What Is Standard Deviation?

Standard deviation is the typical distance of a data point from the mean. If the mean is the center of a data set, what is standard deviation is the measure of how far, on average, every value is from that center. A low standard deviation means the values cluster tightly around the mean. A high standard deviation means they spread out widely. The key question is whether that number is high or low relative to the scale of the data.

Imagine two data sets of exam scores. Set A: 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95. The mean is 90. The scores are all within 5 points of 90. Set B: 70, 75, 80, 85, 90, 95, 100, 105, 110. The mean is also 90, but the scores range from 70 to 110. A dot-plot comparison shows Set A's dots packed around 90; Set B's dots spread from 70 to 110. Both have the same mean, but the standard deviation tells you they are not the same data set. Set A has a low standard deviation; Set B has a high one.

The phrase 'typical distance from the mean' is a direct description of what standard deviation measures. For Set A above, each score is between 1 and 5 points from the mean. The standard deviation pulls that range into a single number that represents the average of those distances. It is not the simple average of the absolute deviations; those sum to zero. Instead, standard deviation squares each deviation, averages the squares (using n-1 for a sample), and then takes the square root to return to the original units.

When you read that a standard deviation is 3.5, it means the typical data point is about 3.5 units from the mean. But 'typical' here is a statistical shorthand, not a guarantee about every point. Some points will be closer, some farther. Only in a perfectly normal distribution does the 68-95-99.7 rule tell you the exact percentages within each distance.

You cannot judge whether a standard deviation is high or low by looking at the number in isolation. A standard deviation of 5 on test scores out of 100 is moderate. A standard deviation of 5 on temperatures in Celsius is huge. The two comparisons that matter are to the mean and to the range.

Compare To The Mean

Divide the standard deviation by the mean to get the coefficient of variation. A coefficient of variation below 15% indicates low dispersion; above 30% indicates high. This is a rule of thumb, not a universal threshold. In analytical chemistry, a CV under 1% is expected. In finance, a CV of 30% on stock returns is normal.

Compare To The Range

In a normal distribution, the range is about 6 standard deviations (3 on each side of the mean). If your range is 100 and your standard deviation is 5, that ratio is 20:1, suggesting a tight cluster. If the ratio is 4:1, the data is spread wide.

Use Chebyshev's Inequality As A Safety Net

Chebyshev's inequality gives you a safety net when the data is not normal. At least 75% of any data set lies within 2 standard deviations of the mean, and at least 89% within 3 standard deviations. If your standard deviation is so large that 2 standard deviations covers more than the entire observed range, the data is extremely spread out.

When the data is approximately bell-shaped, apply the empirical rule. About 68% of values fall within 1 standard deviation of the mean, about 95% within 2 standard deviations, and about 99.7% within 3 standard deviations. This rule only applies to normal distributions. For non-normal data, use Chebyshev's inequality instead, which provides a weaker but universally applicable guarantee.

Standard Deviation Cannot Be Negative

Variance is the average of squared deviations, which is always zero or positive. The square root of a non-negative number is non-negative. A negative standard deviation is impossible. A zero standard deviation means every value in the data set is identical.

Standard Deviation Is Not The Average Deviation

The average absolute deviation from the mean sums the absolute differences and divides by n. Standard deviation squares the differences first, which gives more weight to points far from the mean. This is why standard deviation is more sensitive to outliers than the average absolute deviation.

Standard Deviation Does Not Tell You The Shape

Two data sets can have the same mean and the same standard deviation but completely different shapes: one symmetric, one skewed, one bimodal. The standard deviation only describes spread, not shape. You need skewness and kurtosis for shape.

Sample Standard Deviation Is Biased

The formula with n-1 (Bessel's correction) makes the sample variance an unbiased estimator of the population variance. But it does not make the sample standard deviation unbiased. The sample standard deviation is biased low, especially for small sample sizes. For n below 10, the bias can be several percent. This is a known mathematical fact from Casella & Berger's Statistical Inference.

You cannot compare standard deviations directly when the data sets have different units or very different means. The coefficient of variation solves this by dividing the standard deviation by the mean. It is a unitless ratio. A CV of 12% for height in centimeters and a CV of 12% for weight in kilograms means the two data sets have the same relative variability.

When the mean is near zero or negative, the CV becomes meaningless. In that case, use the interquartile range instead of the standard deviation for comparing spread, or use the range of the data.

Standard deviation tells you two things: the typical distance of data points from the mean, and whether that distance is large or small relative to the scale of the data. To interpret a standard deviation value, compare it to the mean (via the coefficient of variation) and to the range. A standard deviation of 10 on a scale of 0 to 100 is moderate; the same value on a scale of 0 to 20 is extreme.

For normally distributed data, the empirical rule gives you exact percentages. For any distribution, Chebyshev's inequality gives you a minimum guarantee. The most common mistake is to assume a standard deviation is 'high' or 'low' without context. The second most common is to use the standard deviation when the data is skewed: in that case, report the median and the interquartile range instead.

When you see a standard deviation in a report, ask: what is the mean of this data? What is the range? What is the coefficient of variation? If the CV is below 15%, the data is tightly clustered. If the CV is above 30%, the data is widely spread. For a CV above 50%, the standard deviation is more than half the mean, which indicates extreme outliers or a very skewed distribution. In that case, use the median and the IQR instead of the mean and standard deviation.

Source: OpenStax Introductory Statistics 2e, section 2.7.

CV Range	Standard Deviation Relative To Mean	Interpretation
0%	All values identical	No variability. Check for data entry errors.
0% – 15%	Low	Data points tightly cluster around the mean. High consistency.
15% – 30%	Moderate	Noticeable variation, but data still centers on the mean.
30% – 50%	High	Data points widely spread. Mean less representative.
Above 50%	Very high	Extreme dispersion. Look for outliers or skewed distribution.

Standard deviation is the square root of the variance. Variance is the average squared deviation from the mean. Because variance is in squared units (e.g., dollars squared), taking the square root returns the measure to the original unit (e.g., dollars). The sample variance uses n-1 in the denominator; the population variance uses N. The formula is standard deviation = sqrt(variance).

Standard deviation suits: AP Statistics students who need to calculate summary statistics and check their work. College Stats 101 students who need to know which measure to report. Analysts and researchers who need a quick summary of a small data set. Self-taught data enthusiasts who need a clear, worked example.

Standard deviation does not suit: Anyone needing inferential statistics (hypothesis tests, confidence intervals, p-values, regression), probability theory, or data visualization beyond a box plot. Those users should go to a dedicated inferential statistics calculator or a full statistical package. Also, anyone working with highly skewed data: use the median and IQR instead.