Coefficient of Variation Calculator — CV as a percentage

The coefficient of variation calculator above answers one question: how large is the spread in this data relative to its own average? It divides the standard deviation by the mean and reports the result as a percentage. Paste raw values, or switch modes and type in a mean and standard deviation you already have. Sample and population figures appear together.

Arb Digital publishes this alongside the wider statistics set because relative spread is the one dispersion measure that survives a change of units. A standard deviation of 3 tells you nothing until you know whether the mean is 6 or 6,000. The CV supplies that context.

What This Coefficient of Variation Calculator Does

The coefficient of variation, usually written CV, is the standard deviation expressed as a fraction of the mean: CV = sigma divided by mu, conventionally multiplied by 100 and reported as a percentage. Because numerator and denominator carry the same units, those units cancel and the result is unitless. A CV from grams is directly comparable to one from seconds.

In raw-data mode the tool computes the mean, sums the squared deviations, and produces both standard deviations before dividing each by the mean. In direct mode you supply the mean and standard deviation; add the sample size and it converts your figure to the other basis. When the mean sits at or near zero, or the data holds negative values, the page says so rather than printing a percentage that means nothing.

How to Use It

1. Pick your input mode. Raw data is the usual choice; direct entry is for a published mean and standard deviation with no observations behind it.
2. Paste or type your numbers. Commas, spaces, tabs, semicolons and line breaks all work in any mix, and a trailing separator is ignored. Check the parsed count: if a paste brought in a header row, that is where you notice.
3. Choose sample or population. Sample is preselected and divides by n minus 1; population divides by n and is only correct for an entire group.
4. Read the hero figure and any warning. The headline is the CV on the basis named in the label, and a warning outranks it.

The Formula: How the Coefficient of Variation Is Calculated

Take the mean. Subtract it from each value, square the difference, and add the squares: that is the sum of squared deviations. Divide it by n for the population variance or by n minus 1 for the sample variance, take the square root, and divide by the mean.

The n minus 1 divisor is Bessel correction. Deviations from a sample mean are on average smaller than deviations from the true population mean, because that mean is pulled toward the data; the smaller divisor compensates. Our variance calculator works through it in detail. At n equal to 8 the sample standard deviation is about 7% larger than the population figure; at n equal to 200 the gap falls below a quarter of a percent.

The ratio is invariant under a change of unit scale: convert metres to centimetres and both parts multiply by 100, leaving it untouched. It is not invariant when you shift the values by adding a constant, because the standard deviation ignores that shift and the mean does not. That asymmetry drives nearly every misuse.

A Worked Example You Can Check Yourself

The calculator loads with eight values: 2, 4, 4, 4, 5, 5, 7 and 9. It is the classic textbook set for a reason — the population standard deviation is a whole number, so the calculation is checkable by hand.

The values sum to 40, so the mean is exactly 5. The deviations are minus 3, minus 1, minus 1, minus 1, 0, 0, 2 and 4. Squared, they are 9, 1, 1, 1, 0, 0, 4 and 16, adding to exactly 32.

On the population basis, 32 divided by 8 is a variance of 4, whose square root is a standard deviation of exactly 2. The CV is 2 divided by 5, or 40.0000%. No rounding anywhere: the spread is precisely two-fifths of the average.

On the sample basis, divide 32 by 7 instead. The variance is 4.571429 and the standard deviation is its square root, 2.13808993. Divided by 5 that gives 0.42761799, or 42.7618%. The two differ by 2.76 points — the Bessel correction made visible. Switch the basis dropdown to move between them, or use direct mode, where the same pair is preloaded. The NIST/SEMATECH e-Handbook chapter on exploratory data analysis sets out the definitions formally.

Why Relative Spread Beats Absolute Spread

Here is the case the CV exists to handle. One line fills 50 ml vials with a standard deviation of 3 ml; a second fills 500 ml bottles with a standard deviation of 15 ml. Which is less consistent?

The absolute figures say the bottle line, by a wide margin: 15 ml against 3 ml. The relative figures reverse that. Three divided by 50 is 6%; fifteen divided by 500 is 3%. The line with the larger standard deviation is twice the more consistent, its variation being small next to what it delivers.

That reversal is the whole argument. Absolute spread answers how far values stray in the original units, and for that you want the standard deviation calculator. Relative spread answers how far they stray next to their own typical size, and only that question can be put to two scales at once.

Where the Coefficient of Variation Breaks Down

The CV inherits the weaknesses of its denominator. Three failure modes matter.

A mean near zero. As the mean approaches zero the CV diverges toward infinity. Hold the standard deviation at 4. With a mean of 2 the CV is 200%; at 0.5 it is 800%; at 0.1 it is 4,000%. The data barely changed and the statistic multiplied twentyfold. With a small denominator, ordinary sampling noise in the mean swings the ratio wildly, so the CV describes where the mean landed rather than the data. Once the standard deviation approaches the size of the mean, treat the number as unreliable; the calculator flags that on screen.

Data that can go negative. If some values fall below zero the mean can be negative, or sit near zero with large spread either side. A negative CV is not interpretable: the minus sign comes from the denominator, not from anything about variability, so reading minus 30% as less variable than 30% is wrong, and comparing a negative CV with a positive one is meaningless. Percentage changes, temperature anomalies and profit and loss figures all fall into this trap; the tool warns on negative values or a negative mean.

Anything that is not a ratio scale. A ratio scale has a true zero — zero means none of the thing — and no negative values. Length, mass, duration, revenue and counts qualify, and on them saying one value is twice another is a real statement. An interval scale has a zero chosen by convention, and on it the CV is not defined in any useful sense.

The Temperature Example That Settles the Ratio-Scale Question

Temperature is the cleanest demonstration available: the same day gives two different CVs depending on the scale.

Take four readings from one day: 10, 15, 20 and 25 degrees Celsius. The mean is 17.5 and the squared deviations are 56.25, 6.25, 6.25 and 56.25, summing to 125. On the population basis the variance is 31.25 and the standard deviation is 5.590170, so the CV is 31.9438%.

Now write the identical day in Fahrenheit: 50, 59, 68 and 77. The mean is 63.5. Every deviation is multiplied by 1.8, so the standard deviation becomes 10.062306. The CV is 10.062306 divided by 63.5, giving 15.8462%.

Nothing about the weather changed — same air, same four moments — yet one CV is more than double the other. The cause is the shift: Fahrenheit adds 32 after scaling, inflating the mean without touching the spread. Since neither zero marks an absence of temperature, neither denominator has a claim to being the right one, and neither percentage means anything. Kelvin, which does have a true zero, gives a third answer near 1.9%. Wherever your scale has a zero someone chose, do not compute a CV.

Reading a CV Once You Have One

There is no universal threshold separating a good CV from a bad one. A CV of 2% is unremarkable in survey research and alarming in a calibrated laboratory assay, so the statistic supports comparison, not a verdict. Three habits keep it honest. Report the basis, since the two figures differ by a factor of the square root of n over n minus 1. Report n as well: a CV from five observations is an impression, not an estimate. And check the shape first, since one extreme value inflates the numerator and drags the denominator. Run the values through the mean median mode calculator — a mean far from the median warns that a symmetric summary is wrong — or a robust measure from the interquartile range calculator when the tails are heavy. The Penn State Eberly College of Science statistics courses cover descriptive summaries and their assumptions in sequence.

How the CV Relates to the Other Spread Tools

All these measures come from the same sum of squared deviations, each stopping at a different point. The standard deviation calculator gives the absolute spread in the original units; if you want spread relative to the mean, so two scales can be compared, use this coefficient of variation calculator instead. The variance calculator gives that spread in squared units; if you want it back in the original units, use the standard deviation calculator. The z-score calculator locates one individual value; if you want the dispersion of the whole set rather than the position of a single observation, use this coefficient of variation calculator.

One further distinction: the CV describes the spread of the data, not the precision of an estimate made from it. Uncertainty around a sample mean shrinks as n grows; the CV does not. How precisely you know the average is a question for the confidence interval calculator.

Common Mistakes to Avoid

• Computing a CV on an interval scale — temperature and calendar years have arbitrary zeros, so the ratio changes with the scale.
• Reporting a CV when the mean is near zero — the ratio becomes unstable, and a tiny movement in the denominator can double or halve it.
• Reading a negative CV as low variability — the sign comes from the mean, not the spread, and cannot be compared with a positive CV.
• Leaving the basis unstated — the two differ by 2.76 percentage points on the example loaded here, so a comparison across bases is not fair.
• Comparing CVs across very different sample sizes — four decimal places do not turn a handful of observations into an estimate.

Related Free Tools From Arb Digital

Get the absolute spread with the standard deviation calculator, its squared form with the variance calculator, one value positioned with the z-score calculator, the centre with the mean median mode calculator and the shape with the five number summary calculator. The free online tools hub lists the whole set.

Frequently Asked Questions

This calculator performs a descriptive computation on the values you enter. Whether the coefficient of variation is a meaningful summary of your particular measurement scale is a separate question the tool cannot decide for you.