The reason a small sample produces the biggest effects

A list of the highest-performing schools in a region is dominated by small ones. A campaign follows recommending smaller schools. The list of the lowest-performing schools in the same region is also dominated by small ones, and that list is not published alongside the first.

What everyone sees

Topping a ranking is read as evidence of a superior method. The reasoning is standard: find the best performers, identify what they share, recommend it. Size is treated as one of the shared features rather than as the reason the group appears at the extreme at all.

What is actually happening

The variance of a mean falls with sample size, so small groups produce more extreme averages in both directions from identical underlying processes. Wainer and Zwerling demonstrated exactly this in school performance data, where a foundation-funded programme favouring small schools rested on a pattern that also placed small schools at the bottom. Tversky and Kahneman named the general error belief in the law of small numbers: treating small samples as reliable representations of their population.

Why it stays hidden

The artefact hides because rankings publish one tail. A list of the best is newsworthy and a list of the worst is defamatory, so only half the evidence circulates, and from that half the pattern looks causal. Sample sizes are also rarely printed next to rankings, so the one column that would reveal the mechanism is missing from the table.