clock menu more-arrow no yes

Filed under:

Invented arguments and advanced stats

New, 51 comments

How the old guard uses specious arguments to try to maintain their veneer of expertise in light of a new movement that is democratizing analysis.

This is a puck.
This is a puck.
Bruce Bennett

Gare Joyce is one of Sportsnet's lead writers on junior hockey. In the past, he has written some pretty excellent pieces on the subject, including one of my favorite all-time pieces about former-Leaf Alyn McCauley and the impact numerous concussions had on his career. Gare and I follow each other on twitter and we've had a good back and forth from time to time.

All of that said, I really take issue with his piece last week for Sportsnet, titled "Analytics hires are quick fixes to PR nightmares". The gist of the argument, for those who have not read the piece, is that the sudden rush by NHL teams to hire "analytic gurus" this summer has been more to placate or appease the fans and less about developing a competitive edge. I don't have much interest in arguing that point, as much I as I think it's a stretch. However, I do want to address some points Joyce makes that seem to be a common occurrence in such articles.

I was a fan of Bill James back in the day when he had Jesse Barfield and Lloyd Moseby as the worst outfielders in baseball. Back then he was probably working out his stats models on a TRS-80. I always got Sabermetics. I think baseball and hockey are so fundamentally different that Sabermetrics approaches can’t possibly run in lockstep.

Joyce makes the rather ubiquitous comparison between baseball and hockey. That is that the sabermetric approaches "work" for baseball but won't for hockey. We've all seen these arguments before. Baseball is more driven by the individual players and involves discrete events that lend themselves to be easily tracked whereas hockey is a free-flowing game and it's difficult to separate the team effect from the player's effect. That's not necessarily wrong. What is wrong is the conclusion that, as a result, hockey cannot be analyzed empirically. Just because hockey presents greater issues of data collection does not mean data collection yields no advantages. Data can be pulled from damn near anything. The difficulty is in getting the data, not using it.

Go ask Sun Media and Rogers if they employ full-time statisticians; I would bet a pretty penny that they do. Companies long ago discovered that information is a major key to success, and with the kind of computing power now available, data collection and analysis has become absolutely vital. And that's what this is all about: collecting and analyzing information. The more you have, the better prepared you are to make decisions. Hockey, like just about anything else on the planet, presents an opportunity for data collection and interpretation. Data is already being pulled from hockey, and teams are putting money and time into capturing more of it (and in new ways). To suggest that hockey cannot be analyzed empirically due to the nature of the game couldn't be less accurate. No, it may not be as easy as baseball but that doesn't mean there's nothing to gain.

I don’t imagine that Corsi or anything deeper than that will ever have a utility in evaluating draft-eligible prospects, which is my own interest. I don’t know an amateur scout who can really see it that way. Talent is the game’s most precious resource and ranking it at maturity is hard, but a cakewalk compared to evaluating it as it’s emerging. You have to weigh a kid playing high-school hockey in a crap league versus a kid in the Frozen Four versus a kid in Swedish Div-2 versus a kid in the KHL. And then you have to weigh a kid’s chances of growing two inches or more and his attention span or the effects of an injury. I don’t think analytics will provide a unified theory that will neatly sort out a prospect’s chance at the next level. Values in development are just too fluid.

First of all, a metric already exists to help decipher how prospects from different leagues will perform in the NHL. It's called NHLe and it's useful but not perfect. No single number is going to unlock the future of an NHL prospect, just like no single skill is going to get a prospect to the NHL. Yet skeptics of advanced stats often decry their value because they don't provide that single metric that acts as the definitive evaluation of a player's worth. The fact is, it would be pretty rare to find a single value that described an entire system of events. Wins Above Replacement (WAR) is often pointed to in baseball as such a value, but even it has its issues. It's true that there have been attempts at a similar metric for hockey, with none really succeeding to this point. Here's the thing: the goal right now isn't to find some kind of perfect, end-all metric. Even if one did exist that seemed pretty close, it would have to be questioned because the available data kinda sucks. And if we can't do that at the NHL level, it's not happening at the junior hockey level.

So that's why these statements about eliminating scouts always seems to puzzle me. You wouldn't send a single scout to view a prospect. Scouts, as humans, are subjective, unreliable, and subject to bias. It's why highly skilled players always seem to slide when they're under 5'11. It's why future NHL stars sometimes wind up being drafted in the 2nd, 3rd, or later rounds of the draft. NHL teams figured this out long ago. It's why teams send scouts to view prospects outside of their normal region: two sets of eyes are better than one. The NHL even runs a central scouting department, despite the fact that all 30 teams manage their own scouting staffs. The core tenet is: collect as much information as possible. So the idea that teams would even consider abandoning their amateur scouting staffs is not a well thought out one. Advanced stats are simply another tool in the kit. Sabermetrics hasn't killed scouting in baseball. It won't kill scouting in hockey.

What continues to drive myself and others nuts is not so much the skepticism regarding analytics in hockey, but rather the ridiculous invented arguments thrown out as a defense of said skepticism. "Sabermetrics won't work in hockey because the sport is too different!" The only part of that sentence that's accurate is that the sports are, in fact, different. Good job. Or "let's replace all of the scouts and hockey minds with numbers!"  NHL teams already admit that scouts are imperfect. They use layers of evaluation, such as video analysis, in addition to live-viewings. So I fail to see how, if video analysis didn't replace scouts, why analytics would.

Lastly, it continues to perplex me why writers such as Steve Simmons, Damien Cox, and yes, Gare Joyce are so convinced that analytics are a fad. They haven't been a fad in any other discipline. They certainly aren't a fad in science, business, or even other sports. And you would think that, as the supposed intermediate between team and fans, it makes a lot of sense to learn as much as is reasonably possible about advanced stats in order to paint a cleaner picture for readers. "Here's what X team is doing. Here's why they think it might help. This is news." But we don't get that, for whatever reason. And that only hurts hockey and the fans. Just as inventing arguments against analytics does.