Real Madrid vs Real Sociedad: when expected goals and the result disagree
Published
Real Sociedad arrive at the Bernabéu on Wednesday 26 August having lost their opening match, which is the entire story if you read the results column, and roughly half of it if you read anything else. This is a fixture where expected goals and the result disagree about the away side — a disagreement worth quantifying rather than picking a side in. As of our 21 August model run, the split is 67.4% Real Madrid, 18.3% draw, 14.2% Real Sociedad; the current version is always on the match page.
A defeat the underlying numbers dispute
Sociedad lost 1-0 at Real Betis on 21 August to a Rodrigo Riquelme finish just past the hour, according to Sofascore’s match report. The same report has Betis ahead on volume — 18 shots to 10, seven on target to three — while crediting Real Sociedad with the higher expected goals, 1.58 to 1.06, with goalkeeper Álex Remiro rated the outstanding performer on the night.
So: fewer shots, better ones, no points. That is a common shape and it is exactly the case where the scoreline is the least informative thing that happened. It is also, per the same report, a seventh consecutive away La Liga match without a win — which is the argument in the other direction, and belongs in the post as much as the xG does.
Sociedad’s squad news adds context our model never sees. Manager Pellegrino Matarazzo was without Álvaro Odriozola and Jon Gorrotxategi for the opener, with Takefusa Kubo only recently upgraded to group training after a knee problem and Mikel Oyarzabal left out of the starting XI, as OneFootball reported before the Betis match.
The La Liga table is not a ranking this week
There is a second reason to distrust the standings right now. This match is Matchday 1, being played four days after Matchday 2. Real Madrid’s opener against Sociedad was postponed because of the number of Madrid players involved in the closing stages of the 2026 World Cup, so their season actually began at Espanyol on 22 August, with Málaga at home on 30 August, as beIN Sports set out.
Clubs have played different numbers of matches, so points totals are not comparable and positions are noise. Anyone reading form off the table this week is reading a table that has not finished being a table.
What our model did with all of this: nothing directly
Our model does not consume league position, and it does not consume expected goals either. It estimates each team’s attack and defence strength from actual match results, weighted towards the recent ones, and turns the pair into a scoreline distribution — the method is documented here. One 1-0 at Betis is one result, entering with the weight of one result.
What that produces, in our team ratings updated 22 August: Real Madrid rank 2nd of 20 in La Liga for attack and 1st for defence, at 1775 Elo. Real Sociedad rank 7th for attack but 17th of 20 for defence, at 1502. The 273-point Elo gap, plus home advantage at the Bernabéu, is where 67.4% comes from — not from Sociedad having lost on Friday.
Note the shape of Sociedad’s profile, because it matches the Betis performance rather than the Betis scoreline: a mid-table attack and a genuinely weak defence. Our model has them creating enough and conceding too much. That is why the goals lines here read 61.1% for over 2.5 and 53.5% for both teams to score, and why the single likeliest scoreline in the grid is a fairly ordinary 2-0 at 10.7%.
Putting a number on the disagreement
Suppose you side with the expected goals and think Sociedad’s attack deserves a mark-up. How much is that belief worth?
Take our published 1X2 split, approximate it with an independent-Poisson model — our own arithmetic, a simplification of the model rather than its internals — and raise Sociedad’s scoring rate by 10%. Their win probability moves from 14.2% to 16.0%. Push it to a heroic 20% mark-up and you reach 17.8%, with Madrid still above 63%.
That is the honest size of the effect. One over-performing xG line in one match is worth two or three percentage points of away win, not the fifteen that a confident write-up would imply. The reason is not stubbornness; it is that a single match, however you measure it, is a single match. Our guide to expected goals makes the same point at greater length: xG is a better description of one game than the scoreline, and still a small sample of one game.
Where this call is genuinely exposed
The clearest weakness is not the xG argument. It is that our ratings for both sides are built mostly on last season, and both squads have moved on in ways results have not yet revealed — plus a World Cup summer that left elite squads with unusual preparation, none of which the model can read. Madrid playing Matchday 2, Matchday 1 and Matchday 3 inside nine days is a workload the numbers above take no view on whatsoever.
The model’s answer to all of that is to stay at 67.4% rather than pretend to a precision it does not have. Whatever it says at kick-off is recorded and graded publicly, misses included, in the track record. The rest of the round, and the rest of the scrambled early table, sits on the La Liga hub; Real Sociedad’s team page carries their ratings and fixtures as they update.