Islanders sound off about ‘flawed’ analytics in NHL

TAMPA, Fla. — The Islanders don’t want to hear about your math equations.

Though they have acknowledged that analytics has a role in hockey, they still value the eye test and gut feelings over numbers.

“There’s will and structure to our game,” coach Lane Lambert said Thursday. “Those don’t really get analyzed by analytics. We analyze it.”

There has been recent growth in the use of numbers in hockey — every team, the Islanders included, has staffers with analytics as part of their portfolio — but compared to the other three major sports, the NHL is still far behind. That’s in part due to the nature of the game. Hockey is continuous, random and hard to break down into the sort of individual events that are easy to analyze mathematically, such as a pitch in baseball, a snap in football or a possession in basketball.

Most of the publicly available numbers in hockey (individual teams track their own data) use shots as stand-ins for possessions, a reasonable but rudimentary method of estimation. Expected-goals models, which are widely cited, can be useful, but because there is no public puck-tracking data, they fail to account for factors such as whether a shot was preceded by a pass; such shots are harder for a goaltender to stop.

“It’s everywhere nowadays so I guess you certainly see it,” Matt Martin said. “I think it can definitely be flawed as well. I think sometimes you might have a really good game defensively against someone’s top line where you lose the Corsi battle or whatever it is, but you keep them off the board and don’t give up a ton of high-danger chances. And analytically, sometimes that shows up as a very poor game where you’ll get a tap on the back from your coaching staff about how good of a job you guys did.”

The Islanders’ fourth line, which Martin is a part of, is a good test case. Because they spend a lot of ice time in the defensive zone or on the forecheck, their analytics reflect negatively on their play. But the goal of the fourth line is not necessarily to generate offense, but to leave a physical impact on the game, wear down opponents and keep them off the board.

“I think you can walk around a dressing room, someone can tell you whether they had a bad game or a good game,” Martin said. “We know as professionals. … You get matched up against Connor McDavid, you’re probably gonna lose the Corsi battle. But if you limit the amount of high-danger chances he has and keep him off the board, we take that every single night.”

Martin said there isn’t much interaction between the Islanders’ analytics staff and the players. Taran Singleton, a longtime video coach with the Devils when current Islanders general manager Lou Lamoriello was in New Jersey, is one of five staffers devoted to analytics.

“I’m not a big analytics guy,” Zach Parise said. “I just think there’s too many factors in this game that don’t get accounted for. I’ll go as far as shots on goal, cause you give yourself a chance to score. Anything past that, it’s a little difficult to measure.”

Like Martin, Parise cited factors such as the opponent as a reason analytics fall short.

“Are you playing against [Patrice] Bergeron and some Selke [Trophy] winners? Or are you playing against a fourth line all night?” Parise said. “Where are you starting on a faceoff? Are you starting every shift on a controlled breakout? Do you lose the draw? If you lose the draw, you might not touch the puck the entire shift.

“You don’t reset all the time and start over like a pitch [in baseball], you know what I mean? That’s just my theory. I could be wrong. I’m sure there’s value in it, but I don’t stress much about it.”