Why divide by n − 1?

The sample mean is estimated from the same data. Under the standard independent-sampling assumptions, the correction makes the resulting sample variance unbiased for population variance.

Can the sample standard deviation be negative?

No. It is a square root of a nonnegative quantity. A negative value would indicate an error or a different quantity.

Does a larger sample always have a smaller standard deviation?

No. The sample standard deviation estimates the spread of individual values. Increasing sample size does not necessarily reduce that underlying spread.

AI Glossary / Data science and analytics / Intermediate

Sample Standard Deviation

Definition

The sample standard deviation measures the spread of observed numerical values around their sample mean. In its common statistical form, it is the square root of the sum of squared deviations from the mean divided by n − 1, where n is the sample size.

What it means

A small standard deviation means values are relatively close to their mean; a larger one indicates greater spread on the same measurement scale. It is expressed in the original units, unlike variance, which uses squared units.

The statistic describes variability among observations. It is not the same as the standard error of an estimated mean, which concerns uncertainty in that estimate across repeated samples. More observations can improve precision about a mean without making the underlying individual values less variable.

The denominator matters. The n − 1 version is commonly used when a sample informs an unknown population variance. Some software also offers a denominator-n calculation for a different descriptive convention. Reports and code should state which version is intended.

How it works

Calculate the sample mean, subtract it from every observation, square each deviation and add the squared deviations. Divide by n − 1 and take the square root. The formula requires at least two observations for this ordinary sample calculation.

Squaring prevents positive and negative deviations from canceling and gives larger deviations more influence. Consequently, extreme observations can substantially affect the result. Examine unusual values and the distribution rather than treating the standard deviation as a complete description.

Under independent, identically distributed sampling with finite variance, dividing by n − 1 makes the sample variance unbiased for the population variance. Taking the square root does not generally make the sample standard deviation itself unbiased for the population standard deviation.

Weights, grouped observations and dependence require appropriate methods beyond blindly applying the simple formula. If software omits missing values, the effective sample size and missing-data assumptions should also be clear.

Three practical examples

A service analyst reports variation in completion times alongside the mean and a plot of long-running cases.
A manufacturing team uses repeated measurements to summarize observed dimensional variation, after checking the measurement process.
An AI evaluator compares the spread of per-case errors across a defined test set, keeping that spread separate from uncertainty in the average error.

Think of it this way

Think of people standing around a central meeting point. The standard deviation summarizes how spread out their measured positions are, with distant positions contributing strongly.

The analogy has limits: the actual formula uses squared numerical deviations and a specified denominator. It is not simply the average walking distance from the center, and it does not describe every feature of how the group is arranged.

Worked example

Take invented completion times of 2, 4 and 6 minutes. The sample mean is (2 + 4 + 6)/3 = 4 minutes. Deviations from the mean are -2, 0 and 2 minutes; their squares are 4, 0 and 4 square minutes.

The squared-deviation sum is 8. Dividing by n − 1 = 2 gives a sample variance of 4 square minutes. Taking the square root gives a sample standard deviation of 2 minutes.

If the same three values are treated using a denominator-n descriptive calculation, the variance is 8/3 and the standard deviation is approximately 1.63 minutes. The difference reflects the convention, not different observations.

Under the additional assumptions appropriate to the simple independent-sample formula, the estimated standard error of the mean is s divided by the square root of n: 2 divided by the square root of 3, approximately 1.15 minutes. That is a different quantity answering a different question.

Illustrative case study

Illustrative fictional case: Harbour Support reports an average resolution time with “plus or minus” one standard deviation. An AI summary describes that interval as a 95% confidence interval for the mean.

The analyst corrects the interpretation. Standard deviation describes variation among resolution times, while a confidence interval needs an appropriate uncertainty method and assumptions. The team labels each quantity explicitly and shows the distribution because a few long cases make a symmetric summary incomplete.

It also checks whether repeated cases from the same customer affect the uncertainty calculation. This fictional case illustrates reporting distinctions; it does not claim that one confidence-interval method is suitable for every service dataset.

Common misconceptions and mistakes

Do not confuse standard deviation with variance, range, mean absolute deviation or standard error. Do not say that exactly 68% of any dataset must lie within one standard deviation of its mean; familiar percentage rules depend on distributional assumptions.

Avoid assuming the n − 1 correction makes the standard deviation itself unbiased. Do not compare numerical standard deviations across different units without accounting for scale. A zero value means all included observations are equal, but it does not prove that the measurement system captured every relevant source of variation.

Business use

Use sample standard deviation when variation around the mean is relevant and the measurement scale supports the calculation. Report it with the mean, units, sample size and an indication of the distribution when skewness or extremes matter.

Check the software denominator and missing-value treatment. For AI reports, state whether the number describes variability of individual outcomes, variability between repeated model runs or uncertainty in an estimate. These distinct quantities should not share one interchangeable label.

Check your understanding

A report's standard deviation is 12 minutes and its sample size increases from 100 to 400 with a similar distribution. Must individual cases become four times less variable?

Answer: No. Their spread can remain near 12 minutes. Under suitable independent-sampling assumptions, the standard error of the mean may decrease, but that is not the same as reduced individual variability.

Related terms

Learn next

Read sample variance and standard error. Calculate both quantities for a small dataset and keep their units and meanings separate. Then convert every observation to a different unit and explain how the mean, variance and standard deviation change.

Sources

University of Chicago, sample variance and standard-deviation definitions: https://ds1.datascience.uchicago.edu/glossary.html

NIST, measures of scale and the sample standard-deviation formula: https://itl.nist.gov/div898/handbook/eda/section3/eda356.htm

NIST, variance inference under specified assumptions: https://www.itl.nist.gov/div898/handbook/eda/section3/eda358.htm

The three-value calculation and support-report case are original fictional illustrations.

Previous: Sample space Next: Sample statistics