Major flaw in SportVu data?

[pmaymin]

Adam Yudelman just wrote a piece for us at Vantage Sports comparing the optical "contested" numbers with our own definition.

Would love your reactions and thoughts.

http://www.vantagesports.com/story/Vg1Z ... hot-metric

[Nate]

Mostly, I think it speaks to the subjective nature of these stats and the limitations of the data sets: Just like with assists, there's no real consensus about what constitutes a 'contested' shot. How much it takes could easily matter by player too - someone who doesn't regularly take outside shots might be more sensitive than someone who is experienced.

Ostensibly, someone with access to the raw SportVU data could try to train a categorizer that matches up better to the Vantage standards, and distance just happens to be an easy way to do things with the SportVU data set. The issue here is less with the SportVU data proper, but with the interpretation of that data that's being presented to us.

SportVU may also be able to retrodict contested shots from ball tracking: after all, the whole point of contesting is to interfere with the shot.

[bchaikin]

the interpretation of the sportvu data is still in it's infancy - for example in their 2013-14 data there were a couple of frontcourt players who were among the league leaders in backcourt touchs. i came to find this was because they included inbounding the ball after the opponent scored as a touch. i dont know if they still do...

in their 2 dimensional world i'm guessing they attribute the "defender" as being the closest dot of the 5 dots on defense as the defender to the dot having taken the shot, and that may not in fact be the case...

but it is data, it is available, and they do attempt to quantify on defense FGM/FGA allowed for each player, and these can been seen at nba.com. as far as i can tell the data does not take into account the players on offense guarded (i'm sure it will at some point), i.e. it does not differentiate if a defender routinely guards the opponent's best or worst shooters (or #1 vs say #3/4 option on offense), but it does list 2pt and 3pt FG%s allowed on defense, and seems to try to account for all FGM/FGA...

and the 2014-15 data does show low 2pt FG%s allowed on defense for players like tony allen, draymond green, and roy hibbert, and high 2pt FG%s allowed for players like jose calderon, andrea bargnani, and anthony bennett, which for the most part matches perception...

is there a publicly available way to compare the posted defensive 2/3 FG%s allowed by defenders at nba.com to those of Vantage?...

[bondom343]

It makes some inherent sense, because it isn't measuring who is defending the shooter, but "closest player" which is entirely different. I was always a little leery of putting too much faith into the SportVu shot defense data and this was why, and the article backed up some of what I had read/interpreted before.

[schtevie]

My first reaction and thought is where is the corresponding FG% data? That opinions can differ dramatically on the definition of a contested shot is very interesting, but which measure correlates better with the (presumed) negative effect on FG%. That's what is of ultimate interest, no?

[Kevin Pelton]

Yeah, I was curious about that too. I went back to a couple of pieces Krishna Narsu did for Vantage a couple of seasons ago on the topic. First, here's how they define "contested" vs. "pressured," with the latter seeming to square with the shots marked as contested in public SportVU data that are being contested (as in disputed) here:
https://web.archive.org/web/20150318123 ... three.html

Then here is where Krishna broke down shooting percentage by type of shot defense:
https://web.archive.org/web/20150315032 ... fense.html

It appears having a hand up does matter around the hoop but is not so important on jumpshots.

As for distance as measured by SportVU data, Seth Partnow has written about the general bins of nearest defender in terms of its impact on shooting percentages, including in his response to this post: http://nyloncalculus.com/2015/10/01/mou ... tvu-issue/

[pmaymin]

Thank you all for your thoughts and comments.

FYI, some more optical facts and responses in this follow-up article:

http://www.vantagesports.com/story/Vg7l ... o-the-best

[bondom343]

Thank you all for your thoughts and comments.

FYI, some more optical facts and responses in this follow-up article:

http://www.vantagesports.com/story/Vg7l ... o-the-best

Interesting rebuttal, and honestly seems to be good evidence though limited in scope. I expect further challenges from those who disagreed, but digging deeper would be really conclusive and possibly damning.

[Crow]

Optical based data not the simple, firm answer? Huh.

What to do? One option, look at the efg% allowed rapm factor estimates. It is focused not on debatable, imperfect measures of body positions and conjecture about relevance but rather the actual shot defense impact under study using just simple, firm play by play scoreboard data (did it change or not?). Behind that scoreboard measured impact would be body position, where shots came from, time on clock, etc. Anything that appears to have an impact on efg%. It is just an estimate of the impact (randomness involved too) but these other methods are further from the targeted knowledge and quite tenuous in meaning, imo and others.

Best approach is to look at all of it and apply some judgment. Anybody want to line the three measures up comprehensively? Possibly create a metric blend? Just going separate ways in disagreement would be somewhat disappointing but it is a fairly likely outcome.

[Nate]

This sort of question is legitimate and interesting, but it's worth remembering that these articles are being published by Vantage, so some bias is ... inevitable.