Change in distance tracking has screwed up xG stats

[tarheelhockey]

And I say “screwed up” because it’s pretty obvious that the old data was much more accurate than 2019-20 data. It looks to me like the NHL is somehow adding phantom distance to events, which completely changes predictions based on distance and angle.

Click to read more...

 

[morehockeystats]

As I posted on twitter, only shot attempt data (BLOCK, MISS, SHOT, GOAL) seem to be affected. Coordinates for FAC, HIT, GIVE, TAKE remain intact. Likely, the NHL is using a software tracking system for shot attempts now (the one that was showcased during the ASG), and it's bugged.

 

[tarheelhockey]

As I posted on twitter, only shot attempt data (BLOCK, MISS, SHOT, GOAL) seem to be affected. Coordinates for FAC, HIT, GIVE, TAKE remain intact. Likely, the NHL is using a software tracking system for shot attempts now (the one that was showcased during the ASG), and it's bugged.

Not that I envy the person who has to do the legwork on all the games that are impacted, but in theory couldn't this be corrected with an algorithm applied to the existing data?

 

[morehockeystats]

Not that I envy the person who has to do the legwork on all the games that are impacted, but in theory couldn't this be corrected with an algorithm applied to the existing data?

You have to have manual tracking data set first...

I have an idea about what to do here and I hope I've time to do it today.

 

[Filthy Dangles]

As I posted on twitter, only shot attempt data (BLOCK, MISS, SHOT, GOAL) seem to be affected. Coordinates for FAC, HIT, GIVE, TAKE remain intact. Likely, the NHL is using a software tracking system for shot attempts now (the one that was showcased during the ASG), and it's bugged.

Well they were supposed to implement puck tracking this season.....i haven't heart anything about that though so it could be a coincidence

 

[18Hossa]

This thread explores it a bit more and also visualizes some changes throughout the years. TL;DR massive change this year.

 

[morehockeystats]

Just to quote a couple of my tweets from today:

 

[Filthy Dangles]

I was watching some highlights of games and wanted to compare the shot location on video with what was being recorded but......I feel dumb now but where in the play by play are the angle or co-ordinates of shots recorded? I only see the distance listed in PbP sheet.....

 

[morehockeystats]

It's not in PBP html. It's in the JSON.

sprintf("https://statsapi.web.nhl.com/api/v1/game/%s/feed/live", game_id);

 

[Filthy Dangles]

Has this been fixed?

This Phil Danault goal has the coordinates listed as (83.0, -7.0)....which if I understand correctly means he was 6 feet out from the goal line (goal line to red line is 89 feet) and 7 feet to the right of the slot line. Does that look right?

 

[morehockeystats]

Has this been fixed?

This Phil Danault goal has the coordinates listed as (83.0, -7.0)....which if I understand correctly means he was 6 feet out from the goal line (goal line to red line is 89 feet) and 7 feet to the right of the slot line. Does that look right?

Yes, quite so. You can't require more than +-1ft precision for each coordinate. Goalie crease curve is 6' in radius, and goalie crease parallel lines run 1' from each goalpost.

 

[Filthy Dangles]

Yes, quite so. You can't require more than +-1ft precision for each coordinate. Goalie crease curve is 6' in radius, and goalie crease parallel lines run 1' from each goalpost.

Yeah, I checked the William Karlsson goal right after that too which showed (-76, 11). Looks like he releases from somewhere between the top-hash mark and faceoff dot which should be more 20-22 feet than 13 feet. What do you think about that, normal variance in recording or part of same issue this thread was made for?

 

[morehockeystats]

Yeah, I checked the William Karlsson goal right after that too which showed (-76, 11). Looks like he releases from somewhere between the top-hash mark and faceoff dot which should be more 20-22 feet than 13 feet. What do you think about that, normal variance in recording or part of same issue this thread was made for?

This one looks fishier. Although one can argue that the puck had left the stick right past closer hashmark, it's still around -70 or -71.
You might want to tag @EvolvingWild on twitter with this episode. If you don't have twitter, or prefer to stay out of it, I'll do that on your behalf.

The bigger question is that if this is a one-time fluke - but who has the resources to verify?..

 

[Filthy Dangles]

This one looks fishier. Although one can argue that the puck had left the stick right past closer hashmark, it's still around -70 or -71.
You might want to tag @EvolvingWild on twitter with this episode. If you don't have twitter, or prefer to stay out of it, I'll do that on your behalf.

The bigger question is that if this is a one-time fluke - but who has the resources to verify?..

Yeah I'm not on Twitter. This is definitely making me skeptical of xG stats. Not diminishing them, I appreciated the work people put into them and their models but apart from things that can't be accounted for like goal mouth/slot line passes, there seems to be big time issues in the accurate recording.

 

[Filthy Dangles]

This shot from Gabby Bourque was purportedly taken 44 feet from the goal line, which again appears very inaccurate and was given only a 3.7% chance to score which is insane. I don't really know why I keep posting these, just writing down what I am observing, perhaps to refer to a later date but I am really losing faith in these xG metrics the more i dig into them.

I mean he did take a pass and walk into this shot in near full stride, which should be about 20MPH or for an NHL skater at nearly full speed. 20 MPH = 30 feet per second so you do gain ground fast in that instance, it can be hard to accurately record i guess.