Crystal Palace vs Man City: how much one absence changes a probability
Published
“Star winger out” is a headline. “Star winger out, therefore back the other side” is a conclusion, and the step between them is almost never shown. Jeremy Doku misses the start of Manchester City’s season, and this post is an attempt to show that step: how much one absence changes a match probability, using City’s trip to Selhurst Park on Friday 28 August as the worked example. Our model’s 20 August run has it 23.4% Crystal Palace, 26.1% draw, 50.6% Manchester City, and those numbers know nothing about Doku — the current version is on the match page.
What is actually confirmed
Doku picked up a calf injury in the Community Shield defeat to Arsenal and was withdrawn at half-time; Goal reported on 18 August, citing Nieuwsblad, that he could be out for several weeks, ruling him out of the opener against Bournemouth. A Friday-night fixture six days later is inside that window rather than obviously outside it, which is the sort of ambiguity worth stating plainly instead of resolving in whichever direction suits the argument.
On the other side, Palace manager Pierre Sage said before the Everton trip that he was hopeful about both Dean Henderson, who twisted an ankle in a friendly against Freiburg, and Jean-Philippe Mateta, per the club’s own team news.
None of that reaches our model. It estimates team strength from results, so its 50.6% for City was computed with no team sheet, no injury list and no notion that Doku exists.
Four percentage points, not twenty
Here is the arithmetic, and it is deliberately crude so that it can be checked. Take the published 1X2 split, approximate it with an independent-Poisson model — our own back-of-envelope working, not the model’s internals — and then ask what happens if City’s scoring rate drops.
- Scoring rate cut by 5%: City 50.6% → 48.6%, draw 26.1% → 26.8%, Palace 23.4% → 24.6%
- Cut by 10%: City → 46.6%, draw → 27.6%, Palace → 25.8%
- Cut by 15%: City → 44.4%, draw → 28.4%, Palace → 27.1%
A 10% cut in a team’s expected goals is already a strong claim for one wide forward in a squad this deep. It buys Palace 2.4 percentage points and the draw 1.5. The absence is real, it points the right way, and it is worth roughly four points off the favourite — not the twenty that “City without Doku” tends to imply in a preview.
That is the general lesson, and it survives the specific case: single-player absences move probabilities by low single digits unless the player is a goalkeeper, a first-choice centre-forward on a thin bench, or several players at once.
Why City are barely favourites here anyway
The more interesting number is the starting one. Our team ratings, updated 22 August, put City 1st of 20 in the Premier League for attack and 2nd for defence on 1765 Elo, against Palace at 19th for attack, 13th for defence, on 1525. That is a 240-point gap — wider than the 217 points between Bayern and Stuttgart in Friday’s Bundesliga opener — and yet City’s win probability is 50.6% where Bayern’s is 67.9%.
Two things explain the difference. Bayern are at home and City are not. And this is a low-scoring environment: the model gives 52.4% for over 2.5 goals here against 80.8% in Munich, with the single likeliest scoreline a 1-1 at 12.4%. Fewer goals means fewer opportunities for the better team’s superiority to express itself, so the favourite’s edge converts into a smaller win probability. Palace’s rating profile — a weak attack, a mid-table defence — is exactly the profile that drags matches into that territory, which is why the draw sits at 26.1%.
It is worth noting where our ratings sit uneasily with the trophy cabinet. Palace won the UEFA Conference League in May 2026 and finished 15th in the Premier League, per the club season record. Our model weights the 38 league matches, not the cup run, so it sees the 15th-place team. Against City specifically, that read has been unkind and accurate: 0-3 at Selhurst Park in December 2025 and 0-3 at the Etihad in May 2026.
The changes the model has not priced
Two are large enough to name. Rodri has joined Barcelona and Tijjani Reijnders has left for Al Qadsiah, both noted in the same Goal report — the model does not read transfer news, so any strength change from those departures only enters once City play matches and the results update their ratings. And Palace have a new manager: Pierre Sage arrived from RC Lens to replace Oliver Glasner on a deal to 2029, as ESPN reported. A managerial change is invisible to a ratings model until performances move.
This is the part we would rather say out loud than have a reader discover: in late August, every model number carries last season’s information plus a handful of matches. The absence of a winger is the smallest of the uncertainties on this page.
Reading the number honestly
Adjust the 50.6% down a few points for Doku if you want; the model will not, because it has no mechanism to. What it will do is publish the figure, date it to the run that produced it, and let the result be graded in the public track record whichever way it goes. The rest of the round is on the Premier League hub, and Crystal Palace’s team page carries their ratings and fixtures as the season updates them.