Researcher: 95% confidence intervals may really cover just 10-25% of the truth

RexDouglass · x · 2026-09-11

Rex Douglass argues on X that confidence intervals implying 95% coverage of a population estimate may in practice deliver only 10-25% coverage, assuming replications are additional draws from the same population.

He adds that the coverage problem isn't symmetric but systematically biased in whatever direction researchers need to publish — which he calls even worse than the raw numbers suggest.

Related event: Researcher Claims 95% Confidence Intervals Cover Only 10-25% in Practice(2 posts)→

Original post →

More from Research

Research channel →