Advanced Stats vs Don Cherry

[Mimico]

Over the last few years, hockey fans have slowly become introduced to an increasing variety of advanced statistics. The most commonly referenced "advanced stat" used by hockey analysts today is Corsi. Corsi is essentially a possession stat and is calculated as follows, "Shots + shots attempts that missed the net + shot attempts that were blocked". It wasn't long before people realized that Corsi translates to puck possession, and puck possession translates to winning hockey games. However, the fact that Corsi doesn't take the quality of shots into account, it isn't always the best predictive tool (see Toronto Maple Leafs). What I did was put Corsi to the test. Just how strong is the correlation between Corsi and Winning? Well, after putting a quick spreadsheet together, it turns out that Corsi and Winning have shared a correlation of about 0.289 this NHL season thus far. Generally speaking, this is considered a "weak positive correlation". To put the strength of this correlation into perspective, I also compared the % of Canadian players a team has on their roster to winning. The reason for doing so was inspired by Don Cherry and his tendency to emphasize the importance of having Canadian players if you wish to ice a successful hockey team. After testing Cherry's theory, the results showed that the % of Canadian players on a team and winning shared a correlation of about 0.334, topping Corsi's 0.289. Is Corsi irrelevant? Ask James Mirtle and you'll be told that it isn't, but the numbers don't lie...

#defycorsi

Click to read more...

 

[Keeping it Blue]

Heh, interesting find. Check it for last year, or the full season before that.

 

[JAMmer124]

1) paragraphs dude. Use em.

2) Corsi isn't the be all and end all. It's a simple advanced stat, and like any stat, has it's flaws. Having said that, you don't need numbers to tell you that the Leafs get outplayed and outshot, quite badly some nights. Our goaltending has saved our ***** almost every game this season. You shouldn't need Corsi, Fenwick, or any other stat to tell you that. You can see it with your eyes.


All these advanced stats do is back up what we're seeing with our eyes, and allows people with minimal hockey knowledge sound like experts.

Interesting find though, I'll give you that.

 

[Rogie]

Over the last few years, hockey fans have slowly become introduced to an increasing variety of advanced statistics. The most commonly referenced "advanced stat" used by hockey analysts today is Corsi. Corsi is essentially a possession stat and is calculated as follows, "Shots + shots attempts that missed the net + shot attempts that were blocked". It wasn't long before people realized that Corsi translates to puck possession, and puck possession translates to winning hockey games. However, the fact that Corsi doesn't take the quality of shots into account, it isn't always the best predictive tool (see Toronto Maple Leafs). What I did was put Corsi to the test. Just how strong is the correlation between Corsi and Winning? Well, after putting a quick spreadsheet together, it turns out that Corsi and Winning have shared a correlation of about 0.289 this NHL season thus far. Generally speaking, this is considered a "weak positive correlation". To put the strength of this correlation into perspective, I also compared the % of Canadian players a team has on their roster to winning. The reason for doing so was inspired by Don Cherry and his tendency to emphasize the importance of having Canadian players if you wish to ice a successful hockey team. After testing Cherry's theory, the results showed that the % of Canadian players on a team and winning shared a correlation of about 0.334, topping Corsi's 0.289. Is Corsi irrelevant? Ask James Mirtle and you'll be told that it isn't, but the numbers don't lie...

#defycorsi

If you can, (have time and energy), it would be nice to see the correlation (the r-squared would also be nice) for Corsi and winning for last season's regulation games.


I wonder if it isn't higher than .289!

 

[Canada4Gold]

read an article the other day that had Corsi with something like a .6 correlation with winning %

Maybe the small sample size is making the number lower than it usually is?

as for the Cherry stat I could see that, the cbc article the other day which showed the %'s for each team in order, most of the teams that sucked were on one side with the good teams mostly on the other

 

[Rogie]

read an article the other day that had Corsi with something like a .6 correlation with winning %

Maybe the small sample size is making the number lower than it usually is?

as for the Cherry stat I could see that, the cbc article the other day which showed the %'s for each team in order, most of the teams that sucked were on one side with the good teams mostly on the other

That sounds more like what I would have expected - and, I wonder if it's higher and if so, how much higher, in playoff games.

 

[agropop]

If you look at Corsi as nothing more than shot attempt differential it's a useful statistic in terms of quantifying possession. Nothing more. It doesn't have to be predictive to be of value.

Wanna win games with poor possession? Easy, just maintain higher shooting and save percentages than your opponent. Sadly, we know that over time any teams PDO (save + shooting %) will normalize towards 1000.

We may very well maintain our freakishly high PDO but history tells us that more often than not things will regress to the mean.

 

[7even]

I'm curious as to what your motivation was to use one month worth of data when you have literally years worth of data to use. I can't think of a scenario where you would be better off using less data than more.




Unless you just really want your data to say one thing and not another...

 

[colchar]

2) Corsi isn't the be all and end all. It's a simple advanced stat, and like any stat, has it's flaws. Having said that, you don't need numbers to tell you that the Leafs get outplayed and outshot, quite badly some nights. Our goaltending has saved our ***** almost every game this season. You shouldn't need Corsi, Fenwick, or any other stat to tell you that. You can see it with your eyes.


All these advanced stats do is back up what we're seeing with our eyes, and allows people with minimal hockey knowledge sound like experts.

But by winning while being outplayed the Leafs this season are demonstrating that this advanced stat doesn't mean much. All that matters is the score at the end of the game.

Look at it this way - Team A could have the puck for virtually the entire game, pump 100 shots at Team B's net, but not score (say they ran into a hot goalie or something). Meanwhile, Team B could only have the puck on their sticks for a minute or two the entire game, get less than a handful of shots, but score on two of them thus winning the game 2-0. But according to this Corsi silliness, Team A would still be superior to Team B despite having lost.

As I said, all that really matters is the score at the end of the game and maybe we should leave the advanced statistics to those baseball perverts (random M*A*S*H reference there) who need something to keep their minds occupied during those three hour snoozefests.

 

[agropop]

But according to this Corsi silliness, Team A would still be superior to Team B despite having lost.

No, according to Corsi, team A possessed the puck more than team B, that's it.

 

[colchar]

No, according to Corsi, team A possessed the puck more than team B, that's it.

Yes but people interpret that to mean that Team A is better than Team B and extrapolate from there when all they should really be worrying about is the score.

 

[darrylsittler27]

Does Corsi include?

Your ability to hurt versus your ability to protect your stars? A failed stat.

 

[agropop]

Yes but people interpret that to mean that Team A is better than Team B and extrapolate from there when all they should really be worrying about is the score.

An incorrect interpretation does not invalidate the statistic itself.

 

[colchar]

An incorrect interpretation does not invalidate the statistic itself.

If it is being used solely as a measure of possession and shots yeah, it is valid. But no conclusions about results should be assumed.

 

[AlienWorkShop]

Okay, apparently my snarky reference to statistical noise wasn't considered legitimate, so let me expand on that...

For one, you're comparing correlation coefficients between two models and two explanatory variables of different order, neither of which is a good idea. Also, this early in the year you're dealing with a lot of statistical noise for both winning% and Corsi. All Corsi-related arguments are (or at least should be) based on the long-run. I think there are legitimate criticisms that Corsi doesn't "settle down" over a full-season, let alone last year's half-season, but that's simply suggesting one-year outliers can exist yet are still outliers and should be treated with a healthy amount of suspicion.

Secondly, even if it was fine to compare the two correlations as you did, your results are hardly different at a statistically significant level with such a small sample size. Just as a quick demonstration of their sensitivity to model specification, I plugged those numbers into a two-variable regression: winning% = b0 + b1Corsi + b2CDN%

Both variables are positive and statistically significant (which is ultimately all we're concerned about, interesting theories about why Canadian% is important can certainly be spun), but the coefficient on Corsi is 1.62, on Canadian% is .53.

 

[agropop]

If it is being used solely as a measure of possession and shots yeah, it is valid. But no conclusions about results should be assumed.

Yes, and at the same time, the score alone does not indicate the quality of play, it too can be misleading.

 

[Confucius]

If it is being used solely as a measure of possession and shots yeah, it is valid. But no conclusions about results should be assumed.

I fully agree, there is no way it tells you the quality of shots or quality of goaltender.

 

[JAMmer124]

But by winning while being outplayed the Leafs this season are demonstrating that this advanced stat doesn't mean much. All that matters is the score at the end of the game.

Look at it this way - Team A could have the puck for virtually the entire game, pump 100 shots at Team B's net, but not score (say they ran into a hot goalie or something). Meanwhile, Team B could only have the puck on their sticks for a minute or two the entire game, get less than a handful of shots, but score on two of them thus winning the game 2-0. But according to this Corsi silliness, Team A would still be superior to Team B despite having lost.

As I said, all that really matters is the score at the end of the game and maybe we should leave the advanced statistics to those baseball perverts (random M*A*S*H reference there) who need something to keep their minds occupied during those three hour snoozefests.

Its something that can happen a few times in single games, sure, but asking our goalies to stop 35-40 shots a game is ridiculous. It's too much. We dont play good defence most nights, it's as simple as that. It'd be a different story if we were only getting outshot by a couple shots a game, but there's nights where they're doubling us almost.

 

[Duke Silver]

Using the same technique, taking your data and correlating it with points % to take into account extra points gives a correlation of 0.38.

But as AlienWorkShop has noted above, this appears to be a kind of flimsy way of going about the analysis.

 

[Kyle Doobas*]

you don't need numbers to tell you that the Leafs get outplayed and outshot, quite badly some nights. Our goaltending has saved our ***** almost every game this season.

So? That's what goaltenders are for. Lately people are acting like we're supposed to feel bad for having a good goalie or something.

The goalie is a part of the team, so if you can't beat the opposing team's goalie, then how exactly did you 'outplay' that team? By cycling around in their zone, maybe getting some good chances, having them stopped and giving up four goals? Great job. You still lost the game 4-0.

 

[Duke Silver]

I'm seriously interested to know how many games the Leafs have won whilst being out-Corsi'd over the last two seasons (62 games).

 

[agropop]

So? That's what goaltenders are for. Lately people are acting like we're supposed to feel bad for having a good goalie or something.

The goalie is a part of the team, so if you can't beat the opposing team's goalie, then how exactly did you 'outplay' that team? By cycling around in their zone, maybe getting some good chances, having them stopped and giving up four goals?

Yeah, getting outshot 2-1 every night isn't concerning at all... This'll never backfire....

 

[K19*]

1) paragraphs dude. Use em.

2) Corsi isn't the be all and end all. It's a simple advanced stat, and like any stat, has it's flaws. Having said that, you don't need numbers to tell you that the Leafs get outplayed and outshot, quite badly some nights. Our goaltending has saved our ***** almost every game this season. You shouldn't need Corsi, Fenwick, or any other stat to tell you that. You can see it with your eyes.


All these advanced stats do is back up what we're seeing with our eyes, and allows people with minimal hockey knowledge sound like experts.

Interesting find though, I'll give you that.

Agreed. Advanced stats should supplement the eye test, not be your primary method of evaluating talent.

 

[firstemperor]

1) paragraphs dude. Use em.

2) Corsi isn't the be all and end all. It's a simple advanced stat, and like any stat, has it's flaws. Having said that, you don't need numbers to tell you that the Leafs get outplayed and outshot, quite badly some nights. Our goaltending has saved our ***** almost every game this season. You shouldn't need Corsi, Fenwick, or any other stat to tell you that. You can see it with your eyes.


All these advanced stats do is back up what we're seeing with our eyes, and allows people with minimal hockey knowledge sound like experts.

Interesting find though, I'll give you that.

The problem is, we don't get outplayed "most nights". We allow a lot more shots because the defensive system is perimeter-oriented- so we gain less possession out of it (and don't gamble for posession battles)- but the other team often peppers low quality shots on net.

Offensively, we do very little board work compared to a lot of teams, a lot of transition play and we have the creativity to make plays off a flash of brilliance- which is what is needed to score at the end of the day. We get away with this because we have insane talent up front. Generally speaking, this often leads to short leads and spurts of possession.

The point is, the Leafs play a game that isn't predicated purely on posession. Calgary reminds me exactly of those Paul Maurice teams, where we could out-work or out-shoot teams most nights but simply didn't have the talent to get over the edge in terms of the score-sheet.

As for interpretation, Corsi does say we have a lot of room for growth if we improve upon those stats, but saying we're "lucky" is a stretch. We've had good goaltending and we have great special teams. Those are the first elements of a good team. We've deserved at least 80% of the wins in my eyes just based on quality of chances. Edmonton out-shot us by 20+ the other day, we dominated the game from start to finish. The 3rd period was a mis-match-> dump game on a first night of a back-to-back, I wonder how flawed Corsi metrics are for those type of games. On paper, there's way too much talent for us to be considered lucky- coming from a relatively pessimistic Leafs fan.

 

[LogieTrice]

HAHA Mirtle does have a hard on for Corsi.

Good Corsi = taking a ton of bush league shots from the perimeter that basically pads the tendys stats (see Reimer/Bernier).

Quality scoring chances, goals and wins are all that matters. With most emphasis on the latter two.