Sunday, September 13, 2009

Investigating Special Teams Skill

Is penalty kill goaltending a distinct skill from even strength goaltending? We know that skaters show differences in ability on special teams compared to at even strength, which suggests that 4 on 5 hockey is a different game than 5 on 5. If so, maybe there are differences in goaltending results as well. Just anecdotally, there are a few goalies that seem to do better at even strength than on the penalty kill. J.S. Giguere comes to mind as someone I've run across recently who fits that profile. However, we would expect a few goalies in the entire population to have some weird results just by chance or because of possible team factors, and even if a few guys show a difference in ability it does not mean that is typical around the league. We need to look at some overall numbers over a larger sample size to get some better evidence for or against this potential effect.

I checked the correlation between even strength save percentage and penalty kill save percentage at the team level, using the 10 seasons' worth of special teams numbers from The correlation coefficient was 0.71, which is a high result that suggests a fairly close relationship.

With a correlation that high, we should be able to use EV SV% to predict PK SV% with some reasonable accuracy. I used Excel to generate a linear equation to predict PK SV% from EV SV%, and compared each team's results with the expected. The predictions were pretty good. Only 7 out of 30 teams had a difference between actual and expected PK SV% greater than .006. They were the following:

Minnesota: .886 PK, .878 EXP (+.008), .925 EV
Detroit: .875 PK, .866 EXP (+.009), .917 EV
Washington: .875 PK, .864 EXP (+.011), .915 EV
Columbus: .872 PK, .862 EXP (+.010), .914 EV
Boston: .861 PK, .869 EXP (-.008), .919 EV
Phoenix: .858 PK, .869 EXP (-.011), .919 EV
Atlanta: .849 PK, .859 EXP (-.010), .911 EV

There's Minnesota again, with yet another result that suggests they are doing something abnormal shot quality-wise. Another familiar team with some shot quality outlier evidence behind them are the Atlanta Thrashers, who also make their usual appearance at the bottom of the table.

It could be that some teams were excellent or weak for a few seasons, but over a longer time frame that tends to get mostly washed out in the average. A few teams appear to have persistently higher or lower results than expected, which suggests that the penalty kill situation is very similar to the one at even strength: For most of the league there is little shot quality effect, although there are apparently a few outliers at each extreme. It seems that for the most part goalie performance is similar at even strength and on the penalty kill. Either that or teams tend to have similar shot quality against at EV and on the PK. One other factor that would impact PK shot quality is the number of 5 on 3 chances against. I suspect this is one of the reasons Minnesota ranks so high. With their excellent team discipline they were less likely to have to face 5 on 3 situations.

I also checked the correlation between EV SV% and PP SV%, and it was essentially zero (0.01). The sample size for power play shots is a lot smaller, as the most shots any team faced over the last 10 seasons while on the power play was 1,096, which is about half a season's worth of shots for a 70+ game starting goalie. I'm not sure a goalie's save results while his team is on the power play are particularly meaningful.

What these results suggest to me is that we should rate goalies based on their performance both at even strength and on the penalty kill. Focusing on EV only would be underrating the Lundqvist/Luongo types that are likely adding a lot of value through their play on the penalty kill. It does not appear that there is much of a skill difference by game situation for goalies. It is more difficult for all goalies to make a save when their team is down a man, but a better goalie is still generally more likely to make that save than a weaker goalie. We certainly need to take into account the number of power play shots against for each goalie, to level the playing field for goalies on teams with higher or lower penalty totals. Just as at EV, it is likely that in today's NHL most teams around the league are pretty similar in terms of shot quality against on special teams, although there are probably a few outliers that we need to keep in mind. Looking at situational save percentages for even strength and 4 on 5 play only would probably be the best way to go.


Navin Vaswani (@eyebleaf) said...

Great post. Thanks for all the work you do. Love the blog.

Triumph said...

this is some excellent work, but i do not necessarily agree with your conclusion; perhaps i missed something? i'm not sure how we get from ES SV% is 'highly correlated' with PK SV% to saying that PK SV% is a skill. 5 on 3 penalties should certainly affect it, Gabriel Desjardins' work on defensive-zone faceoffs should affect it.

I think it was Henrik Lundqvist's name appearing on the 'PK is a skill' list; for almost his whole NHL career, he's played for one coach, under one system. While I'm the first to discredit coaches and systems, you (and Gabriel Desjardins) have noted Minnesota's doing something wrt shot quality - perhaps New York is as well.

Anyway, this is kind of all over the place and a nitpicky criticism to boot - great work here.

The Contrarian Goaltender said...

Triumph: That's a fair point, I might be concluding too much here, but I'm doing so with a few previous results in mind. This post was born from a discussion with JLikens who has done some investigation into PK SV%. His most important finding was that there was no correlation between the starter's PK SV% and the backup's PK SV%.

I'm not trying to say that we can simply compare PK SV% for different goalies because there are no team effects at all, just that I think PK is similar to EV, i.e. it is likely that most of the league has similar shot quality against. I agree 5 on 3 is an important factor, so I would like to ideally look at 4 on 5 numbers only when evaluating goalies on the PK.

Bruce said...

Interesting stuff, CG.

It could be that some teams were excellent or weak for a few seasons, but over a longer time frame that tends to get mostly washed out in the average.

I agree with this. Would it be worthwhile to do/derive a similar study just for the four seasons since the lockout?

The Contrarian Goaltender said...

Post-lockout only:

Correlation between EV SV% and PK SV%: 0.66

Correlation between EV SV% and PP SV%: -0.32

With the smaller sample, the predictions of PK SV% based on EV SV% were not as good. However, half the league was still within .005 despite the higher expected variance because of fewer PK SA. The evidence still suggests that most of the league likely has little difference in shot quality between 5 on 5 and 4 on 5.

There were four very obvious outliers, two on each end of the scale. Minnesota (+.022) and Edmonton (+.021) both had much better PK SV% than expected, and Atlanta (-.019) and Phoenix (-.019) were much worse than expected. Any ideas on what could explain that Edmonton one?