Tl;dr: If wins come from goals, and goals come from shots, what comes next—and then what comes after that? Front office analytics and game analytics, shooting and passing, puck possession and puck digression. Far better primers to hockey analytics here and here, and widely accessible on Google.
It would be misguided to say that Twitter is a hub of nuance, but as far as sports communities (and niche sports communities) are concerned, Twitter isn’t a half-bad place to discuss things and be prompted to think. This tweet prompted me to think:
I’ve long been interested in hockey analytics, and my Twitter timeline has all the usual staples (Evolving Hockey, Micah Blake McCurdy, Shayna Goldman, Dom Luszczysyn, Corey Sznajder, many others). Hockey analytics are tough, because discrete events—think of a single play in football—are fairly nebulous. Possession changes are frequent, play is fluid, and people move fast.
The above tweet comes from Cam Charron, who worked in the Toronto Maple Leafs analytics department for eight years, and is now a contributor at The Athletic and contributes to the broader hockey analytics community. It’s part of his game tracking work, which involves hand-tracking Maple Leafs and Vancouver Canucks games. It’s that hand-tracking, and its notionally affiliated hand-tracking projects (such as the below, from Corey Sznajder) that got me thinking.
A brief digression
Let’s start by making a distinction between front office and game analytics. Front office analytics, such as Evolving Hockey’s Goals Above Replacement (think baseball WAR), deal with players on a macro level. They try to answer questions like “how much should we pay this player in free agency?” Game analytics, on the other hand, are a little more specific and deal with players on a more micro level. “What percentage of Player X’s zone entries were successful,” one might inquire in terms of game analytics (think points per possession on catch-and-shoot attempts in basketball).
A good way to decide whether a metric falls into the front office or game category is to ask how the metric is actionable. A coach wouldn’t say to a player, with no explanation, “you need to improve your GAR.” A coach might well, however, say to a player (undoubtedly with corroborating video evidence!), “your success rate carrying the puck into the offensive zone in these last few games, 14% hasn’t been as successful as last season, when you got in successfully at 25%. We know you have the ability to improve that, here are some things to work on.” (This is a very rough and simplistic example, but you get the point.) In front offices, however, a player’s pure zone entry success rate is unlikely to dictate team-building decisions, but can be used as part of a larger all-in-one metric, such as GAR, that is actionable for team-building decisions.
Of course, this is just a simple distinction for the time being. The ideal analytics process is cyclical and doesn’t totally separate front office and game analytics, but rather continues to explore how everything is connected. It’s hard to say exactly what any given NHL analytics department is tasked with, but I think it’s fairly safe to assume it deals with front office decision-making AND improving on-ice performance. It should, at any rate.
For this post’s purposes, I want to look at game analytics.
Why that tweet got me thinking
So we know that Cam’s not working on something completely foreign to what your favorite team’s quantitative department (which has more data than publicly available) is doing.
All discussions of NHL game analytics come back to goals, which are the end-all be-all of winning (inclination toward which is the goal of the analytics department, after all). The team that scores more goals, wins. So front office analytics, to put it overly broadly, are about bringing in the right players to score more goals than opponents, and game analytics are about helping those players score more goals than opponents.
That’s not especially rocket science, but what analytics is good at, more theoretically, is looking at the how and why of goals, or “scoring,” to be more universal. In football, as has been much publicized, you might recall that “the analytics” (when I say “the analytics” I mean a very, very broad-strokes melange that squashes all analytics into one, which you really shouldn’t do, but anyways) say that teams should pass more. This is not because “the analytics” are on the payroll of Big Pass, but because the best research available at the moment suggests that passing leads to scoring more points, which, as everyone but Matt Rhule seems to agree, leads to more winning.
In hockey, scoring goals comes from shooting the puck. So the more and better-quality shots you take, the more goals you are likely to score, and thus the more you are likely to win (again, oversimplifying). That preoccupation with shots spawned such “advanced metrics” you may have heard of, like Corsi and Fenwick (explained), which are largely concerned with shot quantity and the share of total shots your team takes, and so on. Also of note are “expected goals” stats (big in soccer analytics too), which essentially try to take a shot and tell you—based on various factors—how likely that shot is to be a goal. For comparative purposes, one might consider Corsi/Fenwick metrics to deal with shot quantity, and an expected goal metric (for any given shot) to deal with shot quality. How far away from goal was the shot taken? Is it a rebound shot? If you look back at a large sample of shots, you begin to see various factors that connect shots to goals. Shots taken closer to the net have a higher success rate. Shots with a good angle are better than shots taken from behind the net (in regions not named Michigan, at least). If the goalie doesn’t have to move to save a shot, the goalie has a better chance of saving it, and so on and so forth.
Thus we have ways to analyze and understand shots and their relation to goals. So, we can state the following: Goals are derived from shots, so to understand goals, we need to work to understand shots.
Now actually back to Cam’s tweet
“The analytics” are all about asking and trying to answer big questions, the next of which is as follows. If goals are derived from shots, what are shots derived from?
Cam’s table is an answer to exactly that question. Cam and Corey’s tweets are the very interesting part of what comes next in the goals > shots > progression. Shots, such a table suggests, come from stuff like zone exits and rush entries, cycles in the offensive zones, and so on and so forth (we’ll call them microstats, and say that they’re not perfect but generally help us make things less nebulous).
Microstats can help us understand shots. For example, you’re likely to get a better shot on a controlled zone entry than on a dump-in. These can lead to the more actionable parts of game analytics. If you have particular microstats, instead of telling a player “shoot the puck more,” a coach can say, “we’re scoring a lot of goals on shots taken after we cycle. So if we cycle the puck more, it’ll help us score more goals” (again, overly simplistic, but you get the picture).
Much valuable discussion is had about what Cam and Corey and others are tracking, and how that leads to shots and whatnot (read their work for more on that). I want to think about something else, and if for some reason you are still reading this, apparently you want to as well.
What I want to think about is this. What’s the next progression? We’ve said goals derive from shots, which derives from things like zone exits and rush chances and whatnot. What do zone exits and rush chances and whatnot derive from? What are the “quantum” stats, as it were, that beget microstats?
More specifically, I think this question is really appealing because it gets us closer to investigating the theoretical tactics and Xs and Os on which sports are built. (This has much to do with film review, which is beyond the scope of this post, but I will just say that the best quantitative analysts watch a ton of film and the best video analysts have a strong grasp of numbers.) If we say that zone exits and rush chances derive from something like team stylistic and schematic decisions, then we’re reaching the point where you can start to quantitatively analyze tactics and coaching. Should teams forecheck heavily (gegenpress, anyone?) or sit back in a neutral zone trap? That’s the sort of important, ultimately actionable question answerable through tracking (automated in the long run) at the level below microstats.
I’ve put such schematic decisions are the lowest level of the pyramid, so really the question has to do with where there exists a level between the microstats of zone entries/rush chances/whatnot and those schematics. I would venture that the so-called quantum realm has a lot to do with publicly-untapped player-and-puck tracking data, and I think passing, and more generally possession, is an interesting part of that link (I make no claim that it is everything, or even the key thing. I do not know; hopefully a more enlightened hockey analyst will have a better answer).
Passing itself is in a sense already tracked. We already consider assists and setup chances to be an interesting form of passing. Assists, by definition, lead to goals (primary assists, at any rate). It stands to reason that “assists” on things that lead to shots would also be interesting and valuable. Passing on a wider level than leading to shots tells you more about the interplay between players, which is of vital importance to front office and game analytics alike. Are there types of dump-ins from a given line that’s more effective than a controlled rush (and other such questions)? Thus the exploration of how passing affects possession, as noted analytics observer William Shakespeare said, is “a consummation devoutly to be wished.” Or, to paraphrase an icon slightly more hockey-related, “you miss 100% of the shots you don’t take, but you don’t have any shots to take if you don’t have possession.” Quantifying that “possession” could be the (really, a) next big frontier to be normalized.
By “passing,” I essentially mean a stand-in for puck possession (shot volume is currently, in certain cases, used as a rough proxy for possession) in the sense that it is varied types and degrees of possession—not necessarily passing—that produces such things as rush chances and zone entries and whatnot. Zone entries and cycles often contain implicit assumptions or conditions of passing. A zone entry is a manifestation of possession, and while it does not strictly allude to a pass, it likely derives from, or is enhanced by, a pass, or passes, of some kind (this is also what makes Connor McDavid rushing end-to-end alone so exciting). Understanding possession overall and how the fluidity of a game works can ultimately be very rewarding. If you can understand (or, frankly, better define) possession, you can understand microstats better, and thus shots and goals and wins follow.
That type of possession and how it relates to schematics is explored a little more easily when it comes to powerplays, where the fluidity of the game is slowed enough for tactics to have a direct influence. With a higher likelihood of scoring at a strength advantage than at even strength, shots become even more important, and how a team establishes itself schematically to take those shots becomes a very measurable and investigable process. If analytics can do for even strength gameplay—which constitutes much more of games than powerplays, especially when Connor McDavid is in the playoffs—what it continues to try and do for odd strengths, the actionable tactical breakthroughs are potentially immense.
So, wins come from goals, goals come from shots, shots come from zone entries and rush chances and whatnot, microstats maybe come from quantum stats (passing/possession, etc.), and microstats and quantum stats come from tactics and Xs and Os. Those, of course, have a great deal to do with the type of players (and coaches) teams add, and so we return to the beginning. Quantitative investigation, just like the subject of those investigations, is fluid, and hardly concrete. The cycle of studying the cycle continues.
Josiah Cohen 
Leave a comment