Avoiding the Statistics Traps

## Cherry-Picking (Data Dredging) If you look at enough statistics, you will always find one that supports any claim. "This team wins 80% of home games on Tuesdays in February when it has rained in the previous week" — technically accurate, statistically meaningless. Data dredging produces patterns that are entirely due to chance. The safeguard: form your hypothesis first, then check the data. Do not browse data looking for patterns and then treat what you find as a discovery. ## Survivorship Bias You hear about systems that worked. You do not hear about the 50 systems tried by the same person that failed. You read about the tipster who has had a great six months. You do not see the 200 tipsters who tried and gave up. Survivorship bias inflates apparent performance of everything you can observe, because only the survivors are observable. ## Correlation vs Causation Two statistics that move together do not necessarily influence each other. Classic sports example: teams that score first win 70% of matches. Does scoring first cause winning? Partly — but mostly, better teams are more likely to score first and more likely to win. The true cause is team quality, not the act of scoring first. Betting on "next team to score" at kick-off based on this correlation misidentifies the mechanism. ## Small Sample Percentages "This referee cards 3.2 players per game" — based on 12 games. A 95% confidence interval around that estimate is enormous. The true rate could be anywhere from 1.8 to 4.6 and still be consistent with the observed data. Treat any percentage derived from fewer than 30 events with significant scepticism.