Corsi, shot quality, and the Toronto Maple Leafs

[Bear of Bad News]

This is all true...

Just because good teams tend to have good corsi does not mean that teams are good because they have good corsi. And even if it did, that does not mean that having a good team without good corsi is impossible.

However, it appears unrelated to this...

Lots of statistics majors in here improperly using statistics.

...since I don't see "lots" of people (let alone lots of statistics majors) claiming what you are arguing against (for what it's worth, my doctorate is in mathematics, not statistics).

The advantage that CORSI has over goal totals is sample size.

The disadvantage that CORSI has over goal totals is an assumed long-term evening out of shooting percentage.

For what it's worth, CORSI wasn't initially developed for the purpose that some are now using it for. It was designed as a proxy for goaltender workload.

 

[Master_Of_Districts]

A single statistic should never be used to make projections, let alone one with such glaring problems and omissions.

And yet, that same single statistic, with all of its glaring problems and omissions, predicts future results more accurately than winning percentage or goal differential.

Perhaps it's not that useless.

 

[Delicious Dangles*]

...since I don't see "lots" of people (let alone lots of statistics majors) claiming what you are arguing against (for what it's worth, my doctorate is in mathematics, not statistics).

The advantage that CORSI has over goal totals is sample size.

The disadvantage that CORSI has over goal totals is an assumed long-term evening out of shooting percentage.

For what it's worth, CORSI wasn't initially developed for the purpose that some are now using it for. It was designed as a proxy for goaltender workload.

I actually wasn't talking about you, I don't think at least lol. But if sample size is the best reason for a statistic, there's probably a problem with it.

Last line is an interesting fact I wish more people knew.

And yet, that same single statistic, with all of its glaring problems and omissions, predicts future results more accurately than winning percentage or goal differential.

Perhaps it's not that useless.

Or maybe it seems like it works better because the majority of the league is made up of teams that have a very similar distribution of talent and/or deploy a system that falls in line with the variables and way that corsi measures things.

For those teams, corsi probably has better accuracy. For those teams that don't follow "the rules of corsi", it would have very little predictive ability.

 

[Master_Of_Districts]

I actually wasn't talking about you, I don't think at least lol. But if sample size is the best reason for a statistic, there's probably a problem with it.

Last line is an interesting fact I wish more people knew.


Or maybe it seems like it works better because the majority of the league is made up of teams that have a very similar distribution of talent and/or deploy a system that falls in line with the variables and way that corsi measures things.

For those teams, corsi probably has better accuracy. For those teams that don't follow "the rules of corsi", it would have very little predictive ability.

I'm not at all convinced that the Leafs are unusual or anomalous to point where established statistical principles wouldn't apply them.

Once again - we're not talking about the 1980s Edmonton Oilers here.

And it's not like we haven't teams with similar profiles most lockout either.

The 2009-10 Capitals, 2008-09 Penguins, 2006-07 Sabres and 2006-07 Predators all posted great numbers in terms of EV shooting percentage.

And when they did, their own fan bases raved about things like shot quality, finishing talent and unique offensive strategies. And maybe those factors had some explanatory role too.

But in a league where 67% of the seasonal variation in EV shooting percentage is explained by luck alone, good fortune was necessarily the largest factor of them all. So not surprisingly, all four teams regressed heavily the next season. And no - personnel change was not the primary cause of that.

 

[Delicious Dangles*]

I'm not at all convinced that the Leafs are unusual or anomalous to point where established statistical principles wouldn't apply them.

Once again - we're not talking about the 1980s Edmonton Oilers here.

And it's not like we haven't teams with similar profiles most lockout either.

The 2009-10 Capitals, 2008-09 Penguins, 2006-07 Sabres and 2006-07 Predators all posted great numbers in terms of EV shooting percentage.

And when they did, their own fan bases raved about things like shot quality, finishing talent and unique offensive strategies. And maybe those factors had some explanatory role too.

But in a league where 67% of the seasonal variation in EV shooting percentage is explained by luck alone, good fortune was necessarily the largest factor of them all. So not surprisingly, all four teams regressed heavily the next season. And no - personnel change was not the primary cause of that.

2006-2007 Sabres = Lost Drury and Briere (#1 and #3 scorers) that off-season, and was probably the last year that afinogenov was any good. Missed the playoffs by 4 points the next year.

2009-2010 Capitals = Lost Morrison in the offseason, and injuries destroyed the next year's team (Fleischmann, Green, Fehr who contributed offensively the year before). Plus, they finished *1st* in the conference the next year. Regress, yeah right. Also huge difference from the Leafs in the make-up of the scoring.

2006-2007 Predators = Lost Kariya (#1 scoring), Sullivan (#4), Forsberg and Hartnell in the off-season. Plus Weber went down with an injury the next season. And still made the playoffs.

2008-2009 Penguins = Lost Sykora and Satan in the off-season, and was the last year all of Letang, Malkin and Crosby were all healthy. Did BETTER the next year. Once again, huge difference from the Leafs in the make-up of the scoring.

So all went through significant line-up changes (some the most significant in recent history), and/or injuries, for the next season. One team improved, One team regressed but still won their conference, and the two teams that lost the most personell, fell the hardest, with one team still making the playoffs the next year, and one missing by 4 points. And none of them played a system like the Leafs do now that is so opposite of corsi.

Looks like you did too much calculating, and not enough research.

 

[Master_Of_Districts]

2006-2007 Sabres = Lost Drury and Briere (#1 and #3 scorers) that off-season, and was probably the last year that afinogenov was any good. Missed the playoffs by 4 points the next year.

2009-2010 Capitals = Lost Morrison in the offseason, and injuries destroyed the next year's team (Fleischmann, Green, Fehr who contributed offensively the year before). Plus, they finished *1st* in the conference the next year. Regress, yeah right. Also huge difference from the Leafs in the make-up of the scoring.

2006-2007 Predators = Lost Kariya (#1 scoring), Sullivan (#4), Forsberg and Hartnell in the off-season. Plus Weber went down with an injury the next season. And still made the playoffs.

2008-2009 Penguins = Lost Sykora and Satan in the off-season, and was the last year all of Letang, Malkin and Crosby were all healthy. Did BETTER the next year. Once again, huge difference from the Leafs in the make-up of the scoring.

So all went through significant line-up changes (some the most significant in recent history), and/or injuries, for the next season. One team improved, One team regressed but still won their conference, and the two teams that lost the most personell, fell the hardest, with one team still making the playoffs the next year, and one missing by 4 points. And none of them played a system like the Leafs do now that is so opposite of corsi.

Looks like you did too much calculating, and not enough research.

Obviously the regression comment pertained to EV shooting percentage specifically and not overall performance.

And in that regard:

0910 WSH: 10.4 EV SH%
1011 WSH: 7.6 EV SH%

0809 PIT: 9.2 EV SH%
0910 PIT: 8.6 EV SH%

0607 BUF: 11.4 EV SH%
0708 BUF: 9.0 EV SH%

0607 NSH: 10.5 EV SH%
0708 NSH: 8.4 EV SH%

Looks like you did too much typing and not enough reading comprehension.

Better luck next time!

 

[Master_Of_Districts]

And on the issue of whether the regression was due to personnel change or simple statistical regression to the mean, those teams regressed 83% percent of the way to the mean in the following season in terms of EV shooting percentage.

If statistical regression was the only factor, we would have expected to see a regression of 67% of the way to the mean.

So they collectively regressed only slightly further than what would have been expected if statistical regression to the mean was the only factor.

So, just like I said, personnel change, in all likelihood, was not the primary causal factor - not by a long shot.

 

[TieClark]

TOR had an average powerplay (5-on-4) last year with respect to scoring rate.

They were one of the worst teams with respect to shot generation, which is the best predictor of future success on the powerplay.

(Just like at even strength! It's amazing the way the world works sometimes).

This year, their shot generation has improved somewhat, but it's still early, and they're only middling in this regard in any event.

Their success has been buoyed by a high shooting percentage. Of course, as smart people have known for a while, the correlation between a team's shooting percentage on the powerplay over a ten game sample and its underlying shooting talent is virtually zero.

So that doesn't really bode well either.

As for the penalty kill, the results were great last year, both in terms of shots against and save percentage. Full marks.

But this year they've just been getting flat out bombed in terms of shots against - 5th worst in the league, in fact.

Given how good they were at shot prevention last year, I suspect that they'll improve in this regard.

And they'll have to - because the logician in me tells me that their goalies may not continue to save over 94% of the shots they face on the PK.

Unless, of course, they're both four standard deviations more talented than Dominik Hasek in his prime. (Not likely!).

Anyway, all of that's to say that your characterization of the Leafs' special teams as "great" doesn't really hold any water. At all.

Do you honestly believe though that it's coincidence the Leafs have the EXACT same tendencies while shorthanded or on the power play than at 5 on 5? Are they really getting THAT lucky? I would think a statistics person would have to agree the statistical likelyhood of them being that lucky are pretty terrible. What is more likely, is the way the team plays and the players they have go against the sh% statistical analysis which isn't exactly fool proof.

 

[Delicious Dangles*]

Obviously the regression comment pertained to EV shooting percentage specifically and not overall performance.

And in that regard:

0910 WSH: 10.4 EV SH%
1011 WSH: 7.6 EV SH%

0809 PIT: 9.2 EV SH%
0910 PIT: 8.6 EV SH%

0607 BUF: 11.4 EV SH%
0708 BUF: 9.0 EV SH%

0607 NSH: 10.5 EV SH%
0708 NSH: 8.4 EV SH%

Looks like you did too much typing and not enough reading comprehension.

Better luck next time!

That was not obvious, but which brings up the obvious question, who the hell cares if their shooting percentage went down if they performed as expected. All 4 teams performed as expected, based on their previous year and personnel changes. If anything, better than expected.

So yes, Leaf players may see a regression in their shooting percentage if they start shooting more, but nothing says this correlates to a regression of the TEAM. Leafs lost none of their key players, and added some new ones.

You basically just said that shooting percentage has no correlation to how the team does.

And on the issue of whether the regression was due to personnel change or simple statistical regression to the mean, those teams regressed 83% percent of the way to the mean in the following season in terms of EV shooting percentage.

If statistical regression was the only factor, we would have expected to see a regression of 67% of the way to the mean.

So they collectively regressed only slightly further than what would have been expected if statistical regression to the mean was the only factor.

So, just like I said, personnel change, in all likelihood, was not the primary causal factor - not by a long shot.

Except you cannot say that they would have regressed 67% to the mean with no changes.

You're basically saying that you expected them to regress X amount, and they regressed slightly more than X amount; Y amount. So the X portion of Y HAS to be because of your prediction of natural regression, and not because of the team, it's players, the personnel changes, the system changes, the changes of OTHER teams (people always conveniently forget this), the way the individual games played out, or where/how opportunities were provided.

Know why teams regress after these types of seasons? Because in a cap world, it is very hard to keep a team that dominant offensively together, because it means they get the big bucks. Especially when it isn't a cap team, or a favourable destination, or you don't have 3 potential PPG forwards on sweetheart deals like the Leafs.

Funny how the team that saw minimal regression was the team that lost the least players that off-season, and have the two best players in the game. It's almost like quality of players DOES matter.

 

[Master_Of_Districts]

That was not obvious, but which brings up the obvious question, who the hell cares if their shooting percentage went down if they performed as expected. All 4 teams performed as expected, based on their previous year and personnel changes. If anything, better than expected.

So yes, Leaf players may see a regression in their shooting percentage if they start shooting more, but nothing says this correlates to a regression of the TEAM. Leafs lost none of their key players, and added some new ones.

Except all of those teams regressed in an overall sense as well.

That fact was just irrelevant to my point.

But to address that specific claim:

0910 WSH: Goal differential: + 86
1011 WSH: Goal differential: + 28

0809 PIT: Goal differential: + 25
0910 PIT: Goal differential: + 14

0607 BUF: Goal differential: + 60
0708 BUF: Goal differential: + 18

0607 NSH: Goal differential: + 59
0708 NSH: Goal differential: + 3

They collectively went from 57.5 to 15.75 in terms of goal differential. A regression of 72% to the league average.

So to claim that the teams in question performed "better than expected" the following year is:

a) Wrong; and
b) Intellectually lazy (if you made the claim without checking); or
c) Intellectually dishonest (if you checked but chose to make the assertion regardless)

You basically just said that shooting percentage has no correlation to how the team does.

And you basically just put words in my mouth.

Except you cannot say that they would have regressed 67% to the mean with no changes.

Certainly, with only four teams, the regression wouldn't, in all likelihood, have been precisely to 67%.

But that is the expected result. The one most likely to occur, probabilistically.

You're basically saying that you expected them to regress X amount, and they regressed slightly more than X amount; Y amount. So the X portion of Y HAS to be because of your prediction of natural regression, and not because of the team, it's players, the personnel changes, the system changes, the changes of OTHER teams (people always conveniently forget this), the way the individual games played out, or where/how opportunities were provided.

I suppose you're right in a technical sense. A team's true talent shooting percentage can only be estimated, rather than known with certainty. So, taking that into consideration, my apportionment was inappropriate.

Still though - I'm not sure it really detracts from my overall point. Statistical principles predicted that a group of teams ought to have regressed 67% of the way to the mean with respect to EV shooting percentage. They actually regressed 83% of the way to the mean. I don't believe that's a point in favour of the "EV shooting percentage is sustainable" crowd.

Know why teams regress after these types of seasons?

Yes.

[See above].

Funny how the team that saw minimal regression was the team that lost the least players that off-season, and have the two best players in the game. It's almost like quality of players DOES matter.

It's interesting that you would characterize a decrease in goal differential from +86 to +28 as "minimal regression."

 

[achtungbaby]

I'm really starting to see a trend where people are starting to complain even after their team wins a game because the corsi stats don't make it seem like a "pretty" win or a "well deserved" win. It's total B.S. I'm not going to let someone on here rain on my mood when the Leafs win. Bring up Corsi all you want you'll just get ignored.

Perhaps this is the wrong thread for you then? Corsi and shot quality are right in the title man. If you want to celebrate, go to the Leafs forum.

 

[Delicious Dangles*]

Except all of those teams regressed in an overall sense as well.

That fact was just irrelevant to my point.

But to address that specific claim:

0910 WSH: Goal differential: + 86
1011 WSH: Goal differential: + 28

0809 PIT: Goal differential: + 25
0910 PIT: Goal differential: + 14

0607 BUF: Goal differential: + 60
0708 BUF: Goal differential: + 18

0607 NSH: Goal differential: + 59
0708 NSH: Goal differential: + 3

They collectively went from 57.5 to 15.75 in terms of goal differential. A regression of 72% to the league average.

So to claim that the teams in question performed "better than expected" the following year is:

a) Wrong; and
b) Intellectually lazy (if you made the claim without checking); or
c) Intellectually dishonest (if you checked but chose to make the assertion regardless)

OF COURSE their goal differentials went down! Not only is goal-scoring decreasing as time passes, and powerplay opportunities going down (surprise, surprise, 2 of the teams are from 2006-2007), but these teams went through MAJOR changes to their offensive cores, that you keep ignoring to fit your narrative.

Let's actually ANALYZE things again, instead of just blindly going off of numbers.

Buffalo lost their #1 scorer (32 goals, 95 points), their #3 scorer (37 goals, 69 points), and essentially lost their #6 scorer (23 goals, 61 points).

Nashville lost their #1 scorer (24 goals, 76 points), their #4 scorer (22 goals, 60 points), their #7 scorer (22 goals, 39 points), and their #10 scorer (15 points in 17 games). They also lost Weber for 1/3 of the season the next year, hurting production.

Washington lost their #8 scorer (12 goals, 42 points), though the most important change was their injuries the next year. Fleischmann for 59 games, Fehr for 30, Mike Green for 33 (and was not the same even when playing), Semin for 17, etc. 2009-2010 was also the last year Ovechkin was his league-crushing goal-scoring self. Having three 100+ forwards and a PPG defenseman every year is somewhat impossible, however it is not like it was only for the one year, so not exactly luck. Either way, this is not a good team to model the Leafs against, as they are built entirely different.

Similarly, the penguins are also built very differently, but let us look at them, just for kicks. Lost their #4 scorer (25 goals, 46 points), their #6 scorer (17 goals, 36 points), and dealt with some injuries the next season.

Now since we are measuring correlation to success and not goal-differential (funny, goal differential is always brought up as a terrible way to judge success by your group), let us look at how, you know, they actually did.

Judging by our basic analysis, we would assume that scoring would go down for each of these teams, to varying degrees. We would expect Pittsburgh to have been affected the least, as they lost less substantial pieces and their great scoring in 08-09 was mostly a result of Malkin and Crosby. As we look at the results, we see that Pittsburgh actually defied the odds, and actually had 2 more points than the previous season.

We would expect Washington to be affected next-least, as they endured bad injuries and their superstars all had bad years, however, they lost the least during the off-season. As we look at the results, we see that Washington did drop off by 14 points, but still finished 1st in their conference.

We would expect Nashville to be affected 2nd-most, as they lost considerable offensive firepower in the off-season, and endured an injury to a main guy in the following year. As we look at the results, we see that Nashville dropped off by 19 points, consistent with our predictions.

Lastly, we would expect Buffalo to be the worst off, as they lost even more offensive firepower, including 2 of their 3 top offensive drivers in one off-season, in what may be the worst off-season in Buffalo Sabre history. Once again, it appears expectations are met by reality, and Buffalo dropped the most with 23 points.

Now if you had read properly, you would notice I said performed as or better than *expected*. I did not say performed better than the previous year (though one team actually did).

So to claim that the results support any theory other than the fact that these teams lost considerable offensive firepower either in the off-season or during the year, and thus ended up with a worse offensive year, is:

a) Wrong; and
b) Intellectually lazy (if you made the claim without checking); or
c) Intellectually dishonest (if you checked but chose to make the assertion regardless)

And you basically just put words in my mouth.

No, if you look again, that it precisely what you are implying.

Certainly, with only four teams, the regression wouldn't, in all likelihood, have been precisely to 67%.

But that is the expected result. The one most likely to occur, probabilistically.

Based off of what, exactly? How did you come to this number, and how can you use this number for what you have already described as irregular teams?

You also brought up a good point. This is 4 teams. How is 4 teams from a sample of at least 240 teams (if only going back to the lockout, which is skewing the numbers enough as it is - only the cap world is represented... hmm...), representative of a trend or whatever you are trying to prove?

It is a serious sample size issue, even aside from how wrong you were about those 4.

I suppose you're right in a technical sense. A team's true talent shooting percentage can only be estimated, rather than known with certainty. So, taking that into consideration, my apportionment was inappropriate.

Still though - I'm not sure it really detracts from my overall point. Statistical principles predicted that a group of teams ought to have regressed 67% of the way to the mean with respect to EV shooting percentage. They actually regressed 83% of the way to the mean. I don't believe that's a point in favour of the "EV shooting percentage is sustainable" crowd.

You are ignoring the underlying reasons for the regression of offensive statistics, which are clear as day, and using numbers that you yourself have admitted are inaccurate.

It is like you have 5 apples. You tell a bunch of people that these apples are going to decay, and by tomorrow, there will only be 2 apples. You sneak back at night, and eat 3 of the apples. The next day, you proclaim to the world, LOOK AT MY BRILLIANCE! FOR I PREDICTED THAT THE APPLES WOULD REGRESS, AND LOOK! ONLY 2 APPLES!

In the numbers, there were 5 apples yesterday, and today, there are 2 apples. You said there should be 2 apples, there are 2 apples. To an outsider, only looking at the numbers, they may think you were right; the apples decayed. But we all know what really happened, and it had nothing to do with the decay of apples.

The apples were taken away. Just as the offensive contributions of key players were taken away for these teams.

It's interesting that you would characterize a decrease in goal differential from +86 to +28 as "minimal regression."

It is interesting that you consider getting 2 more points than last year, substantial regression.

 

[Master_Of_Districts]

OF COURSE their goal differentials went down! Not only is goal-scoring decreasing as time passes, and powerplay opportunities going down (surprise, surprise, 2 of the teams are from 2006-2007), but these teams went through MAJOR changes to their offensive cores, that you keep ignoring to fit your narrative.

Let's actually ANALYZE things again, instead of just blindly going off of numbers.

Buffalo lost their #1 scorer (32 goals, 95 points), their #3 scorer (37 goals, 69 points), and essentially lost their #6 scorer (23 goals, 61 points).

Nashville lost their #1 scorer (24 goals, 76 points), their #4 scorer (22 goals, 60 points), their #7 scorer (22 goals, 39 points), and their #10 scorer (15 points in 17 games). They also lost Weber for 1/3 of the season the next year, hurting production.

Washington lost their #8 scorer (12 goals, 42 points), though the most important change was their injuries the next year. Fleischmann for 59 games, Fehr for 30, Mike Green for 33 (and was not the same even when playing), Semin for 17, etc. 2009-2010 was also the last year Ovechkin was his league-crushing goal-scoring self. Having three 100+ forwards and a PPG defenseman every year is somewhat impossible, however it is not like it was only for the one year, so not exactly luck. Either way, this is not a good team to model the Leafs against, as they are built entirely different.

Similarly, the penguins are also built very differently, but let us look at them, just for kicks. Lost their #4 scorer (25 goals, 46 points), their #6 scorer (17 goals, 36 points), and dealt with some injuries the next season.

Now since we are measuring correlation to success and not goal-differential (funny, goal differential is always brought up as a terrible way to judge success by your group), let us look at how, you know, they actually did.

Judging by our basic analysis, we would assume that scoring would go down for each of these teams, to varying degrees. We would expect Pittsburgh to have been affected the least, as they lost less substantial pieces and their great scoring in 08-09 was mostly a result of Malkin and Crosby. As we look at the results, we see that Pittsburgh actually defied the odds, and actually had 2 more points than the previous season.

We would expect Washington to be affected next-least, as they endured bad injuries and their superstars all had bad years, however, they lost the least during the off-season. As we look at the results, we see that Washington did drop off by 14 points, but still finished 1st in their conference.

We would expect Nashville to be affected 2nd-most, as they lost considerable offensive firepower in the off-season, and endured an injury to a main guy in the following year. As we look at the results, we see that Nashville dropped off by 19 points, consistent with our predictions.

Lastly, we would expect Buffalo to be the worst off, as they lost even more offensive firepower, including 2 of their 3 top offensive drivers in one off-season, in what may be the worst off-season in Buffalo Sabre history. Once again, it appears expectations are met by reality, and Buffalo dropped the most with 23 points.

Now if you had read properly, you would notice I said performed as or better than *expected*. I did not say performed better than the previous year (though one team actually did).

So to claim that the results support any theory other than the fact that these teams lost considerable offensive firepower either in the off-season or during the year, and thus ended up with a worse offensive year, is:

a) Wrong; and
b) Intellectually lazy (if you made the claim without checking); or
c) Intellectually dishonest (if you checked but chose to make the assertion regardless)


No, if you look again, that it precisely what you are implying.


Based off of what, exactly? How did you come to this number, and how can you use this number for what you have already described as irregular teams?

You also brought up a good point. This is 4 teams. How is 4 teams from a sample of at least 240 teams (if only going back to the lockout, which is skewing the numbers enough as it is - only the cap world is represented... hmm...), representative of a trend or whatever you are trying to prove?

It is a serious sample size issue, even aside from how wrong you were about those 4.


You are ignoring the underlying reasons for the regression of offensive statistics, which are clear as day, and using numbers that you yourself have admitted are inaccurate.

It is like you have 5 apples. You tell a bunch of people that these apples are going to decay, and by tomorrow, there will only be 2 apples. You sneak back at night, and eat 3 of the apples. The next day, you proclaim to the world, LOOK AT MY BRILLIANCE! FOR I PREDICTED THAT THE APPLES WOULD REGRESS, AND LOOK! ONLY 2 APPLES!

In the numbers, there were 5 apples yesterday, and today, there are 2 apples. You said there should be 2 apples, there are 2 apples. To an outsider, only looking at the numbers, they may think you were right; the apples decayed. But we all know what really happened, and it had nothing to do with the decay of apples.

The apples were taken away. Just as the offensive contributions of key players were taken away for these teams.


It is interesting that you consider getting 2 more points than last year, substantial regression.

Wow. Where to start?

Firstly, to assert that the regressions are attributable to a decrease in league scoring is flat out laughable.

No one who makes such an assertion should be taken seriously.

I can use goal ratio instead to show you that you're wrong. And I'll do just that, once I get home.

Anyway - for whatever reason - you remain convinced that personnel changes can account for the observed regression.

The problem with that, however, is that if we look at data at the within season level, thereby controlling for offseason roster turnover, we observe the exact same thing - teams with high EV shooting percentages regressing heavily to the league average.

Once again, when I get home I'll be in a position to post the data.

Regarding the 67 percent figure - that was the average percentage of variation attributable to luck for team EV shooting percentage after analyzing each season from 2003-04 to 2010 - 2011. For each season, the percentage was determined on the basis that the amount of observed variance is equal to the amount of variance attributable to skill added to the amount of variance attributable to randomness.

With respect, the apple analogy is stupid. And nonsensical.

And finally, when did I ever characterize the Penguins' regression as substantial? I was clearly looking at those four teams in aggregate.

 

[Delicious Dangles*]

Wow. Where to start?

Firstly, to assert that the regressions are attributable to a decrease in league scoring is flat out laughable.

No one who makes such an assertion should be taken seriously.

I can use goal ratio instead to show you that you're wrong. And I'll do just that, once I get home.

Anyway - for whatever reason - you remain convinced that personnel changes can account for the observed regression.

The problem with that, however, is that if we look at data at the within season level, thereby controlling for offseason roster turnover, we observe the exact same thing - teams with high EV shooting percentages regressing heavily to the league average.

Once again, when I get home I'll be in a position to post the data.

Regarding the 67 percent figure - that was the average percentage of variation attributable to luck for team EV shooting percentage after analyzing each season from 2003-04 to 2010 - 2011. For each season, the percentage was determined on the basis that the amount of observed variance is equal to the amount of variance attributable to skill added to the amount of variance attributable to randomness.

With respect, the apple analogy is stupid. And nonsensical.

And finally, when did I ever characterize the Penguins' regression as substantial? I was clearly looking at those four teams in aggregate.

So basically writing off everything I said with nothing to actually support your side. Cool.

How can you use in-season data? Not only are you shrinking an already tiny sample size, causing the data to be heavily skewed by injuries and streaks, but it also ignores how systems are enforced at different times of the season.

If the Penguins "regression" is not substantial, why are you claiming that to be evidence to support you? So now we're down to 3 teams, an even smaller sample size. And all 3 still point to regression caused for very logical and obvious reasons like injuries and drastic personnel changes, and not "just cause the numbers told me".

Statistics is not just getting a result, it is analyzing WHY you get that result. It is a key part you keep missing.

These numbers you keep spouting, like 67% are an average, a culmination of a multitude of factors. But in a dynamic construct like hockey, if the factors change, the numbers will change. And it's not just because it is "destined" to happen. It is not "destined" to regress to a mean on this small of a scale. There are REASONS the changes come about, and those need to be fully reflected in any prediction being made.

 

[Master_Of_Districts]

So basically writing off everything I said with nothing to actually support your side. Cool.

How can you use in-season data? Not only are you shrinking an already tiny sample size, causing the data to be heavily skewed by injuries and streaks, but it also ignores how systems are enforced at different times of the season.

If the Penguins "regression" is not substantial, why are you claiming that to be evidence to support you? So now we're down to 3 teams, an even smaller sample size. And all 3 still point to regression caused for very logical and obvious reasons like injuries and drastic personnel changes, and not "just cause the numbers told me".

Statistics is not just getting a result, it is analyzing WHY you get that result. It is a key part you keep missing.

These numbers you keep spouting, like 67% are an average, a culmination of a multitude of factors. But in a dynamic construct like hockey, if the factors change, the numbers will change. And it's not just because it is "destined" to happen. It is not "destined" to regress to a mean on this small of a scale. There are REASONS the changes come about, and those need to be fully reflected in any prediction being made.

Regarding the use of in-season data, I have detailed EV shot data for 210 team-seasons (30 teams - 7 seasons).

So - clearly - sample size is a not concern for the purpose of determining how teams regress from one half of the season to the other.

This will be obvious to all but the most analytically clueless of persons.

 

[Delicious Dangles*]

Regarding the use of in-season data, I have detailed EV shot data for 210 team-seasons (30 teams - 7 seasons).

So - clearly - sample size is a not concern for the purpose of determining how teams regress from one half of the season to the other.

This will be obvious to all but the most analytically clueless of persons.

Except it's still over only 7 seasons, so it is still subject to the league-wide biases of those seasons.

And yes, scoring will fluctuate at different and probably even predictable times of the season, but as I said before, there are REASONS for that beyond simple regression.

This should be obvious to everybody.

So before people go off spouting predictions, all these things need to be accounted for.

 

[King Mapes]

Perhaps this is the wrong thread for you then? Corsi and shot quality are right in the title man. If you want to celebrate, go to the Leafs forum.

So you cant speak negatively about Corsi in this thread"?

 

[zeke]

My most significant problem with the current usage of shot-based metrics is that the users are completely ignoring any attempt at individual player analysis, and are simply trying to predict future team success based on a team metric. This seems odd to me, and is something that baseball sabermatricians would never do in their sport. They would always build their team predictions out from their individual player projections.

I would suggest that based on individual player scoring history, that both the Leafs and the Lightning will continue to be two of the highest scoring teams in the league, despite their low shot totals this year and last. And there has yet to be a convincing shots-based argument as to why we should expect the players on these teams to stop scoring at their usual levels.

Now we can do a much more in depth individual analysis than the following, but let's keep it straight up simple for now. Even looking at simply the Career goals per game pace of each individual on the Leafs' roster, for example, would lead to a firm prediction that the team would score in the vicinity of 3 goals per game this year (even firmer when you adjust for age effects). Summing up career goals per game for the Leafs' young team would result in about a 2.80gpg per game projection (and, importantly, this would not be abusing any individual TOI numbers by projecting players to get more minutes than possible). Given the very young age of the players, career numbers would likely be an underrating of their current skill, and summing up their last 3yrs GPG would result in a projection closer to 3.25gpg. If we factor in some regression and injuries in there just because, and cut the projection by 10% or so, we'd still be coming in at around 3 goals per game for the team. And all of that not based on anything other than the actual scoring ability they've shown over their careers so far. The lightning would result in a similar projection as well.

Based on the proven scoring talent that each team has, it seems much more logical that their current low shot totals are the anomaly, not their high goal totals. Saying that the team shot totals over the last ~60gms is more important or predicitve than the scoring ability they have shown over their nhl careers seems, well, silly.

Looking at the Leafs again, we can also look at the goalies' histories, and make a fairly solid prediction that they will continue to receive at least above average goaltending, and given age effects likely well above average goaltending. The Leafs' two 25yr old goalies have combined for a .916sv% on 5282 combined shots in the NHL regular season and playoffs, and before that combined for a .919sv% on 5533 combined shots in the AHL regular season and playoffs. Yes, the goalies are not proven, but based on everything we know about their numbers and ability, projecting a save percentage of above .915 and approaching .920 is a very reasonable projection, and one which would put them at least in the top-10 teams, and maybe much higher than that.

When you build the Leafs' team projections from the individual players out, it's hard not to project a team that will be very good offensively, and receive very good goaltending. None of this is set in stone, of course, but that's what their individual talent and track records clearly suggest.

Trying to project team defense is a little harder, of course, because no reliable individual defensive metrics exist. If we use shots against as a proxy for team defense - which seems fair - then we'd also project the Leafs to be a poor defensive team this year.

In the case of the Leafs, we have two very different methods of projection which in fact contradict each other:

1.The advanced stats literally claim that the leafs team is one of the very worst in all of hockey.
2.Individual projections claim that the Leafs will be very strong in at least 2 of the 3 major areas of team performance.


Now here's the rub - I don't know, but I would guess that in the vast majority of cases, the Corsi/fenwick numbers and the individual projection numbers agree quite closely. I would guess that in most of the cases where the advanced stats predict a significantly different outcome when regression occurs would actually be supported by the individual player projections for that same team. i.e. in most cases where Corsi indicates a team is overacheiving, we can also easily see that the team is overachieving based on their individual player track records (i.e. that a number of individuals are performing at a much higher level than they could have been expected to over the longterm.

But it wouldn't surprise me in the least that in the cases where the advanced stats conflict so blatantly with the individual projections, that the advanced stats lose much of their strength and their outcomes don't come true nearly as often.

So for me, saying that "most every exception to Corsi/Fenwick fade in time" is not a good enough response to this specific team's performance. I'd be much more interested in looking only at the subset of teams in which Corsi/Fenwick sees overachievment while the individual projections see expected achievement - like this current leafs team.

Because, again, bottom line is that based on the Leafs' players track records, their goals for and against numbers over this year and last (which correspond fairly to their win/loss record) are simply not unexpected at all.


When we are dealing with stats that are based on a relatively small sample size AND whose correlation with future wins is noticable but not particularly strong (like the current sum total of corsi/fenwick numbers are), we shouldn't be treating every exception to the rule as just a fluke - we should in fact be looking at exceptions as opportunities to explore weaknesses and fine tune our model.

 

[Master_Of_Districts]

Except it's still over only 7 seasons, so it is still subject to the league-wide biases of those seasons.

.

Name the biases that you think would impact the analysis.

And then explain why you think those biases would impact the analysis.

Because I just don't see it.

Seriously - we're talking about 210(!) team seasons of data.

And you're objecting on the basis that the sample is inadequate.

We've effectively reached the point where you'll tell any lie imaginable in order to avoid admitting that you're wrong.

 

[Delicious Dangles*]

Name the biases that you think would impact the analysis.

And then explain why you think those biases would impact the analysis.

Because I just don't see it.

Seriously - we're talking about 210(!) team seasons of data.

And you're objecting on the basis that the sample is inadequate.

We've effectively reached the point where you'll tell any lie imaginable in order to avoid admitting that you're wrong.

There are an endless number of things that could affect the model, and I've even mentioned some, but you ignore what you can't explain.

How the salary cap affected it, how the lockout affected it, how one of the best drafts in the history of the NHL affected it, how league-age, affected it, how the systems employed over those years affected it, etc, etc, etc.

No, 210 team seasons is not some god-like magic number. It's not that much actually, and even worse when it is confined to a short time period.

I have told no lies. For some reason, you're just unwilling to look at the bigger picture, instead of numbers and equations.

 

[TieClark]

Have to agree... the game changed drastically after the last lockout as well as in the early 90's.

 

[habsfanatics*]

If the leafs keep playing like they're playing, I expect to see a serious drop in results.

 

[Bear of Bad News]

So you cant speak negatively about Corsi in this thread"?

I wouldn't call "Bring up Corsi all you want you'll just get ignored" very conducive to discussion about CORSI. Would you?

If someone went into the Toronto forum and said "Talk about the Leafs all you want - you'll just get ignored," what would you say?

 

[madmike77]

Well looks like the people who said the Leafs can't keep winning if they keep getting outshot may have been on to something.

 

[King Mapes]

Well looks like the people who said the Leafs can't keep winning if they keep getting outshot may have been on to something.

Lol, not really. Every team loses during the season. A few years ago under RW they regularly outshot opponents and lost. So they were unlucky, right?