A couple weeks ago I had a civil discussion on Twitter with WFAN’s Jeff Capellini (@GreenLanternJet) about the application of advanced stats in the NHL. Those of you who know me or Jeff won’t be surprised to see that we were on different sides of the fence when it came to how we felt about the importance of statistics like Corsi (all shot attempts) and Fenwick (unblocked, on-target shot attempts) in today’s game.
Perhaps it’s a generational thing. Perhaps not. Point being: while we agree that the so-called “fancy stats” are now a topic of conversation more than ever, we differ in our opinion of whether or not they provide a true measure of what’s important in a hockey game.
Two things about that exchange: 1) Yes, it’s possible to engage in constructive dialogue on Twitter.com (I only said he was dead to me after he brought up the FOX glowing puck), and 2) It got me thinking about the prevalence of advanced stats in even basic discussions about the NHL. Moreover, it made me wonder why, with the slew of recent stat-centric hirings by NHL teams, there’s still resistance across much of the mainstream media toward fancy stats’ usefulness.
The levels of hot take-itude at which certain mainstream media members operate when voicing their displeasure with advanced stats is rivaled perhaps only by ESPNs First Take’s daily shout-fest. (Which is not a flattering comparison by any means.)
Is this type of mainstream hockey writer against fancy stats solely because those numbers don’t always tell the full story? Or is he against them because they allow people who “don’t watch the game” to analyze and interpret what happens on the ice at a deeper level? Is this just a case of the old boys club being infiltrated by kids with spreadsheets? And if so, why should that be a problem?
To be clear: I don’t consider Capellini among the CORSI IZ DUMB JUST WATCH THE GAME YOU GUYZ group at all. His argument against the acceptance of advanced stats as the be-all, end-all of hockey analysis is simply that: Corsi and Fenwick aren’t the same things as goals and assists. Which, true.
I don’t pretend to know everything—or even most of the things—about advanced stats. But, I do know that denying their importance to how an NHL game is analyzed—or to how an NHL roster is constructed—is shortsighted. In breaking down and reporting on today’s game, an understanding of advanced stats is crucial. You don’t have to accept them as gospel, but you’re doing yourself a disservice if you dismiss them entirely.
Good thing the Leafs don’t play in the CHL. The CORSI hockey league. They’re doing just fine in NHL, though.
— steve simmons (@simmonssteve) October 30, 2013
Take the NBA, for example. The widespread use (and public availability) of data collected by SportVU has enhanced the level of detail at which the game is viewed by players, coaches, general managers, and fans. And the funny thing is, even this type of data collection hasn’t generated the type of war being waged between the mainstream media and bloggers when it comes to the NHL and fancy stats.
OK, then why is this “a thing” in hockey but not basketball? (Or baseball, for that matter?)
The battle over advanced metrics in hockey is based on one argument: Corsi and Fenwick aren’t goals and assists; i.e., they’re not worthwhile measures of a player’s importance to his team because they don’t directly contribute to the box score in a tangible, two-points-in-the-standings type of way.
There are a couple of problems with this argument. First, goals—and by association, assists—are “random” events (i.e., more easily influenced by variation in other areas of the game than shot attempts are). Their occurrences can vary wildly based on fluctuations in shooting percentage, meaning a team with a high shooting percentage can appear better than what its true talent level would indicate. (Conversely, a team with a low shooting percentage will appear to be worse than it really is.) Since shooting percentage regresses to the league average rate—since 2009-10, that average has been approximately 9 percent—over time, the amount of goals scored by a team will increase or decrease in lockstep with that regression.
A quick aside. I get it: since goals are truly the defining statistic in terms of which team wins the game, we, as hockey fans, generally like to analyze a player’s importance by the number of goals he scores. It’s difficult to reconcile that way of thinking with the fact that goals are random events, subject to variations in a player’s shooting percentage that take a while to even out. And in this age of instant gratification, it’s tough to wait for that “evening out” to occur. So, is basing your evaluation of a player’s skill solely on a metric influenced by statistical “noise” really the best (or only) way to get a picture of who that player is on the ice?
Hold that thought. We’ll get back to it.
The second problem with the anti-advanced-stats argument in that Corsi and Fenwick are proxies for puck possession. They’re not designed to replace goals and assists as the means to determine which team won or lost a game. Corsi and Fenwick represent a player’s ability to possess the puck; nothing more, nothing less. And since the team that possesses the puck the most tends to generate the most shot attempts, and because more shot attempts mean more chances to score, it follows that more chances to score generally mean more goals.
Still with me? Good.
Ultimately, puck possession is key to a team’s success in the long run. This is why we’ve seen teams start to shift their tactics away from the dump-and-chase strategy and toward a carry-based approach when entering the offensive zone. (Or exiting the defensive zone, for that matter.) Coaching staffs have become more attuned to the fact that the teams controlling the puck at a higher rate are controlling the flow of the game, and generating more shot attempts as a result. More rubber thrown at your opponent’s net means less rubber directed at your own, etc. etc.
So, if we know that possessing the puck is inherently the game’s single most important aspect, then doesn’t it stand to reason that a proxy for puck possession would be a useful statistic? (Hint: it is.)
I mean, is there truly a good reason to not track puck possession? Since goals are random events—in a statistical way, not in a get-hit-by-lightning way—why not track a less random, more repeatable type of event if you’re looking for a way to parse out which players are better at generating scoring chances? Players can repeat their ability to carry the puck more often than they can repeat their ability to score goals, statistically speaking. Corsi and Fenwick thereby enable teams to better evaluate which players are driving gameplay and improving their team’s chance of winning.
Short of having a guy with a stopwatch assigned to each player on the ice to measure exactly how long each player controls the puck, we’re reliant on Corsi and Fenwick—in conjunction with other such non-box-score metrics—to track puck possession and see which guys are driving the bus (so to speak), and which guys are just passengers.
(Yes, that was a Jack Capuano-ism. It worked perfectly. I will not apologize.)
The counter-argument: “But if Corsi and Fenwick are such perfect indicators of how good a player is, then why does BAD PLAYER X have a better rating than GOOD PLAYER Y ?”
Fancy stats aren’t meant to be stand-alone indicators of a player’s value. By combining a player’s Corsi rel or score-adjusted Fenwick—variations on generic Corsi and generic Fenwick that account for in-game situations like linemates and score effects—with a metric like zone start percentage, or narrowing your dataset to include only minutes played at even strength, you can get a more complete picture of a guy’s value. In short, taking several variables into account allows for a more thorough evaluation of his ability.
Besides, metrics like score adjusted Fenwick are better indicators of a team’s future performance than they are of a team’s current form. Their predictive value is one of their more underrated aspects, and that’s something you’d miss if you took the approach that you couldn’t be bothered to learn about advanced stats. Which, hey: your life, your decision.
The fancy stats ideological war shouldn’t be “a thing.” It shouldn’t be an issue of bloggers impinging upon the mainstream media’s space, it should be an opportunity for both groups to further their understanding of what happens on the ice during the course of a game.
Player tracking technology like SportVU and Sportvision is on the horizon, with a handful of NHL teams slated to test various forms of the technology during their games in 2014-15. If things progress well, a league-wide plan to implement player tracking—and compile the data that comes with it—could be in place for the 2015-16 season.
So what does this mean?
It means that Corsi and Fenwick will be a thing of the past. But, that doesn’t mean the concepts behind advanced stats are going away. It just means that the work pioneered by guys likeTimothy Barnes (née Vic Ferrari), Darryl Metcalf, Cam Charron, Eric Tulsky, and Tyler Dellow will have resulted in a better method of tracking the type of player movement and scoring chance data they’ve been using “fancy stats” to determine for years.
In the meantime though, Corsi and Fenwick are pretty good substitutes.
Your eyes can fail you. Just watching the game sometimes isn’t enough to really understand how a player performed or how a game unfolded. And even though data in a spreadsheet doesn’t necessarily tell the whole story, it tells a compelling one. One which NHL teams (like the Edmonton Oilers, New Jersey Devils and Toronto Maple Leafs, for example) are starting to understand.
I’m not here to attack the “watch the game” crowd, I’m only here to say that advanced stats are coming, their widespread acceptance notwithstanding. In fact, they’re already here.
Advanced stats courtesy of www.BehindTheNet.ca
Follow me on Twitter dot com. That way, I can stop posting all those Hockey Fan 4 Hockey Fan Missed Connections ads on Craigslist.