The Importance of a Goalie's First 15 Games
by Hawerchuk on Feb 23, 2011 7:00 AM EST
in Statistics
Email
Print
After watching ten games from the Leafs' bench, James Reimer has put up stellar numbers in his first 15 games in the NHL. But can we draw any conclusions from his .932 save percentage to start his career?
| Games | 1-15 | 16-30 | 31-45 | 46-60 | 61-75 | 75-90 |
| Reimer | 932 | |||||
| 13 Comps | 931 | 909 | 906 | 903 | 906 | 914 |
We have 13 comps since 1995 - goalies who were called up and posted a .925 save percentage or better in their first 15 games. But over their next 60 games, they posted a very pedestrian .906 save percentage, which would seem to be a huge disappointment for guys who started out so hot. Given that the difference between an All-Star .920 goaltender and a replacement level .900 goaltender is one goal every other game, 15 games is not a large enough sample size to be confident in a goaltender's abilities - one out of every six .900 goalies will out-perform a .920 goalie over a 15-game stretch.
What's interesting is how coaches perceive those hot starts. The typical hot-starting goalie gets 12.2 starts in his first 15 appearances - and after posting those huge numbers, gets 13.5 starts in the 15 following games. And even though his performance is poor over that next 15-game stretch, he gets to keep playing:
| Games | 1-15 | 16-30 | 31-45 | 46-60 | 61-75 | 75-90 |
| Reimer | 13 | |||||
| 13 Comps | 12.2 | 13.5 | 12.8 | 12.6 | 14.2 | 14.3 |
After 60 games, we get some amount of attrition as coaches realize that the goalie's hot start was really nothing, and so the weaker goalies are no longer playing, while the stronger ones play better and keep their starting jobs.
None of this should be even remotely surprising - if you're looking at a goalie's cumulative save percentage, that hot start makes him look above-average even if he's been awful recently:
| Cumul. Sv% |
1-15 | 16-30 | 31-45 | 46-60 | 61-75 | 75-90 |
| 13 Comps | 931 | 920 | 915 | 913 | 911 | 912 |
It's an interesting effect - 15 good games get you another 45 games to prove that you're not good. In terms of determining a goalie's talent level, it doesn't matter when he plays well and when he plays poorly. But if he wants a chance to play, he'd better put up ridiculously good numbers when he gets his opportunity.
22 comments
|
1 recs |
Do you like this story?
Comments
Good news for Leaf fans: .906 is actually much better goaltending than they’ve been getting since the lockout.
I've been looking at the sky
by Back In Black on Feb 23, 2011 10:13 AM EST reply actions 3 recs
Amen!
For the record, I actually quite like what I have seen of him so far – he has a great attitude and his play has been pretty solid. I’m staying positive. I need something to grasp at.
Though, the following quote made me think the kid must read this blog…
“I’m getting the minimum here (US$555,000) and it’s 10 times more than I got down (in the AHL),” Reimer said Tuesday before making his fourth straight start against the New York Islanders. "It’s a lot bigger. It’s weird. It feels like you don’t deserve it sometimes — what am I doing to deserve this?
“There are many people that work a lot harder than I do who don’t get paid half as much.”
by Ricci Handsome on Feb 23, 2011 11:02 PM EST up reply actions
Goalie origin
I’ve been reading your work for a long time so I hope my first post isn’t a foolish one. I’ll be happy if there’s one decent nugget below.
Of the 13 comparable netminders, is there any link between league and level that can be gleaned from this tiny sample size? In other words, does a goalie whose last level was a European pro league tend to start hotter than an AHL one or vice-versa? Could any of these early starts have been projected by league equivalencies?
This ties to my next question: also, you mention the performance across the next 60 games. Since performance can ebb and flow, how many of those goalies ‘bounced back’ (in the eyes of the average fan) and became established in the league past that point and how many fizzled out?
Thanks for reading; more importantly thanks for posting.
Ahh…good ol’ primacy effect.
That and the fact that coaches and GMs really dont know how to judge goaltending performance.
Hey, I thought GM’s knew everything? I mean they don’t use Corsi right, so they must know what’s going on.
Hockey Blogger at Pensburgh.com
It’s an interesting effect – 15 good games get you another 45 games to prove that you’re not good. In terms of determining a goalie’s talent level, it doesn’t matter when he plays well and when he plays poorly.
Couldn’t it also be a fact that you can’t be having that good of goaltending to begin with if you make a new goalie your starter over 15 games?
We’re dealing with such a small sample size of goalies here, that I can see that being the case … even for the goalie under consideration – the heart problems with Gustavsson, the recurring injuries for Giguere, and the initial finger injury for Rynnas have somewhat made Riemer a starter by default more than anything else.
Guess who else could be included in this cohort? Sergei Bobrovsky.
Does he call it Luongo underwear?
Co-Manager at Behind the Net
by Bettman's Nightmare on Feb 23, 2011 5:17 PM EST reply actions
Lies.
.930 ESS% is Bob’s true talent.
HuckNZ recently wrote an excellent post on this topic at BSH. I don’t think you’ll be surprised to find that many Flyers fans think Bob will rival Hasek as one of the best goalies to play the game.
Mourning Gagne forever.
Hey, I’m a Bob apologist as much as anybody, but even I had to balk at the comparisons to Hasek.
Does he call it Luongo underwear?
Co-Manager at Behind the Net
by Bettman's Nightmare on Feb 24, 2011 12:55 AM EST up reply actions
Which part?
I realise that Bob’s true talent isn’t .930 ESS%, but apparently many voters do not.
Mourning Gagne forever.
I thought you were seriously arguing Bob’s true talent was a .930 ESsV%
Hockey Blogger at Pensburgh.com
The point that this sample size is way too small to conclude anything at all is well taken
But your comparison is dodgy.
The problem is that you are specifically selecting the comparisons BECAUSE of their hot start. Given enough goalies that start enough games, sooner or later some of them are going to start off with a solid run of luck, and so if you continue following them they will regress to the mean.
Will Reimer do the same? We’re not interested in him (speaking as a Leafs fan) due because we combed through a large number of goalies looking for one who turned in a brief spectacular performance, so it’s not really fair to say we should expect her performance to recede due to regression to the mean… Really we need to think about this in a Bayseian way.
What is the prior probability that he is a superstar? Probably low, just because most NHL level goalies are not. Now given that he has unexpectedly posted these very solid numbers, do we revise upwards our estimate of his likely quality? Yes. We do. Given the tiny sample size we only up it a smidge… and given how large sample sizes you need for noticeably reliable estimates, that smidge is fairly microscopic… But we DO have just a teeny bit more confidence he’s better than average.
Just to be theoretically correct and all.
not sure I follow. Bayes tells us that any goalie’s first 400 shots against in the NHL are basically meaningless and that he’s probably from the below-average population regardless of how he did.
But people don’t accept Bayes. So this shows what Bayes implies – that a hot start is meaningless.
Sorry if I wasn't clear
but I your summary is about right. The only difference (and it’s a stylistic one, not a substantive one) is that our guess starts out as him being bang-on average before we get any information about him, and then as he adds good performances we update this guess to maybe think he’s better and better. Given that, as you say, a sample size of 400 shots doesn’t tell us much, then our best guess for Reimer is currently just a fraction of a hair over average.
To pull numbers totally out of my ear, if the average goalie saves .915, then that is our first guess for Reimer. After his strong performance over these few games now, our official guess for him is probably something like .91500006. It’s moving upwards, but until the sample size gets a lot larger it doesn’t really have the opportunity to move up very far.
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
