Not to worry; this won't last says Myrtle

[Jay Random]

Originally Posted by Frequitude

Just ran the number for everything since the lockout and you are correct.

Correlation(Shots, Goals) = 0.84
Correlation(Shots, Shooting %) = -0.01

The former is obvious. The latter is surprising to me. There is virtually zero correlation between shots and shooting %.

Surprises me too. Unfortunately, I thought I could trust the source that told me there was a stronger negative correlation between shots and shooting %. Not having access to the necessary data, I couldn't run the numbers myself. My mistake.

[formulate]

Originally Posted by Frequitude

I haven't run any numbers along with watching games with a stopwatch, but my gut tells me those two effects net off and make shots attempted a pretty good proxy for possession across various styles of play.

This was done by JP Nikota at Pension Plan Puppets.

Measured the Leafs attack time with a stopwatch (only at tied situations, for some reason) and compared it to corsi%. Small sample size (twenty something games in tied situations), but looks like a pretty good surrogate.

Spoiler!

[_Q_]

Originally Posted by Street Pharmacist

Stats don't predict. They give probabilities. I know a lady who smoked until she died at age 93. Doesn't mean we throw out what we know about smoking and health

No you're wrong.

One of the first lessons in Probabilities & Statistics 101 is probabilities and statistics aren't the same thing.

For example:

"Smokers have a 1 in 10 chance of dying from cancer" is a probability (just pulled numbers out of nowhere to illustrate a point)

"1000 Canadians die each year from smoking related diseases" is a statistic (again random numbers)

These advanced stats are more akin to the last example than the first and are therefore, as their name suggests, statistics. Statistics are useful in painting a picture, but they by no means can be used to make a prediction.

[Street Pharmacist]

Originally Posted by _Q_

No you're wrong.

One of the first lessons in Probabilities & Statistics 101 is probabilities and statistics aren't the same thing.

For example:

"Smokers have a 1 in 10 chance of dying from cancer" is a probability (just pulled numbers out of nowhere to illustrate a point)

"1000 Canadians die each year from smoking related diseases" is a statistic (again random numbers)

These advanced stats are more akin to the last example than the first and are therefore, as their name suggests, statistics. Statistics are useful in painting a picture, but they by no means can be used to make a prediction.

Ok. Semantics then. My point is completely unaffected by your lesson.

These "advanced stats" (which is a stupid term, they're simple stats) can at most, point to what is likely to happen.

Again, nothing is on absolutes. Can a team with a bad corsi win and make the playoffs? Yes. Can a team keep a high PDO all year? Absolutely.

Are either of those likely? No.

Totally off topic, but holy man Principe looks gross with that mullet and stash

[Enoch Root]

Originally Posted by Regorium

I also apologize if this has been posted, but I think we should set the baseline as to why "advanced" stats are useful.

http://corsihockeyleague.com/2014/11...-the-playoffs/

If you have a score-adjusted Fenwick of >=50%, you have an 80% chance of making the playoffs. <=50% and it's 26%. The author seems to have double counted 50% making the two percentages not add up to 100.

That's a significant deviation from an expected random result of 53% chance of making the playoffs. A sample size of 210 (30 teams from the past 7 seasons) is more than enough to show a significant statistical deviation of greater than 25 percentage points.

It seems pretty obvious that higher possession scores correlate well with success. Why are people having a hard time accepting it?

No one has a hard time accepting that possession scores correlate with success. I don't need to see the numbers to know that they would be fairly strongly correlated.

What people have a hard time accepting is the conclusions some people draw. Too many people are too quick to say 'it won't last, look at the numbers: they have to regress'.

Nothing wrong with the numbers, it's shifting from observing correlation (and not a perfect one) to drawing black and white conclusions that gets people into trouble.

[opendoor]

Originally Posted by _Q_

But see that's just it. It's a surrogate. It's a projection of possession, but it in no way indicates possession.

Teams play different styles of game. To try to project their possession numbers using one metric (which in no way indicates possession, but rather attempts to make a correlation) is simply puzzling.

And last year, Pittsburgh, Montreal and Minnesota all had low Fenwick percentages while Vancouver, New Jersey and Florida were all above average. We all know how the first group did compared to the second group.

In light of the fact that there's no correlation between the volume of shots and a team's shooting percentage, why are you more interested in possession time than attempted shots? Possession time without producing shot attempts is pretty meaningless because it doesn't lead to offense whereas a higher volume of shot attempts does generally lead to more offense.

[SebC]

Originally Posted by Frequitude

Correlation(Shots, Goals) = 0.84
Correlation(Shots, Shooting %) = -0.01

Very interesting. There was one season that showed a pretty clear correlation here, but if this is everything since then the lockout then that was quite clearly an expection. I'm willing to discard "less shots = better shots" as a myth.

[stemit14]

Not sure if anyone agrees with me but a strong quality about the flames is the lack of panic in their own end. The flames are disciplined and well coached so they keep the play to the outside to limit prime scoring chances and, more importantly, they do a great job cleaning up rebounds to avoid 2nd or 3rd chances. They don't freak out cause the other team has possession in their zone... They just play smart, patient defence and don't get caught puck watching or over-commit to one guy. They use quickness, smart sticks in lanes, and shot blocking to frustrate the offence of the other teams. Meanwhile, the oilers panic and end up with 3,4, or even 5 guys trying to check one player on the boards.

Not sure if there are stats to track these things but I've watched every game this season and I will agree there are games where the other teams have more puck possession... But I would say the flames tend to create more quality scoring chances than their opponents. Especially in the last few weeks.

Even last nights game... the ducks played better than the flames in the pest two periods but it's not like they were dominating the flames with overwhelming offensive chances akin to the Chicago game. The ducks just played a good road game for the first two periods and limited scoring chances on both sides. For to he 3rd period and overtime however... Calgary dominated the play and could have scored 6 goals in that time frame with the quality chances get we're generating.

[Itse]

Originally Posted by Street Pharmacist

Stats don't predict. They give probabilities. I know a lady who smoked until she died at age 93. Doesn't mean we throw out what we know about smoking and health

False equivalency. Nobody has suggested shots on goal don't matter. Every team wants to stop shots on their goal and shoot the puck at the other goal. They have always done so.

Getting excited about team corsi and team fenwick is basically just people getting excited about the fact that OMG, you can count the correlation between shots and winning.

Of course you can. Question is, is the result interesting?

Regorium:

If you have a score-adjusted Fenwick of >=50%, you have an 80% chance of making the playoffs. <=50% and it's 26%. The author seems to have double counted 50% making the two percentages not add up to 100.

So no, the result is not interesting. You could probably have ballparked these numbers with an educated guess, based on the number of teams and the number of teams that make the playoffs.

Those numbers are not interesting because they're
1) not surprising
2) not predictive on a previously unseen scale.

Someone here suggested that people deride advanced stats because they "go against common knowledge". In this case it's exactly the opposite. People deride this stat because they just tell us what we know.



Of course all that said, there _are_ things I'd call advanced stats that I think are actually interesting. Individual player stats for example are much more interesting than team stats, because forming accurate opinions on players is really hard to begin with, and there are A LOT more players than teams. You could easily have a pretty good grasp of most teams in the NHL, but having educated opinions on even half of of the players is close to impossible.

Offensive zone starts for example is a a stat that interests me. (EDIT: Not really "advanced, just new. Mostly the same thing really.) It tells me more about the roles players play, and I've noticed that it tells me things I would not have known or guessed, even if we're just talking about the Flames, who I watch closely.

[kehatch]

Debating advanced stats with most posters is like debating the quality of a movie with your buddy that hasn't actually seen the movie.

[_Q_]

Originally Posted by Street Pharmacist

Ok. Semantics then. My point is completely unaffected by your lesson.

These "advanced stats" (which is a stupid term, they're simple stats) can at most, point to what is likely to happen.

Again, nothing is on absolutes. Can a team with a bad corsi win and make the playoffs? Yes. Can a team keep a high PDO all year? Absolutely.

Are either of those likely? No.

Totally off topic, but holy man Principe looks gross with that mullet and stash

Regarding the first part, it absolutely makes a difference what the definition of probability and statistics is and you do touch on it in the second part.

That is illustrated with the 80% of teams with a corsi of +50% make the playoffs study mentioned a couple of pages back.

Do the Flames need to have a corsi of +50% to make the playoffs? No

Will there be any teams that make the playoffs this season with a corsi of less than 50%? Well the probabilities say that 4 out of the 16 teams will.

Can the Flames be one of those teams? Yes.

There's nothing "unsustainable" about statistics. The fact that it's being done means that it can be done. That's what people don't get about stats. Stats are just numbers that quantify certain things they don't predict anything.

Originally Posted by opendoor

In light of the fact that there's no correlation between the volume of shots and a team's shooting percentage, why are you more interested in possession time than attempted shots? Possession time without producing shot attempts is pretty meaningless because it doesn't lead to offense whereas a higher volume of shot attempts does generally lead to more offense.

Because shot attempt stats like corsi and fenwick quantify shot attempts and not possession. Is there a correlation between shot attempts and possession? Yes. However, it's just a correlation based on some assumptions. The model to make that correlation can be broken and that's what makes such an assumption flawed.

For example say a team holds on to the puck for 95% of the time, taking only 10 shots on net the entire game and scoring on all 10 shot attempts. The opponents, during the 5% they have the puck always quickly get the puck to the red line and fire the puck at the general direction of the net. They end up with 20 shot attempts and 0 goals.

The losing team's fenwick and corsi numbers would be off the charts in this game, even though they barely controlled the puck.

While this example is extremely exaggerated, it's not an "impossible" scenario. Go watch a beer league game where one team should clearly be 3 divisions higher and you might see something similar to this.

Anyone that has watched the team this year will tell you that the Flames have been dominant in most games (other than last night and against Chicago), yet for some reason their Fenwick and Corsi numbers say otherwise. So why is it that the Flames can't break this model?

[Frequitude]

Originally Posted by Jay Random

Irrelevant. Fenwick numbers are expressed as a percentage of total shots taken, so if you're not getting credit for part of your possession time, you're going to get a skewed picture from what's left.

I do agree with you. Just saying that it's not entirely hurting you on one side and that the fact it prevents the opposition from shooting counteracts the effect somewhat.

[Frequitude]

Originally Posted by kehatch

Debating advanced stats with most posters is like debating the quality of a movie with your buddy that hasn't actually seen the movie.

That's a little high horse of you. A bunch of people are in here trying to understand them better by engaging in the conversation and learning in the process. I'm one of them.

[kehatch]

Originally Posted by Frequitude

That's a little high horse of you. A bunch of people are in here trying to understand them better by engaging in the conversation and learning in the process. I'm one of them.

It is not meant to be. Its just old.

On one side you have a bunch of people that have strong negative opinions about the subject without the smallest bit of knowledge. Its the internet era people. It isn't that hard to have an educated opinion. Ignorance isn't really defensible.

On the other side you have a group that seems to insist on a maintaining a barrier for entry. Presumably (and they would never admit this) so they can defend their false conclusions to the ignorant masses.

These aren't advanced stats. These aren't fancy stats. Its simple data digestible by the general public. But "they" put fancy terms like CORSI and FENWICK in the title to sound really smart. (How about calling it shot attempts ... gasp). There are dozens of derivatives of the metrics with almost no consistency in how they are used. Metrics like Quality of Competition and Team Mates are calculated differently all over the place with little consensus (What CORSI or QOC metric do people use .. the one that defends their conclusion the best ...) The entire system seems to be designed to be confusing when it really shouldn't be.

They also make false conclusions. Being on the ice for more shots attempts for then against is a good thing (duh) but that doesn't mean that the player with the highest CORSI is the best player. But many of the metrics rely on that conclusion to make any sense. PDO is supposed to be the luck metric. The conclusion being that a high SH% combined with a high SV% means your lucky. But that is just one possibility. Good teams finish above 1.0 all of the time. Plus, the entire think disregards special teams like that doesn't have a significant impact on the results.

I have no doubt that shot attempt based metrics will become a major part of our game. I have no doubt that better player tracking technology will result in new metrics like zone entries. Its information and you can't afford to disregard information. But I seriously doubt these "fancy" stats of tomorrow will resemble the "fancy" stats of today. The final product will be less opinionated, less convoluted, and more accurate.

[Enoch Root]

lol that you think everyone on one side of the argument is ignorant

[SebC]

Isn't a high PDO also known as having gamebreakers?

[PeteMoss]

Originally Posted by kehatch

It is not meant to be. Its just old.

On one side you have a bunch of people that have strong negative opinions about the subject without the smallest bit of knowledge. Its the internet era people. It isn't that hard to have an educated opinion. Ignorance isn't really defensible.

On the other side you have a group that seems to insist on a maintaining a barrier for entry. Presumably (and they would never admit this) so they can defend their false conclusions to the ignorant masses.

These aren't advanced stats. These aren't fancy stats. Its simple data digestible by the general public. But "they" put fancy terms like CORSI and FENWICK in the title to sound really smart. (How about calling it shot attempts ... gasp). There are dozens of derivatives of the metrics with almost no consistency in how they are used. Metrics like Quality of Competition and Team Mates are calculated differently all over the place with little consensus (What CORSI or QOC metric do people use .. the one that defends their conclusion the best ...) The entire system seems to be designed to be confusing when it really shouldn't be.

They also make false conclusions. Being on the ice for more shots attempts for then against is a good thing (duh) but that doesn't mean that the player with the highest CORSI is the best player. But many of the metrics rely on that conclusion to make any sense. PDO is supposed to be the luck metric. The conclusion being that a high SH% combined with a high SV% means your lucky. But that is just one possibility. Good teams finish above 1.0 all of the time. Plus, the entire think disregards special teams like that doesn't have a significant impact on the results.

I have no doubt that shot attempt based metrics will become a major part of our game. I have no doubt that better player tracking technology will result in new metrics like zone entries. Its information and you can't afford to disregard information. But I seriously doubt these "fancy" stats of tomorrow will resemble the "fancy" stats of today. The final product will be less opinionated, less convoluted, and more accurate.

I tend to relate the current state of advanced hockey stats to when baseball people started to realize that on base percentage was more important than batting average. It was such an obvious thing that anyone with a brain would know but sports teams for the longest time only had ex-players in the front office.

Some of the ex-players believe in what they grew up with (batting average/RBIs) and how they could just tell by seeing a guy swing how good a player would become.

It was such an obvious conclusion and there is a 0% chance that most of the teams weren't considering it but it was never spelled out in the public by the front offices or the media that when it started to come out online (and put into the mainstream by Moneyball) people thought they had cracked some code.

Baseball stats have advanced way beyond this now - defensive metrics, base running, park effects, etc to the point where they are useful beyond obvious conclusions.

Hockey stats will reach that point, but for now a lot of it is just obvious points given fancy names so people can't understand what they mean (like you say).

[PeteMoss]

In regard to the Flames. Yes - they need to play better to keep winning. But you're also looking at a small enough sample and Flames being very injury depleted that you can't say they've established themselves as a bad team getting lucky at this point.

And in regards to the Flames shooting percentage this year. It is high, but the Flames have been a fair bit better than average every year since Hartley became the coach with team's with similar to lower talent levels.

You're getting to large sample that shows that a higher than normal shooting percentage should be expected (although very likely not as high as it is now)

[corporatejay]

Just so I'm clear. Fen wick is basically just "shots at goal" is that right? If so, how can anyone not think that is a more useful stat than shots on goal.

[Street Pharmacist]

Originally Posted by corporatejay

Just so I'm clear. Fen wick is basically just "shots at goal" is that right? If so, how can anyone not think that is a more useful stat than shots on goal.

No, it's corsi minus shots blocked. It includes missed shots. May seem weird at first, but it's meant to correct corsi as shot blocking is a skill and not random