Coefficient of Variation Calculator
Two datasets can share the same standard deviation and still behave nothing alike: a spread of 3 around a mean of 14 is wild, while a spread of 3 around a mean of 100 is tight. The coefficient of variation fixes that comparison problem by expressing the standard deviation as a percentage of the mean.
Enter one dataset, or two to compare side by side, choose whether the data is a sample or a full population, and the calculator returns each CV with its mean and standard deviation, plus a verdict on which dataset is more consistent.
CV is only meaningful for ratio-scale data with a non-zero mean. A lower CV means less variability relative to the average, that is, more consistency.
Quick Answer
Key takeaways
- CV = standard deviation divided by the absolute value of the mean, times 100. The units cancel, so datasets in dollars, grams, or seconds can be compared directly.
- Sample or population is a real choice, not a detail: the same data below gives 22.5877% as a sample and 20.2031% as a population.
- CV only makes sense on ratio scales with a genuine zero and a mean comfortably away from zero. Near-zero means make it explode.
- Lower CV means more consistency relative to the average. It does not, by itself, mean better: context decides whether consistency is the goal.
- Quick Answer
- How to use the coefficient of variation calculator
- The formula, and the one decision inside it
- Worked example: the dataset with the bigger SD wins
- Where CV earns its keep
- When CV misleads, and what to use instead
- What competing calculators miss
- Frequently Asked Questions
- Add This Calculator to Your Website
- References
How to use the coefficient of variation calculator
Start with the toggle. If your numbers were drawn from a larger group you actually care about, a week of production sampled from a year, a class sampled from a school, choose sample. If your numbers are the entire group, every part the factory made that day, choose population. Then paste dataset A as comma-separated numbers, and dataset B as well if you want the comparison verdict. The calculator reports the mean, standard deviation, and CV for each dataset and states which one is more consistent.
One formatting note: values must be separated by commas. Spaces after commas are fine, and stray empty entries are ignored, but a value that is not a number stops the calculation with an error rather than silently corrupting the result.
The formula, and the one decision inside it
CV is the standard deviation expressed as a percentage of the mean. Everything else is the question of which standard deviation.
The absolute value around the mean is deliberate. Some sources divide by the signed mean, which makes CV negative whenever the mean is negative and adds no information. This calculator follows the convention documented by the NIST coefficient of variation reference: CV is a magnitude, reported as a non-negative percentage, and it belongs on ratio-scale data where zero means none of the quantity.
That ratio-scale condition rules out a classic mistake. A week averaging 20 degrees Celsius with an SD of 2 has a CV of 10% in Celsius and a completely different CV in Fahrenheit, because both scales have an arbitrary zero. Convert to Kelvin, a true ratio scale, and CV becomes meaningful. NIST makes the same point: on interval scales the CV changes with the scale, so the comparison is an artifact.
Worked example: the dataset with the bigger SD wins
The calculator is pre-filled with both datasets, treated as samples, and every figure below is what it returns.
| Measure | Dataset A: 10, 12, 14, 16, 18 | Dataset B: 100, 102, 98, 105, 95 |
|---|---|---|
| n | 5 | 5 |
| Mean | 14 | 100 |
| Sample SD | 3.1623 | 3.8079 |
| CV | 22.5877% | 3.8079% |
| Verdict | Less consistent | More consistent |
The hand check for A: squared deviations from 14 are 16, 4, 0, 4, and 16, summing to 40. Sample variance is 40 / 4 = 10, SD is the square root of 10 = 3.1623, and CV is 3.1623 / 14 x 100 = 22.5877%. Switch the toggle to population and the same data divides by 5 instead: variance 8, SD 2.8284, CV 20.2031%. Nothing about the data changed; the assumption about what it represents did.
Where CV earns its keep
Quality control. Two machines fill bottles to different target volumes, so their standard deviations are not comparable until each is scaled by its target. A fill line holding a CV under a few percent is a stable process; a rising CV is an early warning before parts start failing inspection.
Finance. CV is risk per unit of return. Two funds with the same average return and different CVs are not the same investment: the higher-CV fund swung more to earn the same average. It pairs naturally with variance-based risk measures such as the beta variance calculator, which looks at market-relative risk instead of raw dispersion.
Laboratory and field measurement. Assay precision is routinely reported as a CV, because a method that reads 100 plus or minus 4 is directly comparable with one that reads 10 plus or minus 0.4: both are 4%. When the question turns from spread to sampling error, the central limit theorem calculator covers how the mean itself stabilises as samples grow, and the area to z-score calculator converts individual values into standard-deviation units for a different kind of comparison.
Comparing published results. Studies report means and SDs in their own units. Converting each to a CV is often the only way to ask whether two studies saw similar consistency, which is also why CV shows up when judging whether a difference is real, alongside tools like the statistical significance calculator.
When CV misleads, and what to use instead
CV has failure modes that a bare number will never warn you about. This calculator blocks the worst one, a mean of exactly zero, with an error, because division by zero is undefined. The near-miss cases are on you.
Use CV when
- The data sits on a ratio scale with a real zero (mass, length, money, time)
- Means differ a lot between the datasets being compared
- Units differ, or one dataset is a rescaled version of the other
- The mean is comfortably far from zero
Use the plain SD when
- Both datasets share the same units and similar means
- The mean hovers near zero, so CV swings wildly with tiny changes
- The scale is interval, like Celsius or Fahrenheit temperature
- You care about absolute spread, for example a tolerance in millimetres
What competing calculators miss
The CV tools reviewed for this page are mostly single-dataset, single-formula boxes, and the gaps are predictable.
Miss 1: population formula only. Several top-ranking tools divide by n and never ask whether the data is a sample. For small datasets that silently understates variability: 20.2031% instead of 22.5877% on the worked example. The sample and population toggle here makes the assumption explicit and switchable.
Miss 2: no comparison. CV exists to compare, yet most tools stop at one dataset and leave you to run the second through separately and eyeball the two percentages. The optional dataset B and the consistency verdict do the comparison inside the tool.
Miss 3: no working shown. A lone percentage cannot be audited. The step-by-step block under the result lists n, the mean, the SD, and the division for each dataset, so a teacher, auditor, or reviewer can reproduce every figure.
Miss 4: signed CV with no explanation. Tools that divide by the signed mean report negative CVs for negative-mean data and leave users thinking they made an error. Using the absolute mean, and saying so, matches the NIST convention.
Frequently Asked Questions
There is no universal cutoff; it depends on the field. In manufacturing, a CV under about 5 to 10% often signals a stable process, while investment returns routinely run at 30% or more. Judge a CV against similar datasets from the same context, not against an absolute rule.
Not as computed here: the formula divides by the absolute value of the mean, so the result is never negative. Some sources use the signed mean, which flips the sign for negative-mean data without adding information. When comparing figures from different sources, check which convention each one used.
Because the formula divides by the mean. A mean of zero makes the division impossible, and a mean merely close to zero makes CV explode into a huge, meaningless percentage. In both cases, compare standard deviations directly instead.
Use CV to compare variability across datasets with different means or different units, where raw standard deviations are not comparable. Use the standard deviation when the datasets share a scale and you care about the absolute size of the spread, for example against a manufacturing tolerance.
Only the standard deviation inside it. The sample formula divides the squared deviations by n – 1, which corrects the tendency of a sample to understate the spread of the group it came from. The population formula divides by n. On the worked example the choice moves CV from 22.5877% to 20.2031%.
It means more consistent relative to the average, nothing more. A factory wants a low CV in part dimensions, but an investor may accept a high CV in exchange for higher expected returns, and a researcher may be studying the variability itself. Consistency is a property, not a grade.
Yes, pairwise. Compare A with B, note the CVs, then compare the more consistent one with C by replacing dataset A. Every comparison uses the same formula and the same sample or population setting, so the verdicts stay consistent with each other.
Add This Calculator to Your Website
References
- National Institute of Standards and Technology, “Coefficient of Variation,” Dataplot reference manual: defines the sample CV as the ratio of the standard deviation to the mean, restricts it to ratio-scale data, and warns it is sensitive when the mean is near zero. NIST: Coefficient of Variation. Accessed October 2026.
- National Institute of Standards and Technology, “Common Coefficient of Variation Confidence Limits,” Dataplot reference manual: population and sample definitions of CV and the conditions for comparing CVs across groups. NIST: Common Coefficient of Variation Confidence Limits. Accessed October 2026.
- National Institute of Standards and Technology, “Sensitivity Coefficients for Measurements on the Test Item,” NIST/SEMATECH e-Handbook of Statistical Methods, Section 2.5.6.1: sample standard deviation with the n – 1 divisor used in measurement uncertainty work. NIST e-Handbook: Sensitivity Coefficients. Accessed October 2026.
- Statistics By Jim, “Coefficient of Variation in Statistics”: CV as a unitless relative measure of variability, also called the relative standard deviation, with guidance on interpretation. Statistics By Jim: Coefficient of Variation in Statistics. Accessed October 2026.
- R-Coder, “Coefficient of Variation in R”: population CV defined with the absolute value of the mean, and the sample form using the sample standard deviation, matching the convention this calculator follows. R-Coder: Coefficient of Variation in R. Accessed October 2026.