How this instrument works
The standard error (SE) of a sample mean measures how much that mean would be expected to vary if you repeated the sampling process again and again — it shrinks as your sample gets larger, following SE = s/√n, where s is the sample standard deviation and n is the sample size. On its own, though, an SE of 3 doesn't tell you whether the estimate is precise or sloppy, because that depends entirely on the scale of what you're measuring.
Relative standard error fixes that by expressing the standard error as a percentage of the sample mean: %RSE = (SE / mean) × 100. An SE of 3 around a mean of 100 is a tight, precise 3% RSE; the same SE of 3 around a mean of 6 would be a much sloppier 50% RSE. This is exactly the statistic government statistical agencies — the U.S. Census Bureau, the Bureau of Labor Statistics, and others — publish alongside survey estimates to flag how much sampling error affects a given number.
As a rule of thumb used by several federal statistical agencies, estimates with an RSE above roughly 30% are often flagged as unreliable or suppressed entirely, because the sampling noise is too large relative to the estimate itself to draw confident conclusions. Lower RSE values indicate a more statistically reliable estimate, all else being equal.
- Enter the sample's spread into Sample standard deviation (s).
- Enter how many observations are in the sample into Sample size (n) — must be at least 1.
- Enter the sample's average into Sample mean — this cannot be zero, since RSE divides by it.
- Read Standard error (SE) first — this is s divided by the square root of n.
- Read Relative standard error (%RSE) — this expresses that standard error as a percentage of the mean, the figure typically used to judge an estimate's reliability.
Worked example — a sample mean of 100, s=15, n=25
Enter 15 into Sample standard deviation (s), 25 into Sample size (n), and 100 into Sample mean. The instrument first computes the standard error: SE = 15/√25 = 15/5 = 3.0000 exactly.
Relative standard error (%RSE) then reads 3.0000%, computed as (3/100) × 100. A 3% RSE is generally considered a tight, reliable estimate — well under the roughly 30% threshold several statistical agencies use as a rough cutoff for flagging an estimate as too imprecise to report with confidence.
Questions
What's the difference between standard error and relative standard error?
Standard error (SE) is in the same units as your original data — dollars, people, degrees, whatever you measured — so a raw SE value only means something once you know the scale of the mean it's attached to. Relative standard error (%RSE) divides SE by the mean and expresses it as a percentage, which strips away the units and lets you judge precision directly, or compare precision across two estimates measured in completely different units.
What counts as a 'good' RSE?
There's no single universal cutoff, but several U.S. federal statistical agencies use roughly 30% as a rule-of-thumb ceiling — estimates with RSE above that are often flagged as unreliable or suppressed in published tables. RSE well under 10% is generally considered quite precise, while RSE in the 10-30% range is usable but should be interpreted with some caution.
Why does RSE use the standard error instead of the raw standard deviation?
Because RSE is meant to describe the precision of an estimate — typically a sample mean — not the spread of the raw data itself. The standard deviation describes how spread out individual observations are; the standard error describes how much the sample mean itself would bounce around if you repeated the sampling, and it shrinks as sample size grows. RSE inherits that sample-size sensitivity, which is exactly what makes it useful for judging how trustworthy a given survey or study's estimate is.
Why can't the sample mean be zero?
Because %RSE divides the standard error by the mean, and division by zero is undefined. In practice, RSE is rarely reported for estimates centered at or near zero for exactly this reason — the percentage becomes either undefined or wildly unstable, so other precision measures (like the raw standard error or a confidence interval) are used instead in that situation.
Does a bigger sample always give a smaller RSE?
All else being equal, yes — the standard error shrinks as sample size n grows (it's divided by √n), which pulls RSE down along with it, assuming the underlying variability and mean stay roughly the same. That's the core reason larger surveys and studies are generally considered more precise: more observations narrow the standard error, and therefore the relative standard error, around the true estimate.