How football prediction models work

A football prediction model turns past results into probabilities for future matches. It does not know who will win. It estimates how often each result would happen if the same match were played many times.

Rating attack and defence

Most models start by rating teams. Scoring three goals against a strong defence says more than scoring three against a weak one, so good models adjust for the opposition. They also account for home advantage and give recent matches more weight.

From ratings to expected goals

Combining one team's attack with the other's defence gives a forecast of how many goals each side is likely to score. Those forecasts are averages, such as 1.7 for one team and 0.9 for the other.

From expected goals to scorelines

Goals are rare and fairly random, which suits a statistical pattern called the Poisson distribution. It turns an average into the chance of 0, 1, 2 or more goals. Combining both teams gives the chance of every scoreline, and every betting market can then be worked out from that grid.

What models miss

A results-based model cannot see injuries, suspensions, a new manager or a team resting players before a cup final until those things show up in results. That is why you should check team news.

How to judge a model

Look for calibration: when a model says 70%, the outcome should happen about 70% of the time. A single good or bad weekend proves very little. You can read how our model works and check its public track record.