How the PreviaXI model works

Every probability on PreviaXI (win, draw and loss chances, over 2.5 goals, both teams to score, expected corners and cards) comes from the same statistical model. This page explains what goes into it, how it works and how well it has done.

The data

The model uses results and match statistics from a licensed data provider: goals, expected goals (xG), corners and cards for every league match of the last two seasons, plus European and domestic cup matches. Fixtures are refreshed every three hours.

Team ratings

Each team gets an attacking and a defensive rating, plus a home advantage for its league. Ratings are fitted with a Dixon-Coles model, the standard statistical approach to football scores, and blend actual goals with expected goals (70% xG), because xG is a steadier measure of how well a team plays than goals alone. Recent matches count more than old ones (a match from a year ago counts half as much as yesterday's), and teams with few matches are pulled towards the league average until there is enough evidence.

From ratings to probabilities

For each fixture the ratings give the expected goals of both teams. From those, the model computes the probability of every possible scoreline and adds them up into the numbers you see: home win, draw, away win, over 2.5 goals, both teams to score and the most likely score. Corners and cards use a separate count model (negative binomial) built from each team's corners and cards for and against.

Cups and European competitions

To compare teams from different leagues in the Champions League, Europa League, Conference League and the domestic cups, the model is fitted on 27 European leagues at once, so a team's rating already reflects the strength of the league it plays in. Other leagues are modelled on their own results, and other cups on the recent matches of the teams involved. Friendlies are not modelled.

How accurate is it?

In a backtest on the 2025/26 Premier League season, the most likely result (home win, draw or away win) was right 49% of the time, the over/under 2.5 goals call 58% of the time and the both-teams-to-score call 57% of the time. Football is unpredictable: a 60% favourite still fails to win four times in ten. Predictions are frozen at kick-off and compared with the final score on our prediction record page.

Limits

The model does not know about late team news, line-ups, weather or motivation, and it can be slow to react when a team changes coach or squad. Treat the numbers as an informed starting point, not a certainty. They are statistics for information and entertainment, not betting advice.