A team’s form: over how many matches should it be measured?
Five matches can be read at a glance and say almost nothing. The right window is wider than intuition suggests.
By The PRIZIFOOT team
The last five results are the usual yardstick, because they fit on one line. That is a display choice, not a statistical one, and the two do not coincide.
The small-sample problem
Over five matches, a single 4-0 shifts a team’s goal rate by nearly a goal per match. That jolt then propagates into everything derived from it: attacking strength, expected goals, final probability. You think you are measuring momentum; you are measuring one match.
Widening the window reduces that noise, up to the point where you start describing a team that no longer exists: squad renewed, manager changed. There is therefore a trade-off, and it is measured rather than guessed. The window is tuned on matches held out, and the one that best predicts those matches is kept.
Weighting rather than cutting off
An alternative to the cut-off is to give each match a weight that decays with age, through a half-life: past a certain age, a result counts half as much. That stops a match from dropping out of the window overnight, and lets distant history weigh a little rather than nothing.
The thing to remember: displaying five matches and computing over five matches are two separate decisions. Nothing forces the second because you took the first.
Move from theory to the match
Pick two teams and see what the model makes of it, without creating an account.
Analyse a match