FanPost

NHL teams decrease their goals against by raising their save percentage

Screen_20Shot_202016-11-29_20at_209.39.26_20PM_zpsyyfth2yo.0.png

Fasth and Scrivens should make us think twice about PDO doctrine

Wading into old waters

There's an old saying: "You can't step in the same river twice." When it comes to hockey analytics apparently this isn't so. We travel in circles, chasing our tails, revisiting spreadsheets and hunting for clues.

In my previous post I showed that teams score more goals by raising their shooting percentage, and that the number of shot attempts (CF60, FF60, SCF60) has a weak relationship to scoring (GF60), much weaker than shooting percentage (Sh%). What of the opposite?? What measures correlate highly with goals against (GA60).

Save Percentage is more important than Corsi for defence

image_206_zpswoieaecc.0.png

The above chart shows NHL team save percentages (SV%) vs goals-against per 60 minutes (GA60) from 2007 to 2016 for 5v5 even strength play (interactive chart here). Teams with higher save percentages give-up fewer goals. The mean team save percentage is 92.19 , with a standard deviation of 0.87. To stop one more goal per hour of even strength play, teams need to increase their save percentage by 4.

image_205_zpswqsjxesi.0.png

The above shows unblocked shot-attempts against per 60 minutes (FA60) against goals per 60 minutes of 5v5 play for NHL teams from 2007 to 2016 (interactive chart here). The relationship between GF60 and FA60 (R2 0.256) is much weaker than the relationship between SV% and Ga60 (0.664). To decrease GA60 by one, NHL teams have to limit more than 20 attempts per 60 minutes of 5v5 play. Of all shot attempt counts (Corsi, Scoring Chances, Fenwick) Fenwick-against 60 has the strongest Pearson correlation to GA60.

How to increase team save percentage

Stopping pucks is a matter of limiting the quality of the opposing team's chances and keeping pucks out of the net. It's complicated. There are a lot of factors involved including the choice of players, their ability to push the opposition to lower-percentage areas, and their ability to read plays and stop odd-man rushes and breakaways. The goalies might have something to do with it too. There are a lot of factors involved that I believe are worth exploring.

PDO is bunk

I believe in Corsi and Fenwick as valuable measures. Ultimately I suppose I also strongly believe in Sv%+Sh% as a valuable measure. But I don't necessarily believe in "regression" of percentages. Clearly, over a few games percentages can be very misleading indicators of what's going on with team defence and team offence. But over long periods of time, small differences in percentage matter a lot. This puts me in the camp of believing in "shot quality", but it's more than just the quality of the opportunity of the shot involved, more than the distance. Many great analysts have been looking at these problems for a long time. But as I see it, the common wisdom in the analytics community, which is to ignore SH% and SV% data and view it as luck (PDO), is very misleading.

Take for example the Oilers of 2014-2015. Many bloggers were convinced that our Sv% under Ben Scrivens and Viktor Fasth would eventually regress to the mean. I was not convinced and I believed there were problems both with our goaltending and our defence. That season we finished with a very low save percentage and it cost us dearly. It was something that I could see clearly on the ice. I didn't believe the extremely poor performance was the product of luck. Maybe I was right. Maybe I wasn't. But one thing's for certain: PDO wasn't a helpful analysis tool.

In This FanPost

Teams

Trending Discussions