I just got doing some data analysis work (I'll have a link up to the actual code soon), but it's worth talking about now because I think the results are interesting. The basic purpose of the work was to assign a score to players that captures how valuable of a shooter they are. The intuition I used is that a shooter's value lies both in how accurate they shoot and how much they're able to shoot when in a game. So if you could come up with a number that blended both of these you'd get a very useful measure of how valuable a shooter someone is. The stat I based my analysis off of is something in data analysis called an F1-Score (which you can read about here). The basic idea is that it calibrates values between 0 and 1 in a way that is more useful than taking their average. The F1-Score applied to shooters is 2 * (standardized # of Attempts) * (standardized Accuracy)/Standardized # of Attempts + Stanardized Accuracy To do the analysis I took 3PTA/40 and 3p% and standardized them to scores between 0 and 1 for every player in the league. If a player led the league in either his value was set to 1 and everyone's value is listed as a percentage of the league leader. Ie, Steph Curry took 13.1 3PA/40, best in the league. So his standardized attempts score is 1. Nikola Mirotic took 8.4/40, so his standardized attempts score is 8.4/40 = 0.64. A nice thing about the F1-Score is that it penalizes people who are very deficient in one area. For example, if you if you have an Attempts score of 0.9 and an accuracy score of 0.1, your F-Score would be 2*0.9*0.1/0.9 + 0.1 = 0.09. But if you score 0.5 and 0.5 in both areas your final score is 2*0.5*0.5/0.5 + 0.5 = 0.5 The sum of both scores is the same in each example, but the player who's more balanced will rate much higher. F1-Scores take a final value that ranges between 0 and 1. I calculated this for every player in the league, and then did some extra statistical analysis that allows compares that persons F-Score to the league at large, among players that take 3 pt shots regularly, and the position that player plays at. If I do say so myself, it's a unique collection of information about how valuable a shooter someone is, and it clarifies how we ought to think about people in some useful ways. So with that said, here's some of the more useful nuggets of information. Shot Taking vs. Shot Making It's illuminating to take a look at the distribution of shot scores and attempt scores: The big takeaway point here is that the accuracy scores bunch together towards the top, and the attempt scores tend to bunch towards the bottom. Ie, being able to shoot at high volume is a more scarce skill than being able to hit shots accurately. So for the best ranking players it's usually the ability to shoot a lot of shots and still make a decent amount of them that sets them apart vs. 1-trick ponies who are great with their feet set but need the table set for them in order to get off their shot. Here is a chart that visualizes the league as a whole (Steph Curry is the dot to the far right): Not surprisingly, the guy to the far right is Steph Curry. As for the Current Bulls roster, this is a breakdown of how they fared: Here's how to interpret this: F-Score: The players score, which I described above Z: The 'z-score', which in statistical parlance is how many stanard deviations they are away from the league average. zShot: The players z-Score amongst players that actually took 3 pointers regularly. Ie, how many standard deviations each player is away from people who are regularly called on to shoot. This is useful because the league averages are thrown off by all the guys who don't shoot threes at all. The absolute value of this number goes up for the guys at the tail ends of the distribution and goes down for the guys that hug the middle. If a player didn't shoot enough to qualify then they get an NaN, which stands for 'Not a Number' zPos: This is the z-score for each player at his position. Ie, Rajon Rondo's F-Score is 0.16 standard deviations above the league average, but is 1.25 stanard deviations below the league average for his position. There are some useful takeaways here, which I'll get into. Nikola Mirotic and Bobby Portis Mirotic gets the label as being inconsistent. However, by the numbers he's actually elite. He shoots 1.95 standard deviations above his position, which is best on the team. The reason why he scores so high is because his 8.4 attempts/40 is very good, and is outstanding for a frontcourt player. It is in fact very difficult to shoot that many times a game at 39% accuracy. To visualize it, this is how Mirotic compares to his peers in the frontcourt: Ie, Mirotic is elite. There's also some talk about what it would mean if Mirotic would be surpassed by Bobby Portis in the rotation, however, based on the differences in their shooting it's very unlikely that would be the case. Bobby Portis is not on the same planet as Nikola Mirotic as a shooter. This is how they compare to one another vs. guys who take more than 1 3PA/40: They are about as far away as you can get. Portis can hit a 3 pt. shot, but his 1.9 3PA/40 is extremely pedestrian, and his mediocre percentage puts him well behind most shooters in the league, and waaaaay behind Mirotic. The comparisons in their jumpshooting prowess need to stop. Doug McDermott McDermott's an excellent shot maker, but he's a mediocre shot taker, which pulls down his score more than you might guess based on his reputation. His standardized Accuracy score is 0.89....very good, but his standardized attempts score is only 0.43, which pulls down his score quite a bit. So while his marksmanship is very good, his F-Score 0.59 standard deviations among people who take more than 1 3PA/40, which puts him at the 33 percentile. Good, but not great. Isaiah Canaan Canaan doesn't have a reputation as being a great shooter, but his ability to take 3pt. shots is elite. Last year he had the 5th highest F-Score in the league, which is waaaay higher than anyone else in our backcourt. To visualize it, here's how Canaan compares to Butler, Rondo and Wade for shooting at their position: What makes our backcourt's shoot especially anemic isn't their modest accuracy, but their incredibly low volume, since backcourt players take more 3 point shots than the other positions. Right now I think there's no way Canaan doesn't have the backup PG spot locked up considering the alternatives are Jerian Grant and Spencer Dinwiddie, two guys who can't throw the barn in the ocean at this point in their careers.
This is really special. You should write this up for fivethirtyeight.com. It's staggering how good Steph Curry is.
Thanks, SST. FWIW I was planning on cleaning it up a bit, pimping it here and on RealGM to see what kind of response it gets, and if that goes well see if 538 or someone else would be interested in picking it up. When I wrote this post I had just finished about 4 hours worth of work on it, and I'm cringing at some of my typos. Yeah, and Curry is 1 SD above his nearest neighbor for both guys who take 3 pt. shots and his position, which is pretty amazing when you consider the level of competition he's being compared against.
This is all very well and good, but how do you reconcile your numbers with the established fact that Mirotic sucks... ...just kidding. All kinds of kudos to you. I'm not enough of a statistician to quibble with your methodology. It clearly measures something that makes sense and is important. Like SST, I encourage you to refine it to your liking and then share it with those who are seriously into advanced sports metrics. Anyway, well done, lad.
Wow what a read. Makes me change my opinion on Mirotic. And I hope Canaan is our back pg. Both Grant and Dinwiddie are not ready for full time use.
I very much enjoyed this post and have been thinking about it a bit off and on. I'm still not sold on shots taken per minute being an out and out skill. Sure, creating ones own shot is a nice skill to have. But so many of these players have other circumstances that affect the shots per minute they can get up. For instance, on a bad team without a pure alpha that has the ball in their hands most of the game, Canaan can get a lot of shots up and his coach allows him to do so. Can he do that on a winning team? Would the coach of a winning NBA team ever want him to do that? Teammates matter too. With Butler, Rondo and Wade on team along with McDermott and Mirotic, how many shots will there be for Canaan to consume? Is he a good enough overall basketball player to warrant taking so many shots on a team with better players, and ball dominant ones, than he played with on the Sixers? Mirotic gets a lot of shots up per minute for his position. Why is that? Stretch 4 is a popular position / skill set in the NBA right now, the coach of the team wants him to fill that role while he's on the court. He's tall enough and athletic enough to get his shot off and taking a lot of threes is how he chooses to play / is asked to play. And he makes them at an average rate. So I guess getting those shots up is a bit of a skill. But does taking a lot of shots on a non playoff team while making them at an average to below average rate (i'm assuming 39% is this without looking it up) difficult to do, or it just doesn't happen a lot since it tends to not lead to winning basketball? I don't know.... Remember when Rose was taking many 3s and not making them at a great clip the season before last? And then last season he cut down on the attempts and saw a small uptick in shooting efficiency, but was still a poor shooter. What Rose would your stat favor? Last years Rose or the year befores Rose? Do you think pace should play any part in this stat? Does it favor players playing in a high pace system? Did you do the coding / charts in R? Python? Something else? Anyway, great work and thanks for providing some thought provoking material. I imagine this stat would have been useful when I was defending the merits of Jamal Crawford!
1). I think the ability to get up a lot of shots is usually one-sided. Guys who can do it have the ability to tone it down, but the reverse is usually not true for guys who have low usage. They're just stuck there because they're not able to create separation when it's not already there for them. 2). Canaan had about the same 3P rate in Philly as he did in Houston. FWIW I don't think Canaan's a great player. My guess is he puts up that many 3 point shots because he has no other choice. If he couldn't do that he'd be out of the league. 3). I think Mirotic gets up a lot of shots for his position because he's a very good player with a quick release. I think the shooting from him is by design. 4). In general it's useful to consider volume when evaluating a player's %'s. Accuracy becomes more valuable the higher the number of attempts it can be maintained at. Steph Curry's shooting last season was more special because of the insane volume he was able to maintain throughout the season moreso than his accuracy alone. It's the interaction of the two that really matters, although I concede there are situations where you'd prefer high accuracy/low volume. However I think in most situations it's more useful to have guys who can shoot at a high volume with at least average accuracy than playing a bunch of 1-trick pony's. 5). Admittedly volume becomes a drawback if you shoot below the team's TS%, with the caveat that some guys' high volume might be because they're the best option a team has when it's in a low % situation. Some guys are tasked with the responsibility of taking a team's most difficult shot, which increases their volume and lowers their accuracy, but their shooting is still an asset because it represents the best option for the team on that particular possession. This is the Michael Jordan/Steve Kerr contrast. But you're right that this measure breaks down when you're looking at guys who have high volume and low accuracy because at that point the volume becomes a drawback and not an asset. I would say DRose from 2014-2015 definitely qualifies as one of those guys. PG's in general are guarded very harshly because the volume curve is much steeper for them than anyone else, so there are quite a few moderate volume/high % guys who get penalized pretty harshly when compared to their position, causing the score to unfairly rewards gunners compared to marksmen. 6). The code was written in Python using Pandas and Matplot lib and the source code can be viewed here: https://github.com/JonathanBechtel/Data-Analysis/tree/master/Shot Data
