For what? Something more sophisticated might be.
United and xG (now that Ole is gone will things change?)
- Thread starter Pogue Mahone
- Start date
How useful it is has to be measured by how accurately it predicts changes (individual players and teams reverting to type, as it were).
Or, more precisely, its value depends on how well it predicts changes that aren't bleedin' obvious to begin with.
In many of the cited examples, where x/G indicated a significant change (positive or negative), it seems that common sense would've sufficed (to reach the same conclusion).
Or, more precisely, its value depends on how well it predicts changes that aren't bleedin' obvious to begin with.
In many of the cited examples, where x/G indicated a significant change (positive or negative), it seems that common sense would've sufficed (to reach the same conclusion).
That isn't a statistic. It is data.
I think that’s true. I guess stats are a way to translate “common sense” to people who lack it. Which is all of us at times, especially with issues we feel passionate about.
In the same way that stats can challenge crazily biased opinions on individual players, they can also be used to correct biased opinions about teams. And even the most “common sense” fan is prone to bias.
Seeing long term trends and scouting players (especially strikers and keepers).
Doesn't seem like very convincing data to me.
As I said, it is the beginning. To be really super-useful it needs to be improved, but even now, I would be surprised if there aren't many managers who at least keep an open mind on them.
When it comes to buying strikers, the xg stat is especially confusing. Do you buy a striker who's over his xg numbers but there's a risk that he might revert down to his xg in the next season or do you buy a striker who's below his xg number but there's a chance that he might not ever reach his xg.
I'd buy the first one, personally.
If a player has a high xG, it means that he has good movements and so puts himself at the end of good chances (unless xG is an accumulation of taking shots from 30 yards). Combined with the number of goals and likely other stats it could be a good indicator of what is to come. Probably better than just looking at number of goals which might be biased from a player being in a purple patch. The diagrams in the previous page illustrated the same point quite well.
I think the influence of psychological factors like confidence or a high team spirit are well beyond the limits of current models. Computational psychology is well on the rise but still far from being able to significantly improve such models.
I might be mistaken but it seems to me that a high team spirit might significantly contribute to, say, a better chance-conversion rate. Such things make the models predictively weak about short-term outcomes or even a whole season. A given manager might be a genius regarding tactics and coaching but if he cannot create a team spirit and motivate the players, he will probably fail. Conversely, the manager might not be particularly sophisticated tactically wise but might be able to bring the best out of the players by helping them play with confidence and freedom. This is why Di Matteo could win the CL with Chelsea whereas Mourinho failed despite being much better. Psychological factors are utterly important, especially short-term.
I might be mistaken but it seems to me that a high team spirit might significantly contribute to, say, a better chance-conversion rate. Such things make the models predictively weak about short-term outcomes or even a whole season. A given manager might be a genius regarding tactics and coaching but if he cannot create a team spirit and motivate the players, he will probably fail. Conversely, the manager might not be particularly sophisticated tactically wise but might be able to bring the best out of the players by helping them play with confidence and freedom. This is why Di Matteo could win the CL with Chelsea whereas Mourinho failed despite being much better. Psychological factors are utterly important, especially short-term.
High xG for a striker probably has more to do with the quality of the creative players on his team than the quality of the striker himself. Especially if he is bad at converting those chances.
Good teams score more and concede less than the expected average.
Well feck... This changes everything.
Well feck... This changes everything.
I don't really understand xG and haven't made an effort to other than the basic inference that it tells you how a team's were better/worse than peformances.
I'd say our results under Ole have been better than the peformances. We started off playing slick stylish football and gradually have become more plucky and jammy with time. Anybody who is honest will admit this. There may be mitigating factors but it is what it is. Those factors being where we were under Jose, the broken mood of the team, the teams we've had to face and a huge amount of injuries.
This is coming from someone that believes he should be our next manager and has done an incredible job thus far.
I'd say our results under Ole have been better than the peformances. We started off playing slick stylish football and gradually have become more plucky and jammy with time. Anybody who is honest will admit this. There may be mitigating factors but it is what it is. Those factors being where we were under Jose, the broken mood of the team, the teams we've had to face and a huge amount of injuries.
This is coming from someone that believes he should be our next manager and has done an incredible job thus far.
As you say, xG is used a lot in the fantasy football community. However, I think it’s a load of BS and I’m arguably the best FF player in the world.
xG is already a better indicator for future performances than actual goals which are better than actual points. That alone should make dismissing it completely naive at best.
Think it was 11tegen11 that showed this. I'll try and find the link later.
Found it: http://11tegen11.net/2015/01/05/the-best-predictor-for-future-performance-is-expected-goals/
That is four years old. The models are only going to get better.
Think it was 11tegen11 that showed this. I'll try and find the link later.
Found it: http://11tegen11.net/2015/01/05/the-best-predictor-for-future-performance-is-expected-goals/
That is four years old. The models are only going to get better.
Last edited:
I think we are probably on the same page then mate.
I’m not saying the concept is poor, merely the crude excecution of it. It’s miles and miles away from being a statistic you can induce anything from.
Just look at any weekend of xG Scoreline vs Real Scorelines and it’s laughably wrong almost every time.
Now apparently we are supposed to be informed by that which team under or over performed in front of goal in a given match.
The very fact it gets it so wrong in almost every instance shows that the model is fundamentally flawed. If it was a trust worthy stat, the vast majority of matches would fall in line with xG and there would be a few outliers where over or under performance occur.
Good post, thanks
Surely this shows the flaw in xG. We were shite all season under Jose so it's no wonder we're scoring goals and conceding less.
Our xG next year will be more accurate.
Our xG next year will be more accurate.
yeah you really need to watch them play. Someone like Cavani got very good movement, but also very good players in PSG. Would you go for him over say a Harry kane who might be more efficient, but not get as many chances?
I think buying a player based on xG stats could be quite stupid.
For a goalkeeper I guess it could help more. If someone is saving a lot of things they should not save then it could indicate a good goalkeeper. Although it is also how you save it. Making a big save only for the other team to score from the rebound might give xG over 1, but you still conceded.
Example is Morata. Morata is a top striker according to xG. In reality, he bottles it whenever he gets a chance and fecks up and this has been an ongoing thing throughout his career.
Morata has excellent movement but poor finishing ability. A great striker has both. Both needs to be considered, the data might just help a scout doing so, to confirm his opinion on the player. If the data and opinion don't match, then if I was a scout, I would reconsider and re-analyze.
Tweet
— Twitter API (@user) date
Imagine when it gets even better and everything in the sport pretty much feels predetermined by data scientists as opposed to managers and players. The only thing at that point that can throw a wrench at the predetermined results will be significant injuries (just like in the NBA now). That will make for fun viewing I suppose...
The variance is too high in football. One goal either way got a huge impact for the result. Big gaps can make league tables boring though.
It doesn't pre-determine anything. It just tells you the likelihood of certain outcomes. But we know that already, just look at the betting odds in efficient markets with high liquidity. I can tell you right now that Arsenal will win 40% of time on Sunday and United 32% of the time. That doesn't take the joy out of watching, does it? Everyone has a rough idea of these numbers in their head anyway, the model is just more precise.
I don't even know what to say to that. "Data good. Statistics bad"? My God.
I'm groping here but do you mean that raw data is good and manly, but if you add it up or make a ratio out of it, it's stupid and evil? So, in your mind, if they gather heart rate data it's ok, but if they analyze it (for example by combining it with distance covered data) then it's dumb and irrelevant?
I mean, WTF?
We have every team in the league paying big money for game data and analytics, and people on here saying that "I bet Ole doesn't" use it, and "data is ok but statistics is a step too far". It's fkn hysterical.
That is true. I encourage everyone to read up on Brentford owner Matthew Benham who has made an absolutely fortune predicting the outcomes of football matches using data such as xG. I posted some links in another thread recently.
Data scientists took over baseball and young people stopped watching.
Data scientists took over basketball and now young people only care about trades and free agency.
Data scientists take over football and young people...
Data scientists took over basketball and now young people only care about trades and free agency.
Data scientists take over football and young people...
The Stattos are revolting
You're looking at a simple difference between two numbers - it cannot be "wrong". You can argue about the interpretation or about how xG is calculated if you like, but not about the result of a simple arithmetic operation. And even the calculation of xG is not very controversial - it's just the summation of a bunch of expectations, which are each themselves the percentage of shots from a particular position that have been scored historically in the matches for which detailed observations are available. That's it, it's just math. So that leaves interpretation - and that's totally in the eye of the beholder.
I always find it bizarre that there isn't more context added to XG.
There should be a breakdown of XG for each situation - The scoreline mainly. People acted like Spurs away this season and Arsenal away last season should have been hammering defeats when in reality we were never even close to losing either game. There's no way we allow the same pressure if we're not holding a lead. I'd imagine in both game we actually had the better of it when it was level. The accumulative total doesn't account for the circumstances.
Also when used accumulatively it should tally points for each game as well as the overall XG. I think some sites may have started doing this. So if you win five games by .5xg you'll still have a better points total than a side that wins four by 5xg and loses one by 1xg.
There should be a breakdown of XG for each situation - The scoreline mainly. People acted like Spurs away this season and Arsenal away last season should have been hammering defeats when in reality we were never even close to losing either game. There's no way we allow the same pressure if we're not holding a lead. I'd imagine in both game we actually had the better of it when it was level. The accumulative total doesn't account for the circumstances.
Also when used accumulatively it should tally points for each game as well as the overall XG. I think some sites may have started doing this. So if you win five games by .5xg you'll still have a better points total than a side that wins four by 5xg and loses one by 1xg.
I have no idea what xG is. 
You’re missing my point.
If the xG model was accurate the majority of actual results would correlate with results predicted by xG and in that scenario you could have confidence in making judgement on the outliers.
As it is, xG has no actual correlation with actual goals so what exactly is it claiming to model? Someone’s definition of when someone should score rather than a statistically accurate prediction of when someone should score. That’s what makes it nonsense.
I posted this link above: http://11tegen11.net/2015/01/05/the-best-predictor-for-future-performance-is-expected-goals/
xG in its current state (which is far from perfect) is already a better indicator for future performance than goals. If anything actual goals are not a very useful metric for evaluating the performance of two teams (if this is what we are talking about here) in a single match due to the low-scoring nature of football.
This is a to each their own debate if there ever was one, but psychologically I am intrigued to know:
What is the primary reason for following or arming yourself with this type of statistic ahead of a match you are about to watch? Is it to make smarter gambling/fantasy football decisions? Is it to pull back emotionally - stop yourself getting too high or too low early in a match? Or is it just that you (this is a general "you" and not you specifically) prefer discussing metrics about an event put on for entertainment?
Again, to each their own, but I am genuinely curious what viewers (and not the data scientists themselves, as this movement is no doubt a jobs creator in that field) are getting out of this. My personal opinion, flawed as it may be, is that the more metrics play a role in sport, the less entertaining they become.
Somebody expects something to happen and when it doesn't they type it down to show their friends. It's all very technical, like playing Monopoly with the board facing down.
I've been reading about it but am still a bit confused as to how you can reliably predict what constitutes a good chance without taking into account player attributes. Left foot? Right foot? Finishing style? Power behind shot? Goalkeeping style? If a player is an FK specialist from a certain area, is that a chance? Or are they considered to be outperforming xG, even if they score 6 in a row? If you're outperforming the xG then is it useful in identifying a higher quality squad than one of a lesser quality?I have no idea what xG is.
What's amazing is that you have the confidence to call it nonsense when you clearly have no idea what it is or what it is trying to do.
There is total correlation between actual goals and xG across the games included in the calculation. That's how xG is calculated: from all the goals scored and from where.
And "a statistically accurate prediction" is exactly what xG isn't. It's an simple mathematical representation of goals scored in the past.
You made two statements and both were literally as wrong as they could be. Maybe you should learn a bit before you tell other posters what is nonsense or not?
Is there an explanation as to what this XG stuff is or what it means anywhere in this thread? I've no idea what you're all going on about.
I think any sane fan knows in the back of the mind that this is not sustainable and we will need to improve as a team to get similar results in future. Having said that people are excited that with time Ole can improve the team and thus keep the level of results up