The coefficient of variation (CV) tells you how spread out data points are relative to the average. It expresses variability as a percentage of the mean, which makes it especially useful when you need to compare consistency across datasets that use different scales or units. A CV of 5% means the typical spread is small relative to the average, while a CV of 30% signals much more variability.
How the CV Is Calculated
The formula is straightforward: divide the standard deviation by the mean, then multiply by 100 to get a percentage.
CV = (Standard Deviation / Mean) × 100
Say you’re looking at delivery times for a shipping company. If the average delivery takes 10 days with a standard deviation of 2 days, the CV is 20%. That single number captures how consistent the process is, scaled to the size of the thing being measured. This percentage format is what gives the CV its power: it strips away the original units (days, dollars, milligrams) and leaves you with a pure measure of relative spread.
Why CV Works Better Than Standard Deviation
Standard deviation tells you how far values typically fall from the mean, but it’s locked to the original units. That creates a problem when you want to compare variability between two things measured on completely different scales. Is a standard deviation of 4 kilograms “more variable” than a standard deviation of 12 millimeters? There’s no way to answer that question with raw standard deviations alone.
The CV solves this by converting both into percentages. You can directly compare the consistency of a production line measuring weight against one measuring length, or compare the volatility of two investments with very different price ranges.
The CV also handles situations where variability naturally scales with size. In laboratory testing, for instance, a test method might show a standard deviation of 4 units at a concentration of 100 and a standard deviation of 8 units at a concentration of 200. Those look like different levels of precision if you only look at the standard deviation. But the CV is 4.0% at both levels, revealing that the method’s precision is actually consistent across concentrations. This is why labs and quality control programs often prefer CV over standard deviation when evaluating measurement reliability.
What High and Low Values Mean
A lower CV means the data points cluster more tightly around the mean. A higher CV means they’re more spread out. There’s no single universal cutoff that separates “good” from “bad” because acceptable variability depends entirely on context. A CV of 10% might be perfectly fine for measuring rainfall across cities but unacceptable for a pharmaceutical manufacturing process.
That said, some general patterns hold. In many scientific and industrial applications, a CV under 10% suggests reasonably consistent data. CVs above 20-30% often indicate substantial variability that warrants closer investigation. When comparing two datasets, two products, or two processes, the one with the smaller CV is more consistent relative to its average.
The CV is also useful for evaluating statistical models. When you fit a model to data, the CV of the residuals (the differences between predicted and actual values) tells you how well the model performs. A smaller CV means the predictions land closer to reality, suggesting a better fit.
Comparing Consistency Across Groups
One of the most common real-world uses of CV is comparing how consistent something is across groups that differ in magnitude. Suppose you manage two retail stores: one averages $50,000 in monthly revenue with a standard deviation of $5,000, and another averages $200,000 with a standard deviation of $15,000. The second store has three times the raw variability. But the CVs are 10% and 7.5%, respectively, meaning the higher-revenue store is actually more consistent relative to its size.
In medical research, the CV helps distinguish between two types of variability that matter for different reasons. Within-person variability (how much a single patient’s measurements fluctuate over time) and between-person variability (how much measurements differ across a population) can both be expressed as CVs. Researchers use intra-individual CVs to evaluate whether a measurement tool gives reliable, repeatable results for the same person, which is critical when tracking patients over time. Because the scales of different measurement tools often differ, the CV serves as a common currency for comparing their reliability.
CV and Relative Standard Deviation
You may also encounter the term “relative standard deviation” (RSD). For practical purposes, RSD and CV are the same thing: both use the formula (standard deviation / mean) × 100. The terminology differs by field. Chemistry and analytical labs tend to say RSD, while statistics, biology, and finance lean toward CV. Both express the dispersion of a dataset relative to its mean, and you can treat them interchangeably.
When the CV Doesn’t Work
The CV has a significant limitation: it breaks down when the mean is close to zero. Because the mean sits in the denominator of the formula, a near-zero average inflates the CV to absurdly high values that don’t reflect meaningful variability. If your dataset has an average of 0.001 and a standard deviation of 0.5, the CV would be 50,000%, which tells you nothing useful.
For the same reason, the CV is unreliable for data measured on interval scales where zero is arbitrary rather than absolute. Temperature in Celsius or Fahrenheit is the classic example: 0°C doesn’t mean “no temperature,” so dividing the standard deviation by the mean produces a number that shifts depending on whether you measure in Celsius or Fahrenheit. The CV is designed for ratio scales, where zero genuinely means “none” (height, weight, concentration, time, money).
The CV also assumes that a larger mean naturally comes with proportionally larger spread. When that assumption doesn’t hold, comparing CVs across groups with very different means can be misleading. If you’re working with data that includes negative values, the CV can produce nonsensical or negative percentages, so it should be avoided in those cases entirely.