The Advanced Stats Thread Episode VIII:

[silverfish]

Pionk is gonna be bad next year but it'll just be a 'sophomore slump'. Remember that stupid save that Hank made against Calgary? Look at Pionk on that play. Totally lost. Shit happens when you ride that 101.79 PDO with your -6.5 relCF% and -7.35 relxGF%

 

[Mac n Gs]

Catch up BB, I was saying that should be our defense for a month now

 

[Blue Blooded]

Issue I have with these pairings is the usage they'll get. Does usage matter? Not as much as we like to think. But I don't want Staal starting in the O-Zone, where I do want Shattenkirk starting. On the same token, while I have no problem with starting Shattenkirk in the D-zone, I don't think it's in his best interest to optimize his game. I might balance it more, something like:

Smith - Shattenkirk
Skjei - Pionk
Staal - DeAngelo
Claesson

Actually, nah, our right side is way too bad to put Smith on the left. Gorton can't be done... can he?

In a perfect world, Staal isn't on the team, and Claesson and Gilly battle for that spot, but here we are.

I also truly believe Smith will go from 'zero' to 'hero' on this board next year. He wasn't even that bad last season. Definitely not bad enough to be exiled to the AHL and a scapegoat.

In terms of discrepancies between the sites, not sure. Are you certain all the parameters are equal? Screenshots of examples?

Considering how Skjei-DeAngelo performed last year and Skjei-Clendening did the year before, I want that pairing to do the heavy lifting offensively. I think DeAngelo is a more skilled player than Shattenkirk and his offensive game is more suited to 5v5 hockey. Shattenkirk is more of a 2-way guy 5v5 and a top-3 PPQB.

Staal-Shattenkirk would be the primary defensive pairing and the third would fill in wherever.

If we bench Staal, which I'm personally very open to, put Smith with Shatty.

I don't want to see Staal with either of DeAngelo or Pionk (or off-side Smith for that matter).

 

[Blue Blooded]

For reference: DeAngelo's 5v5 P60 in the OHL was top-5 pick level, for a forward. (IIRC, I did that research last summer)

 

[Mac n Gs]

Deangelo is a shot assist wizard and the lack of goals seems like an anomaly, especially when you see how much helps drive shots and scoring chances, at an individual and team level.

He basically just needs his on-ice save% to regress to normal.

 

[Blue Blooded]

Deangelo is a shot assist wizard and the lack of goals seems like an anomaly, especially when you see how much helps drive shots and scoring chances, at an individual and team level.

He basically just needs his on-ice save% to regress to normal.

I've missed out on all these fancy new fan-tracked metrics since there isn't enough data to use in my prediction models. Do you have DeAngelo's data?

 

[Mac n Gs]

I've missed out on all these fancy new fan-tracked metrics since there isn't enough data to use in my prediction models. Do you have DeAngelo's data?

I only have percentiles, but I believe they’re available if you donate to Ryan Stimson and Corey Sznadjer on Patreon. Here are some good resouces out there on public tableau:

Tableau Public

Tableau Public

Tableau Public

 

[sbjnyc]

@Harbour Dog Miller was 3/28 on the season. My memory is stupid good. 1/22 with NYR 2/6 with TB.

Namestnikov on the other hand was 0/4 with NYR and 7/24 with TB.

All NYR unblocked shot attempts and goals <= 10' according to NHL PBP

MFW Vesey isn't even that good at the only thing he's good at

If these aren't small sample sizes I don't know what is so I'm not sure how much you can really glean from this. Suppose for example, the league average at 10' is 33%. Then with 32 attempts, a player should score between 8-13 goals 67% of the time or 5-16 goals 95% of the time. Can you really say Vesey can't score at a 33% clip? Even if the league average is 40% Vesey would still be within 2 standard deviations.

But even if they were reliable estimators, how dependent are they on the distance? For example, maybe at 9' Player A is great at 10' he's awful and at 11' he's pretty good. These sorts of artifacts can sometimes pop up (I'm cursed with them at work). You posted something earlier (I think) with more of a distribution by distance. I think that is something more suited to being modeled - ie a league average distribution and how individual players deviate from it.

 

[silverfish]

If these aren't small sample sizes I don't know what is so I'm not sure how much you can really glean from this. Suppose for example, the league average at 10' is 33%. Then with 32 attempts, a player should score between 8-13 goals 67% of the time or 5-16 goals 95% of the time. Can you really say Vesey can't score at a 33% clip? Even if the league average is 40% Vesey would still be within 2 standard deviations.

But even if they were reliable estimators, how dependent are they on the distance? For example, maybe at 9' Player A is great at 10' he's awful and at 11' he's pretty good. These sorts of artifacts can sometimes pop up (I'm cursed with them at work). You posted something earlier (I think) with more of a distribution by distance. I think that is something more suited to being modeled - ie a league average distribution and how individual players deviate from it.

So I feel like this post is a really good example about you drawing a conclusion from conclusions you think I'm drawing which I'm not. And I think this post is great at illustrating why we can't talk about advanced stats anywhere else on HFNYR because of this exact phenomena.

I can't post a thesis every time I post a metric. Just like every time someone posts "Cody Franson sucks because he can't skate", they don't post thousands of minutes of on-ice video showing that he can't skate. Except for some reason, the burden of proof is higher on the person providing the data rather than the opinion. Strange, isn't it?

If your beef was with my eye-roll gif, that is tongue in cheek. I felt that was obvious, but I guess not.

 

[sbjnyc]

So I feel like this post is a really good example about you drawing a conclusion from conclusions you think I'm drawing which I'm not. And I think this post is great at illustrating why we can't talk about advanced stats anywhere else on HFNYR because of this exact phenomena.

I can't post a thesis every time I post a metric. Just like every time someone posts "Cody Franson sucks because he can't skate", they don't post thousands of minutes of on-ice video showing that he can't skate. Except for some reason, the burden of proof is higher on the person providing the data rather than the opinion. Strange, isn't it?

I didn't have any beef. I basically took what you posted at face value. My mistake.

 

[silverfish]

I didn't have any beef. I basically took what you posted at face value. My mistake.

I'm just not sure what else I can do here to make the communication better.

I've given up on the rest of HFNYR in terms of using advanced metrics in any conversation, but at least for the people who still come into this thread, I'd like to be a more productive and better poster. So I understand your criticism, but that was just one post of many in this Vesey analysis. There's the other post you've mentioned. I wrote up a blog post about it.

The thing is, is that I'm not drawing any wild conclusions, here. But if you only read one post, you may think that. That, I suppose is an inherent "bug" of the internet message board. You're not going to scan the whole thread. You shouldn't have to. You saw a post, and you responded to that post. It makes sense. But I think there should be some benefit of the doubt for each poster that let's say if they post a conclusion, like let's say my eye-roll gif was for real, then there needs to be some level of trust that it's not just based on whatever is in that singular post.

 

[Machinehead]

Pionk is gonna be bad next year but it'll just be a 'sophomore slump'. Remember that stupid save that Hank made against Calgary? Look at Pionk on that play. Totally lost. **** happens when you ride that 101.79 PDO with your -6.5 relCF% and -7.35 relxGF%

At least the outlet along the boards once he recovered didn't get picked off.

That's a slight improvement for our defense.

 

[Mac n Gs]

Quinn had an interview in a blogpost on MSG. It was mostly fluff about accountability, but what caught my eye was his comment on trust playing out on the ice using forechecking as an example:

“I’ll use forechecking as an example. If you go forecheck in the right corner and the puck goes to the left, I’ve got one rule: Get above the puck. That means being in position where you’re not chasing the play. You’re in control of the angles, and can dictate the play. I have to trust you will follow that basic rule. Then, once you do that, I have to trust your abilities and that you will make a good hockey decision.”

It’s an immediate improvement from AV where we regularly overcommitted and yielded too much space to make a play against an opponent. I bet if they measured open ice on forechecking/defending zone exits and in-zone defense, we’d be at the bottom of the league.

 

[Blue Blooded]

I only have percentiles, but I believe they’re available if you donate to Ryan Stimson and Corey Sznadjer on Patreon. Here are some good resouces out there on public tableau:

Tableau Public

Tableau Public

Tableau Public

Those were great, thanks!

So DeAngelo's closest comp offensively/in transition seems to have been Karlsson (99/100 across the board) and the closest defensively to have been Pionk (5-10 in all three). Though no Ranger defenceman had pretty defensive metrics and I blame AV for his "back up all the way into Lundqvist's pads"-scheme.

 

[silverfish]

Those Tableau tools are excellent. Just mind your sample sizes.

 

[Mac n Gs]

Those were great, thanks!

So DeAngelo's closest comp offensively/in transition seems to have been Karlsson (99/100 across the board) and the closest defensively to have been Pionk (5-10 in all three). Though no Ranger defenceman had pretty defensive metrics and I blame AV for his "back up all the way into Lundqvist's pads"-scheme.

Yeah, the only negative is the sample sizes are relatively small. I've gone on for ages about AV's system of overpressuring on the forecheck and behind our own net, but the weirdest thing was the complete absence of pressure in the neutral zone.

It just never made sense.

 

[Blue Blooded]

SSS data may not be much in and of itself, but it is a useful part of the puzzle when evaluating a player.

I like the regression models as they can put the puzzle together with large datasets and give good answers as long as you ask the right questions. And while you cannot use the SSS data in them, it can still be part of the larger picture.

 

[Filthy Dangles]

Quinn had an interview in a blogpost on MSG. It was mostly fluff about accountability, but what caught my eye was his comment on trust playing out on the ice using forechecking as an example:

“I’ll use forechecking as an example. If you go forecheck in the right corner and the puck goes to the left, I’ve got one rule: Get above the puck. That means being in position where you’re not chasing the play. You’re in control of the angles, and can dictate the play. I have to trust you will follow that basic rule. Then, once you do that, I have to trust your abilities and that you will make a good hockey decision.”

It’s an immediate improvement from AV where we regularly overcommitted and yielded too much space to make a play against an opponent. I bet if they measured open ice on forechecking/defending zone exits and in-zone defense, we’d be at the bottom of the league.

Staying above the puck and taking a good angle, yeah that's hockey 101. I thought the narrative was always not that they overcommit but rather they were too passive and gave the opponent too much open ice.

It's also a bit of chicken or the egg scenario. Bad offense leads to bad defense. Bad defense leads to bad offense.

If you can't get out of your zone, you'll be caught behind the play a lot and won't be able to apply good forecheck pressure and teams will get easy exits. Bad D to Bad O.

 

[Irishguy42]

Quick hits:

Elite (strong link): Zibanejad
Secondary Star: Kreider/Zucc/Buch/Spooner/Hayes/Names/Fast/Skjei/Shatty/Lundqvist
Replacment: Lias/VZ/Howden/Staal/Smith/Claesson/Georgiev
Negative value: Chytil/DeAngelo

 

[Do you want ants]

Pretty interesting. Chytil and Middelstadt listed as negative and EK listed on Tampa and not Ottawa.

 

[Machinehead]

Some obvious SSS issues, but interesting study.

Carey Price being negative value fills me with flowers.

 

[Irishguy42]

Pretty interesting. Chytil and Middelstadt listed as negative and EK listed on Tampa and not Ottawa.

Chytil/Middelstadt were definitely victims of small sample size here, and while I don't have the data in front of me, I'm going to imagine that they both got pummeled in their on-ice metrics, which led to a negative value GAR/82. Likewise, Andersson had replacement level GAR/82 in the same sample size probably due to the sheltered role he had compared to Chytil/Middelstadt.

As for EK65, this was based off when Tierney made the roster "move" back when the rumors were saying that the deal was pretty much done. So Sean got a little ahead of himself, but probably kept him in there just as a fun "what-if"

 

[silverfish]

Would make a ton more sense to just leave those guys out than rank them, but whatever. From what I remember in terms of relCF% and relxGF%, Chytil got pummeled. Mittelstadt was also pretty bad. Lias, I don't think was positive, but I know had a terrible relxGF%. But like... it was 7 games, so what can you really say about that?

 

[silverfish]

My armchair data-scientist similar players to Skjei:

Severson
Bouwmeester
Mezsaros
Burns
Ranger
Petry
Benning
Pitkanen
E. Johnson
Hickey

 

[SA16]

BURNS