Is a 10% RSE a 10% confidence interval?

No; the interval also depends on a confidence level, reference distribution and suitable approximation.

Does a low RSE prove the survey is unbiased?

No; sampling precision does not establish freedom from coverage, measurement or nonresponse bias.

Can RSEs be averaged across categories?

A simple average usually does not give the RSE of a combined estimate; the combined variance must account for the estimator and relationships among components.

AI Glossary / Statistics and probability / Foundation

Relative standard error

Definition

Relative standard error, usually abbreviated RSE, expresses an estimate's standard error relative to the magnitude of that estimate, commonly as a percentage. For a positive estimate, RSE equals 100 multiplied by the standard error divided by the estimate. It provides a scale-adjusted description of sampling precision. It is not the percentage by which the estimate is known to be wrong, and it does not normally capture every source of uncertainty or bias in the data.

What it means

A standard error describes how an estimator would vary across repeated samples under the relevant sampling design or model. Dividing it by the estimate makes its size easier to interpret across positive estimates with different magnitudes. A standard error of 300 may be modest beside an estimate of 6,000 but large beside an estimate of 1,000. The RSE communicates that contrast. Its usefulness depends on a meaningful denominator and a standard error calculated for the actual estimator and design.

How it works

Obtain the estimate and its associated standard error, ensuring that both use the same units. Divide the standard error by the positive estimate and multiply by 100. For complex surveys, use an appropriate design-based variance estimate that reflects weighting, clustering and stratification. If the estimate is zero or near zero, the ratio is undefined or unstable. For quantities that can be negative, conventions need explicit explanation; an absolute denominator may be used, but relative precision can still be uninformative.

Three practical examples

Example 1: A survey publisher adds RSEs to regional employment totals so readers can identify estimates with substantial relative sampling variability. Example 2: A market analyst compares the precision of positive expenditure estimates across customer groups. Example 3: An AI evaluation team reports an estimated event proportion with its standard error and RSE, while retaining the underlying counts. In each case, the same RSE can accompany very different absolute uncertainty, so the estimate and standard error remain useful.

Think of it this way

Think of comparing the size of a measuring wobble with the size of the quantity being estimated. A wobble of three units is relatively small beside 60 and relatively large beside ten. This analogy concerns variability in an estimate, not proof of its distance from the truth. A consistently biased measuring process can have a small wobble and still miss the target systematically.

Worked example

Suppose a survey estimates 6,000 eligible businesses in a region, with a design-appropriate standard error of 300. The RSE is 100 times 300/6,000 = 5%. Another region has an estimated 1,000 businesses with the same standard error of 300, giving an RSE of 30%. Their absolute standard errors match, but their relative precision differs. If a normal approximation is suitable, an approximate 95% interval for the first estimate is 6,000 plus or minus 1.96 times 300, or 5,412 to 6,588. Separately, a proportion estimated as 0.20 with standard error 0.04 has RSE 20%; its standard error is four percentage points, not 20 percentage points. Expressing the first total in tens gives estimate 600 and standard error 30, leaving its RSE unchanged at 5%.

Illustrative case study

In this fictional case study, Blue Harbor Insights prepares a market-size dashboard for an AI planning service. Small-area estimates initially appear equally authoritative because all are displayed to the nearest whole business. The analyst adds standard errors, RSEs and explanatory notes, then reduces unnecessary numerical precision. A region with a large RSE is not automatically removed; its suitability is assessed against the decision and the survey publisher's guidance. Managers also review coverage and nonresponse because a small RSE would not rule out those problems.

Common misconceptions and mistakes

Common mistakes include calling the RSE the actual percentage error, confusing it with a margin of error, and treating a publisher's caution threshold as a universal statistical law. Another mistake is applying a simple random-sample formula to a complex weighted survey. RSE is different from the coefficient of variation of individual observations: its numerator is the standard error of an estimate. For differences close to zero, a large RSE may be mathematically inevitable and provide little useful perspective.

Business use

Businesses use RSEs to communicate which estimated totals, means or proportions support more precise planning. A reporting policy can combine RSE annotations with absolute uncertainty, sample size and the consequences of an incorrect decision. Comparisons between estimates should use the standard error of their difference, including covariance where relevant, rather than simply comparing their RSEs. AI-generated summaries should preserve reliability notes instead of turning an uncertain estimate into a confident statement about a market or population.

Check your understanding

Which worked-example region has greater relative sampling variability?

Answer: The region estimated at 1,000, because its RSE is 30% rather than 5%.

What is the standard error in percentage points when the estimated proportion is 20% and its RSE is 20%?

Answer: Four percentage points.

Why is an RSE problematic at an estimate of zero?

Answer: The usual calculation divides by the estimate, so the ratio is undefined.

Learn next

Learn next how a survey design affects variance estimation and how confidence intervals differ from point estimates. Practice converting between proportions and percentages without confusing percentages with percentage points. Review the reliability guidance attached to the particular statistical release, then consider whether absolute uncertainty provides a clearer explanation when estimates approach zero.

Variance estimation

Sources

Sources: Australian Bureau of Statistics, Measures of error, https://www.abs.gov.au/statistics/understanding-statistics/statistical-terms-and-concepts/measures-error; ABS, Confidentiality and relative standard error, https://www.abs.gov.au/statistics/microdata-tablebuilder/tablebuilder/confidentiality-and-relative-standard-error. The examples are original calculations; any release-specific reliability rules must be checked in the relevant publication rather than generalised to all datasets.

Previous: Relative risk or risk ratio Next: Relaxation labeling