have been improved though

1) Rubio appears to be very valuable to the Timberwolves based on their performances with and without him
2) Metrics based on his box scores don't seem to capture the true value of Rubio, who is statistically extreme in a couple regards
3) Could these metrics be undervaluing what Rubio does well (steals), especially in relationship with what he does horribly (score)?
4) Investigate that issue
5) Conclude

You can take issue with 4) or 5), but 1-3) are a standard lead-in to a statistical article.

That's the thing there's clearly some chops there, the middle third of the article deserves to be fleshed out (with a more critical eye towards the conclusions), but the Ricky Rubio framing device is just terrible.

also really have no idea how the articles are being edited

that steals article should not have been published in that manner for those reasons