The Nascent Power of Optical Tracking Data

Photo Credit: SLY Photography

Tucked into the Friday morning schedule at the MIT Sloan Sports Conference at the Boston Convention and Exhibition Center, we’re watching Frank Lampard, of Chelsea FC, look around a soccer pitch. He’s not dribbling. He’s not passing. He’s not shooting. Lampard is simple running and looking, his head swiveling back and forth, stretching the human body’s limits for how far the head should turn. Over and over Lampard turns his head, a special camera from Sky Sport focused entirely on him. Eventually the ball finds Lampard and without looking he fires a pass to an open man. He’s already looked, so why look again?

The research, done by a team led by Geir Jordet from the Norwegian School of Sports Sciences in Oslo, tells us something very simple: players that look around and explore their surroundings more without the ball are far more likely to complete a pass than players that don’t look around. This is meant to encourage further like-minded research in other sports, but everyone in that room came out of it with the knowledge, or the confirmation of previous knowledge, that awareness in soccer leads to better passing.

As much excellent, important basketball work as there was at the conference this year, this soccer project represented the sort of hyper-focused, results-oriented research that you may have been more likely to come across before the onset of new, big data in the hoops universe. Results were everywhere, but this was the year for process.

The data in question is being provided to 15 NBA teams by SportVU and STATS LLC via a six-camera system that hangs in the arena rafters, tracking every single movement of the basketball and the players on the court. It can tell you everything from how far a defender is from the ball at any given time to how fast Kenneth Faried is running at the rim as a jump shot hits the highest point of its parabola. It is fascinating technology and featured in three of the four research papers focused on basketball this year for good reason, but it is not without its short-term limitations.

ON THE SHOULDERS OF CAMERAS

If there was a consistent theme among the three SportVU-centric papers, it was ‘Good Work, Keep Working.’ This is less of a comment on the work itself and more on the relative infancy of the data-sets. Only half the league has the cameras installed, and even fewer have had them installed for a full season start-to-finish – leading to a relatively incomplete sample size (primarily based in a lockout season) that, while large enough to get some interesting results is still far too small to use as the foundation for any hard conclusions.

But we don’t always need hard conclusions if we can use the data to support an existing argument. For example, a paper titled ‘To Crash or Not to Crash’ written by a group of MIT graduate students came to the conclusion that sending multiple players to the offensive boards is worth a few extra points per game, while keeping a few back for transition defense was also beneficial – the best combination being to send two rebounders and keep three players back. Considering that this is something you see regularly in every single NBA game, the data here assumes the valuable role of supporting and confirming common logic.

A paper from Kirk Goldsberry and Eric Weiss – the star of the weekend – provided a similar function, using SportVU to track shots taken (and not taken) when particular interior defenders were within a few feet of the attempted shot. The results, from last season, were plentiful, but among the most noteworthy were Larry Sanders allowing a field-goal percentage of just .349 when he was within five feet of an interior shot, David Lee allowing more than half of the shots in that same range to fall and Dwight Howard preventing opponents from taking many interior shots whatsoever.

In other words, Sanders and Howard play highly-valuable, if different, interior defense while Lee struggles have a direct effect on shots around the rim. Again, confirming what may seem like common sense, but in the case of NBA defense common sense has had very little actual data to support it. Ray Allen is a good three-point shooter and that fact is supported by his three-point percentage just as the defense of Sanders and Howard is now supported by tracked field-goal percentage. Moving forward, that sort of information will be revolutionary.

We just can’t take the results of either of these papers as gospel. What if some of the major outliers in offensive rebounding didn’t have SportVU cameras, thus affecting the entire data set? There may be over a thousand shots tracked behind the Larry Sanders numbers, but what if some of his worst defensive games against the best-finishing opponents weren’t available? Synergy Sports has tracked defense using game tape for many years now, and while the resulting numbers have often been misleading, the results of possessions in isolation or post-up defense have shown throughout the years that even the best defenders can have numbers that fluctuate wildly.

Neither paper pretended to be the end-all be-all. The Goldsberry and Weiss paper is very wise in making it very clear that this is all preliminary work, meant primarily to establish the potential for the optical-tracking data as a useful tool for defensive analysis. When SportVU burst onto the scene two years ago it bought the property and cleared the debris off the land in order to make room for an exciting new House of Data. Everything is going as planned, but in terms of what we are going to learn from it, we’re barely putting down concrete in the basement.

Nowhere was that more evident than in Philip Maymin’s study on Acceleration in the NBA, which showed in part that power forwards and centers – among which Joel Anthony was one of the best – are the most frequent accelerators in the half-court offense and that the primary acceleration zones are the areas around the paint and the top of the key (heavy pick-and-roll and rebounding areas). But again, doors are opened for a great deal of questions that we’re probably years away from answering.

We’re also closer than previously thought.

INTRODUCING EAGLE

If the main limitation of SportVU has been the sheer lack of data – solved by all 30 teams buying in and time simply passing to allow for the sample size to mature – then the second obstacle has been figuring out what exactly to do with all the data. SportVU records every action at 25 frames per second, which means that in a single game the cameras spit out enough information to fill a stack of phonebooks. The previously mentioned research papers funneled all those raw numbers into some relatively elegant results, but they weren’t exactly written overnight. While STATS LLC helps each buy-in team with some information, teams that want to get the best return on their investment have had to find programmers capable of writing extensive database language.

Those people don’t exactly grow on trees, nor is one person likely going to be able to, say, come out of a coaches meeting and produce results on a few things an advance scout mentioned in a report?

What if a coach, five minutes before a practice, wants to know exactly how many times Dwyane Wade went away from a screen in his games against teams that start two traditionally large big men? That’s where Eagle comes in.

Built by Rajiv Maheswaran and his team at Second Spectrum (and the University of Southern California), Eagle is a platform, much like Synergy Sports or even NBA.com/Stats offers, that can take complex questions – how many times LeBron James has been within five feet of a jumper taken on the left wings, and the percentages on those shots – and return answers in less than a minute. It not only offers player and team rankings based on the highly-variable filters put into the system, but crafts visuals, like heat maps and animated play-by-play of pick-and-rolls, to go along with everything.

Basically, just as you go to NBA.com to find Shane Battier’s shooting percentage on corner threes, you can use Eagle to find, well, just about anything you can dream up (when the platform is completed).

Whether it is with Eagle or another product, this is the future of SportVU – a combination of speed, power and visuals that makes queries that were unthinkable five years ago almost as simple as looking up offensive efficiency.

SEEING IS BELIEVING

Eagle’s ability to create images and animations highlighted the overarching theme of the weekend in Boston: that at some point, analytics must be applicable. Spreadsheets don’t cut it for coaches and they’ll barely be given a glace from most NBA players. The numbers eventually have to translate on to the court, and in order for translation to occur there must be communication.

Communication. This is what it all comes down to. Numbers are a vessel for communication, but there are more accessible avenues – a panel on infographics at Sloan was dedicated entirely to this topic – for evolving the data into something usable. We won’t stop learning new things from big data, but we can’t stop learning how to communicate those things.