Hello everyone and welcome to Arctic Ice Hockey's very own Friday #FancyStats Mailbag! Once a week I'll be answering a few of your questions on underlying metrics. Also, I will try to explore any concerns about the Jets that can be investigated through data.
Jeff writes:
I was wondering if you'd seen anyone look at goalie strength of opposition. I wanted to find a statistical answer to the question "do back up goalies get easier starts?" The anecdotal evidence says yes, but I wanted something more concrete. I couldn't find it as a stat on any of the analytics sites I frequent. I asked around on twitter too but haven't found anything.
There have been times where this has been looked at before.
In 2012, Ben Wendorf looked at how certain variables predominately outside of a goaltender's control influence a goaltender's win percentage. Although the best way currently to judge goaltenders is by large sample save percentage, the public's and media's opinions tend to follow how often a goaltender is winning. This wasn't exclusive to competition but the results are interesting. (Articles are here and here).
When it comes to competition metrics, goaltenders are a difficult group to create one for. Rob Vollman has previously used shooting percentage and average distant of shots of their opponents as a measure. The issue is that even over a full season shooting percentage and shot location may give a false interpretation of the skill level the goaltender faced, since these numbers tend to regress heavily over time.
One quick way to look at this is by the average 5v5 goal scoring rate a goaltender's opponents have throughout the season when not against them (OppGF60 at stats.hockeyanalysis.com) over the last four seasons. Which goaltender is a starter and which is a back up is fluid over a four year span, so I had to make an arbitrary split at 35 goaltenders, with Braden Holtby being the least used starter and Nikolai Khabibulin being the most used back up. The weighted (by games played) average of OppGF60 was found to be 2.2675, while the weighted average for back ups was 2.259. In other words, the shooters starters and backups tended to face over 4 years was similar in scoring goals per minute.
Here is just last year's data if you want to look at trends, although the determination of correlation between games played and OppGF60 was less than 0.001.
Jonathan writes:
In situations that aren't score close does shot distance increase?
This is a very good question that I haven't been able to fully find the answer to yet, and may re-approach if the data becomes available or get the time to write a script that can scrap the necessary information. As it currently stands, nobody has shot location data by score situation.
However, I would hypothesize that score-effects do indeed effect shot location, and there is some evidence that backs this up. We already know that shooting percentage changes at different score situations (graph by Petbugs). We also already know that shot location is the major driver of shooting percentage (graph by Gabe Desjardins).
Score effects occur due to a change in player behaviour due to having and maintaining a lead or trying to close a gap. It is far more likely that we are seeing significant changes in shot location with score effects rather than dramatic changes in how many shots at the same location go in.
For more on score effects, this article is highly recommended.
Thank you everyone who wrote!
That is all for this week. If I didn't get to your question this week, I promise I will eventually reply. The plan is to answer every single question given to me, although sometimes they may slide a week or two depending on the load of questions I receive.
If you have any questions you want looked at, email me at garrethohlaih (at) gmail (dot) com.
