:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/32320767/486753385.0.jpg)
Finally, let's look at Shootout Save Percentage. I am going to look at 2009-10 through 2013-14. I ran the analysis here two ways. The first way was the way I have done with ES, PK, and PP. 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. In the setting of shootouts, all goalies faced very few shots. A relatively larger number of goalies had save percentages of 1.000 or 0.000. So I also tried looking at unconverted save percentages.
Goalies are all the same. > LinearModel.1 = lm(SOlogit ~ Goalie, data=Shootout, weights=SOSA) > anova(LinearModel.1) Analysis of Variance Table Response: SOlogit Df Sum Sq Mean Sq F value Pr(>F) Goalie 104 285.43 2.7445 1.1639 0.1795 Residuals 207 488.12 2.3581 I was a little worried that I had thrown out too many goaltenders. The logit transform normalizes probabilities and makes for valid regression analysis. Could I just run the regression on the save percentages? How does the distribution of SO Save Percentage look? Ugh! Not a nice bell-shaped curve. Forging on and running the analysis anyway: > LinearModel.2 = lm(SOPct ~ Goalie, data=Shootout, weights=SOSA) > anova(LinearModel.2) Analysis of Variance Table Response: SOPct Df Sum Sq Mean Sq F value Pr(>F) Goalie 111 31.433 0.28318 1.1429 0.2012 Residuals 225 55.751 0.24778 Still no effect there. Conclusions There is no goalie effect, just randomness. Michael Shuckers had looked at this in 2010 and found the same thing. "Based on this analysis," Schuckers said, "I conclude that the NHL shootout is a crapshoot. If you look at all goalies over the 5 years, 1821 goals were scored on 5599 shots, a 0.675 Save Percentage. For their careers, Ryan Miller is 0.710 (81 goals on 279 SA), Jaroslav Halak is 0.690 (40 goals 129 SA), and Brian Elliott is 0.648 (38 goals on 108 SA).