Standard deviation is the square root of the variance: a measure of how far observations typically lie from their mean, expressed in the same units as the data. In statistics, it is the most widely used summary of spread or dispersion, and it appears in data analysis, quality control, experimentation and risk measurement wherever a single average would hide how much individual values differ.
The symbol σ (sigma) denotes the standard deviation of a whole population, while s denotes the estimate computed from a sample. Together with the mean, it describes the center and width of a distribution, and under a roughly normal distribution it allows quick statements about how unusual a given value is.
How standard deviation is calculated
Standard deviation is calculated by averaging the squared distances from the mean and then taking the square root of that average. Each observation contributes its deviation (value minus mean); squaring removes the sign so that values above and below the mean do not cancel, and it weights large deviations more heavily than small ones.
- Compute the mean of the values.
- Subtract the mean from each value to obtain the deviations.
- Square each deviation and add the squares together.
- Divide the sum by n for a population or by n – 1 for a sample.
- Take the square root of the result to return to the original units.
The divisor n – 1 is known as Bessel’s correction. A sample mean sits closer to its own data than the true population mean does, so dividing by n would systematically underestimate the spread. The correction makes the sample variance an unbiased estimator, although the sample standard deviation itself remains slightly biased downward for very small samples.
The following table summarizes the two common forms and when each one applies.
| Form | Symbol | Divisor | When to use it |
|---|---|---|---|
| Population standard deviation | σ | n | Every member of the group is measured |
| Sample standard deviation | s | n – 1 | A subset is used to estimate a larger group |
| Standard error of the mean | s / √n | Not applicable | Describing the precision of an estimated mean |
In software, the default form differs between tools, so analysts should check which divisor a function uses. Spreadsheet functions usually separate the two forms explicitly, while some numerical libraries default to the population form unless a degrees-of-freedom parameter is set.
Interpreting standard deviation in a distribution
A standard deviation is interpreted relative to the mean and the shape of the distribution. A small value means observations cluster tightly around the mean; a large value means they are widely scattered. On its own, the number has no fixed threshold for small or large, because it carries the units and scale of the data.
For data that are approximately normally distributed, the empirical rule links the standard deviation to coverage. The table below shows the share of observations expected within a given distance from the mean.
| Distance from the mean | Share of values inside | Share outside |
|---|---|---|
| ±1 standard deviation | About 68.3% | About 31.7% |
| ±2 standard deviations | About 95.4% | About 4.6% |
| ±3 standard deviations | About 99.7% | About 0.3% |
The distance of a single value from the mean, divided by the standard deviation, is its z-score. A z-score of 2.5 means the value lies two and a half standard deviations above the mean. For distributions that are not normal, Chebyshev’s inequality gives a weaker but universal guarantee: at least 75% of values fall within two standard deviations of the mean, whatever the shape.
How standard deviation compares with related measures
Standard deviation is one of several measures of dispersion, and its main advantage is that it is expressed in the original units while remaining mathematically convenient. Variance carries the same information in squared units, which suits algebra and modeling but is harder to read directly.
The table below places standard deviation among its neighbors in statistics.
| Measure | Units | Sensitivity to outliers | Typical use |
|---|---|---|---|
| Standard deviation | Same as data | High | Describing spread of roughly symmetric data |
| Variance | Squared units | High | Modeling, adding independent sources of variation |
| Interquartile range | Same as data | Low | Skewed data, paired with the median |
| Median absolute deviation | Same as data | Very low | Robust spread estimates with outliers |
| Coefficient of variation | Unitless ratio | High | Comparing spread across different scales |
A practical consequence is that variances add, standard deviations do not. For independent quantities, the variance of a sum equals the sum of the variances, so the combined standard deviation is the square root of the summed variances rather than the sum of the individual standard deviations.
Choosing and reporting standard deviation in practice
Standard deviation is the right choice when the data are roughly symmetric and the mean is a meaningful center. When the distribution is skewed, such as income, response times or order values, analysts usually report the median with percentiles or the interquartile range instead, because a few extreme values can inflate the standard deviation dramatically.
| Situation | Recommended measure | Reason |
|---|---|---|
| Symmetric, bell-shaped data | Mean and standard deviation | Compact and well understood |
| Heavily skewed data | Median and percentiles | Not dominated by the long tail |
| Data with known outliers | Median absolute deviation | Resistant to extreme values |
| Groups on different scales | Coefficient of variation | Removes the effect of units |
| Precision of an estimate | Standard error | Shrinks as sample size grows |
When reporting, analysts should state the sample size, the divisor used and whether the figure describes individual variation or estimate precision. Writing a result as “mean 42.0 (SD 6.3, n = 120)” leaves no ambiguity, whereas a bare plus-minus sign can be read as either a standard deviation or a standard error.
Common mistakes with standard deviation
The most frequent errors come from applying normal-distribution intuition to data that are not normal and from confusing related quantities. The following list covers the mistakes that most often distort conclusions.
- Confusing standard deviation with standard error, which makes estimates look far more or far less precise than they are.
- Using the empirical rule on skewed data, where far more or fewer than 95% of values may lie within two standard deviations.
- Averaging standard deviations across groups instead of pooling their variances weighted by sample size.
- Ignoring outliers, since one extreme value can double the standard deviation of a small sample.
- Comparing raw standard deviations across variables with different means or units instead of using the coefficient of variation.
Numerical precision is a further, less visible pitfall. The naive one-pass formula that subtracts the squared mean from the mean of squares suffers from catastrophic cancellation on large values, so production code typically uses Welford’s online algorithm or a two-pass method to keep results stable.
Stripped to the essentials
An average tells you the typical value, but it does not tell you whether everything is close to that value or all over the place. Standard deviation answers that second question. It is roughly the usual distance between a single value and the average.
Imagine two coffee shops that both average a five-minute wait. In the first, nearly everyone waits between four and six minutes. In the second, some people are served at once and others wait twelve minutes. The averages match, but the second shop has a much larger standard deviation. The key idea is simple: a small number means consistency, a large number means unpredictability.
Where it is applied
A packaging plant fills cereal boxes labeled 500 grams. The quality team weighs a sample of 50 boxes every hour and finds a mean of 503 grams with a sample standard deviation of 4 grams. Under a roughly normal distribution, about 95% of boxes fall between 495 and 511 grams, which means a noticeable share of boxes is lighter than the labeled weight.
Rather than raising the target weight and giving away product, the team investigates the filler. After servicing a worn dosing valve, the standard deviation drops to 1.5 grams. The plant then lowers the target mean to 502 grams while keeping nearly all boxes above 500 grams. The outcome is less underweight packaging and less wasted product from one improvement in consistency, which is the practical value of watching spread rather than only the average.