Up, up, and away
Nestled high up in the rafters of every NBA stadium, six cameras are slowly transforming basketball. In 2005, the Israeli company SportVU converted its missile technology into a soccer player tracking system. STATS, a sports analytics company, purchased SportUV in 2008 and immediately adapted it for basketball. Installed in stadium catwalks, one camera is placed above each basket and two over each sideline, allowing them to capture the entire court twenty five times per second. STATS uses the images to convert every player into an X-Y coordinate and, by triangulating the shot trajectory, can measure the X, Y, and Z coordinates of the ball.
The data SportVU produces is massive. Kirk Goldsberry, a Harvard professor of spatial geography and a prominent sports analyst, wrote, “It was obvious this was the ‘biggest’ data I had ever seen. I’ll always remember my surprise when it occurred to me that everything on my screen amounted to only a few seconds of player action from one quarter of one game.”
The SportVU data set itself is meaningless until translated into “actionable intelligence,” Ryan Watkins, director of basketball products at STATS, told the Independent. It’s like in The Matrix—Neo watches a series of green digits cascade down a computer screen. “I don't even see the code,” says Cypher. “All I see is blonde, brunette, and redhead.” He can read the numbers.
STATS uses algorithms to convert the coordinates of players and the ball into motion—specifically, positioning, speed, and distance travelled. It also records shots, dribbles, and passes. At the end of every game, the company compiles a report to send to teams, which includes statistics like player arc comparison on makes versus misses, field goal percentage based on defender, and total distance run by each player.
SportVU was only adopted league-wide at the beginning of this season, but it is already transforming how organizations analyze the game. The Toronto Raptors Analytics Team developed a way to convert the data into a continuous video of circles moving around the court, just as the players did during the game. As circles take certain paths, a ghostlike unfilled circle moves in a different path—what the Analytics Team’s computer program calculates to be the optimal route.
At this year’s MIT Sloan Sports Analytics Conference, Dan Cervone, Alex D’Amour, Luke Bornn, and Kirk Goldsberry, the former two PhD candidates and the latter two professors at Harvard, presented a way to use SportVU data to compute “how many points the offense is expected to score by the end of the possession.” Applying “competing risk models” generally used in determining a person’s chance of survival, they can determine the expected possession value at every moment of a play.
Over the past quarter of a century, statistics have increasingly become part of the way people analyze basketball, but the statistical approach has been fundamentally limited—the measures in the box score only captured discrete outcomes, like a shot, assist, or rebound. But basketball can’t be reduced to a series of discrete outcomes. As Goldsberry writes, “Unlike baseball or football, [basketball] is a relatively continuous free-flowing sport. The actions within a game are hard to separate because they are chronologically intertwined, and every event in every game is influenced in part by preceding sequences of actions.” Advanced statistics, like Player Efficiency Rating and Wins Produced, are limited by the same problem, as they simply weigh already-measured statistics in complex formulas. These measures can’t capture that made shots are often the result of a teammate’s ball screen, that steals are the result of a teammate’s defensive rotation, that a rebound is the result of a teammate’s box out. The SportVU data set, though, offers the ability to quantify movement, and by extension, every aspect of the game.
Heartbeats per minute x assists = win?
STATS is starting a revolution in the epistemology of basketball. Box score statistics always had an asterisk: basketball is team sport, and since they only measure individual performances, it takes expertise to really understand the game. Now, with the help of technology like SportVU, the previously unquantifiable is being quantified, and there has been a movement towards definite knowledge and away from intuitive understanding.
In an interview with basketball writer Zach Lowe, Brian Kopp, Vice President of STATS, underscored this distinction. An average NBA team scores an average of one point per possession; when James Harden, a guard on the Houston Rockets, drives the ball to the basket, the team averages 1.45 points per possession. “It’s a very complicated way of saying driving to the basket is good,” Kopp told Lowe. The Rockets' coaching staff, based off of countless hours watching practices and games, might intuitively understand that Harden driving the ball to the basket is good. Now they know it is, and by exactly how much.
A traditional understanding of the game will always be necessary, but its importance is fading. “Although big data is very good at detecting correlations, especially subtle correlations that an analysis of smaller data sets might miss, it never tells us which correlations are meaningful,” psychologist Gary Marcus and computer scientist Ernest Davis wrote in a New York Times article in early April. A statistical analysis might show that LeBron James shoots a higher percentage when Chris Bosh is positioned to his left, but there probably isn’t a causal relationship. Bosh’s position doesn’t cause LeBron’s shooting accuracy; they just happen to be correlated. But determining Bosh’s position relative to LeBron’s shot may show how often Bosh obtains an offensive rebound, an important and probably causally related outcome. An understanding of the game is necessary to meaningfully manipulate and interpret the raw SportVU data, but expertise in statistics and computer science is becoming increasingly important.
Intuitive understanding may also be necessary to affect what Bill Russell said is the “only important statistic”: the final score. Statistics—even ones generated by SportVU—aren’t predictive; they can’t determine how certain plays will occur, just how they have in the past. But this is also true for coaches—their decisions are best guesses. Right now, coaches may still be better at predicting outcomes, but as the STATS’ database becomes more extensive and the analytics teams become more comfortable mining SportVU data, this may cease to be true.
Joakim Noah, the Chicago Bulls center, complained that big data can’t measure hustle, drive, or the desire to win. “It’s the guy who’s going to lift you up when you’re down, when things are going on at home,” Noah said. “They really can’t measure that.” Rajon Rondo, the Celtics point guard, echoed Noah’s complaint—“It can’t measure your heart,” he said.
This season the D-League, the NBA’s minor league, made some players wear small devices made by STATS. Attached on their back or chest, they measure, among other things, how hard a player’s heart is working.
ZEVE SANDERSON B’15 thinks SportVU cameras should be installed in more places.