Wikipedia.org

Wikipedia.org

Over the past few years, the debate has grown more intense about the validity and reliance on #fancystats. The concept of quantifying the game has been a theme we have run with around here, albeit with the conceit that there is no perfect, all-knowing stat that can be universally relied upon to demonstrate a player’s ability level.

Statistics trying to quantify human athletic performance are inherently limited. There are very human characteristics in play; such as intelligence, judgment, emotion, situational awareness, etc. It makes it difficult to measure performance as if they were vital signs. I think that to fully expect that level of quantification or to vilify the statistic for being unable to is missing the point.

Much like politics, I think the emergence of these statistics and the resistance to adoption has pushed the two positions out to the extremes. The old school hockey community has written them off or marginalized their effectiveness, citing “games are played on the ice, not on a spreadsheet”, or taking pot shots at the Maple Leafs for hiring Kyle Dubas for their Assistant GM position, and various stats writers to make up a new analytics department.

On the other hand, the #fancystats believers have become more hardened to the holy grail of Corsi and other possession stats, ignoring their flaws. I can completely understand this effect, trying to combat willful ignorance with advanced science, but it has, to an extent, driven a stake through the hockey intelligence community.

That was kind of a long-winded introduction to the topic that I actually want to discuss this morning, and that is conceptual improvements to goaltending analytics. The topic of improving and quantifying the traditional possession stats is a whole different discussion (quantifying the counter attack, turnover locations, systems-based possession, etc.), but a ripe ground for debate and development.

Goaltending analytics is a different can of worms, but I feel like conceptually easier to quantify. There have been some very interesting developments thus far this season, but I don’t think we’ve had that true step forward yet. Now, I am by no means a statistician (I lament the presence of a calculator on my desk every day), but I think there are conceptual measures that someone much smarter (and more mathematically inclined) than myself could put to good use in developing new goalie statistics.

Some of the new mechanisms are heat mapping shot location clusters and shot distance statistics. The heat maps are very cool, the red portions of the map show hot spots of locations from which goals are scored. The green portions are a higher proportion of saves from those locations. They are pretty and informative, but there is no other information taken into account. No shot speed, shot type, presence of coverage, nothing.

The other new advance is looking at goalie save percentage from various distance ranges. Dave was kind enough to share an interesting article looking at this from Canucks Army, which broke down shooting distances from four different zones: 0-10 ft, 10-20ft, 20-30ft and 30+ ft. As you would imagine, save percentages increase dramatically as the distance is greater, and the focus on the analysis is proficiency of each goaltender at the closer ranges.

I feel like location based information is a foundational layer of a much bigger picture. We still need to quantify shot-type at a bear minimum and figure out a way to assign linear weights to each one. I have no idea how to do that, mind you, but it’s a factor that needs to be accounted for, at a bare minimum. Additionally, you need to factor in at least one preceding event (ideally more if we are able to down the road). Was the shot generated from a carry? A pass? A rebound or other deflection? Knowing how much the goaltender had to range laterally or change direction quickly can dramatically increase or reduce save difficulty in those situations.

The other major factor I would like to see quantified would be positional efficiency. The only real methodology I can think of would involve measuring based on the puck as an fulcrum, how efficiently positioned is the goalie related to the puck during any given event. The goaltender’s chest could be the basis for the positional measurement. Since goalie’s don’t cover the physical distance during a period of play that a forward or defender does, it shouldn’t be that difficult to track. I should note, that while positional efficiency doesn’t tell you whether or not the goalie will save a shot, it does answer the question of whether or not he gave himself the maximum available surface area to begin with. Combining this information with shot type and the preceding event could shed some light on how difficult the shot was.

This could be unfairly biased to some, more athletic tenders who don’t have as strong positional skills, but manage to have a similar degree of success by virtue of instinct, strength and athleticism. Think of them as pitchers whose ERA consistently out performs their FIP (for you non-baseball folks out there, see here for an explanation).

While not a perfect statistical model, there are many logical extensions to add to the formula as the model becomes more advanced. I do however, think that being able to quantify the concepts of shot distance, shot type, preceding event and positional efficiency could form the basis for a WAR-like stat that could have great quantitative value for goaltenders. Either way, this type of stat has been sorely lacking for goaltender analysis. Probably why the Leafs’ new analytics department decided to keep James Reimer around.

Share: 

More About: