Why squared deviations are used
Subtracting the mean from each observation gives deviations that sum to zero. Squaring those deviations prevents positive and negative values from cancelling. Variance averages the squared deviations using the chosen denominator; standard deviation takes the square root to return to the original units. [1]
Population standard deviation = √[Σ(x − μ)² ÷ N]
Sample standard deviation = √[Σ(x − x̄)² ÷ (n − 1)]
The population formula describes the complete set you have defined as the population. The sample formula uses n − 1 when estimating population variance from observations whose mean has also been estimated from that sample.
Worked example: five observations
Use the values 2, 4, 4, 4, and 6. Their sum is 20 and their mean is 4.
| Value | Deviation from 4 | Squared deviation |
|---|---|---|
| 2 | −2 | 4 |
| 4 | 0 | 0 |
| 4 | 0 | 0 |
| 4 | 0 | 0 |
| 6 | 2 | 4 |
| Total | 0 | 8 |
If these five values are the complete population of interest, variance is 8 ÷ 5 = 1.6 and standard deviation is approximately 1.2649.
If they are a sample used to estimate a wider population's variability, sample variance is 8 ÷ 4 = 2 and sample standard deviation is approximately 1.4142. The data did not change; the purpose and denominator did.
Choose the denominator before comparing results
Spreadsheet functions often distinguish sample and population standard deviation. Two tools can both be correct and still disagree because they use different defaults.
A single observation has population standard deviation zero when treated as the entire population, but it does not provide a defined sample variance using the n − 1 formula. An empty dataset has neither a mean nor a standard deviation.
The n − 1 adjustment makes the sample variance unbiased under common independent, identically distributed sampling assumptions. It does not make the square root an exactly unbiased estimator of standard deviation, and it does not repair a biased sampling design.
Interpret the number in its original units
If processing times are measured in minutes, their standard deviation is also in minutes. Variance is in squared minutes, which is usually less intuitive for communication.
A larger standard deviation means more spread relative to the mean under this summary, but “large” requires context. A standard deviation of five seconds could be important in one process and negligible in another.
Standard deviation alone does not reveal skew, clusters, outliers, or time trends. Inspect the data or a suitable plot. Two datasets can share the same mean and standard deviation while having very different shapes.
Standard deviation is not standard error
Standard deviation describes variability among observations. Standard error describes the sampling variability of an estimate, such as a sample mean, under a stated model.
For independent observations from a common distribution, the estimated standard error of the mean is often sample standard deviation ÷ √n. More observations can reduce that uncertainty without making individual observations less variable.
Do not label an error bar “standard deviation” when it is actually a standard error or confidence interval. The visual may look similar while answering a different question.
Watch out for extreme values and missing data
Squaring gives large deviations substantial influence. An outlier can therefore change standard deviation markedly. Investigate its origin rather than deleting it merely because it is inconvenient. A measurement error, a genuine rare event, and a separate population call for different responses.
For skewed or outlier-heavy data, report complementary summaries such as the median and interquartile range. State whether missing observations were excluded and whether any transformations were used.
For a single list of finite numeric values, report the count and mean alongside the standard deviation, and state whether you used the sample or population denominator. Arithmetic alone does not determine whether the observations are representative or independent.
Frequently asked questions
Why do sample and population standard deviations differ?
The population formula divides squared deviations by N. The sample variance formula divides by n−1 when estimating population variability from a sample.
Can standard deviation be negative?
No. It is a square root of a nonnegative variance. A negative result indicates a calculation or numerical problem.
Does a small standard deviation mean my estimate is accurate?
Not necessarily. Low spread does not rule out bias, measurement error, or an unrepresentative sample. Accuracy and variability are different issues.
Sources & calculation notes
Primary references are linked below. Dates, limits, and product terms can change; confirm the applicable details before acting.
Use this guide thoughtfully. Examples illustrate a calculation method, not a guaranteed outcome. The usefulness of any result depends on the definitions, measurements, and assumptions used.