Analytics & fancy stats thread

Mr Snrub

I like the way Snrub thinks!
Oct 12, 2016
5,713
2,410
Almost as good as Travis Zajac: Offensive specialist :laugh:

Or Patrick Kane the liability. It's incredible that Chicago won three cups with that absolute anchor on their team. Especially considering Duncan Keith is a liability too :sarcasm:
 

Anglesmith

Setting up the play?
Sep 17, 2012
46,441
14,715
Victoria
Or Patrick Kane the liability. It's incredible that Chicago won three cups with that absolute anchor on their team. Especially considering Duncan Keith is a liability too :sarcasm:

Brent Seabrook is also a liability.

Maybe it's backwards and it means they're a liability for the opposing team?
 

Bjornar Moxnes

Stem Rødt og Felix Unger Sörum
Oct 16, 2016
11,490
3,942
Troms og Finnmark
5hBDYsj.gif


Seriously, after it was blatantly clear how bad that chart was for defencemen (Radko Gudas the juggernaut), why bother posting the same thing for forwards?

For what it's worth, there's nothing wrong with the graph if you take out the labels on the four quadrants. If you take those labels out, all this is is information (though even with that, you still have to do some digging to find out how chances are being categorized in this approach). As soon as those labels are added, the information turns into opinion, and in this case, probably not a very good one.

Except the opinions are created from information. A player that has a positive high danger chance generation and a negative high danger chance allowance is naturally going to have a juggernaut effect on his team, because he's benefitting them on both sides. A player with a negative high danger chance generation and a positive allowance is well a liability aspect on his team's high danger chances.
 

Flames Fanatic

Mediocre
Aug 14, 2008
13,329
2,888
Cochrane
Except the opinions are created from information. A player that has a positive high danger chance generation and a negative high danger chance allowance is naturally going to have a juggernaut effect on his team, because he's benefitting them on both sides. A player with a negative high danger chance generation and a positive allowance is well a liability aspect on his team's high danger chances.

Either you are saying you agree with it, or you agree that it is flawed. If it's the former, many of us have a bone to pick with you and the chart. If it's the latter, why keep posting them?

There are many good ways to look at the individual advanced stats that the chart is presenting. Combined in a GWAR chart is not one of them.
 

Anglesmith

Setting up the play?
Sep 17, 2012
46,441
14,715
Victoria
Except the opinions are created from information. A player that has a positive high danger chance generation and a negative high danger chance allowance is naturally going to have a juggernaut effect on his team, because he's benefitting them on both sides. A player with a negative high danger chance generation and a positive allowance is well a liability aspect on his team's high danger chances.

But this is clearly shown to be an oversimplification, so why still hold faith in this statement?

It's like.... *tries to think of some analogy to make his point*... it's like if I were to state that students who spend more hours studying are smarter. So I do a survey of students and how much they study, and decide who the smartest students are based on that. I call those a form of advanced grades. Actual test scores can fluctuate due to random things like how a student is feeling on a particular day or just blanking. But my metric is better; more advanced. And it's a completely reasonable hypothesis, that has some logic behind it, even if it's simplistic.

However, if it turns out that the kid who everyone knows is the least competent spends every waking hour studying, and the kid who aces every test and answers every question correctly doesn't study at all, it puts a real hole in my metric. And if I continue pushing this metric, it's going to give advanced grades a bad name. And I'm going to lose a lot of credibility.
 

Lunatik

Normal is an illusion.
Oct 12, 2012
56,176
8,337
Padded Room
But this is clearly shown to be an oversimplification, so why still hold faith in this statement?

It's like.... *tries to think of some analogy to make his point*... it's like if I were to state that students who spend more hours studying are smarter. So I do a survey of students and how much they study, and decide who the smartest students are based on that. I call those a form of advanced grades. Actual test scores can fluctuate due to random things like how a student is feeling on a particular day or just blanking. But my metric is better; more advanced. And it's a completely reasonable hypothesis, that has some logic behind it, even if it's simplistic.

However, if it turns out that the kid who everyone knows is the least competent spends every waking hour studying, and the kid who aces every test and answers every question correctly doesn't study at all, it puts a real hole in my metric. And if I continue pushing this metric, it's going to give advanced grades a bad name. And I'm going to lose a lot of credibility.

Best. Post. Ever.
 

Dertell

Registered User
Jul 14, 2015
2,923
474
This is a really bad post, regardless of what I think of the principles of scoring chances. Also, literally anyone who says "X is Y /s. stats suck" doesn't have anything worthwhile to say.
 
Last edited:

Flames Fanatic

Mediocre
Aug 14, 2008
13,329
2,888
Cochrane
This is a really bad post, regardless of what I think of the principles of scoring chances. Also, literally anyone who says "X is Y /s. stats suck" doesn't have anything worthwhile to say.

Do you feel that GWAR charts are useful?

I don't feel like AS's post is purely "advanced stats suck hur derp a durp", more so pointing at that you can't take a single stat in a multifaceted argument as the sole truth.

Just because I vote for candidate A doesn't mean I don't like candidate B in politics sort of thing.

I'm coming around more on advanced stats simply because I've encountered them enough to understand what each means more at first blush, and how to take them into context of other stats such as ice time, quality of comp etc. to paint a fuller picture. I don't feel like GWAR chart is painting a fuller picture, it's painting a picture that isn't fully fleshed out and is highly lacking in some regards, whether it be context, explanation, missing elements, whatever.

I still take issue of people taking one stat such as Corsi and saying "well his corsi is 48%, therefore he sucks. Period. Nope, don't watch him, don't need to". There is far, far too much of that on HF, equally as much as people like MM who despise and argue against advanced stats daily.
 

InfinityIggy

Zagidulin's Dad
Jan 30, 2011
36,068
12,852
59.6097709,16.5425901
Do you feel that GWAR charts are useful?

I don't feel like AS's post is purely "advanced stats suck hur derp a durp", more so pointing at that you can't take a single stat in a multifaceted argument as the sole truth.

Just because I vote for candidate A doesn't mean I don't like candidate B in politics sort of thing.

Literally no one is saying 'advanced stats hurr durr'. It's that we are poking holes in an incredibly flawed graph and since there is no good defence of said stats or graph, we get replies like the quoted.
 

Flames Fanatic

Mediocre
Aug 14, 2008
13,329
2,888
Cochrane
Literally no one is saying 'advanced stats hurr durr'. It's that we are poking holes in an incredibly flawed graph and since there is no good defence of said stats or graph, we get replies like the quoted.

I know, but to me it sounds like Dertell feels like that is what is being said.

Despite the animosity between the advanced stats crowd and the advanced stats haters, I would rather we argue it via intelligent discussions than dropping one liners that are not so subtly insulting. We've got a long summer ahead of us.
 

Anglesmith

Setting up the play?
Sep 17, 2012
46,441
14,715
Victoria
Was Dertell actually referring to my post, though? His reply doesn't really fit what I was saying, but it doesn't make sense referring to VMS' post either, so I'm not sure.

All I was pointing out is that if a particular interpretation is clearly leading to ridiculous results, it should probably be fixed or abandoned altogether. But Dertell seems to be referring to someone attacking advanced stats in general, so hence my confusion. :huh:
 

Bjornar Moxnes

Stem Rødt og Felix Unger Sörum
Oct 16, 2016
11,490
3,942
Troms og Finnmark
But this is clearly shown to be an oversimplification, so why still hold faith in this statement?

It's like.... *tries to think of some analogy to make his point*... it's like if I were to state that students who spend more hours studying are smarter. So I do a survey of students and how much they study, and decide who the smartest students are based on that. I call those a form of advanced grades. Actual test scores can fluctuate due to random things like how a student is feeling on a particular day or just blanking. But my metric is better; more advanced. And it's a completely reasonable hypothesis, that has some logic behind it, even if it's simplistic.

However, if it turns out that the kid who everyone knows is the least competent spends every waking hour studying, and the kid who aces every test and answers every question correctly doesn't study at all, it puts a real hole in my metric. And if I continue pushing this metric, it's going to give advanced grades a bad name. And I'm going to lose a lot of credibility.

Honestly you make a good point AS, I for one have to say analogies are great and are generally comparable, but at the same time it's not the same scenario, there will always be differences. I personally think high danger impact is a flawed method, because it's very team based and some teams don't have the depth and lines as others, which make certain players look better or worse. Nevertheless from a pure team standpoint and when used with other stats, it can be more reliable. For example, Pietrangelo takes on ridiculously difficult competition with a poor partner in Bouwmeester, meaning if he had an easier workload and a better partner he'd most likely have much better stats. Same goes for Brodie, if Brodie didn't have to carry the corpses of Wideman and Engelland, he'd have better stats as well. Lastly, HDCI is good for seeing how effective certain players/pairings/lines are relative to their team mates. Whether they are taking too difficult competition (Especially if certain players taking weaker ones destroy their competition), whether they have no choice but to take difficult competition, etc., etc.
 

Bjornar Moxnes

Stem Rødt og Felix Unger Sörum
Oct 16, 2016
11,490
3,942
Troms og Finnmark
Was Dertell actually referring to my post, though? His reply doesn't really fit what I was saying, but it doesn't make sense referring to VMS' post either, so I'm not sure.

All I was pointing out is that if a particular interpretation is clearly leading to ridiculous results, it should probably be fixed or abandoned altogether. But Dertell seems to be referring to someone attacking advanced stats in general, so hence my confusion. :huh:

He might be taking to both of us lol. Anyways I won't post anymore HDCI on Flame related things anymore, but other sub forums have shown interest (And also extreme criticism).
 

Lunatik

Normal is an illusion.
Oct 12, 2012
56,176
8,337
Padded Room
Was Dertell actually referring to my post, though? His reply doesn't really fit what I was saying, but it doesn't make sense referring to VMS' post either, so I'm not sure.

All I was pointing out is that if a particular interpretation is clearly leading to ridiculous results, it should probably be fixed or abandoned altogether. But Dertell seems to be referring to someone attacking advanced stats in general, so hence my confusion. :huh:
Pretty sure it was a thinly veiled shot at me since I think people who preach nothing but advanced stats are those too stupid to understand the game
 

Anglesmith

Setting up the play?
Sep 17, 2012
46,441
14,715
Victoria
He might be taking to both of us lol. Anyways I won't post anymore HDCI on Flame related things anymore, but other sub forums have shown interest (And also extreme criticism).

There is nothing wrong with the information, as I say. The problem is labelling the quadrants as absolutes like that. As you say, there are many mitigating factors at play. If you take the labels off, the information is realistically above reproach- it's just objective data*. Then if you want to base an argument off of it, you can choose to consider other factors as you wish.

*of course, you need to know how high-danger chances are being defined
 

Bjornar Moxnes

Stem Rødt og Felix Unger Sörum
Oct 16, 2016
11,490
3,942
Troms og Finnmark
There is nothing wrong with the information, as I say. The problem is labelling the quadrants as absolutes like that. As you say, there are many mitigating factors at play. If you take the labels off, the information is realistically above reproach- it's just objective data*. Then if you want to base an argument off of it, you can choose to consider other factors as you wish.

*of course, you need to know how high-danger chances are being defined

It's really just distance, and others have said advanced stats valuing all shots are more valuable because a high danger shot may not be more dangerous than a low danger shot depending on the circumstances. I just hope our top 2 defense pairings all post juggernaut high danger chances, and our third pairing despite having poor HDCI, overall still make one of the best third pairings in the league.
 

Fig

Absolute Horse Shirt
Dec 15, 2014
12,955
8,449
I've been mulling over GAR and WAR and I think I've finally figured out how it's useful. In a nutshell...

The following must occur for GAR/WAR to be useful
1. You cannot compare across positions. It must be separated into F/D/G
2. You cannot compare across teams. It must be within the same organization

What GAR/WAR measures
1. The difficulty of replacement of a player
2. How heavily the team leans on said player to do anything

What charts have shown us
1. CMD has the highest GAR/WAR in the league
2. Klefbom has one of the highest GAR/WAR in the league
3. Draisaitl is up there too


Based on these results, what can we conclude?
1. If CMD, Draisaitl and Klefbom aren't playing, the Oilers are in trouble
2. If an opposing team is so inclined, shut down those guys
3. As players on their team, protect these guys

If these results weren't obvious enough, make sure you read some of the other hilariously bad conclusions other GAR/WAR numbers indicate... Seriously, why do people keep using that garbage metric?

And why do I feel like GAR/WAR is actually useful in comparison to Vali Maki's chart? If Burke's illumination metaphor is apt that chart feels like shining a light inside a house of mirrors. Light everywhere, but mostly not in the places where I expect it.
 

Dertell

Registered User
Jul 14, 2015
2,923
474
What a compelling argument.
not everyone can elaborate like you
Almost as good as Travis Zajac: Offensive specialist :laugh:
What else was I supposed to say?

I mean, yeah, charts have datas in them, some of which must challenge a common positions regarding something or someone. Most of you know this already; "I don't need your fancy stats to tell me X is good", etc. There's no point to charts which do nothing but repeat consensus. People are complaining about 5-6 data points here. Out of what? 150? That's a well within any acceptable margins. If anything, it's not, but that's because 5-6's not enough to be meaningful.

I'm possibly the biggest critic of scoring chances in the forum. I've already discussed why in this thread: it's data binning, you can't WOWY it, it's built on false assumptions, it's built on anti-corsi sentiments and so on. They are completely useless and I long for the days they leave the discourse and are replaced by carry-ins%, stretch passes, pass leading to shots and so on.

Do you feel that GWAR charts are useful?
short answer: I think a WAR model will or at least should be the future of hockey analytics.

long answer:
I mean, winning close games is not a repeatable skill, so predicting GF% is extremely similar to predicting standing points.

However, goal differential is still extremely flawed for a variety of reasons. A WAR (win above replacement) stat for line usage, systems, skaters and goaltenders could be even closer to predicting performance and success than predicting GF%. Essentially, it's less about chemistry and fits; the team with better personnel will have more success than the higher projected GF% team based on one stat or another.

This isn't "the same but better" like score-adjusted for Close or xGoal for scoring chances. It's adjusting forecast to and forecasting changes in the roster and making better predictions as a result. By giving everything a WAR number, you can make a sum and give teams a specific total value, do less data binning than necessary (PP, PK, goaltending) and add or remove variables (players) as things change.
This is what I said half a year ago in this thread and I still feel this way. I used to believe should be, in programming-speak, a double[] instead of simple double variable. However, I feel like I exaggerated the importance of chemistry, roles, needs, fitting systems, etc and underestimed the chaotic nature of the game. I have a more "teams need the best players and that's it" black and white mindset.
 

Flameshomer

Likeaholic
Aug 26, 2010
3,830
1,039
Edmonton
not everyone can elaborate like you

What else was I supposed to say?

I mean, yeah, charts have datas in them, some of which must challenge a common positions regarding something or someone. Most of you know this already; "I don't need your fancy stats to tell me X is good", etc. There's no point to charts which do nothing but repeat consensus. People are complaining about 5-6 data points here. Out of what? 150? That's a well within any acceptable margins. If anything, it's not, but that's because 5-6's not enough to be meaningful.

I'm possibly the biggest critic of scoring chances in the forum. I've already discussed why in this thread: it's data binning, you can't WOWY it, it's built on false assumptions, it's built on anti-corsi sentiments and so on. They are completely useless and I long for the days they leave the discourse and are replaced by carry-ins%, stretch passes, pass leading to shots and so on.


short answer: I think a WAR model will or at least should be the future of hockey analytics.

long answer:

This is what I said half a year ago in this thread and I still feel this way. I used to believe should be, in programming-speak, a double[] instead of simple double variable. However, I feel like I exaggerated the importance of chemistry, roles, needs, fitting systems, etc and underestimed the chaotic nature of the game. I have a more "teams need the best players and that's it" black and white mindset.

Found the oilers fan.
 

Tkachuk Norris

Registered User
Jun 22, 2012
15,600
6,639
Yeah my guess is analytics are very useful when used with extreme detail/multiple contexts. But the simplistic perspectives people provide from things like looking at exclusively shot differentials are just a waste of time, and generally lead people to create some false narrative. It's really just a microcosm of our societies obsession with data, even when that data is providing a limited perspective.

Cody Franson. Stat nerds love him. Coaches hate him. Why? Because he makes a lot of stupid plays that end up in his net. Much like Wideman all those years. A stat page says Wideman and Franson are okay. But the eyeball test shows they are actually junk. Basically the opposite of someone like Stone.
 

Dertell

Registered User
Jul 14, 2015
2,923
474
Found the oilers fan.
huh?

Yeah my guess is analytics are very useful when used with extreme detail/multiple contexts. But the simplistic perspectives people provide from things like looking at exclusively shot differentials are just a waste of time, and generally lead people to create some false narrative. It's really just a microcosm of our societies obsession with data, even when that data is providing a limited perspective.

Cody Franson. Stat nerds love him. Coaches hate him. Why? Because he makes a lot of stupid plays that end up in his net. Much like Wideman all those years. A stat page says Wideman and Franson are okay. But the eyeball test shows they are actually junk. Basically the opposite of someone like Stone.
I agree it's really just a microcosm of our society, but not for those reasons. People hates datas and studies, until they confirm their biases regardless of how they've been conducted. That, and datas should be used to build and change things, not accept the harsh "reality" as it is and develop an easy, convenient cynical mentality. That is when the many people who need help need optimism and can't afford to have those who can help to accept the harsh "reality" as it is...

Wideman was never was analytic darling, unlike Franson. These are two different players. Franson's ability to break up plays and puck battle in the d-zone is one of his best tools.
 
Last edited:

Fig

Absolute Horse Shirt
Dec 15, 2014
12,955
8,449
short answer: I think a WAR model will or at least should be the future of hockey analytics.

I don't think WAR/GAR has more than limited application even if they "fix it". IMO fundamentally, it's idea from baseball is not applicable. Personally, I think the future involves the analytics community needing to agree upon about 5-7 stats/skills to measure. A rock paper scissors format for the stat/skills will allow us to better figure out things like chemistry, match ups, player style, strengths and weaknesses, utility in certain systems etc. and be able to better analyze players from stats.

TL;DR IMO the fundamentals of why WAR/GAR work in baseball are not applicable in hockey. This is due to players adhering to each other in hockey vs players adhering to their position on the ground in baseball.

Player chemistry and pure random chance arguably isn't a thing in baseball. It's a series of repeatable events and each player doesn't stray far from their role. Hockey has chemistry and randomness in spades due to the fluidity of the game. Any player can pretty much go where ever they want.

IMO, WAR/GAR is a calculation of how well a piece fits into a role and can be relied upon. In baseball, a 3rd baseman from one team to another is IMO more connected to the ground he stands on over the players he plays with. That is why a player with higher GAR/WAR can be acquired and plugged into another team reasonably successfully. Dare I say baseball is akin to monochrome colored lego. Swap around any piece and it can still look reasonably ok. Hockey is more akin to colored lego. How a piece looks in comparison to the rest of the pieces is much more important.

I say chemistry isn't a thing in baseball but it's not true. There is chemistry, but far less than hockey to the point it's ok to ignore it. We cannot ignore chemistry because players are more connected to one another than the ice they skate on.

Found the oilers fan.

Not necessary man.

The Oilers use it a lot, but I've been paying attention to it since the Stone deal, a little bit before they started posting it often to sell the narrative Klefbom is great. Klefbom is good, but WAR/GAR isn't doing him any favors and a bad way to support their opinion due to it's horrible flaws.
 

Ad

Upcoming events

Ad

Ad

-->