Coefficient of Variation Calculator
Measure standard deviation relative to the mean for a population or a sample, with the small-sample bias correction applied automatically.
Inputs
Choose Sample to apply the finite-sample bias correction.
Required and nonzero. US decimal point; comma thousands and scientific notation are accepted.
Required. Enter a finite value greater than or equal to zero in the same units as the mean.
Required for Sample datasets. Enter a whole number from 2 to 1,000,000.
Result
Coefficient of variation
7.82%
Sample result includes the factor 1 + 1/(4n).
Calculation trace
(1 + 1/(4 × 5)) × (6.3 / 84.6) × 100% = 7.82%
How to use this coefficient of variation calculator
What this calculator does
This calculator expresses dispersion relative to the mean. For a population it computes CV = (σ / μ) × 100%. For a sample it starts from (s / x̄) × 100% and applies the small-sample correction 1 + 1/(4n). The result is dimensionless and shown as a percentage, which makes relative variability easier to compare when datasets use the same kind of ratio-scale measurement but have different means. It does not decide whether a particular CV is acceptable for your field, and it does not infer the mean or standard deviation from raw observations.
The coefficient of variation is most interpretable for ratio-scale data with a meaningful zero. NIST explains both the basic ratio and the small-sample correction in its coefficient of variation reference, and also notes that values become unstable when the mean is close to zero.
When to use it
Use this calculator when you want to compare relative process variability across measurements with different average levels, summarize repeatability relative to a typical measurement, compare dispersion across groups that use compatible ratio scales, or report a compact variability measure after you have already calculated a mean and standard deviation. It is especially useful when an absolute standard deviation alone is hard to compare because the datasets have very different means.
How to calculate
- The calculator opens with a complete demonstration: Dataset type is Sample, Mean (μ or x̄) is 84.6, Standard deviation (σ or s) is 6.3, and Number of samples (n) is 5. The matching Excel workbook is validated during initialization, so Download Excel is ready immediately.
- Choose Dataset type. Select Population when the statistics describe the full population of interest. Select Sample when the statistics come from a subset and you want the finite-sample correction used here.
- Replace Mean (μ or x̄) and Standard deviation (σ or s) with your values. Both must be finite; the mean cannot be zero, while standard deviation may be zero. Use a decimal point for decimals. Commas are accepted only as three-digit thousands separators, and scientific notation such as 1.2e3 is accepted.
- If Dataset type is Sample, enter Number of samples (n) as an integer from 2 to 1,000,000. The result updates live as you type.
- Read Coefficient of variation in the Result panel. Download Excel exports the current validated inputs and canonical result to a real .xlsx workbook. Reset clears the demonstration data instead of restoring it; after Reset, Download Excel stays disabled until a complete valid state is entered again.
Input guide
Dataset type is required and accepts Population or Sample. Population uses the direct ratio σ/μ. Sample uses s/x̄ and the correction based on n. In the demonstration, Sample is selected. A common mistake is choosing Population simply because the dataset is large; the choice should reflect whether the data cover the entire population you mean to describe.
Mean (μ or x̄) is a required finite number in the original data units. The example is 84.6. The mean cannot be zero because division by zero would make the CV undefined. A mean very close to zero can make the percentage extremely sensitive to small changes, so interpret such results cautiously. A negative mean is accepted to preserve the signed CV identity, but ratio-scale applications normally have a meaningful zero and nonnegative values.
Standard deviation (σ or s) is required, finite, nonnegative, and in the same units as the mean. The example is 6.3. Increasing standard deviation while holding the mean fixed increases the magnitude of the CV; entering zero produces a CV of 0%. Do not mix a population standard deviation with a sample mean or vice versa.
Number of samples (n) is required only when Dataset type is Sample. Enter a whole count from 2 to 1,000,000; the example is 5. Smaller n produces a slightly larger correction factor, while the correction approaches 1 as n grows. Do not enter degrees of freedom or n – 1; enter the actual sample count.
Output guide and worked example
Coefficient of variation is the primary output and is shown as a percentage. Its sign follows the mean because standard deviation is nonnegative. A value of 0% means zero standard deviation. A larger absolute percentage means the standard deviation is larger relative to the mean, but there is no universal threshold for “good” or “bad” variability. The header pills labeled Dataset, Mean, Standard deviation, and CV repeat the current selection, inputs, and primary result so you can verify the active state at a glance.
For the startup example, the raw sample ratio is 6.3 ÷ 84.6 = 0.074468..., or 7.4468%. With n = 5, the correction factor is 1 + 1/(4 × 5) = 1.05. Multiplying 7.4468% by 1.05 gives 7.8191%, displayed as 7.82%. The Calculation trace shows the same operation, and the exported workbook stores the underlying numeric values rather than copying rounded display text.
How should you interpret a coefficient of variation?
CV is a relative measure: it scales standard deviation by the mean. This is why two datasets can have different standard deviations yet still have similar relative variability. The percentage is useful only in context. A process specification, laboratory method, physical measurement, or business metric may have its own acceptable variability range, so domain requirements should determine any threshold rather than a generic rule of thumb.
The method depends on meaningful scale properties. NIST's guide to measures of scale explains how standard deviation summarizes spread and why other scale estimators can be more robust for heavy-tailed data. If outliers dominate the standard deviation, the CV will inherit that sensitivity.
When can the coefficient of variation mislead?
Do not treat CV as universally meaningful for interval scales that lack a true zero. Celsius and Fahrenheit temperatures are classic examples: changing the zero point changes the mean and therefore changes the CV even though the physical spread is unchanged. The same issue appears when a mean is near zero; a small denominator can create a huge percentage that is mathematically correct but practically unstable.
CV also does not replace a distribution check. If data contain strong skew, extreme tails, regime changes, or mixed positive and negative observations, a single mean-and-standard-deviation ratio may hide important structure. For process work, NIST's process variability guidance emphasizes that observed variation can arise from multiple controlled and uncontrolled sources. Use CV alongside plots, distribution diagnostics, and domain-specific quality limits when those details matter.