Mikel Arteta: Aston La Vista To The Title?

[MartiSaka]

One of the soghts had our expected goals at .64 per game multiply that by the remaining goals. If the methodology is correct then we should score 2 goals in 3 matches on current form.

That is a very **** prediction approach to be fair

 

[Farzad Stoned]

All the variables inaccurately short handed by Xg are huge. What player gets the chance, how many defenders block his goal path, how good is the keeper, how good is the ball he receives, how much time does he have on the ball in that position, how close is his marker? Like I said it’s an oversimplification sampling error multiplied times a million.

 

[Farzad Stoned]

Take the Sp**s match, who had more of the better chances? Was Arsenal ever in danger of losing that match? 3 of these sites says Arsenal should expect to win another says Sp**s, all show the match as much closer than it actually was.

 

[Barry]

One of the soghts had our expected goals at .64 per game multiply that by the remaining goals. If the methodology is correct then we should score 2 goals in 3 matches on current form.

The 0.64 number is dodgy. Dunno where that came from but it's not our xG. The real number is around 1.45 from our first 11 matches - something like that

 

[MartiSaka]

It is less useful than shots, shots on target or simply watching to see which team creates most of the best chances. That is the problem I have it has limited usefulness and is more skewed than just looking at a composite of shots, shots on target, and chances created. This supposedly objective rating criteria is more complex and less telling than the base data they use to put in their rating system. The methodology is an oversimplification error multiplied by a million, the percentage chance of any particular shot scoring has a huge number of variables not addressed by the model or over simplified by the model.

But u are suggesting stats can be used to give some indication of chance creation. And these stats are the basis of xG. All xG is using is a large amount of past data about the likelihood a player will score from a shot in a particular context (position on field/angle/defenders around, shot type) and providing a value between 0-1 ( a probability they will score based on past outcomes) so not all shots are treated equally (because they are not). So it's similar to what u said but with a lot more context about if it's highly probable the shot would have gone in based on past outcomes. Surely this will give u a better metric as it's built from the success of a shot based on past matches.

 

[Kav]

I’m not a fan of using stats to judge a team because it does not capture the various elements and factors and context in which phases of play occur. It only looks at the specific actions that it is trying to capture.

That being said it is a tool, nothing more than that just a tool to help us look at the different things that occur in the game. It is by no means some criteria to judge a player on. For example player A takes 10 shots per game while Player B takes 2. I’m sure someone somewhere would use those stats and say that player A is the better player because he is shooting more. The truth of it is that this stat doesn’t give us any information that is useful and without context it is meaningless.

That being said we can’t dismiss all statistical data generated during a game either. For example shots on target or saves made by goalkeepers often have more value than shots taken by team. The stats by themselves won’t tell you much without context.

You can also have a ****ty defender who happens to make a lot of tackles (Mustafi) and the stats with highlight his actions vs a good defender who’s positional play allows him to make less tackles (Varane,Silva). So stats are never some crystal ball to be used to determine the future or the quality of a player.

 

[Rex Stone]

Had Pulisic come in for £58m the summer he joined as well

Didn’t they make Kovacic permanent for 40M as well?

 

[Farzad Stoned]

The 0.64 number is dodgy. Dunno where that came from but it's not our xG. The real number is around 1.45 from our first 11 matches - something like that

Yeah that is my point one of these sites is dodgy, most likely the other 3 aren’t much better at predicting anything because they use the same shot methodology

 

[Farzad Stoned]

But u are suggesting stats can be used to give some indication of chance creation. And these stats are the basis of xG. All xG is using is a large amount of past data about the likelihood a player will score from a shot in a particular context (position on field/angle/defenders around, shot type) and providing a value between 0-1 ( a probability they will score based on past outcomes) so not all shots are treated equally (because they are not). So it's similar to what u said but with a lot more context about if it's highly probable the shot would have gone in based on past outcomes.

Yes the base data is better than the XG rating. So why use stupid Xg?

 

[MartiSaka]

Yes the base data is better than the XG rating. So why use stupid Xg?

Surely a metric that's built from the success of a shot based on past matches is better than number of shots alone. For instance, in your example of the base stats, a shot from outside the box with a stacked defence is treated as likely to score as a one on one with keeper. They are both a value of 1. On the other hand, xG would give a much higher value to the second situation than the first based on the fact that data from past matches has computed the likelihood of scoring in second situation is much higher.

 

[Squillaci]

You guys are extremely bored

 

[Trilly]

Yeah that is my point one of these sites is dodgy, most likely the other 3 aren’t much better at predicting anything because they use the same shot methodology

If the other three used the same shot methodology then why are their numbers different? That literally doesn't make sense. Which site is the dodgy one anyway?

 

[Makingtrax]

Strange, not a single argument on RAWK about the use of xG after this post.

We're separated from them by 2 points.... but 20 goals (in gd terms) and about a goal and a half per 90 in xg terms ... wild

 

[Farzad Stoned]

Surely a metric that's built from the success of a shot based on past matches is better than number of shots alone. For instance, in your example of the base stats, a shot from outside the box with a stacked defence is treated as likely to score as a one on one with keeper. They are both a value of 1. On the other hand, xG would give a much higher value to the second situation than the first based on the fact that data from past matches has computed the likelihood of scoring in second situation is much higher.

I disagree too many subjective calls in Xg where the actual raw data is objective hard numbers. I don’t like stats that vary as wildly from site to site, that tells me a high level of subjectivity is involved. If you want a subjective measure nothing beats watching the actual match in entirety. These type of rating systems are only helpful if they tend to predict something in the future or tell you the form of the team I think if I am not mistaken one of the sites has us in Norwich and Newcastle level attack sorry just not indicative of reality.

 

[Farzad Stoned]

If the other three used the same shot methodology then why are their numbers different? That literally doesn't make sense. Which site is the dodgy one anyway?

Whichever one people were bandying about with a .64 number. No if you see a tight range of results with more uniform methodology then that would cut to this stat being more reliable. 4 wildly different answers to an objective question means almost certainly you got 3 wrong answers or 4 wrong answers

 

[MartiSaka]

I disagree too many subjective calls in Xg where the actual raw data is objective hard numbers. I don’t like stats that vary as wildly from site to site, that tells me a high level of subjectivity is involved. If you want a subjective measure nothing beats watching the actual match in entirety. These type of rating systems are only helpful if they tend to predict something in the future or tell you the form of the team I think if I am not mistaken one of the sites has us in Norwich and Newcastle level attack sorry just not indicative of reality.

The word you want is variance (i.e. spread in xG results) not subjectivity. Just because the different xG models calculate different values, it doesn't mean the numbers are being made up. They are using slightly different methods/data sets to calculate the xG so the numbers will be different. But they shouldn't vary significantly (i.e. something like 50% difference) and pretty sure won't for the reputable models. I don't know anything about the reputability of the models you are comparing.

A good link for posters interested in how xG is calculated:

xG Explained | FBref.com

Find out what expected goals are and how they are calculated

fbref.com

 

[OnlyOne]

@Trilly has gone from potential AMHOF to getting bodied by a brilliant new poster in @farzad4daarse. I could see that Trusted Status switching hands...

 

[Lady_Gooner]

xG has had this forum in a chokehold this intl break. Free us

 

[Sanchez11]

xG has had this forum in a chokehold this intl break. Free us

Blame the usual suspects!!

 

[A_G]

Whichever one people were bandying about with a .64 number. No if you see a tight range of results with more uniform methodology then that would cut to this stat being more reliable. 4 wildly different answers to an objective question means almost certainly you got 3 wrong answers or 4 wrong answers

0.64 was the number of chances created per game, not xG.