The value of Corsi and Fenwick

[Hivemind]

I offered common sense - if that doesn't resonate with you, there isn't much to discuss.

No, you ignored everything I wrote about how to analyze individual players in greater detail and circled back around to your original premise. You failed to address anything I brought up in terms of relative corsi and WOWY.

Taylor Hall is the best player on the Oilers by a good margin, Corsi says he is one of their worst players. That is completely flawed BS. Giving Hall a - when Jeff Petry passes the puck to an opponent in front of the Oilers goal further underlines what crap Corsi is.

Why are you not providing context for your stats? Sounds to me like it's not Corsi that's "crap," it's your application of it. There are numerous metrics that can work with Corsi to help provide context, from zone starts to the afforementioned WOWY analysis. Of course you don't just sort by CF% and claim the player at the top is the best, no more than you sort by Goals and claim that all the defensemen must suck because they never score goals. Trying to use Corsi as a one-stop shop for your player analysis is flawed, but the knowledgable practitioners of hockey metrics don't do that.

The argument that over a large sample size, it works itself out is garbage. Bad data multiplied just becomes worse data. Corsi advocates are just grasping at straws - they are so desperate to claim they have "advanced" stats, they take complete garbage and try to convince the world it provides insight. The naive are fooled but, those with common sense see holes you can drive a truck through.

How about this "common sense hole" in your argument. If it's not "bad data" (which it isn't, as has been shown via multiple methods, including its ability to predict wins), then it doesn't compound to become worse data. Show me a meaningful rebuttal, rather than sticking to your flawed premise.

If you want to measure an individual player's effectiveness, measure the individual for Heaven's sake. Taking a team measurement (that is lille more than a glorified SOG count) and trying to shoe horn it on individual players is wrong, inaccurate and futile. Is it laziness that drives people to use Corsi rather than really evaluating individual performance? I don't know but, I do know that Corsi provides nothing of value as an individual measure.

When the goal of the metric is to evaluate the individual's impact on the team's performance, it absolutely makes sense to measure the end result of the whole team. People do track individual Corsi (iCorsi) and Fenwick (iFenwick) events, but it defeats the purpose of seeing how that individual impacts the total play on the ice.

 

[BeardyCanuck03]

I offered common sense - if that doesn't resonate with you, there isn't much to discuss.

Taylor Hall is the best player on the Oilers by a good margin, Corsi says he is one of their worst players. That is completely flawed BS. Giving Hall a - when Jeff Petry passes the puck to an opponent in front of the Oilers goal further underlines what crap Corsi is.

The argument that over a large sample size, it works itself out is garbage. Bad data multiplied just becomes worse data. Corsi advocates are just grasping at straws - they are so desperate to claim they have "advanced" stats, they take complete garbage and try to convince the world it provides insight. The naive are fooled but, those with common sense see holes you can drive a truck through.

If you want to measure an individual player's effectiveness, measure the individual for Heaven's sake. Taking a team measurement (that is lille more than a glorified SOG count) and trying to shoe horn it on individual players is wrong, inaccurate and futile. Is it laziness that drives people to use Corsi rather than really evaluating individual performance? I don't know but, I do know that Corsi provides nothing of value as an individual measure.

Corsi is only a part of the evaluation. When looking at a players corsi rating you can't just take it as gospel. You have to look at how the player was deployed, who he was playing with and such. The top teams and analysts will not use just Corsi. If you are battling against Corsi you are going to lose because it is a useful stat.

Corsi does provide value as part of how teams measure individuals. There is enough proof out there that denying this is just stupid. Fans who use Corsi as the be all end all are lazy, but fans who use Corsi as a metric to compare what they see during the game along with comparing it to other stats are going to have a much more eductation and accurate idea of the value and skills of a player.

 

[Whileee]

In a nut shell, how do you score goals? You shoot the puck. Corsi tracks shot attempts, which are shots on goal + shots that missed the net + shots that were blocked. Fenwick tracks shots on goal + shots that missed the net.

Well, it's not quite that simple. You score goals by shooting the puck on net, so why not just use shots on goal (for and against) instead of Corsi and Fenwick? How many shots that missed the net or were blocked end up in goals?

Corsi and Fenwick are basic indicators regarding which team has the puck with possession (mostly in the other team's zone). They are fairly good at indicating this because when teams have possession they often end up with shot attempts, though only some of them end up on net. Having more puck possession than the other team (particularly in the opponent's zone) tends to be a pretty good predictor of team success. That's why Corsi and Fenwick tend to work reasonably well.

However, it is almost inconceivable that better indicators or clusters of indicators cannot be developed that will be both more informative and better predictors. Whether the data to formulate those indicators will be readily available to the general public is less likely. The current fascination with Corsi and Fenwick is largely due to the fact that the data are readily available and they are fairly robust indicators. I would hope that NHL teams will be able to develop better indicators with their better resources, even if fans aren't provided with the details..

 

[hatterson]

However, it is almost inconceivable that better indicators or clusters of indicators cannot be developed that will be both more informative and better predictors. Whether the data to formulate those indicators will be readily available to the general public is less likely. The current fascination with Corsi and Fenwick is largely due to the fact that the data are readily available and they are fairly robust indicators. I would hope that NHL teams will be able to develop better indicators with their better resources, even if fans aren't provided with the details..

I think everyone who takes more than a token look at 'advanced' stats understands and wants that. We know that corsi/fenwick aren't perfect indicators of possession, but they are fairly good.

Hopefully some of the SportVu stuff that's coming in will allow us (by us I mean, the sport of hockey as a whole) to take the next step and measure possession, and how it tracks with winning, in a much more accurate and granular way.

However, until then, corsi/fenwick provides a fairly good estimate, so we'll keep using that until something better comes along.

 

[BeardyCanuck03]

Well, it's not quite that simple. You score goals by shooting the puck on net, so why not just use shots on goal (for and against) instead of Corsi and Fenwick? How many shots that missed the net or were blocked end up in goals?

The idea behind these stats is that they are tracking events that have the potential to become a goal. Shots that hit the post are considered missed shots, not shots on goal. The idea of using them as a possession tracking stat comes from the fact that you need to possess the puck in order to shoot it. They are far from perfect in measuring possession IMO, but as a rough, high level measurement they are good enough.

But in a nutshell (as the question which I answered asked), Corsi/Fenwick are important because it shows events within the game that could become goals.

 

[wgknestrick]

The idea behind these stats is that they are tracking events that have the potential to become a goal. Shots that hit the post are considered missed shots, not shots on goal. The idea of using them as a possession tracking stat comes from the fact that you need to possess the puck in order to shoot it. They are far from perfect in measuring possession IMO, but as a rough, high level measurement they are good enough.

But in a nutshell (as the question which I answered asked), Corsi/Fenwick are important because it shows events within the game that could become goals.

They are the "best we have" at the moment, but still only measure quantity.

Goal diff (5v5 GF%) is still the end result we are looking for and more valuable IMO.

 

[hatterson]

They are the "best we have" at the moment, but still only measure quantity.

Goal diff (5v5 GF%) is still the end result we are looking for and more valuable IMO.

Except that 5v5 Fenwick for % has been proven to be a better indicator of future 5v5 GF% than current 5v5 GF%

 

[LaFontaine83]

you have to be very careful with these numbers.

They are facts, but the context of the facts still don't get included... injuries might affect these... morale... staff changes... all these "gut" feeling type of things go into this too.

Corsi is good, but it's not the end all, be all!

 

[LaFontaine83]

yes, might I add that the Corsi Close is a good indicator as to who is smart with the puck.

 

[Doctor No]

Corsi is good, but it's not the end all, be all!

No one (at least, no one who is doing this stuff in practice) is suggesting that Corsi should be used in isolation, or even that it's the best single predictor of anything.

 

[TheDevilMadeMe]

Except that 5v5 Fenwick for % has been proven to be a better indicator of future 5v5 GF% than current 5v5 GF%

Over single season sample sizes, yes. Doesn't looking at goal differential end up passing Fenwick/Corsi as a predictor over a longer sample size (at least multiple seasons)?

 

[TheDevilMadeMe]

No one (at least, no one who is doing this stuff in practice) is suggesting that Corsi should be used in isolation, or even that it's the best single predictor of anything.

But a lot of people on internet forums suggest this. In fact, a lot of hockey fans are first exposed to Corsi by other fans who do seem to think it is the be-all end-all. Which is where you get pushback.

 

[Doctor No]

But a lot of people on internet forums suggest this. In fact, a lot of hockey fans are first exposed to Corsi by other fans who do seem to think it is the be-all end-all. Which is where you get pushback.

A lot of people on internet forums suggest a lot of things.

People generally latch on to anything that supports their preconceived notions - if that happens to be Corsi, expect people to talk about how great it is.

It doesn't mean that we need to take them seriously.

 

[hatterson]

Over single season sample sizes, yes. Doesn't looking at goal differential end up passing Fenwick/Corsi as a predictor over a longer sample size (at least multiple seasons)?

I don't have the numbers in front of me, but IIRC the tipping point for shots v goals metrics on the team level is around the 80-100 game mark.

I haven't seen anything player specific in that regard, however my guess (not based on data, just on hunch) is that it would be longer since it's easier for a single player to ride a hot or cold streak than for an entire team to do it.

 

[wgknestrick]

Over single season sample sizes, yes. Doesn't looking at goal differential end up passing Fenwick/Corsi as a predictor over a longer sample size (at least multiple seasons)?

Tell me which 3 year list shows the best forwards better?

GF%
http://stats.hockeyanalysis.com/rat...000&teamid=0&type=goals&sort=PCT&sortdir=DESC


FF%
http://stats.hockeyanalysis.com/rat...0&teamid=0&type=fenwick&sort=PCT&sortdir=DESC

Just to pick a very important example, Crosby is #3 in GF% and #24 in FF% .

On the other side, Justin Williams is #2 in FF%, but only #29 in GF%.

Which list tells you the proper conclusion to a simple "Who is better question"? Crosby or Williams?

While not running a correlation study between the 2 personally, it only takes logic to conclude that while FF% is a great indicator when you have little history on players, it obviously does not contain "all the variables" that go into the equation (like GF% does) for large samples. Anytime you ignore shooting %, you are throwing away a significant piece of data for some players (the ones at the tail ends of the SH% population).

Just take the past 5years of SH% sorted among forwards. Please try to tell me that this data is something that should be ignored. It's almost as good as FF% (IMO) in sorting the players from best to worst. There is significant value in SH% when evaluating players.
http://stats.hockeyanalysis.com/rat...0&teamid=0&type=goals&sort=ShPct&sortdir=DESC

 

[wgknestrick]

Except that 5v5 Fenwick for % has been proven to be a better indicator of future 5v5 GF% than current 5v5 GF%

This is dependent on sample size. We have many events to track when looking at FF%, while very few when looking at GF%.

Remember that you only get goal data for roughly every 10-12 shots so your sample size is effectively 1/12th as large. As you include more data the indicators diverge with GF% becoming more significant than FF% because it includes another variable.

I am definitely against the "SH% doesn't matter", isn't a skill, is not repeatable, is purely luck, etc, crowd. SH% is like anything else. It follows a natural distribution (bell curve) and for most players, they probably have a small variation between them. This has lead many an analyst to come to the wrong conclusion because the players in their evaluation may have been similar enough in SH% that it could be ignored.

For the players on the tails of the SH% curve though......SH% is much more significant. Anytime you just use FF% or C%, you are assuming that all the players being evaluated have a league average SH%. We all know what happens when you assume. Can you tell if the players being evaluated are "a Crosby" or "a Gomez" with respect to SH%? A shot with Crosby on the ice is twice as likely to become a goal.

 

[BeardyCanuck03]

Over single season sample sizes, yes. Doesn't looking at goal differential end up passing Fenwick/Corsi as a predictor over a longer sample size (at least multiple seasons)?

The problem with using multiple seasons is that there are too many variables that change season to season. Coaches, players, injuries, etc. With all that it's hard to really say that multiple season sample sizes are going to create accurate results to be predicting off of.

There will never be a be all end all stat. Each stat is just a measure that can shed light on different parts of a players game. No matter how good a stat is, there will be outliers that can be used to argue against the stat too. Which again brings me to the biggest part of analytics, it's not about accumulating the stats, it's about analyzing them and understanding why they are happening and how changes in personel and tactics can change them.

 

[Hivemind]

I don't think anyone has ever argued that individual shooting % is not a skill. However, individual SH% isn't a real corollary with Fenwick or Corsi, since individual SH% only measure the shot that player took rather than the shots his team generated while he was on the ice. On-ice shooting % is the more contention issue about it's viability as a predictor of future success. There seems to be some degree of sustainability, however it tends to be less than FF%.

http://hockeyanalysis.com/2012/04/19/on-ice-shooting-percentage-is-sustainable/
http://www.boysonthebus.com/2013/07/18/can-players-sustain-on-ice-shooting-percentages/

 

[CDN24]

Corsi and Fenwick numbers measure possession. If a team has more possession, they are likely generating more scoring chances. It doesn't always correlate in one game or in a handful of games, but the larger the sample size, the more it directly correlates. When you are blocking shots, you are defending. You don't have the puck and arent creating any scoring chances. Its a skill to get in the lanes and block shots, but at the same time you are getting outpossessed when you are forced to block attempts.

I am a firm believer that he who has the puck more wins more. ie posession. Corsi/fenwick attempt to measure posession and are probably a fair approximation of posession but they are a long way from perfect. A lot will depend on coaches systems or team systems. I would be curious to see Corsi numbers on some of the old Red army vs NHL games of the 70's. Those russian teams had the puck for significant portions of the game but did not take or attemt a whole lot of shots, preferring to keep possession rather than attemt a low % shot.

The 3rd line checking specialists of old (carbonneau - Lehtinen -gainey) probably had brutal corsi numbers as they spent all night chasing the top players. Their role of limiting damage probably resulted in brutal possession numbers by any measure but the contributed to team success significantly. yeah QOC attempts to quantify this but it too is based on a less than perfect Corsi stat. Corsi at the end of the day is a glorified plus minus stat so if we are to be fans of advanced stats less not bash plus minus too much.

The inherent weakness in Corsi Fenwick as a measure of possession is an event that likely results in loss of posession ( a shot attempt) improves the measure and an event that may result in your team regaining possesion (blocked shots) are percieved as a bad thing.

I would be curious to see if anyone has looked at hits or hit attempts to measure posession or lack thereof. Only the team/player with the puck can get hit. Look at any box score and the team who got dominated usually has more hits as the other team had the puck all night.

 

[Ogopogo*]

The truth about Corsi:

http://blogs.edmontonjournal.com/category/cult-of-hockey/

 

[Doctor No]

The truth about Corsi:

http://blogs.edmontonjournal.com/category/cult-of-hockey/

Wait, so the "truth" about Corsi is that Dallas Eakins doesn't use it to evaluate players?

That's the "truth"?

And you found this so profound that you posted this (with no additional explanation or justification) in two separate threads?

 

[TheDevilMadeMe]

I am a firm believer that he who has the puck more wins more. ie posession. Corsi/fenwick attempt to measure posession and are probably a fair approximation of posession but they are a long way from perfect. A lot will depend on coaches systems or team systems. I would be curious to see Corsi numbers on some of the old Red army vs NHL games of the 70's. Those russian teams had the puck for significant portions of the game but did not take or attemt a whole lot of shots, preferring to keep possession rather than attemt a low % shot.

The 3rd line checking specialists of old (carbonneau - Lehtinen -gainey) probably had brutal corsi numbers as they spent all night chasing the top players. Their role of limiting damage probably resulted in brutal possession numbers by any measure but the contributed to team success significantly. yeah QOC attempts to quantify this but it too is based on a less than perfect Corsi stat. Corsi at the end of the day is a glorified plus minus stat so if we are to be fans of advanced stats less not bash plus minus too much.

The inherent weakness in Corsi Fenwick as a measure of possession is an event that likely results in loss of posession ( a shot attempt) improves the measure and an event that may result in your team regaining possesion (blocked shots) are percieved as a bad thing.

I would be curious to see if anyone has looked at hits or hit attempts to measure posession or lack thereof. Only the team/player with the puck can get hit. Look at any box score and the team who got dominated usually has more hits as the other team had the puck all night.

The 80s Oilers too - were often outshot by the other team, but consistently put up incredibly high team shooting percentages. Especially when Gretzky was on the ice.

 

[Slats432]

The 80s Oilers too - were often outshot by the other team, but consistently put up incredibly high team shooting percentages. Especially when Gretzky was on the ice.

Don't think this is accurate. In 1984-85 and 1985-86 the Oilers were 3rd in the league for shots for.(Behind Calgary and Philly) I didn't go further because I figured the results would be the same.

 

[BraveCanadian]

Don't think this is accurate. In 1984-85 and 1985-86 the Oilers were 3rd in the league for shots for.(Behind Calgary and Philly) I didn't go further because I figured the results would be the same.

Looking shots against in 85-86, the Oilers gave up more shots than the last place 40 point Dead Wings who had 415 goals against. At least that is what the goaltending totals say..

 

[Slats432]

Looking shots against in 85-86, the Oilers gave up more shots than the last place 40 point Dead Wings who had 415 goals against. At least that is what the goaltending totals say..

How many did they give up...I couldn't find SA.