The value of Corsi and Fenwick

[Freudian]

I see, I had a different formula for GF% in mind. What data source did you use?

Data source for goals for and goals against? I used nhl.com.

 

[charlie1]

gf% = gf / (gf + ga)

x 100

I'd still like to know the source...

 

[Doctor No]

Just butting my head in here...

Some sources include shootouts (as one goal for the winning team) in GF/GA totals, and some don't. That could explain what you're seeing, but I haven't looked at it.

 

[charlie1]

Data source for goals for and goals against? I used nhl.com.

I was talking about all of the data. Did you compile it from several sources?

 

[Freudian]

I was talking about all of the data. Did you compile it from several sources?

I got the corsi and fenwick data from http://stats.hockeyanalysis.com/. The rest from nhl.com.

 

[Freudian]

Just butting my head in here...

Some sources include shootouts (as one goal for the winning team) in GF/GA totals, and some don't. That could explain what you're seeing, but I haven't looked at it.

I used the stats that didn't award a goal for winning the shootout. GF and GA.

I doubt it would alter the results in a significant way either way.

 

[charlie1]

It's funny, if we ask the question, "which stat is most predictable?" we get the exact opposite answer. Using 2012-2013 to predict 2013-2014:

Correlation between CF% in 2012-2013 and CF% in 2013-2014: 0.78
Correlation between FF% in 2012-2013 and FF% in 2013-2014: 0.75
Correlation between GF% in 2012-2013 and GF% in 2013-2014: 0.59
Correlation between Wins in 2012-2013 and Wins in 2013-2014: 0.53

Data attached. R code below.

data <- read.csv("war-on-ice-2014-10-08 20-46-29.csv")

data2013 <- subset(data, data$season==20132014)
data2012 <- subset(data, data$season==20122013)

cor(data2012$Corsi., data2013$Corsi.)
cor(data2012$Fenwick., data2013$Fenwick.)
cor(data2012$TeamGoal., data2013$TeamGoal.)
cor(data2012$Wins, data2013$Wins)

 

[badtakemachine]

It's funny, if we ask the question, "which stat is most predictable?" we get the exact opposite answer. Using 2012-2013 to predict 2013-2014:

Correlation between CF% in 2012-2013 and CF% in 2013-2014: 0.78
Correlation between FF% in 2012-2013 and FF% in 2013-2014: 0.75
Correlation between GF% in 2012-2013 and GF% in 2013-2014: 0.59
Correlation between Wins in 2012-2013 and Wins in 2013-2014: 0.53

Data attached. R code below.

data <- read.csv("war-on-ice-2014-10-08 20-46-29.csv")

data2013 <- subset(data, data$season==20132014)
data2012 <- subset(data, data$season==20122013)

cor(data2012$Corsi., data2013$Corsi.)
cor(data2012$Fenwick., data2013$Fenwick.)
cor(data2012$TeamGoal., data2013$TeamGoal.)
cor(data2012$Wins, data2013$Wins)

Interesting research, nice work. This is the argument that the analytics crowd are trying to make - repeatability. Clearly, the 'Corsi' and 'Fenwick' numbers are more repeatable than the other numbers, largely due to the fact that, to a relatively large degree, these numbers are controllable and have less to do with luck.

Where I think the argument that these numbers matter falls apart: people are using these 'Corsi' and 'Fenwick' numbers as if they are equivalent to puck possession.

IMO, counting shot attempts is a quick and dirty way to go through piles of data to estimate possession for the majority of teams, yes, but it clearly does not account for teams that would rather wait or pass to wait for a higher percentage shot. A better quick and dirty to estimate puck possession would be to count time between zone entry and zone exit. This is assuming actually tracking puck possession time itself is not possible.

 

[Bender66]

Interesting research, nice work. This is the argument that the analytics crowd are trying to make - repeatability. Clearly, the 'Corsi' and 'Fenwick' numbers are more repeatable than the other numbers, largely due to the fact that, to a relatively large degree, these numbers are controllable and have less to do with luck.

Where I think the argument that these numbers matter falls apart: people are using these 'Corsi' and 'Fenwick' numbers as if they are equivalent to puck possession.

IMO, counting shot attempts is a quick and dirty way to go through piles of data to estimate possession for the majority of teams, yes, but it clearly does not account for teams that would rather wait or pass to wait for a higher percentage shot. A better quick and dirty to estimate puck possession would be to count time between zone entry and zone exit. This is assuming actually tracking puck possession time itself is not possible.

It's not hard to understand why CF and FF are so closely correlated from year to year. That's because from year to year, team systems don't change much unless you change coaches. So a Sutter coached Kings team will always throw pucks on net from everywhere and anywhere regardless of personnel changes between years.

 

[badtakemachine]

It's not hard to understand why CF and FF are so closely correlated from year to year. That's because from year to year, team systems don't change much unless you change coaches. So a Sutter coached Kings team will always throw pucks on net from everywhere and anywhere regardless of personnel changes between years.

I mostly agree with this - unless Sacco is your head coach. Then you can probably predict your 'Corsi' rating with a dartboard.

 

[Mubiki]

Interesting research, nice work. This is the argument that the analytics crowd are trying to make - repeatability. Clearly, the 'Corsi' and 'Fenwick' numbers are more repeatable than the other numbers, largely due to the fact that, to a relatively large degree, these numbers are controllable and have less to do with luck.

Where I think the argument that these numbers matter falls apart: people are using these 'Corsi' and 'Fenwick' numbers as if they are equivalent to puck possession.

IMO, counting shot attempts is a quick and dirty way to go through piles of data to estimate possession for the majority of teams, yes, but it clearly does not account for teams that would rather wait or pass to wait for a higher percentage shot. A better quick and dirty to estimate puck possession would be to count time between zone entry and zone exit. This is assuming actually tracking puck possession time itself is not possible.

Counting the shots actually makes sense though. If you possess the puck, the other team can't shoot. It's defense through possession. If you possess the puck but never really shoot, then your possession has zero relevance in an offensive context.

 

[badtakemachine]

Counting the shots actually makes sense though. If you possess the puck, the other team can't shoot. It's defense through possession. If you possess the puck but never really shoot, then your possession has zero relevance in an offensive context.

Yes, if you possess the puck but never shoot, then possession has absolutely nothing to do with winning. However, this is not reality, and the argument is that possession does have an effect on winning. Even the teams most reluctant to shoot the puck will still get 40 shot attempts in a game. It is not unreasonable to think that the Kings may get 70 shot attempts with 15 scoring chances, but the Avs may get 45 shot attempts with 12 scoring chances. If this is repeated over 82 games, I don't see why it is unreasonable to believe that the Avs should have a better shooting percentage. Of course, the ultimate goal is to score goals, which come from scoring chances, which can come from (a) shooting the puck at the net (b) making a pass to an open player or © possessing the puck for a few extra seconds to change the lane or shooting angle, or any combination of the above. Sure, the majority may come from (a), but that doesn't paint the whole picture.

The fact is, possessing the puck is a can be a means for generating scoring chances. Case in point, just watch a good old power play. It is immediately noticeable how much longer teams will wait before shooting. Some teams attempt this at 5 on 5. It is a perfectly normal, albeit rare, style of offence. It can vary player-to-player, or coach-to-coach.

 

[Bender66]

Counting the shots actually makes sense though. If you possess the puck, the other team can't shoot. It's defense through possession. If you possess the puck but never really shoot, then your possession has zero relevance in an offensive context.

No, it doesn't. You just completely contradicted yourself in the span of 2 consecutive sentences. If you possess the puck but never really shoot, then IT IS BY DEFINITION DEFENSE THROUGH POSSESSION. Holding onto the puck for 60 minutes without shooting IS DEFENSE THROUGH POSSESSION. Remember that game in 2011 when Male Tina Fey was still coaching in Tampa, and Pronger was still playing for Philly? Pronger and Cobourn just sat there with the puck for like 5 minutes just to taunt Tampa. THAT WAS DEFENSE THROUGH POSSESSION.

Shooting the puck a lot in a game IS NOT defense through possession. Do you not see the factually incorrect statement you just made?
Shooting the puck a lot has no correlation with preventing the other team from possessing the puck for a longer duration than you.

Counting shots to count possession is a false equivalency. I don't know why this incredible simple fact is so hard to comprehend.

>>

Puck rolls up the side board to Doughty at the point. Without even looking at the net, and trying to handle the puck, Doughty takes a wrist shot from the blue line, it hits the goalie in the chest but takes a crazy rebound at a weird angle off to the side and Jeff Carter slams it home for a goal. The Kings have had possession of the puck for a grand total of maybe .4 secs.

Tell me where in that scenario the 2 corsi events relates in any way to the Kings having possession of the puck.

>>

The idea of keeping possession of the puck is to prevent the other team from having the puck so your team can try to generate high quality scoring chances and not just give the puck away in the neutral zone.

The idea of shooting the puck a lot is to try and generate rebounds and chaos in front of the opposing goalies net. It has nothing to do with how much time you possess the puck.

Trying to correlate the two things is poor application of inferential statistics.

It's like reading 538.com has made every college sophomore who took Statistics 100 into Nate Silver.

 

[BB1133]

I like to point back to this article on "why possession matters".

http://www.habseyesontheprize.com/2013/4/4/4178716/why-possession-matters-a-visual-guide-to-fenwick

Since it was published we've seen Hawks vs Bruins and Kings vs Rangers in the finals - all Fenwick heavyweights.

If you're a poor puck possession team your odds to go on a cup run are slim to none.

 

[hatterson]

No, it doesn't. You just completely contradicted yourself in the span of 2 consecutive sentences. If you possess the puck but never really shoot, then IT IS BY DEFINITION DEFENSE THROUGH POSSESSION. Holding onto the puck for 60 minutes without shooting IS DEFENSE THROUGH POSSESSION. Remember that game in 2011 when Male Tina Fey was still coaching in Tampa, and Pronger was still playing for Philly? Pronger and Cobourn just sat there with the puck for like 5 minutes just to taunt Tampa. THAT WAS DEFENSE THROUGH POSSESSION.

Shooting the puck a lot in a game IS NOT defense through possession. Do you not see the factually incorrect statement you just made?
Shooting the puck a lot has no correlation with preventing the other team from possessing the puck for a longer duration than you.

Counting shots to count possession is a false equivalency. I don't know why this incredible simple fact is so hard to comprehend.

>>

Puck rolls up the side board to Doughty at the point. Without even looking at the net, and trying to handle the puck, Doughty takes a wrist shot from the blue line, it hits the goalie in the chest but takes a crazy rebound at a weird angle off to the side and Jeff Carter slams it home for a goal. The Kings have had possession of the puck for a grand total of maybe .4 secs.

Tell me where in that scenario the 2 corsi events relates in any way to the Kings having possession of the puck.

>>

The idea of keeping possession of the puck is to prevent the other team from having the puck so your team can try to generate high quality scoring chances and not just give the puck away in the neutral zone.

The idea of shooting the puck a lot is to try and generate rebounds and chaos in front of the opposing goalies net. It has nothing to do with how much time you possess the puck.

Trying to correlate the two things is poor application of inferential statistics.

It's like reading 538.com has made every college sophomore who took Statistics 100 into Nate Silver.

It sounds like you're not actually looking at the data that's been compiled and reported on. No one, at least no one rational, is saying that every corsi event means the team had solid possession prior to it. Nor is anyone saying that every time of extended possession results in a corsi event.

Yes, it's entirely possible for a guy to head into the zone 1 on 4 and just loft it at the net before turning to go off on a line change and it's entirely possible for a team to cycle in the opposing end for 3 straight minutes and generate 0 corsi events. No one disagrees with this and presenting and arguing those things as example of how "stat guys" aren't getting it is an absurd strawman.

The simply facts are that
1.) Fenwick close is extremely repeatable. In fact, virtually no measures in hockey are as consistently repeatable as a teams fenwick close.
2.) Fenwick close is highly correlated with winning over the long run (as seen a post or two above mine)
3.) Fenwick close isn't actually a great indicator of who will win a single game. (http://puckprediction.com/2013/11/26/is-fenwick-close-a-consistent-predictor-of-winning/ and http://rinkstats.blogspot.com/2013/12/why-popular-advanced-stats-are-bad-at.html) Arguing that it isn't doesn't show who will win on Saturday, or giving examples of games where a team heavily out Fenwick'd their opponent and lost, doesn't tell us anything about advanced stats that "we" don't already know.
4.) Teams do not realistically take either of a polar approaches of "lets possess the puck for a really long time and not shoot" or "lets chuck a puck at the net every time we touch it" They play in the middle ground, some fading more to one side than the other, but mostly staying near the center.

No one has ever argued that corsi/fenwick are equal to possession. It's argued that it's generally an indicator of possession and ultimately it's a really highly correlated with winning, which is what matters.

 

[charlie1]

Rather than just looking at correlations from year 1 to year 2, this is probably closer to what someone would actually do if they wanted to predict Wins:

Train their model using the last two years of data, and then make predictions about the upcoming year. Doing that using 2010-2012 data to predict Wins in 2013-2014 I found that the model that includes Fenwick outperforms the model that includes only Wins.

Correlation between predicted and actual wins using only Wins: 0.25
Correlation between predicted and actual wins using wins and fenwick: 0.49

So there does appear to be a significant amount of predictive value added when we include Fenwick.

I used a simple linear model. Code below, data attached.

###########
data <- read.csv("/home/c/Downloads/war-on-ice-2014-10-08 20-46-29.csv")

data2013 <- subset(data, data$season==20132014)
data2012 <- subset(data, data$season==20122013)
data2011 <- subset(data, data$season==20112012)

dat_train <- data.frame(dat_year0 = data2011, dat_year1 = data2012)

dat_test <- data.frame(dat_year0 = data2012, dat_year1 = data2013)

model_wins <- lm(dat_year1.Wins ~ dat_year0.Wins, dat=dat_train)
model_wins_fenwick <- lm(dat_year1.Wins ~ dat_year0.Wins + dat_year0.Fenwick.,
dat=dat_train)

pred_wins <- predict(object=model_wins, data=dat_test)
pred_wins_fenwick <- predict(object=model_wins_fenwick, data=dat_test)

observed <- dat_test$dat_year1.Wins

cor(pred_wins, observed)
cor(pred_wins_fenwick, observed)

 

[Rogie]

I posted on Leafs board about Panick.

I mentioned that the addition of Panick might be a bit of test (in a small way) for the value of Corsi and Fenwick.

If we throw out for a second the rather intangible variables like motivation and getting a fresh start, and if we try to look at the players in a purely statistical way, then, I'm going to hypothesize the following about Panick.

For context, I look at his qual of comp and qual of teammates and he ranka about middle of the pack of forwards in TB. Personally, I don't think (jmo) that qual of comp or qual of teammates have as large an effect on a players possession but, in any event, he's about in the middle - roughly.

But, looking at his FF% and CF% and ES zone adjusted (all situations and close as well), he's either the worst or 2nd or 3rd worst ranked TB forward (out of about 15 forwards last year with more than 100 minutes - controlled for).
So, Corsi/Fenwick were very bad - and a better than average possession team.

His GF% ranked lowest (or 2nd lowest) at about 35% - lower than any Leaf last year - Colton Orr and JayMaclellan were 40%.

So, possession wise, statistically at least, (if I've done any justice to context) he ranked certainly as one of the worst players among TB forwards.

If he's going to take a 4th line spot (some Leafs fans envision him as more and some espousing - GIVE HIM A SHOT ON THE FIRST LINE??!!), just for fun comparisons, I'm comparing him to both Frattin and Ashton. Those are a couple of the players he could be trying to compete with to win a spot on the 4th line Leafs.

Both Frattin and Ashton have pretty good possession numbers. Frattin consistently had very good numbers before he left for the year and he continued okay possession numbers with both the Kings and the Jackets - even though he didn't stick and didn't make the cut with those teams. Ashton, (despite his critics) has had okay possession numbers with the Leafs.

But, just BY THE NUMBERS, for me it'll be fun to watch and see how it unfolds. I don't know how Panick plays - I know nothing of him at all NOTHING.

In summary, a poor possession player from a good possession team competing against good possession players from a poor possession team - Ceteris Paribus - well, not really, but anyways.

 

[Mubiki]

No, it doesn't. You just completely contradicted yourself in the span of 2 consecutive sentences. If you possess the puck but never really shoot, then IT IS BY DEFINITION DEFENSE THROUGH POSSESSION. Holding onto the puck for 60 minutes without shooting IS DEFENSE THROUGH POSSESSION. Remember that game in 2011 when Male Tina Fey was still coaching in Tampa, and Pronger was still playing for Philly? Pronger and Cobourn just sat there with the puck for like 5 minutes just to taunt Tampa. THAT WAS DEFENSE THROUGH POSSESSION.

Shooting the puck a lot in a game IS NOT defense through possession. Do you not see the factually incorrect statement you just made?
Shooting the puck a lot has no correlation with preventing the other team from possessing the puck for a longer duration than you.

Counting shots to count possession is a false equivalency. I don't know why this incredible simple fact is so hard to comprehend.

>>

Puck rolls up the side board to Doughty at the point. Without even looking at the net, and trying to handle the puck, Doughty takes a wrist shot from the blue line, it hits the goalie in the chest but takes a crazy rebound at a weird angle off to the side and Jeff Carter slams it home for a goal. The Kings have had possession of the puck for a grand total of maybe .4 secs.

Tell me where in that scenario the 2 corsi events relates in any way to the Kings having possession of the puck.

>>

The idea of keeping possession of the puck is to prevent the other team from having the puck so your team can try to generate high quality scoring chances and not just give the puck away in the neutral zone.

The idea of shooting the puck a lot is to try and generate rebounds and chaos in front of the opposing goalies net. It has nothing to do with how much time you possess the puck.

Trying to correlate the two things is poor application of inferential statistics.

It's like reading 538.com has made every college sophomore who took Statistics 100 into Nate Silver.

You said a whole lot without saying anything.

It's a valid assumption that teams use possession to generate scoring chances. The reason that nobody tracks time of possession is because it's inherently present in shots against. If you possess the puck a lot, you will typically give up less shots, and take more.

Shots for and against are what's truly relevant. Perhaps you simply have a problem with the way the term "possession" is applied in hockey. In general, the term refers to corsi and fenwick in hockey. If you have a petty semantics argument, you should have just made it instead of a useless wall of text.

TLDR: Nobody cares how long you "posses" the puck if you aren't generating shots or preventing them. I didn't coin the term "possession" in hockey, so arguing semantics is pointless.

 

[Rogie]

No, it doesn't. You just completely contradicted yourself in the span of 2 consecutive sentences. If you possess the puck but never really shoot, then IT IS BY DEFINITION DEFENSE THROUGH POSSESSION. Holding onto the puck for 60 minutes without shooting IS DEFENSE THROUGH POSSESSION. Remember that game in 2011 when Male Tina Fey was still coaching in Tampa, and Pronger was still playing for Philly? Pronger and Cobourn just sat there with the puck for like 5 minutes just to taunt Tampa. THAT WAS DEFENSE THROUGH POSSESSION.

Shooting the puck a lot in a game IS NOT defense through possession. Do you not see the factually incorrect statement you just made?
Shooting the puck a lot has no correlation with preventing the other team from possessing the puck for a longer duration than you.

Counting shots to count possession is a false equivalency. I don't know why this incredible simple fact is so hard to comprehend.

>>

Puck rolls up the side board to Doughty at the point. Without even looking at the net, and trying to handle the puck, Doughty takes a wrist shot from the blue line, it hits the goalie in the chest but takes a crazy rebound at a weird angle off to the side and Jeff Carter slams it home for a goal. The Kings have had possession of the puck for a grand total of maybe .4 secs.

Tell me where in that scenario the 2 corsi events relates in any way to the Kings having possession of the puck.

>>

The idea of keeping possession of the puck is to prevent the other team from having the puck so your team can try to generate high quality scoring chances and not just give the puck away in the neutral zone.

The idea of shooting the puck a lot is to try and generate rebounds and chaos in front of the opposing goalies net. It has nothing to do with how much time you possess the puck.

Trying to correlate the two things is poor application of inferential statistics.

It's like reading 538.com has made every college sophomore who took Statistics 100 into Nate Silver.

Kings had possession of the puck - in the first place - when they entered the OZ. So, that POSSESSION needs to be counted, since it led to further possession. Then, after all the back and forth possession that transpired (you haven't provided the details), Kings got possession again. You state something like "the puck rolls up the boards", but, maybe a King rolled it up the boards while the defenders are madly trying to 'regain' possession (most likely scenario here, otherwise, the puck would of been passed or cleared the zone, or some safer option, that simply, rolling up the boards to an opposing D man (esp. one with Doughty's skill). So, you are missing a whole lot of possession that led to this supposed .4 sec. of possession. For however long the puck was in the (Kings Ozone) what was the possession time of the Kings and what was the possession time of your example team? Then we can talk about the .4 seconds are whether or not the shot correlates with possession or not. And even if the Dzone team had more actual possession time than the Kings but lost possession that resulted in a shot, that still doesn't show that possession does not correlate with shots.