Who deserves the Selke strictly based on numbers?

[TomasHertlsRooster]

So rather than have a long strung-out argument, we can simply agree on this statement: there are systemic biases built into the model.

Those systemic biases will, inevitably, conjure statistical phantoms. And identifying those unsupportable results is necessary for refining the model.

All I’m trying to accomplish here is to wave a flag and point out that I see several unsupportable results cropping up on a single team — one that I happen to know very intimately and am (at worst) unlikely to be completely off-base about.

The flag has been waved. Having said my peace, if you still truly think the metrics are correct, that there is no phantom but a real tangible result, then that’s simply a judgment call. Nothing I can do to change it and I’m not bothered by letting it end there.

A mild systemic bias that probably flatters your guy if anything, yes. And there is no systemic bias for goals against, where Aho’s defensive impact this season was equally putrid or perhaps even worse.

 

[tarheelhockey]

A mild systemic bias that probably flatters your guy if anything, yes. And there is no systemic bias for goals against, where Aho’s defensive impact this season was equally putrid or perhaps even worse.

Wow, you're really gonna spoil for this level of fight over this?

 

[tinyzombies]

Yeah, it is pretty strange to see them all that high. However, all 3 of those players have terrible defensive metrics at the career level:

View attachment 329975

It's actually striking how similar they are in EVD GAR at the career level; all atrocious. In addition, Kovalchuk's EV RAPM xGA/60 in 2008-2009 was literally the worst of all time. Nearly 5 standard deviations above average!

View attachment 329977

So, this data isn't telling us these guys were always great defensively. It's actually supporting the conclusion that all of them sucked for most of their careers. It's just also telling us that over the course of this one season, they've been pretty good.

THis matches hte eye test for Kovalchuk in Motnreal - he was very responsible in his play for some reason. No Danault on the list??

 

[emptyNedder]

Just stumbled upon this thread. Have to totally agree with tar heel that the numbers from this season are aberrational.

We should be asking those questions especially in a case where the model is producing fluke results that are obviously incorrect. Like Aho being a terrible defender or... I don't know, Brett Pesce being the team's worst defenseman in his own end

Carolina turned around 10 years of frustration in February last season. During the first 8 days of that month Aho took the large majority of 5-on-5 minutes against top lines. In four games in 8 days none of these players scored an even-strength goal against the Aho-led Canes (William Karlsson, Marchessault, Gaudreau, Monahan, Crosby, Guentzal, Zibanejad, Kreider). In fact, both Pittsburgh and NYR were shutout. To drive home the fact, Aho was on the ice for almost 80% of the time that McDavid was skating on 2/15/2019 and for one of the few times all season McDavid wasn't on the scoresheet.

Jordan Staal was sidelined during this stretch.

I would accept an argument that Aho was playing better than he normally does for those two weeks. However, it is far-fetched to believe that Aho is in the bottom 20% of centers defensively when he shutout several of the best players in the league when his team most needed his defensive efforts.

 

[Trap Jesus]

One thing I've seen referenced a ton online is controlled zone exits. It sounds good in theory, you're just imagining a guy making a defensive play and than safely exiting the zone, but in reality that's more a reflection of players who are given the puck to carry up the ice, wingers that are receiving breakout passes, or players that are up playing closer to the blueline rather than helping out down low.

 

[Byron Bitz]

I found defensive play much harder to measure statistically in hockey than it is in Basketball.

 

[ahmedou]

Danault deserves the Selke strictly based on numbers

 

[supsens]

Actually RAPM takes in to account things like matchups and usage and teammates and things like that. The raw xGA total would definitely have that issue, but RAPM xGA does not. In that metric Aho actually ranks 2nd worst on the Canes, behind only Erik Haula. Even Necas and Dzingel right behind him. I imagine the reasons Necas and Dzingel end up looking better than Aho defensively once usage is taken into account is due to the model blaming Haula for the shitty defense, but regardless, even with usage taken in to account, Aho does not look remotely good defensively.

I was reading that they use ridge regression to cope with guys that play together, when you read about ridge regression it says you will not get the true results, you will get results that behave. So they are forcing what they want to see, that seems extremely flawed.
They also give more weight to longer shifts. Sounds like nonsense.

 

[Henkka]

Yeah, it is pretty strange to see them all that high. However, all 3 of those players have terrible defensive metrics at the career level:

So, this data isn't telling us these guys were always great defensively. It's actually supporting the conclusion that all of them sucked for most of their careers. It's just also telling us that over the course of this one season, they've been pretty good.

Tougher minutes during a career and soft minutes (to look good) at last season.

Qualcomp plays huge role in these and I still haven't seen a model which will put it right.

 

[TomasHertlsRooster]

I was reading that they use ridge regression to cope with guys that play together, when you read about ridge regression it says you will not get the true results, you will get results that behave. So they are forcing what they want to see, that seems extremely flawed.
They also give more weight to longer shifts. Sounds like nonsense.

This is pretty much a perfect example of reading about something for about 2 minutes and then using your own misinterpretation of what you just read to inaccurately support a conclusion that you already had in mind before you started reading.

 

[supsens]

This is pretty much a perfect example of reading about something for about 2 minutes and then using your own misinterpretation of what you just read to inaccurately support a conclusion that you already had in mind before you started reading.

I had nothing in my mind before I read it.
I do not believe a longer shift means your better(why would it)
I do not like when it says it does not represent the true value.

But I’m sure these people who’s cute little graph that’s adjusted and changed until they get what they want to see have a foolproof system that shows all.

 

[Doctor No]

No, it means that reading Wikipedia for two minutes on ridge regression does not mean that you understand what ridge regression is and how it's used (and misused).

Ridge regression isn't perfect and does have flaws, but you're not on point.

You tipped your hand with the "cute little graph" comment, by the way.

 

[supsens]

No, it means that reading Wikipedia for two minutes on ridge regression does not mean that you understand what ridge regression is and how it's used (and misused).

Ridge regression isn't perfect and does have flaws, but you're not on point.

You tipped your hand with the "cute little graph" comment, by the way.

When you need to adjust the numbers of the guys that play with each other a lot and the entire NHL spends most of its time on set lines...you understand what your entire baseline is now correct?
Corsi is a line stat/situation stat, it is not a player stat, people keep cramming it into “advanced stats” and they are turning out junk.

 

[Doctor No]

The only time Corsi has come up in this thread was in an off-handed remark by Mike; why are you bringing it up now as if it's being used here?

 

[supsens]

The only time Corsi has come up in this thread was in an off-handed remark by Mike; why are you bringing it up now as if it's being used here?

I’m just saying, plus minus is also a team stat... anyway a weighted plus minus system is what it is.
Boring

 

[Doctor No]

Let me try this again.

Other than you, who the f*** is talking about Corsi in this thread? Bringing it up is a strawman.

 

[TomasHertlsRooster]

I had nothing in my mind before I read it.
I do not believe a longer shift means your better(why would it)
I do not like when it says it does not represent the true value.

But I’m sure these people who’s cute little graph that’s adjusted and changed until they get what they want to see have a foolproof system that shows all.

Again, you are heavily misunderstanding ridge regression and how it is used. It is not anywhere as simple as something that "does not represent the true value", or something that just provides you with "results that behave".

A longer shift doesn't mean that somebody is better, but longer shifts are weighted heavier because the sample is larger. I'm not sure why that's a difficult concept.

 

[supsens]

Again, you are heavily misunderstanding ridge regression and how it is used. It is not anywhere as simple as something that "does not represent the true value", or something that just provides you with "results that behave".

A longer shift doesn't mean that somebody is better, but longer shifts are weighted heavier because the sample is larger. I'm not sure why that's a difficult concept.

Well it’s hockey if you are an offensive player and a better player the reason you are taking a longer shift will be because you have the puck in deep and are generating scoring chances, there is zero need to give ‘extra weight’ to something you are already gaining max props for.
It’s a double payday and it’s made up because the real numbers don’t make any sense. These people are selling junk because people heard about ‘money ball’

 

[Doctor No]

These people are selling junk because people heard about ‘money ball’

100% false.

 

[GoldenSeal]

Anyone reading this thread deserves the Selke... jeez...

 

[Halla]

No offense since you're approaching this analytically and not subjectively... but..

How are Bozak, Kovalchuk and JVR in the top 40?

Maybe all of them suddenly improved immensely on defense or maybe the fact that they're all there is an indication that one should be skeptical of the numbers in the last column.

James van Riemsdyk could score goals in TOR, but he was anything but good defensively. Seeing him on the chart is basically the equivalent of seeing Kessel on it in terms of shock value.

having watched two of these guys for many years they are among the worst defensively away from the puck.
bozak is good at draws and can get 15g/35pts and JVR can still pop in 30 but selke candidates they are not