2015-2016 Statistical And Underlying Metrics Thread

[garret9]

Jets Nations By The Numbers Profiles
Forwards
Andrew Ladd
Bryan Little
Blake Wheeler
Mathieu Perreault
Mark Scheifeile
Drew Stafford
Alex Burmistrov
Adam Lowry
Chris Thorburn
Matt Halischuk
Anthony Peluso
Matt Fraser
Patrice Cormier

Defenders
Tobias Enstrom
Jacob Trouba
Ben Chiarot
Dustin Byfuglien
Mark Stuart
Tyler Myers
Grant Clitsome
Paul Postma
Jay Harrison
Adam Pardy

Best on Special Teams
Forwards on Power Play
Defenders on Power Play
Forwards on Penalty Kill

Click to read more...

 

[garret9]

Found this interesting:

 

[GJF]

Tyler Myers and Drew Stafford numbers are largely influenced by their Sabres numbers I assume.

 

[garret9]

Tyler Myers and Drew Stafford numbers are largely influenced by their Sabres numbers I assume.

I would hazard more "partly" than largely.

This uses impact on linemates, so the weakness of teammates impact should be severely diminished.

In addition, both had a fairly negative impact on teammates while they were Jets.

Context is always messy though, so my guess is that it is over estimating, slightly.

 

[powder88]

That's an interesting representation...where is it from? I assume we can look at all of the teams in that fashion?

 

[garret9]

Found the app on twitter.
Took me a while to find the post and poster it came from:
https://medium.com/@therobhayes/pro...nish-in-the-2015-2016-nhl-season-783a1742d10f

 

[mcpw]

In addition, both had a fairly negative impact on teammates while they were Jets.

I disagree. I think Myers had a positive impact on teammates.

CF% with and without Myers
Enstrom: with 53.0, without 51.2
Ladd: with 53.9, without 53.6
Scheifele: with 60.4, without 54.5
Frolik: with 57.0, without 54.0
Chiarot: with 56.1, without 53.2

Note: I've used a filter to get these numbers. Only the parts of the season when Stafford was not on the ice count.

 

[Whileee]

Do we really think that Trouba was a net negative, or will continue to be?

 

[GJF]

Somehow this doesn't make sense. This garret,... always putting up wrong numbers. I have no idea how they can feature his tweets on nhlnumbers.com.

 

[jetsforever]

I assume Trouba was dragged down by Stuart?

In addition, both had a fairly negative impact on teammates while they were Jets.

This surprises me - I would have thought it was just the Sabres numbers bringing them down.

 

[garret9]

I disagree. I think Myers had a positive impact on teammates.

CF% with and without Myers
Enstrom: with 53.0, without 51.2
Ladd: with 53.9, without 53.6 (this is essentially the same at that sample)
Scheifele: with 60.4, without 54.5
Frolik: with 57.0, without 54.0
Chiarot: with 56.1, without 53.2

Note: I've used a filter to get these numbers. Only the parts of the season when Stafford was not on the ice count.

You might want to dig deeper than that.

Myers has the lowest Corsi of the Jets defense Feb11-present.
Myers has the lowest relative Corsi of the Jets defense Feb 11-present.


The relative Corsi is the more concerning one in this case than just singling out some WOWYs.

However, for fun, if we do want to dissect some of those WOWYs:
* Enstrom's WOWY without was 1.5 games with Trouba (Myers injured in game) and one game (Jets last game) with Pardy. Also, Enstrom had a 54.5 Corsi with Trouba when looking over the whole season.
* Forward WOWY with D are something you want to be very wary with as you can actually get significant deployment/context dichotomies.
* Chiarots doesn't make sense to me. I get that too from puckalytics but then I look at this: http://thenationnetwork.s3.amazonaw...icle_85970313-f95d-4fcf-a381-607526ee6c05.png

Personally I think this is the better one, looking at isolating with Enstrom when with top line:

Trouba was around 56 IIRC.

 

[ps241]

Rick Ralph quoted Garret on the round table today and discussed this U shaped graph. Thought he explained it well and also emphasised how Huddy is a big Enstrom fan.

Good to see Toby get some love after the post season "who should go" Sun Poll proved how brain dead a small sample size of Winnipeg Hockey fans are.

 

[garret9]

Rick Ralph quoted Garret on the round table today and discussed this U shaped graph. Thought he explained it well and also emphasised how Huddy is a big Enstrom fan.

Good to see Toby get some love after the post season "who should go" Sun Poll proved how brain dead a small sample size of Winnipeg Hockey fans are.

In the end, it's just one model.

While I think it shows some valid things, it's not much that we all haven't seen already.

If you want my positive spin:
* Burmi is set as 0 as an unknown, but I bet he'll be better since when he was in NHL he was slightly more positive impact than Lowry in the same numbers.
* Byfuglien includes his TOI as a forward and defender, and we all know he is much, much better as a defender (or at least hopefully all know this by now)
* Hutch could take more TOI/GP from Pavelec

I expect some players improve, and some get worse, but those above are outliers where I think the model may miss substantially and impact beyond "some better, some worse".

 

[mcpw]

* Enstrom's WOWY without was 1.5 games with Trouba (Myers injured in game) and one game (Jets last game) with Pardy. Also, Enstrom had a 54.5 Corsi with Trouba when looking over the whole season.
* Forward WOWY with D are something you want to be very wary with as you can actually get significant deployment/context dichotomies.

First point: I've used whole season WOWYs.
Second point: I'd argue that that's true for every single WOWY stat.

The rest are good points.

 

[garret9]

First point: I've used whole season WOWYs.
Second point: I'd argue that that's true for every single WOWY stat.

The rest are good points.

1) I misread your odd filter choice haha
2) Not necessarily true by the same extent, it is true in part though. This is why WOWYs are used the way they are currently as trend markers.

 

[garret9]

Another model for this coming season:

http://t.co/JU1lo5NMDg

 

[mcpw]

Found this interesting:

Single-number player evaluations are always flawed (in some way), but they are always interesting to test your model and to look how different models perform.

Let me point out the (in my opinion) biggest flaws of this specific model:
ANA Josh Mason +3.99 (he will struggle to get a roster spot)
ARI Tye McGinn +2.95 (sorry, but he plays for TBL. Also, a 4th liner doesn't have that kind of impact.)
BUF Zach Bogosian -6.39 (ok, he's not too good, but is he that bad?)
CGY Michael Frolik +4.95 vs Mark Giordano +2.81 (Giordano clearly has a bigger impact)
CHI Corey Crawford +6.46 (they're using some weighted 4-year sample size, I think, but I don't believe in that number)
CHI David Rundblad +4.28 (yeah, no.)
COL Erik Johnson -2.64 (really?)
EDM Cam Talbot +9.18 (clearly, goalie sample size is not implemented)
LAK Jake Muzzin +6.04 vs Drew Doughty +1.47 (again: Doughty away from Muzzin played shutdown minutes with close-to-retiring Regehr. Muzzin away from Doughty played offensive minutes with Greene.)
MIN Mikko Koivu +4.15 (not looking at deployment)
MTL Jeff Petry -4.26 (Edmonton effect?! I don't know.)
VAN Jacob Markstrom -13.97 (yikes, what kind of icetime did you assign to that poor guy?)

All things considered, it looks pretty good, though, for a single-number player evaluation model.

 

[garret9]

Always will need to investigate further.

Holistic-stats are good starting points though.

It's interesting to see the differences and similarities between the two.

 

[mcpw]

I like to look at absolutes.

NHL all-star team according to that model:

Tomas Tatar - Patrice Bergeron - Brendan Gallagher
Daniel Sedin - Jonathan Toews - Vladimir Tarasenko
James Neal - Anze Kopitar - Reilly Smith
Brad Marchand - Michael Frolik - Max Pacioretty
Jaromir Jagr, Carl Hagelin, Patrik Elias

Jake Muzzin - Anton Stralman
Olli Maatta - Duncan Keith
Brian Campbell - David Rundblad

Tuukka Rask
Carey Price

 

[Mortimer Snerd]

Another model for this coming season:

http://t.co/JU1lo5NMDg

Once again Tyler Myers looks worse than bad in these "advanced" stats. When you first posted to the effect that he might not have been quite as good as everybody was thinking I was a little surprised but OK, that is what statistical analysis is for. This seems to be a perfect case for examining a difference between advanced statistical analysis and the eye-test. Such a wide difference is a little hard to accept. That he was not quite as good as he seemed OK but this shows him to be our worst D man by a wide margin. How is that possible? If these stats are giving an accurate picture we need to alter our thinking going forward. Keep Buff and trade Myers for instance.

 

[garret9]

Have to remember that worst impact /= worst, especially with these numbers.
ex: Harisson's GAR is likely better than Myers, but his GAR/60 is worse.

And, while people over exaggerate the impact of being on Buffalo, it is still an impact.

I don't expect Myers to ever be Buff good, and Trouba IMO is already better, but I bet that Myers will look better than Stuart, Harrison, or Pardy this year.

 

[Mortimer Snerd]

Have to remember that worst impact /= worst, especially with these numbers.
ex: Harisson's GAR is likely better than Myers, but his GAR/60 is worse.

And, while people over exaggerate the impact of being on Buffalo, it is still an impact.

I don't expect Myers to ever be Buff good, and Trouba IMO is already better, but I bet that Myers will look better than Stuart, Harrison, or Pardy this year.

It is not just this particular analysis though. You have posted several different things showing him to be varying degrees of bad and I'm pretty sure that in some you filtered Buffalo out.

I'm not challenging you or what you have posted on Myers. I'm asking for a reason for the discrepancy between stats and eye-test. I'm unable to come up with a satisfactory one. I hope you can. Maybe larger sample size will eliminate the discrepancy. That is the best I can come up with.

 

[garret9]

It is not just this particular analysis though. You have posted several different things showing him to be varying degrees of bad and I'm pretty sure that in some you filtered Buffalo out.

I'm not challenging you or what you have posted on Myers. I'm asking for a reason for the discrepancy between stats and eye-test. I'm unable to come up with a satisfactory one. I hope you can. Maybe larger sample size will eliminate the discrepancy. That is the best I can come up with.

Just since this is brought up, people need to remember it's not discrepancy between stats and eye-test but discrepancy between stats and YOUR eye-test.

I see how and why Stafford is how he is.

I see how and why Enstrom and Byfuglien are what they are.

I see it with Myers too. Not being terrible, but why he's not to the same level of Trouba or Byfuglien.

 

[Mortimer Snerd]

Just since this is brought up, people need to remember it's not discrepancy between stats and eye-test but discrepancy between stats and YOUR eye-test.

I see how and why Stafford is how he is.

I see how and why Enstrom and Byfuglien are what they are.

I see it with Myers too. Not being terrible, but why he's not to the same level of Trouba or Byfuglien.

OK, rephrase. There is a discrepancy between your evaluation (stat & eye) and a widely held eye-test. It is far from mine alone. In fact I was probably among the less impressed with Myers group (still positively impressed but less so than many). Unless there is a silent majority here it is the consensus opinion that he was great or at least very good.

 

[Chaosmonkey]

Co-worker showed me this infographic today. Between the two of us, we have no idea what the GAR and GAR/60 stat represent. A little help for us less advanced stats guys?