Do advanced stats really work in hockey?
It’s been a good decade since professional hockey started up its bromance with advanced statistics like Corsi and Fenwick, but aside from getting a few more math geeks into the front office, are teams benefiting from the new focus on analytics?
While hockey was slower to catch onto the trend than other sports such as baseball —where every player is now evaluated by a broad spectrum of metrics from OPS+ to WAR to BABIP— Canada’s favourite obsession now comes with its own set of terms that try to root out a player’s real value, beginning with Corsi, a puck possession stat which measures shot attempt differential, as well as others like PDO, Expected Goals and Zone Start Percentage.
The point is, of course, to take more of the guesswork and subjectivity out of judging a player’s worth, thereby producing a better, more predictive look at a hockey team’s chances of winning. Experts say, for instance, that Corsi is more predictive of a team’s success than bare goal differential.
And the stats impact is clearly being felt around the NHL, as teams continue to pull data crunchers from fan websites and analytics companies and put them on the payroll. Last year, ‘Moneypuck’ got one of its biggest endorsements yet when the Arizona Coyotes hired as its new general manager 26 year-old John Chayka, a former junior player whose company Stathletes compiles data on hockey players for NHL teams.
But a good ten years in and not everyone has drunk the Kool-Aid. The hockey media love it, for instance, when coaches get saucy about advanced stats and insist that a player’s worth can’t be summed up in “comparing apples and oranges,” as Winnipeg Jets coach Paul Maurice put it earlier this month. When questioned about Jets forward Matt Hendricks’ poor numbers, Maurice replied, “You can’t win with only one style of player. Corsi doesn’t tell you everything. Or anything, possibly.”
We have businessmen in this league. They will throw pucks from anywhere to get a better Corsi…
Other coaches have complained that advanced stats might actually be changing the way the game is played, by shifting the focus more to puck possession, à la Corsi. Last year, Washington Capitals coach Barry Trotz accused some players of shooting on goal even with little to no chance of scoring, just so they could pad their stats. “We have businessmen” in this league, said Trotz to USA Today, “they will throw pucks from anywhere to get a better Corsi.”
Sports pundits say that hockey still has a long way to go when it comes to finding the most usable set of metrics, with many debating the value of the Corsi in its various guises.
“I think we’re still in the infancy of hockey analytics,” says Johannes Wheeldon, who has written on the history of hockey analytics for The Hockey Writers, in conversation with Cantech Letter. “Today, the most commonly referred to stat is Corsi, but there are serious caveats with Corsi. Do teams with better possession score more goals? The short answer is they don’t, and teams with good Corsi ratings are not guaranteed to be good hockey teams.”
“Corsi is easy to manipulate,” says Salim Valji, statistician for TSN’s Montreal Canadiens broadcasts, to Cantech Letter. “That makes it less reliable. Edmonton is one of the worst NHL teams yet has the second-best Corsi, which goes to show that there’s still work to be done regarding the accuracy of these numbers and what they mean.”
I think hockey’s a hard game to quantify. Baseball is very much a series of events —the pitcher throws the ball, the batter attempts to hit it, the fielder covers a certain amount of ground…
“I think hockey’s a hard game to quantify,” says Valji. “Baseball is very much a series of events —the pitcher throws the ball, the batter attempts to hit it, the fielder covers a certain amount of ground. With hockey, it’s a lot more free-flowing. It’s about play in the neutral zone, it’s about defending beneath the goal line, it’s about zone exits and entries. Those are much harder to quantify.”
But just because Corsi has its flaws doesn’t mean there’s nothing new that analytics and Big Data can bring to hockey — it’s just that much of the new tools and information is currently being mined by private companies and sold to teams through exclusive contracts. Companies like Stathletes, Sportlogiq and HockeyData Inc. have dozens of employees putting in thousands of hours watching video and employing their own metrics, all of which is under proprietary wraps.
“Any of the models that are out there modelling the game through Corsi and Fenwick and PDO and stuff like this, they don’t get down to the root of it, of understanding player performance in my opinion,” says Statlethes CEO Neil Lane, to the Edmonton Journal.
“The fact of the matter is there’s only so much data available to the general public and people are trying to do the best they can with what they have,” says Lane. “That’s why we built this business, to bring more data and do better.”
Even outside of the proprietary domain, though, the general feeling is that hockey’s advanced stats will keep getting better, and that will be good not just for the teams but for the sport, as fan engagement will become that much more sophisticated. “To the extent that commentators can connect credible analysis with what viewers can see for themselves, I expect we will see more reference to analytics in hockey broadcasts,” says Wheeldon.
“You can’t rely on stats to tell you who will win the Stanley Cup, but that doesn’t mean they don’t have value,” he says.