On The Interpretation Of Assists

[rshea]

STATS LLC tracks potential assists using their SportVU technology. This report is from over the summer on some of the more interesting findings: http://www.stats.com/pdfs/NBA_Insidethe ... ations.pdf

[Mike G]

Hmm, that's an interesting link.
However, something is fishy:

Traditional understanding of sharing the ball is confirmed by SportVU metrics. When the ball is being shared and players are shooting and increased percentage within potential assist situations. The most noticeable difference occurs on shots at the rim and the impact drops off as the shooter moves away from the hoop.

Code: Select all

Potential Assist FG% versus Overall
Distance    Overall FG%   Pot'l Ast FG%   Difference
0-5 Feet     .574             .814           +.241  
5-10 Feet    .410             .578           +.168
10-15 Feet   .394             .492           +.098 
15+ Ft 2pt   .393             .433           +.040
3-Point      .347             .385           +.038

We see the overall FG% from each distance, but we don't see the unassisted FG%, nor the fraction which were potentially assisted/unassisted.
So the assisted FG are thrown into the Overall FG%?

Is a potentially assisted 3ptr really just .038 better than one that's unassisted?

If only 1/4 of all 3FGA are unassisted, then these unassisted shots have a FG% that's lower by 4 times that .038 -- .152 lower than .385 is .233

(.233 + .385*3)/4 = .347
That's 3 shots at the assisted % for every one shot at the unassisted rate.

3-point leader Steph Curry has been assisted on 63% of his made 3-pt shots.
#2 Ryan Anderson is assisted on 95%, Klay Thompson 94%, Kyle Korver 98%, Danny Green 97%

[Mike G]

Anyone have insight on this? ^^

[v-zero]

I would interpret the overall FG% numbers as including the assists, and as such the numbers are pretty useless without the fractions of assisted to unassisted shots, but that doesn't surprise me as I find a lot of Synergy/SportsVU stuff is badly done and relies too much on 'logic' and not enough on actual regression analysis. As for the threes - it wouldn't surprise me to find that unassisted three percentage is pretty low, since a contested three will often be an off-balance shot, whereas an assisted three is often a kick-back or fast-break opportunity and is about as close the shooting practice as you get in-game, outside of trips to the free throw line.

[Mike G]

And so a statement like "...the impact drops off as the shooter moves away from the hoop." may not only be inaccurate; but actually the opposite is true?

Or is the error in calling unassisted FG% "overall FG%"?
Are unassisted 5-10' shots made at 41% success? Or does that 41% include assisted shots?

[v-zero]

What I have found on the value of assist types is that the mid-range assist is almost a misnomer - assisting somebody into a mid-range shot isn't really helping much at all - possibly because the interior is usually crowded. Close assists and three/long two assists are all similarly valuable. My interpretation on that is that a long two might've been a three if the shooter had stepped back, and that both long twos and threes are likely to be open shots when assisted as both often come often from kick-backs and fast-breaks where the defensive team is way off being in place to defend an outside shot. Close assists are valuable because they're the easiest shots, at least that's what I believe.

From that I would say there isn't a trend, simply that mid-range assists are very inferior to their cousins at the extremes of distance.

All in all that study is basically worthless without any elucidation as to the ratio of (potentially) assisted to unassisted shots.

I still stand by the interpretation that: number of assists is an indicator of offensive involvement, assist type breakdown is an indicator of offensive acumen.

[Mike G]

What do you say about the FT that accompany certain shot types?
Close-range FGA, even those that don't "count", are associated with more FTA.
A counted assist for a close FG may be accompanied by a pass for 2 FT, on avg.

[v-zero]

I think they're probably represented in part of the value of close assists, and may well be drawn at a pretty stable rate relative to the rate of successful close assists, in which case their impact certainly would be captured by proper regression analysis.

[permaximum]

Averaga assist worth in the nba is 0.37 or 0.38 points. There are way more useless assist counts than useful assists in box scores.

[v-zero]

Averaga assist worth in the nba is 0.37 or 0.38 points. There are way more useless assist counts than useful assists in box scores.

That's the simplistic view of assists that I'm trying to move away from.