How to read our predictions

A statistical prediction does not say what will happen: it says how likely it is to happen. This guide explains the three predictions in every preview, from the most likely to the boldest, and how to read the percentages.

What you find in each prediction

For every game you get three predictions: the most likely, a balanced one and the boldest. Each comes with the probability estimated by the model and the reasoning behind it, written from the season’s numbers and the absences. The three predictions tell the same story about the game: the second and third take more risk on the same reading.

Result and double chance

1 means the home team wins, X a draw, 2 the away team wins. Double chance covers two outcomes out of three: 1X (home win or draw), X2 (away win or draw), 12 (no draw). It is more likely, but it also says less.

Total goals

Over and Under refer to the goals in the whole game: Over 2.5 means at least 3 goals, Under 2.5 two or fewer. The lines go from 0.5 to 4.5.

Teams and both teams to score

Here you find whether a team scores, scores at least twice, keeps a clean sheet or wins by at least two goals. “Both teams to score” means each side scores at least once; “both teams to score: no” means at least one side fails to score.

Combinations, first half and scorers

Combinations join two conditions, for example 1 & Over 2.5: the home team wins and there are at least 3 goals. “Win to nil” means winning without conceding. “Over 0.5 first half” means at least one goal before half-time. In the Serie A predictions you will also find scorers: “to score” means the player scores at least once.

What the percentage means

Next to each pick we show an estimated probability, for example 45% for a home win. It means that, according to the model, in a game like this one the home team would win in about 45 cases out of 100 and not in the other 55. Even the most likely pick can be wrong, and below 50% it is wrong more often than right. An event rated at 70% does not happen 3 times out of 10.

Most likely, balanced, boldest

The level depends on the estimated probability. The most likely prediction usually sits between 65% and 85%, the balanced one between 45% and 66%, the boldest between 28% and 50%. We do not use picks above 90%, such as “at least one goal in the game”: they are almost certain and say nothing about the game. The bolder the prediction, the more often it fails: a 40% prediction is wrong six times out of ten. No prediction is safe.

How the model works

For each game the model estimates the expected goals of both teams from the goals they have scored and conceded this season, pulled towards the league average because a few games are a small sample, with a home advantage. For the result it also blends in an outside estimate of team strength. From the expected goals it computes the probability of every score, and from those the probability of every prediction.

Three mistakes to avoid

• Reading “70%” as “certain”. It means likely, not sure.
• Trusting a handful of results. Early in the season every team has played only a few games: one unusual result moves the averages a lot.
• Forgetting what the model does not know. The percentages do not account for injuries, suspensions and line-ups. We consider absences in the reasoning, but the numbers remain the model’s.

Where to start

Pick a league and open the preview of a game: Serie A · Premier League · La Liga · Bundesliga · Ligue 1. To see how past predictions did, open the archive.