What is this, Fight Club?
Is the first rule of analytics that NHL teams cannot talk about analytics? Ask too many questions on the topic and employees begin to get dodgy. They speak in code or use vague descriptions, running out the clock as if talking longer creates the illusion of a complete answer. They say things like:
“I can’t think of an easy way to do it without getting into things that I probably shouldn’t get into.”
“We probably can’t be that specific.”
“Getting into specifics of an example or two would be a little challenging for me.”
Every NHL team today has an analytics department, and each team protects its statistical secrets as though they belong in the giant warehouse where Steven Spielberg stashed the Ark of the Covenant. That advanced hockey statistics are held in such high regard is quite the coup for the “spreadsheet” community. Ten to 15 years ago, before terms like Corsi, Fenwick, possession, WAR and expected goals went mainstream, advanced stats were a niche for nerds. Hockey’s gatekeepers – the old-guard establishment comprised mostly of former players who’d ascended to roles as scouts, coaches and GMs – brushed them aside.
The most reductive version of the anti-stats argument claimed eyeballs beat spreadsheets 10 times out of 10. But something began to change roughly a decade ago, just as it had in the MLB in the early 2000s when Moneyball came about. Open-minded hockey thinkers began to recognize correlations between winning and having strong peripheral statistics. In a seven-season stretch from 2009-10 through 2015-16, the Stanley Cup winner finished in the top four in 5-on-5 Corsi percentage six times. Journalists started championing the validity of stats websites and bloggers, and NHL teams quietly took notice.
Two of the first to do so: the Toronto Maple Leafs and Carolina Hurricanes. When Brendan Shanahan took over as Leafs president in 2014, he remade the franchise with progressive minds, starting with the hiring of Kyle Dubas as assistant GM. Dubas was known as one of the young hockey executives most interested in analytics at the time.
Within weeks of that hiring, Toronto snapped up Darryl Metcalf and tasked him and writers Cam Charron and Rob Pettapiece with creating a team analytics department. Metcalf had cofounded Extra Skater, one of the most popular advanced-stats hockey websites consumed by the public back then. Hockey stats were a side hobby for him at the time. He had a degree in chemical engineering and worked for an information technology/marketing firm.
In 2015, the Hurricanes hired Eric Tulsky, who was best known for his deep-dive analytical writing on blogs like Broad Street Hockey and FiveThirtyEight. He thought of himself as a conduit between the science and sports worlds, helping convert jargon to relatable terms so readers could understand why specific stats mattered in a hockey context. “A translator role from math to human was kind of how I got started,” Tulsky said. “And it was early enough that, as I was doing that, there were a lot of things that hadn’t been done yet.”
Metcalf ascended from analyst to being the Leafs’ director of hockey research and development, and, today, is special assistant to the GM. Tulsky rose from one-off consulting projects with multiple teams to working exclusively with the Hurricanes as an analyst. He became a manager of analytics, then vice-president of hockey management. Today, he’s assistant GM of the Hurricanes.
So what does a typical day look like for someone tasked with running an analytics department? As Metcalf explains, a data analyst’s job changes along with the calendar. In the middle of the regular-season schedule, his work focuses on studying opponents and player tendencies to gain advantages for the next game. Approaching the trade deadline, he’ll be analyzing potential targets. The research focus will shift again leading up to the draft and again to prepare for signing players to contracts during the frenzied off-season.
Tulsky deliberately stays an arm’s length from the players during day-to-day operations. Many NHLers, from Mark Scheifele to Drew Doughty, have publicly bashed analytics in recent seasons and may be somewhat prickly to being instructed via charts and stats. In Carolina, Tulsky communicates his analysis to the coaching staff, headed by Rod Brind’Amour.
In Tulsky’s mind, the coaches can translate the analytical findings in terms the players understand. He also doesn’t want to deliver mixed messages, emphasizing a given piece of data that a coach doesn’t choose to prioritize. “If the coaches don’t think it will help players to see it in data, if they think it’s better to show them video or to talk about it in the abstract, that’s what they should do,” Tulsky said. “They need to judge their players and have the communications in the way that will be most effective for each player.”
So what exactly are the findings analytics employees share with NHL teams? What are the secret custom statistics teams guard like nuclear launch codes? The analysts don’t just regurgitate numbers anyone can find online. After the 2019-20 season, when asked about then-Leafs defenseman Cody Ceci’s ugly defensive numbers, Dubas remarked Ceci graded out much better through the lens of the Leafs’ private metrics. It’s no secret NHL teams have access to more robust and detailed datasets than the general public, which gets its numbers from what the NHL publishes on its website. Stats like Corsi paint the picture of a game in terms of who controls the shot attempts. “Expected goals” factors in shot quality to create a more accurate representation of which team is threatening to score more. But what new metrics, if any, have teams discovered by investing so much into analytics over the past half-decade or so?
Yeah…that’s where the questioning hits a wall, which is understandable. What team wants to reveal its secrets? Some are kind enough to offer interesting crumbs of information, at least. “I would say that the event data that the NHL publishes is very focused on what happens with the puck in terms of obviously goals and assists, takeaways and giveaways and shots and things like that,” Metcalf said, “but there’s really not a lot of information available about what happens (when) players don’t have the puck.”
Added Tulsky: “It really just does come down to having more detailed data and more information about what assignments were and what people were supposed to be doing on a certain play.”
Perhaps we can glean what teams are looking for based on what type of requests they outsource. Analytics company Stathletes, for instance, has worked with NHL teams as clients for almost 10 years and collects information using every video stream it can find of every on-ice event.
So, might cofounder Meghan Chayka let us in on what teams are after? Just a hint? Please? “It’s a lot easier to quantify offense in hockey, so goalies and defense are always of interest,” she said.
Team analytics departments are clearly unearthing some highly specific and granular hockey stats. A focus so narrow has pitfalls, however. A good explanation of the risks comes from the analytics industry’s new breakout online star, Jack Fraser, best known by his Twitter handle @JFreshHockey. He works in the international politics field for a day job but found himself experimenting more and more with hockey data during downtime because of the pandemic.
Now, he’s setting a standard for rapid player analysis with his graphical representations known as “player cards,” which present easy-to-read summaries of players’ offensive and defensive impacts to give instant snapshots of how they impact the game. He believes teams’ analytics departments drill down on particular skills, which would explain the Ceci-like situations in which a team seems to like a player much better than the public does.
“Let’s say you’re taking a player who would not be considered very good by the analytical community, a defensive defenseman who gets caved in every night,” Fraser said. “If you’re looking at (publicly available) stats, your Wins Above Replacement, things like that, you’ll say, ‘OK, this guy does nothing,’ right? But if you’re looking at the micro stats, you quickly figure out every player is good at something, and that makes it even more challenging to properly evaluate them. That guy might be (rated poorly) across the board, but then his puck-battles-won percentage is super high, or his stick tie-ups in front of the net is super high. And you can see why a GM who has a strong analytics team may see that and say, ‘That brings the element we need.’
“These teams have so much stuff available to them and stuff that is advanced past the level of what we in the public do. The main question really comes down to having people who can figure out, with that data, what will actually make a relevant addition to their team and what is just ramping up (skills) that may be catnip to more traditional GMs, scouts and coaches and actually isn’t making a difference.”
Someday, external sources of data will become as sophisticated as teams’ internal stats are. The analytics world already trends in that direction. Think of how many stats have evolved and changed in relevance over the past decade. Corsi, for instance, is viewed as an oversimplification at this point and is even vulnerable to players gaming the system and puffing up their numbers with low-percentage shot attempts. “Corsi, Fenwick, different types of things that are just using shots without any sort of context around them, those are very good when you only have a box score but may not be as desirable when you have a robust dataset,” Chayka said. “So moving away from that is just natural with the type of data we’re seeing being collected at the league level.”
Fraser says the evolution of data also changes how we perceive certain players even compared to a few years ago. Stat heads vehemently defended Erik Karlsson’s two-way game when he was winning Norris Trophies, for instance, but the more accurate information available in 2021 tells us he wasn’t the player people thought he was. Or he was exactly the player people thought, depending on one’s initial position. “Karlsson wasn’t allowing a lot of shots, but the shots that he was allowing were really, really good,” Fraser said. “And he was allowing these rush chances, 2-on-1s, breakaways, all kinds of stuff because he was so far up the ice. That breakaway, in terms of Corsi, is worth the same as a shot from the point, but in terms of how likely it was to go in the net, it was a lot more dangerous…so that is kind of a useful correction to how, when you get improved stats, you can re-evaluate some takes that you had when you were using rocks and stones instead of bronze.”
The quest for improved stats is now an NHL obsession. All 32 NHL teams have employees devoted to some form of analytics. The Leafs are the leader with eight, while the Seattle Kraken and New York Islanders sit second with six each. Tulsky’s Canes check in with a solid four. The cutting-edge Kraken made analytics-oriented thinkers three of their earliest front-office hires in director of hockey strategy and research Alexandra Mandrycky, senior quantitative analyst Namita Nandakumar and quantitative analyst Dani Chu. Mandrycky had a significant influence on the hiring of Ron Francis because she was tasked with analyzing all the Kraken’s preferred GM candidates.
NHL teams employ 102 people (that we know of) for data-driven roles today. It’s a massive jump from when teams were perceived to be taking chances on people like Metcalf, Tulsky, Sunny Mehta and Tyler Dellow several years back. Hockey analysis has become a viable career. Accrue a large enough following and you could be the latest blogger poached. Is Fraser the next to get an NHL gig? He’s not sure, but he agrees that the analytics community sees NHL jobs as realistic goals in 2021.
“I know that there are a lot of very ambitious people who are dead set on finding themselves in an NHL front office,” Fraser said. “My approach is I’m happy doing what I’m doing right now, especially because this isn’t really my primary field or my primary career. If a team did come calling, I would be interested. I would listen to them. But it’s not something that I’m seeking out so much as I think a lot of people in this space are.”
Fraser should expect a call soon whether he wants one or not. The trend will continue. As trailblazers like Tulsky and Metcalf suggest, the picture Moneyball painted is no longer accurate. In team meetings, crotchety white-haired scouts don’t spit on spreadsheets, nor do analysts roll their eyes at the old dinosaurs. That black-and-white era is extinct. The old and new guards cooperate, and analytics is now permanently entwined in the sport’s decision-making. Once you learn how to better evaluate something, there’s no unlearning that.