^ understandable
And nothing wrong with that.
Fact is, you have a stat like save percentage
It’s easy. How many shots on goal are saved?
The problem is that comparing goalies, maybe not all shots are equal
The improvements on the model involve a first modification. Where are the shots coming from?
Divide the ice in to different areas. Divide the shots taken into buckets, from different places on the ice. Sure, a higher percentage of shots from the slot area are going to go in than shots from the point.
Then they try to work in what happened preceding the shot.
Good. So they are creating multiple buckets of ‘shot situations’, which each have been successful at different statistical rates
Still, there is no measurement of shooter skill, shooter prep time, goalie readiness, shot placement, etc. etc
And per game sample sizes are small, and of notably reduced statistical significance
You have roughly 30 data points (shots) that then get separated in to multiple buckets, which are based on different situations, each with their own statistical probabilities (ex. Based on analyzing 1000 shots, 83 percent of this type of shots score, 19 times out of 20). And that is ignoring material differences distinguishing one shot from another because they aren’t measured
Kudos to them, they are trying. And they are getting better
Sometimes, these stats support what you are seeing. But it’s a stretch to use them to tell you that what you are seeing isn’t actually happening
|