Where sizing inputs come from
Sizing combines clinical choices with evidence about your data; record the source and uncertainty of each input.
Fit
When it matters, and when it does not
It matters when
- You need a target difference, event risk, spread, correlation or loss to follow-up assumption.
- A published calculation uses a different population, outcome definition or follow-up period.
It matters less when
- The sample is already fixed. Assess attainable precision or a meaningful detectable effect instead of inventing a recruitment target. Read: Planning when the sample size is already fixed
- The design needs bespoke simulation. Write its full specification rather than borrowing an independent-groups formula. Read: Simulation-based power and sample-size planning
Before you start
What to check in your study
- The estimate and planning goalsummaryMeasureframingChoose a clinically worthwhile contrast or an acceptable interval width. State the error rate and target power when testing an effect.
- Evidence for each numerical assumptionUse comparable studies or local records. Label an unsupported value as assumed and explain how you will check it.
- Dependence and missing outcomesMatch the correlation to the design. For paired means, use the spread of within-person differences, not the spread of single measurements.
- The source population and measurements resemble your planned study.How to check: Compare eligibility, care, measurement scale and horizon. Examine a plausible range rather than trusting one estimate.
- A pilot estimate is uncertain.How to check: Small pilots give noisy estimates of spread and event risk. Plan sensitivity to that uncertainty.
Plain language
How it works
Clinical importance sets the target contrast; data inform nuisance inputs such as spread and baseline risk. Neither is whatever makes recruitment affordable. A prior study needs the same outcome definition and a comparable horizon.
The precision diagram holds the event proportion near one half and assumes independent observations. Other proportions, clustering and nonresponse change the width. Mallard checks required calculator inputs, but passing those checks does not establish where the values came from.
Worked example
A worked example
- Setting
- A team wants 80 complete outcome records.
- Question
- How many people should it approach under a simple loss assumption?
- Approach
- Assume 80 in 100 enrolled people provide the outcome. Divide the needed 80 by 0.8.
- Result
- Enrolling 100 people gives an expected 80 complete records, or 80 in 100. This is a planning expectation.
- What it means
- The count adjusts for expected loss only. It does not repair bias if missing outcomes depend on unseen values.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A precise sample size can rest on uncertain inputs. Do not choose a larger effect, smaller spread or lower dropout just to fit the available sample.
- Copying the observed treatment effect from a small pilot.Instead: Justify the target difference clinically. Use pilot data cautiously for spread, recruitment and other uncertain inputs.
- Treating a typed number as an established fact.Instead: Keep its source, population and uncertainty beside it. Recalculate over plausible values.
Inside Mallard
How Mallard uses this
Mallard asks for inputs needed by the selected calculation. Missing required values can prevent a headline sample size.
concept.sizing_inputs, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- The size of a pilot study for a clinical trial should be calculated in relation to considerations of precision and efficiency. (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2011.07.011
- Reporting of sample size calculation in randomised controlled trials: review. (opens in a new tab) · BMJ · doi 10.1136/bmj.b1732