Standard Deviation

Standard Deviation: Concept Architecture

Orientation and learning roadmap

This page moves from individual deviations around a mean to the sample and population calculations. It then distinguishes variability among observations from uncertainty about an estimate and shows how that distinction matters in health economic evidence.

What is being measured

Specify the variable, its units, the population or sample, and the observation period before describing its spread. The standard deviation is expressed in the original measurement units because it is the square root of a variance, whose units are squared.

A small standard deviation indicates that observations tend to cluster more closely around their mean than observations with a larger standard deviation on the same scale. It does not describe every aspect of a distribution: groups with equal means and standard deviations can differ in skewness, tails, or distinct subgroups. Comparing standard deviations across unlike scales without further context can be misleading.

Population and sample calculation

The population formula averages squared deviations over every member of the defined population, whereas the usual sample formula uses a denominator of one less than the sample size. This adjustment makes the sample variance an unbiased estimator of population variance under independent, identically distributed sampling with finite variance, but the square root of that sample variance is not generally an exactly unbiased estimator of population standard deviation.

For population values $x_1,\ldots,x_N$ with mean $\mu$, and sample values $x_1,\ldots,x_n$ with mean $\bar{x}$, the corresponding formulas are:

$$\sigma=\sqrt{\frac{1}{N}\sum_{i=1}^{N}(x_i-\mu)^2},\qquad s=\sqrt{\frac{1}{n-1}\sum_{i=1}^{n}(x_i-\bar{x})^2},\quad n>1.$$

If the observed sample is 2, 4, and 6, its mean is 4 and the squared deviations total 8. Thus its sample standard deviation is $s=\sqrt{8/2}=2$ in the outcome's units; if those same three values comprise the entire specified population, the population standard deviation is $\sqrt{8/3}\approx1.63$. The two answers differ because the denominator and inferential purpose differ.

Distinguishing SD from uncertainty in the mean

An SD of patient costs describes between-patient cost variability in the sample; it is not the sampling uncertainty of the estimated mean cost. For independent, identically distributed observations and the usual simple-random-sample setting, the estimated standard error of the sample mean is $SE(\bar{x})=s/\sqrt n$.

For example, if 100 patients' annual costs have sample SD £1,000, the estimated standard error of the mean is £100 under those assumptions. Correlation, clustering, weighting, missing data, or complex sampling can invalidate that simple formula and require a design-aware standard error. The variability of a difference between groups also depends on both groups' data and their dependence structure.

Applications in health economics

Studies often report mean and SD for continuous outcomes such as cost, length of stay, utility, and quality-adjusted life-years. The SD helps describe heterogeneity and contributes to precision calculations, evidence synthesis, and planning a study with a specified detectable difference.

In probabilistic economic models, distinguish an individual-outcome SD from uncertainty in an estimated parameter supplied to the model. Assigning the raw between-patient SD to a distribution intended to represent uncertainty about the mean can substantially overstate parameter uncertainty; the relevant sampling distribution, standard error, covariance, and probability-family constraints need consideration. An SD alone also does not specify a full distribution or capture uncertainty in model structure.

Interpretation checks and limitations

The mean and SD can be useful summaries for symmetric distributions, but highly skewed costs or a mixture of patient groups can make the mean-plus-or-minus-SD range a poor description of typical values. Inspect quantiles and plots, and consider relevant subgroups rather than treating a single pooled SD as a complete portrait of patient experience.

State which SD is reported, the sample size, units, exclusions, and handling of missing observations. When comparing studies or incorporating variability into a model, confirm that the outcome definition, time horizon, and population align.

Evidence base

The population-versus-sample calculation and units are explained in the NIST measures-of-scale handbook. The Cochrane Handbook on continuous outcomes distinguishes SDs from standard errors used in evidence synthesis, while the NICE economic evaluation manual addresses uncertainty in health economic model parameters.