By E. Martin Nolan
The debate over advanced stats in hockey is by now well-established, but it is also still young. The usual positions have been staked out by the usual suspects, with the old school predictably favouring their own intuition and experience while the outsider stats nerds lead a slow insurgency based on numbers and cold method. What we essentially have here, then, is a debate between subjectivity and objectivity. Don Cherry knows a player’s value by what he sees on the ice. He sees the guy block shots. He doesn’t need a complicated calculation to tell him what he sees with his own two eyes, which he can trust because they’re honed by his longstanding and intense interest in the game. The stats people, though, don’t trust their eyes, and they don’t trust Cherry’s. Instead, like Charles Sanders Peirce before them, they rely on a more perfect method, one that is disinterested, separated from human imperfection.
This opposition, between the subjective and the objective, makes the debate over advanced stats essentially a debate between science and mysticism. Mysticism isn’t a much-used word these days, but what is a creationist if not a mystical thinker indebted to a belief in a world beyond the physical realm? But you don’t have to reference an evangelical sermon or a new age microwavable Buddhist self-help book for evidence that science has failed to wipe mysticism off the map. Science cannot eliminate mysticism because it is, by definition, limited. Human beings, not to mention nature and the universe, are incredibly complex and mysterious, making it impossible to reduce everything into testable chunks. So, while we may trust science, we rely on its objectivity only to a certain extent because there are certain movements, certain transitions—both profound and everyday—we experience that simply cannot be tested or quantified.
Likewise, it would seem likely that as the objective perspective on hockey continues to evolve, we will slowly accept its importance, along with its limitations. In fact, this is already happening, but not without major disagreement, even amongst those who profess faith in the objective view. The Web is rife with arguments over Corsi, Neilson, Quality of Competition, Zone Starts, etc. As was clear in the coverage of MIT’s Sloan Sports Analytics Conference, the hockey world is far from settled on what hockey analytics are, let alone whether or not they’re useful. Therein might lie the best argument against the objective view: even if you trust the objective, scientific understanding of the game, how do you select the metrics used to measure it? The inability for the objective crowd to agree only makes their argument for the importance of advanced stats appear weaker.
Meanwhile, it seems to hold that the nature of hockey really does resist quantification. Like soccer, it has long been assumed that hockey is simply too complex to capture with numbers. Consider this: there are 12 players on the ice, all with the ability to move themselves in any direction, while moving a tiny puck that bounces like mad. It has been reported that there are approximately 200 possessions in a hockey game. Those possessions are of a widely variable nature, each subject to multiple decision making processes as well as to the randomness of bounces, misplays, deflections, etc.
Leaving aside the fact that the objective view of the game is way less fun than the subjective view, how could you ever hope to pin this game, with all its myriad potentials, down? Maybe you can’t, maybe you can only keep it from flying off your understanding, like a giant balloon in a Macy’s Day Parade. Or am I not giving the numbers a fair shake? Am I being an uncritical hockey mystic?
I don’t think so. Unlike baseball, which exists in a controlled environment practically begging to be studied using advanced stats, hockey is based in a far more unfixed environment, in which the subject is constantly transitioning. If baseball is like a pre-made laboratory for advanced statistics, hockey like the weather: predict away, but be ready to be wrong. It follows that if you are ready to be wrong, then you are also making more tentative, and less exact, assessments.
Take, for instance, “The Impact of Puck Possession and Location on Ice Hockey Strategy,” a 2006 paper written by Dr. Andrew C. Thomas. Thomas hypothesizes that there are 13 “states” in which a hockey game might exist at any given moment. The purpose of the paper is to determine the likelihood of a goal being scored given the state—or situation—of a game at a particular juncture, but what strikes me is the vagueness of the analysis.
Thomas finds that within five seconds of a transition (meaning play has moved from one state to another), “possession of the puck in the offensive zone is necessary to score goals.” However, if you extend the time frame to 20 seconds, “the expected number of transitions grows to such an extent that scoring a goal beginning in any state is now feasible,” while at 40 seconds, “the scoring rates beginning in each state are nearly identical.” So, give a team 20 seconds, no matter their starting state—meaning they can be in possession or without possession, anywhere on the ice—and they can reasonably be expected to score; give them 40 seconds, and anything can happen. It is no revelation that if you want to score in 5 seconds, you should have the puck in the other team’s zone, and considering the larger time-intervals, the findings are inconclusive. The paper is more conclusive in regard to regrouping vs. dumping it in, dump-and-chase vs. carrying it in, and clearing the zone vs. passing it out, but even there the numbers aren’t all that convincing.
It seems unlikely that advanced stats will ever penetrate the hockey world as it has, and will, in other sports (aside, probably, from soccer). Why? Largely because, as Thomas points out, “in games like ice hockey offence and defence become most difficult to separate, as control of the play is often difficult to determine.” In other words, possession in hockey is fluid and murky (as it is in soccer). So while time of possession is surely a valuable stat to track, it must be noted that a permanent potential for transition persists between and within all hockey possessions, making the value of each ‘possession’ dependent on the widely variable nature of any single possession[i] [footnote]. If the real value of a possession is virtually unknowable, tracking possession can only be marginally helpful to a team, or to the fan attempting to gain a deeper understanding of the game, and in particular the game’s flow.
With that we arrive at a what may be called an educated hunch: hockey is defined by its transitional nature, making it a difficult game to quantify simply because it won’t sit still long enough for you to do so. Sure, you can use your Corsi numbers and the like, but those do little do describe the actual movements of the game, which, like weather, are easily witnessed, but not so easily predicted, due, in both cases, to the high level of variables that might effect the outcome of a given event. That means that, if predictability is certainty, then hockey is a game of uncertain certainties: you are sure what is happening, you are not shocked, but you also could hardly have known it would unfold in exactly this manner. I’ll get into the metaphysical and philosophical ramifications of hockey’s uncertain but ordered nature in my next post. For now, I’ll conclude that barring some unlikely future breakthrough, it seems we’re stuck understanding the flow of a hockey game using those most ancient technologies: the eye, experience and intuition. What do you know, Don Cherry might’ve been right.
[i] The unpredictable nature of a hockey possession led Thomas to use 13 states in his study, as opposed to the four states used in a study cited by Thomas to describe football.
One thought on “Hockey and the measurement of permanent transition: Or, how hockey is like the weather”
Pingback: Hockey simulates the modern world: A speculation « Hockey in Society