Lets talk about Auston Matthews.

[The Hanging Jowl]

I love people that are so selective in the facts they choose for their argument. How significant was Oakland's improvement after using advanced stats? How many championships did the Red Sox get after copying moneyball tactics, vs. the 100 years before? How many teams copied Oaklands moneyball tactics since they started using it?

The point wasn't that one example proved the tactic of moneyball wrong, it just proved it obviously doesn't always work which puts it squarely in the category of one strategy amongst many that may or may not be effective. I wasn't "selective"; I made use of a very handy and apt example directly tied to the birth of the very concept of moneyball.

 

[LeafFever]

MLB has no cap. Oakland was competing against the Yankees with less than 25% payroll and budget. The Leafs are in a capped league with the biggest budget outside of players in the NHL.

 

[The Hanging Jowl]

Coefficient of determination. I think the point you are going for is coefficient of correlation (which you square to get R^2). Correlation does not fit a line. Correlation only tells you how well two variables relate to one another. They are related, but not the same.

Do you guys really have to do this to me? I've been out of university for 25 years and I still have PTSD over crap like this.

 

[PromisedLand]

Coefficient of determination. I think the point you are going for is coefficient of correlation (which you square to get R^2). Correlation does not fit a line. Correlation only tells you how well two variables relate to one another. They are related, but not the same.

For a predictive model to be a bit efficient any variables that are chosen as an explanatory variable better have high correlation with the depending variable otherwise predictions generated from any model is going to be ineffective.

R^2 presents a stat that actually tells you what percentage of your model is explained by the independent variable. Most of the folks that put up stats on their websites do not post any r^2. Would you trust the results of the model that has low r^2?

Your parameters in any regression model in simple terms is nothing but a slope. Th convergence to mean theory only works if model is very well specified.

As an example, in your previous post (not this one) you had mentioned analytics use probabilities to derive the “expected value” (I.e. mean)

However in Hockey the expected outcomes should be modeled using conditional probabilities not simple models that many stats blowhards use

Example(simple)

HDCF%: the model is not allowed to be adjusted for high danger areas where the shooter was actually well defended or left wide open or did the shooter bobble the puck or etc......

There is no dataset that currently exists to allow for such analysis; the results can be heavily biased.

Another example: PDO = SV% + SH%

Both Sh% and Sv% both these metrics are a mathematical fiction of of two variables

Sv% (team defense, opposition offense, goalie)

Sh%(team offense, opposition defense, opposition goalie)

There are no metrics available to quantify which one of these are driving either the sh% or sv% yet people are quick to use PDO as a bottom line without understanding the limitation of the stat and the kind of bias it already has.

I could go on and on and on but there are people who

1. get the math and stat and limitations associated with it,

2. there are people that are completely blind to the limitation because their knowledge of stats is weak (or their ego is massive to accept the limitations-what I say is right screw everyone else)

3. and there are people who completely ignore the stats.

Some fall in the first category (I’m there), majority of the posters who use stats here fall in the second category and the remainder in the third category. It’s a shame that vocal community belongs in the second category (look at mirtle; he is a journalist but using stats like they are TRUTH; if he was a statistician he would exercise caution and if he was a mathematician he would throw the probabilities out of the window)

 

[LeafsFan89]

For a predictive model to be a bit efficient any variables that are chosen as an explanatory variable better have high correlation with the depending variable otherwise predictions generated from any model is going to be ineffective.

R^2 presents a stat that actually tells you what percentage of your model is explained by the independent variable. Most of the folks that put up stats on their websites do not post any r^2. Would you trust the results of the model that has low r^2?

Your parameters in any regression model in simple terms is nothing but a slope. Th convergence to mean theory only works if model is very well specified.

As an example, in your previous post (not this one) you had mentioned analytics use probabilities to derive the “expected value” (I.e. mean)

However in Hockey the expected outcomes should be modeled using conditional probabilities not simple models that many stats blowhards use

Example(simple)

HDCF%: the model is not allowed to be adjusted for high danger areas where the shooter was actually well defended or left wide open or did the shooter bobble the puck or etc......

There is no dataset that currently exists to allow for such analysis; the results can be heavily biased.

Another example: PDO = SV% + SH%

Both Sh% and Sv% both these metrics are a mathematical fiction of of two variables

Sv% (team defense, opposition offense, goalie)

Sh%(team offense, opposition defense, opposition goalie)

There are no metrics available to quantify which one of these are driving either the sh% or sv% yet people are quick to use PDO as a bottom line without understanding the limitation of the stat and the kind of bias it already has.

I could go on and on and on but there are people who

1. get the math and stat and limitations associated with it,

2. there are people that are completely blind to the limitation because their knowledge of stats is weak (or their ego is massive to accept the limitations-what I say is right screw everyone else)

3. and there are people who completely ignore the stats.

Some fall in the first category (I’m there), majority of the posters who use stats here fall in the second category and the remainder in the third category. It’s a shame that vocal community belongs in the second category (look at mirtle; he is a journalist but using stats like they are TRUTH; if he was a statistician he would exercise caution and if he was a mathematician he would throw the probabilities out of the window)

Your post seems reasonable, but I don't recall making the post you said I made though.

Admittedly, I am not into hockey analytics (yet), but as you alluded to, hockey is not yet there in terms of reliable variables. Also, I'd offer a correction to your post in that it is possible to have variables that are not well correlated with a DV alone, but when taken together (X and Z) can explain a lot of variability in Y (suppressor variables). I would also be more interested in unsupervised learning techniques in hockey analytics.

 

[PromisedLand]

Your post seems reasonable, but I don't recall making the post you said I made though.

Admittedly, I am not into hockey analytics (yet), but as you alluded to, hockey is not yet there in terms of reliable variables. Also, I'd offer a correction to your post in that it is possible to have variables that are not well correlated with a DV alone, but when taken together (X and Z) can explain a lot of variability in Y (suppressor variables). I would also be more interested in unsupervised learning techniques in hockey analytics.

My whole concern is that people use stats as TRUTH which they are not; there are limitations associated with it and these are not talked out loud. I could regress apples grown in Mexico to Big Macs consumed in Canada and it will give me a r^2 but what the bell does that even mean? They may be even highly correlated but does that mean when there are more apples in Mexico people in Canada consume more Big Macs? Is apple growth in Mexico a good “predictor” to Big Macs consumed in canada?

There is plethora of info on how these metrics are calculated. Except when they calculate XGf% they never specify their model there in itself is a place for doubt.

Google Natural statrick or Corsica they define variables and what they are trying to approximate if you are interested in it

A goood example of limitations of these metrics: look at team CF% in regular season and then sort it by highest value and then compare how many teams made the playoffs.

I am not saying CF% is useless but on its own without context and other metrics it is not meaningful. But posters here would use CF% as an absolute truth and this is what grinds my gears

EDIT:
Why summary stats can be misleading?
Anscombe Quartet did an amazing job; none of the stats blowhards while running regressions or using summary stats look at that IMO
Anscombe's quartet - Wikipedia

Also; if you get a chance read a paper Achen (2004); pretty neat how the arguments are provided. I haven't read that paper for a while but remember that it had a profound impact on how I should "model" things

Name of the paper I belive was called "Let’s Put Garbage—Can Regressions and Garbage—Can Probits Where They Belong"

 

[Suntouchable13]

Babcock needs to unleash Matthews next season. Enough is enough, it's his 3rd season already. He should be our horse. All the great centres play a lot.

 

[Daisy Jane]

babcock doesn't need to unleash matthews
you want him to.

that is a big difference.

 

[LeafsLegendAkiBerg]

Babcock doesn't "need" to do anything. You could give that response to literally any discussion about lineup/ice time decisions.

 

[Kurtz]

Has anyone here looked at what the Bergeron-Chara combo did to TB's first line of Miller,Stamkov, Kucherov? Tampa's first line had something like a combined 5 even-strength points against that coverage.

This board (and fans in general) always rush to pass judgement on their player's performance, but rarely consider the opponent's role in said performance.

 

[Muston Atthews]

babcock doesn't need to unleash matthews
you want him to.

that is a big difference.

I mean, you can keep telling yourself that but if we want to be successful we need to do it on the back of our big man

 

[Mikeyg]

Guys listen to ole mikeyg for a minute. Imagine there was a team that had one of the best one shot scorers on the planet AND one of the best passers on the planet. Now imagine if I told you that they only play with each other for a handful of minutes each year, and when the shooter is struggling he gets paired with a third line grinder in response. Who here thinks that the coach of that team is the right fit? Theoretically of course, cant see any resemblance to an nhl team.

 

[Daisy Jane]

I mean, you can keep telling yourself that but if we want to be successful we need to do it on the back of our big man

and we are successful without him having to play 20++ minutes
the kid is plenty dominant as he is playing what he is now. so i dont have to "tell myself" anything. it's right there and it's obvious.

i mean if we're losing (or he's feeling it one night) sure throw him over the boards every other shift, since he can game break, but for the most part, Matthews is more than capable than dominating the game, playing less minutes, vs. riding him like a pony.

if he was struggling to do so and it was obvious he needed more time, then i wouldn't say anything.

 

[Muston Atthews]

and we are successful without him having to play 20++ minutes
the kid is plenty dominant as he is playing what he is now. so i dont have to "tell myself" anything. it's right there and it's obvious.

i mean if we're losing (or he's feeling it one night) sure throw him over the boards every other shift, since he can game break, but for the most part, Matthews is more than capable than dominating the game, playing less minutes, vs. riding him like a pony.

if he was struggling to do so and it was obvious he needed more time, then i wouldn't say anything.

Another first round exit won’t be successful. I’d even say losing in the fashion we did doesn’t make this past season a success. I get your stance, “If it ain’t broke don’t fix it”. Just because something work doesn’t mean it can’t be better

 

[Daisy Jane]

Another first round exit won’t be successful. I’d even say losing in the fashion we did doesn’t make this past season a success. I get your stance, “If it ain’t broke don’t fix it”. Just because something work doesn’t mean it can’t be better

how the first round exit happened and Matthews playing time is a comparing an apple to an orange.

I think what a lot of people forget that the minutes etc is worked in tandem with the sports science department to help mitigate wear and tear we are one of the healthier teams in the league. minus actual "in game injuries" (ie blocking pucks (Z) or being smushed (Matthews x 3)" Matthews if not for the injuries he sustained was on pace to better his rookie performance playing the same amount of minutes. So that would imply, that it's better, so to quote you "why fix it?"

And. no. for your information. that is not my stance. i would fix a hell of a lot of things, but this insistent belief that Matthews needs to play 22 minutes to be more dominant than he already is isn't one of them. and it doesn't change the fact that there is a difference between need and want.

Matthews does not need to play more minutes to be dominant. He's dominant already. He was proving he was even better than the year before

People want Matthews to play more minutes for (insert reasons here).

 

[IPS]

Guys listen to ole mikeyg for a minute. Imagine there was a team that had one of the best one shot scorers on the planet AND one of the best passers on the planet. Now imagine if I told you that they only play with each other for a handful of minutes each year, and when the shooter is struggling he gets paired with a third line grinder in response. Who here thinks that the coach of that team is the right fit? Theoretically of course, cant see any resemblance to an nhl team.

I wouldn't expect any reasonable response from the Babcock worshippers. The whole "they both need to drive a line" excuse is just ridiculous. We have the some of the best offensive depth in the league. How come Tampa has Kuch and Stammer together? Don't they they need to drive 2 different lines? What about Stamkos and St. Louis?

We got a different flavor of that kind of combo and a moron coach who refuses to utilize it.

 

[Martin Skoula]

I wouldn't expect any reasonable response from the Babcock worshippers. The whole "they both need to drive a line" excuse is just ridiculous. We have the some of the best offensive depth in the league. How come Tampa has Kuch and Stammer together? Don't they they need to drive 2 different lines? What about Stamkos and St. Louis?

We got a different flavor of that kind of combo and a moron coach who refuses to utilize it.

He doesn't need to do it every game from now on forever. Just.. what's the harm in trying for a full game or two?

 

[Daisy Jane]

He doesn't need to do it every game from now on forever. Just.. what's the harm in trying for a full game or two?

none. which is basically what i said. if he's having a game play him more. if not, play him the regular amount of minutes he's having now. there's no real need for him to do so.

 

[Martin Skoula]

none. which is basically what i said. if he's having a game play him more. if not, play him the regular amount of minutes he's having now. there's no real need for him to do so.

That was about Matthews playing with Marner. No other team splits their top-end forwards like we do, especially when their skillsets complement each other so well. It seems insane not to give it a try for longer than half a period in a game we're already losing.

 

[Daisy Jane]

That was about Matthews playing with Marner. No other team splits their top-end forwards like we do, especially when their skillsets complement each other so well. It seems insane not to give it a try for longer than half a period in a game we're already losing.

i would argue because it makes us dangerous?
like for me, I don't have an issue with a lot of things with the forwards, I'd make a few alterations (ie: move hyman down, maybe move AJ up (see how that works, it would give that line a a lot more speed etc) - but for the most part it's plenty dangerous as it is.

could it be better, arguably, yes - but my argument would be it can get better how it is now (case in point, Matthews was on pace to be better, and I think Marner and Nylander will have better starts ergo have better years next year). if we're losing sure, do all crazy sort of things, i've never quibbled that at all. but if it's just "why not play matthews more." the question is logically going to be why, and I don't think the answer should be "because every centre ____________" same with playing Marner and Matthew. i mean sure play them together (and I'd argue the best way to do that is just to move Matthews to Marner's unit, and that should get that started, and Let Kadri and Will make some magic on the other unit). but do they need to. i don't really think so.

 

[FreeBird]

Well Mathews reach the complete level that Crosby, Toews, and McDavid bring to the table on a consistent basis.

 

[djdev]

Has anyone here looked at what the Bergeron-Chara combo did to TB's first line of Miller,Stamkov, Kucherov? Tampa's first line had something like a combined 5 even-strength points against that coverage.

This board (and fans in general) always rush to pass judgement on their player's performance, but rarely consider the opponent's role in said performance.

ive said this so many times since we were eliminated..they are the best defensive unit in the league imo and are lethal offensively too.

 

[Damisoph]

Well Mathews reach the complete level that Crosby, Toews, and McDavid bring to the table on a consistent basis.

That's quite the eclectic list of players.

 

[biotk]

Has anyone here looked at what the Bergeron-Chara combo did to TB's first line of Miller,Stamkov, Kucherov? Tampa's first line had something like a combined 5 even-strength points against that coverage.

This board (and fans in general) always rush to pass judgement on their player's performance, but rarely consider the opponent's role in said performance.

TB's first line almost never faced Bergeron. They faced Chara, and Chara shut them down, just like he shut down Matthews (who also didn't face a lot of Bergeron - although far more than Kuch and Stamkos did).

 

[The Legend]

P

For a predictive model to be a bit efficient any variables that are chosen as an explanatory variable better have high correlation with the depending variable otherwise predictions generated from any model is going to be ineffective.

Predictive (and regression) models have zero to do with correlation.