There’s an amazing new way to value hockey players, but the NHL isn’t buying

Hockey’s Moneyball moment

2 Premium content image
hockey-main

(Getty; Shutterstock; iStock)

Cameron Brace wasn’t quite sure what the NHL scouts were looking for this June. He just hoped he was it.

Like every other kid in the Ontario Hockey League, the friendly, bright right-winger for the Owen Sound Attack spent his spring awaiting Sunday, June 30, the day of the pro hockey draft. Unlike most of his teammates, he had a chance. Brace is fast, and ranked as one of the OHL’s top scorers. The NHL’s Central Scouting Bureau finally listed him last season—and just in time. It was the 20-year-old’s final year of draft eligibility.

Brace is often told to just be himself and everything will work out. At 180 pounds, he also hears that he’s too small. He has tried to gain weight with a regimen of 1,500-calorie shakes, pounding back one before each workout and another right after. The pressure to prove he’s a physical player is strong. In March, he received a five-game suspension for a hit on another skater. It came during the playoffs—the worst possible time, he says.

“NHL scouts talk to you about getting stronger, getting bigger, bulking up,” he says. “It’s a business, so they want to get the best player they can.”

Actually, scouting is half-business, half-mystical art to the players whose lives hang in the balance. Professional scouts’ routines have been the same for decades—they fly all over the continent, drinking diner coffee and watching small-town games, kibitzing about which boys have a “good face” or fill out their uniforms. They interview juniors about how badly they want it, hoping to sniff something out—instinct, potential, spark.

8 most underrated NHL fantasy picks

Timothy Chan, an assistant professor of industrial engineering at the University of Toronto, thinks there is a better way for the NHL to pick its most precious assets. He hasn’t shaken Cameron Brace’s hand or watched him do sit-ups; instead, the mathematician knows him as a set of numbers. Brace is one of more than 4,000 junior records Chan has fed into an innovative system of player analysis. Small or not, Brace was responsible for an impressive 74 points in 66 games last season. More striking is his “plus-minus,” a fancier stat that shows how his team performs when he’s on the ice. The average junior plus-minus this year was +0.26. Brace’s is +32—only seven other forwards in the OHL scored higher. Ignoring size, age and attitude—too subjective, Chan says—and focusing on the numbers, he ranked Brace 54th among North American skaters and a top-line forward, in contrast to the NHL, which placed him 144th.

With a method cooked up in his spare time, Chan thinks he can take the guesswork out of hockey—what one of his students calls the “crapshoot” that is the draft. His system could ultimately change how the game is monetized, by spotting undervalued players. The harder part may be getting NHL owners—a stubborn breed if there ever was one—to take him seriously.

 

Chan isn’t bucking a trend. In the past few years, teams in every league have been relying less on the hunches of grizzled scouting veterans and more on computer-driven statistical analysis. Call it the Moneyball effect. The 2003 bestseller by Michael Lewis about the Oakland A’s baseball team and its visionary general manager, Billy Beane, who used a data-driven draft process to get to the playoffs, spawned a 2011 Brad Pitt movie and popularized the notion that math could win games. This past March, roughly 2,700 stats buffs, team execs and fans flooded the Massachusetts Institute of Technology for the Sloan Sports Analytics Conference. (It drew about 200 in 2007, its inaugural year.) Attendees were pumped to meet geeky heroes like the New York Times’ stats guru Nate Silver and hear research papers like Acceleration in the NBA: Toward an Algorithmic Taxonomy of Basketball Plays. Who said nerds and jocks have nothing in common?

“The best sports organizations are pursuing analytics more aggressively than the worst-performing organizations,” says J. Michael Boyle, a principal of the Sports Analytics Institute in the U.S. The New England Patriots are almost as famous for their data as for their quarterback Tom Brady; this year, head coach Bill Belichick was given a lifetime achievement award at Sloan.

At the same time, data is becoming a growing sports business. Recent years have seen the rise of firms such as Sportvision, which sell data-collecting cameras that track movements on the field imperceptible to the human eye, such as the precise path and velocity of a baseball pitch. In August, global sports data leaders STATS LLC inked an agreement to place player-position-tracking SportVU cameras in every NBA stadium for a reported $100,000 per camera, per year. The same month, the firm signed another deal in partnership with Intel to analyze massive amounts of football game minutiae such as weather, time of day, the frequency of injuries and player rest durations for the NFL.

For analysts such as Chan, geeking out on data makes sound business sense on two fronts. With the popularity of fantasy leagues, fans crave as much information about their players as possible, and will pay. In the real leagues, team owners want to find potential in players that others can’t see, to pinpoint—in essence—a price-to-earnings ratio for their costliest investment. That doesn’t come from hiring more experienced scouts; it comes from having better data than the other guy.

There’s one notable holdout: hockey. The sport is well behind baseball, football and basketball in the resources it devotes to analyzing its players. Boyle estimates that less than 15% of NHL teams employ even a single dedicated analyst with a math-related background, compared with roughly 60% of NFL teams. Maybe it’s because some in the NHL openly oppose paying math geeks for hockey advice. Former Maple Leafs general manager (and current Calgary Flames president of hockey operations) Brian Burke is so firmly against entrusting sports decisions to computers that he’s become the Sloan Conference’s gadfly, travelling there to denounce it in flamboyant speeches. It’s a little like going to a comic book convention to tell the world Superman sucks.

“Statistics are like a lamp post to a drunk—useful for support, but not for illumination,” Burke quipped this year to a bemused audience (he declined interview requests). Computers might help, but a coach’s wisdom should trump all.

“No one’s ever won a title with Moneyball,” said Burke.

 

Timothy Chan sits at a boardroom table at U of T’s Mechanical Engineering building. His day job is to crunch massive sets of numbers for some of Toronto’s bigger public institutions. Right now, however, he is trying to calculate how old he is.

“Arithmetic, funny enough, is not my strong suit,” says the baby-faced math whiz. “I am . . . 33. No, no, hold on, 1979 is 34, right?”

Originally from Vancouver, Chan has become an expert in solving more complex problems than his age. While still a PhD student, Massachusetts General Hospital pegged him to help optimize their cancer radiation treatments. After he moved to Toronto, St. Michael’s Hospital asked him to work out where in the city the most cardiac arrests take place, to find the most efficient places to locate public defibrillators.

Most people look around and see a messy world filled with illness, war, traffic and shopping. Chan surveys the chaos and sees order in the numbers. “I guess it’s kind of like a disease,” he says. Although health care is his main research area, sports have been his passion since boyhood. A recent paper on baseball stats he co-authored with Douglas Fearing, an operations researcher at the University of Texas at Austin, was presented at the Sloan Conference this year, where it won first prize. But his real passion is the one sport least colonized by nerds. In 2012, Chan and a precocious undergrad named David Novati planted a first flag, publishing a paper on a new way to calculate the value of hockey players. They claim their model is unprecedented.

Hockey players aren’t easy for computers to figure out. In baseball, the Oakland A’s used a single new statistic called on-base percentage to accurately measure a player’s worth. Hockey has no comparable magical stat. To understand why, it’s helpful to know why baseball is so data-friendly. It’s a game of static, separate plays. When a pitcher throws a strike or a batter hits a drive into left field, the event is largely unaffected by which runner happens to be on third base.

A hockey goal might be a simple thing on its surface—a forward shoots the puck and it either goes in the net or it does not—but pan out to the surrounding rink, and goals appear as the final step of a complex, fast-moving dance. Players zip around at varying speeds like fish in an aquarium, having many small, mostly unrecorded interactions. It is, says the Sports Analytics Institute’s Kevin Mongeon, a flow game. Not every player is directly involved in scoring, but just by being on the ice, he affects play.

Instead of trying to evaluate players with a single statistic, Chan smashes them into measurable fragments. He and Novati call it their “Split Personality” model. They begin by graphing players based on a multitude of publicly available figures, among them goals, assists, blocked shots, time on the ice, penalty minutes and plus-minus, creating nuanced portraits. After seeing how they cluster, they classify each player as one of 11 types: four types of forward, four types of defenceman and three types of goalie (from most to least desirable). Sidney Crosby is a top-line forward, a superstar goal scorer. Maple Leafs centre Jay McClement is a defensive forward, a two-way player who blocks shots. Boston’s Shawn Thornton is a physical forward, with lots of hits and penalty minutes—the kind that gets the fewest points. The Canadiens’ P. K. Subban is the best kind of defenceman, an “offensive” one who scores lots of points and logs the most power-play time. They distribute points won by their teams to the players based on their type, giving each player a neat numerical value.

All well and good, but that doesn’t allow for variation within clusters. The group that contains outliers like Crosby also has a lot of very productive, but non-superstar forwards. To refine a player’s worth, Chan and Novati break them apart further. A forward once simply classified as “second line” can now be 50% top line, 30% second line, 15% defensive and 5% physical—just as a person can be two-parts sunny optimist and one-part dire pessimist. Each of their inner “personalities” are assigned points, giving a player a more accurate total value.

When this was done, the whole picture shifted. Crosby was still a top-line forward—worth whatever salary the market was willing to pay—but many skaters with few goals scored, some not yet rookies, moved up the ranks closer to him. Chan was seeing hidden potential in younger, cheaper players who had yet to log enough time on the ice to score.

“That’s the holy grail,” says Chan—prediction. But number crunchers get antsy about words like “predict.” Chan prefers to talk about “reasonable assumptions that the future will be like the past.”

 

Even by cautious standards, it was clear that the young, unknown players atop Chan and Novati’s list were doing surprisingly well. Because they needed forwards to have only 10 pro games under their belts to evaluate them, their first papers swept in green players such as the San Jose Sharks’ Logan Couture and the Canucks’ Michael Grabner, who at that time hadn’t played enough games to be evaluated as rookies, and clustered them near Crosby. Couture hadn’t scored many total goals, but they were high considering he had spent so little time on ice. In Grabner’s case, the Vancouver Canucks traded him to Florida, which waived his contract. He was picked up by the New York Islanders in a larger trade for $765,000. Couture and Grabner would go on to become finalists for Rookie of the Year.

So if Chan’s model is so great, why isn’t the NHL knocking on his door?

Chan and other analysts trying to market their smarts to the NHL are caught in a catch-22. Like hedge funds that protect proprietary algorithms, teams are reluctant to share information. “The advantages of analytics are relative,” says Fearing. “If it were really open, the analytics would be good, but every team would lose its advantage.” And yet without showing teams impressive results, it’s hard to sell the model.

Chan hopes to make waves next spring with the first sports paper he’s ever gotten academic funding for: a breakdown of Canadian junior league players—such as Brace—performed with the help of students Daniel Biancolin and Swapneel Mehta. They looked at every single CHL player from 2005 to 2013 with significant ice time, as many as 2,500 records per year, tracking which players made it to the NHL in order to systematically isolate which traits they share. It’s a canny move. A method to spot valuable juniors would be more attractive to the NHL—and help convert doubters. “We’re where baseball was 20 years ago,” says Chan. “There’s a lot of skepticism.” But change will come.

It might come too late for Cameron Brace. The June 30 draft came and went, and Brace wasn’t selected by any team. “We’re completely perplexed as to why,” says Teresa Brace, who speaks with the measured cool of a diehard hockey mom. “He’s putting up the numbers they’re telling him to.”

Brace plans to keep playing hard in the OHL. With a great over-age year, he could still sign with a pro team as a free agent.

Chan’s next year will be pivotal, too. He hopes to present his work on juniors at Sloan 2014, where he’d rub shoulders with Michael Lewis, Mr. Moneyball himself.

“It’s like Field of Dreams, right?” he says. “If you build it, they will come.”

2 comments on “There’s an amazing new way to value hockey players, but the NHL isn’t buying

  1. Pingback: Buts + passes + quoi encore = victoires… et dollars! | Stéphane Éthier

  2. Pingback: Hockey research highlighted in Canadian Business « Timothy C.Y. Chan

Leave a comment

Your email address will not be published. Required fields are marked *