- Save calculation inputs, scenarios and the design rationale.
Start with the primary analysis
Estimating a proportion, comparing means, fitting regression, cluster sampling and survival analysis have different requirements. A formula for a simple survey is not automatically appropriate for a complex model or a qualitative study. Specify the main outcome and design first.
Choose defensible assumptions
A power calculation needs an appropriate test, target effect, type-I error level, power, sidedness and allocation assumptions. Justify the effect using prior evidence, a meaningful threshold or a scientific scenario. Choosing it only to obtain a cheaper sample is not a scientific rationale.
Consider precision and design
The goal may be a sufficiently precise interval rather than a significance threshold. Repeated measurements, clustering and imbalanced data can require specialist methods or simulation. UCLA’s G*Power examples address particular tests; verify the match to your design independently.
Plan attrition and sensitivity
For n analyzable observations and anticipated attrition r, n/(1−r) is a simple recruitment approximation under explicit assumptions; round upward. Compare plausible effect and attrition scenarios. After observing results, do not substitute observed power for interval estimates or the original design justification.
Report enough to reproduce the calculation
Preserve software and version, test, every input, supporting assumptions and scenario results. Explain for which outcome and assumptions a proposed sample is adequate rather than declaring a universal number. Final study planning requires review of the actual design and resources.
Practical research checklist
- Match the test to the research design.
- Justify effect and precision.
- Account for clustering and attrition.
- Preserve inputs and sensitivity scenarios.
Worked case and implementation decisions
The following is a fictional teaching case. Do not use its numbers or wording as actual study findings.
Suppose an independently justified plan needs 100 analyzable observations and expects 15% attrition. The simple recruitment approximation 100/0.85 rounds up to 118; at 25% attrition it becomes 134. This arithmetic does not justify the initial 100 or account for clustering. Use scenarios to expose risk. If classes or repeated measures are involved, participants are not automatically independent observations. Preserve every calculation input and the effect rationale, not only the software’s final number.
| Stage | Teaching example | Verification question |
|---|---|---|
| Goal | Power or precision for the main outcome | Does it match the design? |
| Analyzable need | 100 is an exercise assumption | Was the actual need properly calculated? |
| 15% attrition | Approximate recruitment 118 | Is the rate justified? |
| 25% attrition | Approximate recruitment 134 | Is the harder scenario covered? |
| Structure | Possible clusters or repetition | Are independence assumptions defensible? |
Exercise output: Save calculation inputs, scenarios and the design rationale.
Sources and further reading
Official sources for verification and further reading