Descriptive studies and precision-based planning
What question does this help answer?
How common is a condition, what is the typical value of a measurement, or how often does an event occur in a defined population, and how much uncertainty is acceptable in that answer?
When to use it
Use precision-based planning when the primary aim is an estimate and no comparison is being tested: a prevalence, an incidence rate, a mean, the proportion meeting a target, or the characteristics of a newly defined group. The same thinking helps comparative studies too, where the width of the interval around an effect often says more than power does.
When not to use it
Do not present a descriptive study as if it compared groups, and do not attach a power calculation to a hypothesis the study is not testing. If the estimate must be judged against a threshold, for example whether uptake exceeds a target, state that comparison explicitly, because it changes the plan.
How it works in plain language
Decide how wide a confidence interval can be while still being useful for its purpose: a service planner may need a prevalence to within a few percentage points, or only its order of magnitude. The number of participants needed then depends on that width, the confidence level, the expected value of the quantity and, for a measurement, its variability. For a proportion, a value near one half needs more participants than a value near zero or one for the same absolute width. Where Mallard offers a precision calculation, the calculator names its method; this page explains the inputs rather than the arithmetic.
Inputs that change the answer
- The target width, and whether it is absolute (a number of percentage points either side) or relative (a fraction of the estimate itself).
- The confidence level, stated rather than assumed.
- The anticipated value or variability, taken from earlier data, with a plausible range, because a guess at the wrong end changes the required number substantially.
- Clustering or complex sampling, which reduces the effective sample size.
- Expected nonresponse, ineligibility and missing data, which reduce the usable sample and may also bias it.
- For rates, the expected person-time or number of events, not only the number of people.
Plan with the interval the analysis will report
For a single proportion, Wilson or Clopper-Pearson intervals behave well near 0 or 1, where the simple normal-approximation interval can be too narrow or fall outside the possible range. Use the same interval method for planning that the analysis will use for reporting.
How precision grows with sample size
Clinical example
A team wants to know what proportion of adults attending a clinic have uncontrolled blood pressure, to decide whether to fund a nurse-led service. They decide an estimate within about five percentage points either way would inform the decision, take an anticipated proportion and a plausible range from last year's audit, allow for incomplete records, and plan how many charts to review from those inputs. No calculation is shown; the example illustrates the inputs, not a result.
What to watch out for
Planning for precision still needs a defined estimate and a defensible sampling frame: a narrow interval around a biased estimate is a precise wrong answer. The width actually achieved depends on the true value, variability, clustering, nonresponse and missing data, so a planned width is a target under stated assumptions, not a guarantee.
Before using this in your workspace
Record the quantity being estimated, the target population and sampling frame, the target interval width and whether it is absolute or relative, the anticipated value with its source and range, the interval method, and the allowances for clustering and missing data.
- Descriptive studies: what they can and cannot do (opens in a new tab) · Lancet · doi 10.1016/S0140-6736(02)07373-7
- Planning Study Size Based on Precision Rather Than Power (opens in a new tab) · Epidemiology · doi 10.1097/EDE.0000000000000876
- The tyranny of power: is there a better way to calculate sample size? (opens in a new tab) · BMJ · doi 10.1136/bmj.b3985