Which New Jersey Devils goalie was the worst this season?

Follow @InfernalAccess on Twitter

Be sure to join the Discord channel to talk hockey, and everything else, with our writers and subscribers.

---

By CJ Turtoro (@CJTDevil)

Sometime during the Sabres game on Thursday – I can’t remember if it was the first shorthanded goal from the neutral zone, the second shorthanded goal that would’ve sailed wide if Andrew Hammond hadn’t kicked it in, or the fourth goal where a whiffed wrister from a rookie defender slid sluggishly through his five hole – I tweeted this:

So, like the good self-flagellating Devils fan I am, I decided to dig into this a little deeper.

Like most questions you ask about player-assessment, this depends on the metrics you associate with the definitions of the words.

Before we do anything though, we need to get a general talent estimate that adjusts for sample size. After all, Akira Schmid only played about a fifth as much as, say, Mackenzie Blackwood has.

Using an empirical Bayesian estimation similar to what was done here, I created an estimate for the posterior distributions of each of our goalies, and ran 10k simulations to show the range of potential performances for each.

It starts with the distribution of all NHL goaltenders, then “bends” towards the specific performance of the goalie being evaluated more and more as it receives more data. That’s why Schmid has a wider range of outcomes than Blackwood – because it’s had less time to pull the estimate towards the new number.

As you can see, Akira Schmid is the worst goalie in overall performance. This isn’t entirely surprising seeing as Schmid has the lowest save percentage in the NHL by 1.5%. In terms of the potential for absolute catastrophe, he’s second to none. Look at all of those results where he gives up literally twice what is expected! Utterly insane.

But that situation I just described is a little different from how it’s visualized. That graph just depicts each goalie ranked by “average” performance. But what if we did want to draw a line and say “Which goalie is most/least likely to _______”? That would be a different way of assessing them which might lead to a different result. 

There are 32 teams and 64 goalies who faced 500+ unblocked shots for a team. That means that the 48th best goalie, Elvis Merzlinkins, is an average NHL backup. He allowed 37% more goals than expected (1.37 goals per xG) so let’s see which of our goalies is most likely to be at least an NHL backup:

As expected, Schmid is bringing up the rear. In fact, every Devils goalie is an NHL backup in at least 50% of simulations except for Schmid (32%). But seeing as Schmid was both the lowest and widest distribution, this isn’t remotely surprising to us. But now, what if we set out sights a bit higher? 

Instead of looking at the average NHL backup, what if we look at the average NHL starter. Well in this case, the median of the top 32 NHL goaltenders is Tristan Jarry who is saving just over expected – allowing 0.98 goals per xG. Which Devils goaltender is least likely to be an NHL starter? 

Of the Devils goaltenders, the uncertainty surrounding Schmid’s small sample still lands 1.7% of his potential seasons in NHL starter territory whereas Jon Gillies’s less disastrous, but more robustly poor results put him there only 1.2% of the time. So if the “worst” goalie is the one least likely the be NHL-starter quality, the answer is Jon Gillies.

Once more, what if we set our sights even higher. The highest. Presumed Vezina winner, Igor Shesterkin has only allowed 78% of the goals one would have expected as of this writing. Which of the Devils goaltenders is least likely to be a Vezina-caliber goaltender. 

Well isn’t that fun! If we are looking for which Devils goalie was least likely to be Vezina-caliber looking at only this year’s performance, “Goalie of the future” Mackenzie Blackwood takes the crown after finishing that strong in only 1 of the 10,0000 simulations half as many often as the sky-high 2-in-10,000 of Jon Gillies.

So if we are defining the “worst goalie” as the “lowest estimated mean performance” then Schmid is our winner(?). If we want to determine who is least likely to be at least an NHL starter, then Jon Gillies has most demonstrated the highest degree of certainty the he doesn’t belong in that role.

And if we define it as who was least likely to manage to put up elite-level goaltending over the course of the season, Blackwood’s consistent mediocrity over a large sample gives us a higher degree of confidence then any of the wild cards we’ve started since his injury.

So there you have it: the answer to multiple potential interpretations of a question that none of you asked. While it’s understandable if the only thing you feel you’ve gained at this juncture is a newfound depression for the Devils current goaltending plight, it is important to train ourselves to think in a more Bayesian fashion because it puts pictures to that thing our brain is normally doing: updating our image of reality progressively as more evidence comes about.

So, if nothing else, take that with you, and, the good lord willing, next year’s evidence will be much more pleasant to read about!

For anyone who is interested in empirical Bayes or how it was visualized, here is a gist of the code used for this piece. Feel free to play around with it to evaluate similar claims!