Sample Size in Research: A Practical Planning Guide

What sample-size calculators can tell you, which inputs matter, and why design, attrition and analysis plans still need methodological judgement.

Sample size is a design decision, not a single universal number

Researchers often ask for the “correct” sample size before defining the outcome, design or analysis. In reality, the required sample depends on what you want to estimate or detect. A prevalence survey, comparison of two groups, regression model and qualitative interview study all use different reasoning. A calculator is useful only when its inputs correspond to the planned design.

For proportions, precision drives the estimate

When estimating a proportion or prevalence, common inputs include confidence level, margin of error and an expected proportion. Using 50% as the expected proportion is conservative when no better estimate is available because it produces the largest variance. If the total population is small and known, a finite population correction may reduce the required sample.

Power-based calculations need an expected effect

Analytical studies often use statistical power calculations. These require an expected effect size, significance level, desired power and details of the planned statistical test. Effect sizes should ideally come from prior evidence, a pilot study or a clinically or practically meaningful difference—not from choosing a value merely to produce a convenient sample.

Account for non-response and incomplete data

The number you need to analyse is not always the number you need to recruit. Surveys may have non-response; longitudinal studies may have attrition; laboratory or digital data may be incomplete. A recruitment target can therefore include an allowance for expected losses. The assumption should be stated rather than hidden.

Complex designs require more than a basic calculator

Cluster sampling, multilevel data, repeated measures, stratification and complex regression models can change sample requirements. Design effects and the number of observations per parameter may matter. In these cases, a simple web calculator can provide an initial orientation, but the final calculation should be linked to the exact model and sampling plan.

Report how the sample-size decision was made

A methods section should state the approach, assumptions, target sample and any allowance for non-response. This makes the design auditable and prevents sample size from appearing arbitrary. If practical constraints limit recruitment, acknowledge them and discuss how they affect precision and generalisability.