I'm curious why Ellis' EVO is so much higher than guys like Hamilton and Makar.
Hamilton has 0.01 less points/60 and much better metrics, but somehow it's less than half of Ellis'?
GAR uses regression of on-ice GF, not points, to calculate EV offense.
At 5-on-5 with Ellis the Predators scored 4.09 GF/60 and without him they scored 2.29 GF/60. It makes sense that any regression model would suggest he had an absolutely insane impact on his team’s on-ice goal rates.
While Ellis is an excellent defenseman, the fact that he ranks so high shows a few of the key flaws with this stat. Using on-ice GF (or individual scoring) for any player in a single year sample will tend to produce some wacky results due to variations in on-ice shooting and GAR is no different in that respect. The even strength offense component of GAR explains a huge portion of the variance observed in skater GAR from year to year, but quite often it can be heavily skewed by luck. This is an even bigger problem for defensemen because conventional wisdom and research both tell us that unlike forwards, defensemen have little control over their on-ice shooting rates to begin with; it’s probably not actually Ellis’ doing that the Predators are shooting 11.28% with him and 7.43% without him, yet a glance at the results just says “Ellis was the best.”
It’s still a very useful metric, but it’s obviously not something that anybody should just take as gospel or a definitive ranking. Ideally, you’d see analytics fans use it the way that “eye test” fans use points today; not a definitive judge of player value, but the list of best players by position will generally look similar to the list of best players by this stat. In practice I think that mostly holds.