Post-Game Talk: Lou unable to beat a team lead by Joe Woll. Leafs win 3-0.

[Notsince67]

My point is 30 -40 -50 years ago a "study" had defined processes, with controls to minimize biases, now a guy builds a website or publishes something in a blog, and it's considered a "study".

Aside for capfriendly, I don't trust many of these pop up "fan stats" websites/blogs.

I understand your sentiment. I spent a lot of time understanding some of this stuff and even ran some of my own correlations. I found a lot of the claims thrown out there to be dubious at best. That said, where as it drives me a bit crazy when analytics are abused to make claims the eyes cant see, there are at least as many people out there (if not more) that wouldn't recognize a good scoring attempt unless the puck went in the net. Goals are a low frequency event and it is useless to discuss the quality of game-play simply by counting goals.
If someone want to counter descriptive quantitative conclusions, it is pretty easy to do it by watching game video. People are both busy and lazy so few would endeavor to to it.

 

[Notsince67]

Yep, we have plenty of examples of bad racecar drivers on this forum.

There are a lot of pundits in the past who were terrible too. Berkshire was a complete hack and Tulloch was only ever looking for clicks.

 

[Bluelines]

I understand your sentiment. I spent a lot of time understanding some of this stuff and even ran some of my own correlations. I found a lot of the claims thrown out there to be dubious at best. That said, where as it drives me a bit crazy when analytics are abused to make claims the eyes cant see, there are at least as many people out there (if not more) that wouldn't recognize a good scoring attempt unless the puck went in the net. Goals are a low frequency event and it is useless to discuss the quality of game-play simply by counting goals.
If someone want to counter descriptive quantitative conclusions, it is pretty easy to do it by watching game video. People are both busy and lazy so few would endeavor to to it.

You captured my sentiment, good post.

 

[zeke]

All stats get correlated to wins. I know corsi correlates poorly at something like .19 which is far less than a flip of a coin. Further parsing the data to exclude medium and low danger shots gives a datapoint that might be interesting but in isolation says little about the game other than the opposing team's intent.

It isn't rocket science...it's data science. BTW...this has zero correlation to being liberal.

All these arguments are nonsense.

Every hockey person when watching the game tries to evaluate which team is controlling the game, and every hockey person understands that the scoreboard doesn't tell nearly the whole story.

Even when we were little we knew well enough to look at other stats beyond the score - shots on goal, save percentage etc. We even knew at a young age that we should adjust for score - that shots taken when you're trying to comeback and the other team is sitting back are less meaningful. We also understand at a very young age that all shots aren't equal, and that getting to the danger zones is crucial to success.

All of the analytics simply reflect things we already know and try to do when we watch games.


Some people are just very upset that their half-assed hot takes from half-drunk half-watched game viewing through the biased fan eye don't stand up to simple objective measurements, so they try their best to pretend that the game of hockey is way too complicated to ever be able to measure effectively.

Hot tip: it's not that complicated. You can measure which teams play better than other teams.

 

[HoweHullOrr]

The Leafs won so you can feel good but I’d reign in the title of this thread until we at least match where the Islanders got playoff wise last year.
Some teams are also average regular season teams but really good playoff teams. We are trying to establish both but haven’t yet.

Agreed. Islanders game seems to have translated well to playoff hockey lately.

Leafs = not so much. Obviously, we hope for something different this year.

I do think that losing Leddy had a pretty big impact on them. They lost their best PMD & distributor on the back end with his departure. I think it changes the complexion of their blueline.

 

[Notsince67]

All these arguments are nonsense.

Every hockey person when watching the game tries to evaluate which team is controlling the game, and every hockey person understands that the scoreboard doesn't tell nearly the whole story.

Even when we were little we knew well enough to look at other stats beyond the score - shots on goal, save percentage etc. We even knew at a young age that we should adjust for score - that shots taken when you're trying to comeback and the other team is sitting back are less meaningful. We also understand at a very young age that all shots aren't equal, and that getting to the danger zones is crucial to success.

All of the analytics simply reflect things we already know and try to do when we watch games.


Some people are just very upset that their half-assed hot takes from half-drunk half-watched game viewing through the biased fan eye don't stand up to simple objective measurements, so they try their best to pretend that the game of hockey is way too complicated to ever be able to measure effectively.

Hot tip: it's not that complicated. You can measure which teams play better than other teams.

No...all data does. Analytics are a whole different kettle of fish.

For a trained quant guy (pleasure to meet you), the concept of selection bias is real...especially when the coefficient of determination (r^2) is so poor. You sub select data and it doesnt predict as well as you thought. Most correlation coefficients are calculated using effective averages and they work better for average players. Poor players and elite players tend to diverge.
Collective behavior in a given night is brutally variable.
I posted it a couple of years back but will reference it yet again...read When Genius Failed: The Rise and Fall of Long-Term Capital Management: Lowenstein, Roger: 9780375758256: Books - Amazon.ca.
...to understand the epitome of statistical hubris.
Stats are useful with a ton of data but systems are never a black box and a single variable (like a trade) changes the population behavior.
I spent a lot of my professional years building risk models for banking. These models requires constant calibration to account for behavior changes. If you think you know stats from the current state of hockey analytics, you really need to get out more. I accept conclusions from huge differences in data but wisdom and experience dictates what is statistically significant enough to rank order what you choose to compare.
There is a difference between a concept which might be absolutely true and a pragmatic application of a concept that makes it useful. I never saw any published error bars in the hockey analytic community but I would bet they're huge. The reason why they arent published is that the number of clickbate hot takes would be significantly reduced.

 

[zeke]

No...all data does. Analytics are a whole different kettle of fish.

For a trained quant guy (pleasure to meet you), the concept of selection bias is real...especially when the coefficient of determination (r^2) is so poor. You sub select data and it doesnt predict as well as you thought. Most correlation coefficients are calculated using effective averages and they work better for average players. Poor players and elite players tend to diverge.
Collective behavior in a given night is brutally variable.
I posted it a couple of years back but will reference it yet again...read When Genius Failed: The Rise and Fall of Long-Term Capital Management: Lowenstein, Roger: 9780375758256: Books - Amazon.ca.
...to understand the epitome of statistical hubris.
Stats are useful with a ton of data but systems are never a black box and a single variable (like a trade) changes the population behavior.
I spent a lot of my professional years building risk models for banking. These models requires constant calibration to account for behavior changes. If you think you know stats from the current state of hockey analytics, you really need to get out more. I accept conclusions from huge differences in data but wisdom and experience dictates what is statistically significant enough to rank order what you choose to compare.
There is a difference between a concept which might be absolutely true and a pragmatic application of a concept that makes it useful. I never saw any published error bars in the hockey analytic community but I would bet they're huge. The reason why they arent published is that the number of clickbate hot takes would be significantly reduced.

you're overcomplicated something very simple.

which hockey teams outplay their opponents.

it's not rocket science.

 

[Notsince67]

you're overcomplicated something very simple.

which hockey teams outplay their opponents.

it's not rocket science.

There is useful information in analytics for sure. The problem is with word outplay. Anaytics cannot determine unequivocable the answer to that question. Is it the offense....a specific line, the defense...a specific pair, the goalie? or is it a combination of a few variables? If we know a team has a better offense than defense and the offense plays at an average level in the nhl but the defense plays like crap, is it the offense's fault for not playing a bit better to overcome the deficiencies of the defense? what if the goalie is good or shit? are they part of a team. Analytics may be helpful in sorting out the pieces but it doesn't sufficiently assist the determination of who should or will win. There is a usefullness but it is contextual and it doesn't end a debate about what the team should do.

 

[zeke]

There is useful information in analytics for sure. The problem is with word outplay. Anaytics cannot determine unequivocable the answer to that question. Is it the offense....a specific line, the defense...a specific pair, the goalie? or is it a combination of a few variables? If we know a team has a better offense than defense and the offense plays at an average level in the nhl but the defense plays like crap, is it the offense's fault for not playing a bit better to overcome the deficiencies of the defense? what if the goalie is good or shit? are they part of a team. Analytics may be helpful in sorting out the pieces but it doesn't sufficiently assist the determination of who should or will win. There is a usefullness but it is contextual and it doesn't end a debate about what the team should do.

You're looking for far too much certainty.

We should all understand that more than most sports, the difference between winning and losing NHL hockey games is tiny.

The thing is we've always used stats to help figure out what we're watching - whether it's goals, points, +/- for players or shots and goals for teams. It's just that those stats really do suck and are near useless, while the current analytics are much more useful and able to be contextualized for game situations.

 

[Notsince67]

You're looking for far too much certainty.

We should all understand that more than most sports, the difference between winning and losing NHL hockey games is tiny.

The thing is we've always used stats to help figure out what we're watching - whether it's goals, points, +/- for players or shots and goals for teams. It's just that those stats really do suck and are near useless, while the current analytics are much more useful and able to be contextualized for game situations.

So look at the games where we got goalied like LA and try to apply your stats to what the team should do to overcome a goalie who is playing hot. That is a real life situation where all other underlying stats indicate that the leafs should have won instead of loosing so badly. The leafs took some risks and the kings were opportunistic is burying the leafs into a deeper hole. No need for analytics there unless it could be used to develop strategy. Strategy is also a large part of quantitative analytics. That is where the power of analytics are useful...not in telling you what your eyes can see. Sure...you can counter some idiots who refuse to acknowledge that certain pieces worked fine but you still have no answer unless you know your game,

 

[zeke]

So look at the games where we got goalied like LA and try to apply your stats to what the team should do to overcome a goalie who is playing hot. That is a real life situation where all other underlying stats indicate that the leafs should have won instead of loosing so badly. The leafs took some risks and the kings were opportunistic is burying the leafs into a deeper hole. No need for analytics there unless it could be used to develop strategy. Strategy is also a large part of quantitative analytics. That is where the power of analytics are useful...not in telling you what your eyes can see. Sure...you can counter some idiots who refuse to acknowledge that certain pieces worked fine but you still have no answer unless you know your game,

It's a good example of how analytics can contextualize exactly what you are talking about, when old school stats had no way of adjusting for context like that.

Look at the massive spike in xG those late-game golden chances/goals gave to LA, resulting in a much closer finish than the old school stats would have guessed:

 

[Polaris1010]

For a trained quant guy (pleasure to meet you), the concept of selection bias is real...especially when the coefficient of determination (r^2) is so poor. You sub select data and it doesnt predict as well as you thought. Most correlation coefficients are calculated using effective averages and they work better for average players. Poor players and elite players tend to diverge.
Collective behavior in a given night is brutally variable.

Myself, I definitely am not a quant guy, but this quote hits the nail right on its head.

The irony is that the analytics crowd in the hockey world, will not fully understand what is being said here by the man.

When the correlation is low, there is not much there in other words. Then more ingredients are added, whose correlations could be even lower.

Then the apples to oranges comparisons. In the old days, we just did not compare 1st line guys to 4th line guys, but now we do. In the threads, some fans pushing for Spezza to be playing higher in the lineup.

This is not a slight against analytics, just an acknowledgement that analytics is a work in progress.

This is NOT saying that analytics bad in itself, it has to be improved, in order for it to get widespread acceptance in hockey.

-------------------- --------------------

When someone, anywhere, is quoting these advanced stats, but are unable to explain how they are derived and the application of it, then the question becomes do they understand what they are using or doing?

That is why I do not like Dubas. He seems like a guy pushing this mathematical model, which he does not fully understand himself. His personnel decisions reflect that. In the financial world, that has lead to spectacular documented failures.

 

[Notsince67]

Myself, I definitely am not a quant guy, but this quote hits the nail right on its head.

The irony is that the analytics crowd in the hockey world, will not fully understand what is being said here by the man.

When the correlation is low, there is not much there in other words. Then more ingredients are added, whose correlations could be even lower.

Then the apples to oranges comparisons. In the old days, we just did not compare 1st line guys to 4th line guys, but now we do. In the threads, some fans pushing for Spezza to be playing higher in the lineup.

This is not a slight against analytics, just an acknowledgement that analytics is a work in progress.

This is NOT saying that analytics bad in itself, it has to be improved, in order for it to get widespread acceptance in hockey.

-------------------- --------------------

When someone, anywhere, is quoting these advanced stats, but are unable to explain how they are derived and the application of it, then the question becomes do they understand what they are using or doing?

That is why I do not like Dubas. He seems like a guy pushing this mathematical model, which he does not fully understand himself. His personnel decisions reflect that. In the financial world, that has lead to spectacular documented failures.

Teams like the Lightning use analytics. Tools are great but they won't replace a good coach. They might enhance him a bit though. Thankfully for the sport, it isn't all cerebral like chess and there will always be moves and countermoves

 

[Polaris1010]

It's a good example of how analytics can contextualize exactly what you are talking about, when old school stats had no way of adjusting for context like that.

Look at the massive spike in xG those late-game golden chances/goals gave to LA, resulting in a much closer finish than the old school stats would have guessed:

View attachment 483328

But dude, not an slight or a criticism, the chart posted does not seem to be a revelation.

When a team is behind, they usually open it up, start taking chances, which gives the other team more chances.

That LA-Toronto game, Campbell probably could have had a couple of those goals. Toronto did not play well, but did limit LA chances, so I got no problem with that chart.

The thing is that that chart does not tell us anything new. The Leafs fell behind, the other team scored on every chance they got, then the Leafs opened it up late when down.

Ideally, we expect more from analytics.

------------- -------------

However, how people do analytics today in hockey, can be erroneous.

For example, the low scoring of the Leafs and the shot percentage is apparently low too.

That does not mean anything other than the Leafs are a low scoring team.

Someone will then say, what about reversion to the mean!?

There is no value in that. Perhaps the Leafs are a low scoring team and the Oilers a high scoring team, and that is where the Leafs belong in the Goals For ranking, in the lower third of the league.

Low shooting percentage could mean they are due, or they suck at scoring like a Ritchie.

I believe the reason is neither. Since they play a defense first system, and most offensive attempts are one and done with no rebound chances because no Leafs is standing in front of the net, then the Leafs will not score many goals, but will still win a lot of games because they are good at their style of play.

But there, did we see that? My explanation probably more accurate, while the analytics totally off base, on issue of low shooting percentage for the Leafs.

This is not saying analytics is bad, but it has to be improved, or contextualize more, if that makes any sense.

 

[zeke]

But dude, not an slight or a criticism, the chart posted does not seem to be a revelation.

When a team is behind, they usually open it up, start taking chances, which gives the other team more chances.

That LA-Toronto game, Campbell probably could have had a couple of those goals. Toronto did not play well, but did limit LA chances, so I got no problem with that chart.

The thing is that that chart does not tell us anything new. The Leafs fell behind, the other team scored on every chance they got, then the Leafs opened it up late when down.

Ideally, we expect more from analytics.

------------- -------------

However, how people do analytics today in hockey, can be erroneous.

For example, the low scoring of the Leafs and the shot percentage is apparently low too.

That does not mean anything other than the Leafs are a low scoring team.

Someone will then say, what about reversion to the mean!?

There is no value in that. Perhaps the Leafs are a low scoring team and the Oilers a high scoring team, and that is where the Leafs belong in the Goals For ranking, in the lower third of the league.

Low shooting percentage could mean they are due, or they suck at scoring like a Ritchie.

I believe the reason is neither. Since they play a defense first system, and most offensive attempts are one and done with no rebound chances because no Leafs is standing in front of the net, then the Leafs will not score many goals, but will still win a lot of games because they are good at their style of play.

But there, did we see that? My explanation probably more accurate, while the analytics totally off base, on issue of low shooting percentage for the Leafs.

This is not saying analytics is bad, but it has to be improved, or contextualize more, if that makes any sense.

It told us something accurate thought, correct? And something more accurate than the traditional stats would have told us, right?

How many of the league's 1312 games do you have the time or inclination to watch closely and thoroughly?

 

[TGB]

I understand your sentiment. I spent a lot of time understanding some of this stuff and even ran some of my own correlations. I found a lot of the claims thrown out there to be dubious at best. That said, where as it drives me a bit crazy when analytics are abused to make claims the eyes cant see, there are at least as many people out there (if not more) that wouldn't recognize a good scoring attempt unless the puck went in the net. Goals are a low frequency event and it is useless to discuss the quality of game-play simply by counting goals.
If someone want to counter descriptive quantitative conclusions, it is pretty easy to do it by watching game video.

Great post. Way too many people dismiss the value of watching gameplay footage. One of the main reasons I seem downright prophetic at times with baseball is purely because of what I observe during gameplay. I look at the stats, sure, but I use them to confirm gameplay performance, not the other way around.

The problem isn't that observation is inherently flawed, but that some people just suck it. For the same reason we don't say math is flawed because the guy you got doing it is bad at math, so, too, can we say observation in itself is not a bad thing, but the observer most certainly can be.

 

[Dekes For Days]

The problem isn't that observation is inherently flawed, but that some people just suck it.

The problem is both.

 

[Gary Nylund]

That is why I do not like Dubas. He seems like a guy pushing this mathematical model, which he does not fully understand himself. His personnel decisions reflect that. In the financial world, that has lead to spectacular documented failures.

I'm not sure I buy this. Can you elaborate and perhaps give examples of personnel decisions that are a reflection of "his model"?

I've heard a lot about how Dubas supposedly likes midget smurfs and so on but his personnel decisions don't support that theory at all as far as I can tell.

Great post. Way too many people dismiss the value of watching gameplay footage. One of the main reasons I seem downright prophetic at times with baseball is purely because of what I observe during gameplay. I look at the stats, sure, but I use them to confirm gameplay performance, not the other way around.

The other way around has something going for it too IMO - if the numbers are telling you something that you're not aware of, maybe the numbers are wrong but on the other hand, maybe they're not wrong and if you take a closer look with those numbers in mind, you might see something that you just haven't noticed before.

The problem isn't that observation is inherently flawed, but that some people just suck it. For the same reason we don't say math is flawed because the guy you got doing it is bad at math, so, too, can we say observation in itself is not a bad thing, but the observer most certainly can be.

I think most people suck at it. and I'd say all observers are flawed, just to different degrees. A good professional scout's observations are worth a lot, but still flawed to some degree. The average Joe watching on TV, his view is obviously much more flawed.

 

[4thline]

Possession stat tracking has fundamentally changed the way hockey is played and evaluated. It's inarguable. It's changed player selection, it's changed tactics.

But to do so it's become public domain, those lessons have become almost universally accepted. The learnings have been adopted, which lead to the margins of performance differences being reduced, which has in turn made them much less useful/harder to use.

 

[TGB]

The problem is both.

Hey man, just because you suck at it doesn't mean the rest of us do.

I think most people suck at it. and I'd say all observers are flawed, just to different degrees. A good professional scout's observations are worth a lot, but still flawed to some degree. The average Joe watching on TV, his view is obviously much more flawed.

Oh, for sure. The vast majority of people can't help being biased in some way...or their vision isn't that great.

But it's also incorrect to suggest NOBODY can do it. Pretty much every systems case study I've ever seen basically boils down to "hundreds of people couldn't figure out the problem until this one guy enters the picture and spots the issue". Hell, I've even seen stories of entire organizations of adults being completely flummoxed, and then the secretary brings her kid to work and the kid solves the problem. If you want a sports example, look at Pete Walker for the Jays. The man apparently spots issues with pitcher deliveries pretty much everyone else in the league misses.

You can accurately suggest these kind of people are rare, but inferring it's impossible based on the collective experiences of a bunch of internet randos posting at some internet forum is just silly.

 

[Dekes For Days]

Hey man, just because you suck at it doesn't mean the rest of us do.

I don't suck at it; I'm actually quite skilled, because I don't ignore the benefit that analytics can provide. But you're still subject to the same inherent biases and limitations as everybody else.

 

[Bluelines]

I don't suck at it; I'm actually quite skilled, because I don't ignore the benefit that analytics can provide. But you're still subject to the same inherent biases and limitations as everybody else.

Based on what metric?

 

[Polaris1010]

I'm not sure I buy this. Can you elaborate and perhaps give examples of personnel decisions that are a reflection of "his model"?

I've heard a lot about how Dubas supposedly likes midget smurfs and so on but his personnel decisions don't support that theory at all as far as I can tell.

Garrett Sparks would be the case study for us to evaluate Dubas on how he does things, or use to do things. In this case, it was really weird, it was like he was using data generated in the minors, then extrapolating into the big leagues. The fans could see with their own eyes something was really off.

Then all these small players with good underlying numbers. This Leafs team no longer has all those small players with good underlying numbers. The problem was these small players did not score, and were not that good at checking. In short, Dubas found small players with good underlying numbers who were able to do neither, score or be responsible defensively. Kerfoot seems to be the only exception, who is at least responsible defensively. Just cycling in all these guys then out, kind of shows Dubas was just fishing, not really knowing what he was doing. The underlying numbers were not working with the small players Dubas brought in. They were quickly shipped out.

The defensemen he brought in, why? Barrie. Ceci. Bougaivine. Holl.


Seems to me, this year is really different. Dubas may not be in charge anymore. Look at the bottom six. All established NHLers, all who do not really score, but the bottom six, everyone is 6 ft tall at least and looks to be 200 lbs. All are solid checkers, except maybe for Ritchie, who was not signed to the bottom six.

My belief is that Dubas was trying to build a very different type of NHL team, which he might not be allowed to do so anymore.

Since they fired Babcock, the overhaul to the roster has been significant. But this year, it is finally settled. They went with size and established checkers in the NHL, which is old school.

It is a general principle with numbers. Misapply the numbers, and the results are not there. That really has been Dubas with the players he brought in and kicked out.

The revolving door for players, goalie, forwards, defense, suggests that the model he was using was wanting, perhaps he did not understand it himself, how it applied to the NHL game.

 

[zeke]

Garrett Sparks would be the case study for us to evaluate Dubas on how he does things, or use to do things. In this case, it was really weird, it was like he was using data generated in the minors, then extrapolating into the big leagues. The fans could see with their own eyes something was really off.

Then all these small players with good underlying numbers. This Leafs team no longer has all those small players with good underlying numbers. The problem was these small players did not score, and were not that good at checking. In short, Dubas found small players with good underlying numbers who were able to do neither, score or be responsible defensively. Kerfoot seems to be the only exception, who is at least responsible defensively. Just cycling in all these guys then out, kind of shows Dubas was just fishing, not really knowing what he was doing. The underlying numbers were not working with the small players Dubas brought in. They were quickly shipped out.

The defensemen he brought in, why? Barrie. Ceci. Bougaivine. Holl.


Seems to me, this year is really different. Dubas may not be in charge anymore. Look at the bottom six. All established NHLers, all who do not really score, but the bottom six, everyone is 6 ft tall at least and looks to be 200 lbs. All are solid checkers, except maybe for Ritchie, who was not signed to the bottom six.

My belief is that Dubas was trying to build a very different type of NHL team, which he might not be allowed to do so anymore.

Since they fired Babcock, the overhaul to the roster has been significant. But this year, it is finally settled. They went with size and established checkers in the NHL, which is old school.

It is a general principle with numbers. Misapply the numbers, and the results are not there. That really has been Dubas with the players he brought in and kicked out.

The revolving door for players, goalie, forwards, defense, suggests that the model he was using was wanting, perhaps he did not understand it himself, how it applied to the NHL game.

There will be just as much turnover next year, and Dubas will once again find great value at the bottom of his roster like he has every year, without fail.

 

[zeke]

Bottom 6 and Bottom 3 year by year


18-19

Johnsson - Kadri - Kapanen
Moore - Gauthier - Brown

Gardiner
Zaitsev
Dermott


19-20

Johnsson - Kerfoot - Kapanen
Clifford - Engvall - Spezza

Barrie
Ceci
Dermott


20-21

Engvall/Galy - Kerfoot - Mikheyev
Thornton - Spezza - Simmonds

Holl
Bogosian
Sandin/Dermott


21/22

Ritchie/Kerfoot - Kampf - Kase
Mikheyev - Spezza - Simmonds

Holl
Sandin
Lilly/Dermott