Reflections from Sloan Sports Analytics Conference

This past weekend, I attended the MIT Sloan Sports Analytics Conference in Boston. While I was there as a member of EPSN’s True Hoop Network of basketball bloggers, I also got the chance to check out a few hockey-related presentations and talk to some people about hockey analytics. I plan on writing another piece or two later in the week off of specific presentations, but today I just wanted to do a stream-of-consciousness style post about the state of analytics in hockey.

The biggest thing that sticks out is how hockey is still in the infancy stage of developing analytics. Compared to baseball, basketball, football, and even MMA and soccer, hockey had a pretty limited presence at the conference. In addition, once people found out I was a hockey fan, I got a LOT of questions about hockey analytics. I might not be an expert, but I seem to have a better understanding than a LOT of people who don’t even realize the hockey world is doing anything.

And I guess it makes sense. Hockey is extremely difficult to analyze quantitatively. It has a few pretty serious constraints relative to other sports:

*There are very few measurable events. In a 60-minute game, there are five or six goals, less than 100 shots and then some hits and penalties mixed in. This is a small number of events to look at, especially since the desired outcome, "goals," is so rare.

*Hockey is extremely fluid. In baseball, and even football, things are pretty stop-and-start. But with hockey, the game is much more continuous and free-flowing, making it extremely difficult to track in real-time.

*Nothing in hockey is binary. The biggest benefit baseball has over other sports is that a large chunk of the analysis is "batter against pitcher." In hockey, perhaps the only binary (yes/no) situation is the shootout – even a penalty shot comes about because of a penalty, which comes about for any number of reasons.

So basically, hockey has just a few events that can be "counted," and even those are very messy since so much is going on. It’s not easy.

But the hockey world is trying. The biggest breakthrough we’ve had is perhaps Corsi and Fenwick, which, as my advanced stats primer explained, do a better job of predicting future team performance than just goals for and goals against.

We also now know that things like Zone Start Percentage and Quality of Competition can have an impact on how a player performs, so we have to look at the context of a player’s performance to better understand their skill.

These are good first steps, but there is a tonne that can still be done. Teams appear to be experimenting with visual tracking system that can create x-and-y-coordinate data about how players move on the ice and what comes of it. This is going to be extremely difficult to analyze when the data set is rich enough to do so, because not all actions on the ice are intuitive. For example, body checks are actually negatively correlated with performance in some cases, but that’s because hitting someone implies your team doesn’t have the puck. This stuff is going to be dirty and messy at first, but it’s necessary to move forward with.

Hockey may never have a "Moneyball" style statistical revolution. But that’s perfectly fine. A big theme through all sports at the conference was that nobody is trying to reinvent the wheel, just make gains at the margin. With the NHL standings constantly extremely tight, it certainly seems enticing to be able to out-analyze an opposing team to gain a slight edge, even if it’s only a win or a point here and there.

Unfortunately, I wasn’t able to speak with any team executives while at the conference to get a feel for exactly what teams are doing in their front offices. I’d imagine it’s still far more qualitative than quantitative, but that’s still analytics. Things you see with your eyes and learn are still analytic in nature. Analytics doesn’t necessarily mean stats, so if you’re anti-stat that doesn’t mean you’re anti-analytics (and neither position would be justifiable at the extreme, really). If you think, for example, the Canucks dump-and-chase too much because that rarely leads to zone control, that’s analytics. The hockey world is just working to try and quantify that and help coaches and players better understand it.

It’s a tough road ahead for the hockey world, for sure. It seems really odd to quote Brian Burke, who was at the conference and was completely trolling stat-heads with this comment, but it’s a fair point to remind the anti-stat crowd what this is all about:

"Stats are like a lamppost to a drunk. They’re good for support, not illumination."

Now, I realize that the quote makes NO sense (seriously, a lamppost would definitely be good for illumination). However, the back half is a nice way of explaining the use of analytics in sport – nobody is trying to reinvent hockey, but advanced stats can definitely support a team with strategic and personnel decisions.

I’d advocate for more stats in the hockey vernacular, obviously. At the same time, there’s no way to quantify something like this:

Sorry, I had to. But seriously, how do you quantify a play like that? Damn.

Follow @BlakeMurphyODC