Form is the most tempting part of analysis. It is easy, it is everywhere and it looks like news. The problem is that five games are far too few to say anything stable.
How much noise a small sample carries
Say a team truly wins 45% of its games. Over n games, the win rate you see has a standard error of √(p × (1 − p) / n). The fewer the games, the bigger the error.
Conclusion: the last five games are not enough. A team "in form" after five wins was often just lucky. Form gets a small weight and never replaces long-run strength.
Home and away
Home advantage is real and shows up over thousands of games. One team’s own home advantage, though, comes from 10 or 15 games, so it carries a large error.
| Team A | Games | W-D-L | Points per game |
|---|---|---|---|
| Home | 10 | 6-3-1 | 2.10 |
| Away | 10 | 3-3-4 | 1.20 |
Strength of schedule
Ten points in five games is something even an average team collects when the opponents are weak. Before you give form any value, ask who it was against. If those opponents would give an average team 1.9 points per game, which is 9.5 over five, then Team A’s 10 points are an overperformance of just 0.5 points. Almost nothing.
- Measure form with goals and good chances, not only results.
- Widen the sample: 10 to 20 games instead of 5.
- Compare the opponents, not only the points.
What to do with this
Put form in its own small box in the analysis. Give it a weight that grows with the number of games and stop when the error is smaller than the difference you are trying to see. The same rule applies to any number with a small sample: goals in the last few matches, corners over a month, cards from one referee. If the gap between two teams is smaller than the standard error, you have not found a gap. You have found noise. One more thing: a team that strays far from its average, up or down, tends to drift back toward it. That is not bad luck that "has to turn". It is just that the extreme values of small samples do not last.