What Corsi really translates to (in numbers that are easy to grasp)

[lomiller1]

That means nothing.

If puck possession is considered to be a good thing, it must correlate to winning hockey games.


Here's a chart plotting wins against CF% for the 2015 season with a miserable .22 r-value.

All that means is that winning an individual hockey game is a highly random thing. This should be no surprise as most games will hinge on plays where a puck bounces over someone’s stick or doesn’t, catches the corner of the goaltenders blocker or doesn’t, catches the post and stays out or doesn’t.

Over larger sample the small edge adds up and 5v5 score adjusted Corsi becomes much more predictive. In a sense it’s like a casino game where the house has a small advantage on individual games but this adds up over time. The r^2 value of CF% over 20 games in predicting the other 60 is ~0.55 and this climes to about 0.6 at 30 games. This is quite a large number given that two huge parts of the game (goaltending and special teams) are not even included.

 

[zeke]

All that means is that winning an individual hockey game is a highly random thing. This should be no surprise as most games will hinge on plays where a puck bounces over someone’s stick or doesn’t, catches the corner of the goaltenders blocker or doesn’t, catches the post and stays out or doesn’t.

Over larger sample the small edge adds up and 5v5 score adjusted Corsi becomes much more predictive. In a sense it’s like a casino game where the house has a small advantage on individual games but this adds up over time. The r^2 value of CF% over 20 games in predicting the other 60 is ~0.55 and this climes to about 0.6 at 30 games. This is quite a large number given that two huge parts of the game (goaltending and special teams) are not even included.

thankyou

 

[Bourdon101]

All that means is that winning an individual hockey game is a highly random thing. This should be no surprise as most games will hinge on plays where a puck bounces over someone’s stick or doesn’t, catches the corner of the goaltenders blocker or doesn’t, catches the post and stays out or doesn’t.

Over larger sample the small edge adds up and 5v5 score adjusted Corsi becomes much more predictive. In a sense it’s like a casino game where the house has a small advantage on individual games but this adds up over time. The r^2 value of CF% over 20 games in predicting the other 60 is ~0.55 and this climes to about 0.6 at 30 games. This is quite a large number given that two huge parts of the game (goaltending and special teams) are not even included.

Im sorry but lol, what even are you plotting???

I understand you are plotting the r squared error of wins versus corsi but what is the dataset? And more importantly, why the hell would you show the dispersion of the data (r correlation-squared) instead of the actual correlation.

This means absolutely nothing, its hot air. The poster you quoted was right: He showed CF% has a negligable correlation to wins in the last few years. You are showing us the dispersion of the correlation (which is irrelevant if you dont know the correlation) of wins vs Corsi ON AN UNKNOWN DATASET.

 

[Machinehead]

What he said was pretty reasonable and you just dismiss it. From reading this thread you seem to just ignore everything that doesn't agree with your original idea. You also talk about "the advanced stat community" like they are a notch above everyone...

Theories and ideas evolve over time. If you continually dismiss attempts at debunking your theory then it will lag behind. Corsi isn't perfect and in fact it's use needs to evolve. Saying that anything is a closed case is just am awful way to go about research

If we're talking about zone starts, and you argue with me using something that isn't a zone start, I'm going to dismiss it. Sorry.

What he said might be reasonable and we can debate his theory, but his theory has absolutely nothing to do with what we were talking about.

I'm talking about zone starts. He's talking about something he literally made up.

 

[gorangers0525]

Everyone knows corsi, especially raw, non score-adjusted, corsi does not line up well with wins. It's correlation with long term success, and ultimate success, is where it holds value.

 

[Machinehead]

Everyone knows corsi, especially raw corsi, non score-adjusted, corsi does not line up well with wins. It's correlation with long term success, and ultimate success, is where it holds value.

The Kings are the poster boys for corsi and they've never been a particularly great regular season team. I don't think they care.

 

[Beukeboom Fan]

People need to stop saying this.

Corsi does operate under the assumption that players are playing the game legitimately, not just trying to rack up shots. However, it's a perfectly valid assumption.

Technically, a team could theoretically just shoot at EVERY opportunity and end up with a phenomenal Corsi, but lose almost every game. But they don't do that, because it would be idiotic. It's a red herring, because we all know full well nobody* would do that.

* Except Dallas Eakins

 

[6 Karlsson 5]

And it's hard to predict who is going to win the possession battle in a playoff series. If you have a 51% CF team going up against a 52% CF team in a series and there's no guarantee that the team with a higher CF% throughout the season does better in the series and it's far from a guarantee that the team with a higher CF% in the series will win the series. Of 135 playoff series since 07/08, 62% of the teams with the higher CF% during the season won the playoff series, while 57% of teams with the better special teams won.

I don't really disagree with any of this

And while that is true, whether the PP or PK is hot or cold certainly has a huge impact on the outcome of a 7 game series.
So basing any future outcome on past performance is a risky project.

how else would you base future outcomes?

 

[Osprey]

I get the feeling that this thread is less about making Corsi easy to grasp and more about making a case defending Fowler. If so, that's fine, but a more honest title would've been "What Fowler's Corsi really translates to," IMO.

There absolutely is no correlation. It's a blind assumption that there's a correlation.

Almost everyone in the advanced stats community universally agrees on this at this point, but yet again, people are trying to educate the advanced stats community on how to use advanced stats.

If the "advanced stats community" doesn't understand the distinction between correlation and causation, then, frankly, they do need to be educated on how to use stats. That's the difference, IMO, between professional statisticians, who do intimately know how to use stats, and amateur wannabes, who think and claim that they do. Whenever someone on a blog or forum talks about "we" or the "community" with regards to advanced stats, I roll my eyes. Even most professionals who are entitled to it don't talk like that, since it sounds elitist. You don't see most professionals going around, saying, "we in the X community." You mostly just see amateurs doing it to feel that they're part of an exclusive, elitist community and make their arguments seem more authoritative. Anyways, I digress...

You could be right that zone starts don't have much effect (causation) on Corsi, but it's quite possible that the factors that affect Corsi also affect zone starts. For example, if you're a stay-at-home defender, you're likely to have one of the lower Corsis in your defense corps and you're likely to have more defensive zone starts. Neither directly has much influence on the other--starting in the D-zone isn't why you have low Corsi and having low Corsi isn't why you're starting in the D-zone--but both are directly influenced by a third factor, which is the fact that you're a stay-at-home defender and your forte is much more defense than offense. That's correlation.

 

[gorangers0525]

If your forte is getting pinned in your own zone you're not a stay at home defenseman, you're a bad defenseman.

 

[Flair Hay]

I think the OP made his point well. The margins are so tight in most cases.

I've learned never to go against a player's worth with a very good corsi or a very bad one.

If you have a great corsi, a lack of scoring isn't such a big deal. Even mediocre scoring indicates a damn good depth player.

If you have a bad corsi, you'd better be putting the puck in the net or you're not helping the team in almost all cases.

High danger scoring chances isn't great either but it's not useless either.

They are just a stat category. Each one helps a bit but none matter all that much by themselves unless a number is very good.

 

[SmellOfVictory]

If the "advanced stats community" doesn't understand the distinction between correlation and causation, then, frankly, they do need to be educated on how to use stats. That's the difference, IMO, between professional statisticians, who do intimately know how to use stats, and amateur wannabes, who think and claim that they do. Whenever someone on a blog or forum talks about "we" or the "community" with regards to advanced stats, I roll my eyes. Even most professionals who are entitled to it don't talk like that, since it sounds elitist. You don't see most professionals going around, saying, "we in the X community." You mostly just see amateurs doing it to feel that they're part of an exclusive, elitist community and make their arguments seem more authoritative. Anyways, I digress...

And now you've just started a tangential argument. I personally use the "analytics community" in discussion pretty regularly, because I don't know how else to describe what I'm arguing, most of the time. I'm not a person who does any of the grunt work at all, but I read quite a bit by people who do, and when there's borderline consensus on something, how else is someone supposed to relay that info than by saying "group x makes this claim/tends to believe this"? Is it pretentious to use the term "scientific consensus" when discussing climate change? Because in effect you're making a similar statement. (noting the differences in the fact that sports stats are not science and climate change is a million times more important/well-studied than hockey)

 

[jw2]

I think the OP made his point well. The margins are so tight in most cases.

I've learned never to go against a player's worth with a very good corsi or a very bad one.

If you have a great corsi, a lack of scoring isn't such a big deal. Even mediocre scoring indicates a damn good depth player.

If you have a bad corsi, you'd better be putting the puck in the net or you're not helping the team in almost all cases.

High danger scoring chances isn't great either but it's not useless either.

They are just a stat category. Each one helps a bit but none matter all that much by themselves unless a number is very good.

I think you're trying to defend advanced stats (??) but you're outlining why they are irrelevant.

 

[Doctor No]

I think you're trying to defend advanced stats (??) but you're outlining why they are irrelevant.

If you think that they're useless, that's your decision.

The people using (truly) advanced stats are fine with that, since the goal is to find useful information that others don't have (and then implement strategies based upon same).

 

[Sojourn]

If your forte is getting pinned in your own zone you're not a stay at home defenseman, you're a bad defenseman.

That's a meaningless statement in the context of this discussion.

 

[jw2]

If you think that they're useless, that's your decision.

The people using (truly) advanced stats are fine with that, since the goal is to find useful information that others don't have (and then implement strategies based upon same).

I said irrelevant. As in, you're writing "blue" in italics, I'm looking at the swatch.

Does looking as "blue" show you the color?

 

[Machinehead]

I get the feeling that this thread is less about making Corsi easy to grasp and more about making a case defending Fowler. If so, that's fine, but a more honest title would've been "What Fowler's Corsi really translates to," IMO.



If the "advanced stats community" doesn't understand the distinction between correlation and causation, then, frankly, they do need to be educated on how to use stats. That's the difference, IMO, between professional statisticians, who do intimately know how to use stats, and amateur wannabes, who think and claim that they do. Whenever someone on a blog or forum talks about "we" or the "community" with regards to advanced stats, I roll my eyes. Even most professionals who are entitled to it don't talk like that, since it sounds elitist. You don't see most professionals going around, saying, "we in the X community." You mostly just see amateurs doing it to feel that they're part of an exclusive, elitist community and make their arguments seem more authoritative. Anyways, I digress...

You could be right that zone starts don't have much effect (causation) on Corsi, but it's quite possible that the factors that affect Corsi also affect zone starts. For example, if you're a stay-at-home defender, you're likely to have one of the lower Corsis in your defense corps and you're likely to have more defensive zone starts. Neither directly has much influence on the other--starting in the D-zone isn't why you have low Corsi and having low Corsi isn't why you're starting in the D-zone--but both are directly influenced by a third factor, which is the fact that you're a stay-at-home defender and your forte is much more defense than offense. That's correlation.

Tanev, Vlasic, Manson, Hjalmarsson, Stralman, Pysyk, Hamhuis, Chara, and Gudas say hi.

The idea that defense-first players or players with defensive deployment are inherently bad possession players is such complete bollocks.

 

[Elvs]

I get the feeling that this thread is less about making Corsi easy to grasp and more about making a case defending Fowler. If so, that's fine, but a more honest title would've been "What Fowler's Corsi really translates to," IMO.

In all honesty, I was already defending Fowler in a topic about Josh Manson. If this was my agenda then I would have posted my study in that thread instead.

Math isn't really my strongest suit so this took me a while to do. And so, to stay interested, I chose Fowler for my study. Unfortunately I ended up going too much off topic in this thread with some of my responses to Fowler. I apologize for that.

However, several other defensemen have come on topic throughout the thread. And my purpose was to enlighten people how small the marginals often are. My theory is that almost any player in the league could be above or below the 50% CF% mark in any given season. It's about the position you put the players in, and thus I think the stat is severely overrated when applied to individual players.

 

[Creativero]

Tanev, Vlasic, Manson, Hjalmarsson, Stralman, Pysyk, Hamhuis, Chara, and Gudas say hi.

The idea that defense-first players or players with defensive deployment are inherently bad possession players is such complete bollocks.

That's only 6 guys. They have minuscule effect on the over average Corsi of defensive vs offensive players. I'm sure there are atleast 6 guys just as bad as they are good to cancel them out. Plus Hjalmarsson isn't a one dimensional defensive player 5v5. His numbers aren't great because he doesn't produce on the powerplay.

 

[4thline]

If you think that they're useless, that's your decision.

The people using (truly) advanced stats are fine with that, since the goal is to find useful information that others don't have (and then implement strategies based upon same).

Interesting distinction in a discussion driven by corsi blogosphere zealots

 

[Osprey]

And now you've just started a tangential argument. I personally use the "analytics community" in discussion pretty regularly, because I don't know how else to describe what I'm arguing, most of the time. I'm not a person who does any of the grunt work at all, but I read quite a bit by people who do, and when there's borderline consensus on something, how else is someone supposed to relay that info than by saying "group x makes this claim/tends to believe this"? Is it pretentious to use the term "scientific consensus" when discussing climate change? Because in effect you're making a similar statement. (noting the differences in the fact that sports stats are not science and climate change is a million times more important/well-studied than hockey)

You've just started a tangential argument to my tangential argument by bringing up that subject . I won't go there, but yes, I believe that it's often pretentious to argue that there's a "consensus," regardless of the topic. Most of the time, people are exaggerating and/or fudging facts to give extra weight to their arguments and end debate. If you're in a technical field and you're trying to end debate because you're so sure that your theory is correct, you're doing your field a huge disservice.

Tanev, Vlasic, Manson, Hjalmarsson, Stralman, Pysyk, Hamhuis, Chara, and Gudas say hi.

The idea that defense-first players or players with defensive deployment are inherently bad possession players is such complete bollocks.

Those players all contribute a fair bit on offense to go along with their defense. They are not "stay-at-home defenders." You changed the language to "defense-first players" to twist my argument, which is rather disingenuous, as is listing a number of players who routinely score 30 or more points. When I say "stay-at-home defenders," that's what I mean, not "defense-first players." I'm talking about guys like Greene and Scuderi, who are/were used in almost exclusively defensive roles.

 

[Canadienna]

A CF% of 55 and a CF% of 45 isn't nearly as big as it may sound. Not over the course of just one season.

You said "the difference between 45% and 55% isn't that big".

I said the difference isn't as big as it sounds.

I know what you said

Clearly.


I agree with OP, Corsi differences are not as extreme as they seem to be once they are investigated.

 

[Bourdon101]

The Kings are the poster boys for corsi and they've never been a particularly great regular season team. I don't think they care.

Are you saying that in the playoffs, corsi correlates with wins? Because that's false.

Corsi has to be the most misused stats of them all. The almighty "advanced statistics community" should concentrate their efforts on finding new models instead of trying to adjust this one until it confirms their theories.

 

[Machinehead]

You've just started a tangential argument to my tangential argument by bringing up that subject . I won't go there, but yes, I believe that it's often pretentious to argue that there's a "consensus," regardless of the topic. Most of the time, people are exaggerating and/or fudging facts to give extra weight to their arguments and end debate. If you're in a technical field and you're trying to end debate because you're so sure that your theory is correct, you're doing your field a huge disservice.



Those players all contribute a fair bit on offense to go along with their defense. They are not "stay-at-home defenders." You changed the language to "defense-first players" to twist my argument, which is rather disingenuous, as is listing a number of players who routinely score 30 or more points. When I say "stay-at-home defenders," that's what I mean, not "defense-first players." I'm talking about guys like Greene and Scuderi, who are/were used in almost exclusively defensive roles.

Good defensive players are going to score a certain amount of points because they can settle the puck down and get it away from their own net.

I can't think of any defenseman who is absolutely hopeless offensively and good defensively. If you're hopeless offensively, you're just bad at hockey at the NHL level.

With the exception of maybe Tanev who has a career high of 20 points. And even that is a far cry from Scuderi.

 

[Machinehead]

Are you saying that in the playoffs, corsi correlates with wins? Because that's false.

Corsi has to be the most misused stats of them all. The almighty "advanced statistics community" should concentrate their efforts on finding new models instead of trying to adjust this one until it confirms their theories.

Again 22 out of the last 24 Cup winners were top 10 in shot differential. Not sure what part of that people can't grasp.