Coventry vs Hull: when a prediction model has no real favourite
Published
No fixture on this Premier League matchday has a lower leading probability than one game in the West Midlands. In our 26 August run, Coventry City vs Hull City comes out at 37.0% Coventry, 27.3% draw, 35.7% Hull — the highest of the three outcomes reaches only 37, and home and away are separated by 1.3 percentage points. This is what it looks like when a prediction model has no real favourite, and the interesting question is not which side to nudge it towards — it is whether a flat number like that is worth publishing at all.
A 1.3-point gap is not a favourite
It is tempting to call Coventry “narrow favourites” and move on. That would be a misreading. A 1.3-point separation is smaller than the uncertainty on either estimate; if the model were refit tomorrow with one more week of data, the order could flip without anything meaningful having changed. The correct summary of 37.0 / 27.3 / 35.7 is “we do not know”, stated in three decimal places.
Our confidence score for the fixture is 37.0 — which, on our scale, is simply the top probability passed through undamped. It is not saying the pick is solid. It is saying the top probability is 37, and 37 is a low ceiling for a three-outcome market. For contrast, the same model run has Barcelona at 72.1% at home to Rayo Vallecano on the same weekend, and Manchester United at 58.8% against Ipswich on the same Premier League card. That is what an opinion looks like. This is not one.
Why the market is 29 points away from us
The prices carried on our match page as of the 27 August feed pull are Smarkets 1.94 Coventry, 3.70 draw, 4.50 Hull. Strip the small overround and that is roughly 51% / 27% / 22% — the market and our model agree almost exactly on the draw, and disagree by fourteen points on each of the other two. Add the two disagreements together and the market’s home-minus-away spread is 29 points where ours is 1.3.
Independent quotes point the same way: EasyOdds’ preview has Coventry at 1.94 and Hull out around 4.07, read on 27 August 2026. The market is not confused. It has a clear view, and the view is Coventry.
The gap is explained by what each side is looking at. The market is pricing opening weekend, in which Coventry lost 0-3 at Arsenal and Hull won 2-0 at home to Manchester United, and then discounting Hull’s result heavily as an upset while treating Coventry’s as a forgivable defeat at the champions — plus a full share of home advantage. Our Dixon–Coles fit weights those two matches like any other pair of scorelines and has no separate machinery for “that result was a fluke”. Everything else it knows about these clubs comes from divisions below this one.
That is a genuine limitation and it cuts against us here as much as for us: the model is carrying second-tier scoring rates into a top-flight fixture and is honest enough about it to produce a flat number, while the market has priced in a judgement the model is not equipped to make. We publish the flat number because the alternative — inventing a favourite to look decisive — is the thing that makes a track record unreadable later.
The one thing both records agree on
There is a strand where the two views converge, and it is goals. Our run has over 2.5 goals at 52.0%, both teams to score at 56.9%, and the single most likely scoreline at 1-1 with a 13.0% share. The market’s most-backed angle, per the same EasyOdds preview, is under 2.5 goals at around 1.95, and the draw at 27/10.
The history behind that is real. KhelNow’s preview records 53 previous meetings — 21 Hull wins, 18 Coventry, 14 draws — and notes that “the last two outings between these clubs ended in a goalless draw”. None of those meetings were in this division: Yahoo’s preview confirms the two clubs have never met in the Premier League, making Saturday a first at this level for both.
Worth being precise about what our model does with that record: nothing directly. It has no head-to-head term. The 1-1 sitting on top of our score grid is not an echo of two goalless draws — it is where two similar, moderate goal expectations put the peak. The two arrive at the same place by different routes, which is more interesting than either would be alone.
Reading a flat probability without flattening it
Team news is the part neither the model nor the score grid contains, and it is lopsided. KhelNow lists Haji Wright out with a muscle injury for Coventry and Ephron Mason-Clark carrying fitness concerns, against a Hull list of Oscar Zambrano, Joe Gelhardt, Eliot Matazo, Darko Gyabi, Matty Jacob, Lewie Coyle, Charlie Hughes and Jack Butland, with Hidemasa Morita not yet confirmed available.
Eight names against two is the kind of asymmetry that would justify moving a probability — and the model has not moved it, because it cannot. If you weight that list, the honest correction is towards Coventry, which happens to be the direction the market has already gone. That does not vindicate the market’s 51%; it means our 37% and the market’s 51% are answering slightly different questions, and the reader who knows about the absences is better informed than either.
So the practical reading of this fixture is: no side has an edge worth acting on from the scoring record alone, the goals market is where both our model and the market have their firmer view, and the team sheets are where the actual information is this week. Every number quoted here is frozen at the 26 August run and the 27 August feed pull; the match page re-renders at every build, the rest of matchday 2 is on the Premier League page, and how calls like this one have settled over time is on the track record.