In statistics, CV stands for coefficient of variation. It measures how spread out data points are relative to their average, expressed as a percentage. You calculate it by dividing the standard deviation by the mean: CV = (standard deviation ÷ mean) × 100. A CV of 10% means the typical spread in your data is about 10% of the average value.
What CV Actually Tells You
Standard deviation tells you how far data points typically fall from the average, but it’s locked to whatever units your data uses. If you’re measuring heights in centimeters, your standard deviation is in centimeters. The coefficient of variation strips away the units entirely by converting that spread into a percentage of the mean. This makes it a pure ratio, a dimensionless number you can use to compare variability across completely different datasets.
The higher the CV, the more dispersed your data is relative to the average. A low CV means your values cluster tightly around the mean. A dataset with a CV of 5% is very consistent; one with a CV of 40% has wide scatter. Think of it this way: if two bakeries both produce loaves that average 500 grams, but one has a CV of 2% and the other 15%, the second bakery’s loaves vary far more in weight even though the average is identical.
Why Not Just Use Standard Deviation?
Standard deviation works well when you’re looking at a single dataset in isolation. But it falls apart as a comparison tool when your datasets have different scales, different units, or very different averages. Suppose you want to compare the consistency of two measurements: one with a mean of 100 and a standard deviation of 4, and another with a mean of 200 and a standard deviation of 8. The second dataset looks twice as variable based on standard deviation alone. But when you calculate CV, both come out to 4%. The spread is actually proportional to the size of the measurements, and both datasets are equally consistent in relative terms.
This proportional relationship shows up constantly in real data. Lab measurements, financial returns, biological samples, and manufacturing outputs all tend to show more absolute variation at higher values. CV accounts for that naturally, which is why it’s often more informative than raw standard deviation when comparing across different concentrations, scales, or categories.
How To Calculate It
The formula is straightforward:
CV = (standard deviation ÷ mean) × 100
Say you recorded five test scores: 78, 82, 85, 80, and 75. The mean is 80, and the standard deviation is roughly 3.7. Dividing 3.7 by 80 and multiplying by 100 gives a CV of about 4.6%. That tells you the scores vary by less than 5% from the average, which is fairly tight clustering.
Now imagine a second class with scores of 55, 90, 72, 98, and 60. The mean is 75, and the standard deviation is roughly 18.2. The CV here is about 24.3%, confirming what your gut already tells you: this group’s performance is far less consistent.
Common Uses of CV
Laboratory and Clinical Science
Labs rely heavily on CV to evaluate how precise their testing methods are. If you run the same blood sample through an analyzer ten times, you want the results to be nearly identical. A CV under 5% generally signals good precision. In many clinical and pharmaceutical labs, a CV above 15% to 20% is considered unacceptably imprecise, and results at that level are often flagged or thrown out. This threshold is somewhat arbitrary, but it’s a widely used benchmark in drug monitoring and analytical chemistry.
One important nuance: at very low concentrations, the same absolute variation produces a much higher CV. Two measurements might have nearly identical standard deviations, but if one is taken at a low concentration, its CV could be 20% (seemingly unacceptable) while the higher concentration shows only 10% (seemingly fine). The actual measurement precision is the same in both cases. This is a real limitation of CV in laboratory settings.
Finance and Investing
Investors use CV as a risk-to-reward ratio. In this context, the standard deviation represents volatility (risk), and the mean represents expected return (reward). A stock with a high average return and low volatility will have a small CV. A stock with modest returns but wild price swings will have a large CV. When comparing two investments with different expected returns, CV helps you see which one gives you more consistency per unit of return. Research on risk sensitivity in both humans and animals has found that neither volatility nor expected value alone predicts how risky a choice feels. Their ratio, the CV, does so reliably.
Manufacturing and Quality Control
Factories use CV-based control charts to monitor whether a production process is staying stable. If you’re making mechanical parts through a sintering process or running a die-casting operation, the CV of your output measurements tells you whether variation is creeping up relative to your target. A sudden increase in CV signals that something in the process has shifted. CV control charts are particularly common in chemical manufacturing, metal fabrication, and textile production, where catching small-to-moderate shifts in consistency matters for product quality.
When CV Doesn’t Work
CV has a critical mathematical limitation: it requires the mean to be meaningfully different from zero. Because you’re dividing by the mean, a mean close to zero inflates the CV to absurdly high values, and a mean of exactly zero makes it undefined. This means CV is useless for data measured on interval scales where zero is arbitrary, like temperature in Celsius or Fahrenheit. A city with a mean January temperature of 2°F would show a wildly different CV than the same city measured in Celsius (about -17°C), even though the actual variability is identical.
CV also breaks down with datasets that include negative values. If your data can go below zero (net profit and loss figures, for example), the mean could be near zero or negative, making the CV misleading or nonsensical. For these types of data, stick with standard deviation or other measures of spread.
Finally, CV works best for data measured on a ratio scale, where zero means “none” and doubling the number means doubling the quantity. Heights, weights, concentrations, prices, and counts all qualify. Test scores on an arbitrary rubric, IQ values, and Likert survey ratings generally don’t, because the intervals between values aren’t proportional in the same way.