How Large Should a Sample Be for Useful Precision?

Replace a universal minimum with a decision target

There is no single sample count that makes every study reliable. The required size depends on what is estimated, how much uncertainty is tolerable, the variability expected and how observations are collected. A rule such as thirty observations cannot guarantee adequate precision, power or representation. Start with a concrete statement such as estimating one proportion within five percentage points at approximately 95% confidence under simple random sampling. That objective can be translated into a calculation and reviewed for usefulness.

See what a five-point target requires

For a proportion with no credible prior estimate, use p=0.5 as the maximum-variance planning choice. At 95% confidence, the local form uses z=1.96. With a five-point margin, the unrounded formula z²p(1−p)/e² gives 384.16, so the target is 385 complete observations for a large population. This number comes from the selected assumptions. It is not a universal minimum for all questionnaires, experiments, measurement studies or populations.

Compare confidence levels without changing precision

At 99% confidence the form uses z=2.576. Keeping p=0.5 and the same five-point margin gives 663.5776, rounded up to 664. The higher confidence requirement costs additional observations because its critical value is larger. Enter confidence 99 after checking the 95% result to see this tradeoff directly. Increasing confidence does not make biased measurements more accurate; it changes the repeated-sampling coverage target within the selected statistical model.

Decide whether narrower precision changes an action

At 95% confidence and p=0.5, halving the margin from five points to 2.5 points multiplies the unrounded sample by four, from 384.16 to 1536.64, rounded to 1537. Before choosing that expense, consider whether the narrower interval would change a decision or conclusion. If the outcome is rare, absolute and relative precision can have very different meanings. A five-point margin around an expected one-percent proportion may be uninformative even though it sounds narrow on a percentage scale.

Distinguish estimation precision from test power

An estimation calculation asks how precisely a parameter can be measured. A power calculation asks how likely a specified test is to detect a particular effect under an alternative model. Power planning needs an effect size, significance level, desired power and design details such as group allocation. The sample-size form here does not take those inputs. Reusing its proportion count as the required size for a treatment comparison would leave the actual power question unanswered.

Plan the units that provide independent information

One thousand repeated records from ten people do not automatically provide the same information as one thousand independently sampled people. Measurements may be clustered within people, classes, sites or time periods. Correlation can change the effective information and standard error. A design effect can describe variance inflation for a specific design and estimator, but it is not a universal multiplier to guess. Determine the sampling unit and model before counting rows as though each were an independent observation.

Protect subgroup and response objectives

If separate conclusions are required for small subgroups, ensure their sample allocations support those conclusions. An adequate overall count can conceal very small subgroup counts. Also distinguish complete usable observations from invitations or initial enrolments. For example, 385 complete responses at an anticipated 80% usable-response rate requires planning roughly 482 invitations. That workload adjustment does not eliminate systematic nonresponse; it only anticipates a number of completed records under the assumed response rate.

Choose and document a defensible compromise

Report the precision or power objective, parameter, confidence or significance level, planning variability, population assumptions and sampling design. Compare plausible scenarios instead of presenting one unexplained number. If the available budget cannot meet the preferred target, state the precision supported by the attainable design or revise the question. A larger sample often reduces random uncertainty, but careful measurement, representative selection and an appropriate analysis remain necessary for the resulting number to support a useful conclusion.

Related guides

Calculate Sample Size from a Confidence-Interval Target

Turn interval width into a margin, derive sample targets for proportions and means, and explain why a published interval may not reveal a unique count.

Calculate Sample Size for a Population Proportion

Choose confidence, precision and an expected proportion, calculate two sample targets, apply a finite-population correction, and round upward.

Sample Size When the Population Size Is Unknown

Plan a proportion sample without a finite population count, distinguish unknown N from unknown prevalence, and test precision and prior assumptions.