Elo ratings, explained (and how Club Elo works)

If you want one number for how good a football team is, you could do a lot worse than the one a Hungarian-American physicist designed for chess in the 1960s. Árpád Elo wanted a fair way to rank players who never all met, and the system he built does something quietly useful: it turns a pile of results into a single figure, and turns the gap between two figures into a probability. That is the whole idea, and it travels from a chessboard to a football pitch almost unchanged.

An Elo rating has no units. On its own, 1893 means nothing; it only means something next to another rating. An average team hovers around 1500, and the strongest clubs in Europe push past 2000. What you actually read is the difference, so the two numbers below matter far less than the distance between them.

The gap is the whole story

Take two ratings and you can price the match. The rating difference goes through a single logistic curve, the difference divided by 400, and out comes an expected score between 0 and 1: the share of the points the stronger side should take on average. Level ratings give an even split. A 100-point edge is worth about 64%. A 400-point gap is roughly 91%, or ten to one. The 400 is just the scale Elo chose; it is why the numbers land where they do.

It is called an expected score, not a win probability, for an honest reason: draws fold into it. In a sport with three outcomes, splitting that expectancy into an explicit win, draw and loss, and then into actual scorelines, is a further step, and it is the step the match model does. Drag the ratings below and watch the gap become the odds.

Ratings move on surprise

The second half of Elo is how a rating changes. After a match, each team’s rating moves by a fixed amount, K, times the difference between what happened and what was expected: a win counts as 1, a draw 0.5, a loss 0. Beat a side you were expected to beat and you gain almost nothing, because you were already priced to win. Lose to them and you drop a lot. The rating only really moves when the result is a surprise.

Elo is zero-sum: the points one team gains, the other loses, which is why the playground above mirrors the two changes. The football versions add one refinement the chess version never needed. They scale K up for a bigger margin of victory and for a more important match, so a 4–0 in a final moves the needle further than a 1–0 in a friendly. Everything else is the same maths.

Club Elo, and Elo for nations

The version you will meet most often in club football is Club Elo, which rates clubs right across Europe and adds the margin-of-victory and home-advantage tweaks above. Its national-team cousin is the World Football Elo Ratings. Both are free and open, and both publish the working, which is exactly why we are happy to lean on them.

We now show a club’s Club Elo rating on its team hub, with a link back to the source, so the number is always checkable rather than asserted. Here are the clubs Touchline covers, lined up by that rating, so you can see the gaps the model sees.

What Elo can’t do

One number is a strength, not a forecast for Saturday. Elo knows only results; it has no idea a first-choice striker is suspended, that a side has not conceded in six, or that one team’s press is built to eat the other’s build-up. And it stops at an expected score. Turning strength into a spread of actual scorelines, with the draw priced honestly, is a different job.

That is where our engine picks up. Elo (among other inputs) helps anchor how strong each side is; the bivariate Poisson model then turns two expected-goals rates into every scoreline, played out 100,000 times. How those rates get set, and why our headline usually sits a shade under the crowd, is in how we set the numbers.