Standard deviation measures spread — specifically, the typical distance of a data point from the mean, expressed in the same units as the original data. This is precisely why it is preferred over variance for interpretation: variance is measured in squared units (dollars squared, centimetres squared), which has no intuitive real-world meaning, while standard deviation converts that back into the original, meaningful unit by taking a square root.
Why sample standard deviation divides by n − 1, not n
Using the sample’s own mean, rather than the true population mean, to calculate deviations systematically underestimates the real spread — because by definition, the sample mean is the value that minimises the sum of squared deviations for that particular sample, meaning it sits slightly closer to your specific data points than the true population mean would. Dividing by n − 1 instead of n is a mathematically derived correction for this bias, known as Bessel’s correction, and it makes the sample variance an unbiased estimator of the true population variance. Use n only when your data genuinely constitutes the entire population you care about — every student in one specific class, every transaction in a closed dataset — rather than a sample drawn from some larger group you are trying to draw conclusions about.
Reading a standard deviation
For data that follows a roughly normal (bell-curve) distribution, standard deviation has a precise, useful interpretation known as the empirical rule: about 68% of values fall within one standard deviation of the mean, about 95% fall within two, and about 99.7% fall within three. This is what makes standard deviation intuitively meaningful rather than an abstract number — a value two standard deviations from the mean is genuinely rare in a normal distribution. The rule fails badly for skewed distributions, though — income is the classic example, where a small number of very high earners stretch the distribution so far that the mean sits well above where most people actually earn, and the median becomes the far more honest summary of a typical value.
Sensitivity to outliers
Because deviations from the mean are squared before being averaged, a single extreme value can dominate the entire result — a squared deviation grows much faster than the deviation itself, so one data point ten times further from the mean than the rest contributes a hundred times more to the sum of squares. Median and interquartile range are the standard robust alternatives when a dataset contains genuine outliers that cannot be justified as errors and removed — they describe the centre and spread of a distribution without being skewed by a small number of extreme values the way mean and standard deviation are.
Coefficient of variation — comparing spread across different scales
Standard deviation alone cannot tell you whether one dataset is more variable than another if the two are measured on different scales or have very different means — a standard deviation of 5 is enormous for data centred around 10, but trivial for data centred around 10,000. The coefficient of variation, calculated as standard deviation divided by the mean, expresses spread as a proportion of the average value, which makes comparing variability across genuinely different datasets — different currencies, different units, different orders of magnitude — meaningful in a way raw standard deviation cannot be.
Population versus sample — the decision that changes your answer
This distinction is not a minor technicality; the two formulas can give noticeably different results, especially for small datasets, where the gap between dividing by n and dividing by n − 1 is proportionally larger. Six data points divided by 6 versus divided by 5 produces a real difference in the final number. Before reading off a result, it is worth explicitly deciding: is this dataset the entire group I care about, or is it a sample I am using to estimate something about a larger group I have not fully measured? That single question determines which formula is the statistically correct one to use.