Sample size in betting

Sample size is how many observations sit behind a statistic. In betting it determines whether a number is information or noise.

Football produces few events per match, so most eye-catching statistics rest on tiny samples. "Won five in a row" is five observations; "scores in 80% of home games" over ten matches means eight events. Neither can distinguish a real effect from chance. The rule of thumb is brutal: shooting and defensive process stabilise within ten to fifteen matches, but goals, results and finishing quality need far more.

Standard error of a proportion ≈ √(p(1-p)/n). Halving the error requires four times the sample.

A tipster shows 60% winners from 50 bets at even money. That looks convincing, but the 95% confidence interval spans roughly 46% to 73% — the honest conclusion is "possibly good, possibly break-even". At 500 bets the same 60% would be near-conclusive.

Nearly every losing strategy started with a small sample that looked like a discovery. Respecting sample size is the cheapest form of scepticism available.

Nigerian tipster channels advertise streaks measured in days. Ask for the number of bets, average odds and full log — the absence of those three numbers is itself the answer.

• Backing a "system" derived from 30 historical matches.
• Dropping a strategy after a 40-bet losing stretch that variance fully explains.
• Slicing data until a subgroup looks profitable, which guarantees a false positive.