Quick answer
Sample size calculation combines expected effect size, outcome variability, significance level, and statistical power. Correct inputs prevent underpowered studies that miss true effects and oversized studies that waste resources. Protocol-ready planning also includes dropout allowance and sensitivity checks.
Key takeaways
- Power the study for its primary outcome and primary comparison.
- Use a clinically meaningful effect, not simply the largest effect reported previously.
- Justify variability, alpha, power, allocation ratio, and design assumptions.
- Inflate the analyzable sample for attrition, clustering, or unequal allocation only after the core calculation.
Core Inputs You Must Specify
Power analysis requires alpha, desired power, anticipated effect, and dispersion estimate. Each assumption must be justified using pilot data, literature, or clinically meaningful thresholds.
Adjusting for Real-World Attrition
Most studies lose participants over time. Add a realistic inflation factor for dropouts, protocol deviations, and subgroup plans. Failure to adjust creates hidden underpower.
Documenting Assumptions
Include all power assumptions in your protocol and report. Reproducible sample-size reasoning improves ethics review quality and publication acceptance.
Practical method
Step-by-Step Workflow
- 1
Define the primary test
Identify the endpoint, comparison, analysis model, sidedness, and significance level. Different tests require different information.
- 2
Source defensible inputs
Use pilot data, a systematic review, or a clinically important threshold to justify the expected effect and variability.
- 3
Calculate and stress-test
Run the calculation under optimistic, expected, and conservative assumptions so decision-makers can see how sensitive the target is.
- 4
Adjust for the design
Account for dropout, nonresponse, clustering, repeated measures, and planned allocation, then state the final number per group and overall.
Worked example
Two independent groups
- Scenario
- A trial aims to detect a 5-unit mean difference, assumes a standard deviation of 12, uses two-sided alpha 0.05, and targets 80% power.
- Approach
- The normal approximation gives about 91 analyzable participants per group. With 15% expected attrition, divide by 0.85 and recruit approximately 107 per group, or 214 participants in total.
- Interpretation
- The calculation is only as credible as the 5-unit target and 12-unit standard deviation. Show alternative totals if either input is uncertain and retain the software output with the protocol.
Common Mistakes to Avoid
- Using an effect size with no clinical justification
- Powering a study for several competing primary outcomes
- Adding dropout incorrectly
- Using post-hoc power to interpret a completed study
Frequently Asked Questions
Can I use published effect sizes directly?
Use them as a starting point, but evaluate population and outcome differences before adopting them in your own calculation.
Should I add margin for missing data?
Yes. Always adjust calculated sample size for expected attrition or incomplete records.