Ilya Bryzgalov and the Shootout

They played very well last night but still have such difficulty scoring. Reimer went down way early like bryz and bob but we did not make him pay for that. we shot low. Also in the shootout bryz has not been good but neither have the shooters. If you are tired don’t shoot or tell me why you are shooting and skating right down the middle with no move. You have to make a move bottom line. And I watched the game closely last night and see a big problem with our team. The flyers offensive players get back but they don’t defend they try to defend space and then chase. I saw three guys on one goal chase one guy and he did not score lol. coburn was owned by lupul, you have to clear the crease boys. The plus in all of this, was the last 6 minutes in the third. That was some of the best hockey i have seen from the flyers all year.

There’s also a lot of pressure on him because we rarely score in shootouts..atleast when I watch. I can imagine he’s over thinking things because he’s always getting burned so bad but if lavy can stop picking Simmonds to shoot and they score more it will take a lot of pressure off Bryz

a rational look at whether he is genuinely bad at shootouts

At “rational view on shootouts,” you lost about 75% of Flyers fans.

Good analysis.

But be glad because this is a playoff team and one of our (perceived) biggest weaknesses will not matter when we are in the playoffs. Winning these games only improves the playoff seed. This team has been better “when it matters” as lower seeds and on the road anyway. Right?

I know it’s completely anecdotal but from the small amount of highlights I’ve seen of Bryz in Phoenix it looks like he was a lot better with the poke. It seems the vast amount of times he has gone for the poke here in Philly it always seems to burn him. Maybe teams have scouted this? I could also be making this entire thing up.. Either/or.

Interesting analysis – Thanks for your work on it.

In looking at the graph, a few things struck me: (1) that there are only 26 data points in the sample and (2) that one or two extreme points can really influence the slope of a regression line with that few of data points. These observations got me thinking of whether a few outliers (of one or two goalies of the 26) might be masking a more general relationship.

Rather than just be satisfied with idle curiosity, I dug into the data a bit more to see what would happen if the sample size was increased, which will decrease the effect of any particular data point. Thus, I sliced the data to look at goalies who have faced at least 50 shootout attempts (instead of the criterion of 100), which increases the sample to an N of 47 (and includes goalies such as Ward, Howard, and Halak that were excluded with the original number of shootout attempts criterion of 100).

When you make this simple change (to 50 attempts instead of 100 attempts), there is a positive relationship between career save percentage and career shootout save percentage – The correlation between these variables is .22. This relationship (r = .22) is small and ultimately doesn’t account for much variance. Further it is not statistically significant, although it approaches significance with a one-tail directional test (p = .07) – This lack of significance is mainly due to the low statistical power given the small sample size.

Overall, I personally think this reanalysis does show that there is some positive relationship (albeit a small one and one that only approaches significance) between career save percentage and shootout save percentage when you expand the data to include a more liberal criterion of 50 shootout attempts. In my opinion, based on this reanalysis, the relationship is indeed small (r = .22), but it is far from "worthless" in understanding shootout performance.

As a postscript – I understand that one could argue that the cut point of 50 shootout attempts is not stable (and that is why 100 shootout attempts should be used); however, this unreliability of the measure could actually work against the magnitude of the relationship such that the correlation would be larger if I corrected for the unreliability. Further, I personally feel that the tradeoff in terms of the positive of almost doubling the sample size and increasing statistical power is worth any potential negative associated with this potential unreliability. I understand that one always needs to make decisions about how to slice the data and reasonable people can disagree as to the best way to do it. Just my attempt at another way to look at things.

The result is interesting.

I think to some extent you are oversimplifying the argument the skill side would make. Fleury is a good example since his best traits as a goalie seem to be quickness, reflexes and the like. His positioning can be bad, his stick handling is dreadful enough that it probably has a real effect and he just has these weird lapses where he lets soft goals so we’ll let’s call that mental focus. If you add it all up he ends up being decentish, but all of his positives seems important to stop a penalty shot and the negatives less so. Also, how sweet was that save on the Ovechkin breakaway in game 7?!?!?!

On the flip side, the correlation is really freaking small. Even with the type of thinking I just went through, if there were pronounced skills that varied across goalies some of that would leak through it seems like you’d at least hit double digits. The only way it would be this extreme and for there to be a big skill component would be for there to be a positive correlation between being lightning quick post to post and having a super strong desire to dick around and lose the puck behind your goal.

It’s worth pointing out that none of this shows that there is no skill, we’d be dreadful defending shootouts, it’s just that NHL level goalies are so close in those skills that they don’t matter.