It's still idiotic.
United and xG (now that Ole is gone will things change?)
- Thread starter Pogue Mahone
- Start date
It never was. It is just people being cavemen. It is by far the best indicator of long term trends.
Yep. It's a perfectly reasonable thing to look at. The trouble is you try and point anything negative out after a win and you get hammered even if it's perfectly valid to have concerns. People were trying to pretend it was all great when we finished 2nd under Jose but we were shit most weeks and massively flattered by it as xg suggested. Arsenal were massively outperforming their xg earlier in the season during that flukey unbeaten run and shock horror the results soon tanked. It's a concern.
Cavemen? The best indicator is to actually watch the games without worrying about some made up stuff. I'm not saying you're not watching, just that xG is a thing made up from people who dont watch games.
Nope, 'feelings' are nowhere as accurate as stats.
Good teams overperform their xG cause they have good players, bad teams underperform it cause they have bad players, best teams like Barca and City were overperforming it by 10% or so, while United was overperforming it by 30%+, which was a very strong indicator that we were quite lucky and soon a return to ("in direction towards") the mean was going to happen.
Made up stats.
Overperforming a little is expected when you have good players, but it isn’t sustainable to be over performing by a lot.
All statistical models are made up. If you go to get a loan, it is a model that will likely decide if you'll get it or not. Soon enough, a model will drive you.
Just cause they are 'made up' it doesn't mean that they aren't useful and that they don't have good prediction ability.
Does anyone have the xG stats for Ole's time at the club and Mourinho's 2nd season?
How do they compare?
How do they compare?
At best, you can draw the conclusion that it means a team are taking their chances. You can't possibly look at it and conclude that "xG says they're being lucky". Having a look at a few previous seasons, I can see that, for example, Chelsea outperfomed it by 37% when they won the league in 16/17 - I suppose they were just lucky all season long?
Even disregarding that, setting a start date to the table of when Ole joined (19/12/18) shows that for those 13 league games, United are overperforming xG by 0.99. In the same time period, Liverpool and Spurs are overperforming it by 3.12 and 5.53 respectively. So then, by your logic, xG says they're actually the lucky feckers and when things normalise, we should be above them... Which ever so slightly goes against your original point.
2017/2018 :
City...........:xG = 91.43 , G = 106, xGA = 24.51, GA = 27
United.......:xG = 59.04 , G = 68 , xGa = 43.53, GA = 28
Liverpool...: xG = 77.49 , G = 84 , xGa = 35.75, GA = 38
2018/2019:
Mourinho United: xG = 28.22, G = 29, xGA = 26.53, GA = 29 // 17 games
Ole United.........:xG = 29.11, G = 29, xGA = 15.22, GA = 11 // 13 games
Total United.......:xG = 56.23, G = 58, xGA = 41.75, GA = 40 // 30 games
City..................:xG = 76.75, G = 79, xGA = 21.68, GA = 21 // 30 games
liverpool...........:xG = 63.12, G = 68, xGA = 22.13, GA = 17 // 30 games
data source: understat
I think this is true but it's more that we need a RW and an RB. Lingard-Young or Lingard-Dalot or whatever else we throw out there looks more like a relegation threatened team than a top 6 one.
Looking at Jose's 2nd season, one could make the following observations:
xG does reflect what we witnessed on the pitch to a fair degree: DDG saved our arses frequently and we were generally pretty shoddy attacking wise. However, the model completely fails to predict what is actually a common scenario in football, i.e. that a team maintain their "unsustainable" form over a full season, finishing higher than they "deserve" based on the sheer numbers. This isn't an anomaly - it happens all the time.
According to the numbers in mid-December (halfway through the season), we should've finished 6th. But that's nowhere near as bad as the model's predictions with regard to Burnley: they should've escaped relegation by the narrowest of margins - but finished 7th. The model also predicts Bournemouth to finish dead last - in reality they ended up very respectably in 12th.
On the plus side, the model does predict the winner and 3/4 of the top 4. It also predicts, specifically, that Tottenham are significantly better than 7th, their actual position in mid-December. And while it misses completely with regard to Bournemouth, it correctly predicts 2/3 of the teams that ended up going down.
xG does reflect what we witnessed on the pitch to a fair degree: DDG saved our arses frequently and we were generally pretty shoddy attacking wise. However, the model completely fails to predict what is actually a common scenario in football, i.e. that a team maintain their "unsustainable" form over a full season, finishing higher than they "deserve" based on the sheer numbers. This isn't an anomaly - it happens all the time.
According to the numbers in mid-December (halfway through the season), we should've finished 6th. But that's nowhere near as bad as the model's predictions with regard to Burnley: they should've escaped relegation by the narrowest of margins - but finished 7th. The model also predicts Bournemouth to finish dead last - in reality they ended up very respectably in 12th.
On the plus side, the model does predict the winner and 3/4 of the top 4. It also predicts, specifically, that Tottenham are significantly better than 7th, their actual position in mid-December. And while it misses completely with regard to Bournemouth, it correctly predicts 2/3 of the teams that ended up going down.
Nope. Still idiotic to me tbh. Football has more factors affecting it to be counted by a stats like xG and luck is an integral part of the sport whether we like it or not.
But then you're admitting it could be down to luck rather than what a lot of people seem think (united magically performing 30% better in front of goal than your average player/team). I think if you speak to any stats people they will be the first to admit that luck plays a MASSIVE part in football. If anything they will admit it more than your average fan who puts everything down to passion or feelings.
Why making it one way or another? Why should it be tactics 100% or luck 100%?
Luck is an integral part of the sport. You can't win by just luck but you can put the best tactics and don't win because you were unlucky. You need to do your best and hope you get some luck on your side, and which is what happened to us. I don't find any this discrediting Ole or the team with this.
All I know is football is a sport with so many immeasurable factors that affect the result, unlike a game like basketball in which stats is an integral part of analyisng everything imo.
Luck is a part of sport, but that's why it's called expected goals. For a given series of inputs(positions, plays, area from which shot was taken, defenders around, ability of goalkeeper), on average, how often does it lead to a output(goal)? This factors in the luck in instances where both a goal is scored and it is not. Reduce it to a simple penalty kick - if Ronaldo was taking a penalty against Romero, you'd expect him to 95 times out of 100(I made this stat up). If Ronaldo suddenly starts missing penalties against Romero every single time 10 times in a row, it means something is off. xG simply does the same thing, but expands it to multiple factors and averaged out over hundreds of thousands of footballers resulting in an average prediction. Luck gets implicitly factored.
It's not right to refute it as being idiotic. The idea is sound - probabilistic predictions power a lot of the technology you use.
There's a possibility that the actual model built for determining xG sucks, but I have no way of knowing that. If it's consistently a leading indicator of upcoming form, it probably means the model is pretty damn good.
Last edited:
xG doesn't make predictions, all it does is add up events in the past.
Last edited:
It doesn't tell you much about future form though. If you just go with xG and ignore goals scored, wins, domination, fixtures etc then you might not know what to expect.
Good teams will keep creating chances as we all know. Weaker teams will still struggle against the big teams. Having very high xG against Huddersfield and Fulham doesn't indicate you will stand a chance against Man City even if you had higher xG than Man City recently.
It shows the quality of chances a team has had. Nothing more and nothing less. Over time things will even out and teams having many chances will score more goals. Good teams might need less chances than the weaker sides. Confidence and form will help teams go over xG and convert those chances more, but also help them create more as well.
Yesterday against Wolves we didn't create much at all so should have low xG. A late goal from Rashford might increase it and make people possible think it was more even than it was. Wolves totally deserved that win though.
Yeah and people use it to suit agendas.
Which people use to make predictions about the future.
If used within context it can be a great stat no doubt. Just like shoot stats in general. Although I would not put too much weight on it.
If I go to a league and only look at xG without ever having watching the games it might not tell me much about how things should go.
Getafe is currently in 4th place in la liga. From xG they should have scored 31 and conceded 30. Based on that people would probably expect them to decline as people would not have expected them to do so well from the start. Although apart from the big 3 teams the rest don't really do that well this season either. I don't follow La liga much outside the big teams so would not know if they have good momentum/confidence (playing well) or have had luck with them.
Luck is kind of integrated in xG. If you take a shoot from a position, xG is just the ratio between goals and all shots from that position (I simplified it a lot). Luck played a part in all those shots taken.
Yes, fair enough - these models weren't made as prediction tools. But as others have remarked, they're clearly being used for that purpose by both journalists (increasingly so) and fans.
As such, it seems reasonable to look at how well they work in that regard.
So today was another game where we massively reversed the xG result, right?
Long may it last...
Long may it last...
No, it was 1.98 - 1.35 in United’s favour. I felt that watching the match. Watford didn’t really create many good chances. Martial scored from about 2 yards out too.
Huh. Well there you go. The fallibility of xG in a nutshell. Watford were by far the better team today.
But they only played it good until our box. They didn't finish their attacks, only one, and that was a brilliant goal. While we had 3 very clear chances (Rashford goal, Martial goal, Martial after the corner).
It's all based on position in the box for shots vs goals.
Obviously strikers in the top teams are better and able to score from positions that other strikers are not able to on average.
xG appears to be an excellent way to rate strikers and keepers but takes no account of individual qualities of players or the characteristics of particular teams. For example, with DDG in goal I would expect us to concede fewer than the xG says we should as he is a far more capable shot stopper than the average.
I would also expect a superb striker of the ball like a Batistuta or a Ronaldo to score far more than the xG says they should. It may be teams like City whose attacking play is based on playing through/around teams to get into scoring positions as often as possible do score similar to what their xG says they should simply because the greater the sample size the greater the chance the at the observations are closer to the expectations.
I would also expect a superb striker of the ball like a Batistuta or a Ronaldo to score far more than the xG says they should. It may be teams like City whose attacking play is based on playing through/around teams to get into scoring positions as often as possible do score similar to what their xG says they should simply because the greater the sample size the greater the chance the at the observations are closer to the expectations.
Yeah. I don’t think we played well and didn’t enjoy the game much, but we did have the majority of the clear chances on reflection after rewatching the highlights
For what it's worth, this model (which is more sophisticated than Understat's) has Watford ahead. 1.52 vs 1.39. Just keep in mind that while expected goals have a higher correlation to performance than actual goals (because goals are such an infrequent event) a 90-minute sample is often not going to tell you much. Not having seen yesterday's match I wouldn't want to draw any conclusions from the numbers below. The larger your sample gets, the higher their significance and predictive power become, though.
No xG model should have Watford ahead of United. They had pot shots and half chances at best apart from Doucoure. We had Martial with an open net, Martials miss off the corner which was bigger then even Doucoures goal, and then Rashford one on one. That alone is all we need to have a higher xG on every single model. Based on quality of chances, we had better chances by a distance.For what it's worth, this model (which is more sophisticated than Understat's) has Watford ahead. 1.52 vs 1.39. Just keep in mind that while expected goals have a higher correlation to performance than actual goals (because goals are such an infrequent event) a 90-minute sample is often not going to tell you much. The larger your sample gets, the higher its significance and predictive power becomes, though.
XG shows simply quality of chances. Watford had loads of shots - but apart from Doucoure what did they have? Pot shots and half chances at best. We had 3 sitters, which gives us the "xG win" easily.
Anyway, we haven't had that many games under Ole where we overperformed xG.
Actually looking at his whole tenure in the league, our stats read:
Goals - 31
XG - 29.99
Goals against - 12
XG against - 16.57
Which is basically par for what youd expect with De Gea in goal and our goals scored basically spot on.
https://understat.com/team/Manchester_United/2018
If you want to look through our games yourself to see. One that sticks out is Fulham being 2-2 on xG when game state entirely dictated how that game went.
Where's that from?For what it's worth, this model (which is more sophisticated than Understat's) has Watford ahead. 1.52 vs 1.39. Just keep in mind that while expected goals have a higher correlation to performance than actual goals (because goals are such an infrequent event) a 90-minute sample is often not going to tell you much. Not having seen yesterday's match I wouldn't want to draw any conclusions from the numbers below. The larger your sample gets, the higher their significance and predictive power become, though.
Lots of low value pot shots will add up though?
0.05-0.15 only I believe. Hardly significant
Wyscout
Thanks