Sloan Paper Reviews: “The Dwight Effect”

In preparation for Sloan, I’ll try and read over the research papers and give a quick review of each. The first paper I’ll be reviewing is The Dwight Effect: A New Ensemble of Interior Defense Analytics for the NBA, by Kirk Goldsberry and Eric Weiss. (Download here)

You’ll recall my last review of Kirk’s work involving Dwight in the title pointed out a few issues. In e-mail discussions Kirk simple said “I’ll have more at Sloan” in what I read as a coy tone. Did he disappoint? Not for a second!

Review

While the introduction to the paper was alright (of course I’ll have a few nits), the best page by far was page 5. It leads with this graphic and has great data!

Kirk_interior_defense

Kirk and Eric start with a simple point – allowing interior shots is bad (or good if you’re on offense). Teams know this, as roughly one third of all shots come from this range. Currently, the only real “proxies” we have for these shots are blocks. But Serge Ibaka, the league’s top shot blocker per-minute last season, only recorded a block once every eight minutes. As interior shots occur much more often that that, trying to figure out which players are responsible for interior defense can be rough. That is, unless you have visual tracking data!

With this, Kirk and Eric go on to ask two questions. When interior defenders are near the interior:

  1. What is the rate of shots the opponent takes (interior, midrange or three point)
  2. How well do these players shoot (with the focus of course being on close range shooting efficiency)

This gives us two useful metrics. If interior defenders can keep opponents from getting off close shots, it’s a huge plus to the team. Kirk calls these “invisible blocks”. I have slight quibbles about the name, but  agree wholeheartedly with the point. Next, how do these players do at preventing (or in David Lee‘s case apparently, enabling) the opponent from scoring? They also examined how often opponents got off shots when the defender was within one foot, three feet, or five feet. They then examined if the defender getting close to the opponent impacted shooting efficiency.

Their findings try to identify which players were good at keeping people out of the paint (spoiler alert, Dwight Howard leads the list), which players are good at reducing opponent interior efficiency (Roy Hibbert is tops here) and which players have the biggest impact when within five feet of opponents’ shots (apparently Andrea Bargnani is really good at this. His problem? He doesn’t often get within five feet of his opponents [editor Devin's note: insert your own joke here]).

I really liked this paper. Many attempts at analyzing individual defense try to include a lot of data (on and off data, +/-, etc). This paper focused on how well opponents shot from close range when the defender was in close range. Additionally, we got some indications of their ability to impact the number of opponents’ shots. The good news? Opponents shoot worse when Kosta Koufos is in the interior! Karl should take note.

Criticisms

mutombo-finger-wave

Of course, you can’t write a glowing review without a few nits. Let’s start with the obvious – sample size. The data is taken from over two seasons (one of which was a lockout, and the other saw flopping banned). Additionally, SportsVU has only been installed in fewer than fifteen arenas. I know this because Kirk and Eric very explicitly state this in their paper. So the key is to be careful about how much to take away from only two seasons of data. This means my hope that Koufos is an excellent defender will have to be tempered until we check the year-to-year significance of this. That said, the authors are upfront with this issue. If more teams support SportsVU, then we can expand this study even further. Hopefully papers like this help convince teams to install SportsVU, because it isn’t exactly cheap – STATS VP Brian Kopp has said that the system costs “mid-five figures to low-six figures per year“.

One thing that did get me was that Kirk mentioned that “blocks are ambiguous to defense” at the start of the paper. I felt this was sloppy. The truth is, when all else is equal, a block is worth around two thirds of a possession. Shooting efficiency is one of the harder things to address in defense, and Kirk and Eric have done great work expanding that. I would say, had Kirk said “not all blocks have equal defensive impact“, I  would have been less irked. That said, this is a very minor criticism.

Another minor nitpick is that Goldsberry and Weiss harp on the idea that “defense is hard to evaluate in the NBA”. I think a subtle tweak is needed: “defensive impact on opponent shooting efficiency is hard to attribute to individual players in the NBA” is really more clear. I think this is mostly obvious given the focus of the paper. Still, it’s a subtle difference that I find many people miss when bringing up defense as a problem in the NBA.

Going Forward

He's not beating Kobe 1 on 1, but he's a great start for a rebuilding effort.

Alright, the key to a great paper to me? At the end of it I have a ton of questions. I’m very excited about where this data can go. Here are just a few things that I’m looking forward to.

  1. The shot locations are broken into three ranges. I can’t wait until we have more data to support going more granular in the ranges.
  2. Currently we examine the frequency of interior shots vs. interior defenders. Is it possible to start figuring out how much of this is the interior defender “scaring” players out vs. weak perimeter defense allowing players/passes to get to the interior?
  3. Something that I also keep wanting to examine is fouls that lead to free throws rates by location. This seems to matter in these as well. For instance, David Lee is poor at interior defense, but his foul rate is lower. What do the same shooting charts look like when we include the impact of fouls?

Summing Up

Check this paper out. At bare minimum, check out page 5. I think this data, mixed with a metric like Ian Levy’s XPPS can actually lend itself well to attributing more of the defense directly to players. At the very least, it would be interesting to start experimenting with this to see if we see any year-to-year consistency in interior defense (for example, Ty Willinganz points out that coaches can change a team’s defensive effectiveness). I’ll definitely be in the audience for the presentation and hope Kirk and Eric bring more amazing images.

-Dre

Comments are closed.