Sunday, January 10, 2016

Are We There Yet? Advanced Stats in Hockey

Hi, my name is Scott (Henry Street on Twitter) and I'm a former minor hockey coach in Oakville. I love hockey and I really embrace open debates about the game. I have always enjoyed stats but I don't identify myself as a "stats guy" by the modern definition. It's very clear that there is a demand for advanced stats in hockey, as we've seen a large contingent of a new generation of fans over the last two to three years really embrace stats as the gospel for defining how good a player or team is. I really enjoy using stats as a tool to reference and help with evaluating players but the current data is not intricate enough to tell the whole story. As a result we see many fans over simplifying by using unreliable, general stats. Kyle Dubas has made a recent statement reflecting how rudimentary the current state of stats is in the NHL. Don't agree with Kyle? A man who was hired by the Toronto Maple Leafs for his innovative use of advanced stats while General Managing an OHL team? Well that's fine, I hope to prove to you the various ways that stats could and probably will be improved in time.

I'm not going to give a Corsi lesson here, or get into all of that, if you are unfamiliar there are a plethora of blogs and sites who can explain the concepts. In theory they do hold some ground, but the main problem I have with this whole wave of thinking lies in that the root of the advanced stats argument relies so heavily on shots - shots on goal and shots thrown towards the goal, which could be blocked, saved, result in goals or whatever. It favours throwing pucks towards the goal over all else. Even if, the best play may have been to make a pass.

Continuing on this line of thought, those who base their entire evaluation of a player on stats rely heavily on the use of shooting percentage. This is a very flawed stat, and relates directly to the comments of Mr.Dubas. I'm going to try to keep this as simple as possible and throw out some factors that may or may not have been considered by everyone in the analytic's community, which are not reflected in the shooting percentage stat:

The basis of my argument can be summed up by this statement:
NOT ALL SHOTS ARE CREATED EQUAL - Shot Quality is Important

To break down the general Shooting % stat into a more accurate, telling stat, we must consider each shot:

What was the velocity of the shot?
Was it a rebound?
Where was the shot taken from? (Shot charts are great for this but the other factors listed have to be taken into account along with location. Location alone doesn't teach us much.)
Was the goalie screened? By how much traffic?
Was the goalie cold coming off the bench?
Did the shooter have a pass option that could have made the goalie cheat off his post on a short side goal?
Did the shot hit a stick?
Did the shooter aim for this particular goalie's weakness? Does this goalie have a weakness? ie. Low blocker.
Was it a one timer? Did the pass come from behind the net forcing the goalie to move out? Or was it across, forcing him to move laterally?
Was it on a 2 on 1?
Was it a breakaway? Was he alone in tight? Was he clear cut from center or was the backchecking dman breathing down his neck?

The more we break down each individual shot and put it into these types of categories, the more we can identify what are high and low percentage shots, which will better define shot quality over a simple location chart. From there we can determine which players are taking high and low percentage shots, and how many of each. We could then identify factors such as which players are prone to waste an offensive zone possession because he fires off too many low percentage shots.

My suggestion is that the NHL hires a few people to watch each game and track all of the types of factors listed above. If you really want to get advanced with your stats, review tape of all of one individual players' shots and start dividing it up into these sub categories. More data could be accumulated league wide from each individual player and maybe we'll find out interesting facts, like for example that "63% of shots go in on 2 on 1's down low when the shooter goes short side, and that number becomes 58% when he goes back the other way..." The less general and more specific we get, the more reliable our stats become.

Hopefully I made a compelling enough argument here and if you are reading this you agree. If you think stats are perfect right now, the way they are, and that you are smarter than hockey people like Kyle Dubas, then good luck to you.  if you are trying to make an argument in the hockey fan community, and you throw out shooting % as your entire argument, without acknowledging these factors (and more), your argument loses all integrity. If you are basing your entire opinion of a hockey player on the data available to us right now, your opinion is only as valid as the limited data we currently have.

I believe hockey stats will evolve and begin to incorporate these types of factors, and in time we will be able to go to or wherever it may be, and use the vast amount of data to form a solid argument, but as of right now it's just not there. I sincerely hope you found this interesting and informative. Criticism is always welcomed. I don't claim to have all the answers, but hey - neither do you!

No comments: