I run a Premier League prediction game. Over seven seasons its players have made 1.2 million predictions. Line those up against the results and a pattern shows up: the crowd gets the smaller clubs about right, and the big ones badly wrong.
Every club with at least 100 matches, sorted by how strongly the crowd backs them.
| Club | Crowd backs them | They actually win | Gap | Matches |
|---|---|---|---|---|
| Man City | 86.3% | 68.4% | +17.9 | 266 |
| Liverpool | 80.2% | 61.7% | +18.5 | 266 |
| Arsenal | 69.0% | 57.9% | +11.1 | 266 |
| Chelsea | 64.9% | 46.0% | +18.9 | 265 |
| Man Utd | 57.3% | 47.9% | +9.4 | 265 |
| Spurs | 55.7% | 43.6% | +12.1 | 264 |
| Newcastle | 41.5% | 39.8% | +1.7 | 264 |
| Leicester | 40.6% | 35.3% | +5.3 | 190 |
| Aston Villa | 40.2% | 42.3% | -2.1 | 265 |
| Brighton | 33.1% | 33.3% | -0.2 | 264 |
| West Ham | 33.0% | 34.8% | -1.8 | 264 |
| Everton | 31.4% | 32.6% | -1.2 | 264 |
| Brentford | 30.2% | 35.8% | -5.6 | 190 |
| Wolves | 27.3% | 30.2% | -2.9 | 265 |
| Crystal Palace | 26.8% | 30.9% | -4.1 | 265 |
| Leeds | 26.7% | 29.8% | -3.1 | 151 |
| Bournemouth | 25.7% | 31.7% | -6.0 | 189 |
| Nott'm Forest | 25.4% | 31.8% | -6.4 | 151 |
| Fulham | 25.3% | 33.3% | -8.0 | 189 |
| Southampton | 20.0% | 23.2% | -3.2 | 190 |
| Sheffield Utd | 15.9% | 21.9% | -6.0 | 114 |
| Burnley | 15.5% | 22.3% | -6.8 | 188 |
Chelsea win 46% of their matches and Aston Villa win 42%, so there is 4% between them. But the crowd backs Chelsea 65% of the time and Aston Villa 40%. That is a 25% gap in belief between two clubs whose records are almost the same.
It is the gap that matters here, not the level. Manchester City are backed at 86% because they usually are the better side. The point is that even then, they win 18% less often than the crowd expects.
The six most overrated clubs in the league are the Big Six. Nobody else is off by more than 8% either way, and the crowd gets Brighton almost exactly right: backed at 33.1%, winners 33.3% of the time.
None of which means the crowd is bad at predicting football. On most of the league it does a pretty good job. It just cannot see the big clubs clearly.
Home advantage is not a constant. The crowd's estimate of it is.
| Season | Crowd picks a home win | Home wins actually | |
|---|---|---|---|
| 2019/20 | 48.7% | 45.0% | |
| 2020/21 | 48.3% | 37.6% | Empty stadiums |
| 2021/22 | 47.8% | 42.9% | |
| 2022/23 | 48.4% | 48.3% | |
| 2023/24 | 49.1% | 46.1% | |
| 2024/25 | 49.0% | 40.7% | |
| 2025/26 | 48.8% | 42.8% |
The crowd's number sits between 47.8% and 49.1% every year, a range of just 1.3% across seven seasons. The real one ranged from 37.6% to 48.3%.
In the empty-stadium season, home teams won 37.6% of their matches and the crowd still predicted 48.3%, the same figure it lands on every other year. Whatever people are doing when they back a home side, they are not reacting to what is actually happening on the pitch.
The most-predicted scorelines, against how often they actually happen.
| Scoreline | Predicted | Happens |
|---|---|---|
| 2-1 | 17.88% | 8.46% |
| 1-2 | 14.16% | 6.92% |
| 1-1 | 13.75% | 11.24% |
| 2-0 | 12.33% | 6.95% |
| 1-0 | 7.23% | 8.35% |
| 0-2 | 6.92% | 5.68% |
| 0-1 | 4.40% | 7.11% |
| 0-0 | 0.92% | 5.51% |
2-1 is predicted more than twice as often as it happens. And 0-0, which comes up in one match in eighteen, more often than 3-1 or 3-0, gets predicted once in every 109 predictions.
It would be easy to put all this down to football fans being irrational. A lot of it is really just the scoring.
In this game you score 10 points for calling the right result and 40 for the exact score. The 10 arrives whichever scoreline you pick, as long as you called the winner or the draw correctly, so the score you choose on top of that is a free shot at the other 40.
That is what leaves 0-0 out. Nothing in the rules is stacked against it, and it scores exactly as any other line does. It is just that if you think a match is a draw, 1-1 pays the same 10 and lands about twice as often, so it is the better way to say so. The split proves it: nearly three quarters of the crowd's draw predictions are 1-1 and under 5% are 0-0, where real draws are 47% 1-1 and 23% 0-0.
Draws get under-called in general too, 18.9% of predictions against a real rate of 23.9%, and the reason there is plainer. You get one prediction per match, and a draw is rarely the single likeliest outcome of a match even though draws are nearly a quarter of all results. The picks go to the narrow wins instead: 2-1 and 1-2 between them account for almost a third of every prediction ever made here.
What the scoring does not explain is the club chart. Backing favourites makes sense under any rules, but being 19% out on Chelsea and 0.2% out on Brighton is not something the maths produces on its own. That one looks like reputation.
Happy for any of this to be quoted or charted. If you want a cut I have not published, like a single club or a single season, just ask and I will run it.
PLPredictor is a Premier League score-prediction game. Every figure here comes from 1,198,201 predictions made by 2,370 players across 2,660 completed Premier League matches, from the 2019/20 season to 2025/26. Predictions are locked at kick-off, so none of them were made with the result known.
No, and that is the point. Everyone in the game predicts every match, so the people backing Chelsea are mostly neutrals with nothing at stake. It is not loyalty to a club, it is a reputation the results stopped backing up years ago.
Partly, and I checked that rather than assuming it. Backing the likelier side does inflate every strong club's number. But it cannot explain the spread: Manchester City are heavy favourites most weeks and Chelsea are not, yet Chelsea come out overrated by more. The crowd is also close to exact on clubs like Brighton, which a general leaning towards favourites would have pulled off too.
Because 1-1 is a better way to back a draw. The 10 points for calling a draw arrive whichever drawn scoreline you pick, so the score itself is a free shot at the other 40 - and 1-1 lands about twice as often as 0-0. Nearly three quarters of the crowd's draw predictions are 1-1, against under 5% for 0-0. Nothing in the scoring is stacked against 0-0 in particular. It is just rarely the best way to say "draw".
No. The same scoring pressure behind these patterns makes the crowd a poor forecaster: it picks 2-1 more than twice as often as it happens, and almost never calls a goalless draw. It is a good record of what people believe, rather than of what is going to happen.
This is the game the data comes from. Predict every Premier League scoreline, run a private league with your friends and react to their picks while the match is still going, and find out whether you are one of the people dragging these averages around. It is free, and it takes about a minute.
Play the Premier League Predictor