Juany8:
The clearest way I can show that the way assists are evaluated is absurd is what happens if you freeze the camera after a shot goes up. Say Kidd just passed to an open Novak from the corner and Novak took the shot. If you simply stopped the game right there, before the shot either hits or misses, the only thing that has happened so far is that Kidd has made a pass and Novak has made an attempt.

Kidd doesn’t have any impact on whether the shot goes in or not at that point, but he still made the same pass regardless. Since Kidd is not responsible for Novak’s shooting, why should the pass he made be evaluated differently after the fact? Making a pass to an open Novak in the corner is a good basketball play, so why is Kidd getting credit for being a good player in one scenario and not the other? He made the same play!

I agree totally. Mainstream basketball statistics are simply flawed representations of observation, as you point out.

Kidd doesn’t have any impact on whether the shot goes in or not at that point, but he still made the same pass regardless. Since Kidd is not responsible for Novak’s shooting, why should the pass he made be evaluated differently after the fact? Making a pass to an open Novak in the corner is a good basketball play, so why is Kidd getting credit for being a good player in one scenario and not the other? He made the same play!

This “reality check” is a necessary test for any model. Physical observation is imperfect, especially for vast quantities of data, but you have to physically be able to see something for it to be true. For example, WP, WS, and PER all give full credit for a shot to the person who took the shot. They also give credit for an assist on any made shot. Therefore, on a possession that results in an unassisted basket, the only player that accumulated value in the possession is the shooter, and he basically accumulated +2 value. On a possession with an assisted basket, however, the shooter is still getting his +2 value, but an extra player is accumulating some value via the assist. For that to make sense, it would have to mean that the individuals on a team are getting more credit for making a basket than the team is. Since I’m pretty sure an assisted basket is not worth 2.8 points or something odd, this means that the way assists are evaluated has no basis on anything resembling basketball.

(cont. on next post)

The Honorable Cock Jowles: Any study using PER is worthless in my book. The consistency of data comes from WP48, which you can read in Wages of Wins or Stumbling on Wins, each of which have a bibliography you can skim at your leisure. I’m not interested in arguing with people who, for example, use five games of James Harden being inversely awesome and average to make a claim about his “little tricks” being figured out by defenses and consequently less effective.

Average?

– Founder of the MIT Sloan Sports Analytics Conference.

http://www.sloansportsconference.com/?p=58

Morey holds a bachelor’s degree in computer science with an emphasis on statistics from Northwestern University.

The Honorable Cock Jowles: And something that, unfortunately, is going to be considered persuasive evidence for those who think that team role and supporting cast are significant factors IF it supports their claim. If his number stay relatively constant, I wouldn’t consider it persuasive evidence one way or the other. He’s just one player. And we sit here and watch Juany8 talk about tiny sample sizes (of one game) between Harden’s first and last games this season, as if that’s indicative of any larger trend.

it’s not about a trend, it’s about accounting for the strength of the opponent when evaluating players. Going up against the Heat is not the same as going up against the Pistons, and I’d argue that the effect is not the same for any player. A heavy pick and roll player like Harden is going to have a more difficult time dealing with excellent trapping teams than an isolation/post up player like Melo. Or if the opponent only has 1 elite wing defender, the top player is going to be bothered more than the second banana. There’s just so much situational context for basketball, you can’t just blankly say a player shot X% over a season and pretend it means anything

The Honorable Cock Jowles: And something that, unfortunately, is going to be considered persuasive evidence for those who think that team role and supporting cast are significant factors IF it supports their claim. If his number stay relatively constant, I wouldn’t consider it persuasive evidence one way or the other. He’s just one player. And we sit here and watch Juany8 talk about tiny sample sizes (of one game) between Harden’s first and last games this season, as if that’s indicative of any larger trend.

Anecdotal evidence is still evidence – especially when you consider that Harden’s specific situation will never have enough of a sample size to satisfy any rigorous statistics — ie. how many times have we seen:
1) super high efficiency high usage bench player
2) in his 4th year in the league (ie. before a bigger sample of his performance is available)
3) who had been playing alongside/with 2 high-usage All-Stars who definitely diverted defensive attention away from him (whether that makes any difference is another question altogether)
4) get traded to a team with no other high-usage players (we’ll see about Lin)?

I can’t think of a single other case off the top of my head. These things just don’t happen. And so for all intents and purposes, this is all the evidence we are going to have for this specific situation. This is totally different from Ray Allen going to Boston, or Shawn Marion going to Miami, or even Steve Nash going to LA.

For cases like this, there shouldn’t be a null hypothesis. Why should we assume he will be the same player in HOU as in OKC? There is no evidence at all to suggest that a player like him that moves from the situation he was in (banana 2a or 3 in OKC) to the situation he’s in now (1st banana in HOU) will be better/worse/same than he was.

jon abbey:
one thing that’s awesome about the Harden move is that it’s a fantastic case study about something we’ve been arguing about for a while now.

And something that, unfortunately, is going to be considered persuasive evidence for those who think that team role and supporting cast are significant factors IF it supports their claim. If his number stay relatively constant, I wouldn’t consider it persuasive evidence one way or the other. He’s just one player. And we sit here and watch Juany8 talk about tiny sample sizes (of one game) between Harden’s first and last games this season, as if that’s indicative of any larger trend.

Frank also brings up a pretty important point, in that averages don’t tell you anything about the distribution of data. If someone gets 20 rebounds one night and 2 the next, it’s not the same as someone who rebounds 11 on consecutive nights. In the case of WP, I buy that player statistics overall stay fairly consistent from year to year, but that could easily mean that a lot players stay really consistent but many don’t. It also doesn’t help WP that most players don’t move teams, and that WP has a lower year to year correlation for players that switch teams than for those that don’t.