A little validation at least.

[Enoch Root]

Originally Posted by colbym72

I am a supporter of advanced stats and when I see things like this and wonder why Calgary hasn't done better, you have to look further. The Flames are 2nd in High Danger Chance % but 24th in High Danger Shooting Percentage. That is pretty glaring that the chances are there but they need to convert them.

Is it though? The problem with these stats is that there is far too much noise.

Maybe the conclusion should be that the Flames' high danger chances aren't as high quality or they take too long to generate them (allowing the D to be more set).

The problem is that we don't know. These stats don't give us the information to know. Which is why the conclusions some people draw are somewhat premature IMO.

[Enoch Root]

Originally Posted by Bingo

I wrestle with this too.

So the Flames generate chances at an elite rate, and prevent them enough to be 2nd in the league in five on five scoring chance splits. That is the definition of hockey success when you script it.

But they can't finish.

So they lack skill? OK ... but in my mind it takes more skill to get the puck into a high danger chance than it does to finish it, so I'm not there on that excuse.

I hear the "all high danger chances are not created equally" argument, but it's tough to quantify, and I don't tend to jump down rabbit holes with an assumption that large on the backs of hard data.

There was that study that had the Flames last in the league in passes into the high danger zone for shot attempts ... and I think there's something to that.

Additionally they are first in the league in missed shots which points to players trying to be too perfect, which I think is a huge issue as well.

No it isn't. This is the fundamental problem with supporters of possession: it's a circular argument that goes something to this effect:

1) obviously the primary goal is to outscore the opposition
2) however, analysing goals for and against is a problem because there are few of them and luck can skew the data
3) so we look for a proxy that gives us a fuller data set
4) CORSI, and its related stats (i.e. possession stats) provide the best available correlation with winning
Therefore strong possession numbers = good team

That is a flawed argument.

The correlation between possession and winning is positive, but it is not particularly strong. Nor is there any reason to believe that it is causal.

More shots does not mean more goals. The two items are correlated, but simply increasing shots does not increase goals. You can't even say that increasing high danger shots increases goals. We can assume it would, due to the correlation, but it is not cause and effect, and it is by no means a certainty.

It is absolutely reasonable to question whether there are degrees of quality to high danger chances, due to style of play.

The problem - from both sides of the argument -is that the data is far too noisy to tell. Which is why some people get their dander up when others draw conclusions from this data. The information is not clear enough to draw the conclusion that the Flames are a great team, based on the numbers, and that the results should follow.

[Bingo]

Originally Posted by Enoch Root

No it isn't. This is the fundamental problem with supporters of possession: it's a circular argument that goes something to this effect:

1) obviously the primary goal is to outscore the opposition
2) however, analysing goals for and against is a problem because there are few of them and luck can skew the data
3) so we look for a proxy that gives us a fuller data set
4) CORSI, and its related stats (i.e. possession stats) provide the best available correlation with winning
Therefore strong possession numbers = good team

That is a flawed argument.

The correlation between possession and winning is positive, but it is not particularly strong. Nor is there any reason to believe that it is causal.

More shots does not mean more goals. The two items are correlated, but simply increasing shots does not increase goals. You can't even say that increasing high danger shots increases goals. We can assume it would, due to the correlation, but it is not cause and effect, and it is by no means a certainty.

It is absolutely reasonable to question whether there are degrees of quality to high danger chances, due to style of play.

The problem - from both sides of the argument -is that the data is far too noisy to tell. Which is why some people get their dander up when others draw conclusions from this data. The information is not clear enough to draw the conclusion that the Flames are a great team, based on the numbers, and that the results should follow.

For sure, but that's not what I'm saying.

I'm saying as a hockey coach my goal is to have my team take the ice and out play the opposition every night.

They do that and I'll take my chances with the randomness of the results.

I'm certainly not implying that there's a formula that guarantees success from underlying numbers.

[Five-hole]

Originally Posted by Flames Fan, Ph.D.

If conventional medical treatments had working correlations on par with hockey's advanced stats, no one would ever visit a doctor.

So?

Literally the only connection between the success of conventional medical treatments and the predictive capacity of hockey statistics is that they are both expressed as a percentage.

[nfotiu]

Originally Posted by FlamesAddiction

I wonder if some teams buy into the Corsi thing too much though. A good team will naturally have a good Corsi rating as their systems and skill level will produce more shots and scoring chances.

If a poor or mediocre team is trying to increase their Corsi by simply taking more shots without the skill or system in place to capitalize, then the Corsi just become a mirage and even a hindrance as it puts the carriage before the horse.

Corsi should be the result, not the goal. This is why I feel the focus on analytics is missing the point.

I agree with this statement. There's a big difference between correlation and causation, and I haven't seen anything from the hockey stats guys to show that their contrived stats do anything except generally correlate with results.

Maybe there are papers out there that prove that they have isolated luck factors, and that team's systems, strategies, and individual talents can affect results over the long term, but I sure haven't seen anything that is convincing.

I'm a strong believer in advanced stats for baseball. Baseball can be quantified much more easily though, since a baseball season is made of thousands of one on one match-ups over a year and allows for isolating factors that players don't have any control over. Maybe hockey can get there some day with technology and sensors on players and pucks, but they really just aren't there yet.

[Erick Estrada]

Even football is easier to quantify than hockey because it operates on a play by play basis and most players match up on specific types of players (DB on WR, DE on Tackle, etc) so each play is a snap shot where you can see which player failed or passed their assignments on that particular play. Hockey it's much more fluid and random so there's a lot more grey area and subjective conclusions you can draw but I don't discount that these methods can evolve over time and improve.

[mikephoen]

Originally Posted by Bingo

I wrestle with this too.

So the Flames generate chances at an elite rate, and prevent them enough to be 2nd in the league in five on five scoring chance splits. That is the definition of hockey success when you script it.

But they can't finish.

So they lack skill? OK ... but in my mind it takes more skill to get the puck into a high danger chance than it does to finish it, so I'm not there on that excuse.

I hear the "all high danger chances are not created equally" argument, but it's tough to quantify, and I don't tend to jump down rabbit holes with an assumption that large on the backs of hard data.

There was that study that had the Flames last in the league in passes into the high danger zone for shot attempts ... and I think there's something to that.

Additionally they are first in the league in missed shots which points to players trying to be too perfect, which I think is a huge issue as well.

The biggest problem I have with advanced stats is this assumption that you have actual/enough hard data. The data currently available is full of noise, errors and outright lies (like edmonton's shot counts from early this season. Or LA's home stats from pretty much every game).

So much of it is subjective and isn't consistently collected from rink to rink. There are so many plays that could be shot attempt, or could be a pass attempt, who is making that call? When a guy takes a shot from around the edge of the high danger area, who is measuring that to see if it should count as high danger or low danger?

There is also the assumption that there are enough shots and shot attempts to constitute a large enough sample size. That has never been proven as far as I know. Sure, there are more shots than there are goals, but just because there are more, doesn't mean there are enough to be statistically accurate.

The data is pretty weak and anything derived from it should come with a HUGE error bar.

[Bingo]

Originally Posted by mikephoen

The biggest problem I have with advanced stats is this assumption that you have actual/enough hard data. The data currently available is full of noise, errors and outright lies (like edmonton's shot counts from early this season. Or LA's home stats from pretty much every game).

So much of it is subjective and isn't consistently collected from rink to rink. There are so many plays that could be shot attempt, or could be a pass attempt, who is making that call? When a guy takes a shot from around the edge of the high danger area, who is measuring that to see if it should count as high danger or low danger?

There is also the assumption that there are enough shots and shot attempts to constitute a large enough sample size. That has never been proven as far as I know. Sure, there are more shots than there are goals, but just because there are more, doesn't mean there are enough to be statistically accurate.

The data is pretty weak and anything derived from it should come with a HUGE error bar.

The building thing has been taken out, first by having venue adjusted numbers, and then I think again because the venues that were outed self corrected to some degree.

So that's no longer a problem.

Beyond that I don't think there's any issue with the data itself it's with the assumptions that come with them.

I wouldn't debate the fact that the Flames have had more shots inside the other team's home plate than they've given up. But I do think it's fair to debate that that's not a great way to measure scoring chances, or that the plate is too big allowing a huge disparity in value within the region.

But the data itself is pretty straight forward.

[jayswin]

I'm not sure how to articulate this properly, especially up against advanced stats but I'll try to explain what I perceive to be an issue using the eye test.

Our team's underlying numbers appear to always suggest we're not finishing and should have more goals and a better record. Could there be anything to my theory that our system is quite rigid and requires perfect (or at least very flawless) execution, which our players struggle to execute game in and game out, making the hardest part (putting the puck in the net) very difficult?

I believe the players basically put everything they have into making the right move with or without the puck at every turn with the end goal being that this flawless system hockey gives us the high danger/corsi/advanced stats to win games. The idea being that if you do every little thing right the goal at the end should be the easy part.


The players look like they squeeze their sticks and constantly focus on exact plays in all situations and struggle to play natural, loose but intense hockey which goal scorers thrive on to put the puck in the net.

Basically our system is so "perfect" that other teams don't appear to need to worry about anything unexpected happening from our offense and our scorers likewise struggle to surprise the opposition with their shot attempts. Sure, on paper we get so many chances, but the eye test looks like the other team (including goalie) is very prepared for everything we throw at them.

Does that make any sense?

[djsFlames]

Originally Posted by madmike

The Flames advanced stats have been solid all season. 5-on-5 I think the team is fine.

But you can't have a PP in the bottom 3rd of the league as well as a PK in the bottom half and expect to make the playoffs.

100% the Flames' largest hindrance to winning at the rate of a top team.

It hasn't been chances generated or level of 5 on 5 play at all. An 8-10% increase in PP production would have turned a good handful of losses into wins and would put them on Vegas' tail at this point.

It's really not that complicated. They had issues closing out game recently and those mental hurdles happen, but a productive PP would have given them enough breathing room to likely close most of those games out as well.

[jayswin]

Originally Posted by Bingo

The building thing has been taken out, first by having venue adjusted numbers, and then I think again because the venues that were outed self corrected to some degree.

So that's no longer a problem.

I mean this respectfully Bingo, but you come off a little biased and defensive when defending stats that you use quite heavily. I mean you state it's been taken out, but then state that you "think" the venues have been self corrected to some degree and then finish by saying "so it's no longer a problem".

What do you mean by you think? Like someone said something? There was a news release or an insider suggested it? You're assuming? Those were some pretty ambiguous points made with very sure statements bookending them.

Also, could you explain and show the "venue adjusted numbers"? (asking sincerely as I hadn't heard that had happened).

[GranteedEV]

I believe the problem is that the Flames play road games at home.

The Flames are 5th in the NHL in road GF% at 52.75% - this is a great stat. For comparision, 18th place Chicago is 46.82% on the road.

However, at home, where the 5th place team in GF% is sitting at a whopping 59.55%, the Flames are sitting at a mediocre 18th place at 51.7%.

In a vacuum, 51.7% is a good overall GF%, but it's not a good GF% for a home team.

I believe this is because Gulutzan is not good at playing the game within the game... dictating forechecking assignments and that sort of thing. All the shots in the world don't matter if Curtis Lazar is taking them. We simply don't dictate the play even when we have the last change.

[EldrickOnIce]

^ without doubt Bingo, the data is the data. The problem is absolutely assumptions.
And the correlation is playing good hockey, which isn't the same as winning, unfortunately.
And there's plenty of reasons why the results might not match the 5v5 play.

[the2bears]

To me it makes sense that the stats from various venues would have been "mostly" corrected over the last few years. Most teams are using stats as an additional way of evaluating their teams which only makes sense if your data is relatively good. With the exception of the Oilers, I would expect most teams to strongly encourage the data gathered at their home venue to be good. Otherwise, all the new analysis they're doing would be for nought.

[Housley4Prez]

I feel like we're going to look back on this era as the dark ages of statistical analysis. Our analytics "experts" are just throwing spaghetti at the wall, to see what sticks. When in reality, they're the medicinal equivalence of medieval doctors performing treatments with leechings and blood lettings. We have no clue if building a team around CORSI numbers wins Cups or not. What we DO know, is that you win games if you score more than the other team... and the Flames haven't been doing that consistently enough for me to think we're actually a good team. If we ain't winning, we ain't a good team.

[Flames Fan, Ph.D.]

Originally Posted by Five-hole

So?

Literally the only connection between the success of conventional medical treatments and the predictive capacity of hockey statistics is that they are both expressed as a percentage.

At their root both systems rely on statistical analyses to determine if there is a reliable cause and effect relationship between the measured variable and the desired outcome. It's not the expressed percentages that matter, but whether those percentages are actually reporting on an effect that reliably grades with outcome. And that fundamental judgment is made by relatively similar statistical analyses across disciplines.

So my point is that at its root, the threshold of "stringency" in hockey's current crop of advanced stats is nowhere near acceptable*. If hockey advanced stats and medicine used similar thresholds, echinacea would qualify as an acceptable treatment for everything that ails us.


* This should be obvious from the discussions themselves. There are smart hockey people on this forum who passionately come out on opposite sides of the analytics debate. That tells you it's basically a subjective judgment and that the underlying correlation ratios are low.

[Five-hole]

Originally Posted by Flames Fan, Ph.D.

So my point is that at its root, the threshold of "stringency" in hockey's current crop of advanced stats is nowhere near acceptable*. If hockey advanced stats and medicine used similar thresholds, echinacea would qualify as an acceptable treatment for everything that ails us.

You imported your own threshold for acceptable stringency and are trying to pass it off as a standard.

It may surprise you to learn that far less time, money, and effort has gone into explaining hockey by way of statistics than the enterprise of science based medicine. One of the disciplines is newer and, ya know, slightly less important. Holding advanced stats in hockey to the same standard as cancer research is pretty silly and doesn’t prove your point.

The question you should be asking is whether the stats being discussed are the best available explanatory tool or not (and why!), not whether the enterprise compares favorably to science based medicine. That would be a productive and interesting discussion.

“It’s not perfect now so we should simply stop” is not a great position.

Or you could continue to be dismissive and spend time expounding upon what a percentage expression of a causal relationship is. Up to you.

[ricardodw]

every single team in the league has a better SAT % when trailing than leading.

If you believe that the shots attempted for and against are what defines a superior team then it would be simple to pull the goalie for the first 2 goals and always be playing from behind.

You would lose a lot of games and not make the playoffs but would have good looking Corsi and fancy stats.


Really hard for teams to keep the pedal down when playing with a lead and always a push from trailing/losing teams.


Common sense test. With 2 competitive teams who will win the game going into the 3rd period with a 2 goal lead? Who will attempt more shots on goal in the 3rd period

[Flames Fan, Ph.D.]

Originally Posted by Five-hole

You imported your own threshold for acceptable stringency and are trying to pass it off as a standard.

It may surprise you to learn that far less time, money, and effort has gone into explaining hockey by way of statistics than the enterprise of science based medicine. One of the disciplines is newer and, ya know, slightly less important. Holding advanced stats in hockey to the same standard as cancer research is pretty silly and doesn’t prove your point.

The question you should be asking is whether the stats being discussed are the best available explanatory tool or not (and why!), not whether the enterprise compares favorably to science based medicine. That would be a productive and interesting discussion.

“It’s not perfect now so we should simply stop” is not a great position.

Or you could continue to be dismissive and spend time expounding upon what a percentage expression of a causal relationship is. Up to you.

The time to run the stats is trivial. On top of it all, it's much easier to accumulate data in hockey, so the timeline comparison to medicine isn't all that relevant.

So how about r > 0.9? Let's scrap any talk of advanced stats that don't meet that threshold.

[Mike F]

Originally Posted by Insane_Flame

http://www.hockeybuzz.com/boards/thr...323&forum_id=1"
"The stat that, over time, correlates most strongly with winning is Corsi. Here are the rankings for the following teams (via naturalstattrick,com)

According to this report, that's not true:

The strongest correlations to win% is goals for (0.853), goals against (-0.817), goal for % (0.974), shots for (0.517), shooting % (0.799), save % (0.704), PDO (0.884), power play % (0.601), penalty kill % (.-0.462), power play goals for and against (0.513 and -0.512). Interesting note is all of the ones with an offensive and defensive side, the offensive side correlates a little higher. This means than in the NHL today, offense correlates more with wins than defense

As far as corsi and fenwick go, fenwick (FF% 0.440) does correlate higher to win % than does corsi (CF% 0.346). But there is some interesting things going on. FF/60 is 0.402 and CF/60 is 0.422, fairly close but an edge to corsi. However with against: FA/60 is -0.325 (again we see defensive stats having a smaller impact) and CA/60 is -0.095, or statistically insignificant. This shows that total shot attempts against doesn't matter that much, but teams that block more win a higher percentage of games