It’s far from a new thing, but there’s been a pretty heavy focus so far this season on the difference, or at least perceived difference, between public and private analytics.
The Chicago Blackhawks have been the biggest, most public case study, where Blackhawks coach Jeff Blashill hasn’t been shy about pointing out that according to his team’s metrics, the Blackhawks aren’t nearly as bad as they actually are.
It’s become a running bit in hockey coverage over the past couple weeks, and my pal John Matisz and I had a good chat about the whole situation last week when we guest hosted the PDOCast for our pal Dimitri Filipovic.
The Blackhawks bit actually took a back seat to the internal/external analytics drama when the New Jersey Devils decided to take it a bit further when they announced that Connor Hellebuyck wouldn’t start in his return to Winnipeg on Tuesday.
And it’s not the decision to not start Hellebuyck that raised eyebrows, but rather that it was phrased this way:
“Connor Hellebuyck was adamant that he felt a duty to play on Tuesday in Winnipeg,” Devils GM Sunny Mehta said in a statement. “However, in our estimation, playing in this game would potentially give us a reduction in our expected win probability.”
Hellebuyck has played three games in his past six nights and the Devils are 5-0 in his starts. Devils coach Sheldon Keefe alluded to potential overuse in his statements, and that his team doesn’t really care about the meaning of a game sentimentally with intra-division games coming up.
“From our perspective, when you zoom out, this is a goaltender that is not going to play every single game of the season. You’re always looking for times to give him a night off. The schedule falling the way that it does, it just made sense.”
I have several problems with how this has played out, and none of them have to do with the decision to start or not start Hellebuyck in Manitoba.
First of all, the messages don’t track.
Mehta is talking about an individual game, the expected win probability on Tuesday, according to the Devils GM, goes down if Hellebuyck is in goal. The head coach is calling it more of a “this game isn’t as important,” because New Jersey happens to play more-important, in-division games later in the week, where, in theory the No. 1 goalie will start.
If we are going to use an analyst decision to make the call, that’s cool, but let’s at least be honest about it or get on the same page about the lie we are presenting.
Second of all, and maybe even more important, I hate what this phrasing and decision does to an analytics conversation that for the most part has improved in recent years.
The Carolina Hurricanes won the Stanley Cup because they married the ideologies of both their more traditional and analytic-minded brain trusts. Rod Brind'Amour and Eric Tulsky work in tandem, they push each other, and both men have altered their views of the game based on data they’ve taken from each other.
Brind’Amour wants his team to play a certain way, he values certain old-school mentalities, but has also embraced the analytic revolution that Tulsky represents. At the same time, Tulsky has done a very nice job accounting for traditional hockey mantras when taking a deeper look and making more data-driven decisions.
The Devils here are doing a bit of the opposite, where through Mehta’s statement are intentionally playing more into the old black-and-white nature of the analytics, where the data says something, so therefore it must be right and if you don’t understand that, well, you must be stuck in the past.
The problem with that is data doesn’t actually say anything. It’s data, it’s a non-living collection of numbers and graphs. Analysts, the human beings who look at the data, actually say things, they make decisions off those numbers, and when simple computer models are used to defend a decision, it’s a shield for the people who actually made the choice on a matter.
It’s also pushed the whole drama of the situation out even further, and in my view adds even more distractions. I’m writing about it today, so are many others in the hockey world, and no the underlying narrative around Hellebuyck is only going to continue that he’s a coward afraid to play in Winnipeg.
As far as I know, there aren’t statistical models that take that into account, which is where this lies more on the human beings looking at the numbers to have some common sense or at least properly own the decision as opposed to spitting out two differing statements.
Again, it doesn’t matter much to me if Hellebuyck starts or doesn’t tomorrow, I’m sure he’ll be swarmed by media members in Winnipeg, so the circus will happen whether he faces a shot or not. But can we at least treat this situation with some honesty about why he is or isn’t starting? Can we stop using analytics as an excuse and actual explain those decisions to better inform an audience and media members?
Just some food for thought. Enjoy your Monday.