The Advanced Stats Thread Episode IX

[TheTakedown]

Pionk's underlying numbers are BADDDD but MSG is praising him because of the 28:12 that he played...

I'm not sure what to think. He has a lot of good moments but I also see a lot of gaffes that I'm not sure are fixable just yet...

This defense needs to more of ADA...

 

[Fvital92]

Pionk's underlying numbers are BADDDD but MSG is praising him because of the 28:12 that he played...

I'm not sure what to think. He has a lot of good moments but I also see a lot of gaffes that I'm not sure are fixable just yet...

This defense needs to more of ADA...

He's been playing with Staal, you have to take that in consideration.

 

[Irishguy42]

He's been playing with Staal, you have to take that in consideration.

Staal has better numbers without Pionk than Pionk does without Staal.

 

[Blue Blooded]

Staal has better numbers without Pionk than Pionk does without Staal.

Significantly so to boot.

 

[JoeThorntonsRooster]

I just came up with a very simple, idiot proof defense of Corsi’s effectiveness that I think is a great retort to people who downplay it. I felt the need to share my beautiful creation with all of you advanced stats nerds:

Corsi is the reason Anze Kopitar and Jonathan Toews have more Cups than Sidney Crosby and Alexander Ovechkin.

 

[SnowblindNYR]

I just came up with a very simple, idiot proof defense of Corsi’s effectiveness that I think is a great retort to people who downplay it. I felt the need to share my beautiful creation with all of you advanced stats nerds:

Corsi is the reason Anze Kopitar and Jonathan Toews have more Cups than Sidney Crosby and Alexander Ovechkin.

Sidney Crosby has as many cups as Toews and one more than Kopitar. And if you add the two groups up there's only a difference of 1 cup. Plus, it's a team sport. Not sure you can just pick one player from 4 different teams. I don't know for sure but Kane who was on the same teams as Toews doesn't have great Corsi. I may be wrong on that though.

 

[JoeThorntonsRooster]

Sidney Crosby has as many cups as Toews and one more than Kopitar. And if you add the two groups up there's only a difference of 1 cup. Plus, it's a team sport. Not sure you can just pick one player from 4 different teams. I don't know for sure but Kane who was on the same teams as Toews doesn't have great Corsi. I may be wrong on that though.

Perhaps the quote isn’t as idiot proof as I thought. (Not saying you are an idiot, but saying it is actually difficult to comprehend in general.)

Toews and Kopitar have 5 Cups, while Crosby and Ovechkin have 4. Crosby and Ovechkin are the two greatest players of our generation, but they have less Cups than these two average elite superstar #1Cs, because those average elite superstar #1Cs were on elite Corsi teams. You can win with mediocre or outright bad Corsi if you have guys like Ovechkin/Crosby playing out of their mind, but more teams have won with merely Toews/Kopitar and great Corsi.

 

[Blue Blooded]

What should be the Rangers' D when healthy:

Skjei-DeAngelo (+12.19 RelCF% in 66 min)
Staal-Shattenkirk (+5.67 in 126 min)
Claesson-Smith (+17.00 in 15 min, extra SSS warning)

What will be the Rangers D when healthy:

Skjei-McQuaid (-6.09 in 90 min)
Staal-Pionk (-7.63 in 382 min)
Smith-Shattenkirk (-0.82 in 140 min)

 

[silverfish]

 

[Blue Blooded]

What should be the Rangers' D when healthy:

Skjei-DeAngelo (+12.19 RelCF% in 66 min)
Staal-Shattenkirk (+5.67 in 126 min)
Claesson-Smith (+17.00 in 15 min, extra SSS warning)

What will be the Rangers D when healthy:

Skjei-McQuaid (-6.09 in 90 min)
Staal-Pionk (-7.63 in 382 min)
Smith-Shattenkirk (-0.82 in 140 min)

After tonight's game we have:
Skjei-DeAngelo (+12.87 RelCF% in 79 min)
Staal-Pionk (-7.84 in 397 min)
Smith-Shattenkirk (-1.48 in 154 min)

I'd switch Smith and Staal with our current injury status, even though Smith-Pionk have been terrible, since:

1. Staal-Shattenkirk have been surprisingly great as a pairing (see quoted post)
2. The sample size of Smith-Pionk being terrible is 8 minutes versus the 400 minute sample size of Staal-Pionk being terrible

 

[sbjnyc]

National Hockey League Announces Landmark Sports Betting Partnership With MGM Resorts

As part of the relationship, MGM Resorts will receive access to previously unseen enhanced NHL proprietary game data that will be generated by the League's state-of-the-art tracking systems currently under development. Access to this data will allow MGM Resorts to provide its customers with specialized NHL game insights, as well as unlocking new and innovative interactive fan engagement and betting opportunities for its U.S. customers wherever legally available.

As always, the path towards advanced metrics is through sports betting.

 

[Blue Blooded]

Chytil-Hayes-Zuccarello have been insanely good as a line, they've outattempted the opposition 39-19, outshot them 18-8, the scoring chances have been 24-6 and the high-danger scoring chances 14-3. Only 1-0 on the scoresheet, but these guys should be kept together as long as they pummel the opposition like they have.

Line Stats - Natural Stat Trick

 

[SA16]

I would bet you could find a lot of lines that have played 30 mins with great results that did not continue going forward.

 

[sbjnyc]

I would bet you could find a lot of lines that have played 30 mins with great results that did not continue going forward.

The NHL will take that bet! Just head to your nearest MGM casino...

 

[pblawr]

Hey guys,

I was wondering if any could anyone give an overview / refresher on why corsi is so highly valued. From my perspective, following hockey stats only loosely, it seems to me like their could be some holes / flaws with Corsi but I'm admittedly naive and curious what I could be missing.

My recollection is that Corsi "rose to popularity" ~3 years ago when they showed that the past ~10 cup winners all had very high team Corsis (with the Kings and Blackhawks comprising half of those cup wins) and has been the accepted hockey stats standard since then. Like I said, I haven't followed hockey stats closely so I could be missing a lot, but 10 seasons doesn't strike me as a sample size that should have a high degree of statistical significance (especially when just two teams drove a significant portion of the results) and my impression is that Corsi hasn't been as predictive in the few seasons since then. Are those valid concerns or is there compelling evidence that Corsi is predictive of a team's success in a statistically significant way (beyond just the correlation with team success over one 10 year period) that I'm not aware of? Broadly speaking, what is the large amount of weight that's placed on Corsi based on and what makes hockey stats people confident that it's a statistic with real predictive value / meaning?

 

[silverfish]

Shot attempts are a better predictor than goals scored for future team performance. Shot attempts also gave way to most expected goals models, which I do believe are even better than shot attempts in predicting future performance. This article seems to back that up: Expected Goals are a better predictor of future scoring than Corsi, Goals

One of the bigger mixups is that people think it is a descriptive metric, when really it's more of a predictive metric. Sure, you can get a grasp on how the game went by looking at shot attempts, but the big thing to not forget is that it gives you an idea of where the team is going.

For me, it always boils down to this. Let's say the Rangers win a game 3-2, but they were outshout 65-40 during the game. Which of those tell you more about the game? Which of those do you have more confidence in being repeated down the line? The event that happened 5 times, or the event that happened 105 times? And you're right, perhaps the shot attempts don't matter for that game, they won, right? But, you have to be wary about what'll happen down the line, which is the most major component of shot attempts.

There are countless other uses for it, as well (drivers vs passengers, with our without yous, etc...), but I'll let someone else come through and post some of their thoughts, too.

 

[pblawr]

Shot attempts are a better predictor than goals scored for future team performance. Shot attempts also gave way to most expected goals models, which I do believe are even better than shot attempts in predicting future performance. This article seems to back that up: Expected Goals are a better predictor of future scoring than Corsi, Goals

One of the bigger mixups is that people think it is a descriptive metric, when really it's more of a predictive metric. Sure, you can get a grasp on how the game went by looking at shot attempts, but the big thing to not forget is that it gives you an idea of where the team is going.

There are countless other uses for it, as well (drivers vs passengers, with our without yous, etc...), but I'll let someone else come through and post some of their thoughts, too.

For me, it always boils down to this. Let's say the Rangers win a game 3-2, but they were outshout 65-40 during the game. Which of those tell you more about the game? Which of those do you have more confidence in being repeated down the line? The event that happened 5 times, or the event that happened 105 times? And you're right, perhaps the shot attempts don't matter for that game, they won, right? But, you have to be wary about what'll happen down the line, which is the most major component of shot attempts.

Thanks for the quick reply. This is very helpful and exactly what I was looking for.

Would it strike you as accurate to say that, while they are better than the R^2's for GF, the R^2's for xG and CF are still relatively low (especially at the individual level) and indicate that the explanatory value of those stats is limited and not significant enough to draw a strong conclusion from by themselves, even if they are best stats that we have? In a nutshell, that Corsi and xG are the best stats we have but they still aren't good enough to rely on as a standalone indicator of player value / performance?

 

[Machinehead]

Shot attempts are a better predictor than goals scored for future team performance. Shot attempts also gave way to most expected goals models, which I do believe are even better than shot attempts in predicting future performance. This article seems to back that up: Expected Goals are a better predictor of future scoring than Corsi, Goals

One of the bigger mixups is that people think it is a descriptive metric, when really it's more of a predictive metric. Sure, you can get a grasp on how the game went by looking at shot attempts, but the big thing to not forget is that it gives you an idea of where the team is going.

For me, it always boils down to this. Let's say the Rangers win a game 3-2, but they were outshout 65-40 during the game. Which of those tell you more about the game? Which of those do you have more confidence in being repeated down the line? The event that happened 5 times, or the event that happened 105 times? And you're right, perhaps the shot attempts don't matter for that game, they won, right? But, you have to be wary about what'll happen down the line, which is the most major component of shot attempts.

There are countless other uses for it, as well (drivers vs passengers, with our without yous, etc...), but I'll let someone else come through and post some of their thoughts, too.

So basically what I learned from this post is: Chytil gonna score some goals.

 

[silverfish]

Thanks for the quick reply. This is very helpful and exactly what I was looking for.

Would it strike you as accurate to say that, while they are better than the R^2's for GF, the R^2's for xG and CF are still relatively low (especially at the individual level) and indicate that the explanatory value of those stats is limited and not significant enough to draw a strong conclusion from by themselves, even if they are best stats that we have? In a nutshell, that Corsi and xG are the best stats we have but they still aren't good enough to rely on as a standalone indicator of player value / performance?

I believe that the "detractors" are right when they say hockey is very random, and unpredictable. It is. This is a game that's predicated on valuing winners and losers by an event that happens, on average, what, like 5.5-6 times a game over 60 minutes? That's not something that's very conducive for good predictive analysis.

However, just because something doesn't explain everything doesn't mean it should be disregarded. It's the best thing we have until another best thing replaces it. What that best thing is? I'm really not sure. People are pretty pumped abotu player tracking. Will that data be public? Maybe eventually. It's something that could really help an xG model. Where the players were when the shot was taken, screen or no screen, player movement prior to the shot, how many passes led to the shot, so many different variables that could be helpful there.

 

[pblawr]

I believe that the "detractors" are right when they say hockey is very random, and unpredictable. It is. This is a game that's predicated on valuing winners and losers by an event that happens, on average, what, like 5.5-6 times a game over 60 minutes? That's not something that's very conducive for good predictive analysis.

However, just because something doesn't explain everything doesn't mean it should be disregarded. It's the best thing we have until another best thing replaces it.

Yeah, that makes sense. It seems like if you are evaluating a team or individual player, Corsi or xG is a better metric to base their performance on than GF. At the same time, neither Corsi or xG are so effective that could really confidently conclude that a player or team is good or bad based on that metric alone. It seems to me like it would be most accurate to say that a high or low Corsi or xG would moderately increase the odds that the player is good or bad. Does that seem reasonable / accurate to you?

 

[silverfish]

Yeah, that makes sense. It seems like if you are evaluating a team or individual player, Corsi or xG is a better metric to base their performance on than GF. At the same time, neither Corsi or xG are so effective that could really confidently conclude that a player or team is good or bad based on that metric alone. It seems to me like it would be most accurate to say that a high or low Corsi or xG would moderately increase the odds that the player is good or bad. Does that seem reasonable / accurate to you?

I'd say it depends on how you feel about the data in general.

The good thing about numbers is that they don't lie, and they have no bias. Kevin Shattenkirk this season, via Corsica.Hockey, when he's on the ice, the Rangers allow 6.38 less shot attempts against per 60, and take 8.39 more shot attempts per 60 than they do when Kevin Shattenkirk is not on the ice. This is an indisputable fact.

What is disputable here, is if you take these numbers and say: "obviously this shows that so far this year 5v5, Shattenkirk is not a detriment to the team on defense, in fact he is suppressing shot attempts against and driving shot attempts for". Now, I wouldn't dispute that analysis, because I value the numbers. However, it is totally fair if someone who does not value these numbers in terms of telling a story about on-ice play to believe that Shattenkirk is a liability defensively. Everyone is free to use their own analysis inputs to get to their output statement.

People are going to say that the shot attempt numbers don't tell the whole story. And they're right. They definitely do not. But it goes back to my previous point where just because something doesn't do everything doesn't mean that it's useless. I'd argue that shot attempts, along with other contextual data, tell a very large portion of the story, but, we still haven't moved everyone off +/-, so, I'm not holding my breath there

 

[pblawr]

I'd say it depends on how you feel about the data in general.

The good thing about numbers is that they don't lie, and they have no bias. Kevin Shattenkirk this season, via Corsica.Hockey, when he's on the ice, the Rangers allow 6.38 less shot attempts against per 60, and take 8.39 more shot attempts per 60 than they do when Kevin Shattenkirk is not on the ice. This is an indisputable fact.

What is disputable here, is if you take these numbers and say: "obviously this shows that so far this year 5v5, Shattenkirk is not a detriment to the team on defense, in fact he is suppressing shot attempts against and driving shot attempts for". Now, I wouldn't dispute that analysis, because I value the numbers. However, it is totally fair if someone who does not value these numbers in terms of telling a story about on-ice play to believe that Shattenkirk is a liability defensively. Everyone is free to use their own analysis inputs to get to their output statement.

People are going to say that the shot attempt numbers don't tell the whole story. And they're right. They definitely do not. But it goes back to my previous point where just because something doesn't do everything doesn't mean that it's useless. I'd argue that shot attempts, along with other contextual data, tell a very large portion of the story, but, we still haven't moved everyone off +/-, so, I'm not holding my breath there

Yeah, that makes sense. It seems like a safe bet that the stats are generally less flawed than posters' eye tests.

Still the context around things like deployment, finishing ability, quality of and fit with teammates, quality of competition, and shot quality is important, difficult to quantify accurately, and leaves a meaningful gap in the ability of those stats to accurately assess player value.

It seems to me like it would probably be a safe bet to say that a player with a significantly better or worse corsi / xGf is better or worse than another player, as long as there aren't significant differences in context. But if the context for two players is materially different, it's just tough to tell, and if you are comparing a player with a 49% Corsi to a player with a 52% Corsi, it might not be that meaningful.

 

[Levitate]

I'd say it depends on how you feel about the data in general.

The good thing about numbers is that they don't lie, and they have no bias. Kevin Shattenkirk this season, via Corsica.Hockey, when he's on the ice, the Rangers allow 6.38 less shot attempts against per 60, and take 8.39 more shot attempts per 60 than they do when Kevin Shattenkirk is not on the ice. This is an indisputable fact.

What is disputable here, is if you take these numbers and say: "obviously this shows that so far this year 5v5, Shattenkirk is not a detriment to the team on defense, in fact he is suppressing shot attempts against and driving shot attempts for". Now, I wouldn't dispute that analysis, because I value the numbers. However, it is totally fair if someone who does not value these numbers in terms of telling a story about on-ice play to believe that Shattenkirk is a liability defensively. Everyone is free to use their own analysis inputs to get to their output statement.

People are going to say that the shot attempt numbers don't tell the whole story. And they're right. They definitely do not. But it goes back to my previous point where just because something doesn't do everything doesn't mean that it's useless. I'd argue that shot attempts, along with other contextual data, tell a very large portion of the story, but, we still haven't moved everyone off +/-, so, I'm not holding my breath there

Honestly I think a lot of people just still haven't gotten around to the idea that it's far better to be on the offense more than it is to be playing defense and that playing actual effective defense means that you get the puck back and go back on offense as quickly as possible. To many people good defense is still all about how you play without the puck and getting in shooting lanes and blocking shots and covering your man...which are certainly all parts of it, but I feel like there's this disconnect especially when it comes to defensemen about defense and where it goes from there.

In my mind the ideal defensive play is to not allow the other team to possess the puck in your end and be able to transition that quickly to offense for your own team. You take the puck away, you start the transition, you go on the offense. Perfect defense. This is not something Neil Pionk does well right now, but since to a casual eye test he seems to not make major blunders out here on the ice, people think he's a good defender. Yet he spends all his time defending and struggles to transition the puck well but people are unwilling to see that as "bad defense" so far.

I dunno, now I"m rambling, I feel like using stats to get a clearer view of the impact players are having should be a no brainer but instead some people seem to think you're saying they know nothing about hockey and don't know how to watch a game.

 

[silverfish]

Honestly I think a lot of people just still haven't gotten around to the idea that it's far better to be on the offense more than it is to be playing defense and that playing actual effective defense means that you get the puck back and go back on offense as quickly as possible. To many people good defense is still all about how you play without the puck and getting in shooting lanes and blocking shots and covering your man...which are certainly all parts of it, but I feel like there's this disconnect especially when it comes to defensemen about defense and where it goes from there.

In my mind the ideal defensive play is to not allow the other team to possess the puck in your end and be able to transition that quickly to offense for your own team. You take the puck away, you start the transition, you go on the offense. Perfect defense. This is not something Neil Pionk does well right now, but since to a casual eye test he seems to not make major blunders out here on the ice, people think he's a good defender. Yet he spends all his time defending and struggles to transition the puck well but people are unwilling to see that as "bad defense" so far.

I dunno, now I"m rambling, I feel like using stats to get a clearer view of the impact players are having should be a no brainer but instead some people seem to think you're saying they know nothing about hockey and don't know how to watch a game.

I don't think it's a coincidence that the guys who play a lot, and the guys who are in the defensive zone a lot, are often referred to as 'good shutdown d-men'.

 

[DanielBrassard]

Honestly I think a lot of people just still haven't gotten around to the idea that it's far better to be on the offense more than it is to be playing defense and that playing actual effective defense means that you get the puck back and go back on offense as quickly as possible. To many people good defense is still all about how you play without the puck and getting in shooting lanes and blocking shots and covering your man...which are certainly all parts of it, but I feel like there's this disconnect especially when it comes to defensemen about defense and where it goes from there.

In my mind the ideal defensive play is to not allow the other team to possess the puck in your end and be able to transition that quickly to offense for your own team. You take the puck away, you start the transition, you go on the offense. Perfect defense. This is not something Neil Pionk does well right now, but since to a casual eye test he seems to not make major blunders out here on the ice, people think he's a good defender. Yet he spends all his time defending and struggles to transition the puck well but people are unwilling to see that as "bad defense" so far.

I dunno, now I"m rambling, I feel like using stats to get a clearer view of the impact players are having should be a no brainer but instead some people seem to think you're saying they know nothing about hockey and don't know how to watch a game.

I think this is very true. That's not to say that certain players aren't better at the traditional "defensive" attributes people are familiar with, but if you are really good at those things, but a detriment everywhere else, it's not that valuable of a skillset.