The value of Corsi and Fenwick

[Doctor No]

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

My data have them with 2776 non-empty net shots on goal, second most (to Chicago's 2793).

For Fuhr, he was facing 35.6 shots/60.

 

[Doctor No]

One thing that I haven't seen explicitly mentioned here is the difference between using Corsi as a descriptive measure (how good is this team?) and using Corsi as a prescriptive measure (how can I make this team better?).

Stated differently, there's a correlation between Corsi and winning, but it's likely that winning teams have good Corsi levels as a result (since they're outshooting their opponents).

If one were to take this and say that existing teams should shoot more, because it will make them a better team, that part could not follow. In other words, being good causes shot differentials, not the other way around - and this may be what Drew Doughty was trying to say when he called Corsi "crap" (he's got a gift for brevity). Players don't go out and try to have good Corsi rankings - they go out and try to win, and a result of that is good Corsi rankings.

Thoughts?

 

[Ogopogo*]

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?

The truth is: Corsi is worthless as a measure of individual performance. The article clearly states that, why should I re-type it?

 

[Doctor No]

The truth is: Corsi is worthless as a measure of individual performance. The article clearly states that, why should I re-type it?

The article states that Dallas Eakins believes that it's worthless, not that it is worthless.

Happy to help!

 

[Ogopogo*]

The article states that Dallas Eakins believes that it's worthless, not that it is worthless.

Happy to help!

Unfortunate you didn't understand the article.

 

[Doctor No]

Unfortunate you are unable to understand the article.

Fine. "And some blog author agrees with him."

And apparently you do, too.

This all assumes that I'm reading the correct article, since your link was not to any article at all, but to the main blog page. Or did I misunderstand that, too?

 

[Ogopogo*]

Fine. "And some blog author agrees with him."

And apparently you do, too.

This all assumes that I'm reading the correct article, since your link was not to any article at all, but to the main blog page. Or did I misunderstand that, too?

Keep your chin up. If you aren't able to understand this stuff, you're probably good at something else.

 

[Doctor No]

Keep your chin up. If you aren't able to understand this stuff, you're probably good at something else.

Thanks - I appreciate it.

Now I see why you disappeared from the thread in mid-September.

 

[BraveCanadian]

One thing that I haven't seen explicitly mentioned here is the difference between using Corsi as a descriptive measure (how good is this team?) and using Corsi as a prescriptive measure (how can I make this team better?).

Stated differently, there's a correlation between Corsi and winning, but it's likely that winning teams have good Corsi levels as a result (since they're outshooting their opponents).

If one were to take this and say that existing teams should shoot more, because it will make them a better team, that part could not follow. In other words, being good causes shot differentials, not the other way around - and this may be what Drew Doughty was trying to say when he called Corsi "crap" (he's got a gift for brevity). Players don't go out and try to have good Corsi rankings - they go out and try to win, and a result of that is good Corsi rankings.

Thoughts?

I think there is a lot of truth to this..

 

[The Mars Volchenkov]

I know the Avs had low possession numbers in relation to Corsi and Fenwick, but this video has some interesting possession stats for the Avs. It starts like halfway through.

[nhl]637723[/nhl]

 

[Rogie]

One thing that I haven't seen explicitly mentioned here is the difference between using Corsi as a descriptive measure (how good is this team?) and using Corsi as a prescriptive measure (how can I make this team better?).

Stated differently, there's a correlation between Corsi and winning, but it's likely that winning teams have good Corsi levels as a result (since they're outshooting their opponents).

If one were to take this and say that existing teams should shoot more, because it will make them a better team, that part could not follow. In other words, being good causes shot differentials, not the other way around - and this may be what Drew Doughty was trying to say when he called Corsi "crap" (he's got a gift for brevity). Players don't go out and try to have good Corsi rankings - they go out and try to win, and a result of that is good Corsi rankings.

Thoughts?

Maybe it can help in determining who plays on what lines and which players might be more effective together. All players have tendencies of course, some players hold the puck better, some are distributors, etc etc (you know all that stuff). So, analytic's guys, especially, if they are also hockey guys, might gain an edge by putting better combinations of players together, as well as then playing them against the other teams lines (as much as line matching can permit).

So, even if Drew Doughty doesn't think much about Corsi stuff, it still is possible the reason the Kings are such a good possession team is more than just because they are first and foremost simply a good team. Maybe that is a large part of it, maybe, most of it. But, maybe it's also due to how they put their lines together. Maybe management, in consultation with stats people convince Sutter, hey, you've got to keep trying these guys together, give it a chance. Sutter still decides who plays with whom, but, maybe there is a lot more discussion and convincing reasons that are based (just partly) on some of the stats. Also, as teams get better at identifying what it is that good possession players do better and better, they can look for those players for specific positions they need filled.

Matt Frattin comes to mind. He had (relatively) good Corsi numbers with a terrible possession team - the Leafs. He hits hard, skates hard, shoots hards, can play along the boards, etc, seems like the making of a good possession player. Kings tried him out, gave him some chances, even some opportunities with offensive players. In the end, the Kings moved on from Frattin, but, I remember them saying in a presser, that they had been trying to get Frattin for a while.

My point is that they were honing on a type of player - at least to me Frattin looked like the possession type player that would fit with them. It didn't turn out that way. The guys that were competing against Frattin (I remember he was in and out of the line-up alternating with 2 or 3 players) beat him out.

Stats guys might find tendencies in players, and often times players have tendencies that really aren't easily seen by the coaches or by simply watching the game. There is such an incredibly small difference between winning and losing. After 82 games, and how many rounds of playoffs, then, often, it comes down to the 7th and final game AND even to overtime. So, the small tendencies and little differences here and there when added together maybe can make a difference.

And the chicken and egg question is a fair question. It's just all part of learning how to figure out the game better. So, okay, we're a good team (the Kings), WHY? How did we keep possession, why did we out possess the other team? And is that the reason that analytics has entered the game, because, it you want to know how to get better, you have to know what makes you better. Analytic's is going to allow a much much deeper look into the game, especially as the technology advances. Just getting the best players isn't as easy with salary caps and the player pool being diluted. Aside from occasional superstars that come along, players are very closely matched in talent, that it might not even be possible to buy the best team even if there was no salary cap.

 

[hatterson]

One thing that I haven't seen explicitly mentioned here is the difference between using Corsi as a descriptive measure (how good is this team?) and using Corsi as a prescriptive measure (how can I make this team better?).

Stated differently, there's a correlation between Corsi and winning, but it's likely that winning teams have good Corsi levels as a result (since they're outshooting their opponents).

If one were to take this and say that existing teams should shoot more, because it will make them a better team, that part could not follow. In other words, being good causes shot differentials, not the other way around - and this may be what Drew Doughty was trying to say when he called Corsi "crap" (he's got a gift for brevity). Players don't go out and try to have good Corsi rankings - they go out and try to win, and a result of that is good Corsi rankings.

Thoughts?

I said something similar in that Doughty thread on the main boards which seems to mirror your thoughts.

How Doughty feels about corsi is irrelevant, it's not a concern of his as a player.

Players don't get better by heading out on a shift thinking "man I'm gonna pump up my corsi number" You can't become a more effective player by just chucking pucks at the net.

It's been shown time and time again that corsi does not cause good play, good play results in good corsi numbers (on average). Given that Doughty's job is to play well, not to worry about player personnel decisions, corsi should be irrelevant to him.

And as rogie said, I believe that the role of analytics in the day to day operation of the team is more along the lines of coaching decisions. Perhaps assisting with potential line combinations or who sits in the press box. I can also see analytics being used to determine what areas of the game a team needs work on. Perhaps you can see that one of the reasons your team is performing so poorly is because you're allowing the other team to enter the zone very easy compared to other teams, or perhaps you're rarely gaining offensive zone possession because your dump and chase game is off balance.

Instead of just hearing the tired old adages of "win the battles" or "up the compete level" you may instead realize that you'll have much more success by carrying the puck in versus a dump and chase strategy.

Now, as many have said, I don't believe that analytics (of which corsi is only a small piece) are a replacement for traditional coaching, GMing and scouting. You'll never get a spreadsheet to tell you how to coach, but it can provide additional insight and help to hone those disciplines.

 

[Hivemind]

Doctor No's thoughts are pretty spot on. Firing a bunch of shots from the blue line isn't going to make you a better player or better team. However, success and corsi correlate well. Good teams tend to have better possession metrics.

Doctor No is also spot on about Norm Ullman's appeal to authority. My favorite irony here is that Eakins just hired Tyler Dellow to work with Edmonton's coaching staff.

 

[Doctor No]

I said something similar in that Doughty thread on the main boards which seems to mirror your thoughts.

And as rogie said, I believe that the role of analytics in the day to day operation of the team is more along the lines of coaching decisions. Perhaps assisting with potential line combinations or who sits in the press box. I can also see analytics being used to determine what areas of the game a team needs work on. Perhaps you can see that one of the reasons your team is performing so poorly is because you're allowing the other team to enter the zone very easy compared to other teams, or perhaps you're rarely gaining offensive zone possession because your dump and chase game is off balance.

Instead of just hearing the tired old adages of "win the battles" or "up the compete level" you may instead realize that you'll have much more success by carrying the puck in versus a dump and chase strategy.

Now, as many have said, I don't believe that analytics (of which corsi is only a small piece) are a replacement for traditional coaching, GMing and scouting. You'll never get a spreadsheet to tell you how to coach, but it can provide additional insight and help to hone those disciplines.

Figures I'd miss something of value in that thread.

Good call as always - and I especially agree on the last paragraph. There are a lot of folks in the debate who think that (for some reason) they can only choose "statistics and predictive analytics" or "traditional scouting and coaching". The answer is always both. Always choose both.

 

[Thesensation19*]

Its a very simple stat that you cant help but to enjoy... All teams should look at this.

See where your teams at and where it is heading by seeing if they are creating more opportunities than they allowing. See if the lines you put together, work. See if the players you have are doing what is needed.

No one should say that Corsi and Fenwick are the end of discussion for a players worth. Or a teams worth. A team can still win with low corsi. A player can still be effective and valuable for a team or their line or their role, with having a negative Corsi #.

Doenst mean it shouldnt be looked at either.
I mean, I for one still think quality of shots is important.

There are many games we all watch where a team has more shots than the opposing team. The game looks very close though and in the end the other team wins. Rangers vs Pens last playoffs - We lost a game at home but we hit 2 posts, and outshot the Pens by like 15 or 20 shots. Fans asked, why did we lose? How? Closer than score shows. Might have been 4-1 actually.

Anyway, I watched that game and the possession and the quality of shots favored the Pens. Easily. Yea, we shot more. But a lot of our shot count was from the point with no one in front, hardly a body around for a rebound either. We shot a lot from poor angles. We even shot a few from the neutral zone because we couldnt penetrate well so we just shot in on net, or dumped it in and got one lousy shot fluttering towards Fleury.

The Pens had only 15 shots less, but they still hammered in 3 or 4 goals off of like 25 or 30 shots. Why didnt Rangers score 3 or 4 and doing over 40 shots? Quality. Quality of shots comes from a quality of play. The system in place. Its why quick break outs, quick attack is so favored these days. It generates a lot of odd man advantages.

 

[TOML]

I know the Avs had low possession numbers in relation to Corsi and Fenwick, but this video has some interesting possession stats for the Avs. It starts like halfway through.

[nhl]637723[/nhl]

Interesting in that Duchene's quotes, for example, certainly devalue corsi events as a measure of success. ie. Context of shots taken is very important. Then they go on to show some play-driving moves made by their forwards in the highlights. Considering the Avs had a successful season, but their corsi numbers were weak, you have to consider other factors when measuring success.

Where's 5-on-5 defensive zone possession time per game measured? The Avs being ranked 5th there, behind LA, Chicago, Boston, St. Louis, it seems to be an interesting way to measure success, at least team-wise.

 

[Freudian]

The predictive value of Corsi and Fenwick is being exaggerated by the advanced stat community.

This whole off-season we have seen predictions that team x will do better/worse in 2014/15 because of their shot attempt based statistics in 2013/14. But the correlation between past shot attempt based statistics and future results is low.

Here is the correlation between some stats and the point total for a team the year after (2008/09 to 2013/14).

Correlation between ES close CF% and point total the year after: 0,35
Correlation between ES close FF% and point total the year after: 0,41
Correlation between point total the year before and the year after: 0,39
Correlation between GF% the year before and point total the year after: 0,41
Correlation between goal differential the year before and point total the year after: 0.4

As you can see, neither stat have all that great predictive power. You might as well use goal differential or point total in 2013/14 to predict how teams will do in 2014/15. They're as good at predicting the future as Fenwick is and better than Corsi is.

 

[charlie1]

The predictive value of Corsi and Fenwick is being exaggerated by the advanced stat community.

This whole off-season we have seen predictions that team x will do better/worse in 2014/15 because of their shot attempt based statistics in 2013/14. But the correlation between past shot attempt based statistics and future results is low.

Here is the correlation between some stats and the point total for a team the year after (2008/09 to 2013/14).

Correlation between ES close CF% and point total the year after: 0,35
Correlation between ES close FF% and point total the year after: 0,41
Correlation between point total the year before and the year after: 0,39
Correlation between GF% the year before and point total the year after: 0,41
Correlation between goal differential the year before and point total the year after: 0.4

As you can see, neither stat have all that great predictive power. You might as well use goal differential or point total in 2013/14 to predict how teams will do in 2014/15. They're as good at predicting the future as Fenwick is and better than Corsi is.

Looks like you're missing a digit on that last one.

Did you find correlation between goal differential the year before and point total the year after: 0.40?

 

[Freudian]

Looks like you're missing a digit on that last one.

Did you find correlation between goal differential the year before and point total the year after: 0.40?

0.4 and 0.40 is the same thing.

 

[badtakemachine]

Looks like you're missing a digit on that last one.

Did you find correlation between goal differential the year before and point total the year after: 0.40?

An admittedly nitpicky comment, but even if the correlation was 0.49999999, it would still be weak.

 

[charlie1]

Looks like you're missing a digit on that last one.

Did you find correlation between goal differential the year before and point total the year after: 0.40?

0.4 and 0.40 is the same thing.

Yes, but 0.41 and 0.40 are not the same thing. And since GF% and Goal Differential are perfectly correlated, their correlation with anything else should be exactly the same. But they're not, apparently.

 

[Freudian]

Yes, but 0.41 and 0.40 are not the same thing. And since GF% and Goal Differential are perfectly correlated, their correlation with anything else should be exactly the same. But they're not, apparently.

They aren't perfectly correlated. Take for example a team that scores 280 goals and allows 260 and one team that scores 200 goals and allows 180. Both have a goal differential of +20, but the GF% is different (51.9% and 52.6%).

 

[badtakemachine]

Yes, but 0.41 and 0.40 are not the same thing. And since GF% and Goal Differential are perfectly correlated, their correlation with anything else should be exactly the same. But they're not, apparently.

They are not perfectly correlated, but yes, they should be very closely correlated. Two teams can have the same goal differential, say, +20. But if one team does it by scoring 265 and allowing 245, and the next team does it by scoring 215 and allowing 195, the second team has the superior GF% by a non-insignificant margin at 52.4% to 51.9%. Of course they are very close, but that would be enough to cause the minor discrepancy you mentioned in comparing those numbers to win percentage.

Edit: What he said.

 

[charlie1]

They aren't perfectly correlated. Take for example a team that scores 280 goals and allows 260 and one team that scores 200 goals and allows 180. Both have a goal differential of +20, but the GF% is different (51.9% and 52.6%).

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

 

[badtakemachine]

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

gf% = gf / (gf + ga)