The Colorado Avalanche and the Assumptions of Hockey Analytics

Photo by Kerri Polizzi

The Avs accomplished a remarkable feat by going from the bottom of the Western Conference a year ago to the top of the Central Division with a whopping 112 points in 2013/14. The amount of change between those two seasons – behind the bench, in the front office, and in the dressing room – was massive, and confounds most simple attempts to point to a reason for the rapid change in fortune. 

Early in the season I said that they were building on the Blackhawks model based on how they used their speed in defensive transition. It looked like a wholly new team with only laundry to connect them to their recent failures, and Matt Duchene agreed, noting that team was finally playing a style that fit their personnel. Apparently others saw it differently as the narrative about their success has been organized around two factors: an amazing season by Semyon Varlamov and pure luck.

On Friday, well-known pessimist Ryan Lambert wrote about the expectations of the Avs to regress. ‘Varlamaov can’t repeat, and their percentages can’t either’ goes the reasoning. They were not a strong possession team (26th in 5 on 5 corsi %), and when you give up a lot of shots and your goalie gets worse, the end of that algebra equation is obvious. We can throw circumstantial evidence around that should make us uncomfortable with such a firm position on which way the wind is blowing – Montreal was 25th in corsi %, Anaheim 17th. Toronto was dead last and for all the obscene failures of that club, they still scored the 14th most even strength goals of any team.

Ultimately, however, the failure of using a regression analysis here is more fundamental. 

We found a way to quantify one way of winning hockey games, and teams are able to use that data to make efficient and insightful use of their limited funds when building a possession style club. But since when was there only one way to win a hockey game?

In Ryan Lambert’s column, he predicts the Avs overachieved by 22 points in the standings – making them a 90 point group of spring golfers. He does so by giving them an average save and shooting % and looking at a difference in scoring differential. He lumps their special teams and even strength numbers together and we get a huge drop in goals for and a huge spike in goals against. 

Of course, every analyst knows to separate out our even strength data. And when we apply the same analysis to just their 5 on 5 play, the difference is more slight. Had we played the same season 1,000 times, it’s likely the Avs would have managed closer to 100 points. They tied for 8th in powerplay shooting percentage, and tied for 7th in PK save percentage with Buffalo and just behind Edmonton. Not so outrageous, given they have a Selke finalist and the runaway winner of the Calder on the club (not to mention Matt Duchene and his 70 points in 71 games). 

That’s not really the point, though. Had we played the same season 1,000 times, perhaps Buffalo wouldn’t have scored an historically low goal total, and Vancouver wouldn’t have posted a sub-200 goal season for the second time in franchise history, and maybe Florida wouldn’t have had a 7.2% shooting rate on the powerplay (worst in the league by a large margin). 

The problem with assuming any team should get an average rate of goals per shot is that the NHL is not a perfectly equal league. Buffalo, Florida, Edmonton, Vancouver, Calgary… these were legitimately poor clubs this year. They each had some bad luck as well, but no one is picking them for a major step forward next year. 

The critique of the Avalanche as a favourite to regress rests on the subjective belief that they are not deserving of higher than average percentages. After all, Boston had a higher save percentage. The Blues and Bruins both had a shooting rate just 0.2% below that of Colorado. Is every playoff team going to see their percentages regress? No, just the ones who don’t also possess the puck a lot, the thinking goes.

We know one way to build a team that is effective – the Kings, Bruins, and Blues are great at it. Be big, violent, and control the boards. We know that leads to good possession results. We also know of a player type – whether Backes or Bergeron or Kopitar – that succeeds at possession. Smart, two-way players who manage the centre of the ice and have the vision to make plays from the turnovers they create. We’ve found ways to measure those players, and it led to a revolution in what makes a ‘star.’

We still don’t know how much skill matters in determining things like shooting percentage. We’re woefully unprepared to answer questions about shot quality, relying on coaches and players to adapt to any given defence and thus normalize our data set over a season or so. We haven’t found a stable, objective way to quantify the ability to create offence. We haven’t figured out how to map the geography of an ice hockey game.

The very brilliance of hockey is that you can solve all problems many different ways. We just don’t yet have the data to explain how to optimize those other methods. If a team wants to build a possession club, they can point to all the data analysts have generated about that. If a team wants to build a skill club with speed and creativity, they’re doing it the old fashioned way. That doesn’t mean one is better than the other, it means we can’t evaluate the efficiency of those clubs as well. We don’t have the tools to provide a quantitative reason Toronto failed and Colorado succeeded. All we have conjecture and subjectivity. But it doesn’t mean the difference was luck (at least not necessarily).

The Colorado Avalanche don’t have the roster they want. We’ll see them add another defender this off-season, Tanguay may be healthy enough to rejoin them, and the coaches will hope for and expect steps forward from the various young players that form the team’s core. No one from Colorado stood up and said they were a perfect team and infallible. 

But that doesn’t mean they were purely lucky. Their powerplay moves the puck extremely well, and in a dangerous area beside the net. O’Reilly, Duchene, Stastny, Erik Johnson, Varlamov, Landeskog, and now MacKinnon have all be considered stars or prospective stars are various times in their careers. Their extremely innovative and odd-ball coach took known research on the effectiveness of pulling the goalie early and used it to capture important points with late goals. That they didn’t out-shoot their opponents may not be the most insightful way to understand them as a hockey club.

I find Kings/Bruins/Blues hockey extremely boring and awful to watch. It’s violent and slow and it relies on all sorts of rules-bending and boundary-pushing about interference. It’s frustrating to play against the style of hockey, but it’s equally frustrating to watch it happen, knowing how beautiful the game can be under different circumstances. 

As a fan, I don’t want to encourage the NHL to give jobs to the Kyle Clifford’s of the world over the David Desharnais’s. It’s incumbent upon us as fans and bloggers to create analysis that leads to a game we actually want to watch. It’s also important to think about hockey as a game with many solutions and more variables than can be captured in any single data point. Let’s embrace the creativity of the game and do our best to reflect that in our data.