How Screwed are the Flames?

[CorsiHockeyLeague]

Originally Posted by Street Pharmacist

Yes and no. I agree it's the best stat for determining who's good, but the purpose of the other statistics is to help predict who will win later and they do that better than wins/losses

I don't even agree it's the best stat for determining who's good. How many games do you need to determine this? IIRC, there was a point about 1/4 of the way through the season where the Leafs had a better record than Chicago. Were the Leafs a better team? Did they suddenly get worse at hockey when Chicago passed them? Right now, the Kings have fewer wins than the Canucks, Wild and Flames, but in my opinion at least they're better than all three of those teams.

When you say a stat is more useful at "predicting who will win later", that to me is a better indicator of who is good at hockey, whatever that stat may be. I don't think reducing stats to that descriptor is a useful process because it's misleading (no stat can tell you who will win the NYR / CHI game today with any certainty), though.

Originally Posted by heep223

My biggest issue is that "possession" stats don't actually track possession. A team could be holding the puck for 2 minutes cycling in the o-zone and have some great scoring chances that don't end up in actual shot attempts. Then the other team rushes back with a weak shot attempt from the outside and they are winning the Corsi battle.

This is another common refrain. In theory, you're right, but any effort made to track this has determined that it's not what happens. Remember, the Leafs in 2012 were actually saying to the press (coaches AND management) that this is what they were doing. When checked against actual O-zone time of possession:

[EldrickOnIce]

Originally Posted by Street Pharmacist

I agree and disagree here. While it may or may not track "possession" (small sample size says it's pretty close), it's really irrelevant. If it's predictive power (which is really the end value we're looking for) is good or bad, who really cares about actual possession?

Because the predictive power is not all that spectacular, from a pure statistical standpoint.

[heep223]

Originally Posted by Street Pharmacist

I agree and disagree here. While it may or may not track "possession" (small sample size says it's pretty close), it's really irrelevant. If it's predictive power (which is really the end value we're looking for) is good or bad, who really cares about actual possession?

It sounds a bit silly but I honestly wonder if they would be taken more seriously and have more of a widespread impact if they were called "shot differential, shot attempt differential" as opposed to "Corsi, Fenwick, advanced stats, analytics, possession stats, etc".

[Jay Random]

It depends what you're after, doesn't it?

If you're satisfied with correlations, and don't give a damn about understanding how the correlations work or what the causative factors are, by all means content yourself with possession proxies.

If you want to figure out how hockey games are actually won and lost, and learn something from data analysis that coaches and managers don't already know, you need more than correlation. You need to do some actual science. To begin with, when there are significant exceptions to a correlation, you can't just classify them as exceptions and shrug them off – because the exceptions disprove your model. You need to look at the exceptions, study additional factors, and refine the model.

The next job for hockey stats people is to explain why there are exceptions that Corsi and Fenwick do not adequately predict. And no, writing it off as ‘luck’ or ‘unsustainable’ is not an explanation; it is a refusal to explain. These people are staring at a golden opportunity right here in Alberta, with two teams that consistently defy their expectations, one in each direction. Instead of studying these phenomena and trying to figure out what additional factors are involved, they screw their eyes shut and chant, ‘Corsi and Fenwick! Corsi and Fenwick! Everything is explained by Corsi and Fenwick!’

[EldrickOnIce]

Originally Posted by Resurrection

the thing is though that historically, corsi has been a very good predictor of playoff teams. of course exceptions exist: its not perfect. no stat is. but its another great tool to determine whats going on with teams on the ice.

Yes, corsi is about a 65% predictor of playoff teams.
There is a reasonable chance maggie the monkey could get 65%. Or maybe even I could predict that, before the season started.

Goal differential is about 90% predictor. it terms of statistics, that is a remarkable difference.

[Street Pharmacist]

Originally Posted by heep223

It sounds a bit silly but I honestly wonder if they would be taken more seriously and have more of a widespread impact if they were called "shot differential, shot attempt differential" as opposed to "Corsi, Fenwick, advanced stats, analytics, possession stats, etc".

Oh, hell yes. The only reason, and I mean only reason, they have those names is for self aggrandizing. Jim Corsi, Matt Fenwick and some Internet dude who goes by PDO may all be swell people (and by all accounts they are), but their names were used to obfuscate the terms so only people "in the know" would understand it. It's the equivalent of a password to get in their tree fort.

Why are so many of them against the new terms? Because then it becomes normal and that can't act like they're in some special "smart people club"

[CorsiHockeyLeague]

^Matt Fenwick is heavily in favour of losing the old names. Jim Corsi probably doesn't give a damn, he wasn't even involved in naming the stat.

Originally Posted by heep223

It sounds a bit silly but I honestly wonder if they would be taken more seriously and have more of a widespread impact if they were called "shot differential, shot attempt differential" as opposed to "Corsi, Fenwick, advanced stats, analytics, possession stats, etc".

Well the NHL has actually done that and there's been a call for that for a while, but the NHL's stat site is both useless and wrong so until they fix it this likely won't catch on.

Second, it doesn't really matter because in my view once SportVU tech comes in we will have actual time of possession data down to the second, and corsi will no longer add much if any value.

Third, "corsi", "fenwick" "advanced stats" "analytics" and "possession stats" all mean different things. Corsi and fenwick are types of possession stats, which in turn are types of what some people call "advanced stats" (along with PDO, ONSH%, P/60, etc). All of these together can be used to engage in analytics. Analytics happens when you take statistics and use them to either explain something that's happened, or predict something that will happen, on the ice.

[heep223]

Originally Posted by CorsiHockeyLeague

Third, "corsi", "fenwick" "advanced stats" "analytics" and "possession stats" all mean different things.

Yes but the terminology is all kind of used interchangeably for the layman. It kind of shoots itself in the foot.

[OutOfTheCube]

Originally Posted by CorsiHockeyLeague

You can also enjoy hockey by meticulously obsessing over these details to a borderline crazy degree, watch each breakout ten times to determine what seems to produce controlled zone entries, then track the controlled zone entries and try to figure out how those are generally best transitioned into scoring chances, while mapping out shot locations for your team, etc etc etc.

I can't believe that there's actually real people that do this. Unless they're some kind of guy like Dustin Hoffman in Rain Man.

[CorsiHockeyLeague]

Originally Posted by EldrickOnIce

Goal differential is about 90% predictor. it terms of statistics, that is a remarkable difference.

In what sense is it a predictor? If I take goal differential at the ten game mark, will it help me figure out who will make the playoffs this year? How about the 20 game mark? How about next year?

This chart demonstrates usefulness and reliability of different stats in predicting wins and losses.


EDIT: from Steve Burtch's article here - http://www.pensionplanpuppets.com/20...-blocked-shots

I kind of suspect that what you mean when you say "90% predictor" is that if you look at the standings at the end of the year, there's a 90% chance that if you're in the top 16 in goal differential you'll make the playoffs. That's not any more useful than saying that a team's win loss record is a 100% predictor of whether they make the playoffs. It's rear-view-mirror stuff.

[CorsiHockeyLeague]

Originally Posted by OutOfTheCube

I can't believe that there's actually real people that do this. Unless they're some kind of guy like Dustin Hoffman in Rain Man.

Corey Snazjder watched every single NHL game played last year (something like 1230) for the purpose of tracking zone entries / exits, and a few other things.

[EldrickOnIce]

No more or less rear view mirror than corsi.

sample size and all that

[CorsiHockeyLeague]

I think the column labeled "reliability" there contradicts that. You were right earlier when you said the predictive power is by no means staggering; it's just better than everything else.

[MrMastodonFarm]

Originally Posted by CorsiHockeyLeague

If your general view is that people who like to talk about this stuff are smarmy and arrogant, I'd just say re-read your own posts here. The sneering derision towards analytics on this board (and as has been pointed out, any board dedicated to any hockey team that has a season like this one) is at least as grating - partly because there's absolutely no substance underlying it. Those kind of cheap drive-by lines do seem to generate thanks clicks, though, so I guess you're getting the pats on the back you set out for.

Everyone knows I'm smug and arrogant, that's why I can recognize it so easily!

The loudest voices, and even a founding guy like mudcrutch79 are supremely arrogant, this isn't new information and I'm not the first to say so. Even you yourself come across as a little smarmy saying that you want an intellectual discussion about hockey, almost suggesting that analytics is the only way to do so.

As for the thanks part at the end, ha. There was a time on Calgarypuck when thanks didn't exist and anyone from around that time would say I haven't changed much, I'm probably less of a jerk now. But I'm super handsome make funny jokes, so women and men alike fawn to me, like a moth to a Flames, like a cheetah pouncing on the proverbial juicy rabbit.

[MrMastodonFarm]

Originally Posted by CorsiHockeyLeague

Corey Snazjder watched every single NHL game played last year (something like 1230) for the purpose of tracking zone entries / exits, and a few other things.

That's crazy, how did he have time for Game of Thrones?

[Street Pharmacist]

Originally Posted by CorsiHockeyLeague

I think the column labeled "reliability" there contradicts that. You were right earlier when you said the predictive power is by no means staggering; it's just better than everything else.

Lots of holes to be poked in corsi, which again doesn't mean it's useless, also doesn't make it as strong as Lambert or the other opinion guys would suggest.

I did some analysis on Dec 1st as the Flames were below 45% Corsi 5v5 mark and a blogger suggested it was "impossible to make the playoffs" below 45%.

14 teams were below that mark by Dec 1st since shot attempts started being tracked ('06/'07. Ignored lockout season as there was no hockey be Dec 1st). That's the bottom 14 teams of 270 teams. That's the bottom 5% of corsi 5v5. Yet somehow, 4 of those 14 still made the playoffs (28% or almost a third).

This tells us again that while Elite teams always have great corsi and terrible teams always have terrible corsi, the reverse isn't true.

[Ryan Coke]

Originally Posted by Street Pharmacist

Yes and no. I agree it's the best stat for determining who's good, but the purpose of the other statistics is to help predict who will win later and they do that better than wins/losses

So win percentage over let's say 40 games is a less reliable predictor of win percentage for the next 40 than Corsi? It doesn't look like win percentage is in the chart above and ranked for its predictive ability.

Either way, my point was more about the determination of what team is better at a given moment. To me, if a team has more wins, then they are the better team. Beyond that we are talking about a teams potential to perform better in the future than they have in the past. But it becomes pretty hollow IMO, like saying we lost the 7 game series, but were the better team. It can make us feel better if we were on the losing side, but ultimately the team that won was better over that stretch.

[Inglewood Jack]

Originally Posted by Street Pharmacist

In my opinion Toronto and Colorado ate the worst things to happen to the analytics movement. They are the trump cats for those who don't understand analytics.

i'm so sorry

Spoiler!

as the 2012-2014 Leafs under Carlyle used to say..."Keep calm and Corsi's wrong."

[EldrickOnIce]

Originally Posted by CorsiHockeyLeague

I think the column labeled "reliability" there contradicts that. You were right earlier when you said the predictive power is by no means staggering; it's just better than everything else.

Goal differential is a better predictor.
Perhaps one team a year with a negative goal differential makes the playoffs. So yes, you can use smaller sample sizes to predict with it, just like you can with small corsi sample sizes.

Goal differential is just far more accurate.

As has been echoed a thousand times here, by many. You are far less likely to win being consistently out shot and outplayed. There is nothing advanced about that. But far more teams will make the playoffs who fall into this category that will make it when they are consistently outscored.

[Rerun]

Originally Posted by EldrickOnIce

As has been echoed a thousand times here, by many. You are far less likely to win being consistently out shot and outplayed. There is nothing advanced about that. But far more teams will make the playoffs who fall into this category that will make it when they are consistently outscored.

I remember back in the '70s and '80s when you'd watch those Canada vs Russia games where Canada would consistently out shoot Russia 2:1 but they'd still lose.

Shot quality and quality goaltending made a huge difference.