The technology was originally developed to track missiles. Now, SportVU systems hang from the catwalks of 10 NBA arenas, tiny webcams that silently track each player as they shoot, pass, and run across the court, recording each and every move 25 times a second. SportVU can tell you not just Kevin Durant’s shooting average, but his shooting average after dribbling one vs. two times, or his shooting average with a defender three feet away vs. five feet away. SportVU can actually consider both factors at once, plus take into account who passed him the ball, how many minutes he’d been on the court, and how many miles he’d run that game already.
It’s big data in a relatively small pool, and it has the potential to impact everything about basketball, from how it’s coached, to how it’s recruited--even to how we calculate a player’s worth. Sportvision, another sports data collection system based on the same underlying big data idea, has already massively impacted baseball since it came into play in 2006. Now SportVU is generating more basketball data than anyone ever has. And its potential has only begun to be tapped--health care researcher Kirk Goldsberry, who recently wowed the stats geeks at MIT Sloan Sports Analytics Conference with his spatial analysis to determine the best shooters, has begun mining SportVU’s data for new insights. But only 10 teams in the NBA are currently using SportVU. Four of them made the playoffs. One even made it to the finals: The Oklahoma City Thunder.
You could call SportVU the new Moneyball, but that would probably sell SportVU short. “What’s interesting about the Moneyball analogy is that they were using data everybody else had and putting a new twist on it," says Brian Kopp, a vice president at Stats, the company that owns SportVU. "We’re doing that, but also entering into the equation data no one had before. It’s almost Moneyball Plus.” Stats pretty much owns the IP on player stats across sports. Whether it’s the NBA or the NFL that you’re reading about on ESPN or CBS, all those player metrics are being provided by Stats (which is oddly enough, half owned by News Corp and AP). And what they don’t track themselves, they license exclusively from the pro sports themselves.
Kevin Durant’s closely guarded attempts happen all over the court …
But technology has been getting smarter. A few years back, Stats realized the importance securing the future technologies in stats tracking, so they acquired an Israeli company called SportVU, that had already repurposed military tracking technology for use in international soccer. “It was an offensive and defensive play,” Kopp says. “The defensive was, another company could automate what we did. The offensive was, we can collect all sorts of new data.”
Simultaneously, this shifted Stats’ relationship with the industry overnight. Whereas they’d once paid leagues for their data, Stats began approaching the teams themselves to supply it. They converted SportVU to work in the NBA--it’s currently the only professional U.S. sport offering they have, though they’ve dabbled in NFL development--and for a fee, they offered to place six tiny, webcam-like cameras in stadium rafters, connected to a laptop. Each game, an operator would just need to show up to calibrate the system (tell the cameras which player was which), and they’d have access to massive amounts of new data.
Kevin Durant’s open shot attempts …
… don’t really line up to his open-shot makes.
Their system captures the X/Y coordinates of all the players and refs--along with the X/Y/Z (3-D) coordinates of the ball--25 times every second (or 72,000 times a game). Algorithms take into account all sorts of variables to keep the system accurate, from the lines on the court to the reflections of flashing billboards. Another layer of software at a central server puts this raw data together into something meaningful. Information as specific as player ball touches and dribbles can be calculated within 60 seconds of being spotted by SportVU cams. Stats can generate these values in simple, automated reports.
And then there’s a third layer of what’s going on: a layer of deep connections. NBA staffs have access to all their own raw data (think huge spreadsheets), and in an information sharing agreement, they have access to everyone else’s raw data, too. That means every team can mine all of the information collected in 10 courts worth of home games across the NBA. This layer is where the teams get very quiet about what’s really going on. Because if sports are about getting an edge, no one wants to broadcast any edge they’ve discovered.
One insider offered several comparisons in the tech world--from Google leveraging big data to choose between 50 shades of blue, to Target knowing, by a customer’s shopping habits alone, whether she was newly pregnant, and customizing the experience for her. But how do font colors and diaper coupons relate to paint penetration and high efficiency play? For some, it’s about separating elite skills from the just sub elite, then rolling that into player evaluations. Maybe we can all agree on the best rebounder in the NBA from their grabs per game. But who is the second best rebounder if several people are tied in stats? And how much better is the best rebounder from the second best? Is that one skill elite enough to offset other weaknesses?
Russell Westbrook’s attempts after five dribbles: A little wild.
A perfect example of this principle is in how SportVU can track assists. Traditionally, the NBA will award a player with the assist if they pass the ball to a player who immediately scores a field goal. It’s a human calculation, a judgement call based more on feel than science (much like a foul). Aside from clear issues of accurate counts, think about everything that generalized "assists" stat misses: What if the best passer in the world is on a team of people who can’t shoot? They’re making the correct play but getting no credit because they’re surrounded by mediocrity. But put this playmaker on a good team, and that team might excel.
It’s big data in a relatively small pool, and it has the potential to impact everything about basketball, from how it’s coached, to how it’s recruited--even to how we calculate a player’s worth. Sportvision, another sports data collection system based on the same underlying big data idea, has already massively impacted baseball since it came into play in 2006. Now SportVU is generating more basketball data than anyone ever has. And its potential has only begun to be tapped--health care researcher Kirk Goldsberry, who recently wowed the stats geeks at MIT Sloan Sports Analytics Conference with his spatial analysis to determine the best shooters, has begun mining SportVU’s data for new insights. But only 10 teams in the NBA are currently using SportVU. Four of them made the playoffs. One even made it to the finals: The Oklahoma City Thunder.
You could call SportVU the new Moneyball, but that would probably sell SportVU short. “What’s interesting about the Moneyball analogy is that they were using data everybody else had and putting a new twist on it," says Brian Kopp, a vice president at Stats, the company that owns SportVU. "We’re doing that, but also entering into the equation data no one had before. It’s almost Moneyball Plus.” Stats pretty much owns the IP on player stats across sports. Whether it’s the NBA or the NFL that you’re reading about on ESPN or CBS, all those player metrics are being provided by Stats (which is oddly enough, half owned by News Corp and AP). And what they don’t track themselves, they license exclusively from the pro sports themselves.
Kevin Durant’s closely guarded attempts happen all over the court …
But technology has been getting smarter. A few years back, Stats realized the importance securing the future technologies in stats tracking, so they acquired an Israeli company called SportVU, that had already repurposed military tracking technology for use in international soccer. “It was an offensive and defensive play,” Kopp says. “The defensive was, another company could automate what we did. The offensive was, we can collect all sorts of new data.”
Simultaneously, this shifted Stats’ relationship with the industry overnight. Whereas they’d once paid leagues for their data, Stats began approaching the teams themselves to supply it. They converted SportVU to work in the NBA--it’s currently the only professional U.S. sport offering they have, though they’ve dabbled in NFL development--and for a fee, they offered to place six tiny, webcam-like cameras in stadium rafters, connected to a laptop. Each game, an operator would just need to show up to calibrate the system (tell the cameras which player was which), and they’d have access to massive amounts of new data.
Kevin Durant’s open shot attempts …
… don’t really line up to his open-shot makes.
Their system captures the X/Y coordinates of all the players and refs--along with the X/Y/Z (3-D) coordinates of the ball--25 times every second (or 72,000 times a game). Algorithms take into account all sorts of variables to keep the system accurate, from the lines on the court to the reflections of flashing billboards. Another layer of software at a central server puts this raw data together into something meaningful. Information as specific as player ball touches and dribbles can be calculated within 60 seconds of being spotted by SportVU cams. Stats can generate these values in simple, automated reports.
And then there’s a third layer of what’s going on: a layer of deep connections. NBA staffs have access to all their own raw data (think huge spreadsheets), and in an information sharing agreement, they have access to everyone else’s raw data, too. That means every team can mine all of the information collected in 10 courts worth of home games across the NBA. This layer is where the teams get very quiet about what’s really going on. Because if sports are about getting an edge, no one wants to broadcast any edge they’ve discovered.
One insider offered several comparisons in the tech world--from Google leveraging big data to choose between 50 shades of blue, to Target knowing, by a customer’s shopping habits alone, whether she was newly pregnant, and customizing the experience for her. But how do font colors and diaper coupons relate to paint penetration and high efficiency play? For some, it’s about separating elite skills from the just sub elite, then rolling that into player evaluations. Maybe we can all agree on the best rebounder in the NBA from their grabs per game. But who is the second best rebounder if several people are tied in stats? And how much better is the best rebounder from the second best? Is that one skill elite enough to offset other weaknesses?
Russell Westbrook’s attempts after five dribbles: A little wild.
A perfect example of this principle is in how SportVU can track assists. Traditionally, the NBA will award a player with the assist if they pass the ball to a player who immediately scores a field goal. It’s a human calculation, a judgement call based more on feel than science (much like a foul). Aside from clear issues of accurate counts, think about everything that generalized "assists" stat misses: What if the best passer in the world is on a team of people who can’t shoot? They’re making the correct play but getting no credit because they’re surrounded by mediocrity. But put this playmaker on a good team, and that team might excel.
Comment