Fair enough. More stats to amend the original stat leads to a more nuanced pic.

Flyers dominated in game other than OT2.
But some of that domination comes from pp effectiveness disparity.
And once that is taken out and OT2 is added in, domination stats improve considerably.

See - stats add value and more stats continue to add value and highlight things.


- PT21

Yeah I wasnt posting that to dispute you or MJL. More so that the xG stat kind of jibed with my eye test that the Flyers weren't being dominated as much as the shot count may have indicated.

I hated all these silly stats at first because I felt like too many writers relied on them almost exclusively as if they were a magic elixir.

But as I started to understand what they were, and what they represented, I learned to just incorporate them in to part of what I see.

They have merit, as do our eyes, and guts.

And that's all I have to say about that.

I will make it spectacularly thrilling for you. You will be enthralled.

You are a dude in an foreign country. You are standing at a bus stop. You happen to know that in that country, all families are man-woman-2 kids. Each kid is a boy or a girl with equal probability.

A beautiful woman comes by. You have no compunctions with chatting up a married woman, so you do your thing. And this woman asks you to guess the gender (same as sex here) of her 2 kids.

You find this question stupid, but you want to play along (no pun intended). But before you go make your guess, a little girl comes by and hugs the woman, and the woman says:" this is my little girl. Now you know the gender of one of my kids. Whats your guess about the gender of the other?"

What should you say?

EDIT: Forgot to add critical piece: no twins in this country.
- PT21

Sorry, barely made it past the first paragraph.

The actual game. The strategies of the opposition. The other players. Time, place and score. Shall I go on?

When confronted, you retreat.
- MJL

Retreating like seeing an annoying acquaintance at the grocery store, bruh.

That acquaintance is you. And you leave no stone unturned on this thread, there is no hiding from your Banya-level annoying ass.

The largeness of the sample cuts down on the noise. You have it in reverse.

And the value of any stat is correlation. If what you said was true, and it was indeed a mediocre stat for the purpose it was being used, it would weed itself out.

Because the correlation would fail.
- PT21

No, that's incorrect. Again, you're not including the nature of the game. A larger sample doesn't cut down on the noise. It just builds up more and more. Getting a larger sample does not remove false data. Here is what you're not understanding. When you look at the analytic data for JVR from a game, you aren't looking at his data. You're looking at the data of the team while JVR was on the ice. You can have good players who have lousy analytics. Simply because he plays with bad players on a bad team. Vice versa you can a have a mediocre player with really good numbers because he plays with good players and or on a good team.
The only way to tell the true story, is to actually look at the game and watch it. See who is carrying play and teammates, and who is a passenger. There are a percentage of players who truly have a large effect on their own numbers but overwhelmingly, the majority of the players in the league are greatly affected by numerous variables such as teammates, opposition and usage.

If the breeze is relevant, it will show up as a factor in the empirical data set that generates the probabilities. If not, it is not relevant.
- PT21

That's false. The breeze is certainly relevant but it's not there all the time. Sometimes the table is tilted, affecting the role of the dice. Sometimes it's tilted one way, sometimes it's the other way. Again, you're too focused on the science and ignoring th e nature of the game and how that effects the data collected which they can't possibly account for.

Fair enough. More stats to amend the original stat leads to a more nuanced pic.

Flyers dominated in game other than OT2.
But some of that domination comes from pp effectiveness disparity.
And once that is taken out and OT2 is added in, domination stats improve considerably.

See - stats add value and more stats continue to add value and highlight things.


- PT21

The value of the stats is that they can help tell you what happened in the game. However they can't tell you why, or who was responsible for it.

You would guess a boy. And you still have a 50-50 shot at being wrong to several levels of precision...assuming there is a fairly large population in this country.
- jmatchett383

Kudos to you for going first.

For many years I taught a class to undergrads that fulfilled a college requirement. Since it was a mathy class I had to always work hard to make it remotely interesting, as the requirement option meant I got plenty of non-mathy inclined students.

I would devote the first class entirely to such riddles to introduce ideas, and this was one of them. And one problem I would face is that some of the students would not dare say their guess publicly at first, simply because they were scared it would be wrong.

Then, as cellphones became the menace of teaching, I had to change the parameters of the story so that the answer was not searchable.

No, that's incorrect. Again, you're not including the nature of the game. A larger sample doesn't cut down on the noise. It just builds up more and more. Getting a larger sample does not remove false data. Here is what you're not understanding. When you look at the analytic data for JVR from a game, you aren't looking at his data. You're looking at the data of the team while JVR was on the ice. You can have good players who have lousy analytics. Simply because he plays with bad players on a bad team. Vice versa you can a have a mediocre player with really good numbers because he plays with good players and or on a good team.
The only way to tell the true story, is to actually look at the game and watch it. See who is carrying play and teammates, and who is a passenger. There are a percentage of players who truly have a large effect on their own numbers but overwhelmingly, the majority of the players in the league are greatly affected by numerous variables such as teammates, opposition and usage.
- MJL

You should look up what noise means in statistics and how it is related to size of sample set.

These things you mention are the kinds of things that the stat needs to survive to retain usefulness: the time of the game, the nature of the coaching, the linemates, the minutes used, the quality of opponents, the style they play, the time of day and so on and so forth. This is the noise. Like your breeze.

If you take a large enough sample set, and you find a correlation between your stat and your desired predictive outcome, then your stat has survived all this noise.

There is a saying in Statistics: A map on a scale of 1:1 is no map at all. There will always be individual variations between the map and the reality - that's what distinguishes the former from the latter. However, over a large enough sample, those variations should cancel out. Else, it is not a useful map.

In our case, if we find that say, ixG is an useful stat, what it means is that it is correlated with actual goals scored in most games.

Does it mean that in every game it will tally? No. Does it mean that it will provide more useful than a random guess? Yes.

Thus, the correct way to look at Giroux's iXG score is to say: on average, players who have such low IxG tend not to score goals. Thus, it has some diagnostic value, which is all that I was saying.

Kudos to you for going first.

For many years I taught a class to undergrads that fulfilled a college requirement. Since it was a mathy class I had to always work hard to make it remotely interesting, as the requirement option meant I got plenty of non-mathy inclined students.

I would devote the first class entirely to such riddles to introduce ideas, and this was one of them. And one problem I would face is that some of the students would not dare say their guess publicly at first, simply because they were scared it would be wrong.

Then, as cellphones became the menace of teaching, I had to change the parameters of the story so that the answer was not searchable.
- PT21

I would guess a boy. But them I already knew there were (for the vast majority) more boys born every year than girls.


I see every birth as a new event. Just because you had a girl or a boy already, has no bearing on the next birth. So its safest to choose boy.

Im weird I know.

They could also potentially look to trade Gostisbehere in a swap of defenseman.
- MJL

I'd rather not do that. Let Ghost earn his spot back. Niskanen's gone in a year. Veteran puck-movers are gonna be a need again soon. If they trade him now for a 'Justin Braun', it would be a shame.

Every single NHL team has a robust analytics dept. The hockey + stats debate was over like 5 years ago. Can't believe you guys are still going on about it.

Occam’s Razor: The simplest explanation is most likely the right one. Parsimony in problem solving in that entities should not be multiplied without necessity.

That's false. The breeze is certainly relevant but it's not there all the time. Sometimes the table is tilted, affecting the role of the dice. Sometimes it's tilted one way, sometimes it's the other way. Again, you're too focused on the science and ignoring th e nature of the game and how that effects the data collected which they can't possibly account for.
- MJL

These things are all included.

Lets say I am looking at whether car accidents occur due to drinking.

The time of day, the gender of driver, the kind of car, the kind of roadway, the nature of drinks, the age of car, and zillions of other things are all relevant factors.

But If I find a correlation between blood alcohol level and car accidents, then those factors are noise for the purpose of my analysis.

Those individual vagaries only refine the conclusion: that alcohol causes accidents. They do not negate it.

You should look up what noise means in statistics and how it is related to size of sample set.

These things you mention are the kinds of things that the stat needs to survive to retain usefulness: the time of the game, the nature of the coaching, the linemates, the minutes used, the quality of opponents, the style they play, the time of day and so on and so forth. This is the noise. Like your breeze.

If you take a large enough sample set, and you find a correlation between your stat and your desired predictive outcome, then your stat has survived all this noise.

There is a saying in Statistics: A map on a scale of 1:1 is no map at all. There will always be individual variations between the map and the reality - that's what distinguishes the former from the latter. However, over a large enough sample, those variations should cancel out. Else, it is not a useful map.

In our case, if we find that say, ixG is an useful stat, what it means is that it is correlated with actual goals scored in most games.

Does it mean that in every game it will tally? No. Does it mean that it will provide more useful than a random guess? Yes.

Thus, the correct way to look at Giroux's iXG score is to say: on average, players who have such low IxG tend not to score goals. Thus, it has some diagnostic value, which is all that I was saying.
- PT21

That is the fallacy of analytics as applied to the game of hockey. The noise is not filtered out just because of a larger sample. The noise is just piled higher and higher. They cannot isolate one player from another on the ice.

You weren't just saying it has some diagnostic value. You used it to say that Giroux wasn't good. If you used it as one tool and gave it the proper weight relative to the quality of the stat, which is a small percentage in your analysis, then maybe you might have had a chance.

Every single NHL team has a robust analytics dept. The hockey + stats debate was over like 5 years ago. Can't believe you guys are still going on about it.


- Tomahawk

I was waiting for you. What gender should you guess?

You weren't just saying it has some diagnostic value. You used it to say that Giroux wasn't good. If you used it as one tool and gave it the proper weight relative to the quality of the stat, which is a small percentage in your analysis, then maybe you might have had a chance.
- MJL

I will give you a proper legal phrasing then.

Statistics that are correlated with point producing performance suggest that Giroux was not likely to be involved in goals last night.

These things are all included.

Lets say I am looking at whether car accidents occur due to drinking.

The time of day, the gender of driver, the kind of car, the kind of roadway, the nature of drinks, the age of car, and zillions of other things are all relevant factors.

But If I find a correlation between blood alcohol level and car accidents, then those factors are noise for the purpose of my analysis.

Those individual vagaries only refine the conclusion: that alcohol causes accidents. They do not negate it.
- PT21

No, all of the variables are not included. Just look at what they use for the expected goal stats.

Quantify this for me. What is the on ice difference between Konecny's xGF% of 40.94 and Niskanen's xGF% of 53.53?

I will give you a proper legal phrasing then.

Statistics that are correlated with point producing performance suggest that Giroux was not likely to be involved in goals last night.
- PT21

Suggest being the key word. Does it also suggest that Giroux did not have a good game?

Suggest being the key word. Does it also suggest that Giroux did not have a good game?
- MJL

You sound like Jim Jordan interviewing Fauci.

Let me rephrase.

Markers which are predictive of outcomes imply that Giroux's presence on ice was unlikely to be causative of positive increases in the Flyers' offensive tally.

No, all of the variables are not included. Just look at what they use for the expected goal stats.

Quantify this for me. What is the on ice difference between Konecny's xGF% of 40.94 and Niskanen's xGF% of 53.53?
- MJL

I am telling you the same thing again and again. All the individual vagaries you mention are not irrelevant. They may well be relevant.

But on average, they do not serve to diminish the usefulness/correlation of the central link.

You sound like Jim Jordan interviewing Fauci.


- PT21

Outstanding. Now I know I'm doing it right.

Let me rephrase.

Markers which are predictive of outcomes imply that Giroux's presence on ice was unlikely to be causative of positive increases in the Flyers' offensive tally.
- PT21

You can rephrase it countless ways, you'd still be wrong.

By the way you commented earlier in the thread on JVR's Expected goals versus Giroux

JVR xGF% 28.61
Giroux xGF% 59.56

I am telling you the same thing again and again. All the individual vagaries you mention are not irrelevant. They may well be relevant.

But on average, they do not serve to diminish the usefulness/correlation of the central link.
- PT21

Yes they do. To the point where it could easily cause a players ratings to fall on either side of the ledger.

Quantify the on ice difference between Konecny and Niskanen

I'd rather not do that. Let Ghost earn his spot back. Niskanen's gone in a year. Veteran puck-movers are gonna be a need again soon. If they trade him now for a 'Justin Braun', it would be a shame.
- Tomahawk

I'd rather not also but there is an expansion draft coming up. I would want a more mobile Justin Braun with a little more puck skill.

Every single NHL team has a robust analytics dept. The hockey + stats debate was over like 5 years ago. Can't believe you guys are still going on about it.


- Tomahawk

I know that you're well aware that the analytics that NHL teams use are not the same that is available publicly.

You are mistaken about the predictive value of probabilities.

Lets say you flip a die one hundred times. You notice that 2/3 of the times, the face shows a # 3 or more.

Does that mean that the next roll will definitely produce a value between 3-6? No.

Does that mean the past pattern you discovered has no value in predicting the future? Also nope.

The value lies in the following: think not only of a yes or no or correct/incorrect answer, but degrees of likelihood.

How likely is it that the next roll will result in a value between 3-6? The answer is twice as likely as the alternative.
- PT21

I hate probabilities. The chances remain the same, always. If you have a 10 sided die, you have a 10% chance to get any number. It never changes. I hate the idea that it does based on previous rolls.

I've lost many DnD characters with that predictive logic. I just rolled a 1, no way its gonna happen again! Boom roll another 1 and die lol