Coefficient of Variation (CV)

You need to compare the spread of two data sets on completely different scales. The standard deviation won't work; it stays in the original units. The coefficient of variation (CV) solves that: it is the ratio of the standard deviation to the mean, expressed as a percentage. The formula is CV = (standard deviation / mean) × 100%. A CV of 10% means the standard deviation is one-tenth the size of the mean. That unitless number lets you compare variability across different tests, different labs, or different years.

The coefficient of variation formula is straightforward: CV = (σ / μ) × 100% for a population, or CV = (s / x̄) × 100% for a sample. The result is a percentage. Because it is a ratio, it is only meaningful when the mean is positive and not close to zero. If the mean is zero or negative, the CV becomes undefined or misleading. The NIST/SEMATECH e-Handbook of Statistical Methods defines the coefficient of variation as the ratio of the standard deviation to the mean, expressed as a percentage.

Compare the variability of two processes: process A produces bolts with a mean length of 50.0 mm and a sample standard deviation of 1.5 mm. Process B produces bolts with a mean length of 2.0 mm and a sample standard deviation of 0.2 mm. The standard deviations alone (1.5 mm vs 0.2 mm) make process B look more precise. But the scales are different.

Calculate the CV for process A: (1.5 / 50.0) × 100% = 3.0%. For process B: (0.2 / 2.0) × 100% = 10.0%. Process A, with the larger absolute standard deviation, actually has lower relative variability. The CV tells you that process A is more consistent relative to its own mean. Without the CV, you would draw the wrong conclusion.

The CV is only valid for data measured on a ratio scale, a scale with a true zero point, like length, weight, time, or concentration. If the zero point is arbitrary, as in temperature measured in Celsius or Fahrenheit, the CV is meaningless. A temperature of 0°C does not mean 'no heat', so the ratio SD/mean has no physical interpretation. Stick to the standard deviation or IQR for interval-scale data.

The CV also fails when the mean is near zero. A small mean inflates the CV, making a stable process look wildly variable. For example, a lab measuring a concentration near the detection limit will get a CV over 50% even if the measurements are tight in absolute terms. In that case, report the standard deviation on the log-transformed data or use a different metric.

There is no universal threshold for a 'good' coefficient of variation. A CV under 5% might be excellent in analytical chemistry but impossible in psychology. Field-specific benchmarks exist, and you should use the one from your discipline.

Clinical Chemistry and Immunoassays

The Clinical Laboratory Standards Institute (CLSI) guideline EP12, 'User Evaluation of Precision', 2nd edition (2014), provides benchmarks. For clinical chemistry assays, aim for a CV of 5% or less. For immunoassays, a CV of 10% or less is typical. These are guidelines, not hard limits. Check the specific test requirements from the CLSI or your regulatory body for the exact values in force.

Finance and Economics

In finance, the CV is used to compare the risk-to-return ratio of investments. A lower CV means less risk per unit of return. There is no fixed 'good' CV; it depends on the asset class. A CV of 30% might be acceptable for a high-growth stock but terrible for a bond fund.

When the CV Is Not the Right Tool

If your data set is small (n < 10), the sample standard deviation is biased low, and the CV computed from it will be too low as well. The bias in the standard deviation is typically small for n > 30 but can be several percent for n < 10, as noted in Casella & Berger, 'Statistical Inference', 2nd ed., section 7.2. For small samples, use the CV with caution and report the sample size alongside it.

Excel does not have a single CV function. You compute it in two steps: get the standard deviation and the mean, then divide.

For a sample data set in cells A1:A50, enter =STDEV.S(A1:A50) to get the sample standard deviation. Enter =AVERAGE(A1:A50) for the mean. In a third cell, divide the SD by the mean and multiply by 100: =STDEV.S(A1:A50)/AVERAGE(A1:A50)*100. The result is the CV as a percentage.

If your data is the entire population, use STDEV.P instead of STDEV.S. Using STDEV.P on a sample underestimates the population variability, which makes the CV too low. That is the most common error. Google Sheets uses the same function names: STDEV.S and STDEV.P.