Introducing Detailed SportVU Defensive Tracking

Over the past few years, we’ve seen the tip of the iceberg with SportVU data, yet it has already transformed the way we talk about basketball and analysis. The SportVU section on the NBA’s stats site has given us a truly useful new stat in rim protection and a handful of others, like catch-and-shoot versus pull-up jumpers and touches with time of possession. The dialogue on players like Roy Hibbert and Rudy Gobert is changed, utilizing defensive FG% from an optical tracking system.  We also have a better understanding of the wrath of Stephen Curry’s shooting and how well he does even with a defender an arm’s length away. We can break down a team’s open three-point “defense” to separate skill from luck in post-season runs. We have shotlogs showing who was the nearest defender for every single shot along with corresponding data like exactly how far away and the time left on the shot clock.

Yet it’s not enough. We don’t have public information on where that defender was, as “nearest defender” can mean the guy three feet behind with his head turned. There’s no information on who the second nearest defender was, which is important for double teams and the fact that apparently many shot-blockers don’t even get credit in the rim protection and shotlog stats. We’re also only getting a sliver of this deep defensive information: defender locations at the time of the shot attempt. It’s like peering into the heavens with a telescope and proclaiming you’ve mapped the universe even though you’ve only seen 0.0024% of it.

Thankfully, there is some access to the raw movement data, and it’s a ton of data. In the typical game, there are 2880 seconds of game-time and once you divide that into hundredth of seconds, as is done with SportVU, you begin to understand how detailed the information is. Players are given X-Y coordinates  corresponding to where on the court they are as well as a time-stamp, while the ball is also given a Z-coordinate for height. You can see some ball-specific analysis here for free throws to understand the source better.

Thus, definitive defensive tracking is possible, and the results could prove truly useful since defense is still an uncharted territory in the public NBA statistics universe. The question then is, what do we do with it?

The recent Warriors-Rockets series seemed pertinent with the coming finals, the presence of the MVP and the runner-up, and looming problem league-wide of how to defend Curry. Grabbing all the data for the series, I applied a fairly simple method where I identify who the nearest defender is for every relevant time segment and then compiled how often a guy was guarding Curry for every possession and synthesized that information with points Curry scored for those possessions. For instance, if two defenders had 50-50 splits for the amount of time they guarded a player, and that player scored two points, then those two defenders would each be branded with one point scored against.

However, the nearest defender method is not without its flaws. The most glaring issue is that it uses simple distance to the defender as the measurement. A defender five feet behind Stephen Curry paying attention to someone else might be the nearest by a few inches, but he’s not really guarding Curry. I created an adjusted nearest defender method using a simple geometric tweak so that defenders in front of Curry would be credited more. I’m not sure how useful this is, and it needs to be tested somehow, but I’d rather try it and keep both methods than just go with the obviously flawed nearest defender system.

Looking at the summary of the results below, Corey Brewer was the most successful defender in limiting Curry’s scoring. His size and svelte body, probably the skinniest guy in the league, means he has the quickness to track smaller guys and bother them with his length, if he’s locked in. Jason Terry, interestingly, looks quite good here too, which is surprising given his age and reputation but I thought he did a pretty decent job considering guarding Curry is like trying to hold a liter of water in your open hand. Among the perimeter defenders, Prigioni fared the worst, and given his lack of quickness that should be expected. Even the “eye test” would single out Pablo, as Zach Lowe here called out Prigioni for a major lapse off the ball. James Harden didn’t look great either, and his critics are the first to point out his problems on defense even after his improvement. Lastly, Ariza looks like a capable defender, which is certainly true even against much smaller point guards, and I wouldn’t put too much stock into Nick Johnson’s numbers given the low number of possessions.

Table: Stephen Curry’s defenders, western conference finals, 2015

*Adj. is an adjusted method that looks for both a defender that’s physically near by distance and in front of Stephen Curry.

I’m not exactly sure what to make of the numbers for big men yet because if a big man is guarding Curry, something went wrong, like a missed rotation or that Curry is going in for a layup. However, I can tell my adjusted method worked because big men saw more possessions through the adjusted method than the standard one; this means they are being identified as the defender more often because they are more frequently in between Curry and the basket.

There’s still some work to be done with the methodology. Who is guarding whom? is a tough question to parse because someone can duck under a screen for a moment while tracking another player. Plus a few things can be added, like efficiency, as I’m only looking at points per possession, assists, turnovers, and other measures of impact. But this is a big improvement over what we have, and one useful device is that it doesn’t see a possession in black-and-white terms: a player isn’t guarded by one person per possession. There are switches, soft double teams, and numerous screens to mix things up. If Curry scores on Terrence Jones, we can’t forget that he first lost Prigioni in the backcourt. We need to assign the blame appropriately.

Pulling this data is tough because there’s so much of it for an entire season. But there’s some real potential here for the public in showing how players score versus certain defenders, how far away they are, and what it ultimately means. Any amount of useful information on who the best defender is for someone like Curry, or LeBron or Harden, is sorely needed. We’ve had the subjection evaluation for eons; it’s time to progress further into the data-ball era, not dropping the use of our eyes and intuition, but integrating them with hard data for the most informed decisions.

The next thought should be, what else can we figure out?