Thursday, October 28, 2010

Game 5: Deflection

It was sometime in the second period, and the Montreal was down 3-1. Milan Michalek was having a good night, the Habs’ power play was still offline, and Jaques Martin had sent Lars Eller to bed with no supper. In other words, it was yet another of the many, many totally unremarkable moments that fill the long and storied history of the Canadiens. Jeff Halpern and Maxim Lapierre are fussing around the Ottawa net, doing a respectable fourth-line approximation of offense, insofar as ‘offense’ is defined as ‘not being in your own zone’. Halpern is behind the net with the puck, and being a real life NHL hockey player and therefore knowing that this is not the most useful place to have the puck, he sensibly tries to toss it back out in front, although there are no particular teammates in that area at that time. What is in that area at that time is Chris Campoli, who is not paying close attention to the angle of his skate blade as the puck approaches, and… well, now we have ourselves a one-goal game.

Moments like that, it’s hard to give either credit or blame. You gotta stretch your logic like viili. You have to believe that Halpern saw, in some split second revelation of offensive precognition, that he could bank it blind off Campoli’s skate. Or you can blame Elliot in a fit of uncontrollable goalie-loathing because IT’S HIS JOB TO STOP ALL PUCKS AT ALL TIMES. Or, my personal favorite, you can just go suddenly, inexplicably vague: the Canadiens were in the offensive zone, so they were obviously doing something right. You can come up with rationales, if you want to pretend it makes sense, but you’re more likely to shrug, and call it luck

If you’re tallying, though, the system forces you to credit someone and blame someone. Halpern was the last Canadien to touch the puck before it went in the net, therefore, he scored a goal. Elliot was playing net when the puck went in, therefore failed to make a save. There is luck in hockey, but there is no luck on the game sheet, no luck in the event log. To those who count countables, every event is caused by someone. Somewhere down Quebec way, Olivier is counting the scoring chances for this game. I like to imagine he does so with grid-ruled Rhodia notebook, a giant click pen featuring at least five different colors, and some kind of disreputable facial hair. While the NHL counters are busily assigning credit for this strange goal, Olivier's multicolored pen hasn't even been clicked. Because there is no reasonable universe where no-look backhands from behind the net constitute dangerous offensive plays.

The Scoring Chance Initiative, which is what it would be called if it were a thing so orderly as to have a name, is the next big thing in hockey analysis. Of all of the things that have ever been hyped as CHANGING THE WAY YOU SEE THE GAME, this is the one that might actually do it. This is the thing that might, a few years down the road, become the indispensible counting number. As it is, it’s still a bit of a guys-in-basements-with-blogs thing, but that’s a step up from where it was a few years ago, when it was an assistant-assistant-coach-with-a-notepad thing that you were never going to hear one word about.

What makes the scoring chance such a brilliant countable is that it stands at the intersection point between the objective and the subjective. It takes the expertise of knowledgeable, attentive hockey-watchers and translates it into data that can be processed. Unlike other numbers in a game- shots, Corsi, hits- it doesn’t treat all similar actions as equally significant. A chance is only credited if the observer believes the attackers had a high chance of scoring on a play, filtering out the white noise of weak, inaccurate, desperate shots. And the counting, apparently, works pretty well. Yes, various counters have different standards for a chance- some will count a breakaway even if the shooter fans the shot, for example, while others won’t- but generally speaking, different chance-counters watching the same game will independently come up with similar numbers, usually within two or three of each other. Definitions are difficult to legalize, but people who know hockey know a chance when they see one.

“But,” says the crotchety old-fashioned straw-hockey-fan who lives in my imagination, who looks sort of like Red Fischer fused with the fatter of those two guys who used to heckle the Muppets, “Why count something new at all? The games are won and lost on goals, not chances. If it doesn’t go in, it doesn’t matter.” Oh, straw-hockey-fan, you are ridiculous, but thank you for providing me the opportunity to move on to my next point by reintroducing a core paradox of the game as if it was something new: to understand hockey offense, you have to filter out not just the least successful attempts (i.e. shitty shots) but also the most successful non-attempts. You have to ignore goals.

Hockey is a chronically low-scoring game. Goals are tough to come by, but ironically that makes them highly susceptible to luck. At the level the game is played at now, it’s extremely difficult to beat another team on raw skill, to outplay in the pure sense. More often, you play a little bit better for a few seconds, and then get lucky, or vice-versa. Generally, the luck is a few gestures removed from the goal itself, but without a bit of it somewhere, very few goals would be scored at all. To track superior play, you need to ignore the result and focus on the few seconds of playing-better that surround it. Those seconds are the currency of good teams, and over time, they do result in more goals. By treating scoring chances with goals and scoring chances without goals as equal, the hope is to get a clearer idea of who drives offense (or compromises defense), and a clearer view of the real balance of power in a game. To filter out the luck and see only the skill.

It’s not perfect, of course, but what is?

A peculiarity of the system, however, is that it completely ignores non-chance goals, as if they didn't even happen. Because they are nothing more than somebody’s luck or unluck, they are exactly the sort of thing the counters hope to filter out. They will ignore all the torturous viili-arguments contending that all goals represent something done well, but you can do a dozen right things a minute and not be close to a scoring chance, and if one of those right things happens to ricochet off a defenseman’s face and into the net... it still wasn’t a chance. It was just luck.

Including the unscoring chance goals as if they were chances wouldn't really fuck the results, because those sort of goals tend to even out over time, and over the course of the season there probably aren’t enough of them to drastically alter the profile of a given team or player’s performance. Presuming they are as randomly and more or less evenly strewn as they seem to be, they favor no one.

But the aggregate of all those lucky goals would be something considerable (by which I mean something that would support and possibly reward consideration). There needs to be a separate column for the flukes and the deflections and the weird bounces, a whole category of uncredited goals. It would be illuminating to track luck itself as though it were a player, and see what share of league-wide offense it contributes. One goal in a hundred? One in fifty? One in ten? How clutch is luck? How often does it get the game-winner? And how random is the distribution anyway? Does it hit all players equivalently, regardless of team, caste, or game state? Or does it have favorite haunts? Does it have habits?

Once upon a time, a few posts (and therefore a few months) back, I said the single most difficult and significant problem facing any further study of the game is the cartography of luck. I still believe that to be true (more every day, actually), but it’s not an especially helpful thing to say. It’s far easier to look at a negative space, what remains after you have understood the understandable things, and call that luck, rather than to take luck itself as an object.

But, if one was going to try, a good place to start would be chanceless goals.


Olivier said...

Oooooh, me being cited on AToI. I am famous!


On the cartography of luck:

Nowadays, I'm stuck with a decision I made when I started that project last season: a goal is almost automagically a scoring chance. A goal really have to be scored the Halpern way against Ottawa not to be recorded as a chance.

Hal Gill slings a soft wrister from the blueline and, seeing-eye, the pucks ends in the net? A chance. Halpern backhands a beachball from the left wing wall that deflects on a defender's knee? A chance.

I'm not terribly comfortable with this way of doing things, but others are doing so too, plus that's how I scored stuff for the last 110 games, and I don't feel like fucking up with the consistency of the dataset. So there.

But, knowing this, and knowing a given team's shooting % over a season is mostly the same as most of the other teams in a given season, one could flush out a whole bunch of luck by simply removing goals from my dataset. Sure, the percentages will skew a bit, but Gionta, Gomez, Pleks and Cammalleri will still end up atop of the pile.


- I'm in Montréal, most of the time
- I use lined notebooks. Cheap ones so I can spill salsa over them without crying. No uses for grids.
- No click pen with different colors. I always ue blue ink, mostly Pilot UniBall and Pilot Hi-Tecpoint 0.5. I am oggling those fancy gell-ink pilot pens. They look nice.
- I do have disreputable facial hairs.
- I never work from the basement. Bad wireless reception.

You got me thinking about adding a note for those false chances I keep recording. They would enlarge the "chanceless goal" sample and, over 70 games, could give us an interesting insight.

saskhab said...

I don't record chances like Olivier, but I can't see how a breakaway could not be deemed a scoring chance even if the guy spontaneously combusted before getting a shot off. That's an important event regardless of making the shot. To me, any breakaway or two on one should count. They are dangerous. 3 out of 10 breakaways result in a goal (in common parlance). That doesn't mean a goalie makes a save 7 out of 10 times, it means the puck doesn't go in 7 out of 10 times. Actual shooting percentage on breakaways is higher than 30%, probably above 40%.

A 3 on 2 is when you start to get picky about the shot.

I've always thought some goals shouldn't be classified as shots on goal. And I used to play goal.

Julian said...

My idea of a scoring chance is any situation or action that makes my heart jump when it occurs against my team. After the thousands of hours of hockey watching I've done in my life, I have a decent idea what's worth worrying about and what isn't.

I agree with saskhab (Bruce, right?) that a breakaway or 2-1 is almost automatically a scoring chance, barring something crazy like going offsides. And even then... I figure it's a situation the defending never is never happy with, so someone messed up somewhere and the team deserves to be punished by getting a minus.

That said... I wonder if you could start awarding scoring chances on a basis of $.25 chance, $.50 chance, $.75 and full dollar chance. Then you just add them up and the end of the game, Team A outchanced Team B 8.5 to 5.5. Or something like that, to account for the disparity of quality of chances. A breakaway or well-established 2-1 or fanned on tap-in from two feet out are full dollar chances, a wrister through a screen from the top of the circle is a quarter. An unscreened slapshot from the point isn't a scoring chance at all.

Anyway, nothing much further to add on the luck part, except that my guess would be about 10-15% of goals would be non-scoring chance goals, by my definition. The trickiest ones are the point slapshots into a screen. When my team is defending against them, they don't really cause my heart to race enough for me to consider them chances, but I would consider them if I were to use a 1-4 rating system. On a binary system though, I wouldn't count them.

E said...

olivier, this blog is a looooong way from the big time, but for what it's worth, i have tremendous respect for what you're doing. the scoring chance thing is pretty thankless work for any team, but it strikes me that doing it for the habs, about whom only seven or eight rational thoughts have ever been thunk, must add a creeping sense of existential despair from time to time. and you know, i sort of thought you were counting unchances as chances, but then i didn't see the halpern goal on your list, so i revised accordingly. anyway, if it's not too much extra work, a notation for chanceless goals would be a super-cool starting point, and if i was still in montreal i would buy you a cake from the cute cake girl with the shop near parc and mont-royal.

as to defining scoring chances... the more you try to draw it out, the messier it gets. it's got as much to do with the positions of the defenders as the movements of the attackers, so something can easily be a chance one moment when a similar thing in another moment is not. there's an element of irreducible subjectivity where you just have to trust the counter.

that said, i think where it gets interesting is the point when, down the line, somebody tries to shift from using scoring chances descriptively to using them prescriptively. it's one thing to use them to analyze the game-that-was, but if it gets to the point where they're supposed to inform coaching tactics, shit's gonna hafta get a little more precise. those point shots may never be chances, but that doesn't make them not worth the taking, from time to time... or does it?

doug schlenker said...

i've thought about this concept of scoring chances before.. or another statistic of "shots AT goal" since someone it seems silly that a d-man that takes a slapshot, misses the net by half an inch doesn't register as a shot on goal.

this is the first time I've read about it in detail though and I have to say it's brilliant. this would over time be *the* killer statistic when comparing one team to another from every point of view. It's really a complete team statistic. If the defensemen are doing their job, scoring chances against their team should be consistently lower. If the forwards and defensemen are offensively strong, their scoring chances should be higher.

Tom Benjamin said...

I think scoring chances is an okay metric, but I don't think it is anything near a holy grail. There is no good metric for hockey and there never will be because the essence of the game is the fight to control the ice.

The team that does the best job of denying open ice on defense and finding or creating open ice on offense will usually (but not always) get the most chances. The team that gets the most chances will usually (but not always) score the most goals. Unfortunately, controlling the ice is an unmeasurable team activity.

I keep a rough count of chances when I watch a game, but it is often misleading and not just because of chanceless goals. Chances come in clusters. Gionta rips a shot off the post. The rebound goes right to Gomez who shoots. Great save. Another rebound. Goal!

How many chances? Three? If Gionta scores on the initial shot there's only one chance. A team can get several chances on a single power play before they score. If they are good enough to score on the first chance, they don't get more than one.

Team A can carry the play, get the most distinct chances and score the most goals, while team B is behind the play most of the night, gets fewer distinct chances, but gets the most actual chances because they did not score with a flurry of shots.

E said...

doug- yeah, it's interesting stuff, but it's still a bit spread out over the interwebs. i know they were hoping to get somebody for every nhl team this year, but not sure how that went. behind the net has a link on the sidebar to all the various counters. provokes some interesting discussions as well.

tom- "there is no good metric for hockey". hmmm... i'm not persuaded of that yet. i would agree that there's no perfect metric for hockey, in that any metric needs to be applied in an informed, thoughtful way. there's never going to be a chart, anywhere, that represents hockey exactly enough to do your thinking for you. however, i do think there are such things as useful metrics for hockey, numbers that can pull out dynamics our eyes miss, or that we will simply never see at all because no one can watch every game every year with perfect attention. the scoring chance thing can do both sometimes, but especially the latter- it can give you a much better picture of a game you can't see than any other mechanism i've come across. it seems to me that the times the chance-count gives an honest representation of the flow of a game outnumber the ones where it fails to do so (as in the case you present). but, since we're still more or less in the raw data stage of the project, we'll have to wait to see what comes of it once someone starts crunching the numbers into conclusions.

Tom Benjamin said...

it seems to me that the times the chance-count gives an honest representation of the flow of a game outnumber the ones where it fails to do so (as in the case you present).

No doubt, but the same thing is true for goals. What chances can do is the same thing shots or the Corsi number can do - it can indicate to us whether the score was a just result, or whether the hockey gods were funning with us.

but, since we're still more or less in the raw data stage of the project, we'll have to wait to see what comes of it once someone starts crunching the numbers into conclusions.

Fair enough although I don't really understand what specifically we want to learn. My prediction is that these numbers will have the same problems all of the other numbers do. It is an outcome and randomness will rule outcomes. It will just delay the question for a click. "What activities lead to a goal?" is really no different than "What activities lead to a scoring chance?" Those activities, I submit, are tangible, but they are not measurable.

It is the flow - and who controls it - that matters. That is the challenge. That's what coaches try to accomplish. If the team controls the ice surface, they will get chances (goals) and opponents will not. Usually, control passes back and forth and nobody really understands why or how momentum can shift. But the basic objective of every team must be to cover the ice surface, to be the first to the loose pucks, to apply pressure on the puck all over the ice. That's how scoring chances are prevented and generated.