Player Discussion: Winnipeg Jets Defense

[GumbyCan2]

1. I just think of it this way, Jets waiving Beaulieu or putting him in the pressbox and never letting him on the ice helps their on ice product by similar amount.

Keeping him on the roster though does help:
* expansion draft
* guy that some teammates like
* veteran presence bs

I believe there is only 6 days left that a current roster player with an expiring contract has to be resigned for next season, in order to qualify as ED eligible.
Beaulieu is Not ED keepable unless, or until signed for next season. This "point" has zero merit until he is signed through the 21/22 season.
I think Boo boo's contribution is roster filler, tough guy, enforcer if needed. PK time is basically a time filler and the fact he is tough on opponents in our zone. Saves Morrissey and Pionk from having to get over-burned out and beat-up on PK, or injured from array of shots bombarding them.

 

[Whileee]

I believe there is only 6 days left that a current roster player with an expiring contract has to be resigned for next season, in order to qualify as ED eligible.
Beaulieu is Not ED keepable unless, or until signed for next season. This "point" has zero merit until he is signed through the 21/22 season.
I think Boo boo's contribution is roster filler, tough guy, enforcer if needed. PK time is basically a time filler and the fact he is tough on opponents in our zone. Saves Morrissey and Pionk from having to get over-burned out and beat-up on PK, or injured from array of shots bombarding them.

Beaulieu is already signed for 2021/22. This was likely specifically to make him eligible as expansion draft fodder.

 

[garret9]

Two public models. Note the results are not realtive to ice time but overall impact.

Let me know any thoughts or questions.

 

[DRW204]

id generally chalk this up as merely "coach-speak" or maurice protecting his guy

however - his action of continuously trotting him out as a top-pair Dman says enough of what maurice thinks of his play

 

[Adam da bomb]

Two public models. Note the results are not realtive to ice time but overall impact.

Let me know any thoughts or questions.

So are those models suggesting Demelo is having a better season than Pionk?

 

[Imcanadianeh]

Two public models. Note the results are not realtive to ice time but overall impact.

Let me know any thoughts or questions.

This is the problem with advanced stats and different models, look at the huge difference between the two for Demelo, Pionk and Poolman

 

[puck stoppa]

Two public models. Note the results are not realtive to ice time but overall impact.

Let me know any thoughts or questions.

This matches my eye test. When Maurice keeps saying they have their own advanced stats whats he looking at?

 

[WolfHouse]

Two public models. Note the results are not realtive to ice time but overall impact.

Let me know any thoughts or questions.

I suppose this would be skewed by matchups but its tie for DeMelo to be back as top pairing

 

[Calcutta]

Can anyone provide some insights into our D prospects past our two studs Samberg and Ville? Samberg and Ville I see as top 4 defenders (Samberg in the Pesce mould and Ville as a Heiskanen-lite). Also, I see Stanley as a full time NHLer soon. But what do the others project as?

Here’s my take:



Tyrel Bauer: Who knows. People were grilling this kid on draft day, but I like his mean streak and he’s worth a flyer late in the draft. The rest of his junior career will determine whether he makes the AHL or not, and then we’ll go from there. Verdict: Project.


How off-base am I?

I don't know much about stats and skating but I know a fair bit about getting punched in the face. My god I wouldnt want to fight that lad. Punches like a jackhammer. A young man who can fight like that, on skates, and still do a decent job at hockeying is worth his weight in gold.

 

[Romang67]

This matches my eye test. When Maurice keeps saying they have their own advanced stats whats he looking at?

 

[puck stoppa]

That’s what I thought!

 

[WaveRaven]

I'm starting to wonder if we should be looking for another 2-3 type dman and try to solidify the group. Nobody is going to give us a #1 and we could get the d a bit more solid for now.

It would give the kids a chance to grow which it seems the Jets believe they need.

We are fine upfront.

 

[MardyBum]

Can't do much till after the expansion draft, which is why guys like Samberg, Heinola and Stanley should be getting ice time to see what we have going in to next year.

Otherwise we'll be back in "but we can't ice two rookies in our d-core!" next year.

Also Demelo probably needs to go play with Morrissey again.

 

[Adam da bomb]

So is the narrative demelo is as good as Pionk or our best D?

 

[garret9]

So are those models suggesting Demelo is having a better season than Pionk?

One is, one isn't... and that's 5v5 performance only.

 

[garret9]

This is the problem with advanced stats and different models, look at the huge difference between the two for Demelo, Pionk and Poolman

Well if you understand how the models work it no longer becomes a problem but just about understanding how the models work and then employing nuance.

Oversimplification:
EW starts everyone at 0 at the start of the season, and your performance pushes you up or down throughout the season.
HV uses the previous season as a prior, so you start somewhere between 0 and how you ended last year, and your performance pushes you up or down throughout the season.

There's pros and cons to both methodologies:
1) EW is more sensitive to players changing due to things like injuries, or large changes in usage, player development, etc.
2) HV is less sensetive to variance and error, so it's more confident on players true talent on average if the above changes don't exist. It also gets fooled less when you have extremes (sheltered rookie, a team with vastly unusual F to D skill differences, etc.) or smaller samples.
3) EW is better at telling what happned.
4) HV is better at telling how good a player is.

There are a few more but that's the start.


Strong and important note: Better at different situations does not mean always right in that situation. It means that one is right more often than the other.

Ex: Let's say we take 100 cases where they disagree big time on players with big changes like injuries, extremely different usage, etc. EW is likely a better model for discerning that difference due to those issues. So maybe EW is more right 60% of the time and HV is right 40% of the time.
(Made up numbers to display a point)

 

[garret9]

This matches my eye test. When Maurice keeps saying they have their own advanced stats whats he looking at?

Well, when Maurice says something... it doesn't necessarily mean it's true... but here's some things that could be going on.

This are some of many possibilies...

1) Maybe their models are intended to be more descriptive or more predective than these public models. Moving in one direction may cause you to lose ability in the other. Ex: if you have a more descriptive but less predictive model, you may make the wrong call in who deserves more ice time.

2) You may lose forest for trees. Microstatistics like puck tracking and such can distract people from the overall reason why you care about those things: overall impact. Ex: everyone knows this one but blocking shots is good, but having lots of blocked shots may be a symptom of doing other things poorly. You can have similar situations in puck tracking statistics.

3) Over dependence on what is rare. Ex: the royal road passes became a huge rage when SV started talking about it with his tracking stats. A shot from a royal road pass was WAAAAY more likely to score than one that was not. But these passes are rare, so over stressing on these may make you make mistakes. This is related but not the same as the last one.

4) I thought of another good example but forgot about it when I left my computer... I'll keep this up just in case hahah.

 

[garret9]

I suppose this would be skewed by matchups but its tie for DeMelo to be back as top pairing

It in essence accounts for the quality of match ups and linemates, that said it does not count for qualitative aspects of those variables, like chemistry.

Example:

I'm a good defensive defender but bad puck mover.
I change partners half way through the year.
The new partner (B) is not as good as my old partner (A).
Results as expected go downward.
Model uses change in those results (my new results vs old, my new partners results vs old (B), my old partner's results vs old (A), and also my new partner's old partner who is also my old partner's new partner (C)) to make inferences of how much to divvy out credit/blame.

Now... let's say Player A is a good puck mover, and Player B is a defensive defender but bad puck mover.
In theory, our results (B+me) together may actually be worse than one would predict as now which ever one of us has the role of moving the puck is worse than when it was A + me.

I talked about this here:
Behind the Numbers: Theory on Environmental Impacts and Chemistry

 

[Whileee]

Well if you understand how the models work it no longer becomes a problem but just about understanding how the models work and then employing nuance.

Oversimplification:
EW starts everyone at 0 at the start of the season, and your performance pushes you up or down throughout the season.
HV uses the previous season as a prior, so you start somewhere between 0 and how you ended last year, and your performance pushes you up or down throughout the season.

There's pros and cons to both methodologies:
1) EW is more sensitive to players changing due to things like injuries, or large changes in usage, player development, etc.
2) HV is less sensetive to variance and error, so it's more confident on players true talent on average if the above changes don't exist. It also gets fooled less when you have extremes (sheltered rookie, a team with vastly unusual F to D skill differences, etc.) or smaller samples.
3) EW is better at telling what happned.
4) HV is better at telling how good a player is.

There are a few more but that's the start.


Strong and important note: Better at different situations does not mean always right in that situation. It means that one is right more often than the other.

Ex: Let's say we take 100 cases where they disagree big time on players with big changes like injuries, extremely different usage, etc. EW is likely a better model for discerning that difference due to those issues. So maybe EW is more right 60% of the time and HV is right 40% of the time.
(Made up numbers to display a point)

One issue with using priors, especially from previous seasons, is that it makes a lot of assumptions about the stability (reliability) of shot metrics from season to season, context to context. On balance, I'd prefer to consider these metrics as descriptive, and focus more on segmenting by season, context, etc. I also have a tendency to focus on exploratory data analysis, including data visualization, before developing and using a multivariate model. Even then, I seldom trust complex models that try to do too much and end up with lots of stratification along with somewhat more simple models. I am very wary of machine learning unless I really understand the underlying data and have a strong conceptual framework. I've seen many fail badly in terms of face validity when the underlying causal mechanisms are known.

 

[Whileee]

It in essence accounts for the quality of match ups and linemates, that said it does not count for qualitative aspects of those variables, like chemistry.

Example:

I'm a good defensive defender but bad puck mover.
I change partners half way through the year.
The new partner (B) is not as good as my old partner (A).
Results as expected go downward.
Model uses change in those results (my new results vs old, my new partners results vs old (B), my old partner's results vs old (A), and also my new partner's old partner who is also my old partner's new partner (C)) to make inferences of how much to divvy out credit/blame.

Now... let's say Player A is a good puck mover, and Player B is a defensive defender but bad puck mover.
In theory, our results (B+me) together may actually be worse than one would predict as now which ever one of us has the role of moving the puck is worse than when it was A + me.

I talked about this here:
Behind the Numbers: Theory on Environmental Impacts and Chemistry

This is such an insightful and important post for anyone who wants to be a better consumer of data analyses.

One "pair" that I think had great "chemistry" was Scheifele and Byfuglien. Their shot metrics together were very good, and I think some of it was a good match of styles and strengths. Buff was uniquely good at stopping an opponent in his zone, turning the puck over, and moving it. Scheifele has been weak at that in his own zone, so playing with Buff essentially allowed him to play how he liked - looking for a transition / break opportunity, instead of digging in to help his D force a turnover.

That's just a theory, but I completely agree with Garret that there are some matches / pairings that work better than you might expect, and others that are surprisingly bad. Nuance / chemistry matters.

 

[garret9]

One issue with using priors, especially from previous seasons, is that it makes a lot of assumptions about the stability (reliability) of shot metrics from season to season, context to context. On balance, I'd prefer to consider these metrics as descriptive, and focus more on segmenting by season, context, etc. I also have a tendency to focus on exploratory data analysis, including data visualization, before developing and using a multivariate model. Even then, I seldom trust complex models that try to do too much and end up with lots of stratification along with somewhat more simple models. I am very wary of machine learning unless I really understand the underlying data and have a strong conceptual framework. I've seen many fail badly in terms of face validity when the underlying causal mechanisms are known.

No matter what you are making a prior and an assumption.
Just one is average and one is a mix between average and last year.
Is it weirder to assume everyone is average or someone is likely between last year and average?

That said I still think Dawson built the best public model ever, but he's in COL now.

 

[Teemusalami204]

All we have to do is get morrisey of the first pairing and first pp. get a player specifically to shelter morrisey and we will be a much better team for it

He’s almost in bogo territory

 

[Whileee]

No matter what you are making a prior and an assumption.
Just one is average and one is a mix between average and last year.
Is it weirder to assume everyone is average or someone is likely between last year and average?

That said I still think Dawson built the best public model ever, but he's in COL now.

Unless I have good reason to use a specific prior, I think it's better to go without. That way, you are more likely to be providing a descriptive set of metrics, rather than trying to assess and compare players based on disparate settings. Generally, I think we too often use these methods to try to assess and compare players simplistically, rather than digging into a more nuanced analysis. I don't it all the time, but I don't think it's a great idea for the most part.

I prefer an approach that you have been advocating - trying to assess what different players do well, and then trying to understand why, and how that might fit with other players or styles / systems.

 

[GumbyCan2]

3 rookie dmen? maurice would never.

What have you done with Morrissey, traded him to Calgary for Sean Monahan, or?
With these pairings, we actually might not miss Morrissey?

 

[GumbyCan2]

Beaulieu is already signed for 2021/22. This was likely specifically to make him eligible as expansion draft fodder.

Sorry, I guess i missed that? I haven't checked in with Cap Friendly W Jets site for a while.
Sorry for the irrelevancy in my "regency"? Or something like that.