:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/32322833/467422823.0.jpg)
Once again let's look at Even Strength Save Percentage. I am going to look at 2009-10 through 2013-14. I have shown previously that goalies are different at Even Strength. I have shown in the past that teams are the same and we hope that is still true. From some old studies, I know that there was a year effect. Even Strength Save Percentage was 0.914 in 1997 and was 0.920 by 2010. Even Strength Save Percentage has been stable for several years, so I expect there is no Year effect in this data. Once again, in order to do regression analysis I constructed logits (ln odds of a save). I threw out the goalies whose logits were zero or undefined. All the regressions are weighted by number of shots faced. Lastfirst is the name of the goalie variable.
There is no year effect in this data. > LinearModel.4 = lm(ESLogit ~ Year, data=Goalies5, weights=ESA) > anova(LinearModel.4) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) Year 1 23.4 23.400 1.3826 0.2403 Residuals 447 7565.4 16.925 Goalies differ, teams don't. > LinearModel.3 = lm(ESLogit ~ lastfirst + Team, data=Goalies5, weights=ESA) > anova(LinearModel.3) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) lastfirst 145 3208.1 22.125 1.5977 0.0004795 *** Team 29 586.5 20.226 1.4606 0.0650010 . Residuals 274 3794.2 13.847 Just to show this isn't due to the choice of years, I also looked at the last 4 years and 6 years. > LinearModel.1 = lm(ESLogit ~ lastfirst + Team, data=Goalies4, weights=ESA) > anova(LinearModel.1) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) lastfirst 135 2759.64 20.442 1.4177 0.01265 * Team 29 540.88 18.651 1.2935 0.15589 Residuals 198 2854.94 14.419 > LinearModel.2 = lm(ESLogit ~ lastfirst + Team, data=Goalies6, weights=ESA) > anova(LinearModel.2) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) lastfirst 157 3858.1 24.574 1.7970 4.382e-06 *** Team 29 437.9 15.101 1.1043 0.3285 Residuals 343 4690.5 13.675 Goalie and Year There's no overall year effect, but there might be a year effect within each goalie. For example, Brian Elliott might go up, up, up and Cam Ward might go down, down, down. > LinearModel.5 = lm(ESLogit ~ lastfirst/Year, data=Goalies5, weights=ESA) > anova(LinearModel.5) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) lastfirst 145 3208.1 22.125 1.4001 0.01348 * lastfirst:Year 99 1157.1 11.688 0.7396 0.95408 Residuals 204 3223.7 15.802 First of all, this is an overparameterized model. We are trying to estimate 145+99 = 244 variables with only 204 residual degrees of freedom. You might see a significant result just because of that. But there is nothing here. So the upshot is, if you are trying to predict goalie performance, knowing which goalie makes a difference. Knowing last year's result doesn't add useful information. I will get into that more in a subsequent post. Goalie and Age There might be a relationship between ES Save Percentage and Age. I had found a small relationship in 2010, but it was being driven by a handful of goaltenders who could not possibly continue to get better (Cam Ward was one of them and he certainly didn't). > LinearModel.6 = lm(ESLogit ~ lastfirst/Age, data=Goalies5, weights=ESA) > anova(LinearModel.6) Analysis of Variance Table Response: ESLogit Df Sum Sq Mean Sq F value Pr(>F) lastfirst 145 3208.1 22.125 1.4001 0.01348 * lastfirst:Age 99 1157.1 11.688 0.7396 0.95408 Residuals 204 3223.7 15.802 Once again, this is an overparameterized model. Once again, despite that, there is nothing here. ES Save Percentage doesn't vary with age in a meaningful way. Discussion There is no year effect. There is no age effect. There is no team effect. There is a significant goalie effect. There may actually be a small team effect, just one that is too small to see in 5 years of data. This data suggests that if such an effect exists, the magnitude is around 0.001 or 0.002. Over the course of a single season this is of no practical (or statistical) significance. In the 5 years, goaltenders faced 273598 shots and made 251863 saves for a 0.92055 ES Save Percentage on average. The way goalies differ is a bit unexpected. No goalie in the database is significantly better than average, and no goalie is significantly worse than average. The top goalies are better than the bottom goalies.