This was a great read, thanks for sharing! Interesting that they highlighted the play-by-play becoming more accurate right around when I think the chip technology started being integrated.
I'm not a denizen of the hockey play by play world, so what's with NYR? Are they like the Buffalo Bills or the NE Patriots of the NFL, with play by play scorers that are a little out over their skis? Or are they just awful at their jobs LOL?
Also, I think I lost the hypothesis a little bit. Are you claiming shot locations haven't truly changed that much, as this is merely a data quality issue, or are you claiming that a real change is happening here, the scope of which is being confounded by a data quality issue? Whether either is true, why does that mean expected goals need to be era adjusted? Are you discounting the quality of the models based on these shot location issues?
Forgive me for being such a novice. I don't even know what the eGoal trend is right now.
Yeah, I'm not sure what the deal was with NYR but for a long time the average shot was like 5-10 feet closer to the net in Madison Square Garden than in any other arena. To be fair, it seems like there wasn't much of a home team bias, every team was closer to the net in MSG.
The idea is that each season there is some natural variation from changing playing styles plus random noise, but in 2022-23 there was a large difference from the previous year and in 2023-24 there was an even larger one. My theory is that the NHL changed how they record the shot location and shot outcome (on net/missed) by integrating or maybe even completely relying on the puck tracking technology.
Maybe I should have left out the bit about xG, because it's a little beside my main point. I'm not really discounting the quality of xG models, because they've generally been fairly accurate in the aggregate, but rather saying 10 goals above expected in 2021-22 isn't the same as scoring 10 goals above expected in 2023-24 because the baseline has changed. The trend the last couple of seasons has been for xG to underestimate the number of actual goals. Like how passing or rushing stats are sometimes adjusted to a common baseline to make comparisons decades apart.
I’d estimate about a 25/75 split between fundamental change and data collection method change based on what was typical in other seasons and what we’ve seen so far this season.
Great read! Related to this, there were a couple students in our summer program two years ago that started investigating drop in save percentages, wrote about it and linked to their work here https://open.substack.com/pub/statthinksportsanalytics/p/cmsacamp-2023?r=a5tci&utm_medium=ios
This was a great read, thanks for sharing! Interesting that they highlighted the play-by-play becoming more accurate right around when I think the chip technology started being integrated.
The NHL is tracking less data manually using humans and using more and more via chips in pucks. That’s a big part of it.
I'm not a denizen of the hockey play by play world, so what's with NYR? Are they like the Buffalo Bills or the NE Patriots of the NFL, with play by play scorers that are a little out over their skis? Or are they just awful at their jobs LOL?
Also, I think I lost the hypothesis a little bit. Are you claiming shot locations haven't truly changed that much, as this is merely a data quality issue, or are you claiming that a real change is happening here, the scope of which is being confounded by a data quality issue? Whether either is true, why does that mean expected goals need to be era adjusted? Are you discounting the quality of the models based on these shot location issues?
Forgive me for being such a novice. I don't even know what the eGoal trend is right now.
Yeah, I'm not sure what the deal was with NYR but for a long time the average shot was like 5-10 feet closer to the net in Madison Square Garden than in any other arena. To be fair, it seems like there wasn't much of a home team bias, every team was closer to the net in MSG.
The idea is that each season there is some natural variation from changing playing styles plus random noise, but in 2022-23 there was a large difference from the previous year and in 2023-24 there was an even larger one. My theory is that the NHL changed how they record the shot location and shot outcome (on net/missed) by integrating or maybe even completely relying on the puck tracking technology.
Maybe I should have left out the bit about xG, because it's a little beside my main point. I'm not really discounting the quality of xG models, because they've generally been fairly accurate in the aggregate, but rather saying 10 goals above expected in 2021-22 isn't the same as scoring 10 goals above expected in 2023-24 because the baseline has changed. The trend the last couple of seasons has been for xG to underestimate the number of actual goals. Like how passing or rushing stats are sometimes adjusted to a common baseline to make comparisons decades apart.
So do you think there was any fundamental change in the real on-ice play at all? Or is it impossible to tell with the data issues?
I’d estimate about a 25/75 split between fundamental change and data collection method change based on what was typical in other seasons and what we’ve seen so far this season.