These models have never been able to filter out noise and allow for context.
Again, I had people telling me that Bo Horvat was sub-replacement level in 2019-20 based on WAR calculations. Two years ago, people were telling me that Chris Tanev was garbage based on this stuff.
I don't like Tyler Myers. His contract sucks. But I'm convinced he's a pretty average #4-5 defender. If something is spitting out that he's one of the worst defenders in the NHL, to me that statistic is a bad statistic. It *MAY* mean that his ability : usage ratio is amongst the poorest in the NHL, but again I'm not even convinced of that.
If you took a league average mid-pairing defender and put him on a bad team with a bad partner in tough minutes, these WAR metrics will tell you that that player is one of the worst players in the NHL. And when that's the case, the data is worthless.
I love statistics. I love WAR in baseball. But I also hate bad statistics, and think that people making bad judgements off bad statistics are every bit as bad as bad dinosaur GMs making bad decisions off bad data like boards go boom.
I'm not really looking at WAR/GAR. Those are goal-based metrics, and thus are really susceptible to swings in shooting/save percentage. For a descriptor over a large amount of time, I think it's okay, but I don't think it's the best metric either and I haven't referenced it here. I've referenced RAPM, which I think is a lot more useful.
I'm mean, before Myers was signed, I also thought he was # a 4/5, despite what the numbers said. Then watching him, I've adjusted my opinion downward more in line with some of the numbers. We could be wrong, is what I'm saying. I disagree with the view that "oh, this model very clearly contradicts me so it must be wrong". We could be wrong.
Your example of a "putting a mid-pairing defender and putting him on a bad team in tough minutes" will mean he's automatically the worst player is not true. Someone like Jake McCabe from Buffalo is something of an analytical darling, based on his defensive results. Guys in tough situations can have good results. So perhaps there is something else about Myers we're not getting at.
Of course, you can easily find examples of the models "missing" on players. Just like you'll find innumerable examples of our eye tests missing. That doesn't mean "throw out the models". I mean, by shot metric models, Tanev used to be one of the elite defensive defensemen in the league. Honestly, I'd trust a flawed but systematic and
consistent model of player evaluation far more than my own eyes at real-time.
In sum, please don't let Rathbone be attached to this guy for the next 3 years.