News Article: Dellow: Jets' record is great but long-term success requires better shot generation

[Ducky10]

My eye test is telling me that the Jets are playing mostly defensively-enabled hockey, while waiting on and capitalizing on mistakes generated by the other team. They're trying to play behind the puck almost exclusively. It's leading to the lopsided possession numbers we're seeing.

Yes and no, I think his videos show great examples of very big gaps between the D and the forwards on face off breakouts. I've seen them do this continually in games and Maurice has mentioned their spacing numerous times.

We're talking face offs here by the way, I do agree that they have a tendency to collapse too low when teams gain the zone, which hampers their transition breakout, this happens to most teams in every game at various stages though. I think their biggest problem possession wise comes from not being able to gain the other teams zone with control often enough.

 

[garret9]

This article looked at shots. Shots are an issue for the Jets because they allow too many and create too few. I can look at xGoals to show why this *is* a red flag for the Jets despite their record in two ways.

1) Shot Quality is fleeting
Corsi is shot quantity only.
Corsica xGoals (v1.o) uses shot quality (shot location and situational factors like rush shots and rebounds) added to shot quantity.
Dawson's xGoals used all this but also added *regressed* finishing talent.
In terms of predictive relationship to goals, it went Dawson > Corsi > Corsica. This shows you that teams struggle to consistently perform well in the shot quality measures without the inclusion of *regressed* finishing talent.
The fact that Dawson needed to regress finishing talent to improve his model shows you that sh% is fleeting.

2) Jets even if elite finishing team are over performing
dFenSh% is the gap between a the team's shooting percentage on unblocked shots and their expected given shot quality factors.
Jets have a dFenSh% of 1.8 on Corsica 2.0.
Finishing 1.8 shots per 100 more than expected is an unrealistic finish for the end of the season, here are the last 4 full seasons:

 

[JetsFan815]

This is not a surprise. Many of us have been pointing out the Jets struggles at shot generation for weeks. The response to such posts often ends up being "something something negativity". :

Like I posted in another thread the other day, before I start watching every Jets game I wonder if this is the one that will start the regression.

BTW, his point about Jets being really bad relative to league average after d-men change is not supposed to imply that "bad changes" are a cause of bad shot numbers in these situations. He is just using the d-men change as an indicator of a on the fly line change and using that to bucket Jets shot numbers in OTF, DZF, OZF etc type situations. The line changes are unlikely to be a problem, it's what they do after the line changes that is a problem.

I also don't buy this argument that "Jets are passing up on shots and that explains their shot numbers". That doesn't lineup with my eyetest. The Jets are generating fewer shots than the league due them not being in a position to generate a shot from what I see. Based on an article I read I have an idea on how I can get the numbers to backup to reject that, waiting on @ShutdownLine to track a few more Jets games before I get into that

I admittedly did not. No subscription, maybe after Xmas.

Am I wrong in saying that the article was only speaking of corsi%?

Personally, I've found some of Dellows stuff (that im able to see) to be lazily written.

It would be considered reasonable to actually read someone's work before declaring it "lazily written".

 

[Guffman]

I thought this was interesting on how Schiefele reduced his shots in favour of quality, which improved his shooting %.



Hopefully, that’s not a fleeting metric for Schefele!

@garret9 - sorry, what is regressed finishing talent mean?

 

[Aavco Cup]

This is not a surprise. Many of us have been pointing out the Jets struggles at shot generation for weeks. The response to such posts often ends up being "something something negativity". :

Like I posted in another thread the other day, before I start watching every Jets game I wonder if this is the one that will start the regression.

BTW, his point about Jets being really bad relative to league average after d-men change is not supposed to imply that "bad changes" are a cause of bad shot numbers in these situations. He is just using the d-men change as an indicator of a on the fly line change and using that to bucket Jets shot numbers in OTF, DZF, OZF etc type situations. The line changes are unlikely to be a problem, it's what they do after the line changes that is a problem.

I also don't buy this argument that "Jets are passing up on shots and that explains their shot numbers". That doesn't lineup with my eyetest. The Jets are generating fewer shots than the league due them not being in a position to generate a shot from what I see. Based on an article I read I have an idea on how I can get the numbers to backup to reject that, waiting on @ShutdownLine to track a few more Jets games before I get into that


It would be considered reasonable to actually read someone's work before declaring it "lazily written".

Dellow has been using this OTF CF% for D men all season. The theory is that D will usually only change on the fly when their own team has the puck or the puck is dumped in deep in the other team zone. This is usually a recipe for higher than normal CF. The D coming on the ice contribute very little to the play for the first 10 seconds after the change. Some D around the league are getting their CF% artificially inflated by mostly getting OTF shifts.

The question to ask about the Jets in this area. Why do we not get the same CF bump for the first 10 seconds after a D change OTF? Are we changing at inappropriate times? Or is something else at play?

It's also why I postulated this one stat OTF CF% is likely why our second periods look so bad relative to the other two. Changing OTF is much harder for the D in that period.

 

[Channelcat]

Helly is hiding a lot of flaws, that's no secret. Here's hoping it continues. Not sure if we're allowed to go macro, but we're 3rd last in SA

 

[Whileee]

Obviously, the Jets need to really tighten up aspects of their play, but I think that the lack of adjustment for game situation is a serious flaw in many of the analyses.

Amount of time leading really skews a lot of things, and I've noted before that I think the Jets' style of play is to cycle a lot rather than shoot a lot, and their skilled players in particular don't take a lot of low-percentage shots. They probably should shoot more.

Shooting percentage is highly affected by score, because teams that are trailing take more shots and more low-percentage shots. Teams that are leading get more high-quality chances, and have a higher shooting percentage.

An old analysis from Cam Charron... Score effects and you - NHL Numbers

Score Situation, Shot Share, Shooting%
Down 2, 55.5%, 7.6%
Down 1, 53.3%, 7.7%
Tied, 50.0%, 7.5%
Up 1, 46.7%, 8.4%
Up 2, 44.5%, 9.2%

And this article gives a wider analysis of score effects (hint: talented shooters have a shooting percentage around 9-10% when their team is leading)... Examining The Drivers of Score Effects I: Shooting and Save Percentage

Jets' distribution in time (5v5) by score...

Down 2 or more, 10.6%
Down 1, 5.7%
Tied, 35%
Up 1, 21.6%
Up 2 or more, 27.3%

By the way, it's not just shooting percentage that is skewed by score effects, it's also save percentage (obviously converse to sh%).

We should expect the Jets to have a higher than average save percentage, because they have spent about 84% of their 5v5 time tied or in the lead.

 

[JetsFan815]

Dellow has been using this OTF CF% for D men all season. The theory is that D will usually only change on the fly when their own team has the puck or the puck is dumped in deep in the other team zone. This is usually a recipe for higher than normal CF. The D coming on the ice contribute very little to the play for the first 10 seconds after the change. Some D around the league are getting their CF% artificially inflated by mostly getting OTF shifts.

The question to ask about the Jets in this area. Why do we not get the same CF bump for the first 10 seconds after a D change OTF? Are we changing at inappropriate times? Or is something else at play?

It's also why I postulated this one stat OTF CF% is likely why our second periods look so bad relative to the other two. Changing OTF is much harder for the D in that period.

I don't see the "bad line changes". My guess on why the Jets CF% is so much lower than the league on OTF shifts is that it's probably a combination of them making more "dump and change" and "chip and change" type of line changes as opposed to one/two of your players having control of the puck when the change happens and also the Jets general malaise at shot generation.

Line changes tend to happen in 3 ways- 1) Dump and change 2) Player takes the puck behind his own net or deep in his zone and waits for everyone to change 3) Players have control in the offensive zone and change on the fly and maintain pressure. Dellow already showed in this article how the Jets are so bad at generating shots after a defensive zone win, it seems they have trouble breaking out, getting through the neutral zone and taking the offensive zone with ease, those same problems would apply to 2) above as well. That combined with overdoing 1) and not doing enough of 3) could explain why they are so bad relative to league in CF% during OTF shifts

 

[Whileee]

Helly is hiding a lot of flaws, that's no secret. Here's hoping it continues. Not sure if we're allowed to go macro, but we're 3 last in SA

Hellebuyck is also benefiting from his team playing with the lead, and seldom trailing... (see above).

 

[Whileee]

This article looked at shots. Shots are an issue for the Jets because they allow too many and create too few. I can look at xGoals to show why this *is* a red flag for the Jets despite their record in two ways.

1) Shot Quality is fleeting
Corsi is shot quantity only.
Corsica xGoals (v1.o) uses shot quality (shot location and situational factors like rush shots and rebounds) added to shot quantity.
Dawson's xGoals used all this but also added *regressed* finishing talent.
In terms of predictive relationship to goals, it went Dawson > Corsi > Corsica. This shows you that teams struggle to consistently perform well in the shot quality measures without the inclusion of *regressed* finishing talent.
The fact that Dawson needed to regress finishing talent to improve his model shows you that sh% is fleeting.

2) Jets even if elite finishing team are over performing
dFenSh% is the gap between a the team's shooting percentage on unblocked shots and their expected given shot quality factors.
Jets have a dFenSh% of 1.8 on Corsica 2.0.
Finishing 1.8 shots per 100 more than expected is an unrealistic finish for the end of the season, here are the last 4 full seasons:

Is that adjusted for game state (score effects)?

 

[garret9]

I thought this was interesting on how Schiefele reduced his shots in favour of quality, which improved his shooting %.



Hopefully, that’s not a fleeting metric for Schefele!

@garret9 - sorry, what is regressed finishing talent mean?

No worries. I'll do a fake example with numbers that are easy for me to work with hahah.

Example:
Player Bob enters a league with a 10% shooting percentage.
Player Bob scores with a 20% shooting percentage with 100 shots.

Odds are Bob is probably a better finisher than the average player (although there is still a chance he is actually worse than average).
Odds are also that Bob is not that extremely better than average (although there is still a chance he is).

The most likely truth is Bob is somewhere in between 10 and 20. Regressed shooting is just a methodology to best estimate future performance knowing that extremes are unlikely and most as samples expand the outcomes tend to regress towards a mean (ex: graph).

Here's two ways you can regress shooting percentage for Bob...

Method One:
Let's say history shows us that we should regress to 80% of the mean at 100 shots. That means the most likely (but not only) true talent value for Bob is 12%.
(Maths: 10 is average. Bob is 20. Difference between two is +10. 20% of +10 is +2. 10+2=12.)
The issue with this method is you'd have to know the amount to regress by at each sample size.

Method Two
Another way to do it is just assuming league average for remainder of shots. Let's say we know that 500 shots are what we need for 95% confidence. We would then assume a 12% shooting percentage (this number being the same is actually accidental... I'm making this example up as I go).
(Maths: 20% sh% on 100 shots = 20 goals. 10% on 400 shots is 40 goals. That's 60 goals in 500 shots, or 12% sh%.)

Hope this is helpful.


EDIT:
As an aside, there is one problem I have with this video, which I brought up to Mike Kelly and he didn't get it.
"Luck" and unsustainability doesn't just mean random bounces.
Players garnering more high quality chances per shot one season than previous doesn't mean he necessarily will again.
(That and I hate the binning continuous variables with things like shot quality)

 

[Aavco Cup]

I don't see the "bad line changes". My guess on why the Jets CF% is so much lower than the league on OTF shifts is that it's probably a combination of them making more "dump and change" and "chip and change" type of line changes as opposed to one two of your players having control of the puck when the change happens and also the Jets general malaise at shot generation.

Line changes tend to happen in 3 ways- 1) Dump and change 2) Player takes the puck behind his own net or deep in his zone and waits for everyone to change 3) Players have control in the offensive zone and change on the fly and maintain pressure. Dellow already showed in this article how the Jets are so bad at generating shots after a defensive zone win, it seems they have trouble breaking out and taking the offensive zone with ease, those same problems would apply to 2) above as well. That combined with overdoing 1) and not doing enough of 3) could explain why they are so bad relative to league in CF% during OTF shifts

They also have had a bunch of exhausted OTF changes. Not to say your theory is wrong but it would be interesting to find out why we are so much of an outlier. Data based not eye test based.

 

[Jimby]

The sample size is also quite small and easily skewed. In that one game where the Jets whipped the Penguins the adjusted shooting percentage was a whopping 32.7%. If you remove that game suddenly their shooting percentage is a more sustainable number - 8.39% (adjusted).

Obviously, the Jets need to really tighten up aspects of their play, but I think that the lack of adjustment for game situation is a serious flaw in many of the analyses.

Amount of time leading really skews a lot of things, and I've noted before that I think the Jets' style of play is to cycle a lot rather than shoot a lot, and their skilled players in particular don't take a lot of low-percentage shots. They probably should shoot more.

Shooting percentage is highly affected by score, because teams that are trailing take more shots and more low-percentage shots. Teams that are leading get more high-quality chances, and have a higher shooting percentage.


We should expect the Jets to have a higher than average save percentage, because they have spent about 84% of their 5v5 time tied or in the lead.

 

[garret9]

Is that adjusted for game state (score effects)?

No, and for a reason. We're talking about "should the Jets expect to score as many goals as they have per shot" not about "should the Jets expect to score as many goals adjusted for score effects as they have per shot adjusted for score effects."

That said, score adjustment actually makes it look MORE unstable. Jets dFSh% increases to +2.05.

 

[Mortimer Snerd]

The article is quite detailed and includes a lot of video to illustrate the problems being talked about at 5v5. The Jets raw Corsi numbers are really bad, so it's at least a good starting point to say "what's going on here?"

"They get outshot by about five shots per 60 minutes of play at 5-on-5. Only a 9.8 percent shooting percentage and a .927 save percentage is keeping them above water."

"The Jets are currently outscoring opponents 28-25 at 5-on-5. If they'd shot, say, eight per cent — still excellent — they'd have been outscored 25-23. If they're worried about what happens going forward rather than what happened in the past, which they should be, and eight percent is a more likely shooting percentage than 9.8 percent, then they've got a problem they need to fix. Shooting percentage is very unlikely to save you when the other team is getting five more shots per 60 minutes of 5-on-5 play."

"It's still fairly early in the season. And wins buy you time. Leading a lot will suppress your possession numbers (although the Jets aren't great regardless of whether they're leading, tied or trailing). We've seen enough that Winnipeg's inability to generate shot attempts is a concerning trend though. When the shooting percentage cools off — and it will — you need something else to keep you in the game at 5-on-5. Right now, Winnipeg doesn't seem able to generate enough shots to have that."

Two thoughts provoked by the bolded bits.
1) Does Maurice make the mistake of looking at what has happened too much instead of thinking about what likely will happen? Goes to use of vets, use of goalies, not making changes while winning, etc. Just a question.
2) If the Jets low shot attempts is a result of striving for more high quality shots and they are succeeding at that then there is no reason to expect a drop in SH%. At least not necessarily, unless that effort is, in some way unsustainable.

 

[garret9]

PS:
I've been active a lot lately but I'll disappear soon again.
Here is a good resource for finding research: Publications

 

[Whileee]

No, and for a reason. We're talking about "should the Jets expect to score as many goals as they have per shot" not about "should the Jets expect to score as many goals adjusted for score effects as they have per shot adjusted for score effects."

That said, score adjustment actually makes it look MORE unstable. Jets dFSh% increases to +2.05.

Thanks.

I expect that these models assume too much with respect to the accuracy and precision in measuring shot quality.

 

[Whileee]

For future reference... (though it will soon become out of date)

That is highly skewed....

Note, labeling should be Down 2 or more, and Up 2 or more...

 

[garret9]

Thanks.

I expect that these models assume too much with respect to the accuracy and precision in measuring shot quality.

If they were more accurate and precise (albeit I view the former not as much of an issue as the latter), do you think the Jets would be more or less than +2.05 dFenSh%?

My guess is about the same.

 

[garret9]

Obviously, the Jets need to really tighten up aspects of their play, but I think that the lack of adjustment for game situation is a serious flaw in many of the analyses.

Amount of time leading really skews a lot of things, and I've noted before that I think the Jets' style of play is to cycle a lot rather than shoot a lot, and their skilled players in particular don't take a lot of low-percentage shots. They probably should shoot more.

Shooting percentage is highly affected by score, because teams that are trailing take more shots and more low-percentage shots. Teams that are leading get more high-quality chances, and have a higher shooting percentage.

An old analysis from Cam Charron... Score effects and you - NHL Numbers

Score Situation, Shot Share, Shooting%
Down 2, 55.5%, 7.6%
Down 1, 53.3%, 7.7%
Tied, 50.0%, 7.5%
Up 1, 46.7%, 8.4%
Up 2, 44.5%, 9.2%

And this article gives a wider analysis of score effects (hint: talented shooters have a shooting percentage around 9-10% when their team is leading)... Examining The Drivers of Score Effects I: Shooting and Save Percentage

Jets' distribution in time (5v5) by score...

Down 2 or more, 10.6%
Down 1, 5.7%
Tied, 35%
Up 1, 21.6%
Up 2 or more, 27.3%

By the way, it's not just shooting percentage that is skewed by score effects, it's also save percentage (obviously converse to sh%).

We should expect the Jets to have a higher than average save percentage, because they have spent about 84% of their 5v5 time tied or in the lead.

Whileee... I know this doesn't change the results... but why are you using a line graph :|
This is like a stats 101 faux pas! EDIT: It wasn't you!

On a more serious note:

Jets leading by one: 38% Corsi (28th), 136 mins
Jets tied: 50% (17th), 221 mins
Jets down by one: 45% (30th), 36 mins

Jets leading: 41% (24th), 310 mins
Jets trailing: 48% (31st), 103 mins​

Jets are doing poorly in shot differentials overall, despite the score. I don't think we can account things as being predominately driven by score effects.

I mean, even qualitatively we've had multiple post-games where people are talking about Hellebuyck stealing games.

 

[Guffman]

No worries. I'll do a fake example with numbers that are easy for me to work with hahah.

Example:
Player Bob enters a league with a 10% shooting percentage.
Player Bob scores with a 20% shooting percentage with 100 shots.

Odds are Bob is probably a better finisher than the average player (although there is still a chance he is actually worse than average).
Odds are also that Bob is not that extremely better than average (although there is still a chance he is).

The most likely truth is Bob is somewhere in between 10 and 20. Regressed shooting is just a methodology to best estimate future performance knowing that extremes are unlikely and most as samples expand the outcomes tend to regress towards a mean (ex: graph).

Here's two ways you can regress shooting percentage for Bob...

Method One:
Let's say history shows us that we should regress to 80% of the mean at 100 shots. That means the most likely (but not only) true talent value for Bob is 12%.
(Maths: 10 is average. Bob is 20. Difference between two is +10. 20% of +10 is +2. 10+2=12.)
The issue with this method is you'd have to know the amount to regress by at each sample size.

Method Two
Another way to do it is just assuming league average for remainder of shots. Let's say we know that 500 shots are what we need for 95% confidence. We would then assume a 12% shooting percentage (this number being the same is actually accidental... I'm making this example up as I go).
(Maths: 20% sh% on 100 shots = 20 goals. 10% on 400 shots is 40 goals. That's 60 goals in 500 shots, or 12% sh%.)

Hope this is helpful.


EDIT:
As an aside, there is one problem I have with this video, which I brought up to Mike Kelly and he didn't get it.
"Luck" and unsustainability doesn't just mean random bounces.
Players garnering more high quality chances per shot one season than previous doesn't mean he necessarily will again.
(That and I hate the binning continuous variables with things like shot quality)

Yes, it’s helpful but what if Player Bob is someone who is more apt to shoot from more desireable locations? It is reasonable to expect that he would

No worries. I'll do a fake example with numbers that are easy for me to work with hahah.

Example:
Player Bob enters a league with a 10% shooting percentage.
Player Bob scores with a 20% shooting percentage with 100 shots.

Odds are Bob is probably a better finisher than the average player (although there is still a chance he is actually worse than average).
Odds are also that Bob is not that extremely better than average (although there is still a chance he is).

The most likely truth is Bob is somewhere in between 10 and 20. Regressed shooting is just a methodology to best estimate future performance knowing that extremes are unlikely and most as samples expand the outcomes tend to regress towards a mean (ex: graph).

Here's two ways you can regress shooting percentage for Bob...

Method One:
Let's say history shows us that we should regress to 80% of the mean at 100 shots. That means the most likely (but not only) true talent value for Bob is 12%.
(Maths: 10 is average. Bob is 20. Difference between two is +10. 20% of +10 is +2. 10+2=12.)
The issue with this method is you'd have to know the amount to regress by at each sample size.

Method Two
Another way to do it is just assuming league average for remainder of shots. Let's say we know that 500 shots are what we need for 95% confidence. We would then assume a 12% shooting percentage (this number being the same is actually accidental... I'm making this example up as I go).
(Maths: 20% sh% on 100 shots = 20 goals. 10% on 400 shots is 40 goals. That's 60 goals in 500 shots, or 12% sh%.)

Hope this is helpful.


EDIT:
As an aside, there is one problem I have with this video, which I brought up to Mike Kelly and he didn't get it.
"Luck" and unsustainability doesn't just mean random bounces.
Players garnering more high quality chances per shot one season than previous doesn't mean he necessarily will again.
(That and I hate the binning continuous variables with things like shot quality)

Is it more likely than not that Player Bob’s shooting percentage will regress towards an average based on limited metrics for Bob? Does this account for how shooting percentages for players of his position are distributed to determine this regression?

It sounds like an interesting theory and plausible if someone is recording results near the extremes of the distribution.

 

[Whileee]

No, and for a reason. We're talking about "should the Jets expect to score as many goals as they have per shot" not about "should the Jets expect to score as many goals adjusted for score effects as they have per shot adjusted for score effects."

That said, score adjustment actually makes it look MORE unstable. Jets dFSh% increases to +2.05.

How did you do the score adjustment?

 

[Whileee]

Whileee... I know this doesn't change the results... but why are you using a line graph :|
This is like a stats 101 faux pas!

On a more serious note:

Jets leading by one: 38% Corsi (28th), 136 mins
Jets tied: 50% (17th), 221 mins
Jets down by one: 45% (30th), 36 mins

Jets leading: 41% (24th), 310 mins
Jets trailing: 48% (31st), 103 mins​

Jets are doing poorly in shot differentials overall, despite the score. I don't think we can account things as being predominately driven by score effects.

I mean, even qualitatively we've had multiple post-games where people are talking about Hellebuyck stealing games.

I agree about the graph (painful). It's not my graph... that should have been obvious from the context. I cited the source, too.

I've commented on the Jets' style before. I think they tend to be conservative in shot selection (which is different from teams like the Canes). I've watched every shift this season, and I don't see them being hammered as much as the Corsi makes them look.

As I've noted in many post-game posts, I think the notion that Hellebuyck is "stealing" games has been greatly exaggerated. He's been solid and hasn't given up many second chances, but he's actually faced a lot of shots that he should save, and not that many that required him to be heroic.

 

[garret9]

Yes, it’s helpful but what if Player Bob is someone who is more apt to shoot from more desireable locations? It is reasonable to expect that he would

It is suggesting Bob is likely a player who is more apt to shoot from more desirable locations (and/or finish better). Otherwise his true talent would be expected to regress 100% to the mean and we'd just say his true talent was 10%.

Is it more likely than not that Player Bob’s shooting percentage will regress towards an average based on limited metrics for Bob? Does this account for how shooting percentages for players of his position are distributed to determine this regression?

It sounds like an interesting theory and plausible if someone is recording results near the extremes of the distribution.

1. Yes, it is more likely than not for Bob.
2. I used league shooting percentage, but using position would be superior, but the methodology does not change. If Bob was a forward and in a league where forwards tend to shoot at 10% with the average player regressing by 80%, then the same method would show Bob to be 12% most likely a finisher.
3. It's applicable not just to the extrema but the entire distribution range of talents. You could look at it as saying: "We expect many people to be different than what their true talent level is simply because of sample size, variance, and outliers. The better you do in a short sample, the more likely it is you are above average, but doesn't mean you are more likely at that spot."
4. The Jets ARE an extreme to the distribution. They currently have scored 1.8 goals per 100 shots more than what we expect given where their shots are from. That's three standard deviations above the mean.
   
   

 

[garret9]

How did you do the score adjustment?

Similar to this:
Adjusted Possession Measures
And this:
Score-Adjusted PDO - NHL Numbers

I agree about the graph (painful). It's not my graph... that should have been obvious from the context. I cited the source, too.

I've commented on the Jets' style before. I think they tend to be conservative in shot selection (which is different from teams like the Canes). I've watched every shift this season, and I don't see them being hammered as much as the Corsi makes them look.

As I've noted in many post-game posts, I think the notion that Hellebuyck is "stealing" games has been greatly exaggerated. He's been solid and hasn't given up many second chances, but he's actually faced a lot of shots that he should save, and not that many that required him to be heroic.

Ya, I edited that but I was too late.

"The things Colorado does well results in a 106.7 PDO, in my opinion."

I have similar qualitative observations in Jets shot selection, but I do not see it as causing a successful team perform better than their shot differentials in a sustainable manner.

As to the Hellebuyck thing, I think a bit of this has to do with "big and boring" and also how goals both impact heuristics, IMO.

Do you really think Jets will finish this season with being 3 st dev above the typical season mean in dFenSh%?