Effect Size, Power & NNT
From statistical significance to clinical meaning — ARR, RRR, NNT, Cohen's d, and power calculations.
Absolute / Relative Risk & NNT
The Hierarchy of Effect Measures
ARR (most honest) → NNT (most actionable) → RR (most used in research) → OR (case-control studies). RRR sounds the largest and is the most frequently presented — always demand ARR alongside it.
Mean difference (most interpretable in clinical units) → Cohen's d (standardised, for cross-study comparisons). A d of 0.5 is medium. Context always matters more than the benchmark.
Effect size, power, and sample size are mathematically linked. Knowing any two determines the third. Sample size calculations before a trial are not bureaucracy — they are the honesty of specifying what your study can and cannot see.
A large trial can find a tiny effect statistically significant. A small trial can miss a clinically important effect. p-value and effect size must always be reported together. One without the other is incomplete.