College Basketball

Jacobs: How data help Duke refine approach on court

Anyone who has watched Duke basketball this season recognizes that Jahlil Okafor’s teammates have tailored their games to complement the big man’s strengths. You might think the need for such deference is obvious, but you probably weren’t a high school all-star like the other Blue Devils.

“Adding a number to it makes the argument a whole lot more convincing,” says Kevin Cullen, Duke’s basketball’s director of information technology. “I think that helps people buy into that philosophy.”

That’s where analytics – discovering and evaluating patterns in data – has helped Duke’s 2014-15 squad define and accept what constitutes a good shot, both in absolute terms and in making the best use of Okafor’s skills and talents. “I think coaches for years have always tried to coach good and bad shots by their players. And one of the more simple things is the catch-and-shoot three versus a 3-pointer off the dribble,” Cullen explains.

The Blue Devils made 42 percent of their catch-and-shoot 3-pointers (51-121) in their first seven games, all wins. Given that only 42 long-distance tries were created off the dribble – with 28 percent accuracy – it appears the lesson sank in. “You try to look at where you can be your best,” Cullen says. “That’s one area that we’ve made a conscious effort to improve.”

Refining similar performance-related messages can be accomplished by conventional video analysis, a process that requires endless parsing of images into desired categories. Cullen has earned praise from coach Mike Krzyzewski for his work preparing scouting videos for USA Basketball squads, along with motivational riffs prior to each medal game in international competition.

Tailoring statistics

Back home, Cullen also handles a more exacting tool, one that automatically links basketball act and player, accompanied by corresponding video. SportVU, sold by STATS LLC, uses the concept behind missile-tracking technology to code and record player and ball movements 25 times per second. The software creates algorithms that follow objects without requiring attachment of a tracking device. The images thus provided come from six triangulated, high-resolution cameras nestled near the ceiling of 75-year-old Cameron Indoor Stadium and six more at the team’s adjacent practice facility. The same system is also in place at all 30 NBA arenas.

Statistics acquired through SportVU can be tailored to a team’s needs. In its second year of use at Duke, the staff is still exploring the possibilities. “We’re getting more of an understanding of what we’re seeing,” says Cullen, a former team manager who graduated from Duke with a degree in computer science in 2007. “I think we understand better what they’re presenting and how to use it.”

Duke is one of the few university programs employing SportVU (along with ACC member Louisville), and might be the only team, college or pro, whose practice floor as well as home arena are equipped with the system. The idea is to get as much useful data as possible.

Through Dec. 11, Duke had played eight games, all except one in a building where SportVU is available. Meanwhile, the Devils had 42 practices; including those greatly expanded the universe of data. That enlarged trove of information is then subject to analysis that generates data in a range of unfamiliar but useful categories – touches per game, rebound chances (calculated when a player is within 3.5 feet of the ball), secondary assists (passes leading to a basket, as in hockey), potential assists (resulting in free throws).

SportVU also measures speed, distances run, shot trajectories, player alignment relative to ball location, where drives originate and how successfully they conclude when launched by particular players moving in particular directions.

“NBA analysts say their proprietary work with SportVU’s raw numbers fundamentally alters their understanding of basketball – one reason why most progressive teams protect their SportVU insights like state secrets,” the Wall Street Journal noted last year.

The newspaper said NBA folks, who’ve employed SportVU since 2010, found “all kinds of hidden truths” revealed by the technology. The accolades are more muted at Duke, where Krzyzewski long has been enamored of intensive video study – into the night immediately after games surrounded by his staff, sequestered at the beach with a compilation of every contest from a season recently concluded.

“It’s amplifications of principles that I think are generally true,” Cullen says of SportVU data. “Some of them are surprising. None of them are earth-shattering. None of them are going to win the game.”

Feel vs. numbers

Jeff Capel, Duke’s associate head coach, finds the intricate analysis “really confirms some of the things that you may think. It’s really cool, especially for people that love numbers.” But the son of a coach, who turns 40 in February, sees fascination with numbers as a “younger coaches” orientation. He relies more heavily on a “feel” for the game, basketball common sense and multi-game stats to evaluate players.

During the 2014 season, SportVU’s data placed Amile Jefferson among the nation’s 25-best rebounders, based on percentage of available rebounds he grabbed while on the floor. Capel thought the numbers only confirmed what the Duke staff already knew. “Not trying to be funny, but we knew as coaches, Amile’s going to rebound,” he says. “He should be a good rebounder.”

Last year the 6-foot-9 Jefferson was the Blue Devils’ second-best rebounder, with 6.9 per outing. More revealing, he outstripped team leader Jabari Parker with a board every 3.3 minutes played. As a junior playing alongside Okafor, Jefferson leads the squad in rebounds, his per-minute (2.8) and per-game (8.3) averages better than last season.

Cullen, the point person for Duke analytics, sees limits to applying advanced metrics to basketball, especially compared with baseball, where a Moneyball-style approach is increasingly part of the conversation. “Baseball is so easy,” Cullen says. “Baseball is like addition. This is like calculus, trying to put all the moving parts together. Nobody has ever accounted for a back pick that led to a lob, or a hammer screen (a San Antonio Spurs maneuver) that led to an open three in the corner.” In Cullen’s estimation, similar limitations mark efforts to develop comprehensive measures of defensive acuity in basketball.

He sees a better chance that SportVU’s tracking of movements will yield new insights into player health and performance. Coaches and trainers are always looking for ways to minimize injuries, increase endurance and ease the strain caused by practice and play over the course of a long season.

But so far Nick Potter, Duke’s assistant director of athletic rehabilitation, views the prospect of SportVU producing performance-enhancing information as more “intriguing” hope than real possibility. “The data they give us from a performance standpoint is not very relevant and not very clear right now,” he says. “If we get valuable information, I’ll share it and the coaches will decide what to do with it.”

Intriguing can certainly be entertaining. The Big Ten has used SportVU during its postseason tournament, offering TV viewers a savory stew of uncommon statistics such as the average number of touches before a key player scored. A dose of real-time game data would surely enliven the ACC’s generally formulaic basketball telecasts; let’s hope SportVU is tapped in 2016 to give viewers a taste of off-beat insights when the conference tournament moves to the Verizon Center, home of the NBA’s Washington Wizards.