NBA HD: Dismantling the Assist

If I were a stat guy, here is what I would track instead of the "assist" stat as we have it now.

1. Quality Shot Setups - Every pass that leads directly to a quality shot (a quality shot is basically an open shot or a shot that results in a shooting foul). You would also want to include passes that were mishandled by an obviously wide-open teammate in position to score. These don't happen a lot, but certainly a few times per game.

2. Count assists for FTs if the pass led to the shooting foul.

3. If you want to get crazy like this article, you need to keep track of the player who took the shot on every potential assist. If the player took a bad shot, this is partly the passers fault for passing him the ball when it wasn't a good scoring opportunity(he should know if his teammate has a tendency to take bad shots). You also need to track where that player was setup, so you know how good of a position it was for them personally to score. Using per player stats you can get a perfect picture of the shooting % gains.

A discussion about this over at APBR metrics:
http://sonicscentral.com/apbrmetrics/viewtopic.php?t=2532

hey,that's not Gasol, that's Radmanovich.

So, I'm not sure I'm totally getting this here. Here seem to be the two different ways to count assists.

1. Only count potential assists when they actually lead to made baskets.

2. Count any potential assist, but weight it according to the expected number of points scored on the resulting shot.

So, let's say Player A passes to Player B, who fires up a two-point shot that has, for the sake of argument, a 55 percent chance of going in. By reckoning number 1, Player A will get a full assist 55 percent of the time. By reckoning number 2, Player A will get 1.1 assist all of the time. On average, reckoning number 2 yields a result that is exactly twice as high (as a result of being counting expected points rather than expected baskets).

It seems to me that (outside of counting treys as 1-1/2 times as valuable as deuces) what this does is just reduce the variance on the result. Over the course of a season, for instance, you'd expect the two reckonings to be pretty well correlated (modulo the treys again, natch). Is that so, or am I missing something else?

It also seems that what one could do here is account for the opportunity cost of making the pass. Player A could have shot the ball. In some cases, Player A's shot might even have been better. You don't want players who, whatever happens, DON'T take the shot, but always pass. Accounting for the opportunity cost would reflect that. Of course, this would be a heck of a lot of work to encode in processing video--and highly-paid human work, too, is my guess. But maybe some of it could be done by grunts. :)

great assist articles...finally, someone who feels the same way i do....exceptional work!

I like the idea but the result of the calculations seems irrelevant. of course dunks are going to be high percentage shots no matter if they were assisted or not. i'd bet more dunks are assisted than not though. if you can't take the dunk into account, and you can't take dribble drive kick-outs into account accurately than the formula is flawed from the get-go. assists are hard to measure. how about giving screeners assists for getting guys open? rip hamilton runs around screens all day and all it takes is a 5 foot pass from the top of the key to him to get an assist. Who created that shot? Hamilton, the passer, or the screeners? Its just too hard to quantify.

Jon, this is a followup to the article last week where he re-weighted assists to correspond to the league average field goal % from the area of the court they came from.

In addition to what the above poster said, it seems to me that this metric is just as dependent on teammates as regular assists, and perhaps more so.

Replace Dwight Howard with Chuck Hayes, and suddenly all of Jameer Nelson's dunk assists are "close shot" assists. As the above poster points out, this would actually hurt the team, since those close shots aren't converted at anywhere near the rate that dunks are.

I'm not sure I can really trust the rough metric used in this article because it devalues assists that lead to at rim shots and three pointers, the two most effective shots in the game. At rim shots are the easiest baskets you can get and three pointers have the bonus of higher scoring reward for making it. How much stock can you really put in a system that rewards playmakers for making passes to teammates for tough shots (thus the disparity in potential versus actual for 2pt jumpers and close shots) while punishing them for creating easy buckets?

While I really admire your work, Haberstroh, I think your system of valuing assists needs a lot more work before it can be taken seriously. The true measure of an assist is whether it creates an easier shot for your teammate, and just how much easier you've made the shot. Simply rewarding point guards for potential shots is silly since it is much easier to create lots of bad shots that are potential assists rather than getting a few easy buckets.

Shouldn't there be a way to factor in the extra worth of three point shots into this assessment when considering the value of assists? While some kick-and-drive passers like LBJ and D-Wade do pass to the three point shooter quite a lot, should we ignore the added value eFG% gives to such shots?

Also, the three point % difference (3.7%) is probably skewed by the fact that most players are not good at getting a clean look from three unassisted (thus why 81% of those get assisted), and by the fact that the players who take the most unassisted threes are guys like Nash and LBJ who are great at getting themselves space and creating a good shot for themselves from three.

So I think it's a little unfair to dismiss the value of assists from kick-and-drive players based purely on this analysis.

The series on passing has been phenomenal, and that article on 82games was a great find.

The 82games data takes a large view on something that we could also take a micro view of if we had a serve (to run with Jason's idea) statistic. Right know we can track how many of a players shots are assisted. With serves we could also see how many of an individual’s attempts are assisted, and how that impacts their shooting percentage. The implications for assessing player value with potential assists make the assist stat look outdated. I hope the potential assist takes hold in stats.

On a side note I believe that the dunk numbers aren't an ‘apples to apples’ comparison, because on assisted dunks, there is little chance a man with the ball would have gotten to a position to take the shot. The same can be said for drive and kick (or weak side) treys. If someone doesn’t force the defense to collapse first (or doesn’t force the defense to give help strong side, leaving the weak side corner open), there’s no way the shooter gets the open look for the shot.
There are examples of bad three shooters who will jack up contested threes consistently, and those players tend to drive their fan bases crazy. They are the exception in the league (as the 81% assist rate on 3’s shows) but they are a good counter example for minimal value of the assist on a threes. I think that both the three point and dunk numbers are undervalued, because in those cases the ball handler typically does more to set up the shot than just pass the ball to the shooter. I’m not sure how to take that into account, but I’m sure someone smart out there has some ideas.

In addition to assists, which record passes that lead directly to made baskets, why can't we also track passes that lead directly to shots? I don't know what we would call this statistic (serves?), but it would seem to capture what we're missing: passes that lead to scoring opportunities, as opposed to passes that lead to successful scoring opportunities.

I would assume that, in general, if a pass leads directly to a shot, then it's probably a decent scoring opportunity, or at least an improved scoring opportunity. This obviously isn't universally true — players take bad shots every game — but it's probably true on the whole, and anyway the goal here isn't to measure whether any given shot is good or bad. After all, good shots can miss and bad shots can go in. The goal here is to track scoring opportunities, and to reward those who set them up, regardless of the outcome.

I would love to be able to look at a player's serves and assists side-by-side, and compare players' serve/assist ratio, from which I'm sure we could draw all sorts of conclusions. I would also be curious to find out if there is "serve parity" — that is, if all guards (or forwards, or centers) have roughly similar numbers of serves per 48 minutes, simply as a virtue of playing their given position. For example, it might be the case that Jason Williams and Steve Nash each average 23 serves per 48 minutes — yet, Steve Nash leads the league in assists per 48 at 16.1, whereas JWill only averages 8.5 per 48. This would imply that Steve Nash, although he creates the same amount of scoring opportunities as JWill over the course of 48 minutes, in fact creates better scoring opportunities. Or it might tell us that Steve Nash has better shooters for teammates. On the other hand, maybe we'd see that certain players like Chris Paul and Steve Nash have a significantly higher number of serves per 48 than other guards in the league, but roughly the same serve/assist ratio as other guards, suggesting that their high assist numbers are the result of their high serve numbers, and not because the scoring opportunities they create are somehow more efficient. On the contrary, the conclusion might be that their ability to create any scoring opportunity (not just a more efficient one) is better than that of the average guard. I obviously have no idea what we would see, and obviously the serve statistic wouldn't be perfect, but it would be fairly illuminating, and very interesting.

Great post.

Actually the 82games post discusses my beef with your points and more.

Why aren't passes that lead to foul shots given an assist?

One way I think this is fairly flawed is the likelihood of getting a certain shot. I'd think it's difficult to get a dunk opportunity if you're not set up by an assist. So, even though your FG% increases only slightly (because dunking is such a high % shot to begin with), I'd think that the alternative to an assisted dunk is probably a lower percentage shot. The three-ball is much more interesting, because you can get an unassisted three point look nearly any time you want, so there I can see how if it's only a minor increase in FG% that assist may be less valuable.