What Championship matchday 2 told us about defence ratings in August
Published
The second Championship round of 2026/27 finished at The Hawthorns on Sunday lunchtime, and the thing worth taking from it is not who is top. It is that three of the division’s weaker attacks, by our own ratings, put eight goals past three sides our model rated above them. A defence rating is the slowest, thinnest thing a goal model estimates, and matchday 2 is a good place to show why that matters in August specifically.
All model figures below come from our 20 August run, before any of these matches kicked off. All ratings are as of 23 August, ranked within the 24-team division.
Three bottom-half attacks, eight goals
Start with the clearest case. Our model ranks Burnley’s defence 1st of 24 — the strongest in the division — and rates them top on Elo overall. They went to West Brom, a side our model ranks 16th for attack, and conceded three. Sky Sports reported that Isaac Price set up Harry Whitwell’s equaliser and then scored twice himself after the break, with Burnley — missing Zian Flemming to a calf injury amid interest from Lille — failing to match their billing as promotion favourites. Their own goal in the game was a Callum Styles own goal, so the division’s third-ranked attack did not score at all.
Then Blackburn against Middlesbrough. Blackburn’s attack is ranked 24th of 24 in our ratings — dead last. Middlesbrough’s defence is ranked 3rd. Our model made Middlesbrough 39.3% away favourites against Blackburn’s 31.4%. Jeremy Sarmiento put Boro ahead on 12 minutes; Andri Gudjohnsen headed in on 49 and Todd Cantwell added the winner within six second-half minutes, per Sky Sports. The worst-rated attack in the division scored twice against the third-best defence, in six minutes.
And West Ham against Charlton, where the model’s own warning was sitting in plain sight and its headline number buried it. West Ham are ranked 19th of 24 defensively, Charlton 23rd for attack, and the model still made it 47.0% West Ham against 25.4% Charlton, because attacking rank and home advantage outweighed the defensive one. Charlton won 2-1, VAVEL reporting that Jobes opened the scoring with a bicycle kick inside the opening minutes.
Why a defence rating is the slowest thing a model learns
An attack rating is estimated from a common event. Teams score in most matches, so a season produces plenty of evidence about how many goals a side generates, and the parameter settles reasonably quickly.
A defence rating is a claim about the absence of an event. Clean sheets are rare, one-goal games are noisy, and the difference between conceding 38 and conceding 46 across a season — the difference between a top-four defence and a mid-table one — is eight goals spread over nine months. That is a thin signal, and it means defensive parameters carry more of last season’s context and shed it more slowly.
Now add the August problem on top. Two rounds in, almost everything in these ratings is inherited: last season’s results, mostly from a different division for some of these clubs, from squads that have since been rebuilt. Burnley’s division-best defensive rating and West Ham’s poor one are both statements about a period that ended months ago, made about squads that have changed since. The model has no mechanism to know that, which is the same limitation we set out for promoted clubs — it just points the other way for clubs arriving from above.
Where the ratings did hold
This is not a round in which everything broke, and it is worth being precise about what did not.
Southampton beat Stoke 3-1 — our model’s most confident call of the round at 57.5%, and Southampton’s attack is ranked 1st of 24. The rating said they would score, and they scored three. Millwall beat Norwich 3-0, and Millwall’s defence is ranked 2nd of 24; they kept the clean sheet the rating implied, in a fixture where the model had put over 2.5 goals at only 45.6%. West Brom’s own defence, ranked 4th, conceded only through their own player.
The pattern in that is not comforting so much as clarifying. Where an extreme rating belonged to the side already favoured, it did what it said. Where an extreme rating was the reason a side was favoured against an opponent nobody expected to score, it did not.
The draw the model rated highest was the one that wasn’t
Five of the twelve fixtures finished level: Birmingham 2-2 Bristol City, Wrexham 1-1 Watford, QPR 0-0 Bolton, Swansea 0-0 Sheffield United and Derby 2-2 Cardiff.
The detail worth keeping is that the highest draw probability our model produced anywhere in the round — 30.3% — belonged to West Brom against Burnley, the one fixture with a three-goal winner. In the two goalless draws it had the draw at 26.9% and 27.9%, ranked third of three outcomes both times. That is not a scandal; a goal model built on Poisson-style scoring rates structurally struggles to make the draw its top pick, and the Championship is the division where that bites hardest. But it is a concrete instance of it, on a Saturday, in a division where more than a third of the card finished level.
The two remaining fixtures both went against the home side the model favoured: Lincoln lost 1-3 to Portsmouth from a 47.2% home figure, and Preston lost 1-3 to Wolves from a much narrower 38.6% against 33.9%.
What matchday 3 asks
The useful thing about a round like this is that it is a dated, checkable test of a specific claim rather than a mood. Every figure above was published before kick-off and carries its model version; how the calls actually graded, across every league and every round, lives on the public track record rather than in a paragraph we wrote afterwards. The division’s fixtures and ratings sit on the Championship hub, and they will keep moving — a defence rating built on the wrong division is exactly the kind of number that a few rounds of real evidence repairs.
The matches, at a glance
ChampionshipFT
West Brom3
Burnley1ChampionshipFT
Blackburn2
Middlesbrough1ChampionshipFT
West Ham1
Charlton2ChampionshipFT
Southampton3
Stoke1ChampionshipFT
Millwall3
Norwich0ChampionshipFT
Lincoln City1
Portsmouth3ChampionshipFT
Preston NE1
Wolverhampton3ChampionshipFT
Birmingham2
Bristol City2ChampionshipFT
Wrexham1
Watford1ChampionshipFT
QPR0
Bolton0ChampionshipFT
Swansea0
Sheffield Utd0ChampionshipFT
Derby County2
Cardiff2