Expected Goals - what a bunch of garbage

[Midnight Judges]

You guys can correct me if I'm wrong but xGoals does not assess the quality of individual player's shots and factor that into predictions, right?

Somebody mentioned to me a few years ago that they were considering that. Not sure if it has changed since then.

Okay yes it does, but in a sort of wonky one-size-fits-all shot multiplier method.

 

[supsens]

Ok so you're not here for a conversation at all you're here to miss the point entirely and keep talking about math you don't grasp.

Here is from what you posted
——————————————————————————————
Put it simple, what the percentage of shots with same variables were goals.
And that’s your xG value.
When you take 10 or 20 shots, you can probably determine yourself just by looking on those shot features/attributes and say which ones were repetitive for shots that led to goals, but when we talk about half a million of shots you need an algorithm to do it for you.
————————————————————————————————
Now they have taken the average and that’s all it is. I have no idea what you think you are seeing or how you think they come up with the magic number but in the end it’s average number of goals scored. Average shooting %
Why do you insist that I don’t understand when clearly the basics of the entire thing is lost on you.

 

[ResilientBeast]

Here is from what you posted
——————————————————————————————
Put it simple, what the percentage of shots with same variables were goals.
And that’s your xG value.
When you take 10 or 20 shots, you can probably determine yourself just by looking on those shot features/attributes and say which ones were repetitive for shots that led to goals, but when we talk about half a million of shots you need an algorithm to do it for you.
————————————————————————————————
Now they have taken the average and that’s all it is. I have no idea what you think you are seeing or how you think they come up with the magic number but in the end it’s average number of goals scored. Average shooting %
Why do you insist that I don’t understand when clearly the basics of the entire thing is lost on you.

OMG it's not the bloody average

The next line in the article

is "After that they can categorize (classify) which of the features had bigger weight in shot to become a goal"

They explain how they figure it out with "Then using statistical modeling (usually in Python or R, programming language) they create mathematical algorithm, (known as Logistic Regression or Gradient Boosting, these are the most common approaches because they tend to give the best “out of the box” results compared to the amount of time it takes to design them) to determine in how many instances given similar features (attributes) the vector Y value ended being 1."

They take the percentage of goals with a certain attribute and use regression to find the weight of each value and how they impact the outcome (goal/no goal). Then multiply the various factors like I listed up thread of a particular shot by the weighting determined in their regressions.

Again this is not the average

 

[supsens]

OMG it's not the bloody average

The next line in the article

is "After that they can categorize (classify) which of the features had bigger weight in shot to become a goal"

They explain how they figure it out with "Then using statistical modeling (usually in Python or R, programming language) they create mathematical algorithm, (known as Logistic Regression or Gradient Boosting, these are the most common approaches because they tend to give the best “out of the box” results compared to the amount of time it takes to design them) to determine in how many instances given similar features (attributes) the vector Y value ended being 1."

They take the percentage of goals with a certain attribute and use regression to find the weight of each value and how they impact the outcome (goal/no goal). Then multiply the various factors like I listed up thread of a particular shot by the weighting determined in their regressions.

Again this is not the average

Yes it is the average. That is what the entire point is, just because the are using computer programming to 'learn' there is no way to get an answer without averages.
Machines only understand yes and no._ on and off.
Find out total shots vrs total goals. Average shooting percentage
Break it down to percentage of goals scored from each area of the ice. Average of where goals come from. The part of ice where the most goals come from is a high danger area.
Of the high danger area what percnt were passes what percent were not.
Again it is just finding the average shooting % for any number entered.

What kind of voodoo magic do you think is going on?

 

[Morbo]

Worse than it being a worthless stat, it's adhered to like it was the laws of gravity to a lot of people.

It's very unscientific that these crude measures are never questioned at all by those that supporft them.

I think it's a good idea to try to develop these kinds of data, but the execution is lacking.

 

[Doctor No]

Worse than it being a worthless stat, it's adhered to like it was the laws of gravity to a lot of people.

It's very unscientific that these crude measures are never questioned at all by those that support them.

I'll grant your thesis (depending on what you mean by "a lot of people") - but how is that different than any other hockey statistics, or even the "eye test"? A large percentage of the population - whether on here, or in sports bars, or at arenas - takes whatever information supports their preconceived notions and uses them like they were handed down on stone tablets.

You could pretty much take "these crude measures are never questioned at all by those that support them" and apply it to any sort of argument that's offered on HFBoards.

 

[Doctor No]

And the people who do this work professionally certainly understand the limitations of what they're looking at - I can't talk specifics because of non-disclosure agreements, but there's a lot of talk about limitations because that's where the next advancements are trying to be made.

 

[hockeyes]

The "problem" with advance stats in hockey is goals are infrequent and almost no two shots, even if from the same location, are identical. Sports like Baseball don't have someone hooking the hitter just before the pitch or screening the pitcher just before they throw so it's a lot easier to put numbers to the game.

I put problem in quotes because it's not really a problem if you understand what is going on and take it for what it is.

 

[supsens]

And the people who do this work professionally certainly understand the limitations of what they're looking at - I can't talk specifics because of non-disclosure agreements, but there's a lot of talk about limitations because that's where the next advancements are trying to be made.

It is somewhat interesting but the limitations are tough to get around I was a bit of a dick in here the other day, I get annoyed seeing the one game expected goal stat posted by a fan of the loosing team every second day because they are 'clearly better'
or five page debate how a 3% higher expected goal rate after 10 games makes one guy way better when the guy is only getting 40 points for the entire year anyway.
A massive problem plotting goals like this is that teams already identify trends with players and take away the bread and butter money shots so you end up with a model that is so last year.

 

[jc17]

It's nice to see an actual conversation about frequently used metrics here, rather than the generic stats good/bad.

I use xg to help evaluate players and teams, but I certainly think there's also a misunderstanding of it among many fans that use it to draw conclusions from tiny sample sizes, because it's supposed to be a better indicator than goals.


With an r2 of .33, which I also do think is lower in predictive value than I'd hope for, even if there is a bit of luck involved in the sport, many people don't use it like that. Most people see xg as something that will predict future results with high amounts of accuracy, like 1st half of the season xg predicting 2nd half actual goals.

The issue is less with the stat and more with how people use it. There are a lot of people happy to be on the stats side of things because it's considered the "smart side", but then misinterpret many numbers

 

[Iapyi]

While this is true, does OP strike you as one of these people? If we're being honest the majority who are adamantly against them just don't understand them

Interesting. I have found that many who are adamant against them are more so against being told they have to accept them and being called stupid if they don't.

 

[Mickey Marner]

It's nice to see an actual conversation about frequently used metrics here, rather than the generic stats good/bad.

I use xg to help evaluate players and teams, but I certainly think there's also a misunderstanding of it among many fans that use it to draw conclusions from tiny sample sizes, because it's supposed to be a better indicator than goals.


With an r2 of .33, which I also do think is lower in predictive value than I'd hope for, even if there is a bit of luck involved in the sport, many people don't use it like that. Most people see xg as something that will predict future results with high amounts of accuracy, like 1st half of the season xg predicting 2nd half actual goals.

The issue is less with the stat and more with how people use it. There are a lot of people happy to be on the stats side of things because it's considered the "smart side", but then misinterpret many numbers

It's definitely the misuse and misunderstanding of these analytics that breeds the most contempt, with maybe an assist going to the hubris from some of the people creating them. The two things that baffle me the most are when people -some of whom are even the authors of the statistic(s)- post single game samples sizes of xg, rapm etc. as you'd expect simply understanding the statistic would preclude someone from doing that. The second is seeing them treated as if the conclusions are unbiased when either one person or a very small number of people have autonomy over the inputs and obviously created the statistic based on their own preferences.

 

[mcdingdong]

It's definitely the misuse and misunderstanding of these analytics that breeds the most contempt, with maybe an assist going to the hubris from some of the people creating them. The two things that baffle me the most are when people -some of whom are even the authors of the statistic(s)- post single game samples sizes of xg, rapm etc. as you'd expect simply understanding the statistic would preclude someone from doing that. The second is seeing them treated as if the conclusions are unbiased when either one person or a very small number of people have autonomy over the inputs and obviously created the statistic based on their own preferences.

THANK YOU. Especially for the crowd that always talks about 'unbiased' statistical models, etc. Anyone with a modicum of understanding in the field (or any data-driven field) knows that the bias is built in and you have to examine the inputs carefully. Crap data in, crap data out.