3 of the top 4 Corsi teams in the league are in the Final Four.

[PenguinSuitedUp]

Once again, the advanced stats nerds clearly know nothing about statistics.

There are always biases.

The only bias comes from the user, not the numbers themselves. Do user inputs and user interpretations vary? Yes. But someone who understands statistics will know this and take this into consideration. Everything needs context, but anecdotal eye tests do not provide that reliably.

 

[PenguinSuitedUp]

You do make a great point there. I'm a stats proponent, don't worry. I do think eye test matters, but stats helps influence eye test much more efficiently than vice versa.

In a world where we are being asked to do more and more with less and less time, we need to find more efficient ways to use our time. That’s where these models can help ina sport like hockey. Tell the scouts what to look for. Tell your coaches how their systems translate to the bottom line. Tell your players what part of their game to focus on in development.

 

[PenguinSuitedUp]

Yes - didn't agree on this either. There were 4 Grade A chances from Edmonton on these goals. Bouchard was stoppable, but he also has a 100mph shot...

Klingbergs was weak...

Also, there is a difference between guys like McDavid, Hyman, and Kane vs. average NHLers, which is certainly not modeled...

Dallas had more pressure, but gave up odd-man rushes and grade As all game when Edmonton countered. Edmonton had better scoring opportunities despite the shot difference.

I think if you had NHL Teams advanced stats, it would have told a different story.

1. Bouchard goal was stoppable, but also has one of the hardest most accurate shots in NHL.
2. McDavid was on a 3-1. That is a grade A chance.
3. McDavid 2nd was at top of hashmarks.
4. Hyman 2-1 (with Kane)
5. Hyman breakaway
6. Klingberg - weak goal.

I think the key thing to identify is what assumptions is this model making? IE what is identified as a high danger chance in this model? And how does that compare to other models’ assumptions? What data is the model using to make these assumptions? And has the model been ‘back tested’ to understand how consistent its analysis is?

There’s validity in the questions you ask - they’re the right ones- but those again, to the point of this thread - point to user interpretation and input rather than ‘numbers are for nerdzzzzz and they’re garbage’.

 

[stealth1]

It doesn't seem like the advanced stats nerds do.

They love pointing out how the eye test is biased while completely ignoring the fact that advanced stats are also biased.

They also tend to be the ones who understand hockey the least and require numbers to tell them what they are watching.

The one advanced stats nerd even said that "advanced stats teach your eyes what to watch"

Said the person who has no idea what they're watching.

Advanced stars aren't everything but they do help help tell the story. They do have a place in hockey.

As bad as some fans are pushing advanced stars, the old boys fans that dismiss them are worse. Look at how many of you are quickly dismissing the OP of this thread.

 

[Pucklington]

I think the key thing to identify is what assumptions is this model making? IE what is identified as a high danger chance in this model? And how does that compare to other models’ assumptions? What data is the model using to make these assumptions? And has the model been ‘back tested’ to understand how consistent its analysis is?

There’s validity in the questions you ask - they’re the right ones- but those again, to the point of this thread - point to user interpretation and input rather than ‘numbers are for nerdzzzzz and they’re garbage’.

True. But considering some models were showing Edmonton under 3 XGF and Dallas above consdering Edmonton had had a break away, semi break away, 2 on 1, and a 3 on 1.

Saw that it had high danger 20 - 19 in favor of Edmonton on HD chances. So how they do XGF is a little strange.

Conner Brown on a 3-1 is not McDavid on a 3-1 as example.

Advanced stats are good and relevant, regardless. But should correlate with eye test from unbiased viewing.

 

[Golden_Jet]

As bad as some fans are pushing advanced stars, the old boys fans that dismiss them are worse. Look at how many of you are quickly dismissing the OP of this thread.

Lots of bias in OP like has been pointed out already, just look at the sample size. Then goes on to say how predictable it is.

 

[bert]

You can’t quantify ANYTHING with your eyes.

And here’s a part everyone is missing. You can GAIN context from statistics. You want to talk about using your eyes? Proper statistical models will train your eyes on where to look. You can’t possibly watch every game out there, or even everything that occurs in one game. But you can use models to tell you where to look, so that you can prioritize what you are looking for/watching more efficiently. In order of operations, analytics first, eye test second.

Absolutely absurd comment. Proper experience analysis for the sport that comes from playing it, being coached in it. Understanding situational decisions players make because you have physically been learning it your entire life. Knowing in game systems that are being played and which players are executing it. These are simply aspects of the game that you cant pick up unless you have played it and lived it.

The ability to use this type of experience and knowledge first hand then secondly adapting it and using advanced analytics as a secondary tool is how the most successful teams are run.

This is an outrageous attempt to try and justify someone who never played the sport has some sort of superior outlook of what they are evaluating solely using statistics and its laughable.

Please point out anything fictitious in what you quoted.

They are a worse team. They traded an elite player for a 3rd liner and two mystery boxes. Now they have no game breakers that can get through against Florida. Teams that are rebuilding make those trades not teams that have all their elite players in their primes and they are trying to win the stanley cup. They are a worse team there is absolutely no debate. Necas would be the most creative and biggest game breaker on that team and they dont have him. To suggest he made them better is about as laughable as your ridiculous comment right above this post.

True. But considering some models were showing Edmonton under 3 XGF and Dallas above consdering Edmonton had had a break away, semi break away, 2 on 1, and a 3 on 1.

Saw that it had high danger 20 - 19 in favor of Edmonton on HD chances. So how they do XGF is a little strange.

Conner Brown on a 3-1 is not McDavid on a 3-1 as example.

Advanced stats are good and relevant, regardless. But should correlate with eye test from unbiased viewing.

Bingo.

 

[Edmonton East]

You can’t quantify ANYTHING with your eyes.

And here’s a part everyone is missing. You can GAIN context from statistics. You want to talk about using your eyes? Proper statistical models will train your eyes on where to look. You can’t possibly watch every game out there, or even everything that occurs in one game. But you can use models to tell you where to look, so that you can prioritize what you are looking for/watching more efficiently. In order of operations, analytics first, eye test second.

See, again, folks with this mindset hurt the hockey analytics community far more than they help it. I'm not gonna regurgitate my dozen+ previous points in this thread (that you can't / won't logically refute), so I'll just address the absurdity of this post.

1) Very ironically, many advanced stats are derived off of subjective eyes to some degree. One indisputable example: For decades now, everyone knows certain arenas are MUCH friendlier to shot totals than others. Meaning, there is subjective bias. No, shots aren't an "advanced" stat, but many advanced metrics incorporate shots in some way. The issue of subjectivity then becomes compounded as you get more granular with, say, unscreened shots, traffic shots, HD/LD, etc. Are base stats more programmatic in recent years? Sure, but there is still subjectivity in the puck tracking or counting data/software/formulas used....nevermind the actual weights applied to variables in many "advanced" metrics.

2) Making an all encompassing claim that order of operations should always be analytics first, eye test second is laughably absurd. Like most things in life, it depends. Context is everything. Your stance is so ridiculous I don't even know where to begin. So I guess I'll just ask yet again, if that was the case, why did 42% of the top 12 Corsi teams miss the playoffs entirely? Do you honestly believe a person that religiously follows hockey (THE EYE TEST you mock) would do worse in predicting top performing playoff teams? YOU corsi bros keep claiming this stat alone offers some far superior predictive insights that imbeciles that just watch the game couldn't possibly derive on their own. EXPLAIN. From another post of mine:

Do you honestly believe most people that watch and follow hockey religiously couldn't predict the top 12 contenders with better accuracy than Corsi? Who that watches hockey would say the Preds (11), Pens (10), Utah (6), Flames (7), or my Devs (8....play fell off pre injuries btw)...are top teams in this league? Or even above average? Who would say those teams are better than Corsi #29 Toronto or #18 Dallas? Who would say the Canes are completing their dynasty this year since they've far and away dominated corsi 3 straight years?

 

[JaegerDice]

Absolutely absurd comment. Proper experience analysis for the sport that comes from playing it, being coached in it. Understanding situational decisions players make because you have physically been learning it your entire life. Knowing in game systems that are being played and which players are executing it. These are simply aspects of the game that you cant pick up unless you have played it and lived it.

Even if you believe this, which is fine, that doesn't make any of the skills or values you're listing quantifiable by the definition of the word.

You can make the argument that there is expertise that is valuable and unquantifiable, but nothing you've listed is quantifiable.

 

[Kamus]

There are a variety of expected goals models, and they are all constantly evolving. So to answer your question, we would need to understand how this expected goal model works. But essentially, whatever the input is into this model, it’s quantifying Robertson’s goal as ‘higher danger’ than the other opportunities. But this is where individual talent can play a role, too. Because a guy like McDavid is going to be a whole lot more likely to score on ‘lower danger’ opportunities than other players. So even if you add up the xG of all of McDavid’s scoring opportunities versus the average player and they are equal to the average player’s xG, he’s still going to score more ACTUAL goals. That’s where something like shooting percentage begins to matter, because it will create some variability from the norm.

I understand but my issue with these three specific events is that what type of flawed statistical regression model are they using if the Dallas shot results in more goals then a 3-1 and a breakaway? It doesn’t matter who the players were, even if it was Janmark on both Oilers plays, it would result in more goals as a breakaway and a 3-1 should be one of the highest probabilities of scoring. If you get a chance watch all three goals

 

[Kamus]

Also do these models only take into account where the shot was taken from? Not the situation it was in? And if so, I would argue would be a massive flaw

 

[JaegerDice]

Also do these models only take into account where the shot was taken from? Not the situation it was in? And if so, I would argue would be a massive flaw

Pretty much every xGF model, from the kid working in his basement to the guys paid 7 figures in Florida take in AT LEAST the data provided by scraping the NHL event tracking. That's shot type, shot distance, shot angle, and a bunch of other stuff. Many now also include the 'Edge' data, so they can include context like speed, how long players have been out on shift, etc.

There's a reason public xGF models only have an r-squared of about 0.50 at best (out of 1.00).

That's simply roughly double or triple any other singular stat or combination of 'traditional' stats put together. Not to mention xGF models consistently embarrass pundit predictions.

Different xGF models weigh inputs differently, add different inputs, etc. Every NHL team is working on their on xGF model with their own blackbox of weighting and inputs.

Generally speaking, the finer you slice an event (type of shot, distance of shot, pass before shot, offensive context of shot, defensive context of shot, goalie position on shot), the fewer instances you have to add to your sample per game, per team, per season. Which means you need more time to accrue events until you have enough to provide a statistically reliable value foe each finely sliced scenario.

 

[Kamus]

Pretty much every xGF model, from the kid working in his basement to the guys paid 7 figures in Florida take in AT LEAST the data provided by scraping the NHL event tracking. That's shot type, shot distance, shot angle, and a bunch of other stuff. Many now also include the 'Edge' data, so they can include context like speed, how long players have been out on shift, etc.

There's a reason public xGF models only have an r-squared of about 0.50 at best (out of 1.00).

That's simply roughly double or triple any other singular stat or combination of 'traditional' stats put together. Not to mention xGF models consistently embarrass pundit predictions.

Different xGF models weigh inputs differently, add different inputs, etc. Every NHL team is working on their on xGF model with their own blackbox of weighting and inputs.

Generally speaking, the finer you slice an event (type of shot, distance of shot, pass before shot, offensive context of shot, defensive context of shot, goalie position on shot), the fewer instances you have to add to your sample per game, per team, per season. Which means you need more time to accrue events until you have enough to provide a statistically reliable value foe each finely sliced scenario.

I’m not going to pretend I understand half of what you said…. But I’m am curious for you to comment on the two shots. McDavids 3-1 goal and Dallas lone goal. One has a .19 xgf and the other 0.09 xgf. I generally do support advanced stats but am limited in the technical aspect to them. I just can’t understand the discrepancy between them and feel there is a flaw in the calculation if the result is 0.10 xgf difference between the two. Maybe Moneypuck is using a very basic regression??? Idk

 

[TLEH]

I’m not going to pretend I understand half of what you said…. But I’m am curious for you to comment on the two shots. McDavids 3-1 goal and Dallas lone goal. One has a .19 xgf and the other 0.09 xgf. I generally do support advanced stats but am limited in the technical aspect to them. I just can’t understand the discrepancy between them and feel there is a flaw in the calculation if the result is 0.10 xgf difference between the two. Maybe Moneypuck is using a very basic regression??? Idk

The xG models take that stuff into account but the stat this thread is based on does not take shot quality into account.

 

[KillerMillerTime]

Not shots on goal.

Shot attempts. So it includes blocked shots, missed shots, shots on goal.

The shots themselves aren't as important as the fact that the team has the ability to consistently try shooting demonstrates that they have the puck all the time.

And so long as there's one puck, having the puck all the time not only allows for the team in question to apply pressure, it prevents the other team from scoring goals as well.

It's not everything. I love Carolina, but I think they will lose to Florida by virtue of Florida having more game-breakers (ie, players that need fewer attempts to make big plays and score goals).

Who would have thought that the best defense is a good offense.

 

[JaegerDice]

I’m not going to pretend I understand half of what you said…. But I’m am curious for you to comment on the two shots. McDavids 3-1 goal and Dallas lone goal. One has a .19 xgf and the other 0.09 xgf. I generally do support advanced stats but am limited in the technical aspect to them. I just can’t understand the discrepancy between them and feel there is a flaw in the calculation if the result is 0.10 xgf difference between the two. Maybe Moneypuck is using a very basic regression??? Idk

I don’t use moneypuck as a source for statistics in general. Not familiar with the details of their xGF model or if its even public.

EDIT: When I have a spare moment, Ill look through the output you’re referencing to see if I can provide a better answer regarding their model.

As a general rule though, xGF is a predictive measure used for larger samples. So using the stat as a descriptive stat for what happened in a single completed game is kinda using the wrong tool for the job.

 

[PoutineSniper]

Once again, here is a person who insults useful
mechanisms to teach us things because they lack either the patience or intelligence to understand them.

Once again, here is a person who has no idea what they're watching and so resort to using "advanced stats" to tell them what they're seeing.

Its okay that you dont understand hockey as well as most.

But don't expect us to believe your numbers are so predictive when 42% of the top 12 teams missed the playoffs.

 

[PoutineSniper]

See, again, folks with this mindset hurt the hockey analytics community far more than they help it. I'm not gonna regurgitate my dozen+ previous points in this thread (that you can't / won't logically refute), so I'll just address the absurdity of this post.

1) Very ironically, many advanced stats are derived off of subjective eyes to some degree. One indisputable example: For decades now, everyone knows certain arenas are MUCH friendlier to shot totals than others. Meaning, there is subjective bias. No, shots aren't an "advanced" stat, but many advanced metrics incorporate shots in some way. The issue of subjectivity then becomes compounded as you get more granular with, say, unscreened shots, traffic shots, HD/LD, etc. Are base stats more programmatic in recent years? Sure, but there is still subjectivity in the puck tracking or counting data/software/formulas used....nevermind the actual weights applied to variables in many "advanced" metrics.

2) Making an all encompassing claim that order of operations should always be analytics first, eye test second is laughably absurd. Like most things in life, it depends. Context is everything. Your stance is so ridiculous I don't even know where to begin. So I guess I'll just ask yet again, if that was the case, why did 42% of the top 12 Corsi teams miss the playoffs entirely? Do you honestly believe a person that religiously follows hockey (THE EYE TEST you mock) would do worse in predicting top performing playoff teams? YOU corsi bros keep claiming this stat alone offers some far superior predictive insights that imbeciles that just watch the game couldn't possibly derive on their own. EXPLAIN. From another post of mine:

Do you honestly believe most people that watch and follow hockey religiously couldn't predict the top 12 contenders with better accuracy than Corsi? Who that watches hockey would say the Preds (11), Pens (10), Utah (6), Flames (7), or my Devs (8....play fell off pre injuries btw)...are top teams in this league? Or even above average? Who would say those teams are better than Corsi #29 Toronto or #18 Dallas? Who would say the Canes are completing their dynasty this year since they've far and away dominated corsi 3 straight years?

Stop they're already dead!

But seriously, this is just getting embarrassing at this point.

@Penguin Suited Up clearly can't touch your points and yet the poster still acts so smug.

Posters like that give the analytics community a bad rep.

 

[PoutineSniper]

Absolutely absurd comment. Proper experience analysis for the sport that comes from playing it, being coached in it. Understanding situational decisions players make because you have physically been learning it your entire life. Knowing in game systems that are being played and which players are executing it. These are simply aspects of the game that you cant pick up unless you have played it and lived it.

The ability to use this type of experience and knowledge first hand then secondly adapting it and using advanced analytics as a secondary tool is how the most successful teams are run.

This is an outrageous attempt to try and justify someone who never played the sport has some sort of superior outlook of what they are evaluating solely using statistics and its laughable.


They are a worse team. They traded an elite player for a 3rd liner and two mystery boxes. Now they have no game breakers that can get through against Florida. Teams that are rebuilding make those trades not teams that have all their elite players in their primes and they are trying to win the stanley cup. They are a worse team there is absolutely no debate. Necas would be the most creative and biggest game breaker on that team and they dont have him. To suggest he made them better is about as laughable as your ridiculous comment right above this post.


Bingo.

Imagine thinking you can't qualify anything with your eyes

Like you said, what an absurd comment.

 

[KaseMeOutside]

Breaking: good teams with good players are good in the playoffs

 

[rielledup]

I have a question for the people who think analytics are useless, how do you explain situations like Gudbranson being traded for McCann and Pittsburgh signing Jack Johnson? In both of those situations there was a big divide between analytics people and eye test old school people and the analytics people were right in both cases.

In the Gubranson trade analytics people were saying he has terrible possession numbers and can't get the puck out of his own end and old school people were talking about how he's big and physical and blocks shots. He ended up being terrible in Vancouver and they got rid of him and I remember Ray Ferrero talking about how terrible Gudbranson was with the puck in Vancouver.

The analytics people were very critical of the Jack Johnson signing and he ended up being terrible and Pittsburgh got rid of him as well.