Skip to content
Scintria
ToolsGrant writerFunding
HomeLiteratureClinical trialsEvidence LabProtocol packData analysisLearnProjectsLibraryReferencesApplications
Start free
ScintriaStart free

Learn / Biostatistics / Sample size and power

Why sample size matters: the four ingredients of power

12 minute read · 6-question quiz

Download this lesson as a colour Word or PDF study guide with the figures and answers. Included in paid plans; reading and quizzes stay free.Sign in

What you will learn

  • Explain why an underpowered study is both a scientific and an ethical problem

  • Name the ingredients of a sample size calculation and how each moves the answer

  • Choose a defensible effect size (the minimal clinically important difference)

A trial that is too small cannot detect a real benefit, and a trial that is too large exposes more patients than necessary to an inferior treatment and wastes money. Sample size calculation is the discipline of getting between them. Ethics committees, funders and journals expect it, and a flawed calculation is among the commonest reasons for protocol rejection.

Power is the probability that the study will detect (reject H0 for) an effect of a stated size, if that effect truly exists. Conventional targets are 80% or 90%.

The ingredients

For comparing two groups, every sample size formula combines the same four elements.

1. Significance level (α)

The false-positive risk you accept, usually 0.05 two-sided. Smaller α needs a larger sample.

2. Power (1 − β)

The chance of detecting the effect if real. 80% is the minimum for most trials; 90% for important or hard-to-repeat trials.

3. Effect size

The smallest difference worth detecting (the minimal clinically important difference). Halve the effect size and you need about four times the sample.

4. Variability

SD for continuous outcomes; the control-group proportion for binary outcomes; the event rate for survival. More noise needs more participants.

The most important number is the one you choose

Sample size is not 'found'; it is calculated from an assumption about the effect. If your pilot suggests a 10 mmHg difference but the smallest clinically meaningful difference is 5 mmHg, plan for 5. Using the optimistic pilot estimate underpowers the study for the effect that really matters, because small pilots overestimate effects. Justify the effect from clinical meaning, prior trials or meta-analyses.

How power rises with sample size

Consider a trial comparing mean SBP between two arms, with SD 15 mmHg and a true difference of 5 mmHg, at α = 0.05 (two-sided).

0255075100205080100142190250300Participants per armPower (%)Power (%)
Power climbs steeply at first and then flattens: 80% needs 142 per arm, 90% needs 190, and 300 per arm buys only another 8 points.
025507510034567True difference (mmHg)Power (%)Power (%) with 142 per arm
With 142 per arm, a true effect of 3 mmHg would be detected only 39% of the time: the study is powered for 5, not for smaller effects.

Why underpowered studies hurt

  • They miss real effects (type II error), so useful treatments are abandoned.
  • The significant results they do produce tend to overestimate the effect (the 'winner's curse').
  • They waste participants' contribution, since an inconclusive study answers nothing.
  • Pooling many small studies can rescue some information (meta-analysis), but the original waste is permanent.

Why overpowered studies hurt too

  • More people than needed receive the inferior arm.
  • Trivial, clinically meaningless differences become 'significant'.
  • Resources are consumed that could answer another question.

The formula's family tree

1Define the primary outcomecontinuous, binary, time-to-event, count2Choose the comparisonsuperiority, equivalence, non-inferiority3Set α, power and effect sizejustify from clinical meaning4Estimate variabilityfrom earlier studies or a pilot5Calculate, then inflatefor attrition, clusters, interim analyses
The steps of a sample size calculation for a protocol.

Where to get the SD or control-group rate

Published trials in a similar population, national survey data, routine facility records, or a pilot. A pilot's estimate of the SD is useful, though not its effect estimate. Use the upper end of a plausible range to stay safe.

'Post-hoc power' is not a thing to report

Calculating power after the study from the observed effect only restates the p-value. If a result is non-significant, report the confidence interval instead and discuss what effects it excludes.

Summary

  • Power is the chance of detecting a real effect of a stated size; 80% to 90% is the norm.

  • Four ingredients: α, power, effect size and variability.

  • Halving the effect size roughly quadruples the sample size.

  • Choose the effect from clinical importance, not from an optimistic pilot.

  • Both underpowered and overpowered trials are ethically problematic.

Check your understanding

Question 1
1

Halving the detectable difference changes the required sample size by about:

Question 2
2

Which change would INCREASE the required sample size?

Question 3
3

An effect size taken from a small, positive pilot is risky because:

Question 4
4

Power of 80% means:

Question 5
5

A study is powered at 80% to detect 5 mmHg. If the true difference is 3 mmHg, power will be:

Question 6
6

Post-hoc power calculated from the observed result:

Draft study guide. Spotted an error or something unclear? Tell us at hello@scintria.com. This is educational material, not clinical or regulatory advice.

The Cox proportional hazards modelCalculating sample size for means, proportions and events