2013-2014 Adjusted Defensive Impact by Court Location

I have finally completed work on my visual NBA stat: Adjusted Defensive Impact by Court Location.1 I first explained how this stat works here, but in a nutshell this is a way to visualize how a player defends shots in the NBA adjusting for the other defenders on the court with him, the expected probability of the shot being made, and the other (non-shooting) offensive players on the court.2 This new model also adds the possibility of a team-wide effect that you might attribute to coaching (this is not visualized in any way just yet). I had many requests to also include something about how players affect the location of a shot. You can now see this at the bottom of each player’s chart. This is a simple regression that controls for defensive players only and shows you how a given player affects the volume of shots that are close (<8 feet), midrange, or 3 pointers. I have lots more to say below the jump but here's the widget, have fun poking around! Warm colors mean that opponents are more likely to hit a shot when the player is defending, cool colors mean opponents are less likely to hit their shots.

2013-2014 Adjusted Defensive Impact by Court Location

I’m going to keep the caveats pretty short this time, because I plan to have another post in the next couple of days that elaborates on things, but here are a couple of important notes on the whole project.

How did you make this? What’s new?
Most of the details are explained here. Since then, I have added a number of new model features. Specifically, I now control for offensive players and teams. Offensive players are only included in the regression if they were not the shooter (I already controlled for the probability of the shooter making the shot). I am not currently doing anything with the team effect (indeed, I haven’t even retrieved the coefficients so I couldn’t look at it if I wanted to) but the overall impact on player coefficients seems small. In fact, model diagnostics suggested that the team effects brought little, if anything, to the model. The inclusion of offensive players, on the other hand, has a very positive effect on the model’s fit and predictive capabilities. I will have another methods-y post soon that updates my old post with the new model. I should also have a nifty graph showing how good the model is at predicting at different court locations. Another ‘new’ thing is that it turns out I had a little math error in my previous Javascript implementation that had all boxes about 12 pixels from where they should be. Whoops.

What this is not
This is not a holistic way of looking at defense. It only applies to plays where a shot was actually made. That means it misses plays that ended in a foul or a turnover. Drawing offensive fouls and creating turnovers can be a pretty big component of a player’s defensive game! A player could easily look bad by this measure and be a good defender nonetheless, because creating a turnover is a better defensive play3 than allowing the opposing team to take a shot (even a low percentage one). Moreover, each chart needs to be interpreted carefully. Take Jonas Valanciunas, for example. There’s a ton of orange and red on his chart. That seems bad but I would argue that there is a substantial silver lining for Valanciunas. As you can see, Valanciunas’s presence increases midrange shots by 4%. Those shots are easier to make when Valanciunas is on the court, but this may actually be preferable to a player who is good at defending the midrange but does not force shots there. This is simply because on average, a close shot is a much, much better shot than a midrange shot. So if you force shots to the midrange, that’s good, even if you allow those shots to go in at an above average rate. You would literally have to push FG% in the midrange up by 10-20% before a midrange shot would be better than a close shot.

Feel free to use this!
One feature that did not make the final cut was an ‘export to .png’ to make it easy for basketball writers to feature this on their site. This is entirely due to my incompetence with Javascript. However, if you would like to use a chart on your site, please feel free to do so. Printscreen or screencap the chart and use it as you wish. Please do provide a link back to this post or to the front page of my blog somewhere near the chart.

  1. This is all part of my master plan to call it “RADICL” when I get around to using ridge regression.
  2. Yes, this last bit is a new feature
  3. And usually it’s a much better offensive play too.
Share on FacebookTweet about this on TwitterShare on LinkedIn

10 thoughts on “2013-2014 Adjusted Defensive Impact by Court Location

    • I’ve been waiting for this comment… I don’t really have a rejoinder except to say that I think there is general face validity but a few confusing cases. James Harden is one of them.

    • Well that could be due to the Rockets hiding him on an opponent’s worst player and some overlap playing with the defensive monster that is Dwight Howard

  1. Two questions:

    1. How do you explain Anthony Davis’ red boxes in the paint?

    2. Does your answer to # 1 justify choosing Noah as the DPOY over Anthony Davis? The data appear to suggest Noah wasn’t the best defender in the league. In fact, I thought a decrepit Kevin Garnett’s results looked preferable to Noah’s.

    • I’m going to take the safe, oft repeated line on Davis here, which is that he has all the tools to be a defensive beast, but he’s not there yet. One thing we’ve known for a while is that blocks aren’t always a great indicator of defense near the rim. Davis is unpolished and needs some work, but the NBA is on notice–he’s a scary guy. You can see looking at adjusted APM stats for example, though, that his defense right now is maybe a slight plus at best: http://stats-for-the-nba.appspot.com/ratings/xRAPM.html

      For 2, I probably wouldn’t have voted for Noah. I don’t think this makes his defense look bad by any means. He looks pretty decent on average around the basket despite some problem areas. He forces shots into the midrange, which is exactly what you’re supposed to do in today’s NBA. He’s a pretty solid defender but… Hibbert was to me the obvious DPOY, playoff struggles and late-season struggles aside. He would have easily won it had the Pacers not fallen off a cliff after the all-star break.

      edit: I also would have voted for Bogut and Gasol above Noah, just to pick two others. And Garnett played pretty good defense for a big chunk of the year. He’s still got it.

  2. Wondering if there aren’t a couple more things you could factor in: Fouls and offensive rebounding.

    Adding in a factor like ORebound(1 or 0) * Team FG% might clean up some of the discrepancies between eye test and centers looking bad on your chart, as good defenders not only cause opponents to miss but leave their team in position to secure the rebound after the miss.

    Also factoring in fouls as well will show things like Harden allowing folks to blow by him forcing other players to cover and foul.

    • Yeah, I’d like to include all these kinds of non-shooting events. The immediate problem is that when I was assembling my data I didn’t really think about this (hindsight is 20/20) so now I have no data on them. The bigger problem is that it’s not clear how you would integrate them into the court location scheme–I don’t know where things like turnovers happen geographically.

  3. Really Interesting that the notable shot blockers all seem to see an increased shot probability from mid-range, is there an underlying theory to why that is the result? Are those defenders less willing to step out to mid-range, leaving the shots more open? Are wings more willing to let players into the area assuming help from the shot blocker?

Comments are closed.