I’m a big believer in striking while the iron’s hot. With Zach Harper’s breakdown of Darren Collison’s assists lighting a fire under the Hardwood audience, I thought I would take a deeper look at the assist variety.
I’ll begin with the assertion that not all assists are created equal, no matter what the box score tries to tell you. The traditional box score specifies the number of shots coming from beyond the three-point line for each player but the assists column offers no such detail. This shields the basketball world from a more complete understanding of ball distribution for each contest. Let’s pull the veil on that part of the game.
In Tuesday night’s Utah Jazz massacre over the Chicago Bulls, Deron Williams dropped 17 dimes which is a season-high for the former Illini. But did you know that seven of those led to treyballs? Probably not, unless you picked up your daily advanced box score at Hoopdata. On the same night, Raymond Felton dished out 11 assists, with 8 of them setting up for gimmes (layups and dunks). These two point guards aided their teammates to score from high efficiency areas on the floor but this valuable information hadn’t been unlocked in the public domain. Until now, of course.
Depending on a combination of skill set, fellow personnel, and X-and-O’s, every point guard varies on how they compile their assists. Some earn their dimes through dribble penetration and dishing to a cutter to the basket after the defense collapses. This type of distributor tends to yield a high percentage of at rim buckets for his team. Others make a living instead by patiently waiting for the perfect moment to hit their sharpshooting teammates on the perimeter rather than exploiting the porous help defense on dribble drives. These types tend to tally a slew of assists leading to long twos and three-pointers.
So, who are these guys? Let’s compare two of the best assistants out there, LeBron James and Jason Kidd. One is a 6-8 dominant scorer and the other is a 36-year old Hall of Fame distributor. Both average over 8 assists per game. The pie charts display the shot location share of their assists in terms of the five zones: at rim, short (<10 feet), mid (10-15 feet), long (16-23 feet), and threes.
As you can see, these two ball-handlers get their high assist totals in much different ways. Over 75 percent of LeBron James’ assists yield a three pointer or a bucket at the rim, highlighting just how devastating LeBron James can be as a ball-hander. If he’s not scoring himself, he commands the help defense and double-teams, opening up the high efficiency areas for his teammates. Actually, 3.9 at rim assists per game understates how many buckets he yields at the rim. In his last ten games, the Chosen One has dished out 5.7 assists per game at the basket, nevermind every other area on the floor. Consider for a moment that Kevin Durant, Carmelo Anthony, Kobe Bryant, and Dirk Nowitzki each average fewer than 5.7 total assists per game. And LeBron leads the League in scoring.
Conversely, fellow All-Star Jason Kidd splits his assists between the mid-range and the high efficiency areas, which says as much about him as a distributor as who receives the delivery. Dirk Nowitzki, the most frequent shooter on the Mavs, lives in the mid-range and chucks up a league leading 8.5 attempts per game from 16-23 feet.
Moving along, when we look at assist numbers, it might be worth it to break down each assist according to their expected point value. The numbers tell us that at rim shots and 3-point shots have a higher expected value than mid-range jumpers which only go down 39.5 percent of the time and only have a 2-point payoff but each assist is treated as equal to one. To account for this value distribution, I created a simple metric that assigns a weight to each assist zone, giving point guards who distribute the ball to high efficiency areas a boost in the bottom line.
The weighting doesn’t arbitrarily assign point values that I plucked out of thin air. Rather, I found it appropriate to weight the assists according to the expected pay off of the shot location. The average non-fouled shot at the rim converts at a 60.8% percent rate so the expected payoff, fouls aside, is the product of probability and point value or .608 * 2 = 1.216. Shots taken from 10-15 feet have the lowest conversion rate (.396 field goal percentage) and thus, the expected value from this area amounts to just .792. Threes have a higher expected value than the mid-range because of the added point bonus.
Taking LeBron James as an example, his weighted assists per game (wAPG) would be calculated as:
wAPGLeBron = (At rim APG * 1.216) + (Short APG * 0.880) +
(Mid APG * 0.792) + (Long APG * 0.794) +
(Threes APG * 1.062)
wAPGLeBron = (4.8*1.216) + (0.5*0.880) + (0.2*0.792) + (0.8*0.794) +(2.8 *1.062)
wAPGLeBron = 9.1
Without accounting for the types of assists, LeBron has a 8.5 assists per game so the net difference is only 0.6 assists per game. As is, at rim shots have about a 50% premium on the mid-range shots inside the three-point line. Admittedly, the model would be improved if we accounted for shots that led to free throws but my powers are limited at this point in time. If that were the case, distributors like LeBron would see even more of a boost since most fouls occur around the basket.
This adjustment derives its theoretical foundation from Dean Oliver’s “Basketball on Paper”, where he argues that assists to the perimeter should be discounted relative to shots at the basket because the outside shooter holds the most responsibility for nailing the shot. The passer only changes that likelihood slightly. However, since defenses rarely opt to allow big men to get open around the basket, the point guard receives more credit for creating that opportunity. Put another way, do you think Chris Paul should receive the same credit for his patented alley-oop floater pass as he does for a routine pass to Peja Stojakovic who hits a fallaway mid-range jumper? This model doesn’t think so either.
Who isn’t excited about this adjustment? As you might of guessed, Jason Kidd loses some of his point guard moxie with the tweak. While most point guards see a higher wAPG, Jason Kidd, Derrick Rose, and Jerryd Bayless each have a lower wAPG than raw APG. Here are the leaders and trailers as measured by percentage difference, among those who average 4 ast per 40 minutes.
If you’d like to see the full spreadsheet, I’ve uploaded here on Google Docs.
Cheers.
Rate this post!
This is a really cool idea, but I have a few problems with it… or maybe just ideas on how to improve it. Because essentially it doesn’t really tell you much more than the regular assist numbers. For instance, I understand the percentages at different spots on the floor, but in this instance 100% of those assist-to-field goals counted as a basket (obviously). For a guy like Jason Kidd, I think you would have to take into account Dirks and Jason Terry’s field goal percentages from those spots on the floor – in those cases Kidd knows he’s getting the ball to those players in their favorite spots. I’d be interested to see players wAPG from teams with different playing styles – such as JKidd in New Jersey. Basically, I could be wrong, but it seems to me like this metric says more about a team’s efficiency and attack style than the point guards’ ability to facilitate. Really interesting read though
JKidd’s stats 2007 with the nets were
9.2 Assists Per Game
3.5 At the Rim
2.2 from Three
.5 < 10
.4 10-15
2.5 16-23
So the difference between 2007 and 2010 is that basically Dirk stands closer than whoever was shooting the 3s on the Nets.
Thanks Rob. I view this as Step 1 in the process. I’m interested in assigning credit to distributors for their passing. A ball distributor is more valuable if he’s feeding high percentage shots rather than setting up 40% shots, but I understand the effect of personnel. Perhaps comparing their assist variety to the team’s shot location distribution would be a logical next step.
Ooo I like your new idea Tom. Team’s highest assister’s assists distribution vs the shot loc distribution of team, very sexy idea.
Great start to a potentially hugely important evolution in how we weight assist numbers. I’d love to see an in-depth analysis of how ball usage by an individual player affects his assist numbers. I’m convinced that a large percentage of an individual’s assist numbers are determined simply by the team’s offensive and the associated amount of time that player has the ball in his hands when the offensive is run. Is there any way to account for this? Is, say an assist by Lebron more a product of him having the ball almost exclusively? Look at his percentage of team assists compared to say the Spurs, who average as many assists as a team as Cleveland, but whose leading passer, Tony Parker, averages 3 assists less per game. Interested to hear your thoughts and read more articles like this. Rebounds is also a stat I think we could use some more analysis of. Cheers.
Just to reiterate what I think is the point of comment #1, shouldn’t the weighting be done according to the expected conversion of the player’s teammates, not the whole league? If your man shooting the mid-range jumper is Dirk, you’re going to be much more confident making that pass, and Dirk shoots appreciably better than the league average from that range (though not as well as the last few years). That might alter things a little bit, but probably not that much; it’s also a lot harder to do, I’m sure. Very cool information in any case.
Interesting idea, but I would also like to see how a PG changes with new teamates. For instance, what happens to Chris Paul’s mix when he loses a perfect ally oop partner like Tyson Chandler? Does he find a new ally oop partner, or does his assist mix shift to jumpers? This could be incredibly interesting, because it will essentially point out who ‘created’ the basket opportunity- if it was all Chris Paul then losing Chandler shouldn’t matter. Also interesting would be to see how Derrick Rose #’s would change if he had a big who could pick and pop consistantly like D-Will has in Boozer, or to see Rose’s change in assist mix after losing a 3 pt specialist like Ben Gordon (and subsequently playing on a team with almost no 3pt shooters).
Very cool idea and write-up… and greatly appreciate you sharing the stats!
The fact that two Orlando PGs are in the top-4 got me thinking a bit about cause and effect though. What I’m thinking is that, in many cases, the PGs might not be creating the “gimmies” – Dwight Howard’s monstrous presence does. And if I think about their offense overall (Dwight inside surrounded by 3-point shooters), it’s not surprising Orlando PGs fly up the list.
But of course, cause and effect hits traditional assists metrics too. Just thought I’d share – thanks again for the interesting read.
I’d suggest taking “wAPG” – (league or team average expected points per shot * assists) and comparing assist makers by the per game amount they helped elevate teammate production over average as opposed to mere involvement.
Another criticism – albeit one that I do not have any data for.
Midrange percentages are so low partially because they factor in so many contested shots and shots off the dribble. A three-pointer off the dribble is one of the least efficient shots in basketball, but that reflects the number of off-the-dribble threes taken at the end of the shot clock/while contested more than an inherent difficulty for NBA players to shoot after dribbling.
Open midrange shots, on the other hand, can be high percentage looks. The Mavs set offense often runs plays designed to pop Dirk or Terry free for an open midrange J – a shot they hit at a good percentage. But their overall midrange percentage is lower because of the shots they take in iso situations, where they are guarded closely, moving while shooting, or otherwise off balance.
Also, the wAPG should be weighted according to the expected wAPG, not merely against raw APG numbers. This would account for the larger number of distributors with increases over their respective APG – wAPG are in general higher than actual APG.
But, since assists only count the shots that go in, aren’t you in effect double-counting for higher-percentage shots? Or, to put it another way, shouldn’t you include all the passes that would have been assists, if only the shot had gone through?
I agree with Robert, this is a very interesting article but the same things came to mind. Specifically for Derrick Rose, doesn’t he suffer because he is setting up Luol Deng who like to shoot long 2s and the lack of a strong inside finisher? The change could partially be attributed to what system the player is in as well as who they play with. I still really like the concept. Its a great idea, but like any metric, it has to be seen in context.
I like the idea of comparing this data to the team’s overall shot chart (or perhaps even reducing that data set to only include assisted baskets). It might also be valuable to cross-reference this with time left on the shot clock when the basket is scored. Other readers of this site are probably better at finding the data than I am, but if you added the time remaining on the shot clock to the weighted percentages it will give you a more accurate estimate.
You can’t compare apples and oranges… In the half court, Lebron gets his assists from drive and kick, Kidd gets his assists from Pick-N-Pop while Kobe gets his from the post position. In the process, Lebron’s targets are always wide open with no defense to rotate. Kidd’s targets are in a shooting rhythm and open when they receive the ball. While Kobe’s targets usually has to pass the ball once again because the defense rotates quick enough to cover the receiver.
Where the players end up scoring has to do with the spacing of the recipients and the location of the pass. The math formula can’t account for it. Now if you would show comparison of Kidd and D. Williams, that would be interesting.
While this is an interesting idea in concept, I think it is flawed in execution.
1. The actual numbers of assists are used to weight assists, but those same areas of the floor are already infuencing the number of assists the player has. So your weighting method counting an at-rim assist as worth more than a long-range jumper double penalizes JKidd and overweights Lebron’s assists.
2. Point-weighting the assists is possibly useful, because obviously a 3-point shot is a more useful assist than a 2-point shot. So you can try to derive “true assist” value using just the breakdown of 2s and 3s.
3. You could back out the number of potential assists a player had based on the numbers of actual assists and the shooting locations. For Lebron, the at-rim potential assists is 3.9/0.608 = 6.4 and the 3-pt potential assists is 2.6/0.354 = 7.3. If you were really ambitious you could use the actual percentages from individual teams to weight these things. Now, I’m not sure what value the potential assist metric has, but it might be a useful number to compare to turnovers to get an adjusted player turnover ratio.
This is interesting, but I don’t think it helps the actual goal of evaluating players very much. A player as good at seeing the floor as Deron is will find open guys all the time, but it’s not like he will hesitate in passing to a wide open mid-range jumpshooter. The perentage of assists for good shots I believe is much more a product of the system and the players utilized in it, and should be looked at on a team assist basis. I bet a large percentage of the Houston Rockets and San Antonio Spurs assists are for high value shots, because that’s where the system tells the players to be. It doesn’t necessarily mean that the person credited for assists is performing better than another point guard who’s SG is sitting 2 ft in front of the 3pt line when he could just as easily step back, or for say a piston’s PG who gets a large % of assists on Rip Hamilton screen curls. So my point is, I see this stat as something new that wasn’t observed before, but does not assist an analyst in determining how good a player is.
[...] Crooked Timbernimlas.org » Day Seven point OThe Better Point of View – O Driscoll ImagingHardwood Paroxysm » Blog Archive » NBA HD: Adjusting How We Measure and View AssistsSingle Point of Failure: The 10 Essential Laws of Supply Chain Risk Management | download free java, [...]
Wouldn’t the best stat be “assisted eFG%” where you looked at all passes that directly preceded a shot and calculated the eFG% for those shots. First, it would account for exactly variables you’re interested in here more accurately (i.e. it would allow us to tell the difference between a pass that sets up a contested jumper and a free jumper, a dunk and an awkward layup in the lane over two defenders etc.). Second, it would account for who the ball was passed to. Passing the ball to a big who is open at the three point line is different than passing the ball to a shooter. And the passer should be cognizant of this fact.
The only downsides I see are that it is might be difficult to collect the data — you need stats on passes leading to missed shots as well as made ones. And it would retain the flaw that exists in all assist stats — the passer is dependent on the shooter’s quality.
The first commenter made a good point: “in this instance 100% of those assist-to-field goals counted as a basket (obviously).” In other words, doing this kind of analysis on assists rather than all passes seems a little off to me. I think it would be good (if possible) to know how many of a player’s “possible assist” passes (i.e., passes that lead to shots or fouls) go to each of these zones. That removes the variable of the shooter from the equation. As it stands, you’re counting only those passes on which the shooter converted. Being a Bulls fan, there are plenty of passes that go to a three-point shooter that don’t get converted (because the Bulls lack deep threats). It stands to reason that if you put Derrick Rose (or whoever) on a team with better shooters, his numbers would go up. The same can be said for passes to an open big man or a cutter who then gets fouled. Each of those is a good pass that is not accounted for, even in this methodology. If we truly want to rate how well a PG (or other player) “distributes,” we need to find a way to take into account good passes that lead to scoring *opportunities*, not just made baskets.
I like the idea, but i have a few issues with it.
#1. If you are going to have wAPG based on shooting percentage, shouldn’t you base that on that player’s team % (minus his own fg%). after all he is passing to his team not another team and shouldn’t be penalized or credited by better/worse shooters around the league…
#2. i dont think you can take away points for mid-range assists. The shot went in, so its a full 2 points. You should give extra points for 3′s that go in. If you want to prove who is a more efficient passer then…
#3. Efficiency in assists should be based on how many times the player in question passed to a SHOOTER (Make or Miss) in certain areas of the court.
For example, Lebron might pass to the mid-range shooter 10 times a game, but only 1.8 of them went in. If we are going to start tracking quality of assists, we should also track assists that aren’t finished/converted into points.
If Lebron or JKidd find a guy wide open 10 times and he misses all 10 shots, we have no way of crediting Kidd or James for their efforts…
This would be a pain in the a$$ to track but if we are going to adjust assists, lets figure it out correctly.
twitter: @suga_shane
I just realized someone else had made some of the same points as me. carry on…
Like the commenters above said, I don’t get why you’re decreasing the weight of a successful assist because shots from the area that assist went to often don’t go in. That shot did go in. It’s like saying, “he hit a home run on this ball in the dirt, but let’s weigh that less than a home run off a bad pitch because usually people don’t hit balls on the dirt.” Well that guy did.
I agree with Robert in comment #1 and Andy in #19 (and possible more…I didn’t read every comment). Applying an expected value to an assist mismatches the values. One value pertains to time period before the shot. The other value applies to the shot itself. If one is going to multiply an expected value by another value, that other value should be total number of passes to that spot on the floor. That will tell you how good the passer is at finding the open shooter. OK…now who is going to collect all that data?
I like #15′s idea of point-weighting the assists. Maybe giving a bonus point to an assist on a game-winning field goal (the money ball). And while we’re at it, we should count passes that result in two foul shots (which would be weighted according to the player’s free throw percentage). There’s a big difference between passing to a wide open Shaq and a well-covered Shaq
Most of these suggested adjustments aren’t necessary, and just overcomplicate things. What people want to adjust for – the quality of the shooters – is already indirectly reflected in the weighted assist factors. A midrange shot is only worth .792 for good reason, and while some guys can make those shots at a better clip than others, it’s on the distributor to get the ball to the right shooter in the right situation – i.e. if he makes better personnel decisions, he gets more assists. So making an adjustment based upon the shooting percentage of the shooter for each assist would really be double-counting.
Here’s another thing to consider – LeBron’s assist numbers haven’t changed much over the years (unless forced to play point), but the quality of his teammates certainly has. I’d be willing to bet that the quality of the shooters only makes a negligible difference in a player’s assist numbers, compared with this zone weighing system.
Thanks for the ‘assist’ guys. You all put it a little more eloquently than I could. Lol
To the author: “might’ve” or “might have” not “might of”…
[...] Over 75 percent of LeBron James’ assists yield a three pointer or a bucket at the rim, highlighting just how devastating LeBron James can be as a ball-hander. If he’s not scoring himself, he commands the help defense and double-teams, opening up the high efficiency areas for his teammates. Actually, 3.9 at rim assists per game understates how many buckets he yields at the rim. In his last ten games, the Chosen One has dished out 5.7 assists per game at the basket, nevermind every other area on the floor. Consider for a moment that Kevin Durant, Carmelo Anthony, Kobe Bryant, and Dirk Nowitzki each average fewer than 5.7 total assists per game. And LeBron leads the League in scoring.” [Tom Haberstroh/Hardwood Paroxysm] [...]
Possible Assists would be an invaluable stat for this. It would help describe how many times a teammate misses a short the passer set up for him, which in turn could help value a passer independantly of the quality of his shooting.
Weighing the potential assists versus those that were actualized would allow the creation of an Assist%, which I believe is what you are after: a stat that describes how effectively a given passer creates points.
[...] and Staples High School) is doing to come up with new ways of understanding assists. Last week Haberstroh calculated which players’ assists most often led to easy inside baskets and three-pointers. In a [...]