A team scores four goals in its opening match. Is it suddenly a great attacking side? Probably not, or at least not as great as that one afternoon suggests. This is the problem of small samples, and it affects every football statistic.
Luck is loud when data is scarce
Every match contains chance: a deflection, a goalkeeper error, a missed open goal. Over many matches these random events cancel out and the true quality of a team shows through. Over two or three matches they do not. The average of a short run is mostly noise with a bit of signal inside it.
Regression to the mean, explained simply
Here is a teaching example. Imagine the league average is 1.4 goals per team per match. A team scores 3.0 goals per match over its first two games. Does it deserve to be rated as a 3.0-goal attack? Common sense says no, since two games is too little evidence.
Regression to the mean formalises that instinct. The estimate is pulled back towards the league average, and the fewer matches there are, the stronger the pull. With two games, the adjusted figure might land closer to 1.7 or 1.8. With thirty games, the team's own average would dominate, and the pull would be tiny. The more evidence, the more we trust the team's own numbers.
The reverse works too: a side that fails to score in its first two matches is not assumed to be hopeless.
Where it matters most
- Early in the season, when each team has played only a few matches.
- Newly promoted teams, who arrive with little or no record at the new level.
- Short history windows, such as the last 5 matches, where noise is highest.
Predicta applies this idea to its attack and defence strengths. When a team has few matches behind it, the strengths are shrunk towards the league average, which prevents extreme values from distorting the estimate. As matches accumulate, the team's own record gradually takes over.
This makes early-season estimates deliberately cautious. It may feel less exciting than a bold number, but it is the more honest answer when evidence is thin.
Key takeaways
- Short runs are dominated by chance, not by true quality.
- Regression to the mean pulls extreme estimates towards the league average, more strongly when data is scarce.
- Early seasons, promoted teams and short windows need the most caution.
- Predicta limits extreme values this way so that few matches cannot produce wild estimates.