With Jaro Halak's injury on Tuesday, there's been a lot of commentary lately about how the Isles will likely not be that hurt by the injury due to the amazing performance of Thomas Greiss. Greiss leads the NHL in overall save percentage this season amongst qualifying starters. In terms of 5v5 save percenage, Greiss comes out #2 (tied with Brian Elliott) in the league, behind only Henrik Lundqvist.
You may also have seen Blueshirts Banter and Hockey-Graphs writer Nick Mercadante tweet out this picture:
showing Greiss as the #2 goaltender in the league this year, once the difficulty of shots is taken into account.
But the question is: How much of this is real? The Isles got Greiss this offseason on a cheap two year deal, after the Pens used him as a backup for one year. Is he bound to fall back? And if so, how much?
Goalie Performance and Sustainability
Goalie performance is HIGHLY variable - good goalies will go through streaks of seemingly long duration of terrible play and bad goalies will do the opposite. Even a goalie playing a full single season doesn't give us a large enough sample in order to determine whether a goalie's performance is reflective of his true talent. So how many shots are needed?
3000 is a number frequently banded out to answer this question - see this old Behind The Net post or this The Book post. But the answer isn't that simple: In both those posts for example, 3000 is simply the answer essentially to a good number where we're pretty sure that a goalie's performance isn't completely luck - in fact, in the Book Blog post, it's suggested that 3000 shots is roughly where you'd guesstimate 50% of the goalie's performance is based on luck!
Thomas Greiss has faced 3123 shots in his career, but many of those have been from many years ago - and players DO change over time. So even if 3000 shots was a good enough number, this still wouldn't be a particularly great sample to use as is.
Projecting Thomas Greiss
A quick answer to this problem is to use Hockey Marcels to project Greiss going forward. Hockey Marcels are a very very basic way of projecting player performance: We take a players' #s over the last few years, weight the more recent years more heavily, and then regress the player performance based upon the sample size (specifics can be found in the link). If we were projecting beyond this year, then, we'd also add an adjustment for aging - but let's simply talk about this year.
If we just weighted Greiss' performance the last 4 years per the Marcel formula, we'd expect Greiss to have a .9227 SV% going forward. This is very good - well above average - but a decent drop from where he is right now at .929. However, we can't simply do that - we need to regress as well, for the sample size for this calculation is only 1563.5 shots - far lower than we need to be confident in Greiss' true talent.
Thus, after a large regression, we'd project Greiss to put up a .9186 save percentage over the rest of this year. This comes in at slightly above average for an NHL goaltender (NHL Save Percentage has trended upwards a bit this year and previous years). So playing Greiss won't hurt the Isles. Should we expect him to keep standing on his head and carrying the team? Not as much.