Aston Villa finished fourth in the Premier League with 65 points and a positive goal difference of seven.
Their expected-goal numbers told a very different story.
Villa scored 56 goals from 48.46 xG and conceded 49 from 55.75 xGA. Put those numbers together and their expected goal difference was -7.29. Their actual goal difference was therefore 14.29 goals better than their underlying chance numbers suggested. StatMuse Premier League
That sounds like the most extreme xG overperformance in the league.
It was not.
Manchester City finished 15.58 goals better than their expected goal difference, narrowly exceeding Villa’s gap. StatMuse Premier League
But that does not make City the more interesting case.
Their underlying numbers already described an elite team. Villa’s did not.
And once we look across the rest of the Premier League, another problem emerges. The phrase “xG overperformance” makes several completely different seasons sound like the same thing.
City exceeded their expected numbers at both ends.
Villa did too, but with a particularly unusual attacking story.
Everton’s positive gap was almost entirely defensive.
Crystal Palace’s enormous negative gap came overwhelmingly from scoring fewer goals than their chances suggested.
Wolves went backwards at both ends.
Tottenham beat their attacking xG, only for the opposite to happen defensively.
The useful question, then, is not simply which Premier League team most exceeded its xG.
It is how they did it.
There isn’t one xG table
Expected goals are most useful when they are treated as a description of chances rather than an alternative league table.
For every team, we can start with two basic numbers.
xG estimates the quality of the shots a team created. xGA applies the same idea to the chances it conceded.
Subtract xGA from xG and we get expected goal difference: whether a team’s overall chance balance was positive or negative.
We can then compare that with what actually happened on the scoreboard.
Manchester City, for example, finished with a +42 actual goal difference. Their xG and xGA produced a +26.42 expected goal difference.
That leaves a positive gap of 15.58 goals.
Villa finished at +7, despite an expected goal difference of -7.29.
Their gap was 14.29. StatMuse Premier League
Figure 1 — Which teams most exceeded their underlying goal difference?
Behind them came Sunderland at +8.69, Arsenal at +7.85 and Everton at +6.46.
At the other end, Wolves finished 17.52 goals worse than their expected goal difference. Crystal Palace were 12.68 worse, Newcastle 8.84 worse and Chelsea 8.25 worse. StatMuse Premier League
That gives us a league-wide measure of the gap between chances and outcomes.
It does not tell us why those gaps appeared.
For that, we need to split the numbers apart.
City actually had the biggest raw discrepancy
Manchester City are the useful place to begin because they show why simply ranking xG discrepancies can be misleading.
City scored 77 goals from 70.96 xG, putting their attacking output 6.04 goals above the StatMuse model. They conceded 35 goals from 44.54 xGA, another positive difference of 9.54 goals at the defensive end.
Together, those two gaps explain why their +42 actual goal difference finished 15.58 goals ahead of their +26.42 xGD. StatMuse
But there is an important distinction.
City did not need those favourable differences to transform a poor underlying profile into a good one.
A +26.42 xGD already says that City created substantially better chances than they conceded over the season. Their actual results widened an advantage that was already there. StatMuse
There is also some evidence of goalkeeper contribution within the defensive gap. Opta’s completed-season goalkeeper rankings placed Gianluigi Donnarumma second in the Premier League with 5.7 goals prevented according to its own post-shot model. Premier League
That does not mean Donnarumma personally explains City conceding 9.54 fewer goals than their StatMuse xGA.
Those are different models measuring different things, and the gap between xGA and actual goals conceded is not a goalkeeper statistic. FIVDA’s goalkeeper analysis explains why post-shot measures such as PSxG or xGOT evaluate a different stage of the chance from traditional pre-shot xG.
It does, however, reinforce the broader point: the difference between chances conceded and goals conceded can contain information that a basic xGA total does not capture.
City had the league’s biggest overall discrepancy.
Villa had the more consequential one.
Villa turned a negative chance balance into fourth place
Villa’s underlying numbers were unusual precisely because they did not describe a conventional top-four season.
They scored 56 goals from 48.46 xG, an attacking difference of +7.54.
At the other end, they conceded 49 from 55.75 xGA, another positive difference of 6.75.
Combine the two and Villa transformed a -7.29 expected goal difference into a +7 actual goal difference. StatMuse
That distinction matters because Villa’s discrepancy was not merely making a strong underlying team look even stronger.
It changed the entire picture of their season.
They finished fourth with 65 points, while Opta’s separate Expected Points model placed them 12th with 48.7 xPts. Premier League
Those two expected measures should not be confused.
Villa’s -7.29 xGD here comes from StatMuse’s season totals.
Opta’s 48.7 expected points come from a different model that simulates individual matches using the chances created and conceded in each fixture. Opta Analyst
Neither number says Villa “should” have finished somewhere else.
They tell us something narrower and more interesting: Villa’s actual results ran substantially ahead of what two different chance-based approaches would lead us to expect.
And unlike a vague explanation about luck, there is at least one identifiable football mechanism behind part of that gap.
Fourteen goals from distance
Villa scored 14 Premier League goals from outside the box in 2025/26. FIVDA had already identified Villa’s unusual finishing at the halfway stage, when their goals-minus-xG figure was the highest in the league.
Premier League analysis using Opta’s own model had them scoring 7.8 goals more than expected, the second-largest attacking overperformance in that model behind Tottenham. Premier League
That long-range output matters.
Shots from distance are generally lower-probability opportunities than shots from more dangerous positions. When a team repeatedly turns those attempts into goals, its actual scoring can separate quickly from its xG total.
That is exactly the type of distinction hidden by the phrase “Villa overperformed xG”.
The numbers establish that Villa scored more often than the quality of their shots would ordinarily suggest, and the 14 goals from outside the box give us a concrete feature of how that happened. Premier League
What they cannot tell us is how much of that difference represents repeatable finishing quality and how much reflects outcomes that would be difficult to reproduce next season.
That uncertainty matters.
Calling the entire difference “luck” would pretend we know more than we do.
Figure 2 — Who finished chances above — and below — expectation?
Villa actually had the largest positive goals-versus-xG gap in the StatMuse data at +7.54.
City followed at +6.04, Tottenham at +5.47 and Arsenal at +4.87.
At the opposite extreme was Crystal Palace.
And their season shows how different the same calculation can look when it goes the other way. StatMuse
Palace were the finishing outlier
Crystal Palace generated 57.64 xG.
They scored only 41 goals.
That leaves a gap of -16.64 goals — comfortably the largest attacking shortfall in the league using this model. StatMuse
This was not merely a late-season statistical quirk.
By February, Opta had already identified Palace as an extreme finishing outlier. At that stage they had scored 14 fewer non-penalty goals than their 37.0 non-penalty xG suggested. The next-largest deficit belonged to Nottingham Forest at six. Opta Analyst
Yet Palace’s overall discrepancy was smaller than their attacking one.
Why?
Because their defensive outcomes moved in the opposite direction.
Palace conceded 51 goals from 54.96 xGA, meaning they allowed 3.96 fewer goals than their xGA total. That defensive difference partially offset the enormous attacking shortfall, leaving their overall goal-difference gap at -12.68. StatMuse
This is why one headline xG number can conceal as much as it reveals.
Palace were not simply “12.68 goals below expectation”.
Their attack was much further below its chance output, while their defensive results pulled the overall number back in the other direction.
Everton show the opposite problem
Everton provide an almost mirror-image lesson.
They scored 47 goals from 47.45 xG.
That is about as close as a team can come to matching its attacking expectation across a full season: a difference of just -0.45. StatMuse
Yet Everton still finished with an actual goal difference 6.46 goals better than their expected goal difference.
Almost all of that came at the other end.
They conceded 50 goals from 56.91 xGA, a positive defensive gap of 6.91. StatMuse
Figure 3 — Which teams conceded fewer — or more — than their xGA?
Calling Everton an “xG overperformer” without specifying the defensive component would therefore tell the reader almost nothing useful about what happened.
Their finishing was essentially in line with their chances.
Their defensive outcomes were not.
Again, that does not allow us to assign the whole difference to one cause. xGA is measured from the chances conceded, before the eventual shot outcome is known. The difference between xGA and goals conceded can reflect more than goalkeeper performance alone.
The useful finding is the location of the discrepancy, not an invented explanation for it.
Wolves went the wrong way at both ends
If Everton demonstrate a concentrated defensive gap, Wolves show what happens when the two components reinforce one another negatively.
Wolves scored 27 goals from 35.66 xG.
That put their attack 8.66 goals below its expected total.
They then conceded 68 goals from 59.14 xGA, meaning their defensive outcome finished another 8.86 goals worse than the chance model suggested. StatMuse
Put the two together and Wolves finished with a -41 actual goal difference against a -23.48 xGD.
Their actual goal difference was therefore 17.52 goals worse than their expected one — the largest negative discrepancy in the Premier League. StatMuse Premier League
But even here, the important point is not simply that Wolves were the biggest negative outlier.
Their underlying chance numbers were already poor.
The discrepancy made the outcome worse.
It did not create the underlying problem.
That distinction also develops FIVDA’s mid-season analysis of Wolves, which had already found that poor underlying attacking numbers were being compounded by defensive outcomes running beyond their xGA.
Tottenham show how two anomalies can disappear
Tottenham offer perhaps the clearest example of why the decomposition matters.
They scored 48 goals from 42.53 xG, beating their attacking expectation by 5.47 goals.
Look only at goals versus xG and Spurs appear among the league’s strongest positive attacking outliers. StatMuse
Then look at the defence.
Tottenham conceded 57 goals from 50.90 xGA.
That is 6.10 goals more than their expected total.
The two effects almost perfectly cancelled one another.
Tottenham’s actual goal difference was -9. Their expected goal difference was -8.37.
Across the whole team, the final discrepancy was only -0.63 goals. StatMuse
Figure 4 — The Premier League’s xG outliers were built in different ways.
That makes Tottenham a useful warning.
A team can substantially exceed xG in attack and still finish almost exactly where its overall expected goal difference points because something completely different happened at the other end.
Looking only at attacking xG would miss half the season.
Sunderland show why expected points are different
There is one more distinction worth making.
Sunderland finished seventh with 54 points, despite Opta’s final Expected Points model placing them in the bottom three. Opta Analyst
Their season-total goal numbers also show a positive discrepancy, but it is much less dramatic.
Sunderland scored 42 from 39.06 xG and conceded 48 from 53.75 xGA.
That produced an expected goal difference of -14.69 against an actual goal difference of -6: a positive gap of 8.69 goals. StatMuse
So how can a team with the league’s third-largest positive goal-difference discrepancy end up in the bottom three of an Expected Points model?
Because they are answering different questions.
Season-total xGD adds together every expected goal scored and conceded across 38 matches.
Expected Points works fixture by fixture. Opta uses the xG of the shots in each match to simulate that match thousands of times, then calculates expected points from how often each side wins, draws or loses those simulations. Opta Analyst
That means the distribution of chances matters.
Ten expected goals spread across ten tight matches do not necessarily produce the same points picture as ten expected goals concentrated in a handful of dominant victories.
It is why an xGD ranking should never simply be presented as an “expected league table”.
And it is why Sunderland’s seventh place can look particularly unusual through an Expected Points model even though their season-total goal discrepancy was smaller than City’s or Villa’s.
Why “overperformance” does not mean “luck”
There is a temptation whenever actual goals move away from expected goals.
Call the difference luck.
Move on.
FIVDA’s analysis of Brighton’s finishing made the same distinction from the opposite direction: falling below xG identifies a gap between chances and goals, not its cause.
The 2025/26 Premier League shows why that is inadequate.
Villa’s positive discrepancy included 14 goals from outside the box. Premier League
City’s defensive results included a goalkeeper whom Opta rated among the league’s best by its goals-prevented measure. Premier League
Palace had already developed an enormous non-penalty finishing deficit months before the season ended. Opta Analyst
Those observations help describe what happened.
They still do not tell us exactly how much was skill, tactical design, goalkeeper performance, shot placement or ordinary football variance.
Expected-goal models also have limitations of their own. Opta notes that its Expected Points approach is affected by game state and does not capture threatening possession that never produces a shot. Opta Analyst
There is also no universal xG number.
Different providers use different models, which is why FIVDA has kept the league-wide xG decomposition here within the same StatMuse dataset and treated Opta’s figures as separate supporting evidence rather than combining the two.
The models are useful precisely because they give us a baseline.
They are not a replacement for the football.
What the xG gaps actually teach us
Manchester City had the Premier League’s largest positive gap between actual and expected goal difference.
But that is not the most important finding.
City’s +26.42 xGD already described an elite underlying team. Their actual +42 goal difference made that superiority look even greater. StatMuse Premier League
Villa are different.
They took a -7.29 xGD and finished with an actual goal difference of +7, fourth place and 65 points. Their long-range scoring gives us one identifiable reason their attacking results separated from their chance quality, without telling us how repeatable the entire gap will prove to be. StatMuse Premier League
Elsewhere, the same calculation tells completely different stories.
Everton were almost exactly on expectation in attack but well ahead defensively.
Palace’s negative gap was overwhelmingly about goals scored.
Wolves fell below expectation at both ends.
Tottenham’s positive attacking and negative defensive differences almost erased one another. StatMuse
That is the real lesson of the 2025/26 xG table.
There was no single type of overperformer.
There was only a gap between chances and outcomes.
The interesting football begins when we ask where that gap came from.
Stay tuned for future posts and please send us a message if there is a specific topic you would like to see covered.
As always, thanks for taking the time to read these posts!
JC
Sources for the data presented are referenced within the article.
The main image is from David Bayliss on Unsplash with the article title overlay.