Some suprising (and not so suprising) Habs player stats.

[Talks to Goalposts]

too much emphasis on stats, not enough on watching games.

I'd imagine I do much more watching game then most, because I do a lot of rewatching video so I can look over what happened dispassionately and figure out how things work.

But I can't exactly bring you over and watch 20 hours of tape to illustrate my points so when I'm going to formulate and support an argument I'm going to use something you can check for yourself if you want rather than taking my word for it.

But for this thread were really just talking about some elementary concepts used in statistical analysis that capture the broad strokes of what goes on in an NHL season. And an NHL season is too large a collection of events to talk about in anything other than statistical terms. You can describe a family qualitatively, but a village requires numerical terms.

 

[ECWHSWI]

I'd imagine I do much more watching game then most, because I do a lot of rewatching video so I can look over what happened dispassionately and figure out how things work.

But I can't exactly bring you over and watch 20 hours of tape to illustrate my points so when I'm going to formulate and support an argument I'm going to use something you can check for yourself if you want rather than taking my word for it.

But for this thread were really just talking about some elementary concepts used in statistical analysis that capture the broad strokes of what goes on in an NHL season. And an NHL season is too large a collection of events to talk about in anything other than statistical terms. You can describe a family qualitatively, but a village requires numerical terms.

chances are, you'd do as good if you were just to explain your thoughts

 

[Talks to Goalposts]

chances are, you'd do as good if you were just to explain your thoughts

You can't separate it as much as you'd think. I believe strongly that you need to check assessments against concrete information. And by studying how the numbers fit together you get a much better idea of what's important when it happens on the ice.

For example, when Gionta flubs what could have been a tap in goal the first time you see it you're inclined to say he sucks and throw high balls at the screen. Looking over it on the broad scale, the key thing actually was that his line generated that chance and that over time that's going to result in good things for the Red, White and Blue. Or that while Cole isn't scoring he's still the same guy in terms of gaining the zone clean to set up the attack and he's still generating offense, just potential goals are getting chipped away by goal posts and saves and minor screw ups. Thinking probabilistic rather than mechanically is counter-intuitive, put its key to understanding how a inherently chaotic sport.

 

[ECWHSWI]

You can't separate it as much as you'd think. I believe strongly that you need to check assessments against concrete information. And by studying how the numbers fit together you get a much better idea of what's important when it happens on the ice.

For example, when Gionta flubs what could have been a tap in goal the first time you see it you're inclined to say he sucks and throw high balls at the screen. Looking over it on the broad scale, the key thing actually was that his line generated that chance and that over time that's going to result in good things for the Red, White and Blue. Or that while Cole isn't scoring he's still the same guy in terms of gaining the zone clean to set up the attack and he's still generating offense, just potential goals are getting chipped away by goal posts and saves and minor screw ups. Thinking probabilistic rather than mechanically is counter-intuitive, put its key to understanding how a inherently chaotic sport.

not really, more like if he does it over and over again...

 

[Timelord]

or maybe they are giving up too many quality chances...

And all conventional early season stats (percentages, averages) tend to revert to a mean value, how much it does depending on the team's/player's abilities.

 

[DAChampion]

Some may not understand,

But the statistics are themselves a game.

Those of us who are nerds like to be able to solve things and to understand things. The challenge of sabremetrics is to be able to measure hockey players and hockey teams to the limit of statistical precision, and to bet able to adequately state the size of this precision.

Is this a solved thing? No, but there's been tremendous progress on the issue and I'm really impressed by the people involved, for the most part. They don't know everything of course, they don't know as much as somebody who watches a lot of games, but they know a lot and are gaining ground fast.

I'm guessing the big problem in hockey stats now is the lack of proper counting. For example, some stats like takeaways are counted differently in every building, that's a pain.

 

[Kjell Dahlin]

Correct me if I am wrong but the main flaw in behindthenet.ca et al., is that they rely heavily on shots on goal. I don’t blame them: it is an easy (a shot is a shot; interpretation plays an insignificant role) stats to record and it is also correlated to scoring chances. The more scoring chances you have, the higher your chances are at winning games.

However the correlation between “shots for” and “team win” is not 100%. In fact I just calculated it and, thus far this season, the correlation factor between “team win” and “shots for” is... 0,224. It is very low.

What I am saying here, is that inherently, at his very core, the system includes a significant “margin of error”.

I guess that over a long period of time, the correlation would be higher (though it would never reach 1) but it does underline the fact that micro stats are a decision helping tool. Nothing less, nothing more.

That said, am I right in assuming that they rely heavily on shots on goals?


EDIT:

I just noticed that I did not use AVERAGE shots on goal per game... give me one minute and I will update the correlation factor! I just did and the correlation factor is still very low. Based on last season, the correlation factor between “Team points” and “Shots for” is 0,454.

 

[Talks to Goalposts]

Correct me if I am wrong but the main flaw in behindthenet.ca et al., is that they rely heavily on shots on goal. I don’t blame them: it is an easy (a shot is a shot; interpretation plays an insignificant role) stats to record and it is also correlated to scoring chances. The more scoring chances you have, the higher your chances are at winning games.

However the correlation between “shots for†and “team win†is not 100%. In fact I just calculated it and, thus far this season, the correlation factor between “team win†and “shots for†is... 0,224. It is very low.

What I am saying here, is that inherently, at his very core, the system includes a significant “margin of errorâ€.

I guess that over a long period of time, the correlation would be higher (though it would never reach 1) but it does underline the fact that micro stats are a decision helping tool. Nothing less, nothing more.

That said, am I right in assuming that they rely heavily on shots on goals?


EDIT:

I just noticed that I did not use AVERAGE shots on goal per game... give me one minute and I will update the correlation factor! I just did and the correlation factor is still very low. Based on last season, the correlation factor between “Team points†and “Shots for†is 0,454.

There is this thing called "score effects" on shot totals, the team that leads plays conservatively and shots less while the losing team plays aggressively and shoots more. So paradoxically while long term shot differentials are a great measurement of team play, who wins and loses a particular game doesn't line up well with who out shot who.

Which is why when measuring results its preferred to look at what happens when the score is tied or close to counteract this effect. Long term who is doing well in shots during close situations is stable and collrelates strongly with success but that is in the long run while short term goaltending and scoring streaks predominant.

 

[Kjell Dahlin]

There is this thing called "score effects" on shot totals, the team that leads plays conservatively and shots less while the losing team plays aggressively and shoots more. So paradoxically while long term shot differentials are a great measurement of team play, who wins and loses a particular game doesn't line up well with who out shot who.

Which is why when measuring results its preferred to look at what happens when the score is tied or close to counteract this effect. Long term who is doing well in shots during close situations is stable and collrelates strongly with success but that is in the long run while short term goaltending and scoring streaks predominant.

Thanks! Do you know the correlation factor in such situations? 0,7? 0,9?

Unless you are telling me that the correlation is really high (> 0,95), inherently, at its very core, the system does include a significant “margin of error” right?

 

[ECWHSWI]

Some may not understand,

But the statistics are themselves a game.

Those of us who are nerds like to be able to solve things and to understand things. The challenge of sabremetrics is to be able to measure hockey players and hockey teams to the limit of statistical precision, and to bet able to adequately state the size of this precision.

Is this a solved thing? No, but there's been tremendous progress on the issue and I'm really impressed by the people involved, for the most part. They don't know everything of course, they don't know as much as somebody who watches a lot of games, but they know a lot and are gaining ground fast.

I'm guessing the big problem in hockey stats now is the lack of proper counting. For example, some stats like takeaways are counted differently in every building, that's a pain.

I'd suggest you start watching more games then, stats may tell you a few things but they sure wont tell you who's playing well and who's not.

for example some systems have teams letting the opponent shoot from the outside (like we did vs WSH and PIT in the po a few years back), other teams will have their players block a lot of shots, and so on...

some players are more of the passer type, others are more of the "parking his ass in front of the net" type, these guys wont get that many shots in games, sometimes none, doesnt mean they're playing a bad game at all...

if anything, stats will help you understand a very specific aspect of a player's game but it will never tells you about the overall game of said player...

the only way you get an "understanding" of how a player is doing using stats is by doing what the Goalpost user is doing, watch games 2, 3, 4+ times...

 

[Talks to Goalposts]

Thanks! Do you know the correlation factor in such situations? 0,7? 0,9?

Unless you are telling me that the correlation is really high (> 0,95), inherently, at its very core, the system does include a significant “margin of error” right?

By my recollection of the studies that have been done on it they tend to come up with a ballpark correlation of about .7 of long term team success explainable by ability to out-shoot the opposition at even strength. The remaining third is explainable by special teams play and goaltending with a pretty tiny 5-10% based on shot quality. The shots ratios tend to correspond extremely closely with things like offensive/defensive zone time and puck possession (.8-.95 in the studies I've seen) so these numbers are often referred to as possession statistics. They aren't the only thing that goes into a team winning but they are such a strong influence that its very hard to be good without them, especially as the number of penalties has so sharply declined. Prime Markov era Montreal was a poster child for being able to do better in the standings than the shots on the basis of a lethal power play and a succession of strong goaltending performances but that was only viable when penalties were called a lot more.

The big problem with offensive shot quality is that while it does exist on the team level it's so heavily effected by variance that by the time you've put together a large enough sample to detect it conditions will have changed so it doesn't tell you anything useful for the future. So over the long term team shooting will tend strongly towards league average of around 8%, with forwards taking about 75% of the shots and scoring on 9% of them while defensemen take about 25% of the shots and scoring on 4-5%. Individual forwards routinely beat that 9% average but its pretty rare for a line to consistently do so.

 

[DAChampion]

I'd suggest you start watching more games then, stats may tell you a few things but they sure wont tell you who's playing well and who's not.

for example some systems have teams letting the opponent shoot from the outside (like we did vs WSH and PIT in the po a few years back), other teams will have their players block a lot of shots, and so on...

some players are more of the passer type, others are more of the "parking his ass in front of the net" type, these guys wont get that many shots in games, sometimes none, doesnt mean they're playing a bad game at all...

if anything, stats will help you understand a very specific aspect of a player's game but it will never tells you about the overall game of said player...

the only way you get an "understanding" of how a player is doing using stats is by doing what the Goalpost user is doing, watch games 2, 3, 4+ times...

There is no dichotomy between watching games and analyzing and designing stats.

 

[Habset]

I don't think stats are the tell all but get enough of them together and you can create yourself a nice picture of how a team/player is doing.

Being able to understand the stats is also another thing. For example POD can be said to be "luck" but by comparing shot % to save% while on the ice tells us a lot more then luck. It tells us how many goals per chances on both sides of the ice while that player is on the ice which is good to know but once again not tell all!

As an example of how important interpretation of the stats is look at the guy above who saw White and said he's a liability. Yes the penaltys part does show that but if I looked at those stats with the high POd, Save % and corsi with a low shot % I'd say he's a good defensive player who keeps the puck in the offensive zone but isn't very offensive and needs to be better with his penalties.

 

[Miller Time]

You can't separate it as much as you'd think. I believe strongly that you need to check assessments against concrete information. And by studying how the numbers fit together you get a much better idea of what's important when it happens on the ice.

For example, when Gionta flubs what could have been a tap in goal the first time you see it you're inclined to say he sucks and throw high balls at the screen. Looking over it on the broad scale, the key thing actually was that his line generated that chance and that over time that's going to result in good things for the Red, White and Blue. Or that while Cole isn't scoring he's still the same guy in terms of gaining the zone clean to set up the attack and he's still generating offense, just potential goals are getting chipped away by goal posts and saves and minor screw ups. Thinking probabilistic rather than mechanically is counter-intuitive, put its key to understanding how a inherently chaotic sport.

exactly where this type of approach falls terribly short...

the object of a hockey game isn't to "generate scoring chances", it's to score goals.

the lower level professional leagues are littered with athletes who, were they inserted into an NHL lineup, could generate ample scoring chances. the ones who stick, and then the ones who excel, and then the ones who excel consistently, are separated by intangible qualities that are far more accurately observed by an expert then they are by a statistical analysis.

that gionta is drifting more towards that territory, than continuing to efficiently execute, is troubling, and while many factors could be contributing, one of them may very well be that the tools he relied on in his youth to succeed are no longer as effective, and he failed to put in the necessary time/focus to make the improvements in his game that those players who maintain their level of play into the later years of their careers do.

When we played the devils, I caught the U.S feed, and they interviewed Gionta's parents... in describing the two boys, they pointed out that Brian was far more talented/athletic, whereas Stephen (the journeyman who made it to the NHL much later in his career) never had the same "natural" skill.

despite B.Gionta's small size, he was absolutely dominant in his jr & NCAA career (right from his first 18-year old season)... he wouldn't be the first "natural talent" who dropped off sharply as his physical tools ceased to be the advantage they were early in his career.


One would think that after using this same statistical approach to understanding a player's career progression failed so miserably, as it did with the rebuttals once applied to Gomez & his inevitable return to form given the #'s, that it would at least open the door to a consideration that maybe...just maybe, the numbers DO lie

 

[ECWHSWI]

There is no dichotomy between watching games and analyzing and designing stats.

missed the point.

 

[ECWHSWI]

I don't think stats are the tell all but get enough of them together and you can create yourself a nice picture of how a team/player is doing.

Being able to understand the stats is also another thing. For example POD can be said to be "luck" but by comparing shot % to save% while on the ice tells us a lot more then luck. It tells us how many goals per chances on both sides of the ice while that player is on the ice which is good to know but once again not tell all!

As an example of how important interpretation of the stats is look at the guy above who saw White and said he's a liability. Yes the penaltys part does show that but if I looked at those stats with the high POd, Save % and corsi with a low shot % I'd say he's a good defensive player who keeps the puck in the offensive zone but isn't very offensive and needs to be better with his penalties.

not much.

 

[Habset]

not much.

Well you my friend are negligent of the fact that there are is more then 1 way of analyzing something.

I admit that viewing is the best way of analyzing a player but stats are also helpful.

 

[ECWHSWI]

Well you my friend are negligent of the fact that there are is more then 1 way of analyzing something.

I admit that viewing is the best way of analyzing a player but stats are also helpful.

there is obviously, but putting a bunch of stats together and trying to paint the overall picture of a player isnt one of them. Not when some of those stats are redundant.

White isnt a "good" defensive player, he's a 4th liner after all - if he was that good, he'd be higher up the depth chart, same for offense, just have to look at his offensive numbers over the years to know that.

Stats can be useful, but having multiple stats analyzing more or less the same things is not.

 

[Kjell Dahlin]

exactly where this type of approach falls terribly short...

the object of a hockey game isn't to "generate scoring chances", it's to score goals.

the lower level professional leagues are littered with athletes who, were they inserted into an NHL lineup, could generate ample scoring chances. the ones who stick, and then the ones who excel, and then the ones who excel consistently, are separated by intangible qualities that are far more accurately observed by an expert then they are by a statistical analysis.

that gionta is drifting more towards that territory, than continuing to efficiently execute, is troubling, and while many factors could be contributing, one of them may very well be that the tools he relied on in his youth to succeed are no longer as effective, and he failed to put in the necessary time/focus to make the improvements in his game that those players who maintain their level of play into the later years of their careers do.

When we played the devils, I caught the U.S feed, and they interviewed Gionta's parents... in describing the two boys, they pointed out that Brian was far more talented/athletic, whereas Stephen (the journeyman who made it to the NHL much later in his career) never had the same "natural" skill.

despite B.Gionta's small size, he was absolutely dominant in his jr & NCAA career (right from his first 18-year old season)... he wouldn't be the first "natural talent" who dropped off sharply as his physical tools ceased to be the advantage they were early in his career.


One would think that after using this same statistical approach to understanding a player's career progression failed so miserably, as it did with the rebuttals once applied to Gomez & his inevitable return to form given the #'s, that it would at least open the door to a consideration that maybe...just maybe, the numbers DO lie

I just want to comment on “... the object of a hockey game isn't to "generate scoring chances", it's to score goals...â€.

Long terms team success is strongly (0,7 according to Talks to Goalposts) correlated to the ability to out-shoot the opposition at even strengths when the score is tied or close. The correlation is not 100% (inherently, at his core, it does include a “margin of error) but micro stats remains a very useful decision helping tool.

I do find it annoying when posters use micro stats without understanding its limitation; the op is a perfect example. On the other hand, if I learn that our Canadiens are not using them, I would be seriously annoyed!

 

[ECWHSWI]

Long terms team success is strongly (0,7 according to Talks to Goalposts) correlated to the ability to out-shoot the opposition at even strengths when the score is tied or close. The correlation is not 100% (inherently, at his core, it does include a “margin of error) but micro stats remains a very useful decision helping tool!

actually, it's obvious that outshooting the opponent will bring success more often than not. You need advanced / micro stats to know that ???

 

[Kjell Dahlin]

actually, it's obvious that outshooting the opponent will bring success more often than not. You need advanced / micro stats to know that ???

The objective here was to find an easy to track stats that is strongly correlated to long terms team success. Once you find that stats (even strength shots when the score is tied or close), you can use that info to “refine” or throw a different light on your players (individually) analysis.

That said... wise guy!


EDIT: I never put that exclamation point (helping tool!) in the quote you provided... wise guy x 2!

 

[JoelWarlord]

Just to clarify what PDO actually is (it has nothing to do with +/-), it's the sum of a players on ice shooting percentage and save percentage. It has a heavy tendency to regress to 1.000 in the long run, and if a guy is having a dynamite start and you see that his PDO is 1100, you can expect him to fall off as the averages play out. Conversely, if a guy is having a rough start, and if you check and his PDO is stting at 900, you can expect his production to increase over time. Some players do carry a PDO of higher than 1.000 over a course of a few seasons, which is where you start to see players who are good at creating chances and such (finishing on a higher percentage of shots).

 

[ECWHSWI]

The objective here was to find an easy to track stats that is strongly correlated to long terms team success. Once you find that stats (even strength shots when the score is tied or close), you can use that info to “refine†or throw a different light on your players (individually) analysis.

That said... wise guy!


EDIT: I never put that exclamation point (helping tool!) in the quote you provided... wise guy x 2!

understand that, but when you're creating stats to show the obvious (ex: outshooting opponent = more chances of success) you're not refining anything.


haha, make it X3!!!

 

[Miller Time]

I just want to comment on “... the object of a hockey game isn't to "generate scoring chances", it's to score goals...â€.

Long terms team success is strongly (0,7 according to Talks to Goalposts) correlated to the ability to out-shoot the opposition at even strengths when the score is tied or close. The correlation is not 100% (inherently, at his core, it does include a “margin of error) but micro stats remains a very useful decision helping tool.

the inherent problem, however, is that "scoring chance" in itself is a subjective observation, not a scientific fact...

was Pacioretty's goal vs the Canes a "scoring chance"? how could it not be since it ended up in a "score".

Gionta, like Gomez before him, gets credited with "scoring chances" for events that about as likely to result in a goal as Max's shot from half.

I do find it annoying when posters use micro stats without understanding its limitation; the op is a perfect example. On the other hand, if I learn that our Canadiens are not using them, I would be seriously annoyed!

agreed

 

[Kjell Dahlin]

understand that, but when you're creating stats to show the obvious (ex: outshooting opponent = more chances of success) you're not refining anything.


haha, make it X3!!!

Example: player x only has 8 points after 70 games. His ice time is limited and he is on the verge of falling through the cracks at the risk of flourishing elsewhere. The coach, while consulting a micro stats analysis report, realises, among other things, that when this player is on the ice, on even strength when the score is tied or close, his team is outshooting the opponent. He decides to give this guy a chance.

There is nothing “obvious” about the bolded part btw: you need to process a lot of infos to write the aforementioned report and, though you can observe the phenomenon (on even strength when the score is tied or close, his team is outshooting the opponent when player x is on the ice), chances are you won’t be able to record and process it all. The speed of the game, your own personal bias... a lot of things can get in the way of a proper evaluation.

Micro stats offer you a complementary... lighting. Key word: complementary.

EDIT:

the inherent problem, however, is that "scoring chance" in itself is a subjective observation, not a scientific fact...

was Pacioretty's goal vs the Canes a "scoring chance"? how could it not be since it ended up in a "score".

Gionta, like Gomez before him, gets credited with "scoring chances" for events that about as likely to result in a goal as Max's shot from half.



agreed

Though I studied and work with stats, I am no expert on micro stats applied to sports so the following should be taken with a grain of salt. I guess that “shots on goal” are strongly correlated to “scoring chances”. And shots on goal, contrary to “scoring chance”, “hits” or “takeover”, are an easy stats to track: a shot is a shot, interpretation does not play a role.