Dont quite follow, praps you could expand?
xG And Analytics Under Amorim
- Thread starter bosnian_red
- Start date
-
Nik (@Rood) and Imran (@Annihilate Now!) discuss United's fantastic win away against Athletic Club in the semi-final of the Europa League.
They also pick out the positives of United's 4-3 loss away at Brentford, and Nik reviews a day out at Old Trafford watching Manchester United Women qualify for the Champions League.
It doesn't necessarily 'show' either, you have to understand the stat.
For example, very quick exercise using understat, look at where United vs Arsenal shoot from. If one team is purposefully having less shots from outside the area and more from within the six yard box, how much of that is finishing versus tactics? How much is simply bad decision making?
Also just using a bit of context on the second part, pre Saka injury Arsenal were top of the league for xG and xPoints, Jesus then got injured, Havertz followed, and Martinelli was out for a few games. Would we not expect their xG to then drop off (that is their entire starting front 3 out at the same time for 10+ games) considerably?
I'm beginning to think that xG overvalues/undervalues certain types of chances to an extent even though it comes as a reasonably decent model overall.
The better chances slightly undervalued, the lesser chances overvalued. Could well be player dependent though.
Edit: Would require a deep delve, but a player like Semenyo is well known for taking lots of low xG shots and he is underperforminh his xG, Eze very much in the same boat now. Wissa, Haaland, Isak, Brennan Johnson with some of the highest xG's per shot and have overperformed or come out around even. There are counter examples too that would go against that narrative, so like I say would take a deep dive.
Team-wise suppose you could at look at Forest and how they've conceded fewer goals than their xG against. They have the lowest xG per chance against them so could be a factor. When it comes to non-penalty xG per shot taken, the best 3 are Brentford, Liverpool and Newcastle, and they have all scored more than their xG.
The players on the end of the chances count too, but maybe the modelling is slightly off as well.
Last edited:
The last paragraph yes hence why they have not been as good as last season as an attacking side which is reflected by the xG.
I hope this makes sense.
Chris Wood is massively overperforming his xG this season, the quick response people will have is that he is a 'good' finisher but it's really not that simple.
xG on an individual level is quite good, in my opinion, with a bit of extra context alongside it. In Wood's case, you can make a strong argument to say that, compared to the historical data, Wood is elite at finishing a certain, quite specific, type of chance looking at his goal map (right footed or headed finish from left of centre, close to GK. I think he has one goal which bucks this trend and was a long shot vs Palace and terrible keeping).
xG on a team level however is not great if you try and use it as some have above. Forest get battered on xG in many games and, if you only looked at xG and nothing else, you would believe they had the most insane luck both in terms of somehow not conceding but also scoring, the truth is simply they plan to play in a certain way. I imagine as @Chipper says you can't really escape from the fact that every shot on a football pitch is completely player dependent (in fact every action must be) and then the frequency/type of shot will mostly be down to the tactical battle between both teams.
For example, Forest go a goal up, they might sit in even more and shut down the one or two players they deem the greatest threat. The opponent will keep creating half chances given they are sitting in and their xG will keep rising but the context of who is shooting is not part of the calculation. At the end of the game it will seem like Forest got super lucky, but actually they played for that exact scenario. Just thinking of this alone, you can see how warped xG at a team level becomes over 38 games with near countless variables in every game of football (cards, injuries, weather, pitch sizes, the list is near endless).
For me, it is becoming a great stat to use individually alongside others (for example who is actually the best all round penalty box finisher in the league, you could work out every season with a decent amount of accuracy) and a cool way to look at a broad overview once a season is done, but it's very hard to accurately compare two teams solely using xG.
As far as Forest are concerned, some of the free models, like Understat, actually indicate that they have been over-performing both their xG and xGA when their games are level. Additionally, they have been over-performing their xG most between 1–15 minutes, and their xGA between 16–45 minutes. Which, of course, could be a huge factor to them being able to play "their game". They have spent fewer minutes behind in games than City. Don't call it "luck" if you don't, but you wouldn't expect it to continue next season. Good for them, if it does.
You can never outperform xG with a bunch of poor finishers with no striker instinct.
They're certainly overachieving on the 2 expected points tables I'm aware of which has them in 13th and 14th. With negative "luck" (think I prefer variance) they might well be lower than that if the models are correct.
Said in another thread that any downturn is likely to be blamed on Europe next season, when they could play exactly the same way as they have this and suffer one anyway should the models be right.
Throw in those European demands and a key player in Chris Wood a year older and I might even be looking at the relegation odds for them over the summer. Do I think they'll go down? No, but if the odds are big enough any bet can be tempting.
I know I said that they might be doing something that the models don't quite pick up accurately in my last post, but that's not a definite.
Last edited:
If I understand your argument correctly, you're saying that good teams make chances that suit their players, which means they're more likely to finish them? And they defend by making their opponents attack in ways that don't suit them, eg by forcing teams to cross into the box when they have tiny strikers?
So immediately I'd point out that you're conflating two different things here - your xG (chances), and whether you over or under-perform it (finishing). Those are different metrics. There isn't really a strong correlation between the two metrics, so I don't think you can make conclusions about xG based on whether you're over or under-performing it.
If you rank the current table on xG, its very similar to the current table. Top 5 the same group, bottom 3 the same group, just a few notable changes, like Spurs & Forest almost swapping places. However if you rank the table by over/under-performance against xG, totally different. Forest, Wolves, Arsenal, Newcastle and Spurs are the top 5. Liverpool are 7th, City 8th and Chelsea 15th. If xG and over-performing it were synonymous you'd expect to see a clear correlation between the best teams and the greatest over performance. Liverpool have been a very balanced and consistent side this season, with a style of play that presumably matches their players, since they've done so well. Yet they're not near the top in terms of overperforming xG.
So while I understand your argument and can see merit in thinking it through, the examples you gave don't support the premise. Fundamentally, xG is about chance creation, not finishing, and stats show strong correlation between xG and good a team's attack is. So that's what you'd need to address.
Not really, I gave an example of why xG on a team level is somewhat useless on its own + gave an example of how it can be useful on an individual level.
The point was more that there's a lot more nuance than a team having better finishers or attackers because how much is tactical, how much is luck, how much is individual? You can't use xG in isolation as proof, you have to dig into it.
I assume you mean xPoints, not xG (as the table shifts more on xG)? As said, I think on a team level it is a good overview when you zoom out and use it as you just did i.e. over a long period. I'm not sure how you have come to the bolded? Forest were just an example as they have a distinct style.
Re your point about chance creation, that is a given where does it say otherwise? The point was simply the chance value will not take into account the person who is taking the shot when xG is calculated at the end of the game and, subsequently, added up in the xG league table so maybe you have misread or I have not explained it properly.
No, the example you gave was about over- or under-performing xG (or GF-xG if you prefer), not xG. They're different things. You talked about Chris Woods this season looking elite. Chris Woods has 11 xG this season, sandwiched between Raul Jimanez and Evanilson. There is nothing elite about his xG. What's impressive is his over-performance against xG, ie how many more goals he scores than you would expect based on his xG, which is the highest in the league.
But as I pointed out, extrapolating the effectiveness of xG from over-performance of xG doesn't work. There's little correlation between the two.
For me, it shows that a bit of good fortune in team sports can help you build momentum, and a bit of momentum can raise your confidence. It might continue next season, the underlying metrics aren't predictors, just "pictures" of certain time periods.
On a more general level, what they "show" is that "the table doesn't lie" is one of football's worse clichés. Of course, it does. Leicester weren't an 80-points side in 2015/16. NF aren't really top-5 quality this season. Conte's Chelsea finishing 90+ points was a massive over-performance. United's 81-points season was the same. Klopp's biggest achievement is that he actually turned Liverpool from a team that needed to massively over-perform to compete with City to a team "deserving" of their tallies. And now, Slot is enjoying the fruits of that.
Teams and reporters can pay and gain access to much more detailed stats than the ones available to us. We're trying to get the bigger picture from glimpses, and often enough we get things wrong. But, from my experience, what most people "rebel against" when it comes to the underlying stats is that said stats often challenge "long-held truths". See the discussions on this forum about the "transformative" striker or the "clown" Onana. xG won't tell you that new additions won't improve us. Titles have been won by great footballers playing out of their skins. It would tell you, though, that, in our current predicament, it will most likely be a short-lived joy. The most prolific goalscorers are also the ones who miss big chances more than anyone else. Ronaldo's career xG matches his actual goals. That's why a team's total xG, as a single number, matters. You want to score more goals? Find ways to increase your xG, don't look for "good finishers". Mourinho's second place was, more or less, the product of De Gea having his last truly world-class season at club level. But you can't expect to go far with that. When you concede about 1.50 xGA per game, you need more than a goalie doing crazy stuff to improve your defence.
Please reread what I said and rephrase the question as something is getting lost here. I wrote about xG and two aspects of it, the individual (you are now taking about) vs the team.
I read it very carefully, but Im evidently not getting something. What are the "two aspects" of xG?
Using it to assess an individual player vs a team.
I think this season ETH had already greatly improved our defense. Remember last season at one point we were averaging 20 shots against us. We ended the season at 17.6 shots against us which was 2nd most in the league to Sheffield United with 17.8. So we were right there as the worst in the league at stopping the opposition shooting.
This season iirc it was down a few, lets say 13-14 and today its 11.1 which is 5th in the league after Arsenal, Liverpool, City and Chelsea
ETH's problem was similar to Amorim's problem. Our attacking players couldnt create lots of clear chances and the clear chances we hdid have our players were finishing poorly. In between we had Ruud, and for his games we were finishing better. Then Amorim came and its been the same problem.
We have the joint 9th amount of shots in the league but 11th most shots on target. So we're poor at getting those shots on target let alone beating a goalkeeper
As in, using additional details on understat, you could use xG alongside them to really delve into the detail of who could be a good finisher. On its own though it wouldn’t necessarily mean one player was a better finisher than the other without the added context.
xG is not about finishing, nor does it even strongly correlate with finishing, so you can't use xG in a conversation about how good a finisher someone is. You certainly can talk about the role of a striker in creating chances, making good runs etc, but finishing is a separate conversation.
Well the basic idea of how many people look at it, which is why I wrote the original explanation, is that many do simply do that i.e. overperform = good finisher and underperform = bad finisher.
On an individual level I disagree though, as said, when it is used with other stats. I guess this becomes a conversation about what is 'good' finishing because, for me, xG is very revealing of decision making and that is a big part of what constitutes good finishing in my opinion.
Yes, I did notice that. But our inexcusable profligacy apart, I also didn't really see serious cohesive development from the collective in the first 2-3 months of the season. Maybe the players had already checked out.
We're more defensively solid now, and have thankfully sorted out our set piece defending somewhat (there was a short period where it was worse than anything I can remember by any team, ever), but once this attack is sorted out and we have more than one threatening player (sometimes none), and more pace and power, there'll be less pressure on the collective defensively because teams will be more cautious, whereas now it seems absolutely no one is afraid of us (and up until recently I'd say rightly so).
| # | Team | MP | xWins | xDraws | xLosses | xPts | Actual Pts | xP v Actual | |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Manchester City FC | 35 | 25 | 7 | 3 | 82 | 64 | -18 | |
| 3 | Arsenal FC | 35 | 22 | 10 | 3 | 76 | 67 | -9 | |
| 2 | Liverpool FC | 35 | 23 | 7 | 5 | 76 | 82 | +6 | |
| 4 | Chelsea FC | 35 | 20 | 10 | 5 | 70 | 63 | -7 | |
| 5 | Brighton & Hove Albion FC | 35 | 16 | 11 | 8 | 59 | 52 | -7 | |
| 6 | Fulham FC | 35 | 14 | 13 | 8 | 55 | 51 | -4 | |
| 7 | AFC Bournemouth | 35 | 14 | 12 | 9 | 54 | 53 | -1 | |
| 8 | Manchester United FC | 35 | 12 | 16 | 7 | 52 | 39 | -13 | |
| 9 | Newcastle United FC | 35 | 13 | 12 | 10 | 51 | 63 | +12 | |
| 10 | Tottenham Hotspur FC | 35 | 14 | 8 | 13 | 50 | 38 | -12 | |
| 11 | Crystal Palace FC | 35 | 10 | 12 | 13 | 42 | 46 | +4 | |
| 12 | Everton FC | 35 | 8 | 16 | 11 | 40 | 39 | -1 | |
| 13 | Aston Villa FC | 35 | 8 | 15 | 12 | 39 | 60 | +21 | |
| 14 | Nottingham Forest FC | 35 | 8 | 13 | 14 | 37 | 61 | +24 | |
| 15 | Wolverhampton Wanderers FC | 35 | 7 | 15 | 13 | 36 | 41 | +5 | |
| 16 | Brentford FC | 35 | 7 | 11 | 17 | 32 | 52 | +20 | |
| 17 | West Ham United FC | 35 | 5 | 12 | 18 | 27 | 37 | +10 | |
| 18 | Southampton FC | 35 | 4 | 9 | 22 | 21 | 11 | -10 | |
| 19 | Leicester City FC | 35 | 3 | 10 | 22 | 19 | 21 | +2 | |
| 20 | Ipswich Town FC | 35 | 3 | 9 | 23 | 18 | 22 | +4 |
Expected Points Table for England - Premier League 2024/25 season. xWins = Expected Wins, xDraws = Expected Draws, xLosses = Expected Losses. The expected results are based on the xG of each match. xG of each match is rounded to the nearest round number and treated as the result. For example, if Manchester City FC's xG was 0.95 and Liverpool FC's xG was 1.17, then the result will be rounded to 1-1, and therefore, treated as a draw. If the xG were 0.95 vs 1.55, then it would be rounded to 1-2, and therefore a win for Liverpool FC.
Based on the Expected Points, the 1st place on the table would be Manchester City FC with 82 expected points.
It’s gas that for my entire United supporting life finishing 8th would be an utter disaster. But now I look at that table and think “Ooh! 8th place. Not bad. Not bad at all!”
Where's that from, is that your own method? I have used the that same method myself before to be fair, as a very shorthand way at looking at things.
The 2 I usually look at are markedly different from that, but are similar to each other. The methods they use will be better, as for example a team with 0.79 xg that only has 1 shot all game which was a penalty can only score 1 goal in that match. A team that 2 shots for 0.79 xG which were a 0.50 xG chance plus a 0.29 xGchance can score 2. Think they run monte carlo simulations or similar to come up with the expected points taking into consideration the chance of each indvidual shot going in.
Last edited:
Where's that from, is that your own method? I have used the that same method myself before to be fair, as a very shorthand way at looking at things.
The 2 I usually look at are markedly different from that, but are similar to each other. The methods they use will be better, as for example a team with 0.79 xg that only has 1 shot all game which was a penalty can only core 1 goal in that match. A team that 2 shots for 0.79 xG which were a 0.50 xG chance plus a 0.29 xGchance can score 2. Think they run monte carlo simulations or similar to come up with the expected points taking into consideration the chance of each indvidual shot going in.
https://footystats.org/england/premier-league/xpts
Interesting. Thanks for sharing. That rounding feels odd to me though. Especially since the majority of chances are way smaller than 0.5 .
Cheers mate, appreciate it!
A lot of draws with that method isn't there? Suppose that's partly the nature of football and xG. It's a low scoring game without of ton of xG either. Close games on xG showing up as draws on that method when in reality one team bags the win with a 1 goal victory.
It's not completely without value using that as a shortcut even though there are better methods out there. It shares 7 of the top half teams compared to reality, the top 3 and bottom 3 are "correct" albeit in different orders. Could be a lot worse, and is much better than randomly shuffling the teams.
https://danny.page/expected_goals
That link is the other kind of method I was talking about. Quite fun to play around with.
Team A 0.79 xG (1 shot of 0.79 xG)
Team B 0.79 xG (2 shots of 0.50 & 0.29 xG )
28.11% chance Team A wins by 1 goal (can only be 1-0 in this situation)
46.78% chance it's a draw, could either 0-0 or 1-1.
21.72% chance Team B wins by 1 goal, could be 1-0 or 2-1.
3.39% chance Team B wins by 2 (could only be 2-0)
Expected points = 1.31 for Team A, 1.22 for Team B.
Last edited:
Feels about right. We aren't as terrible as people think (whereas last year I thought we were far worse even though the point tally was much better), but we are severely underperforming in points due to our horrid attack/finishing.
City's position is pretty surprising. I haven't watched them as much (abhore the way they often play and none of their players excite me now that De Bruyne declined), but I was under the impression that they were far worse than that table shows. I assume it's based on some poor finishing luck?
All through our 2nd / 3rd placed finishes and when we were overperforming our xG, no one really cared. I don't know what's better - the table is truth gang or the underlying stats are all that matter gang. I'm more in the latter camp.
The tables/underlying metrics aren't gospel and I don't believe in them strictly, but they absolutely give a good idea of how a team is performing that doesn't involve looking at surface level stats or results. In those 2nd/3rd place years we often struggled to really impress in many games that we would end up winning, but we'd have enough individual quality to bail us out and get 3 points more often than we do now. Players like Bruno/Pogba/Rashford/Martial/Cavani/Greenwood etc. could all come up with something special to decide the game, whereas now you watch us and it's really only Bruno and Amad. That's ultimately the decider in a lot of these metric discrepencies, which is why Liverpool are the only team outperforming because they got a special Salah season that won them points when an average player in his spot would have led to a draw or loss in some of those games.
Have we turned into Arsenal fans? feck me.
Amorims on the xG train now...
What does the xG table look like ? Perhaps we’re challenging for top half there ?
Poor finishing is frustrating. Suppose the counter argument is that it's just as much a part of the game as everything else. The counter-counter argument is that teams usually trend towards their xG longer term.
Just for the hell of it, using Fotmob's data for a few recent ones:
United 1.32-0.23 Wolves (xG)
United 0.09-0.72 Wolves (xGOT aka PSxG)
Bournemouth 0.53-2.53 United (xG)
Bournemouth 0.69-1.84 United (xGOT)
United 2.23-1.76 West Ham (xG)
United 1.20-2.98 West Ham (xGOT)
For those who don't know xGOT = expected goals on target, based on shot placement. It attempts to measure finishing. Good from a goalie's perspective too to see how difficult the saves he's making are.
xG is just how good the chance was. You get nothing on xGOT if a shot is off target, even if from a good chance. You also get nothing from it on a blocked attempt, and very little if your shot results in an easy save.
Last edited:
Nah we still suck in it, just that we have had some games recently that we just haven't finished chances. 12th in the xGD/90 table on fbref, 13th in the xPts table on understat, both of them virtually identical with Spurs too. 11th in the last 10 league games in xPts, 13th in last 5, 13th since Amorim took over.
Its shit whatever way you look at it, yet we're still underperforming our metrics. Last season we over performed them by a ton and finished 8th, this season underperformed them by a lot and will be 16/17th, while the metrics ranked us 15th last year and 13th this year. And really this is the value of underlying metrics. Our general play for 2 full seasons now has been that of a bottom half side. Last season we may have finished 8th, but it was clearly a fluke, and while we aren't quite 17th level bad, it's still a bad situation. The metrics have stayed consistent for a while now despite results fluctuation.
Yeah honestly that question was a bit tongue in cheek but those stats are interesting. The problem with xG and analytics is that it never provides the whole story. Context is necessary. The value here as you say is that there is an obvious trend the past couple seasons which is pretty much right in line with where we are.
I think it’s a stretch to say we have underperformed this season though. Or perhaps the better way to say it is I expect us to continue to underperform from an xG perspective so long as Amorim persists with trying to play out from the back. Weird to say we are underperforming when we’re gifting the opposition clear cut chances every match.
I was thinking that the performances were better in the last few games without being rewarded mainly due to the crossing positions and the quality of the chances being created but I wanted to be sure that it wasn't just wishful thinking. If xG is picking it up then it sounds encouraging.
Alright he's dead to me now. What's Steve Bruce up to these days? let's hire him.
I flip flop on xG regularly but this sort of state makes a mockery of it if people actually watch the games.
Man. City were dismal over the winter period. Can only guess xG is making a case they should've won at Bournemouth and Brighton when 1-0 up in those but I really fail to see how they should be on nearly 20 more points than they currently are. Even this year they comfortably lost at home to Liverpool and away to Arsenal and won most of the games you'd have expected them to win over last three months.
Also us (Villa) on 13th place?! Been a running joke on our forums this season how often we've conceded from opposition's first shot on target and all those weren't just tap ins. Also had poor winter run when we were constantly losing away and drawing games at home we should've won so again failing to see how we've effectively fluked 21 more points than we've deserved.
I'm sure it is a stat useful for recruitment officials to decide which players are more productive in final third when scouting signings but to actually define seasons it has its flaws and think that table shows it.
I flip flop on xG regularly but this sort of state makes a mockery of it if people actually watch the games.
Man. City were dismal over the winter period. Can only guess xG is making a case they should've won at Bournemouth and Brighton when 1-0 up in those but I really fail to see how they should be on nearly 20 more points than they currently are. Even this year they comfortably lost at home to Liverpool and away to Arsenal and won most of the games you'd have expected them to win over last three months.
Also us (Villa) on 13th place?! Been a running joke on our forums this season how often we've conceded from opposition's first shot on target and all those weren't just tap ins. Also had poor winter run when we were constantly losing away and drawing games at home we should've won so again failing to see how we've effectively fluked 21 more points than we've deserved.
I'm sure it is a stat useful for recruitment officials to decide which players are more productive in final third when scouting signings but to actually define seasons it has its flaws and think that table shows it.
The particular methology they (footystats) used to compile that table and dish out points is very different and quite frankly the wrong way to go about it all compared to Opta, Understat, Statsbomb etc. Just look at all the draws!
It's a method I've used myself for shorthand looks at xG, it is better than randomly assigning results to teams too, but compared to how you really should do it it's not that good. Some, but very limited usage.
Here's Opta's current take and how it really should be done:
