2019/20 Roster Thread XXIX

[Lord Defect]

This is totally a re-enactment of target tori.

 

[Ghosts Beer]

Ah yes the stats you used wrong.

It's amazing. You spend post after post whining about people taking a single stat and using it definitively, and then do the exact same thing.

What you've failed to observe is all the people who combine many stats and pair it with usage and attempted honest observation.

Ah, my bad. I thought you used a different stat.

 

[Beef Invictus]

Congrats on proving nothing. Keep up the hypocrisy of shrieking at people for allegedly using one or two stats while you make sweeping generalizations with one or two stats.

I understand that all you can do is throw a tantrum now that you've been called out. I forgive you.

 

[TB87]

Corsi is a very predictive stat at the team level and at the individual level.


Anyways, EvolvingWild’s Regularized Adjusted Plus-Minus (RAPM) does as good a job as any metric to measure skater performance via: GF/60, xGF/60, CF/60, xGA/60, & CA/60 (at Even-Strength primarily):

Full article explaining this metric: Reviving Regularized Adjusted Plus-Minus for Hockey


Even-Strength RAPM in Chart form below:

HockeyViz’s isolated shot impact Viz’s are pretty interesting as well:

Explained in detail here: Model Description: Magnus 2



- All that orange in Hagg’s PK isolated shot suppression chart throughout the entirety of the slot is the stuff of nightmares.



Hagg is still very much a net-negative player via shot-attempt generation & suppression.

 

[Striiker]

It's really crazy that we still have to point out how bad Hagg is in 2020.

You'd think that even the most hardcore contrarians would've given up on that one by now and that this team would have traded the trash away.

 

[Curufinwe]

Further clarity is an understatement when two of the forwards are guys who will missed most/all of 19-20. What's for certain is that the Flyers will lose a better player than Bellemare this time. Speaking of PEB, he has already set a career high with 18 points in 48 games.

Mock expansion draft: What Seattle's NHL team might look...

Philadelphia Flyers (7-3-1)
F: Sean Couturier, Claude Giroux, Travis Konecny, Kevin Hayes, Nolan Patrick, Oskar Lindblom, Jakub Voracek
D: Ivan Provorov, Travis Sanheim, Philippe Myers
G: Carter Hart

if Gostisbehere were available and maybe the best option for power-play QB (next to Subban), then Seattle could go in that direction. Or, on a team that figures to be desperately short of natural scorers, select Voracek or van Riemsdyk. Tough call today – though there could be further clarity by June 2021.

 

[Starat327]

Congrats on proving nothing. Keep up the hypocrisy of shrieking at people for allegedly using one or two stats while you make sweeping generalizations with one or two stats.

I understand that all you can do is throw a tantrum now that you've been called out. I forgive you.

Still? This is STILL happening?!?

 

[flyersnorth]

Corsi is a very predictive stat at the team level and at the individual level.

Predictive of what? The regression models I've seen have shown that Corsi has a weak r2 in relation to a team's performance (pts), and as a result has a weak predictive value on team performance.

There is more predictive value at the individual level (statistically significant).

 

[deadhead]

Both the stats and eyeballs say the same thing, Provorov, Niskanen and Braun are heads and shoulders above the others defensively.
Myers has potential as a shutdown defenseman once he cuts out the mindfarts that leave him out of position, his size, speed and reach could make him the best defender on this team in a few years if he continues to develop.

Sanheim and Ghost are mediocre defensive defenseman and have to produce offensively to have real value.

Hagg is a body. Like other "bodies" (L Schenn, for example) he will only succeed if he learns to improve his positioning and recognize what's happening on the ice - but he's only 24. Sanheim and Myers also struggle with recognition, but have the make speed to cover some of their mistakes, Hagg doesn't. Hagg has about a two year window to step up his game or he'll end up like Bigras and Witherspoon.

 

[Striiker]

Sorry, here's another chart for everyone.

But this is a key one, because it includes one of the biggest factors of the defensemens play, which many on here do everything in their power ignore.

Time with Couturier at 5v5​
Provorov​	333:01​
Niskanen​	293:18​
Sanheim​	253:37​
Braun​	184:15​
Myers​	138:28​
Ghost​	124:04​
Hagg​	87:19​

[TBODY] [/TBODY]

Now look at the "top 4" who are getting the best usage and how much it boosts their stats...

Provorov with Couturier	55.77 CF%	58.95 xGF%
Provorov without Couturier	49.54 CF%	48.96 xGF%
Niskanen with Couturier	55.37 CF%	59.75 xGF%
Niskanen without Couturier	48.48 CF%	48.76 xGF%
Sanheim with Couturier	55.05 CF%	54.05 xGF%
Sanheim without Couturier	50.05 CF%	47.42 xGF%
Braun with Couturier	55.04 CF%	52.11 xGF%
Braun without Couturier	51.27 CF%	50.48 xGF%

[TBODY] [/TBODY]

That top pair gets carried crazy hard. No wonder they have better individual on-ice statistics when they're playing almost 3x as much with their hard carry as the bottom pair.

 

[deadhead]

So you're saying Hagg's bad metrics are due to having the fewest minutes with Couts?

I'd point out that Braun doesn't get that much more time with Couts than Ghost, which simply shows that the "rel" numbers don't lie, Braun >> Ghost this season.

 

[TB87]

Predictive of what? The regression models I've seen have shown that Corsi has a weak r2 in relation to a team's performance (pts), and as a result has a weak predictive value on team performance.

There is more predictive value at the individual level (statistically significant).

predictive of future shot attempt differential and goal scoring. You’re right though, It’s a little more pronounced at the individual level than at the team level.

 

[deadhead]

predictive of future shot attempt differential and goal scoring. You’re right though, It’s a little more pronounced at the individual level than at the team level.

You have to be a little careful with the kind of regression analysis that used in all these hockey models.
1) it assumes a linear relationship, sophisticated statistical modeling (see Hendry) can test functional forms instead of just assuming one.
2) most of these models are pretty naive, multi-collinearity is probably a big issue that's generally ignored
3) there's a lot of left out variable error in hockey models, some of this will be resolved when tracking data becomes widely available

In general, if you have the puck in the O-zone and shoot more, you'll score more, don't need a regression model to figure that out.
Players who shoot more may simply be players who have good scoring skills and are fed the puck.
But a five man unit is generally chosen for an offensive philosophy, some coaches stack their top offensive groups, others want balance, some teams take a lot of shots, others play for high quality shots. Hard to capture these influences.

All hockey stats should be taken with a large bag of salt.
They have value as suggestive metrics, but they're far from conclusive.

 

[Curufinwe]

Predictive of what? The regression models I've seen have shown that Corsi has a weak r2 in relation to a team's performance (pts), and as a result has a weak predictive value on team performance.

There is more predictive value at the individual level (statistically significant).

Man, I used to know all about R2 and other statistical stuff. But 20 years later I've forgotten almost all of it.

I do remember this always being a problem.

Heteroscedasticity: Simple Definition and Examples - Statistics How To

 

[Striiker]

Braun's two most common teammates: Sanheim and Couturier
Ghosts two most common teammates: Hagg and Hayes

Ultra-super-hyper-massive difference there.

And even when he's not with Couturier, at least he's still with Sanheim instead of Hagg.

Meahnwhile, when he's stuck with Ghosts most common teammates...

Braun with Hayes	49.69 CF%	47.24 xGF%
Braun with Hagg	28.57 CF%	27.76 xGF%

[TBODY] [/TBODY]

Ghost is better than Braun, without a doubt. But usage/teammates matter. Even without stats, that's clear as day to see.

 

[Ghosts Beer]

Corsi is a very predictive stat at the team level and at the individual level.


Anyways, EvolvingWild’s Regularized Adjusted Plus-Minus (RAPM) does as good a job as any metric to measure skater performance via: GF/60, xGF/60, CF/60, xGA/60, & CA/60 (at Even-Strength primarily):

View attachment 313225


Full article explaining this metric: Reviving Regularized Adjusted Plus-Minus for Hockey


Even-Strength RAPM in Chart form below:
View attachment 313227



HockeyViz’s isolated shot impact Viz’s are pretty interesting as well:

View attachment 313231

Explained in detail here: Model Description: Magnus 2



- All that orange in Hagg’s PK isolated shot suppression chart throughout the entirety of the slot is the stuff of nightmares.



Hagg is still very much a net-negative player via shot-attempt generation & suppression.

Goal-based Metrics Better Than Shot-based Metrics at Predicting Hockey Success

Interesting read, though his analysis is retrospective & doesn’t necessarily prove “predictive.”

 

[TB87]

This may be just my perception, but I imagine that there’d be far fewer outsized reactions and disagreements on here (social media, other sites, etc) if people took the time to understand exactly what people are saying instead of putting words in other people’s mouths. Reading and responding to the actual words being shared and not what the words “might be” saying or what they “might be” referring too.

 

[deadhead]

Braun with Hagg is a bad fit, we saw the same issue with Gudas and Hagg. Fit matters as much as talent.
Cherry picking Braun with Hayes misses that Braun without Couts still had far better metrics than Ghost without Couts.

No matter how you try to spin it, Braun has clearly outplayed Ghost this season.
At some point you'll run out of excuses and admit the obvious.

Braun is an elite defensive defenseman with very limited offensive skills, his best offensive skill is his ability to pinch without getting beat.
Ghost is a mediocre defensive defenseman who was a good offensive defenseman (never great at 5x5) and top PP QB two years ago. Since then, meh.

 

[deadhead]

Goal-based Metrics Better Than Shot-based Metrics at Predicting Hockey Success

Interesting read, though his analysis is retrospective & doesn’t necessarily prove “predictive.”

This is a prime example of "data mining," run a bunch of different models and see which ones generate good t-stats.
Problem is that if you are going by the book, each successive model has to incorporate the previous models in determining the significance of its results.

Any large enough data set, if you throw enough different functional forms at it, will generate "significant results."

Which is why results from "Big Data" are often just spurious correlations, give me enough astrological models and I'll find one that predicts stock market prices.

Goal based models should predict success for obvious reasons.
Shot based models will also predict success for similar reasons.

The key is teasing out which factors are significant and robust, and to understand the mechanics behind the prediction - is it shots that matter, quality of shots, defensive strategy (lots of shots without players setting screens have less value, how do you measure goalie visibility?). What kind of shots generate rebounds against what defensive schemes and what goalies? And so on.

 

[Ghosts Beer]

This is a prime example of "data mining," run a bunch of different models and see which ones generate good t-stats.
Problem is that if you are going by the book, each successive model has to incorporate the previous models in determining the significance of its results.

Any large enough data set, if you throw enough different functional forms at it, will generate "significant results."

Which is why results from "Big Data" are often just spurious correlations, give me enough astrological models and I'll find one that predicts stock market prices.

Goal based models should predict success for obvious reasons.
Shot based models will also predict success for similar reasons.

The key is teasing out which factors are significant and robust, and to understand the mechanics behind the prediction - is it shots that matter, quality of shots, defensive strategy (lots of shots without players setting screens have less value, how do you measure goalie visibility?). What kind of shots generate rebounds against what defensive schemes and what goalies? And so on.

I agree with all of this.

 

[Curufinwe]

It sure would nice if Fletch could pull off a trade like when the Pens got rid of Pronman favorite Daniel Sprong for a young #4/5 dman.

Marcus Pettersson - CapFriendly - NHL Salary Caps

Pettersson is having a bit of an uneven season, apparently, but the value gap is still huge right now.

 

[Striiker]

@Beef Invictus Hey, did you ever get an answer to the question about whether or not it was good coaching to make the team worse on purpose?

 

[bobbythebrain]

Maybe when people disagree with you, they're not lying.
Maybe they're just smarter and more experienced than a 26 year old kid with no statistical background.

And never played hockey

 

[kudymen]

And never played hockey

And doesnt do 284.574 squats a day

 

[bobbythebrain]

And doesnt do 284.574 squats a day

Who works out legs?

It's curls for the girls