In one sentence A standard deviation says how far apart the individual observations are; a standard error says how uncertain a sample statistic is.
Standard deviation describes individuals; standard error describes an estimate
Both numbers come from the same sample, yet they answer two different questions. A standard deviation (s) belongs to the patients: it measures the typical distance between one individual measurement and the mean, and it carries the unit of the measurement, so a spread of pulse rates is expressed in beats/min. A standard error (SE) belongs to the estimate: it measures the uncertainty, or sampling variability, of a statistic such as the sample mean. It answers the question “if this study were repeated on another sample, how far might the new mean land from this one?” The whole section turns on that distinction. One number describes people; the other describes the precision of a number calculated from people.
Pulse rates 70, 72, 75, 78 and 80 beats/min give s ≈ 4.12 beats/min: patients differ from the mean of 75 by roughly 4 beats/min. The standard error of that mean is only 1.84 beats/min, because it describes the mean, not the patients.
A ward mean is 75 beats/min with s = 4.12 beats/min. Which number would you quote to describe how different the patients are from one another?
Show the answer
The standard deviation, s = 4.12 beats/min, because the question is about the spread of individual observations. The standard error of 1.84 beats/min would describe how precisely the mean itself is estimated, which is a different question.
Read detailed explanations
- Why the standard error shrinks as the square root of n
The rule SE = s ÷ √n is not a convention; it follows from how variances behave when observations are combined. Work on the variance first, because variances add where standard deviations do not. If one independent observation has variance σ², then the sum of n independent observations has variance nσ². The mean divides that sum by n, and dividing a quantity by n divides its variance by n², so the variance of the sample mean is nσ² ÷ n² = σ²/n. Taking the square root returns the standard error, σ ÷ √n. This is the square-root law of sampling, and it has a hard practical consequence: precision improves with the square root of effort. To halve the standard error you must quadruple the sample; to make it ten times smaller you need a hundred times as many observations. Doubling recruitment therefore buys only about 41% more precision, since √2 ≈ 1.41. Reversing the formula answers the question a nurse researcher asks when planning a study — the sample size needed for a target standard error is n = (s ÷ target SE)². If s = 12 mg/dL and the target is 1.5 mg/dL, n = (12 ÷ 1.5)² = 8² = 64 patients. The same law explains the confidence interval usually printed beside a mean, roughly the estimate ± 1.96 × SE: halving that interval costs exactly what halving the standard error costs.
- Describing individuals and describing an estimate: two questions, one sample
Why is the standard error a standard deviation at all? Because the sample mean is itself a random quantity. Imagine repeating the study on a thousand different samples drawn from the same population. Each sample produces its own mean, and those means are not all equal; they scatter. Plot them and you have the sampling distribution of the mean. The standard error is simply the standard deviation of that sampling distribution, which is why it measures wobble rather than patients. Two named results keep this picture honest. First, the sampling distribution of the mean is narrower than the distribution of the observations and becomes more bell-shaped as n grows — the central limit theorem. Second, about 95% of sample means fall within roughly 1.96 standard errors of the true population mean, which is where the familiar confidence interval comes from. The practical warning for nursing is sharp: a ward dashboard reporting mean length of stay as mean ± SE can look impressively stable while individual patients stay anywhere from three days to three weeks. The statistic also depends on what is being estimated. A proportion has its own standard error, and a difference between two means has one too, so “the standard error” is meaningless until the statistic is named. When you read a mean, ask two separate questions: how variable are the patients, and how many patients were there? Only the second of those controls the standard error.
- The coefficient of variation: what the percentage buys, and where it fails
Dividing the standard deviation by the mean makes the spread dimensionless, and that is the whole value of the CV: it survives a change of unit. Medical laboratories lean on this. An assay’s analytical imprecision is routinely quoted as a CV so that a glucose method in mg/dL can be judged against a sodium method in mmol/L, and quality-control charts flag an assay whose CV crosses a locally agreed limit. The same percentage lets a ward compare biological variation between patients with the technical variation of the machine that measures them. The CV also has limits that a beginner must respect, and they all come from the arithmetic of the ratio. First, it needs a ratio scale with a true zero. Temperature in °C fails: a mean of 20 °C with s = 2 °C gives a CV of 10%, but 40 °C is not twice as warm as 20 °C, so the ratio has no physical meaning. Second, a mean near zero makes the CV explode: an s of 2 with a mean of 0.5 gives 400%, and an s of 2 with a mean of 0.01 gives 20,000%. Third, variables that can take negative values or cross zero — a fluid balance, a change score, a temperature difference — have no sensible CV at all. Fourth, for strongly skewed data the mean and the standard deviation describe different aspects of the shape, and their ratio can mislead. Used inside those limits, the CV answers a question the standard deviation cannot: variability as a fraction of the typical size. A rough rule of thumb in laboratory work is that a CV under about 10% signals a comparatively stable measurement, but that threshold is a local convention, not a law of nature.