Premier League Power Rankings vs Final Table 2025/26 – Where the Model Got It Wrong

1st June 2026

The season’s done, the data’s in, and it’s time for a proper reckoning. Statz runs a power rankings model throughout the Premier League season – crunching xG, defensive solidity, attack output and everything in between to rank every team by underlying performance. Sometimes the model and the actual table agree. Sometimes they’re miles apart.

This year? The model nailed the top three perfectly. Arsenal, Man City, Man United – 1-2-3 in both the power rankings and the final standings. Job done, take a bow. But further down? It got properly weird.

Let’s dig into where the numbers and reality parted ways.

The Top Three: Model Says Yes, Table Says Yes

Arsenal finished on 85 points with the best overall xG profile in the league – 1.74 attack xG per game against just 0.83 conceded. Dominant at both ends. The power rankings had them first and that’s exactly where they ended up. No arguments.

Man City were second in both, generating the best attack xG in the division at 1.95 per game. They conceded more than Arsenal (1.16 xG against) but that front line is still absurd. 78 points, second place, model agrees. Man United completed the clean sweep at third – 71 points, third in the power rankings, third in the table. Three for three.

When the model works, it really works.

Sunderland: The Season That Broke the Algorithm

Here’s where it gets spicy. The power rankings had Sunderland 18th. Eighteenth. Their underlying numbers were grim – 1.14 xG created per game, 1.48 conceded, giving them a negative overall xG differential of -0.34. By every metric the model tracks, they looked like a relegation side.

They finished seventh. With 54 points.

Their first full Premier League season back and they’ve made an absolute mockery of the expected goals model. How? Grinding. Every single week, finding ways to nick results that the data said they shouldn’t be getting. Set pieces, late winners, defensive heroics that don’t show up neatly in xG. Whatever it was, Sunderland were the ultimate “vibes over data” team this season.

The gap between 18th in power and 7th in the table is the biggest disagreement across the entire division. Eleven places. The model will tell you they were lucky. Sunderland fans will tell you where the model can shove it. Both have a point.

Nottingham Forest: When Good Data Goes Bad

The power rankings had Forest eighth – a solid, respectable mid-table outfit with a positive xG balance (1.35 attack, 1.28 defense). The underlying numbers said Forest were creating enough and defending well enough to be comfortably in the top half.

They finished 16th. Just 44 points. Eight places lower than the model expected.

This is the kind of gap that makes data people lose sleep. Forest were doing enough on the pitch to be an eighth-place team but the actual results just didn’t come. Whether that’s individual errors not captured by xG, a lack of composure in key moments, or just plain bad luck with finishing – the model saw a team that should’ve been fine, and instead they spent half the season looking over their shoulder.

Aston Villa: Winning Ugly, Finishing Pretty

Villa’s numbers were thoroughly average. Attack xG of 1.52, defense xG of 1.48 – barely net positive. The power rankings placed them 10th. A nothing-special, mid-table performance in underlying terms.

They finished fourth. 65 points. Champions League places.

Six places higher than the model predicted, and they did it through sheer clinical efficiency. When the chances came, Villa buried them. When they needed a clean sheet from a scruffy 1-0, they got it. This is the classic overperformance case – the question is always whether it’s sustainable or whether it regresses next season. For now though, Villa don’t care about your xG models. They’ve got European football.

Chelsea: All the xG, None of the Points

Chelsea’s attack xG of 1.62 per game was good enough for the power rankings to place them seventh. They were creating plenty – the underlying offensive numbers were genuinely strong. But they finished 10th with just 52 points.

The conversion problem strikes again. Chelsea generated chances all season but couldn’t turn them into goals consistently enough. Three places might not sound like much, but the gap between 7th-place xG production and 10th-place reality tells you everything about their finishing. Lots of shots, lots of Expected, not enough Actual.

The Rest of the Pack

Some other notable gaps worth a mention. West Ham were 15th in the model but finished 18th and went down – the power rankings actually thought they were better than those three places suggest, which is cold comfort in the Championship. Everton were rated 16th but finished 13th, quietly overachieving in that Everton way where nobody really notices.

Leeds (12th power, 14th table) and Bournemouth (4th power, 6th table) both underperformed their data slightly but nothing dramatic. At the very bottom, Burnley and Wolves were always going to struggle – Burnley’s defensive xG of 2.01 conceded per game was the worst in the league by a mile. You’re not surviving that.

What Does This Tell Us?

The Statz power rankings model is built on underlying performance – what teams should be doing based on the chances they create and concede. It’s not trying to predict final position directly. It’s measuring quality of play.

When the model and table disagree this much, it usually means one of two things: either the team is significantly over or underperforming their data and will regress, or there’s something the model isn’t fully capturing – mentality, set-piece quality, game management.

Sunderland finishing 7th with 18th-rated underlying numbers is either the story of a team that found something the data can’t measure, or a correction waiting to happen. Forest looking like an 8th-place team but finishing 16th suggests either terrible luck or fundamental issues the xG model is too kind to spot.

Either way, the beauty of the model is that it gives you a second opinion. The table tells you what happened. The power rankings tell you what probably should have happened. The gap between the two? That’s where the interesting stories live.