Calgarypuck Advanced Stat Summary

[polak]

Originally Posted by Bingo

The verdict is in!

Lambert says the Flames suck http://flamesnation.ca/2015/2/12/fiv...medium=twitter

I was pretty harsh in my comments but I'm so sick this

[PeteMoss]

Advanced stats are what they are. They measure what they say they do.

Acting like something with a 0.5 (or whatever it is) correlation to points means nothing is dumb, but so is acting like something with that type of correlation is a sure-fire solution to determining which team is better.

[Flames Fan, Ph.D.]

Originally Posted by PeteMoss

Advanced stats are what they are. They measure what they say they do.

Acting like something with a 0.5 (or whatever it is) correlation to points means nothing is dumb, but so is acting like something with that type of correlation is a sure-fire solution to determining which team is better.

Actually, that's the point of the measurement. The whole intent is to determine the likelihood that action A means something in relation to action B.

[PeteMoss]

Worth noting the Flames have been a relatively average team by the advance stats over the last 10 games. http://www.puckon.net/?s=2014-09-01&...f=0&ld=-1&l=10

[tknez16]

Here's a little rant I prepped about advanced stats. I thought about a new thread, but this fits nicely here.

People greatly misunderstand “advanced stats” for hockey. This derives, mostly, from wanting to simplify what they mean. “Is this team/player good?” is not a question that’s answered by the current stats that are being tracked. Generally, they track something more specific. Here are the two that really bother me in how they’re interpreted.

PDO (Shooting % + Save %) – This isn’t a great stat, but it’s the only one that really says “Is this team playing well?” If you have a high shooting percentage and a high save percentage, then you’ll win lots… and that’s obviously not a coincidence.

What this isn’t, is a proxy for luck. Sure, a team that’s lucky will have a higher PDO, but this also reflects skill, both offensive and defensive. If a team has a very high PDO, then it could be either luck or skill (if it’s high enough, then likely both) that cause it. For the Flames, many pundits will say it’s all luck because the story lines going into the season said that the team would be bad. This take is simply bias, just as all of us Flames fans hating their take is our own bias. Media types don’t want to admit they were wrong about the skill level of the team, while fans will tend to overrate their team’s skill level.

Corsi (Shot Attempt Differential) – The biggest thing that bothers me about this stat is how it’s interpreted. This does not say whether a player/team is good. It says how much a player/team controls the puck. Generally speaking, teams that control the puck more tend to be better. This is not a rule. Different styles of play will greatly affect corsi ratings.

The misconception that corsi directly relates to team quality comes from the trend in play style in the league. Most teams play a puck possession game. If you play a game that’s centered on puck possession and you have a bad corsi, chances are that you’re a bad team because you aren’t achieving your goal of controlling the puck.

On the other hand, there are teams that don’t play a puck possession game, but instead concentrate on hard puck pursuit and collapse in the defensive zone. We Flames fans are familiar with this. When a team collapses in the defensive zone and concentrates more on shot blocking, it tends to control the puck less and have more shot attempts against. When a team doesn’t have large amounts of offensive zone time, instead scoring on the rush and on turnovers, they will not have a large number of shot attempts for. When both of those are the same team, they will have a terrible corsi. This has nothing to do with them being a bad team, but simply their style of play.

Tl;dr – Advanced stats aren’t bad, they’re just misused.

[PeteMoss]

Originally Posted by Flames Fan, Ph.D.

Actually, that's the point of the measurement. The whole intent is to determine the likelihood that action A means something in relation to action B.

Yes I know, my point is that it isn't a perfect correlation (or anywhere close to it) so shouldn't act like it is.

[polak]

Originally Posted by PeteMoss

Advanced stats are what they are. They measure what they say they do.

Acting like something with a 0.5 (or whatever it is) correlation to points means nothing is dumb, but so is acting like something with that type of correlation is a sure-fire solution to determining which team is better.

Isn't it more like .25?

http://www.sportingcharts.com/articl...tatistics.aspx

That's not even acceptable, let alone sure fire. You wouldn't use a coefficent like that for any sort real world predictions. I can't remember my stats classes from University but I'm pretty sure the minimum acceptable coefficient for a statistic to be valid as a predictor would be around .6 -.8 and even then a .6 coefficient means you're barely right more than 50% of the time.

Again, these stats are a great insight for more in-depth analysis, but they are useless when it comes to predicting wins on their own.

[GranteedEV]

Originally Posted by Bingo

I wonder about sustainability all the time. I look into advanced stats, and then I marry that with my eye test and try and come up with what my blended theory of things are. The Flames have bad advanced stats, and that is troubling.

I can't help but wonder why there's no advanced stat for "time spent with puck in your opponent's zone"

I mean, time-with-ball is THE statistic used in soccer. Yet in hockey it's all about volume shooting, where patience is punished.

[EldrickOnIce]

Originally Posted by polak

You wouldn't use a coefficent like that for any sort real world predictions.

You absolutely could use an R^2 value in the 25% range.
But,it is exactly what that R^2 says: the model explains 25% of the variation in Y (points). 75% is still unexplained.

[polak]

Originally Posted by EldrickOnIce

You absolutely could use an R^2 value in the 25% range.
But,it is exactly what that R^2 says: the model explains 25% of the variation in Y (points). 75% is still unexplained.

Which is useless for a real world use. No organization would use a coefficent of correlation that low for any sort of prediction.

[Ashartus]

Originally Posted by GranteedEV

I can't help but wonder why there's no advanced stat for "time spent with puck in your opponent's zone"

I mean, time-with-ball is THE statistic used in soccer. Yet in hockey it's all about volume shooting, where patience is punished.

I think that's because time spent with the puck isn't tracked, so shot attempts are used as a surrogate for time spent with the puck on the assumption that if you have the puck more you'll have more opportunities to attempt shots.

[EldrickOnIce]

This is from 2009, but it's interesting (before 'advanced stats' became the flavour of the day).
A regression analysis initially using PP%, PK%, 5 on 5 goals for and against, goals against per game, goals scored per game, shots per game and shots against per game.

The interesting part is the first thing the guy eliminated was shots for and against:

I was quickly able to eliminate the two shot categories thanks to their high P-values, leaving me with a formula with a r-squared of .843. Not too shabby. I was happy enough with this and decided to move forward. My formula is 22.771 + 0.714PP% + 0.480PK% + 23.960fiveonfive + 16.703 GPG -19.541 GAPG

Would be an interesting comparison to the corsi based predictors.

http://www.dailyfaceoff.com/2438/a-n...t-team-success

[polak]

Originally Posted by Ashartus

I think that's because time spent with the puck isn't tracked, so shot attempts are used as a surrogate for time spent with the puck on the assumption that if you have the puck more you'll have more opportunities to attempt shots.

And this is where it all breaks down.

Hockey isn't nearly as much of a possesion game as a sport like soccer. That's why I can't ever get on board with this. In soccer, it's almost impossible to score if you don't have control of the ball. The distances involved and the number of players on the field mean you have to systematically be able to control possesion to move up the field and be successful. Even if you are counter-attack team, you still need to have decent possesion.

In hockey you can let the other team have the puck all day if you can keep them to the outside and have reliable goaltending. Why? Cause if the team with possesion bobbles the puck in a matter of moments you can break free and score. Relistically you have a handful of strides and one or two players between you and the net almost every time you break out.

[EldrickOnIce]

Originally Posted by polak

Which is useless for a real world use. No organization would use a coefficent of correlation that low for any sort of prediction.

R^2 values like that are used all the time, in real world regression equations.
Surprising, but true.
But in the spirit of what you are referring to, agreed. It does not explain enough

[getbak]

Originally Posted by GranteedEV

I can't help but wonder why there's no advanced stat for "time spent with puck in your opponent's zone"

I mean, time-with-ball is THE statistic used in soccer. Yet in hockey it's all about volume shooting, where patience is punished.

There's currently no way of measuring actual possession stats without someone sitting with a stopwatch while watching every game.

I believe that someone did sit down with a stopwatch for a few games and figured out that the number of pucks directed at the net by each team related pretty closely with the actual time of possession during the game. As a result, shot attempts has become the go to analog for possession.

The NHL is testing actual player and puck tracking technology (it was used at the All Star Game). Once that starts being used in every game (likely within the next couple of years), no one will care about Corsi or Fenwick ever again.


When that does happen, it will be interesting to see how well Corsi and Fenwick actually match real possession numbers in all games. Will it link up pretty well in all situations, or will teams with a different playing style have actual possession stats that differ greatly from the Corsi numbers?

[MrMastodonFarm]

Originally Posted by Bingo

its stuff like this that drives me nuts



That's a person that wishes to discount opposing views rather than just standing on your two feet and defending your position.

I wonder about sustainability all the time. I look into advanced stats, and then I marry that with my eye test and try and come up with what my blended theory of things are. The Flames have bad advanced stats, and that is troubling. However they aren't getting absolutely killed every night or scoring flukey goals from center ice. Anyone who watches the games knows that.

Lambert is a complete fool.

[Street Pharmacist]

Originally Posted by polak

Which is useless for a real world use. No organization would use a coefficent of correlation that low for any sort of prediction.

True, but this isn't real world. We constantly are looking for ways to simplify an incredibly complex mix of tangible and intangible parameters within the game. Therefore anything that correlates with winning is useful. That R^2 value simply means, while useful, it only paints part of the picture. Unfortunately, Colorado and Toronto are the two examples constantly thrown out to rebut that argument. I'm going to find examples of successful teams that had consecutive sub 50% corsi.

Here we go
Pittsburgh:
Year corsi%5v5 pts
06/07 46.2% 105
07/08 45.5% 102
08/09 48.1% 99

Pretty sure there's two consecutive Stanley Cup appearances there along with a Cup.

[Resolute 14]

The most useful nugget of info from Travis Yost's article on TSN a couple days ago is this chart:



Of 128 playoff teams (or in playoff position for this year), the vast majority had positive Corsi ratios. It looks like about 38 made the playoffs below that line. So good Corsi teams make the playoffs just over 70% of the time. Only about 18 of those did so with bad possession ratios, but that is still 14%

What that tells us is that overall, what the Flames are doing is an exception, but not necessarily exceptional. And advanced stat people should probably take that to heart. Predicting that we can't keep it up indefinitely is fair. But analytics guys should not act so offended because Calgary is bucking the trend. By the same token, anti-stats people shouldn't be offended that people don't think it will last indefinitely.

[Bingo]

Originally Posted by PeteMoss

Worth noting the Flames have been a relatively average team by the advance stats over the last 10 games. http://www.puckon.net/?s=2014-09-01&...f=0&ld=-1&l=10

I see them something like 24th in your link, which stat are you looking at in there?

Better than the season where they are worse than only Buffalo and Colorado, but still pretty bad.

[Tiger]

In all the games I have watched it looks like the flames try to make the extra pass for the better scoring chance. which means lower shots, but still good possession. I find Corsi hard to believe when there are games like the NYR game. Where everyone who watched it would tell you it was a ####ty game (Rangers played bad too), but that is the flames "most dominate game". It just doesn't pass the smell test. Correlation not causation and therefore you cannot make firm conclusions from it.