Has countering puck possession been successfully figured out?

[CCM19]

When "corsi" became a buzz word in hockey, there was definitely a revolution in the types of players teams would employ. Gone were the lumbering defensive defensemen who couldn't move the puck - it was adapt or get out. Gone were many floating goal scorers who let their linemates do all the work.

But it seems that we've hit a point where the league has "homogenized" in those aspects. In 2008, the best possession team in the league had a CF% of 58.84% and the worst had 42.85%. This was about a 16% swing from top to bottom and nearly 9% swing from top to middle - which makes sense because some teams were oblivious to how they were handicapping their rosters. This year, the best team is at 53.84% and the worst at 45.71%. That cuts those previous numbers in half. What this also means is that teams have isolated the pure handicap players and these possession numbers seem more to do with coaching styles than anything. Players who can drive possession are still important - but it isn't an advantage because everyone has them on most lines.

If you look at the last two years and there just doesn't seem to be any real advantage to being a so-called possession team.

In last year's playoffs:

- Pittsburgh had a playoff CF% of 47.23% yet won the cup. Without Kris Letang they just had no puck possession but countered their way through.
- St. Louis out defeated the wild 4-1 by being outcorsied 39-61
- Ottawa had a playoff CF% of 48.45% and a regular season CF% of 48.35%, yet made the ECF, toppling the league's corsi leading Bruins en route
- NYR had a first round CF% of 47.77% and a regular season CF% of 47.96% yet advanced to the second round
- Edmonton had a playoff CF% of 48.03$ yet made the second round and even won some games there.
- Anaheim was outcorsied 49-51 yet converted that into a series sweep


Okay, so those are ALL small sample sizes. This year seems to be more of the same carrying right through into the regular season though. Carolina, Calgary, Chicago, and Dallas are the #2,#3, #4, and #9 possession teams in the NHL yet all are about to miss the playoffs with similar issues of being unable to score. What really stands out is that it seems rush scoring is far more dependable these days with the emphasis on pure speed, while all these teams appear overly focused on cycle-based "half-court" offense.

I'm not saying possession is irrelevant because it clearly is not either in terms of overall correlation to ES goal differentials, but I am wondering if it's been successfully countered by possession-parity and coaching strategy.

How come you don’t put this much effort in your predictions?

 

[Bounces R Way]

The opposite is true if you look at the recent history of the Stanley Cup.

Lundqvist, Luongo, Bobrovsky, and Price put together - 0 Cups

Top 5 corsi teams - 7 Cups since the lockout and 3 in the last 5 years.

The team with the better SV% won every playoff series last year except 2(both the Oiler series). The team with the more shots put on net won their series about a 1/4 of the time.

Especially in a shorter series goaltending is the bigger difference maker. Corsi is fine as a predictor of future success when looked at over a large sample size, but as the OP illustrates it's far from perfect in that regard as well.

The way defense is played currently invites a lot of poor quality shots. Teams just collapse in front of their nets and protect the high danger areas by getting their sticks and bodies into shooting and passing lanes. They clog up the neutral zone minimizing the chance of a scoring rush or what I would deem as meaningful possession. Defenders are fine giving up a shot attempt when it's most likely going off their stick into the netting.

Advanced stats are still in their relative infancy, just how vital they are and to what degree they should be prioritized is still very much in flux.

 

[Machinehead]

The team with the better SV% won every playoff series last year except 2(both the Oiler series). The team with the more shots put on net won their series about a 1/4 of the time.

Especially in a shorter series goaltending is the bigger difference maker. Corsi is fine as a predictor of future success when looked at over a large sample size, but as the OP illustrates it's far from perfect in that regard as well.

The way defense is played currently invites a lot of poor quality shots. Teams just collapse in front of their nets and protect the high danger areas by getting their sticks and bodies into shooting and passing lanes. They clog up the neutral zone minimizing the chance of a scoring rush or what I would deem as meaningful possession. Defenders are fine giving up a shot attempt when it's most likely going off their stick into the netting.

Advanced stats are still in their relative infancy, just how vital they are and to what degree they should be prioritized is still very much in flux.

Goaltending is only the better indicator over small samples. The better possession team is more likely to advance through all four rounds.

I've given you the data. You can't deny that elite possession teams have several more Cups than elite goaltenders.

Yes, defense is played that way, but it doesn't exactly work when you consider how many goals in the NHL today are deflected.

I can buy the argument that individual analytics have some developing to do. Team analytics are pretty cut and dry. Good teams outshoot their opponents. Successful teams outshoot their opponents. We've known this for a century.

 

[NickH8]

Over the past several years my stance has put me on both sides of the corsi argument (depending on who was grinding the axe). Here is that stance:



As OP pointed out the range in performance in the bolded as narrowed, primarily due to teams being aware of its importance and how how to measure it. A and K have always been understood, now the emphasis/ grey area/ differentiation is on B&J. The game evolves as we understand it better.

Could we somehow figure out quality scoring chances using the mean distance and angle of shots?

 

[4thline]

Could we somehow figure out quality scoring chances using the mean distance and angle of shots?

That's one way, another has been undertaken by The Passing Project where they have found that one of the greatest quality factors is the play leading up to the shot ie, pass from behind the net, pass across the royal road, etc. It's the kind of thing that teams are doing that makes the whole "which model that can be spreadsheeted in people's spare time from publicly available data is best" argument kind of funny

 

[JoeThorntonsRooster]

The top 4 teams in Corsi last year missed the playoffs as well. Also 7 times in 12 years is not exactly that convincing.

Like I said I agree 100% with you philosophically. I just don't agree with focusing on the Corsi standings because they're way too tight, and there's too much fluctuation in them throughout the season.

Focusing on the Corsi standings isn't the way to go - totally agreed. If #4 and #7 CF% meet up in the first round and their rosters are totally identical except #4 CF% has Joonas Donskoi (elite CF% RW who can't crack 40 points) and #7 CF% has Patrick Kane, and their CF% is separated by 0.2%, I'm putting my money on the #7 CF% team, no question.

What CF% does is it gives you the green light to discount teams that are near the bottom of the league as Cup contenders, and it also gives you the green light to discount teams near the middle of the league unless they have elite ice tilters that can defy the laws of PDO, like Crosby/Malkin/Kessel in 2017 and Tim Thomas in 2011.

 

[JoeThorntonsRooster]

Could we somehow figure out quality scoring chances using the mean distance and angle of shots?

I don't know exactly how HDCF% and xGF% are calculated, but those do that to a certain degree.

 

[Bounces R Way]

Goaltending is only the better indicator over small samples. The better possession team is more likely to advance through all four rounds.

3 of the top 5 and 6 of the top 15 CF% teams aren't making the playoffs. Meanwhile 9/10 of the top SV% goalies(min 40 GP) are probably in.

 

[Machinehead]

3 of the top 5 and 6 of the top 15 CF% teams aren't making the playoffs. Meanwhile 9/10 of the top SV% goalies(min 40 GP) are probably in.

And when they get in, we'll see what happens to teams like Minnesota, Anaheim, LA, and Toronto who are relying way too much on their goaltending, like we always do. There's almost no evidence of a goaltender taking an otherwise bad team to the Stanley Cup. Tim Thomas is the only Vezina winner with a Stanley Cup as a starter since the lockout. Seven top 5 CF% teams have Cups since then.

 

[Machinehead]

Focusing on the Corsi standings isn't the way to go - totally agreed. If #4 and #7 CF% meet up in the first round and their rosters are totally identical except #4 CF% has Joonas Donskoi (elite CF% RW who can't crack 40 points) and #7 CF% has Patrick Kane, and their CF% is separated by 0.2%, I'm putting my money on the #7 CF% team, no question.

What CF% does is it gives you the green light to discount teams that are near the bottom of the league as Cup contenders, and it also gives you the green light to discount teams near the middle of the league unless they have elite ice tilters that can defy the laws of PDO, like Crosby/Malkin/Kessel in 2017 and Tim Thomas in 2011.

Exactly this.

 

[Finlandia WOAT]

I don't think there is a single team in the League that plays trying to game corsi.

 

[SladeWilson23]

Focusing on the Corsi standings isn't the way to go - totally agreed. If #4 and #7 CF% meet up in the first round and their rosters are totally identical except #4 CF% has Joonas Donskoi (elite CF% RW who can't crack 40 points) and #7 CF% has Patrick Kane, and their CF% is separated by 0.2%, I'm putting my money on the #7 CF% team, no question.

What CF% does is it gives you the green light to discount teams that are near the bottom of the league as Cup contenders, and it also gives you the green light to discount teams near the middle of the league unless they have elite ice tilters that can defy the laws of PDO, like Crosby/Malkin/Kessel in 2017 and Tim Thomas in 2011.

Totally agreed.

Even with the Pens having 3 top 10 forwards, and Tim Thomas being god in 2011, I sure as hell don't think it's wise for any GM or coach to try and replicate that and thinking that it's a secret formula to winning Cups. Building a team's foundation around Corsi is the most reliable way to build a SC contending team.

Carolina seems to love proving me wrong though.

 

[JoeThorntonsRooster]

Totally agreed.

Even with the Pens having 3 top 10 forwards, and Tim Thomas being god in 2011, I sure as hell don't think it's wise for any GM or coach to try and replicate that and thinking that it's a secret formula to winning Cups. Building a team's foundation around Corsi is the most reliable way to build a SC contending team.

Carolina seems to love proving me wrong though.

I mean, building around Corsi probably isn't the way to go either. But I think you should always be looking at acquiring good hockey players - if one of them is a very bad Corsi player, and clearly drags their teammates down, you should probably stay away. If you notice a guy like (yes, I'm being a homer) Joonas Donskoi, who is at over 55 CF%, and known stars like Joe Pavelski seem to rely on him to maintain good CF% after checking WOWYs, it's probably a good idea to take note of that player and say "if he does ever become available, I'd be willing to pay a higher price than other GMs, given the fact that he is an elite possession guy".

Corsi doesn't take goaltending into account though. I think Carolina would be a playoff team if they had league average goaltending. I think LA over the last 3 years is a bigger indictment of CF% but then they won 2 cups in 3 years before they started sucking. They're also the biggest endorsement though with their two SCF wins as #1 CF%

 

[me2]

Goodhart's Law

"Goodhart's law is an adage named after economist Charles Goodhart, which has been phrased by Marilyn Strathern as: "When a measure becomes a target, it ceases to be a good measure."[1] One way in which this can occur is individuals trying to anticipate the effect of a policy and then taking actions which alter its outcome."

Seems like teams started 'gaming' corsi when it's 'value' was revealved.

I don't think Corsi (at best a very basic stat) is now useless but is and always was a bit limited. If you abandon possession metrics (and there is more to possession than raw numbers like corsi) you still put yourself at risk. Most teams are trying to build their own metrics, much more advanced than corsi. Even still corsi has some use, at least to us fans. Ditching possession metrics because they are in vogue doesn't make sense.

Example

If turbo chargers make one team's race cars faster and they go from losing badly to winning a lot, soon the other teams will start adding turbos. When 100% of cars are using turbos, whether a car has a turbo or not loses it's predictive power. Does that mean turbos are useless and the first teams should stop using them? Of course it doesn't.

 

[puckIuck]

teams that have a high CF% and their position in the standings doesn't reflect it means they lack finishers or have suspect goaltending.

 

[Machinehead]

I mean, building around Corsi probably isn't the way to go either. But I think you should always be looking at acquiring good hockey players - if one of them is a very bad Corsi player, and clearly drags their teammates down, you should probably stay away. If you notice a guy like (yes, I'm being a homer) Joonas Donskoi, who is at over 55 CF%, and known stars like Joe Pavelski seem to rely on him to maintain good CF% after checking WOWYs, it's probably a good idea to take note of that player and say "if he does ever become available, I'd be willing to pay a higher price than other GMs, given the fact that he is an elite possession guy".

Corsi doesn't take goaltending into account though. I think Carolina would be a playoff team if they had league average goaltending. I think LA over the last 3 years is a bigger indictment of CF% but then they won 2 cups in 3 years before they started sucking. They're also the biggest endorsement though with their two SCF wins as #1 CF%

Also, people say it never happens but I think Daryl Sutter legitimately reached a point where he was gaming corsi.

 

[Bjornar Moxnes]

The Flames are one of the top Corsi teams this year. Tells you about all you need to know about the stat.

Their high danger chances are solid too. Advanced-stats-wise the Flames should easily be a playoff team. Fact is they’re not even close. I’ve pretty much jumped off the advanced stats bandwagon at this point.

Or maybe because the Flames legitimately have not been lucky and had to deal with extreme inconsistency?

 

[Bjornar Moxnes]

Focusing on the Corsi standings isn't the way to go - totally agreed. If #4 and #7 CF% meet up in the first round and their rosters are totally identical except #4 CF% has Joonas Donskoi (elite CF% RW who can't crack 40 points) and #7 CF% has Patrick Kane, and their CF% is separated by 0.2%, I'm putting my money on the #7 CF% team, no question.

What CF% does is it gives you the green light to discount teams that are near the bottom of the league as Cup contenders, and it also gives you the green light to discount teams near the middle of the league unless they have elite ice tilters that can defy the laws of PDO, like Crosby/Malkin/Kessel in 2017 and Tim Thomas in 2011.

Yup, it's the same with relative corsi on two teams with similar corsi. Let's say team A has a CF% of 52.3, and team B has a corsi of 52.5. Team A has a DMan that plays 19 mins on average and faces 45% of elite comp. Team B has a Dman that plays 18 mins on average and faces 28% of elite comp. Dman A takes 45% Ozone starts while his team takes 47%. Dman B takes 58% of Ozone starts while his team takes 53%. Dman A has a CF% Rel of 2%. Dman B has a CF% Rel of 2.5%. In this scenario I'd still easily take Dman A even though Dman B has higher raw and relative corsi. But if Dman A had a relative corsi of -5%, give me Dman B easily. That's why while quality of competition and zone starts matter, if the differences are drastic, the Dman who is destroying his comp is still better in my opinion.

 

[CanadianPensFan1]

I'd say the key to killing puck possession for your opponent is.. to have the puck more than them :p

 

[JoeThorntonsRooster]

Also, people say it never happens but I think Daryl Sutter legitimately reached a point where he was gaming corsi.

I don't know about that...I'm sure he was aware of how his team was dominating shot attempts and tried to put a system in place to sustain that, but I can't imagine him gaming Corsi.

Yup, it's the same with relative corsi on two teams with similar corsi. Let's say team A has a CF% of 52.3, and team B has a corsi of 52.5. Team A has a DMan that plays 19 mins on average and faces 45% of elite comp. Team B has a Dman that plays 18 mins on average and faces 28% of elite comp. Dman A takes 45% Ozone starts while his team takes 47%. Dman B takes 58% of Ozone starts while his team takes 53%. Dman A has a CF% Rel of 2%. Dman B has a CF% Rel of 2.5%. In this scenario I'd still easily take Dman A even though Dman B has higher raw and relative corsi. But if Dman A had a relative corsi of -5%, give me Dman B easily. That's why while quality of competition and zone starts matter, if the differences are drastic, the Dman who is destroying his comp is still better in my opinion.

Right, there are a lot more things to consider than just raw CF%. I knew Evander Kane was going to be an elite possession player as soon as he came to San Jose, even though his raw CF% was just a hair under 50%. I also took it with a massive grain of salt when Buffalo fans said that he was a volume shooter whose great possession numbers were a case of him shooting from anywhere, because his HDCF% and xGF% numbers were beast. I checked his WOWYs and saw that he didn't appear to rely on Jack Eichel all that much - the stats actually said that Eichel was more reliant on Kane. These numbers are like +/-, you have to consider the teammates. Kane had the 3rd best CF% on Buffalo, best HDCF%, and he played by far the most minutes of any forward - he also had the most individual HDCF and was on the ice for the most HDCF as a whole, despite the fact that he played forward. His relative possession numbers were all top-3 on his team and in a lot of them he was the best on the team.

I don't mean to disrespect Buffalo fans, but after I looked at these numbers, I already knew their opinions of him couldn't be entirely accurate. There's no way that a guy who is on the ice for the most high danger chances for, created the most high danger chances, was on the ice for the highest percentage of high danger chances could not be a guy that just shoots from anywhere.

You pair these stats with the fact that the guy has actually scored 30 goals before, had 28 in 70 last year, and it didn't take a genius or a hockey expert to tell you that he was going to be great on a remotely decent team San Jose.

 

[lomiller1]

The team with the better SV% won every playoff series last year except 2

Even a single goal moves the needle pretty far on sv%, so a team that is loosing invariable has a lower sv% than one that is winning when you look at games that have already been played. IOW in large part this result is a an effect of winning/loosing not a cause of winning/loosing.

The real test of statistical model comes when you try to use it to predict out of sample outcomes. IOW use past results to see which team is most likely to win in the future. It turns out that "out goaltending" the other team isn't a very reliable plan for winning. Swings in goaltending performance are too large. Even really good goaltenders can be outplayed by bad goaltenders on any given night.

 

[lomiller1]

Could we somehow figure out quality scoring chances using the mean distance and angle of shots?

That's part of the formula for xGoals (expected goals), xGoals also includes things like shot types rebounds and even captures some rush chances (which typically have higher shooting %). The methodology was described nearly a decade ago, the problem is that for all the additional complexity it doesn't actually work as well as Corsi. It wasn't a case a of people not thinking of this stuff, when it was looked at and shelved because the simpler metric worked better.

A couple years ago xGolas was revisited by Ryan Stimson and he added a couple things to it. Primarily he integrated historical data on who was taking the shot. His model does outperform Corsi, however he was hired by the Avs at the start of this season and his data has been removed. Some people may still be getting his data but AFAIk it's no longer public domain. The xGoals most people cite is from Corsica.hockey, but but while their model is improving AFAIk it still underperforms Score Adjusted Corsi.

 

[lomiller1]

I don't think there is a single team in the League that plays trying to game corsi.

Some pretty clear well poisoning going on here. How do you propose a team could even "game Corsi" to begin with? The notion that teams can make their Corsi "look better" by taking lost of outside shots was silly to begin with and has already been thoroughly debunked in this thread.

NHL teams are certainly paying attention to Corsi, the fact that there are no longer any dominate or terrible Corsi teams shows us that. They are doing this by fixing basic problems that cause bad Corsi, things like D-men who's go to move is off the glass and out because they can't make good players with the puck. This isn't "gaming Corsi" this is using what you learn from Corsi to improve your teams performance.

 

[BahlDeep]

I think Corsi is an overrated stat, not because of the stat itself, but by the way people use it to try proving a point.

 

[Mike Lowry]

NHL teams are certainly paying attention to Corsi, the fact that there are no longer any dominate or terrible Corsi teams shows us that.

But yet, we've had countless NHLers (players and coaches) come out and say they know nothing about corsy, it doesn't get talked about within the team, or that it's a terribly stupid stat.

These aren't Atom House League players who don't understand the game, and have parents as coaches. These are the best players in the world who spend nearly every day of the season immersed in the game. And yet none of them have come out and offered any comments that would support your theory.

Cosry has never been taken seriously by anyone with an ounce of experience or knowledge about the sport. It was made up to make the people who can't figure things out, believe they have an understanding. But you guys pushed it too much, so people pushed back, and the failure of the stat has been exposed. Please just let it die. It will go down as the worst idea that has been brought to hockey since the glow puck