A Baffling Expected Goal Measurement

[majormajor]

Have a look at this breakaway goal by Eric Robinson where he has a clear shot from the hashmarks. According to Moneypuck it is a 4.9% chance of scoring or .049 xG, which seems absurdly low to me.



Does their expected goal model not take into account breakaways?

And even if it is not a breakaway, doesn't a shot from that location have a better than 4.9% chance of going in?

Is shooting talent incorporated now? I could understand a low xG if shooting talent is a part of it, it is Eric Robinson after all, but I did not think that was involved.

Natural Stat Trick's xG is calculated a little bit differently, I can see the difference at the team level, but this Robinson shot is weighted similar to Moneypuck, about 5% chance of scoring.

Click to read more...

 

[Filthy Dangles]

Change in distance tracking has screwed up xG stats

 

[majormajor]

Change in distance tracking has screwed up xG stats

I recall that thread, I didn't realize this was that extremely off though.

Like where within 10 ft of the hashmarks do you have less than a 5% chance of scoring?

 

[Filthy Dangles]

This shot, uncontested, right at the top of the crease, one-timed after making a royal road pass, had an 8% of going in per Moneypuck. I think it's safe to say just throw out all xG metrics (assuming they all use the same data source as MoneyPuck). They simply do not reflect reality in any way. I feel like I check every goal and they are just way off...




8 / 100....

 

[tarheelhockey]

I think it's safe to say just throw out all xG metrics (assuming they all use the same data source as MoneyPuck).

Do they all use the same data?

If so, word needs to get around because I see xG metrics posted all the time and this is a pretty good reason to shut it down.

 

[Filthy Dangles]

Do they all use the same data?

If so, word needs to get around because I see xG metrics posted all the time and this is a pretty good reason to shut it down.

I don’t know for sure but I’m pretty sure they and everyone else who does XG get the coordinates from the official NHL play by play. Perhaps MP assigns the scoring probabilities differently though? I don’t know how much of it is an actual incorrect distance problem vs bad interpreting of the quality of the shot (assuming distance and coordinates are close)

seems like a bit of both...



E: There is where MP had the shot (presumably straight from NHL PbP) bottom hashmark instead of top of the crease. Still 8% is rather ridiculous from that spot on average, and that's not even beginning to account for the quality of that specific chance itself where the goalie was moving and the shooter had no one contesting the shot.

 

[Filthy Dangles]

In their ability to reflect the quality of chances? There's been some noise about this on the Number board, but feel like it's worthy of the mainboard.

Here's a good example. Moneypuck, one of the more noteable hockey analytics source, gave this shot an 8% chance to score. Insanity.

It was also incorrectly recorded the shot as taking place at the bottom hashmark instead of the top of the crease. Even with that error, 8% on average from that spot seems pretty crazy, not even accounting for how good that specific chance was.

 

[DownIsTheNewUp]

This is a good example of why without puck tracking and player tracking, most of the advanced stats publicly available are pretty useless. xGF doesn't know the puck crossed the slot leaving the goalie helpless.

 

[aufheben]

This is a good example of why without puck tracking and player tracking, most of the advanced stats publicly available are pretty useless. xGF doesn't know the puck crossed the slot leaving the goalie helpless.

Expected Goals aren’t a singular metric. Different sites track them differently.

MoneyPuck:

Variables In Shot Prediction Model:

1.) Shot Distance From Net
2.) Time Since Last Game Event
3.) Shot Type (Slap, Wrist, Backhand, etc)
4.) Speed From Previous Event
5.) Shot Angle
6.) East-West Location on Ice of Last Event Before the Shot
7.) If Rebound, difference in shot angle divided by time since last shot
8.) Last Event That Happened Before the Shot (Faceoff, Hit, etc)
9.) Other team's # of skaters on ice
10.) East-West Location on Ice of Shot
11.) Man Advantage Situation
12.) Time since current Powerplay started
13.) Distance From Previous Event
14.) North-South Location on Ice of Shot
15.) Shooting on Empty Net

 

[Filthy Dangles]

Expected Goals aren’t a singular metric. Different sites track them differently.

MoneyPuck:

Variables In Shot Prediction Model:

1.) Shot Distance From Net
2.) Time Since Last Game Event
3.) Shot Type (Slap, Wrist, Backhand, etc)
4.) Speed From Previous Event
5.) Shot Angle
6.) East-West Location on Ice of Last Event Before the Shot
7.) If Rebound, difference in shot angle divided by time since last shot
8.) Last Event That Happened Before the Shot (Faceoff, Hit, etc)
9.) Other team's # of skaters on ice
10.) East-West Location on Ice of Shot
11.) Man Advantage Situation
12.) Time since current Powerplay started
13.) Distance From Previous Event
14.) North-South Location on Ice of Shot
15.) Shooting on Empty Net

I guess they try their best with that they got but it's just not good enough. Things like if the shooter was bothered/contested and had traffic to deal with is ignored. And whether the goalie had to move because of a pass. Those are way more pertinent than most of the things on that list.

Do you know if there's a way to see what other sites had that specific Jarnkrok shot/event at? Like NaturalStatTrick or those RAPM Charts?

 

[Gurglesons]

Do you know if there's a way to see what other sites had that specific Jarnkrok shot/event at? Like NaturalStatTrick or those RAPM Charts?

Using specific singular events to discredit statistics makes me doubt your grasp on predictive analytics.

Most writers using xG statistics are the first to say they are limited in their scope and notoriously favor untalented teams. Thus why Washington consistently finishes badly in them, yet given their talent they often vastly outproduce them.

 

[aufheben]

Do you know if there's a way to see what other sites had that specific Jarnkrok shot/event at? Like NaturalStatTrick or those RAPM Charts?

Who took the shot and roughly what time was it?

 

[Filthy Dangles]

Using specific singular events to discredit statistics makes me doubt your grasp on predictive analytics.

Most writers using xG statistics are the first to say they are limited in their scope and notoriously favor untalented teams. Thus why Washington consistently finishes badly in them, yet given their talent they often vastly outproduce them.

This isn't based off one event or goal...You can see these threads for more info and examples.

Change in distance tracking has screwed up xG stats
A Baffling Expected Goal Measurement

Who took the shot and roughly what time was it?

Jarnkrok with 4:12 left in the 1st.

 

[Gurglesons]

This isn't based on one event or goal...You can see these threads for more info and examples.

Change in distance tracking has screwed up xG stats
A Baffling Expected Goal Measurement



Jarnkrok with 4:12 left in the 1st.

Well shame on me for not fully reading the thread. That is bizarre. Didn't this happen last season too? Was it with goaltending stats?

 

[Machinehead]

What's wrong with 8%?

That's a .920 save percentage.

At 5v5, .920 would place you about 45th in the league.

Even breakaways are stopped 75-85% of the time.

 

[aufheben]

This isn't based off one event or goal...You can see these threads for more info and examples.

Change in distance tracking has screwed up xG stats
A Baffling Expected Goal Measurement



Jarnkrok with 4:12 left in the 1st.

I don’t see this shot on MoneyPuck.

 

[Filthy Dangles]

What's wrong with 8%?

That's a .920 save percentage.

At 5v5, .920 would place you about 45th in the league.

Even breakaways are stopped 75-85% of the time.

What's wrong is that shot gets deposited into the net way more than 8/100. You think that's an average quality of shot? That's one of the best chances you can get in the league, arguably better than a breakaway which is 31% I belive.

 

[TheDoldrums]

NHL shot tracking is very bad. This makes xG not as reliable as it should be. I think it's possible that Corsi can still be more useful than xG in some situations. Both should probably be considered rather than treating one as gospel.

 

[Machinehead]

What's wrong is that shot gets deposited into the net way more than 8/100. You think that's an average quality of shot?

The average quality of shot at 5v5 is more like 2 or 3%.

I think you're overestimating how often any given shot goes in.

Unless it's something like a tap-in right on the line, 15% is about as good as a scoring chance gets. Most goalies run at about .850 on high-danger chances.

 

[Filthy Dangles]

I don’t see this shot on MoneyPuck.

MoneyPuck.com

 

[Machinehead]

I don't know where you got 31% on breakaways, but if your goalie is running at .690 (nice!) on any kind of shot, he better find a day job.

 

[aufheben]

MoneyPuck.com

Dude why would you post about a goal from a NSH-FLA game that happened yesterday when they played again tonight?

 

[Filthy Dangles]

I don't know where you got 31% on breakaways, but if your goalie is running at .690 (nice!) on any kind of shot, he better find a day job.

I was thinking of shootouts. League average Sh% on shootouts in 32%. Breakaways can't be that much lower...?

 

[aufheben]

Are you saying 8% was too high or too low?

Location doesn’t seem wildly inaccurate.

 

[Machinehead]

I was thinking of shootouts. League average Sh% on shootouts in 32%.

That makes more sense.

The thing is, 8% is about average for a shot on-goal. xG is measuring the likelihood of a shot attempt going in.

Half of attempts don't get to net. Teams miss the net 1,500-2,000 times a year.