Sample Size and Statistical Power
Study size should reflect the research goal, desired precision and explicit assumptions rather than a universal numerical rule.
#The question determines what size must achieve
The Field Trials of Health Interventions chapter distinguishes choosing trial size for precision from choosing it for statistical power. The choice depends on the trial's objective. NIH's sample-size resources likewise connect the calculation to the research question, intervention, design and analysis plan.
Read what the study was sized to estimate or detect, and for which outcome. A participant count without its purpose is hard to interpret. These intervention-trial examples do not provide one acceptable sample size for every health-science study.
Evidence: Field Trials of Health Interventions (2015): trial size / NIH: sample-size calculators and design assumptions
#Power is conditional on a specified effect
The field-trials chapter describes power as the probability of obtaining a statistically significant result when a difference of the specified size truly exists. NIH's calculators ask for desired power, a type I error rate, the outcome distribution and design-specific parameter estimates.
Read the assumed effect, variation and testing plan alongside a power statement. Power is not the probability that a hypothesis is true or a promise that a particular study will find significance. This page supplies no calculation for an individual proposed study.
Evidence: Field Trials of Health Interventions (2015): trial size / NIH: sample-size calculators and design assumptions
#Precision is a separate planning goal
The field-trials chapter connects precision to the width of a confidence interval around an effect estimate. It warns that a design with high power to detect an effect can still estimate its size imprecisely. It also separates sampling error, which decreases with study size, from bias, which generally does not.
Read the estimate and its uncertainty after the study, not just the planned power. A larger study does not automatically repair selection or measurement problems. Statistical detectability and a worthwhile clinical effect are different questions.
Evidence: Field Trials of Health Interventions (2015): trial size
#Design assumptions need to be visible
NIH explains that familiar sample-size methods need modifications for group- or cluster-randomized and related designs because observations within a group can be correlated. The necessary changes depend on the design, outcome distribution and analytic plan. It recommends working with a methodologist familiar with these issues.
Read the assumptions rather than treating a calculator output as a certificate. A number produced for one design may not apply to another. Mynd has not designed, powered or approved a clinical study on this page.
Evidence: NIH: sample-size calculators and design assumptions
Source note
The sections above were checked against the linked sources. No clinical review has been performed. This is general research education, not a clinical guideline.