Shot Location Efficiency
A couple weeks ago, I wrote an article using data from basketballgeek showing shot location visualizations. The logical next step from visualizing the data is to use it for more analytical purposes. So I set about to build a model to predict points based on shot location.
Here is the expected field goal percentage based on shot location. The data set runs from 2006-2007 to this year's All-Star Break and contains over 600,000 shots.
That is the starting point for my model. I take the expected field goal percentage for a given spot on the floor, and multiply it by either two or three, depending on whether the shot is an attempted two pointer or three pointer.
Another part of my model is offensive rebounding rate. From the field goal percentage chart, you can see that some three point locations are as high percentage shots as some two point locations, yet the value of a three pointer is inherently higher. Offensive rebounding rate on three pointers as compared to long two pointers is another reason that mid-range jumpshots are inefficient plays.
The value of an offensive rebound is contested in the basketball analytics community, as I recently learned. I understand why player evaluations based on linear weights don't work at all in basketball, but I'm not sure why they wouldn't work on the team level. Why can't we say that the average value of an offensive rebound is roughly equal to the average value of adding another possession. If somebody can enlighten me on if and why this assumption is faulty, I would appreciate it. Regardless, the average possession yields something like 1.05 points, so for each shot location, I multiplied the expected missed field goal percentage by the expected offensive rebounding percentage and again multiplied that by 1.05.
Then, I found the shooting foul rate based on shot location. This was a challenge, since the play by play files don't chart foul locations. I therefore used three resources to try to predict shooting foul locations. Ryan Parker collected data that tracks the locations of nearly every event over ten games, including 200 or so shooting fouls, which definitely helped. 82Games has charted shooting fouls, though the data isn't very granular, and they don't mention the magnitude of the study. Lastly, I found the shot locations of all made baskets where there was an and1. Here's what I came up with.
I think the above graph reasonable. It's too smooth, since I think there is probably a steep breaking point where players stop taking mainly jump shots and start playing with their backs to the basket. Jump shots are much less likely to draw fouls than post-ups, however my model can't capture that since I use smoothing techniques. The play-by-play data does include shot type information, so if I had a do-over, I would do some testing based on jumpers vs. other shot types. Anyway, what I do with my shooting foul model is multiply the rate of missed shots at a given location by the shooting foul percentage at that location, and then multiply that by either 2 or 3, and again by either 0.76 or 0.81, depending on whether the respective shot was a 2 or a 3, which represent the number of free throws a player earns for a shooting foul on a missed shot and the made free throw rates on those shots. I also multiplied the rate of made shots by the expected And1 percentage, which is much lower than the shooting foul percentage.
Put that all together, and here's my ultimate point expectancy model.
The average is up around 1.25. That's about 0.2 points better than the average possession, since plays that don't result in shots either end up as personal fouls or turnovers, mainly turnovers, which net 0 points. I applied the model on five-man units as well as individual players.
First, the top and bottom five five-man units in shot location efficiency, or expected points per shot. Ideally, some of the shooting, free throw, and rebounding percentage would be customized but I'm using league average rates for this entire study. Minimum 500 shots.
I'm happy to see that the Eastern Conference Champion Magic are the top team on this list because I'd always assumed that their offense last year was extremely efficient. The Magic had two options on offense. Dwight Howard took shots at the rim, while Hedo Turkoglu and Rashard Lewis hoisted threes. That unit was also by far the best in effective field goal percentage in the league, so they were getting high percentage shots, making high percentage shots, and though I can't include their free throw rates or offensive rebounding rates since those would be pains to calculate, I'm sure that with Dwight Howard, the Magic were successful at getting to the line and grabbing rebounds. The Suns, of course, are one of the top five teams.The Bobcats, surprisingly, take highly efficient shots, but don't make many of them. On the other end, we already knew the Bulls run an inefficient offense, and I'm not surprised to see the Pistons do too. That Thunder offense last year must have been absolutely brutal.
Now turning to defense, teams that force the least efficient shots.
It's no surprise that the Rockets force teams into low percentage shots, as they boast three of the top five five-man units. That defensive lineup containing Chuck Hayes, Shane Battier, and Yao must be impregnable. And what do you know, but the Magic offense that generated the most efficient shots also had the defense that allowed the second most inefficient shots. Interestingly, the Bobcats offense that ranked second in shot efficiency actually allowed the most expected points per shot on the other end of the floor. I don't think I've watched a Bobcat game this year, but I'd be interested to know what's going on with that unit. A couple surprises on the bottom five list. The Thunder have made noise throughout the league for their much-improved defense, yet it's not a matter of holding opponents to inefficient shots. Instead, their opponents have gotten quality shots off, but have not made them, which would point to an impressive ability to contest shots. Also, the Thunder might do a good job of defensive rebounding and not fouling, which wouldn't appear in the numbers I'm showing.
The next table includes defensive stats for individual players, but still uses data based on the entire five-man opposition. I raised the minimum to 1,000 shots.
I could've guessed that the top defenders at forcing low percentage shots would be centers, since preventing shots at the rim is the best way to force inefficient jump shots. Dikembe Mutombo, even at (insert whatever made-up hilarious age here), remained an astonishingly good defender. He forced opposing teams into inefficient shots, and no player held rivals to as low an effective field goal percentage as Deke. I'm not sure if any of the guys who show up on the bottom five have reputations as poor defenders. Basketballvalue exhibits poor defensive ratings for Russell Westbrook and Lous Williams and says that by adjusted +/- Sam Young has been a flat-out awful player in general this year, though the guy who runs basketballvalue is the stats guy for Sam Young's team, the Grizzlies.
This table shows how a player's five-man unit performed while he was on the court.
The top four players were all Knicks during this time frame, as were three of the next eight on the leaderboard. All this is telling us is that Stevie Franchise, Starbury, and Baby Shaq all excel at hanging and banging, and that Isiah is attracted to that type of player. Sam Cassell, on the other hand, can't get to the rim. So I decided to take out a player's own shots, and include only shots by a player's teammates while he was on the floor.
At one end are players who spread the ball around and at the other end are players who inhibit floor spacing. Steve Nash's teammates had easily the highest effective field goal percentage, and oh by the way, Nash's own eFG% beats out that of his his teammates. Erick Dampier and Joel "Prezbo" Pryzbilla clog the paint like a hot fudge sundae clogs one's arteries.