How football statistics are analysed
Raw match data has to be cleaned, weighted and tested before it can support a prediction. This is how FootIQ does it.
Last updated:
Collecting finished matches
For every covered competition, all finished matches of the current and previous season are stored with full-time and half-time scores and, where available, xG.
Weighting by time
Recent matches matter more. Each match is weighted so that one played a year ago counts about half as much as one played today.
Home and away
Teams usually perform differently at home and away, so strengths are estimated with a home advantage, and one model uses separate home and away strengths.
Shrinking small samples
A team with few matches can look extreme by chance. The empirical-Bayes model pulls such estimates toward the league average until enough data exists.
Testing out of sample
Accuracy is only meaningful on matches the model has not seen. FootIQ predicts past weeks with models fitted on earlier data only and keeps the most recent weeks aside purely for measurement.
Knowing when not to predict
If there is too little reliable data for a match, no prediction is published. See football statistics.