A3Z Player Comparison Tool

[Rempe73]

I was just wondering if anyone could help me out with the A3Z player comparison tool. I’m still kind of new to advanced stats, but I want to learn. I’m looking for the best and most efficient way to judge a player’s passing, shot, defensive play, skating, hockey sense, and overall skill. Will this tool be helpful at all? Or does it only tell you how good/bad a player is offensively and defensively? Thanks and go f*** yourself (kidding, Bill Burr reference).

Click to read more...

 

[oilerbear]

As creator of 40+ hockey theories/theorems like ( homeplate) from a 40-50 yr old period of time watching WHL hockey.

Theorem: Current binary data does not have the analysis resolution level to create accurate measure for players.

Just looking at Oiler Dmen shift count and shift on the Fly (SOTF) starts last year.

Dman; games; shifts; on the fly starts, % of shifts on the Fly (SOTF)
Nurse, 82gm, 2306 shifts, 1164 sotf, 50.48%
Larsson, 79gm, 2243 shifts, 1141 sotf, 50.87%
Russell, 72gm, 1858 shifts, 997 sotf, 53.66%
Klefbom 61gm, 1695 shifts, 759 sotf, 44.78%
Benning, 70gm, 1499 shifts, 895 sotf, 59.71%
Gravel, 36gm, 752 shifts, 455 sotf, 59.71%
Sekera, 24gm, 548 shifts, 322 sotf, 58.76%
Jones, 17gm, 434 shifts, 242 sotf, 55.76%
Manning, 12gm, 239 shifts, 138 sotf, 57.74%

Theory: 50-60% of Entry & corsi against is affected by (SOTF) shift on the fly with or without pocession.
Theory: Entry & Corsi is also dependent on variance from NZ trap & No NZ trap.
What!
Currently a 60-75% inaccurate Measure of corsi.
That is a 60-75% error that translates to forward: corsi for, Fenwick for,Closed shot for, open shot for.

That is a 60-75% error that translates to the 2 GA structures (based on HD area location) entry and corsi numbers.
Theory: rover is (off D outside HD area) thevabandoning provides a free path to on side of the HD area yielding High % Open HD shot density.
1R - 1D - 1G
2D - 1G ( all defending HD area)
The 1-1-1 and 2-1 structures dictate Goal for per corsi success.

Theorem: goal diff 3-1-1-1 vs 3-2-1 structures give varied affects.
Dpair: the Fenwick against, Closed shot against, open shot against.

Theory: Dpairs establish the save% baseline based on open HD shot density yeilded.
Theory: Goalie performance: is a +ve/-ve measure of save% above or below the D pairs established save% baseline based on open shot HD success density.

Theory: the dynamics of a GF and GA situation requires 3 diffrent performance levels with difrent situational avg (team, Comp, ZS , NZ trap, 3-1-1-1 or 3-2-1 structure) they regress to for any given player.
The 3 difrent performances are.

1. Theory: Forward Open HD penetration shot density. Each forward has a diffrent.
2. DA dman establishes an expected save% baseline ( expected goals) to their side. The average theybrehress to is difrent dependent on the 5 large factor situation variables.
3. Goalie performance is measured by +ve or -ve save% relative to the expected save% baseline.

Theorem: PDO is a low resolution ( GA/GF in simplest form) statistical fluke with no real value at a deeper analytical resolution.
It has no analytical value at an single player form.

When you get to the 2nd level of resolution
Their are 3 variables identified above.
3 x

Then you go to 3rd level 5D area graph of situational permutations.
Based on distinct avg variance groups.
A. Team 4 x ( 21st, 2nd, 3rd, 4th)
B. Comp 4x (1st, 2nd, 3rd, 4th)
C. ZS 4x but currently 50-60% inaccurate
D. NZ trap 2x ( yes/ no)
E. GA structure 2x (3-1-1-1 yields high open Highndanger shit success density)(3-2-1 yields Lowe rate of open high danger shot success density)

3rd resolution level is at the first point individual analysis has accurate value.

3 x 4 x 4 x 4 x 2 x 2 =768 difrent non complex averages to regress to.
Not 2

All current low resolution binary anslysis you see on here does not take it to a higher resolution ( variable identification) to even come close to being accurate.

It is important to know that dmen generate (G, 1A, 2A) Even Goals at 4 times less rate than forwards.

I developed the 2 critical goal diff sections of analysis.
1. HD area (homeplate) shotsuccess density chart and HD:LD ratio. Watching Flin Flon bombers and SAskatoon blades of late 60’s and early 70’s and continued watching the Tier2 PA raiders.
Their are elite Dmen out their who keep a high % of corsi released to perimeter.

2. Identification of shot quality ( Open/ Closed shots) ( theorem)
Watching a 7 yr old Ron Gunville move like a table hockey goalie.( theorem)
Their was a high rate of hit goalie balls.


Open shot is a shot that hits open space in net elevation requiring
A goalie to make a save. The have a shot success density >0

Closed shot is a shot that hits a goalie moving with the puck like Ronnie Gunville did. 0% chance of going in. Density =0

Their was a jump in save% when goalies started to copy the 2 goalies that first imitated Table Hockey goalie movement.
JVB & ROY.

Ron Gunville is the current Director of Player Personel of the WHL champ PA Raiders.

Theory: Their are Elite Dmen ( who cause the highest % of 0% corsi ( 0%chance) per corsi faced.

Theory: 0% Corsi = ( blocks + forced misses + closed shots)
Theory: Only open Corsi which become open shots are scoreable.

Elite 0% Corsi Dmen like Kris Russell, Calvin DeHaan.

Theory: So all Closed shots must be removed from analysis to get a true Open HD shot density Chart ( map) 768 averages must regress to.

I could go into further multivariable levels of resolution.

Suffice to say.
Theorem: war works in Baseball cause they have a series of binary result actions. All teams try to perform the large win % factors, the same. So the small margin affects have a more valued value in Baseball.

WAR & GAR does not work cause the 5 large win% factors are not played the same by each GM.
Which causes the high group of averages. At higher resolution.
Until hockey starts to play those situations all the same.
Marginal measures used in baseballs war will not be available.

I ran into D. Sutter at Cactus corner truck stop Corner of Highway 9 and south highway 36 near Hanna, Alberta.
He was headed to his nieces wedding and wanted a Calgary sun.
I was putting 40k into kids education doing the 2 hr paper distribution run daily.

I congratulated him on winning his 2 cups and replace Mitchell and Reghr minutes with the best Open HD shot D prospect in the game. Brayden Mcnabb.

Sutter said “ you like analytics.”
“ yes, come up with a lot of theories”
Sutter, “ we are turning them into ........ robots”

I am trying to get everyone to play the high % win factors the same.
Turning them into Robots.

All these tools are a nice ways to learn the simple binary analytical process.
But most are well above 60% in accurate.

Some embarrassing bad at an individual player analysis.

My Wife is a award winning sports page editor for Post Media.
The only paper chain that asks their reporters to be accurate.
She told me they were taught to write at a grade 3 readers level of comprehension.

Their are grade 3-5 level mistakes in the hockey analytic community.

Continue to try to learn.

But remember 60+% inaccurate for most.

This one gets its own chapter.

 

[Filthy Dangles]

As creator of 40+ hockey theories/theorems like ( homeplate) from a 40-50 yr old period of time watching WHL hockey.

Theorem: Current binary data does not have the analysis resolution level to create accurate measure for players.

Just looking at Oiler Dmen shift count and shift on the Fly (SOTF) starts last year.

Dman; games; shifts; on the fly starts, % of shifts on the Fly (SOTF)
Nurse, 82gm, 2306 shifts, 1164 sotf, 50.48%
Larsson, 79gm, 2243 shifts, 1141 sotf, 50.87%
Russell, 72gm, 1858 shifts, 997 sotf, 53.66%
Klefbom 61gm, 1695 shifts, 759 sotf, 44.78%
Benning, 70gm, 1499 shifts, 895 sotf, 59.71%
Gravel, 36gm, 752 shifts, 455 sotf, 59.71%
Sekera, 24gm, 548 shifts, 322 sotf, 58.76%
Jones, 17gm, 434 shifts, 242 sotf, 55.76%
Manning, 12gm, 239 shifts, 138 sotf, 57.74%

Theory: 50-60% of Entry & corsi against is affected by (SOTF) shift on the fly with or without pocession.
Theory: Entry & Corsi is also dependent on variance from NZ trap & No NZ trap.
What!
Currently a 60-75% inaccurate Measure of corsi.
That is a 60-75% error that translates to forward: corsi for, Fenwick for,Closed shot for, open shot for.

That is a 60-75% error that translates to the 2 GA structures (based on HD area location) entry and corsi numbers.
Theory: rover is (off D outside HD area) thevabandoning provides a free path to on side of the HD area yielding High % Open HD shot density.
1R - 1D - 1G
2D - 1G ( all defending HD area)
The 1-1-1 and 2-1 structures dictate Goal for per corsi success.

Theorem: goal diff 3-1-1-1 vs 3-2-1 structures give varied affects.
Dpair: the Fenwick against, Closed shot against, open shot against.

Theory: Dpairs establish the save% baseline based on open HD shot density yeilded.
Theory: Goalie performance: is a +ve/-ve measure of save% above or below the D pairs established save% baseline based on open shot HD success density.

Theory: the dynamics of a GF and GA situation requires 3 diffrent performance levels with difrent situational avg (team, Comp, ZS , NZ trap, 3-1-1-1 or 3-2-1 structure) they regress to for any given player.
The 3 difrent performances are.

1. Theory: Forward Open HD penetration shot density. Each forward has a diffrent.
2. DA dman establishes an expected save% baseline ( expected goals) to their side. The average theybrehress to is difrent dependent on the 5 large factor situation variables.
3. Goalie performance is measured by +ve or -ve save% relative to the expected save% baseline.

Theorem: PDO is a low resolution ( GA/GF in simplest form) statistical fluke with no real value at a deeper analytical resolution.
It has no analytical value at an single player form.

When you get to the 2nd level of resolution
Their are 3 variables identified above.
3 x

Then you go to 3rd level 5D area graph of situational permutations.
Based on distinct avg variance groups.
A. Team 4 x ( 21st, 2nd, 3rd, 4th)
B. Comp 4x (1st, 2nd, 3rd, 4th)
C. ZS 4x but currently 50-60% inaccurate
D. NZ trap 2x ( yes/ no)
E. GA structure 2x (3-1-1-1 yields high open Highndanger **** success density)(3-2-1 yields Lowe rate of open high danger shot success density)

3rd resolution level is at the first point individual analysis has accurate value.

3 x 4 x 4 x 4 x 2 x 2 =768 difrent non complex averages to regress to.
Not 2

All current low resolution binary anslysis you see on here does not take it to a higher resolution ( variable identification) to even come close to being accurate.

It is important to know that dmen generate (G, 1A, 2A) Even Goals at 4 times less rate than forwards.

I developed the 2 critical goal diff sections of analysis.
1. HD area (homeplate) shotsuccess density chart and HD:LD ratio. Watching Flin Flon bombers and SAskatoon blades of late 60’s and early 70’s and continued watching the Tier2 PA raiders.
Their are elite Dmen out their who keep a high % of corsi released to perimeter.

2. Identification of shot quality ( Open/ Closed shots) ( theorem)
Watching a 7 yr old Ron Gunville move like a table hockey goalie.( theorem)
Their was a high rate of hit goalie balls.


Open shot is a shot that hits open space in net elevation requiring
A goalie to make a save. The have a shot success density >0

Closed shot is a shot that hits a goalie moving with the puck like Ronnie Gunville did. 0% chance of going in. Density =0

Their was a jump in save% when goalies started to copy the 2 goalies that first imitated Table Hockey goalie movement.
JVB & ROY.

Ron Gunville is the current Director of Player Personel of the WHL champ PA Raiders.

Theory: Their are Elite Dmen ( who cause the highest % of 0% corsi ( 0%chance) per corsi faced.

Theory: 0% Corsi = ( blocks + forced misses + closed shots)
Theory: Only open Corsi which become open shots are scoreable.

Elite 0% Corsi Dmen like Kris Russell, Calvin DeHaan.

Theory: So all Closed shots must be removed from analysis to get a true Open HD shot density Chart ( map) 768 averages must regress to.

I could go into further multivariable levels of resolution.

Suffice to say.
Theorem: war works in Baseball cause they have a series of binary result actions. All teams try to perform the large win % factors, the same. So the small margin affects have a more valued value in Baseball.

WAR & GAR does not work cause the 5 large win% factors are not played the same by each GM.
Which causes the high group of averages. At higher resolution.
Until hockey starts to play those situations all the same.
Marginal measures used in baseballs war will not be available.

I ran into D. Sutter at Cactus corner truck stop Corner of Highway 9 and south highway 36 near Hanna, Alberta.
He was headed to his nieces wedding and wanted a Calgary sun.
I was putting 40k into kids education doing the 2 hr paper distribution run daily.

I congratulated him on winning his 2 cups and replace Mitchell and Reghr minutes with the best Open HD shot D prospect in the game. Brayden Mcnabb.

Sutter said “ you like analytics.”
“ yes, come up with a lot of theories”
Sutter, “ we are turning them into ........ robots”

I am trying to get everyone to play the high % win factors the same.
Turning them into Robots.

All these tools are a nice ways to learn the simple binary analytical process.
But most are well above 60% in accurate.

Some embarrassing bad at an individual player analysis.

My Wife is a award winning sports page editor for Post Media.
The only paper chain that asks their reporters to be accurate.
She told me they were taught to write at a grade 3 readers level of comprehension.

Their are grade 3-5 level mistakes in the hockey analytic community.

Continue to try to learn.

But remember 60+% inaccurate for most.

This one gets its own chapter.

Some pretty hot takes here...

Let's start with this

Theory: 50-60% of Entry & corsi against is affected by (SOTF) shift on the fly with or without pocession.
Theory: Entry & Corsi is also dependent on variance from NZ trap & No NZ trap.
What!
Currently a 60-75% inaccurate Measure of corsi.
That is a 60-75% error that translates to forward: corsi for, Fenwick for,Closed shot for, open shot for.

What does this mean? Yes, most players start most of their shifts on the fly without the puck, so it should have a big impact on Corsi....

And yes, neutral zone forechecks (trap) obviously impact entries and possession but again what's the point you're making?

 

[Mickey Marner]

Some pretty hot takes here...

Let's start with this



What does this mean? Yes, most players start most of their shifts on the fly without the puck, so it should have a big impact on Corsi....

And yes, neutral zone forechecks (trap) obviously impact entries and possession but again what's the point you're making?

He's saying treating line changes as binary (on-ice; off-ice) is problematic when the majority of shift changes are during live action. This is one of my main criticisms of RAPM.

Zone entry and corsi numbers will be affected by where the defending team chooses to challenge the offensive team. In the case of the trap, the goal is to force a dump-in. Where as, say, Shea Weber allows the zone entry and challenges the shooter in closer proximity to the net. I'd have to see some unencumbered data to draw any conclusions on this point.

 

[Filthy Dangles]

He's saying treating line changes as binary (on-ice; off-ice) is problematic when the majority of shift changes are during live action. This is one of my main criticisms of RAPM.

Zone entry and corsi numbers will be affected by where the defending team chooses to challenge the offensive team. In the case of the trap, the goal is to force a dump-in. Where as, say, Shea Weber allows the zone entry and challenges the shooter in closer proximity to the net. I'd have to see some unencumbered data to draw any conclusions on this point.

Thanks for clarifying, so as for the first paragraph, is the implication that not all line changes are equal and/or players can make a bad play and leave the ice before the subsequent evens and not be charged for them? I understand the logic but you’d think this is just noise that evens out pretty quickly over an adequate sample size ( like a few games). Please correct me if I misunderstood.

As for the second paragraph,don’t the player(s) who comes onto the ice impact that significantly? They need to read the play and decide if it’s worth it to pursue the puck in the offensive zone or fall back into a trap. Better players will know when there’s a chance to forecheck and will force less dangerous plays when there isn’t. What I’m getting at is isn’t that largely skill based? Or your saying team philosphies and systems impacts this too much and skews the numbers?

 

[Mickey Marner]

Thanks for clarifying, so as for the first paragraph, is the implication that not all line changes are equal and/or players can make a bad play and leave the ice before the subsequent evens and not be charged for them? I understand the logic but you’d think this is just noise that evens out pretty quickly over an adequate sample size ( like a few games). Please correct me if I misunderstood.

I don't consider this noise that evens out. I think the difference generally compounds because people use corsirel & the like. Using an example I'm knowledgeable of: JVR while with the Leafs was given both on-the-fly and post-whistle offensive zone starts & also had a habit of gnawing on his Brick mouthgaurd as he went for a line-change when the puck went the other way. Effectively, this made him a one-zone player. This was incorrectly reflected in CFrel% because he would be rewarded for not participating in winning the puck and abandoning his team once they lost it. Where as Komorov -being his most common off LW- was punished when he came on the ice 100 ft out of defensive position and was instructed to get of the ice if/when the puck went the other way and a scoring chance became available. On the whole, I believe this level of line-matching exists to the detriment of the team, but that's just my opinion.

As for the second paragraph,don’t the player(s) who comes onto the ice impact that significantly? They need to read the play and decide if it’s worth it to pursue the puck in the offensive zone or fall back into a trap. Better players will know when there’s a chance to forecheck and will force less dangerous plays when there isn’t. What I’m getting at is isn’t that largely skill based? Or your saying team philosphies and systems impacts this too much and skews the numbers?

That's a whole lot of info you're expecting to be instantaneously processed. I personally believe team philosophies have a greater affect on worse players than they do on better players, but again, this is just my opinion. So, a player that requires an extremely compartmentalized role -Micahel Grabner- will see far more coaching-related variance than a player -Matthew Tkachuk-that has demonstrated himself to be capable of being the 1st, 2nd or 3rd best player on a competent line.

Again, whether the trap specifically has any impact of this I don't know.

 

[Filthy Dangles]

I don't consider this noise that evens out. I think the difference generally compounds because people use corsirel & the like. Using an example I'm knowledgeable of: JVR while with the Leafs was given both on-the-fly and post-whistle offensive zone starts & also had a habit of gnawing on his Brick mouthgaurd as he went for a line-change when the puck went the other way. Effectively, this made him a one-zone player. This was incorrectly reflected in CFrel% because he would be rewarded for not participating in winning the puck and abandoning his team once they lost it. Where as Komorov -being his most common off LW- was punished when he came on the ice 100 ft out of defensive position and was instructed to get of the ice if/when the puck went the other way and a scoring chance became available. On the whole, I believe this level of line-matching exists to the detriment of the team, but that's just my opinion.

Respect your analysis and opinion but I just can't behind the bad change theory, unless I see some kind of proof showing there is something there. IMO, their cfrels% are better expalined by their relative zone starts. JVR much higher, LK much lower. I recall there being a definite relationship between those 2 proven statistically.