https://twitter.com/OptaSuit/status/971031834040635392
1 Like
Even I think thatâs cool. 
1 Like
http://www.football-observatory.com/IMG/sites/b5wp/2017/220/en/
CIES Football Observatory analysis of the best players by position in the last three months.
Interesting that Draxler shows up as a box-to-boxâŚhas he been playing that way for PSG lately?
Is that Oxlade Chamberlain that I see is it ?
1 Like
These nonces seem to have too much time on their hands after robbing us.
In all seriousness, Iâd like to see their methodology in the article, because thatâs the main way to see how good this model really is.
Arriving arbitrarily at these numbers is pointless.
let go of former captain Laurent Koscielny to Bordeaux for more than ÂŁ20m under his projected fee.
I canât remember the last time I read an article from the BBC and enjoyed it.
1 Like
Which mentally subnormal cunt projected that a crocked thirty four year old would go for ÂŁ25 million 
4 Likes
Probably on mountains of coke tbh
Since there has been increasing interest in advanced stats like xG and xGA, etc., I thought I would take a look at last yearâs team by team stats and overall metrics on these - in particular how teams âover performedâ or âunderperformedâ their xG and xGA.
So a few things I noticed⌠one is that teams higher up on the table tend to over perform these metrics - not a huge shocker, but interesting.
For example, if you take totals of xG for top 10 teams in league, that amounts to 16.73 positive xG. Put another way, these teams collectively scored nearly 17 more goals than their xG would indicate.
If you look at bottom 10 teams, it is even more dramatic - they underperform their xG by total of 30.12 goals. Again, this means they scored collectively just over 30 fewer goals than xG would predict.
So:
Top 10 Teams Overperformance: 16.73
Bottom 10 Teams Overperform: -30.12
Same pattern holds for xGA.
If you look at the teams that top the charts in terms of over performing xG and xGA, you see the following:
- Liverpool: combined 16.69 goals (9.54 xG and 7.15 xGA)
- Spurs: combined 15.40 goals (5.25 xG and 10.15 xGA)
- WHU: combined 14.69 goals (4.04 xG and 10.65 xGA)
- AFC: combined 14.50 goals (8.2 xG and 6.30 xGA)
- NU: combined 11.64 goals (2.09 xG and 9.55 xGA)
Teams that were disasters of underperformance arenât really shocking:
Huddersfield, Fulham, Soton, Bâmouth, Cardiff notable ones.
xG is not an advanced stat, itâs some dude/dudetteâs opinion on whether or not a player shouldâve scored lol
Curious what you consider an advanced stat then⌠not sure you have an understanding of xG if that is how you describe itâŚ
Stats in football are very difficult to apply b/c of the nature of the game, but they do have their place.
2 Likes
nah, itâd take too long. But it is not that it is just difficult to apply stats, it is that the thing theyâre attempting with xG is ridiculously difficult, and so it is off by ridiculous amounts all the time and I would argue it holds no value. And I know how the algorithm for xG is calculated in a few places, they explain it themselves if you want to look it up, and as you probably also know, it isnât even calculated the same. There isnât one xG. You can make up your own if you think optaâs regressive model is shite for example.
You should also be able to tell that if a game (recent example) ends 2-2 and xG says 3.3 - 0.2 youâre not off by a little. xG is for rough suggestions, itâs algorithm is coded by humans who guess what things are worth and put it in and they donât even agree with each other between companies. If you and I decide to separately count how many oranges are in a kitchen it doesnât really make any sense if half way through I feel like you start adding apples to your count and then olives and I end up with 7 and you end up 65 and that would probably still be less off a deviance than that xG stat above in the City - Spurs game where the real value was a thousand percent higher than what xG suggested (which I remember you know how to handle from some previous discussion). And there if we canât even agree on what weâre counting it isnât agreeably a stat (or atleast not a meaningful one).
So you can take it as a suggestion of how a game plays out, what it says about tactics vs. execution etc. Itâs all good and I am not arguing against it having its place if you think it does, I just feel the need to add the reminder that it is an algorithm based on whether or not people think you should score or not, it leaves out a billion things (as you said if you allow me to paraphrase âthe nature of the gameâ) so it is without any doubt parametrised wrong from the absolute beginning and that is often lost in the discussion.
What is a âbig chanceâ.
2 Likes
To the Man City- Spurs game, you will have large variances in something like xG because you donât have enough events, football is a goal-shy sport so outliers like that game will happen and will have an effect. Itâs pretty much agreed that xG shouldnât be used to analyze single games, but it does have predictive result during larger periods of time, which is why companies use them. For single games one should use pass-position charts, shot-plots and xG time plots.
And yes xG models will differ because theyâre models, and I understand that youâre more use more objective empirical models, in other fields subjective models are the norm and there a lot of things that must be done to validate the model. While xG algorithm varies from company to company, the value on the long term have converged, because if your model doesnât represent reality well itâs a shit model cough xPts cough.
4 Likes
Just to be clear, âit is an algorithm coded by humans who guess what things are worthâ is wrong. I donât necessarily disagree that the challenge they are confronting is daunting; but it has also been around long enough to develop some maturity.
Anyway, I have covered xG and other stats other times on these boards and while I agree they have limited value, they along with other stats can be useful in proper context.
As SpecialCunt says too - all of these stats are pretty meta⌠most relevant/interesting and âusefulâ football stats are relatively rare event stats, so you need a lot of games (or a season say) to get through the variance.
1 Like
no that is exactly how it works. Exactly like that. 20 degrees off the normal to the goal, 7 meters out, what is that worth to you. Stop saying no when that is exactly how it works. Do you know they value the playerâs chance based simply on what team he plays for quite often for example. They can base it on history going back further than that player has even played for that club lol
The human interpretation comes into what is considered analogous/similar, after that they are just following the results⌠xA btw is much more complicated and thus I would imagine suffer from another level of abstraction to validity.
Now you could argue that it is a foolâs errand to attempt to create samples that are similar by simply using positioning, and I might agree with that argument. How they do the sampling beyond that is evolving and heterogenous. Your example of using one team sample is one that I would agree with you isnât terribly good one.
But as far as stats go for football, they all suffer from some major downsides.
yes, agree because it is ridiculously hard to do to any significance lol
look at this for example:
https://understat.com/league/EPL
this weekendâs xG etc. according to some model. Itâs like they just threw some numbers in there. Wouldnât be surprised if you rand() from 0 and 3 to two figures for each team in each game and compared it to those numbers and the deviation isnât really significantly worse lol