The Leafs Collapse And Hockey Analytics

Happy Leafs Elimination Day! This year’s edition is especially entertaining, and of interest beyond NHL fans, because it represents another disaster for old-school media troofers who, as part of their identity, feel it necessary to attack even the most obvious insights derived from analytic methods.

Sean McIndoe’s Grantland piece at the beginning of the season did an excellent job laying it out. The core of the contemporary analytic view of hockey is an insight similar to Voros McCracken’s transformative insight about pitching. McCracken, and those that applied his findings, found that a pitcher’s strikeout rates and HR rates were better predictors of his ERA going forward than ERA itself, because what happens to balls in play is mostly beyond a pitcher’s control (essentially, it’s a combination of luck and defense.) Similarly, in hockey analysts found that a teams ability to maintain the possession of the puck predicts goal differential better than goal differential itself. In part, this is because shooting percentage isn’t really a skill, but is a combination of luck and goaltending. If you keep getting outshot when the game is on the line, unless you have Dominik Hasek in his prime or something you’re likely to lose in the long run.

One thing worth emphasizing is that the statistics that analysts use as a proxy for possession aren’t complicated or esoteric. Fenwick looks at unblocked shots for and against; Corsi includes blocked shots. The differentials, however, are most meaningful even strength when the games are close. When you see someone discuss a team’s Fenwick or Corsi close” this reflects the fact that the stats are only meaningful if you eliminate the garbage time. because when the game isn’t close it affects the shot patterns. (Just as, when analyzing football, you can’t put a lot of weight on yards gained by an offense in the fourth quarter of a blowout; it’s in the interest of the defense to trade yardage for time, so accumulating yards doesn’t prove much about your ability.)

The Leafs were a key test case because they made the playoffs last year (and made it to Game 7 of the first round) but were a terrible possession team. (I’ll present it ordered by Fenwick close because it was developed by a Flames fan, but as you can see the difference between the two are marginal.) The verdict of the analysts was clear: if the Leafs didn’t make actual improvements they were going to regress substantially. Adding to fuel to the controversy is that 1)the Leafs organization is as notably hostile to analytics as their media sycophants, and 2)this was reflected in their widely-mocked offseason moves.

For much of the year, there was a lot of crowing from the Murray Chass equivalents in hockey, since the Leafs seemed safely in a playoff spot. And this wasn’t because they actually improved. Their Fenwick close is 29th in the league: Worse than the disastrously rebuilding Oilers, substantially worse than the Flames, a poor organization in year one of a rebuild. Worse than everyone but a Sabres team with a historically bad offense. Their Corsi is also 29th. Did this prove analytics wrong? Nope — the Leafs’ success, such as it was, was built on exceptional goaltending and luck in the skills competition that the NHL idiotically uses to award points in regular season games. Their massive collapse at the end of the year, taking them all the way out of the playoffs, is what happens when the luck runs out.

So remember this when you hear that the Avalanche – a young team in the playoffs ahead of schedule despite terrible possession numbers — are defying the NERDS because of Patrick Roy’s coaching wizardry or whatever. Either they’ll actually improve, or the odds are overwhelming that they’ll be picking in the lottery again. But I, for one, hope the Leafs never change…