2017-18 stats and underlying metrics thread [Mod: updated season]

[Daximus]

Well the Leafs certainly use Corsi....what exactly is a heavy shift? Sounds "fancy"

Heavy shifts is a prolonged shift in the offensive zone. I don't think anyone knows the exact time a heavy shift is counted as in TML land but I'd imagine it's around 40 seconds or more. It's counted both for and against.

 

[WPGChief]

This recent thread by Dellow is interesting, in which he was looking at shots below the dots by defenceman before going on this tangent. Some images because it's a long thread:

TL;DR - There has been posts before on how this, most notably "that defensemen’s control of save percentage is minimal enough that adding on-ice save percentage effects do not add value in player evaluations other than looking at those likely to regress to the mean". Cumulatively, however, Buff is noticeably off. Why, and is this fixable? EDIT: And maybe a third question: if your solution is to make him roam less, does that take away from is uniqueness and contribution on offence?

 

[ps241]

This recent thread by Dellow is interesting, in which he was looking at shots below the dots by defenceman before going on this tangent. Some images because it's a long thread:

TL;DR - There has been posts before on how this, most notably "that defensemen’s control of save percentage is minimal enough that adding on-ice save percentage effects do not add value in player evaluations other than looking at those likely to regress to the mean". Cumulatively, however, Buff is noticeably off. Why, and is this fixable? EDIT: And maybe a third question: if your solution is to make him roam less, does that take away from is uniqueness and contribution on offence?

This is worth a deep dive if I had the talent.

 

[Whileee]

I recall asking whether Buff's on-ice save percentage might indicate his propensity to give up too many good scoring chances. I think the response from Garrett and others was that save % was random, so can't be pinned on Buff. But I might be remembering this wrong.

 

[WPGChief]

I recall asking whether Buff's on-ice save percentage might indicate his propensity to give up too many good scoring chances. I think the response from Garrett and others was that save % was random, so can't be pinned on Buff. But I might be remembering this wrong.

It's generally true season-to-season, in which a defenceman's impact on SV% is generally unrepeatable (Hockey-Graphs post #1, post #2, post #3). It's why you see the difference in SV% of the rest of the Jets defence whether they are on or off the ice is like ~0.3%.

Looking at totals and the numbers cumulatively, however, Buff is a consistent negative outlier - and considerably so. Hell, even Mark Stuart's SV% averages out, but Buff's doesn't. That is where Dellow's (and my) interest is piqued. I do not think it was affected much by his time when he played forward (only 30-40 games or so?), so it's most likely is defensive capabilities. Is it because he jumps up in the play so often, creating 2-on-1s?


My quick theory is that Maurice's (and to a small extent Noel's) system is a very hard and rigid 3F-2D, circle around the boards for a period of time, see if their players are tired from skating around and pass it into the slot for a shot. Buff (and sometimes Trouba) are the only D on this team that like to become active all the time, and that is why we see their great offensive numbers, but the system doesn't adjust or play into this well - if they fail when they get active, the forwards trying to cover don't know what to do other than putting their stick out and skating backwards.

If you want your special players like Buff to be active, your whole team has to have a system designed around that, not just allowing him to do it every once in a while.

 

[Whileee]

It's generally true season-to-season, in which a defenceman's impact on SV% is generally unrepeatable (Hockey-Graphs post #1, post #2, post #3). It's why you see the difference in SV% of the rest of the Jets defence whether they are on or off the ice is like ~0.3%.

Looking at totals and the numbers cumulatively, however, Buff is a consistent negative outlier - and considerably so. Hell, even Mark Stuart's SV% averages out, but Buff's doesn't. That is where Dellow's (and my) interest is piqued. I do not think it was affected much by his time when he played forward (only 30-40 games or so?), so it's most likely is defensive capabilities. Is it because he jumps up in the play so often, creating 2-on-1s?


My quick theory is that Maurice's (and to a small extent Noel's) system is a very hard and rigid 3F-2D, circle around the boards for a period of time, see if their players are tired from skating around and pass it into the slot for a shot. Buff (and sometimes Trouba) are the only D on this team that like to become active all the time, and that is why we see their great offensive numbers, but the system doesn't adjust or play into this well - if they fail when they get active, the forwards trying to cover don't know what to do other than putting their stick out and skating backwards.

If you want your special players like Buff to be active, your whole team has to have a system designed around that, not just allowing him to do it every once in a while.

I don't think it's just activating offensively for Buff. He plays way too casually too often. I commented on this often last year.

 

[Weezeric]

I don't think it's just activating offensively for Buff. He plays way too casually too often. I commented on this often last year.

I think buff simply plays too much. He would be optimum playing 22 mins a night in a purely offensive role and let trouba take the toughs. He could put up 20 goals, 60 points imo

 

[Bob E]

Have said it before and will say it again, this team will take a step forward defensively and move into a serious playoff contender once Buff is moved. This team doesn't seem to be able to compensate for some guys not being solid defensively. Ehlers and Wheeler have a ways to go defensively too.

Trouba being retained is very important to this team.

 

[Festinator]

I could almost believe that Buff is a major problem, except that the charts also claims that guys like Stuart are the solution sooo...

To me it's simple. Buff plays a lot more than any of our other defencemen. Buff is the most offensive of all of our defencemen. He's always going to allow more goals while he's on the ice, but he's going to produce more than he allows.

 

[WPGChief]

I don't think it's just activating offensively for Buff. He plays way too casually too often. I commented on this often last year.

That's part of his "offensive uniqueness" though. Him playing casually is him skating around the OZ with the puck on his stick and barely anyone can get it from him. But I think part of that is the allegory of the "dog chasing the car" - he (and his teammates) doesn't know what to do next once he skates once or twice around the opposition's goalie.

I think buff simply plays too much. He would be optimum playing 22 mins a night in a purely offensive role and let trouba take the toughs. He could put up 20 goals, 60 points imo

Could be part of it. His average ice time (all strengths) has gone from 22:41 in the 14/15 season, to 25:12 in 15/16, to 27:27 in 16/17. Part of that could be Maurice's reliance on his top guys once the team goes down early, and of course injuries never help. But then how do you split it when Myers gets healthy, 20/20/20 for Buff/Trouba/Myers? No one is gonna be happy about that - unless they're winning.

Have said it before and will say it again, this team will take a step forward defensively and move into a serious playoff contender once Buff is moved. This team doesn't seem to be able to compensate for some guys not being solid defensively. Ehlers and Wheeler have a ways to go defensively too.

Trouba being retained is very important to this team.

I fail to see the reasoning behind this. You're essentially saying "once you remove a perceived defensive liability like Buff, these other perceived defensive liabilities of the team won't be liabilities anymore." If the team is unable to compensate activating their D (and I will mention, which is what Nashville rode all the way to the Finals this season), is that not a statement on coaching?

I could almost believe that Buff is a major problem, except that the charts also claims that guys like Stuart are the solution sooo...

To me it's simple. Buff plays a lot more than any of our other defencemen. Buff is the most offensive of all of our defencemen. He's always going to allow more goals while he's on the ice, but he's going to produce more than he allows.

It's not so much the charts saying Stuart is the solution, but rather in the vein of the Hockey-Graph posts that everyone - including Stuart - usually averages out. But Buff is a significant and consistent outlier.

That's generally true, and I believe that he does produce more than he allows, but I don't have access to the numbers at the moment to back it up. Looking at Dellow's work, you would think that it is almost contrary to this belief: he might be scoring more, but he's actually allowing significantly more. Plus, when looking at the team as a whole, one might say that it's time to sacrifice his somewhat increased scoring, and that our very talented forwards are now experienced enough to do that on their own. If Buff focuses on his defensive part of the game and gets back to a league average, sure maybe he individually doesn't score as much, but maybe it's worth the sacrifice.

 

[Whileee]

That's part of his "offensive uniqueness" though. Him playing casually is him skating around the OZ with the puck on his stick and barely anyone can get it from him. But I think part of that is the allegory of the "dog chasing the car" - he (and his teammates) doesn't know what to do next once he skates once or twice around the opposition's goalie.



Could be part of it. His average ice time (all strengths) has gone from 22:41 in the 14/15 season, to 25:12 in 15/16, to 27:27 in 16/17. Part of that could be Maurice's reliance on his top guys once the team goes down early, and of course injuries never help. But then how do you split it when Myers gets healthy, 20/20/20 for Buff/Trouba/Myers? No one is gonna be happy about that - unless they're winning.



I fail to see the reasoning behind this. You're essentially saying "once you remove a perceived defensive liability like Buff, these other perceived defensive liabilities of the team won't be liabilities anymore." If the team is unable to compensate activating their D (and I will mention, which is what Nashville rode all the way to the Finals this season), is that not a statement on coaching?



It's not so much the charts saying Stuart is the solution, but rather in the vein of the Hockey-Graph posts that everyone - including Stuart - usually averages out. But Buff is a significant and consistent outlier.

That's generally true, and I believe that he does produce more than he allows, but I don't have access to the numbers at the moment to back it up. Looking at Dellow's work, you would think that it is almost contrary to this belief: he might be scoring more, but he's actually allowing significantly more. Plus, when looking at the team as a whole, one might say that it's time to sacrifice his somewhat increased scoring, and that our very talented forwards are now experienced enough to do that on their own. If Buff focuses on his defensive part of the game and gets back to a league average, sure maybe he individually doesn't score as much, but maybe it's worth the sacrifice.

To be clear, my observation of his casual play has nothing to do with his play in the offensive zone. It was all related to his casual play in his own zone. Lazy and ill-advised passes, lack of attention to positioning, etc. Those were the bad signs I saw a lot in his game from early last season, and commented on it from the early games.

 

[MrBoJangelz71]

"Make him a coach then"

Maybe he has more influence as a player. The best coach's assistant many times are players.

 

[garret9]

Maybe he has more influence as a player. The best coach's assistant many times are players.

It was a joke.

I don't agree with the idea of giving roster spots to players for off-ice value that is complete guess work on the extent of impact.

There's no one intangible. There are hundreds all interacting and cancelling many of each other out. What's the intangible of having a weaker player get scored on more and score less than a stronger player would? Winning feels good and is an intangible.

You have no idea how much his off ice value is worth, nor do even the most informed individual of the room. It's not that I suggest to be atheistic of intangibles, but I'm extremely skeptical of these people running hockey teams understanding the true value of intangible and are instead view it merely as nice guys who play hard.

If teams were serious about intangibles they would be doing more work like the empirical studies you see in work place on things like leadership, perseverance, etc.

What I would suggest is that you do not actively seek out intangibles but make minor, tie braking like, adjustments of a player's worth given some guess work on intangibles.

I'd also suggest teams to actually do a better job in understanding these things with work of people like Stephan Wolejszo.

 

[MrBoJangelz71]

It was a joke.

I don't agree with the idea of giving roster spots to players for off-ice value that is complete guess work on the extent of impact.

There's no one intangible. There are hundreds all interacting and cancelling many of each other out. What's the intangible of having a weaker player get scored on more and score less than a stronger player would? Winning feels good and is an intangible.

You have no idea how much his off ice value is worth, nor do even the most informed individual of the room. It's not that I suggest to be atheistic of intangibles, but I'm extremely skeptical of these people running hockey teams understanding the true value of intangible and are instead view it merely as nice guys who play hard.

If teams were serious about intangibles they would be doing more work like the empirical studies you see in work place on things like leadership, perseverance, etc.

What I would suggest is that you do not actively seek out intangibles but make minor, tie braking like, adjustments of a player's worth given some guess work on intangibles.

I'd also suggest teams to actually do a better job in understanding these things with work of people like Stephan Wolejszo.

Not sure what this all means, but I can tell you from my own experiences, I have had more success with teams of lesser talent but better characters in the dressing room, that kept accountability and structure, than I have had with much more talented teams of skilled players, but some marching to their own beats.

But of course, if you can get high character guys and the skill, take it every day.

And I know exactly what his Off ice value is, it's worth the contract they just awarded him. Cause it sure wasn't for his scoring and high end skill.

 

[garret9]

Not sure what this all means, but I can tell you from my own experiences, I have had more success with teams of lesser talent but better characters in the dressing room, that kept accountability and structure, than I have had with much more talented teams of skilled players, but some marching to their own beats.

But of course, if you can get high character guys and the skill, take it every day.

And I know exactly what his Off ice value is, it's worth the contract they just awarded him. Cause it sure wasn't for his scoring and high end skill.

It should be simple.

If you are having success than you are garnering good results. Do not confound skill with results.

We're talking about off ice value of someone who does not garner positive on ice results.

Okay? If you know exactly how much his off ice stuff is worth:
How much does his off ice value improve the teams goal differential through his intangibles relative to the average alternative?

I can estimate how much his playing minutes does.

 

[MrBoJangelz71]

It should be simple.

If you are having success than you are garnering good results. Do not confound skill with results.

We're talking about off ice value of someone who does not garner positive on ice results.

Okay? If you know exactly how much his off ice stuff is worth:
How much does his off ice value improve the teams goal differential through his intangibles relative to the average alternative?

I can estimate how much his playing minutes does.

I think it was simple, you complicated it.

Do not always confound results with skill. There is a human element that you are not able to measure, but that doesn't mean it doesn't exist, or that it doesn't play a role in on ice results.

Leaders are not always your most skilled players. Just because you do not have any way of quantifying what value leaders bring to a team, does not negate it's value.

Please tell me, what does at teams goal differential tell you about how the kid is going to practice in the middle of the season when they're tired and they don't feel like working. How will Corsi dictate accountability and professionalism for a 20 year old kid, or what stat do you have that kicks teammates buts when they are slacking and mailing it in?

 

[garret9]

I think it was simple, you complicated it.

Do not always confound results with skill. There is a human element that you are not able to measure, but that doesn't mean it doesn't exist, or that it doesn't play a role in on ice results.

Leaders are not always your most skilled players. Just because you do not have any way of quantifying what value leaders bring to a team, does not negate it's value.

Please tell me, what does at teams goal differential tell you about how the kid is going to practice in the middle of the season when they're tired and they don't feel like working. How will Corsi dictate accountability and professionalism for a 20 year old kid, or what stat do you have that kicks teammates buts when they are slacking and mailing it in?

I did not confound skill for results. You were the first person to say skill. You confounded it. If your human element impacts results directly, which many do (perserverance, grit, hard work, adaptability, etc.). It is something that will separate results from skill.

So while it is an intangible in how I cannot directly measure how a player's percerverance impacts results, it impacts results so they account for them among the aggregate of other things.

I also never said the human element does not exist or impact results. In fact I distinctly said the opposite.

I'm not scrupulous to the human element as it is a real variable.

What I'm scrupulous to is yours or anyone's ability to actually appropriately determine a scalable value to that.

If you cannot, you are merely making guesses in the air to a player's value from that area, which you are.

My problem isn't on the impact of the human element.
My problem is that the human element is misunderstood.

I've had a social scientist whose whole life work revolves around studying intangibles in the work place and is a stout defender of their importance in hockey present at a conference I ran. He still laughs at people's poor judgement on the area.

I can tell you about how many wins in a season a player's Corsi is worth.

I nor you nor the hockey execs can tell me how many wins a player's off ice value is worth.

Your practice example is another one. Yea we know it's a negative thing. That player is less likely to reach their potential as they are not working hard. A guy working hard may be contagious. How many wins is it worth though? 1? 100? 1/100?

 

[Whileee]

I did not confound skill for results. You were the first person to say skill. You confounded it. If your human element impacts results directly, which many do (perserverance, grit, hard work, adaptability, etc.). It is something that will separate results from skill.

So while it is an intangible in how I cannot directly measure how a player's percerverance impacts results, it impacts results so they account for them among the aggregate of other things.

I also never said the human element does not exist or impact results. In fact I distinctly said the opposite.

I'm not scrupulous to the human element as it is a real variable.

What I'm scrupulous to is yours or anyone's ability to actually appropriately determine a scalable value to that.

If you cannot, you are merely making guesses in the air to a player's value from that area, which you are.

My problem isn't on the impact of the human element.
My problem is that the human element is misunderstood.

I've had a social scientist whose whole life work revolves around studying intangibles in the work place and is a stout defender of their importance in hockey present at a conference I ran. He still laughs at people's poor judgement on the area.

I can tell you about how many wins in a season a player's Corsi is worth.

I nor you nor the hockey execs can tell me how many wins a player's off ice value is worth.

Your practice example is another one. Yea we know it's a negative thing. That player is less likely to reach their potential as they are not working hard. A guy working hard may be contagious. How many wins is it worth though? 1? 100? 1/100?

Not to downplay stats too much, but saying you can tell how many wins a player's Corsi is worth is stretching it. You'd need to add the caveat that your model only accounts for a certain percentage of the variance (usually less than 50%, right?) and you'd need to add a caveat around type 1 error in the estimate.

 

[garret9]

Not to downplay stats too much, but saying you can tell how many wins a player's Corsi is worth is stretching it. You'd need to add the caveat that your model only accounts for a certain percentage of the variance (usually less than 50%, right?) and you'd need to add a caveat around type 1 error in the estimate.

I didn't say there wouldn't be confidence intervals.

Tell me the win value and confidence intervals of Cormier playing hard at practice?

 

[Whileee]

I didn't say there wouldn't be confidence intervals.

Tell me the win value and confidence intervals of Cormier playing hard at practice?

I made no claim about "intangible" variables. I commented on your assertion that you could tell us the value of a player's Corsi in terms of team wins.

When was the last time you saw confidence intervals in public discourse about hockey advanced stats?

Tight confidence intervals won't help if the r-square for a model is 0.4. You still have 60% of the variance in the outcome prediction to explain...

 

[garret9]

I made no claim about "intangible" variables. I commented on your assertion that you could tell us the value of a player's Corsi in terms of team wins.

When was the last time you saw confidence intervals in public discourse about hockey advanced stats?

Tight confidence intervals won't help if the r-square for a model is 0.4. You still have 60% of the variance in the outcome prediction to explain...

The context of it being Corsi compared to intangibles matters in the discussion. The whole point is you have some marking in one and none in the other.

Yes, I'm well aware.

And there's not much work with that because know your audience.

The general fan doesn't understand these things. When you dumb down chemistry to the bohr model, it is wrong... but we do it for high schoolers because it explains a lot of the concepts and ideas efficiently enough to give them a general idea.


As an aside, since typically the prediction variable is goals, we know goaltending has a major part to do with the variance output, otherwise goaltending wouldn't really matter. 60% variance in a models output is one thing, but you'd have to be careful then saying the variance would be the same for the hypothetical Corsi to player impact.

Ie: part of the variance pie isn't the player Corsi->player impact but a known confounding in goaltending.

 

[Whileee]

The context of it being Corsi compared to intangibles matters in the discussion. The whole point is you have some marking in one and none in the other.

Yes, I'm well aware.

And there's not much work with that because know your audience.

The general fan doesn't understand these things. When you dumb down chemistry to the bohr model, it is wrong... but we do it for high schoolers because it explains a lot of the concepts and ideas efficiently enough to give them a general idea.


As an aside, since typically the prediction variable is goals, we know goaltending has a major part to do with the variance output, otherwise goaltending wouldn't really matter. 60% variance in a models output is one thing, but you'd have to be careful then saying the variance would be the same for the hypothetical Corsi to player impact.

Ie: part of the variance pie isn't the player Corsi->player impact but a known confounding in goaltending.

I agree that objective statistical analysis >>> particular qualitative observations. However, there are almost certainly qualitative observations that matter, but we don't systematically document and evaluate them, and perhaps couldn't adequately.

I think most people could understand the concept of statistical uncertainty. Political pollsters do it all the time, even though they often misrepresent concepts like confidence intervals. But I only teach grad students so I can tend to be a bit pedantic.

You could add goaltending to the models to improve the fit. What are the r-squared values of the best predictive models you've seen with goals as the dependent variable?

 

[MrBoJangelz71]

I did not confound skill for results. You were the first person to say skill. You confounded it. If your human element impacts results directly, which many do (perserverance, grit, hard work, adaptability, etc.). It is something that will separate results from skill.

So while it is an intangible in how I cannot directly measure how a player's percerverance impacts results, it impacts results so they account for them among the aggregate of other things.

I also never said the human element does not exist or impact results. In fact I distinctly said the opposite.

I'm not scrupulous to the human element as it is a real variable.

What I'm scrupulous to is yours or anyone's ability to actually appropriately determine a scalable value to that.

If you cannot, you are merely making guesses in the air to a player's value from that area, which you are.

My problem isn't on the impact of the human element.
My problem is that the human element is misunderstood.

I've had a social scientist whose whole life work revolves around studying intangibles in the work place and is a stout defender of their importance in hockey present at a conference I ran. He still laughs at people's poor judgement on the area.

I can tell you about how many wins in a season a player's Corsi is worth.

I nor you nor the hockey execs can tell me how many wins a player's off ice value is worth.

Your practice example is another one. Yea we know it's a negative thing. That player is less likely to reach their potential as they are not working hard. A guy working hard may be contagious. How many wins is it worth though? 1? 100? 1/100?

Did I ever say it was measurable.? Nope, distinctly said the opposite, not quantifiable but prevalent.

You mentioned skill prior, and assuming "stronger" players are usually more skilled players, you introduced it prior. All semantics really, the stats you love, are they not all out puts that skill influences?

My guess, it's based off the contract they awarded Cormier. It wasn't for his ability to positively effect shot differential, it was for his leadership.

So if a professional organization feels it is important to have an older vet that carries the C, controls the dressing room and keeps the younger players accountable, must mean the see that "value" in it as well.

Once again, just cause you are unable the measure does not mean it does not exist.

 

[garret9]

Did I ever say it was measurable.? Nope, distinctly said the opposite, not quantifiable but prevalent.

Obviously you didn't say measurable, because my whole point. If it were measurable, you wouldn't be wildly guessing the impacts significance. Because you cannot measure it, you do not know the extent of impact. I can estimate the win value and the $ value and the win per $ value of Corsi, shooting pucks at the net, points production, drawing penalties, winning face offs, etc, etc.

That is powerful for decision making. It allows me to make informed decisions on how much a Player A is worth more than Player B specific to those points. Not just that Player A is better than Player B, but by how much and how much that is worth. You may know Cormier is better at certain off-ice intangibles, but you do not know by how much and how much it is worth.

You mentioned skill prior, and assuming "stronger" players are usually more skilled players, you introduced it prior. All semantics really, the stats you love, are they not all out puts that skill influences?

I went back and my first mentioning of skill was my correcting your conflation of results in my argument for skill. Skill and intangibles are all inputs, yes. Results, though, are the output of those things interacting with the environment. This is an important semantic given your false assumption that I'm suggesting the human factor or intangibles do not exist.

I'm pointing out that some intangibles (grit, perseverance, tenacity, physicality, etc) are inherently accounted for in the results for a player. Not all are obviously, like the ones we are discussing with Cormier (good in the room, leadership, pushing the pace at practice), but some are. If intangibles had no impact, then skill would = results and I would have no problem with you conflating my earlier mentioning of results as player skill.

My guess, it's based off the contract they awarded Cormier. It wasn't for his ability to positively effect shot differential, it was for his leadership.

I'm sure it was for his leadership and other qualities. Those qualities have value and they are real. The issue I have is that how much value they have is complete and utter guess work.

As I mentioned earlier, we can make some solid estimates of the win and $ value of Player A's on-ice results vs Player B's on-ice results. But, if Player B is better in some intangibles, we have no idea how much better those intangibles would need to be to tip the scale in Player B's favour.

Note: Some intangibles are confounding variables to the results, so those intangibles (grit, perseverance, tenacity, physicality, etc) would be accounted for in part by the on-ice results of Player A being better than Player B.

So if a professional organization feels it is important to have an older vet that carries the C, controls the dressing room and keeps the younger players accountable, must mean the see that "value" in it as well.

Obviously they see value in it. The issue is they do not see how much value is in it. How many dollars it is worth.

On Twitter a few weeks back I once said:

Being agnostic on intangibles or simply recognizing the unknown is the unknown is not being atheistic on such potential variables.
There are many variables to the human factor. Many will counter act; some may be significant and some may be insignificant.
The best practice is
1) understand that guessing on how the many unknowns combine is a wild and speculative guess so take with lbs of salt
2) reduce the amount of unknown with empirical practices, see @StefanWolejszo's work for examples​

Once again, just cause you are unable the measure does not mean it does not exist.

And once again, I never said that because something not measured means it does not exist. I have never once made that argument. If I did, there would be no reason for me to correct your conflation of my argument for results as skill.

My argument is and always will be that because you don't know how much it impacts results you do not know how much it is worth. That is merely a fact.

 

[garret9]

I agree that objective statistical analysis >>> particular qualitative observations. However, there are almost certainly qualitative observations that matter, but we don't systematically document and evaluate them, and perhaps couldn't adequately.

I think most people could understand the concept of statistical uncertainty. Political pollsters do it all the time, even though they often misrepresent concepts like confidence intervals. But I only teach grad students so I can tend to be a bit pedantic.

You could add goaltending to the models to improve the fit. What are the r-squared values of the best predictive models you've seen with goals as the dependent variable?

I'm just about to head out, but my issue with intangibles in hockey is that hockey people do a bad job with dealing with them, when they don't need to be. Teams could do much better here but are highly resistant, more so than even to regular analytics.

Stefan Wolejszo at his blog "Stories Numbers Tell" does a great job in breaking down stuff like this. I had him speak at VanHAC.

Here is my own account of Stefan at VanHAC:

After a short break, watching the Canuck's pre-game skate, came Stefan Wolejszo and his talk on rank ordering some intangibles. Wolejszo is a social scientist working with the Government of Canada. Over the past few years he has spent a lot of his free time writing about intangibles in hockey, specifically how they work and how they can be measured. In his talk, Wolejszo dove into specific intangibles and ranked them based on five factors; these were conceptual clarity (whether the intangible was clearly defined), empirical research (whether there is evidence of the intangibles value outside of hockey), elite occupation study (whether or not there is research on elite populations, since NHL hockey is one), practicality (whether teams could in some manner collect meaningful data), and applicability (whether teams could use this data to improve their team). (Slides)​

Link to my breakdown of VanHAC <- this has a link to his slides too