I wish I had a more substantive comment, because this is deep stuff. I wish I could add anything. It does seem weird to me that having faster players on the ice would result in worse shots.
What is your interpretation? Is it as simple as faster players tend to be worse, therefore writing the reduction in shot quality off as more Omitted Variable Bias than anything, in the absence of a variable for shooting talent? Or is it something deeper than that?
That's true. Can an xG model differentiate the different types of shots (rush shot, one time, etc.)? I never even thought of that. I do suppose your second point could perhaps explain why it feels like all the fast players are bottom six. Maybe they ought to slow down some.
BTW, forgive my amateur tone. My knowledge of hockey analytics is almost nil. My qualifications for being in this discussion are: Being Canadian, and that's all.
Some xG models attempt to account for rush shots, but the way that’s do that depends on there being another event (hit, faceoff, shot, etc) outside the offensive zone ~3 seconds before the next shot. Evolving-Hockey’s model, which I use, just has predictors for distance from last event and time since last event.
One-timers aren’t accounted for because the NHL doesn’t track passes at all, except to say who assisted on a goal (no time stamp for the assist though).
I think there’s are probably some confounding factors around expected role for depth players playing a role, e.g., don’t do anything too fancy, lol.
I wish I had a more substantive comment, because this is deep stuff. I wish I could add anything. It does seem weird to me that having faster players on the ice would result in worse shots.
What is your interpretation? Is it as simple as faster players tend to be worse, therefore writing the reduction in shot quality off as more Omitted Variable Bias than anything, in the absence of a variable for shooting talent? Or is it something deeper than that?
Thanks for the question! I’m not sure I have a good answer for that. My best guesses are:
1. The shots are more dangerous than xG can capture because current xG models can only infer player movement from prior events
2. More frenzied skating pace results in poorer decision-making in shot selection
It could be something else I’m not thinking of though.
That's true. Can an xG model differentiate the different types of shots (rush shot, one time, etc.)? I never even thought of that. I do suppose your second point could perhaps explain why it feels like all the fast players are bottom six. Maybe they ought to slow down some.
BTW, forgive my amateur tone. My knowledge of hockey analytics is almost nil. My qualifications for being in this discussion are: Being Canadian, and that's all.
No problem! I’m happy to explain!
Some xG models attempt to account for rush shots, but the way that’s do that depends on there being another event (hit, faceoff, shot, etc) outside the offensive zone ~3 seconds before the next shot. Evolving-Hockey’s model, which I use, just has predictors for distance from last event and time since last event.
One-timers aren’t accounted for because the NHL doesn’t track passes at all, except to say who assisted on a goal (no time stamp for the assist though).
I think there’s are probably some confounding factors around expected role for depth players playing a role, e.g., don’t do anything too fancy, lol.
Should I continue commenting here if I'd like to ask you some general hockey questions, or should I DM you or something?
Either is fine, DMs might be a bit easier.