Re: 5v5 CF% Is Bad, Right?

4
T.C. wrote:please interpret this for the people that may be too stupid to understand what they're looking at. not me, obviously, but i'm asking for those that are too proud to to do so. i think the URL is a bit deceiving.
Wait, you're asking the guy who claims to provide statistical analysis to actually provide that analysis?

Insanity!

Re: 5v5 CF% Is Bad, Right?

5
This is a discussion board. I like discussions. I especially like discussing advanced stats and learning more and more. I get that you're trying to drive people to your site, but is this the right vehicle for driving up your website traffic? Since this is a discussion board, can you just discuss instead of posting a link? I respectfully ask because I access this discussion board from my work computer and it blocks any site with the word "blog" in it.

Re: 5v5 CF% Is Bad, Right?

6
SickMittsRomKnee wrote:This is a discussion board. I like discussions. I especially like discussing advanced stats and learning more and more. I get that you're trying to drive people to your site, but is this the right vehicle for driving up your website traffic? Since this is a discussion board, can you just discuss instead of posting a link? I respectfully ask because I access this discussion board from my work computer and it blocks any site with the word "blog" in it.
I can totally understand your issue with the site being blocked. Perhaps he can link the URL to the words "Click Here" or something like that. I like the site and appreciate the link being posted.

Re: 5v5 CF% Is Bad, Right?

7
HockeyFan85 wrote:
T.C. wrote:please interpret this for the people that may be too stupid to understand what they're looking at. not me, obviously, but i'm asking for those that are too proud to to do so. i think the URL is a bit deceiving.
Wait, you're asking the guy who claims to provide statistical analysis to actually provide that analysis?

Insanity!
not trying to be a dick, i am just clearly too dumb to know wtf i'm looking at. i'd like to know so we can discuss.

Re: 5v5 CF% Is Bad, Right?

8
I didn't mean to be overly cryptic.

Bottom line, one of the gold-standard "advanced stats" does a terrible job of predicting winning in the regular season.

Many examples by team, "low" Corsi and lots of points (Montreal and Rangers). "High" corsi and not many points (Carolina).

And almost all of the correlation, to the extent that there is one, is driven by one data point (Buffalo).

If Corsi fails like this at the team level that means that all of those by-player Corsi analyses are mostly worthless.

Yes, mostly worthless.

Take home: Forget about Corsi. Or at least understand that it correlates not very well with actually winning hockey games. Kind of like analyzing the numbers on the winning lotto balls?

I'll try to be more expansive going forward.

Thanks.

Re: 5v5 CF% Is Bad, Right?

9
As for the Kings and last season, flipping a coin to predict who will win the Cup will be right sometimes.

Many of the "advance analytics" crowd are in a tizzy trying to explain, post hoc (after the fact), why the Corsi darling's the Kings didn't even make the playoffs this year.

And, this is a worthwhile endeavor if, and only if, the results of these analyses are used to create a new model of how to win hockey games and not as one-off after the fact explanations.

This new model could then be back-tested against previous data to see what the predictive value is.

This does not seem to be the focus.

It appears that such an industry has grown up around Corsi that the focus instead is to defend it.

Re: 5v5 CF% Is Bad, Right?

10
HockeyFan85 wrote:
T.C. wrote:please interpret this for the people that may be too stupid to understand what they're looking at. not me, obviously, but i'm asking for those that are too proud to to do so. i think the URL is a bit deceiving.
Wait, you're asking the guy who claims to provide statistical analysis to actually provide that analysis?

Insanity!
Are you always this unpleasant?

Or is it just towards me?

Re: 5v5 CF% Is Bad, Right?

12
WebSant wrote:I didn't mean to be overly cryptic.

Bottom line, one of the gold-standard "advanced stats" does a terrible job of predicting winning in the regular season.

Many examples by team, "low" Corsi and lots of points (Montreal and Rangers). "High" corsi and not many points (Carolina).

And almost all of the correlation, to the extent that there is one, is driven by one data point (Buffalo).

If Corsi fails like this at the team level that means that all of those by-player Corsi analyses are mostly worthless.

Yes, mostly worthless.

Take home: Forget about Corsi. Or at least understand that it correlates not very well with actually winning hockey games. Kind of like analyzing the numbers on the winning lotto balls?

I'll try to be more expansive going forward.

Thanks.
I'm going to respectfully disagree with you here. If I read this graph correctly (quite possible I'm not), it looks to me like Corsi is very beneficial. The teams below 50% in 5v5 Corsi are: BUF, COL, CGY, TOR, ARI, CBJ, EDM, NJ, PHI, VAN, MTL, NYR. Only 4 of those teams made the playoffs, while 8 did not. 2 of those teams (NYR and MTL) have arguably the 2 best goalies in the world to cover up for below average possession numbers (just below average in the case of those 2). On the flip side, if LA would've had even a mediocre record in the shootout this season (they were dead last in the league at 2-8) they would've been in the playoffs this year.

Of course these advanced stats don't paint the entire picture, but there certainly does seem to be ample evidence that Corsi 5v5 is a useful stat. Again, it doesn't tell you everything, no single statistic does, but it's certainly worth paying attention and can have some predictive worth.
...but whatever, the Blues won the Cup!!!!!

Re: 5v5 CF% Is Bad, Right?

13
Dave's a mess wrote:
WebSant wrote:I didn't mean to be overly cryptic.

Bottom line, one of the gold-standard "advanced stats" does a terrible job of predicting winning in the regular season.

Many examples by team, "low" Corsi and lots of points (Montreal and Rangers). "High" corsi and not many points (Carolina).

And almost all of the correlation, to the extent that there is one, is driven by one data point (Buffalo).

If Corsi fails like this at the team level that means that all of those by-player Corsi analyses are mostly worthless.

Yes, mostly worthless.

Take home: Forget about Corsi. Or at least understand that it correlates not very well with actually winning hockey games. Kind of like analyzing the numbers on the winning lotto balls?

I'll try to be more expansive going forward.

Thanks.
I'm going to respectfully disagree with you here. If I read this graph correctly (quite possible I'm not), it looks to me like Corsi is very beneficial. The teams below 50% in 5v5 Corsi are: BUF, COL, CGY, TOR, ARI, CBJ, EDM, NJ, PHI, VAN, MTL, NYR. Only 4 of those teams made the playoffs, while 8 did not. 2 of those teams (NYR and MTL) have arguably the 2 best goalies in the world to cover up for below average possession numbers (just below average in the case of those 2). On the flip side, if LA would've had even a mediocre record in the shootout this season (they were dead last in the league at 2-8) they would've been in the playoffs this year.

Of course these advanced stats don't paint the entire picture, but there certainly does seem to be ample evidence that Corsi 5v5 is a useful stat. Again, it doesn't tell you everything, no single statistic does, but it's certainly worth paying attention and can have some predictive worth.
You have reiterated several of my points for me.

Corsi 5v5 is a useful stat to the extent that it predicts 29% of regular season points. 13% if you take Buffalo out of the analysis. If some deem that to be "useful" then great.

The post-hoc explanations, e.g., MTL and NYR, should be used to build a better model. Perhaps add in GA/GP in a multiple regression.

Similarly with shootout record. If it explains the results this past season then build it into the model and back-test against previous seasons and see how it predicts going forward.

I'll have to check to see whether these post-hoc "explanations" were the same or different last year.

Re: 5v5 CF% Is Bad, Right?

14
WebSant wrote:
Dave's a mess wrote:
WebSant wrote:I didn't mean to be overly cryptic.

Bottom line, one of the gold-standard "advanced stats" does a terrible job of predicting winning in the regular season.

Many examples by team, "low" Corsi and lots of points (Montreal and Rangers). "High" corsi and not many points (Carolina).

And almost all of the correlation, to the extent that there is one, is driven by one data point (Buffalo).

If Corsi fails like this at the team level that means that all of those by-player Corsi analyses are mostly worthless.

Yes, mostly worthless.

Take home: Forget about Corsi. Or at least understand that it correlates not very well with actually winning hockey games. Kind of like analyzing the numbers on the winning lotto balls?

I'll try to be more expansive going forward.

Thanks.
I'm going to respectfully disagree with you here. If I read this graph correctly (quite possible I'm not), it looks to me like Corsi is very beneficial. The teams below 50% in 5v5 Corsi are: BUF, COL, CGY, TOR, ARI, CBJ, EDM, NJ, PHI, VAN, MTL, NYR. Only 4 of those teams made the playoffs, while 8 did not. 2 of those teams (NYR and MTL) have arguably the 2 best goalies in the world to cover up for below average possession numbers (just below average in the case of those 2). On the flip side, if LA would've had even a mediocre record in the shootout this season (they were dead last in the league at 2-8) they would've been in the playoffs this year.

Of course these advanced stats don't paint the entire picture, but there certainly does seem to be ample evidence that Corsi 5v5 is a useful stat. Again, it doesn't tell you everything, no single statistic does, but it's certainly worth paying attention and can have some predictive worth.
You have reiterated several of my points for me.

Corsi 5v5 is a useful stat to the extent that it predicts 29% of regular season points. 13% if you take Buffalo out of the analysis. If some deem that to be "useful" then great.

The post-hoc explanations, e.g., MTL and NYR, should be used to build a better model. Perhaps add in GA/GP in a multiple regression.

Similarly with shootout record. If it explains the results this past season then build it into the model and back-test against previous seasons and see how it predicts going forward.

I'll have to check to see whether these post-hoc "explanations" were the same or different last year.
Adding PDO and making a multiple regression might increase the predictive power of the model. I would think that would control for some of the things, since PDO takes into account save percentage. Isn't PDO the only "advanced stat" you've seen that predicts winning relatively well?

Re: 5v5 CF% Is Bad, Right?

15
Perfect timing.

Even Strength 5 v 5 PDO R-squared = 0.51.

Even Strength 5v5 Goals Scored R-squared = 0.63.

Score four, baby!

http://hockeysimplified.blogspot.com/" onclick="window.open(this.href);return false;

And, how do you score goals?

At it's most basic level you put players on the ice who have shown that they have the ability to score goals and you let them do their thing.

And, scoring goals does not result in more goals allowed.

Slight negative relationship, in fact.

Turn 'em loose!

Re: 5v5 CF% Is Bad, Right?

16
WebSant wrote:Perfect timing.

Even Strength 5 v 5 PDO R-squared = 0.51.

Even Strength 5v5 Goals Scored R-squared = 0.63.

Score four, baby!

http://hockeysimplified.blogspot.com/" onclick="window.open(this.href);return false;

And, how do you score goals?

At it's most basic level you put players on the ice who have shown that they have the ability to score goals and you let them do their thing.

And, scoring goals does not result in more goals allowed.

Slight negative relationship, in fact.

Turn 'em loose!
Well, PDO is at least better than Corsi! The fancy stat people got one stat (sorta) right.

I think the lot of the issue with these stats is that people aren't looking at them like you do-they just think about it logically. With corsi for example-they just assume that "more shots means more possession which means more wins" when thats not necessarily the case (as the data shows).

Re: 5v5 CF% Is Bad, Right?

17
Haha, I think you might be giving me too much credit for my statistical knowledge, but everything you've said in your last two replies makes sense to me if you're trying to find a predictive statistic.

You bring up PDO, and I wonder what your opinion of it is. I hear it get used a lot to describe how lucky a team is, cliff notes being that a PDO over 1000 for a certain period mean's you've been lucky, under 1000 means you're unlucky. I've always had a problem with that for the same reason I mentioned above before, some teams just have goalies who have a much higher save percentage than others every year. I understand shooting percentage can go hot and cold over the course of a season, but don't some teams just legitimately through either skill or system just have higher shooting percentages as well? Am I missing something, or is the PDO=luckiness thought a little oversimplified?
...but whatever, the Blues won the Cup!!!!!

Re: 5v5 CF% Is Bad, Right?

18
Dave's a mess wrote:Haha, I think you might be giving me too much credit for my statistical knowledge, but everything you've said in your last two replies makes sense to me if you're trying to find a predictive statistic.

You bring up PDO, and I wonder what your opinion of it is. I hear it get used a lot to describe how lucky a team is, cliff notes being that a PDO over 1000 for a certain period mean's you've been lucky, under 1000 means you're unlucky. I've always had a problem with that for the same reason I mentioned above before, some teams just have goalies who have a much higher save percentage than others every year. I understand shooting percentage can go hot and cold over the course of a season, but don't some teams just legitimately through either skill or system just have higher shooting percentages as well? Am I missing something, or is the PDO=luckiness thought a little oversimplified?
Over short stretches a lot of people try to equate it with luckiness, yes you are correct with that. However like you say, if a team's PDO is high over the course of a season then there is obviously something they are doing right to have that high PDO.

My thought with adding PDO into a multiple regression with CF% is that adding PDO into the model will help control for some of the variation with CF and result in a better model overall.