## Penalty Kill Save Percentage

For this analysis, I used all the goaltender data from 1997 through 2012. In order to run linear regression on the Save Percentages, I calculated logits (log odds, or ln (Saves/goals) ). I removed goalies with logits that were infinity or undefined. This is of little practical consequence, as the regressions are weighted by shots against and all of the goalies removed faced at most a handful of shots. Compared to Even Strength, the Penalty Kill data is a real muddle. The best model explains only about 10% of the variability seen. About 90% of the variability remains unexplained.

Penalty Kill versus Even Strength

There is a small correlation between Even Strength Save Percentage and Penalty Kill Save Percentage. Penalty Kill Save Percentage is converted to "PKLogit" and Even Strength Save Percentage is converted to "RawLogit".

> cor.test(RawLogit, PKLogit, alternative="two.sided", method="pearson")

Pearson's product-moment correlation

data: GoaliesAll\$PKLogit and GoaliesAll\$RawLogit

t = 7.3362, df = 1173, p-value = 4.092e-13

alternative hypothesis: true correlation is not equal to 0

95 Percent confidence interval:

0.1541061 0.2634815

sample estimates:

cor

0.2094489

Multiple Effects

Looking at Team, Year, a Team:Year interaction, and Goalie (here called lastfirst).

> LinearModel.4

> anova(LinearModel.4)

Analysis of Variance Table

Response: PKLogit

Df Sum Sq Mean Sq F value Pr(>F)

Year 1 13.7 13.745 1.3272 0.249636

Team 38 747.1 19.662 1.8985 0.001001 **

lastfirst 252 2802.9 11.123 1.0740 0.233835

Year:Team 31 396.0 12.774 1.2335 0.179541

Residuals 852 8823.7 10.357

---

Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Goalie doesn't seem to make any difference. Rerunning the regression without the goalie.

> LinearModel.3

> anova(LinearModel.3)

Analysis of Variance Table

Response: PKLogit

Df Sum Sq Mean Sq F value Pr(>F)

Year 1 13.7 13.745 1.3228 0.2503467

Team 38 747.1 19.662 1.8922 0.0009791 ***

Year:Team 31 551.1 17.777 1.7108 0.0093333 **

Residuals 1104 11471.6 10.391

---

Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Comparing the two models shows we have not lost anything.

> anova(LinearModel.4,LinearModel.3)

Analysis of Variance Table

Model 1: PKLogit ~ Year * Team + lastfirst

Model 2: PKLogit ~ Year * Team

Res.Df RSS Df Sum of Sq F Pr(>F)

1 852 8823.7

2 1104 11471.6 -252 -2647.8 1.0146 0.4361

Conclusions

Penalty Kill Save Percentage is best predicted by Team and a Team*Year interaction term. Goaltenders do not seem to differ significantly and do not add resolving power to the model. The model predicts only about 10% of the total variability seen in the data.

Please make sure that any content you post is appropriate to Game Time, which means that it pertains to hockey, the Blues, frosty adult beverages, or puppies.

## Trending Discussions

forgot?

We'll email you a reset link.

Try another email?

### Almost done,

By becoming a registered user, you are also agreeing to our Terms and confirming that you have read our Privacy Policy.

### Join St. Louis Game Time

You must be a member of St. Louis Game Time to participate.

We have our own Community Guidelines at St. Louis Game Time. You should read them.

### Join St. Louis Game Time

You must be a member of St. Louis Game Time to participate.

We have our own Community Guidelines at St. Louis Game Time. You should read them.