Boston Bruins Advanced Stat Thread

[Dennis Bonvie]

Me neither, I don't put Ketchup on anything it's disgusting.

So my take away reading here.
If the advanced stats don't say what we want them to say then we chalk it up to bad luck or poor goaltending.

If however they say what we want them to say then no such excuses are offered up for that player.

I don't subscribe to advanced stats because I see way to many variables for them to tell the whole and accurate story. Proof of that is when we attribute results we don't like to bad luck and poor goaltending . Which can actually be factual but the stats don't tell us that.
For me they complicate what
I enjoy the most which is watching the game. Only to be told I didn't actually see what I thought I did.

Stats are but one tool among a tool box full and way to much work for the miniscule advantage they provide IMO.

Ketchup on burgers, please, where it belongs.

 

[RussellmaniaKW]

Me neither, I don't put Ketchup on anything it's disgusting.

So my take away reading here.
If the advanced stats don't say what we want them to say then we chalk it up to bad luck or poor goaltending.

If however they say what we want them to say then no such excuses are offered up for that player.

I don't subscribe to advanced stats because I see way to many variables for them to tell the whole and accurate story. Proof of that is when we attribute results we don't like to bad luck and poor goaltending . Which can actually be factual but the stats don't tell us that.
For me they complicate what
I enjoy the most which is watching the game. Only to be told I didn't actually see what I thought I did.

Stats are but one tool among a tool box full and way to much work for the miniscule advantage they provide IMO.

it's pretty impressive that you're able to observe and perfectly analyze every single event that happens on the ice in every single game of a 32 team league with 82 games per team, but the rest of us mere mortals need to rely on stats to help us out with the 99% of events we didn't have time to see. Please allow us this minor concession.

 

[RussellmaniaKW]

thats why this is here as I said in first post.

this thread is slowly getting flooded with people who clearly have no interest in discussing analytics and are simply here to dig in their heels on arguing the worth of certain players with virtually no data, advanced or otherwise, to back up their arguments. Is that what this thread is for?

several of us are genuinely happy to discuss this topic in good faith with people. contrary to appearances this is absolutely not about idealism. It's just about understanding hockey better and understanding some of these modern tools that help us do that. But clearly there are people here who are joining this thread to make purely idealistic arguments and it's really frustrating that it's being allowed in the one thread that is supposed to take idealism out of the equation.

 

[RussellmaniaKW]

So the Bruins should fire all their scouts?

this makes as much sense as asking if people should stop seeing doctors because WebMD exists.

We live in a time where technology and crowd-sourcing has allowed us to aggregate more information on various topics than ever before. That information requires context and in most cases requires specialized knowledge (scouts, doctors, mechanics, etc) to most effectively extract value from that information.

Obviously scouts have an important place in hockey, even in a world with lots of available analytics. But even the most well-funded, devoted scouting staff will only be able to view a fraction of the hockey happening in the world in any given season. Advanced stats are not perfect, but they provide more baked in context and higher fidelity information about hockey events than just the basic stuff like goals and +/-.

Saying the Bruins should fire all their scouts would make no more sense than saying the scouts shouldn't use any stats to supplement their scouting.

 

[Gee Wally]

this thread is slowly getting flooded with people who clearly have no interest in discussing analytics and are simply here to dig in their heels on arguing the worth of certain players with virtually no data, advanced or otherwise, to back up their arguments. Is that what this thread is for?

several of us are genuinely happy to discuss this topic in good faith with people. contrary to appearances this is absolutely not about idealism. It's just about understanding hockey better and understanding some of these modern tools that help us do that. But clearly there are people here who are joining this thread to make purely idealistic arguments and it's really frustrating that it's being allowed in the one thread that is supposed to take idealism out of the equation.

as long as people are civil they can enter. Anyone is free to use the ignore option if desired.

 

[RussellmaniaKW]

one of the things that keeps bugging me about this debate is just how fundamentally crazy it is that people will reject something like expected goals in favor of something like +/-.

+/- tells you literally one thing and one thing only: was this player on the ice when a goal happened at even strength? Goals are obviously important, it's like the whole point of hockey, but in a sport with 12 guys on the ice most of the time, just knowing if a goal was scored tells you next to nothing about whether that guy is any good or contributed to that goal in any meaningful way.

^ this is what we mean when we say +/- is a flawed stat. It's just extremely limited in its usefulness in evaluating individual players on its own. Even over a long period of time if a guy has a really positive +/- it doesn't tell you with much certainty if that player actually played well. Maybe he's just the 5th wheel on a really excellent team.

On the other side you have something like expected goals (xG). First, it's not contingent on whether an actual goal was scored, which means you have a lot (orders of magnitude) more data points to work with. This makes the data less prone to "luck". Additionally xG models include information about the type of shot, where was it taken, how dangerous it was, etc.

Is xG perfect? Does it tell you everything you could want to know about a scoring chance without seeing it? Obviously not. It doesn't tell you about the passing play before the shot, or the competence of the shooter, for example. It also doesn't tell you if they were scared of a big, tall defenseman nearby or if they had bubble-guts from the wings they had last night. But even just the basic shot quality & location information still tells you a hell of a lot more about the individual player's performance than +/- does.

So given all of the above, why would anyone ever reject the use of something like expected goals when it simply gives you more, higher fidelity information than another stat like +/-. And this is especially baffling to me when you are allowed to use both! this doesn't have to be a choice. It's not religion, people! It's just information!

 

[neverwatchthegames]

one of the things that keeps bugging me about this debate is just how fundamentally crazy it is that people will reject something like expected goals in favor of something like +/-.

+/- tells you literally one thing and one thing only: was this player on the ice when a goal happened at even strength? Goals are obviously important, it's like the whole point of hockey, but in a sport with 12 guys on the ice most of the time, just knowing if a goal was scored tells you next to nothing about whether that guy is any good or contributed to that goal in any meaningful way.

^ this is what we mean when we say +/- is a flawed stat. It's just extremely limited in its usefulness in evaluating individual players on its own. Even over a long period of time if a guy has a really positive +/- it doesn't tell you with much certainty if that player actually played well. Maybe he's just the 5th wheel on a really excellent team.

A few more pot shots at +/-:

• counts empty net and special teams goals the same
• obscures good players on bad teams
• counts if you are in the middle of a line change

 

[sarge88]

one of the things that keeps bugging me about this debate is just how fundamentally crazy it is that people will reject something like expected goals in favor of something like +/-.

+/- tells you literally one thing and one thing only: was this player on the ice when a goal happened at even strength? Goals are obviously important, it's like the whole point of hockey, but in a sport with 12 guys on the ice most of the time, just knowing if a goal was scored tells you next to nothing about whether that guy is any good or contributed to that goal in any meaningful way.

^ this is what we mean when we say +/- is a flawed stat. It's just extremely limited in its usefulness in evaluating individual players on its own. Even over a long period of time if a guy has a really positive +/- it doesn't tell you with much certainty if that player actually played well. Maybe he's just the 5th wheel on a really excellent team.

On the other side you have something like expected goals (xG). First, it's not contingent on whether an actual goal was scored, which means you have a lot (orders of magnitude) more data points to work with. This makes the data less prone to "luck". Additionally xG models include information about the type of shot, where was it taken, how dangerous it was, etc.

Is xG perfect? Does it tell you everything you could want to know about a scoring chance without seeing it? Obviously not. It doesn't tell you about the passing play before the shot, or the competence of the shooter, for example. It also doesn't tell you if they were scared of a big, tall defenseman nearby or if they had bubble-guts from the wings they had last night. But even just the basic shot quality & location information still tells you a hell of a lot more about the individual player's performance than +/- does.

So given all of the above, why would anyone ever reject the use of something like expected goals when it simply gives you more, higher fidelity information than another stat like +/-. And this is especially baffling to me when you are allowed to use both! this doesn't have to be a choice. It's not religion, people! It's just information!

I'm going to go out on a limb here and open myself up to ridicule and explain why I struggle with expected goals for, and a lot of the other "projections", while at the same time fully stating a willingness to be objective about them and wanting to learn more.

The problem I (and I'm guessing many people have) is this ----

Hockey is so free flowing an unpredictable that literally a stick blade being 8 inches to the left or right or a shot hitting the bottom portion of a 4 inch cross bar, as opposed to the top portion provides entirely different results.

Every single player could do everything right during a shift and McDavid can snipe a top shelf goal from an impossible angle and that somehow leads to a prediction regarding the players that were on the ice at the time and what will happen in the future, even when players less skilled than McDavid are the opponent.

To me, what McDavid does when a certain line is on the ice is going to be different when Draisaitl is with him....which will be different when the two of them are paired with Pulijarvvi vs. when they're paired with Hyman.

Sure, each of them have tendencies, but do they have tendencies as a line? (I'm sincerely asking -- do they track these stats based on who they are playing with and can those stats be pulled up immediately during a game?)

For instance, I think it stands to reason that the Bergeron - Marchand pair plays differently when Pasta is with them vs. JDB.

Do the stats show that Marchand shoots less with Pasta than JDB? Does Bergeron do something different on the faceoff when JDB is on the ice with him?

Projecting or predicting what is going to happen in a given game or season just seems to be incredibly tenuous simply because of the speed of the game and immense variables that exist.

I get that the predictive stats don't exist to guarantee anything, but they also don't/can't take certain things into consideration such as a players health, effort, ability to perform under pressure etc.

If I can jump to baseball for a second. I'm sure that there are/were predictive models that we could have looked at to determine the likelihood of David Ortiz getting a hit or making an out in a playoff game vs. a particular pitch or pitcher based on what happened throughout that season, his matchups with that pitcher or his career in general.

I don't think there's a person alive who watched that guy play here for 15 years or so that would put one iota of credibility in anything that suggested he'd make an out with the game on the line.

And, IMO baseball has fewer variables than hockey.

Lastly, I'm not trying to be argumentative -- just being honest about the things that I think about with predictive stats and I'm happy to learn more about them.

 

[JOKER 192]

it's pretty impressive that you're able to observe and perfectly analyze every single event that happens on the ice in every single game of a 32 team league with 82 games per team, but the rest of us mere mortals need to rely on stats to help us out with the 99% of events we didn't have time to see. Please allow us this minor concession.

Never made that claim but whatever

This is a really poor representation of what several posters have taken time to carefully write out in many posts in this thread and elsewhere.

Is that what the stats say?

 

[neverwatchthegames]

I'm going to go out on a limb here and open myself up to ridicule and explain why I struggle with expected goals for, and a lot of the other "projections", while at the same time fully stating a willingness to be objective about them and wanting to learn more.

The problem I (and I'm guessing many people have) is this ----

Hockey is so free flowing an unpredictable that literally a stick blade being 8 inches to the left or right or a shot hitting the bottom portion of a 4 inch cross bar, as opposed to the top portion provides entirely different results.

Every single player could do everything right during a shift and McDavid can snipe a top shelf goal from an impossible angle and that somehow leads to a prediction regarding the players that were on the ice at the time and what will happen in the future, even when players less skilled than McDavid are the opponent.

To me, what McDavid does when a certain line is on the ice is going to be different when Draisaitl is with him....which will be different when the two of them are paired with Pulijarvvi vs. when they're paired with Hyman.

Sure, each of them have tendencies, but do they have tendencies as a line? (I'm sincerely asking -- do they track these stats based on who they are playing with and can those stats be pulled up immediately during a game?)

For instance, I think it stands to reason that the Bergeron - Marchand pair plays differently when Pasta is with them vs. JDB.

Do the stats show that Marchand shoots less with Pasta than JDB? Does Bergeron do something different on the faceoff when JDB is on the ice with him?

Projecting or predicting what is going to happen in a given game or season just seems to be incredibly tenuous simply because of the speed of the game and immense variables that exist.

I get that the predictive stats don't exist to guarantee anything, but they also don't/can't take certain things into consideration such as a players health, effort, ability to perform under pressure etc.

If I can jump to baseball for a second. I'm sure that there are/were predictive models that we could have looked at to determine the likelihood of David Ortiz getting a hit or making an out in a playoff game vs. a particular pitch or pitcher based on what happened throughout that season, his matchups with that pitcher or his career in general.

I don't think there's a person alive who watched that guy play here for 15 years or so that would put one iota of credibility in anything that suggested he'd make an out with the game on the line.

And, IMO baseball has fewer variables than hockey.

Lastly, I'm not trying to be argumentative -- just being honest about the things that I think about with predictive stats and I'm happy to learn more about them.

My crude understanding is that the basis for expected goals comes from scoring chances generated. Boiled down... It looks at the shot attempt situations/locations that a player is involved with, and--based on weighted values for the quality of the shot location and constituent play components that immediately precede it --creates an approximate idea of the rate at which one should generally expect goals if the same plays were run over and over.

 

[MarchysNoseKnows]

Is that what the stats say?

Yes.

 

[JOKER 192]

Yes.

Cool

 

[Dennis Bonvie]

A few more pot shots at +/-:

• counts empty net and special teams goals the same
• obscures good players on bad teams
• counts if you are in the middle of a line change

It does?

 

[RussellmaniaKW]

It does?

yes. A shorty is a plus. an EN goal is a plus.

Plus–minus - Wikipedia

en.wikipedia.org

 

[Dennis Bonvie]

yes. A shorty is a plus. an EN goal is a plus.

Plus–minus - Wikipedia

en.wikipedia.org

But a power play goal is not.

I believe a power play goal is a special teams goal.

 

[RussellmaniaKW]

I'm going to go out on a limb here and open myself up to ridicule and explain why I struggle with expected goals for, and a lot of the other "projections", while at the same time fully stating a willingness to be objective about them and wanting to learn more.

The problem I (and I'm guessing many people have) is this ----

Hockey is so free flowing an unpredictable that literally a stick blade being 8 inches to the left or right or a shot hitting the bottom portion of a 4 inch cross bar, as opposed to the top portion provides entirely different results.

Every single player could do everything right during a shift and McDavid can snipe a top shelf goal from an impossible angle and that somehow leads to a prediction regarding the players that were on the ice at the time and what will happen in the future, even when players less skilled than McDavid are the opponent.

To me, what McDavid does when a certain line is on the ice is going to be different when Draisaitl is with him....which will be different when the two of them are paired with Pulijarvvi vs. when they're paired with Hyman.

Sure, each of them have tendencies, but do they have tendencies as a line? (I'm sincerely asking -- do they track these stats based on who they are playing with and can those stats be pulled up immediately during a game?)

For instance, I think it stands to reason that the Bergeron - Marchand pair plays differently when Pasta is with them vs. JDB.

Do the stats show that Marchand shoots less with Pasta than JDB? Does Bergeron do something different on the faceoff when JDB is on the ice with him?

Projecting or predicting what is going to happen in a given game or season just seems to be incredibly tenuous simply because of the speed of the game and immense variables that exist.

I get that the predictive stats don't exist to guarantee anything, but they also don't/can't take certain things into consideration such as a players health, effort, ability to perform under pressure etc.

If I can jump to baseball for a second. I'm sure that there are/were predictive models that we could have looked at to determine the likelihood of David Ortiz getting a hit or making an out in a playoff game vs. a particular pitch or pitcher based on what happened throughout that season, his matchups with that pitcher or his career in general.

I don't think there's a person alive who watched that guy play here for 15 years or so that would put one iota of credibility in anything that suggested he'd make an out with the game on the line.

And, IMO baseball has fewer variables than hockey.

Lastly, I'm not trying to be argumentative -- just being honest about the things that I think about with predictive stats and I'm happy to learn more about them.

@burstnbloom would surely do a better job mitigating some of these concerns than I could, but I'll give a high level reaction anyway:

First of all, I think a big thing about expected goals that people don't seem to get is that it has been shown repeatedly that the top xG models show a very strong correlation between expected goals & actual goals in aggregate. There will always be outliers, but over time, the larger the data set gets the more expected & actual goals converge. So much so, that when they diverge you can reasonably assume that you are observing an outlier and that with enough time that outlier (a player finishing way above their xG rate, for example) will regress back to their xG rate.

Secondly, I don't think anyone professes or expects that any xG model can be used to reliably predict short-term results. Like I said before, you can reasonably expect that a player will score at approximately their expected goal rate over time, but that doesn't mean you can accurately predict that they'll score at that rate over the next month of games for example. There will always be variance. That variance smooths out over time, but the existence of that variance (and the inability to quantify every possible variable that goes into a scoring chance) is what makes playing and watching the sport still worthwhile.

Lastly, it's totally valid to have a hard time with putting so much stock in analytics models that still have so many fundamental blind spots. I've had a hard time wrapping my head around it too (and still do). This will surely get easier over time as the NHL makes more tracking data available and the amateur stats guys don't have to infer so much from shot attempt data alone. With that said, expected goals is still a reasonable approach to try and quantify quality of play based on the types of events that are intrinsically tied to success & failure in hockey. It's the best thing we have today and it's way better than just having nothing or even just looking at actual goals. Nobody is saying it's perfect or even close to it, but it's still way more valuable than just being limited to watching a fraction of all the possible NHL minutes and moments and having to draw conclusions from that very limited perspective.

But a power play goal is not.

I believe a power play goal is a special teams goal.

yes, but I don't think @neverwatchthegames was saying all special teams, just that pointing out that broadly different situations still count +/- the same.

 

[neverwatchthegames]

yes, but I don't think @neverwatchthegames was saying all special teams, just that pointing out that broadly different situations still count +/- the same.

Sorry. Yeah, there's a bunch of atypical situations that count. Not whistle in whistle out power plays, but all the in-betweens. Like extra skater from a delayed penalty, pulled goalie, player down, no stick, etc...

 

[burstnbloom]

@burstnbloom would surely do a better job mitigating some of these concerns than I could, but I'll give a high level reaction anyway:

First of all, I think a big thing about expected goals that people don't seem to get is that it has been shown repeatedly that the top xG models show a very strong correlation between expected goals & actual goals in aggregate. There will always be outliers, but over time, the larger the data set gets the more expected & actual goals converge. So much so, that when they diverge you can reasonably assume that you are observing an outlier and that with enough time that outlier (a player finishing way above their xG rate, for example) will regress back to their xG rate.

Secondly, I don't think anyone professes or expects that any xG model can be used to reliably predict short-term results. Like I said before, you can reasonably expect that a player will score at approximately their expected goal rate over time, but that doesn't mean you can accurately predict that they'll score at that rate over the next month of games for example. There will always be variance. That variance smooths out over time, but the existence of that variance (and the inability to quantify every possible variable that goes into a scoring chance) is what makes playing and watching the sport still worthwhile.

Lastly, it's totally valid to have a hard time with putting so much stock in analytics models that still have so many fundamental blind spots. I've had a hard time wrapping my head around it too (and still do). This will surely get easier over time as the NHL makes more tracking data available and the amateur stats guys don't have to infer so much from shot attempt data alone. With that said, expected goals is still a reasonable approach to try and quantify quality of play based on the types of events that are intrinsically tied to success & failure in hockey. It's the best thing we have today and it's way better than just having nothing or even just looking at actual goals. Nobody is saying it's perfect or even close to it, but it's still way more valuable than just being limited to watching a fraction of all the possible NHL minutes and moments and having to draw conclusions from that very limited perspective.


yes, but I don't think @neverwatchthegames was saying all special teams, just that pointing out that broadly different situations still count +/- the same.

Well done, Russell. @sarge88 just so you know, I will only ridicule people who show willful ignorance. If more people were interested in learning, we'd all be better off.

Russell has this right. There are a billion variables in hockey. The xG models are trying to isolate the chance that a certain shot and a certain situation goes into the net. They literally use millions of data point to do that.
Whether or not you agree with the methodology or not is frankly irrelevant because there is a statistically significant correlation between xG models and actual goal for models over a long enough timeline. They will always intersect. There's no way to tell when they will intersect but eventually they will. that uncertainty can be unsettling but the correlation is real.

I'll use another baseball example back at you. In the first half of last year, Alex Verdugo had the batted ball profile of a player with an .800 OPS but his results were a .678 OPS in the first half. Guess what he did in the second half? .803 OPS. there wasn't a major change in xslug% the lines just intersected.

The next frontier for xG is to isolate shooting talent. Right now these guys know that shooting talent exists. Auston Matthews and Pasta and other goal scorers with elite shots consistently outperform their xG by 3-5 goals per year. It's enough that we can see there's something there but without player tracking and shot metrics, we cant be sure of their significance so they aren't tackled.

TLR - yes they are imperfect but what you need to know is that there is a reliable correlation between xG performance and G performance over a large enough sample.

 

[burstnbloom]

To be honest he's not as bad as anti-forbort people make it seem.

He's an NHL defenseman with a niche. On this team in particular we just project higher expectations (You'd think for 3 mil this would be the case).

Orlov since he's not played with Forbort, dominated.

Clifton's stats away from Forbort, fantastic.

No players stats with Forbort are better with than without haha. Every other defenseman on this team is carrying him when he's on the ice at even strength.

Just my opinion but he's the weakest link on a team with not very many weaknesses. And to me that's ok to point out.

I legit think hes one of the 10 worst 5v5 d in the league. He made Charlie McAvoy bad.

 

[burstnbloom]

How does that demonstrate the plus/minus is flawed? It tells us exactly where Forbort fits on the defense.

He played RD some early on this season before McAvoy came back. Played there last season also when paired with Riley or Zboril or Vaaka. But that's by memory of eye test so it could be wrong.

They always turned the other guy around and left Forbort on the left because apparently potato hands can't turn towards the boards.

 

[Bruinswillwin77]

I legit think hes one of the 10 worst 5v5 d in the league. He made Charlie McAvoy bad.

He has a decent shot but his skating lacks.

 

[neverwatchthegames]

This would be a good spot to collect some player cards for general use.

 

[NDiesel]

Well done, Russell. @sarge88 just so you know, I will only ridicule people who show willful ignorance. If more people were interested in learning, we'd all be better off.

Russell has this right. There are a billion variables in hockey. The xG models are trying to isolate the chance that a certain shot and a certain situation goes into the net. They literally use millions of data point to do that.
Whether or not you agree with the methodology or not is frankly irrelevant because there is a statistically significant correlation between xG models and actual goal for models over a long enough timeline. They will always intersect. There's no way to tell when they will intersect but eventually they will. that uncertainty can be unsettling but the correlation is real.

I'll use another baseball example back at you. In the first half of last year, Alex Verdugo had the batted ball profile of a player with an .800 OPS but his results were a .678 OPS in the first half. Guess what he did in the second half? .803 OPS. there wasn't a major change in xslug% the lines just intersected.

The next frontier for xG is to isolate shooting talent. Right now these guys know that shooting talent exists. Auston Matthews and Pasta and other goal scorers with elite shots consistently outperform their xG by 3-5 goals per year. It's enough that we can see there's something there but without player tracking and shot metrics, we cant be sure of their significance so they aren't tackled.

TLR - yes they are imperfect but what you need to know is that there is a reliable correlation between xG performance and G performance over a large enough sample.

The bolded greatly interests me, I believe I asked in another thread if any models existed like this.

In my head it doesn't seem like it would be impossible to do using a combination of individual shooting% for each different shot type/positions and league average shooting% for those same types of shots/positions. I would think by doing this you can then start to apply a sort of "weight" on each shot based on what each individual does with each shot type and position on the ice compared to what the average player does from that same spot.

Obviously there is a lot of work to hone in the accuracy of such a model, but I look forward to some of these much smarter people working towards this.

 

[MarchysNoseKnows]

The bolded greatly interests me, I believe I asked in another thread if any models existed like this.

In my head it doesn't seem like it would be impossible to do using a combination of individual shooting% for each different shot type/positions and league average shooting% for those same types of shots/positions. I would think by doing this you can then start to apply a sort of "weight" on each shot based on what each individual does with each shot type and position on the ice compared to what the average player does from that same spot.

Obviously there is a lot of work to hone in the accuracy of such a model, but I look forward to some of these much smarter people working towards this.

The teams have it already. As soon as it’s public the analytics sites will be going crazy.

 

[NDiesel]

The teams have it already. As soon as it’s public the analytics sites will be going crazy.

That's very cool to hear, hopefully it comes out sooner rather than later