The Winnipeg Jets have started the season rocky in the goals against department. They have allowed three goals in their most recent game, and four each in the remainder.
Of course, this has ultimately brought up the conversation on who to blame.
Since the move to Manitoba, the topic of goaltending has been both a hot and controversial topic. Over the previous three seasons the Jets have a +416 Corsi differential, +290 Fenwick differential, +86 shot on goal differential, and -59 goal differential over all minutes in 380 games, where their starting goaltender posted a save percentage below that of his back-ups combined.
So, let’s take a dig into the small samples and really suss out who is to blame.

# An Introduction: Expected Goal Model

Expected goals are a shot differential, much like Corsi, Fenwick, and shtos on goal. The difference is that expected goals are weighted by the probability that the shot was likely to be a goal, given shots of similar characteristics.
Most expected goal models use things like shot location, shot angle, and whether the shot came from on-the-fly rush, rebound, or sustained pressure. While this doesn’t perfectly grab every single detail, it covers the bulk of the pie. Over the past five seasons there have been 54 goals more than the model has predicted, an error of less than 0.2%.
We will use the difference between the Jets’ expected goals against from league average as a measure of how much the Jets’ defensive performance plays blame to the Jets’ above average goals against. Then the difference of the Jets’ actual goals against versus their expected goals should suggest how much falls on goaltending.
As of the morning of October 25th, 2016, the Jets have allowed 19 goals against, one of which was an empty net goal. For this exercise, we will be looking at the Jets and other teams only when goalies are not pulled.

# The Numbers

Now right of the top, you can easily see that the Jets have allowed more goals than their expected goals suggest they should. Now, these are averages of the number of goals one should expect given shot quality and shot quantity, so of course you get things like Michael Hutchinson allowing one and a half more goals than one would expect in his first game.
Overall, we see that Hutchinson has allowed 1.52 goals over two games more than we would expect, while Connor Hellebuyck has allowed 1.27. This averages out to 0.76 and 0.42 goals per game. Combine the two and you get about 2.79 goals more than expected, or about three goals.
On average, NHL teams have paced about 2.57 expected goals against per 60 minutes. Again, you can easily see that the Jets have paced over that in each game, except in Hutchinson’s first start. Expected goal models suggest the Jets should have allowed 15.21 goals on average given their shot quantity and quality. At the league pace over the same number of minutes, the expected goals against would be 12.73.
We then see the Jets worse than average defensive issues have been about 2.48 goals worth.
In summary, we get about 50/50 blame on team defense (47%) and goaltending (53%) for the Jets high numbers of goals.
(Note: If you were referred here, by Gary Lawless from the day prior of this article being published, you will note the numbers are different. This is because the numbers I provided Lawless included empty nets out of my own error, and I did not realize until early the next day (the day of this article’s publishing). When fixing this error, I have also added the one extra day’s worth of games to the NHL’s dataset. The distinction is my own fault and I am sorry.)

# Some Caveats

There are some issues with this data.
League scoring is up this year. On average, NHL teams are allowing one more goal against than the expected goal model predicts. This is a big difference. Given how probability works, it is not unusual for this to sometimes happen under small samples like the first bit of the season. However, there are some who feel this is due to an increase of inexperienced youth in the NHL.
It is unclear whether if the NHL environment has changed or if it is simply just natural variance, but when you change the environment a model based on the old environment does not hold up as well.
Another big caveat is the Jets inability to carry any lead in the game. The Jets have led the game for a grand total of 17.04 minutes. No team has carried the lead less often. The Jets have also been the fourth worst in trailing, with 162.60 minutes.
Score effects do not impact shot differentials only, but shooting and save percentage as well. The Jets should be expected to have a greater xGA per hour than league average simply from trailing early and often. Just like inexperience can cause an increase in goals against, so can taking risks when trailing.
The above numbers are not score adjusted for the reason that trailing is still the Jets’ own fault, both their team and their goaltending, but that doesn’t mean that the risks they take should be

# The Jets goaltending “problem”

Drawing was quick and poorly done, so image is not remotely to scale
I can hear the complaints already. Some accusing me of being harsh to goaltenders. Some crying that the Jets are cursed to always have bad goaltending. Some stating this is proof the Jets save percentage issues were never on the netminder. Some even calling out for the return of Ondrej Pavelec.
All of these things are overly hasty and wrong.
There is a non-zero probability that either, or even both, of the Jets current netminders are Ondrej Pavelec quality or worse; however, what is possible is not what is most probable.
Thus far, the career NHL save percentage for each of the Jets’ netminder is 0.914 for Hellebuyck, 0.911 for Hutchinson, and 0.907 for Pavelec.
Some will rightfully point out that shot quality, sampling size, and variance can cause a goaltender’s save percentage to skew slightly relative to “true talent” levels, but it would contrary to all evidence to call save percentage a team stat.
Some will also rightfully point out that both Hutchinson’s and Hellebuyck’s save percentage has been on the down trend recently, but it would be contrary to all evidence to assume that the trend will persist.
So what do we make of the Jets goaltenders? Well, the above quick (and poorly drawn) graph above should give you some idea.
The red line represents Pavelec. For the most part, Pavelec is a known commodity. He has had his ups and downs, but if you were to take essentially any section of 3000 or more shots, you find him posting almost exactly his career average save percentage.
The blue lines then represent the spread of possible “true talent” level for Hutchinson. Hutchinson has faced just over a tenth of the shots of Pavelec’s career, and is still 1050 shots shy of 3000. This is why he is represented as a spread of outcomes rather than a line. His current save percentage is above that of Pavelec, and so there is a higher probability that he is better than Pavelec versus that he is worse, but both are possible. The down trend in Hutchinson’s recent save percentage impacts the graph through the slopes, and is the reason the top blue line is steeper than the lower blue line.
The green line then represents the spread of possible “true talent” level for Hellebuyck. Hellebuyck’s is much like Hutchinson’s except he has the best save percentage thus far and the smallest sample. For this reason, the spread is much larger but also he is more likely than not better than both Hutchinson and Pavelec.