Match Thread [FT] Rangers 4 (Lammers 10'; Danilo 78'; Sima 84'; Dowell 90') - 0 Livingston

[Rousseau 10,611]

5 minutes ago, Bluedell said:

You said you'd love to see a new manager...

I would - mainly because I didn't want him in the first place. That's not the same as calling for him to be sacked, though, is it? 

 

I also said it would be harsh to sack MB. 

 

(And the '[Manager X]'s still available. Just Saying' posts are tongue in cheek.)

[Bluedell 5,624]

11 hours ago, Rousseau said:

 

Yes, low blocks are sitting deep. 

 

I've tried to explain xG above:

 

They take hundreds of thousands of past shots, going back decades, from every conceivable position on the pitch, and check how many times a player actually scored. 

 

In practice, that means if a chance has 0.2xG, it should be scored 20% of the time, because out of those hundreds of thousands of past shots they've looked at, the player actually scored 20% of the time. 

 

They take into account many more variables than just position on the pitch, e.g.:

 

- Type of Assist

- Distance to the goal

- Angle to the goal

- Did the player strike it with his feet or was it a header?

- In what passage of play did it happen? (e.g. open play, direct free-kick, corner kick, counter-attack)

- Has the player just beaten an opponent?

 

We do it in our heads all the time, using our judgement and past experience of watching games, when we say, 'that's a sitter - he has to score that'.

 

The xG tells you how many times players in the past have actually scored from that position - and taking into account all the other variables. 

 

An xG measurement can be generated for both teams as a whole and individual players, giving an indication as to how well they should be performing in front of goal.

 

Combining a player or team's xG ratings during the course of a season can give an approximation of how many goals a they should have scored.

 

Not only can that be used to evaluate a particular performance, it can also be used in the projection of a future or long-term performance.

Is it done using AI? How can they possibly take account of all these variables, and do it so quickly? Is it done using a TV feed or someone at the game? 

 

Once you start thinking about it, it's mind-blowing as to how it's possible to do that.

[Rousseau 10,611]

6 minutes ago, Bluedell said:

Is it done using AI? How can they possibly take account of all these variables, and do it so quickly? Is it done using a TV feed or someone at the game? 

 

Once you start thinking about it, it's mind-blowing as to how it's possible to do that.

It is indeed mind-blowing. 

 

I don't know, to be honest. 

 

Surely it's just another small step up from gambling sites? They track a mind-blowing amount of stuff, too.

 

If it's not done by AI, is soon will be.   

 

Edit: From the Opta site: 'Opta data is generated in real-time though a combination of human annotation, computer vision and AI modelling. Through this parallel human and computer fused process, we are able to generate more in-depth and detailed performance data than ever before.'

[BlackSocksRedTops 3,752]

I'll have to be honest and say that I'm relatively new to this level of statistical analysis but I'm fairly open to learning new things about the game I love. I have found Joshua Barrie's contributions on the daily Rangers Review podcast very insightful and informative. That boy loves a good diagram. 

[Gonzo79 15,293]

87% of statisticians are likely to bore you into a coma.  

[Rousseau 10,611]

1 minute ago, BlackSocksRedTops said:

I'll have to be honest and say that I'm relatively new to this level of statistical analysis but I'm fairly open to learning new things about the game I love. I have found Joshua Barrie's contributions on the daily Rangers Review podcast very insightful and informative. That boy loves a good diagram. 

What I like is the xG graphs. It's the xG for and xG against, for every match. 

 

The red line is xG against and the white is xG for. 

 

It's a bad image, but it shows that when Arteta took over in 2019/20, he increases the chances they create and reduces the chances they concede. 

 

If you go with the xG it's no surprise that they pushed City last year. 

 

I'd love to see something like this for Rangers, but I've yet to see it. 

 

 

[Blue Moon 1,289]

What xG doesn't take account of when it comes to an individual is the human factor, a stud caught in the surface, a slip, a tug of the jersey, a great tackle, an outstanding save, a lucky deflection, an accidental hand ball so someone could be judged by AI as failing to convert a sitter when it was outwith their control.

[Rousseau 10,611]

46 minutes ago, Blue Moon said:

What xG doesn't take account of when it comes to an individual is the human factor, a stud caught in the surface, a slip, a tug of the jersey, a great tackle, an outstanding save, a lucky deflection, an accidental hand ball so someone could be judged by AI as failing to convert a sitter when it was outwith their control.

It's just the quality of the chance. It's got nothing to do with the player or the defence.

 

Like I've said before, if you give have Haaland and Sebo taking the same chance, one will score, one won't (Haaland  ). 

 

The best players will score from low xG chances; that's why they are the best. 

 

The quality of the chance remains the same, though: on average and player in x position scores y% of the time.  

[Tannochsidebear 2,406]

7 hours ago, Rousseau said:

There are several data modelling companies, Opta being one. 

 

Football clubs will pay for the service - which includes far more data than I get from my wee preferred App. (The App I use gets its data from Opta, so they must be paying for that.)

 

It should? I don't think so. Even Sima's goal at the weekend was similar, I'd say, but easier to score than Lammers' chance. It has an xG of 0.41 (I think - I can't remember exactly off the top of my head). 41% of the time players score that. It's quite high. That's what the data tells us. 

 

Remember: it's based on tens of thousands of shots. 

 

You say, 'it should be scored every single goddam time', but how many games is that based on? Can you say that with absolute objectivity. 

 

I actually think it was a tricky chance because it comes at him quite fast, from behind. It's difficult to get your legs sorted for that type of pass. 

Thanks for that R. 

 

The bit I have put in bold, it is based on one game, last Wednesday at Ibrox, and it should have been scored every single time because its a sitter! I dont need Opta to tell me how good a chance it is, I could see it live from the stadium with my own eyes! I dont accept it was tricky (just my opinion), not at this level, he sees it coming and fluffs it past the post. It shouldnt happen but it did. I only used it as an example to ask the question on the stat. If Opta or whoever are saying that chance should only be scored 41 times out of 100, then they are clearly counting girls football, juniors, and the Albanian 9th division in amongst their 1000's of games that make up the data, not professional football at this standard. I know you love your stats and I'm not going to (or trying to ) convince you otherwise, but it makes a mockery of the whole xG business when they say that .99 (i'll concede it's not a 1.0) sitter is a .41  

 

Saying they have taken data from 1000's and 1000's of games is one thing, but there are levels in football, and as I have outlined above, what is a sitter at this level is perhaps not so many levels down, but if Opta are picking all of football to generate their stats then thats not accurately reflecting chances at the level which concerns the viewer.

[Rousseau 10,611]

23 minutes ago, Tannochsidebear said:

Thanks for that R. 

 

The bit I have put in bold, it is based on one game, last Wednesday at Ibrox, and it should have been scored every single time because its a sitter! I dont need Opta to tell me how good a chance it is, I could see it live from the stadium with my own eyes! I dont accept it was tricky (just my opinion), not at this level, he sees it coming and fluffs it past the post. It shouldnt happen but it did. I only used it as an example to ask the question on the stat. If Opta or whoever are saying that chance should only be scored 41 times out of 100, then they are clearly counting girls football, juniors, and the Albanian 9th division in amongst their 1000's of games that make up the data, not professional football at this standard. I know you love your stats and I'm not going to (or trying to ) convince you otherwise, but it makes a mockery of the whole xG business when they say that .99 (i'll concede it's not a 1.0) sitter is a .41  

 

Saying they have taken data from 1000's and 1000's of games is one thing, but there are levels in football, and as I have outlined above, what is a sitter at this level is perhaps not so many levels down, but if Opta are picking all of football to generate their stats then thats not accurately reflecting chances at the level which concerns the viewer.

I meant how many games have you watched to come to that judgement? I can guarantee the xG model has watched more games. Your judgement is subjective. The xG model objectively tells us how many times a player has scored a goal in that situation: 41%. 

 

We've both watched the same shot; you think it's a sitter, I think it's a little tricky. We're both using our own judgement. Are we both right? Both wrong?

 

I'm revising my judgement after seeing such a high xG. 

 

There are many variables that cause a player to not score: they're shit, the 'keeper pulls off a worldie, doesn't connect properly, slips, etc, etc. It happens. In a perfect situation he'll score, but football is an imperfect game, played by flawed humans. 

 

He definately should score. The xG says so: its a high xG. A penalty, for comparison, is 0.79. Players should be scoring penalties, but they don't. That's just a fact. 

 

Like I said, I've seen players sky a shot from one yard out. It's inconceivable that they could miss, yet they did. It happens. 

 

They won't be including Sunday league football. Having a range of levels is beneficial, though. You need a spectrum of data to find the most accurate average. 

 

I disagree that it won't 'accurately reflect [...] chances at the [top] level', but even if we accept that, you'd just factor in a little tolerance. Even then it won't fluctuate wildly from the numbers, IMO.