Coefficient of Variation Calculator
Compute the coefficient of variation (CV), a dimensionless measure of relative variability defined as the standard deviation divided by the mean, expressed as a percentage. Used to compare variability across datasets with different units, magnitudes, or context — investment risk, biological variability, measurement precision.
Last updated: May 2026
Compare with similar
About this calculator
The coefficient of variation is CV = (σ / μ) × 100% for a population or CV = (s / x̄) × 100% for a sample, expressed as a percentage. By dividing standard deviation by the mean, CV becomes a unitless number that can compare variability across datasets with completely different scales — annual returns on a stock (mean 8%, SD 15%, CV ≈ 187%) versus body weights in a study (mean 70 kg, SD 10 kg, CV ≈ 14%). Without this normalisation, comparing ‘spread’ between such different quantities is meaningless. CV is widely used in finance (relative risk: assets with the same expected return but higher CV are riskier), biology (relative variability across organisms), engineering (manufacturing process consistency), and analytical chemistry (precision of repeated measurements, where CV < 5% is good and CV > 15% suggests method problems). Variables: σ or s is the standard deviation in the same units as the data; μ or x̄ is the mean in the same units; the ratio is unitless and is multiplied by 100 to express as a percentage. Edge cases: CV is undefined when the mean is zero (divide by zero) — and is meaningless when the mean is small relative to noise, because tiny means produce huge ratios that don’t reflect real variability. CV is also poorly behaved for variables that take both positive and negative values (portfolio returns can be negative), because the sign of the mean depends on the sample. Use CV only when the variable is strictly positive on the natural scale (concentrations, masses, times, ratios) or when both mean and SD are unambiguously well-defined for the application.
How to use
Example 1 — Comparing two manufacturing lines. Line A produces bolts with mean length 50 mm and SD 0.5 mm. Line B produces washers with mean diameter 20 mm and SD 0.3 mm. Which is more precise relative to its size? Line A: CV = (0.5/50) × 100 = 1.0%. Line B: CV = (0.3/20) × 100 = 1.5%. ✓ Line A is more precise on a relative basis — its variability is 1% of its mean, versus 1.5% for Line B. Both are excellent precision for manufacturing (CV < 5% is typical for good processes); the comparison would be impossible without normalising by the mean, since the absolute SDs are misleadingly close (0.5 vs 0.3 mm). Example 2 — Comparing investment risk. Stock X has mean annual return 8% with SD 12%. Stock Y has mean annual return 15% with SD 25%. Which carries more risk per unit of return? Stock X: CV = (12/8) × 100 = 150%. Stock Y: CV = (25/15) × 100 ≈ 167%. ✓ Stock Y has slightly higher CV — slightly more risk per percentage point of return — so the risk-adjusted comparison disfavours Y despite its higher absolute expected return. In finance the inverse (mean/SD = Sharpe-like ratio) is more common; CV gives the same ranking but with the higher number meaning worse rather than better.
Frequently asked questions
Why use the coefficient of variation instead of just standard deviation?
Standard deviation has units (kg, $, °C) and the same absolute SD can mean very different things depending on the scale of the data. A 10-kg SD is huge for adults (mean weight ~70 kg, so CV ~14%) but tiny for elephants (mean ~6000 kg, so CV ~0.17%). Without dividing by the mean, you cannot compare ‘how variable’ two completely different datasets are. CV solves this by being unitless — it tells you the percentage of the mean that the standard deviation represents. This makes CV ideal for comparing precision of measurement methods (a CV of 2% is good regardless of whether you measure mg/L or mol/L), comparing biological variability across species, comparing financial volatility across assets at different price levels, and any context where the scale of the variable is not fixed.
What is a ‘good’ or ‘high’ CV?
It depends on the context. In analytical chemistry, CV < 5% is considered good precision, CV > 15% is usually unacceptable. In biological measurement, CV up to 20% is normal for many traits. In financial markets, CV well above 100% is common (stocks routinely have SD > mean). In manufacturing, CV < 1% is the standard for high-precision processes. There is no universal threshold for ‘good’ or ‘bad’ CV — each field has its own conventions established by what is achievable for that type of measurement. When CV exceeds 100%, the SD is larger than the mean, indicating either very high variability or a near-zero mean; either way, CV is no longer a useful summary because relative variability becomes nonsensical. Always interpret CV against the conventions of your specific field.
When is CV unreliable or misleading?
CV is unreliable when the mean is close to zero — a tiny mean inflates CV regardless of how small the SD is. For example, daily stock returns averaging 0.05% with SD 1% give CV = 2000%, which makes the variation look extreme but actually just reflects the fact that mean daily returns are tiny. CV is also problematic for variables that can take negative values: portfolio returns, temperature changes, growth rates — the sign of the mean depends on the sample, so CV becomes sign-ambiguous and uninterpretable. CV assumes the variable has a natural zero (mass, time, concentration); it doesn’t work for interval-scale variables like Celsius temperature where zero is arbitrary. And CV breaks down for highly skewed distributions where mean and SD don’t summarise the spread well anyway. In all these cases, use absolute SD with appropriate scaling, percentile-based spread measures (IQR), or a different relative-variability metric.
What are the most common mistakes people make with CV?
The first is using CV on variables that can be negative (portfolio returns, temperature anomalies) — the sign of the mean affects the interpretation, often making CV nonsensical. The second is computing CV when the mean is very small relative to the SD; this produces enormous CV values that look impressive but actually just reflect the small denominator. The third is comparing CVs across measurements with different transformations — log-transformed data has a different CV than the raw data, and the two are not directly comparable. The fourth is using CV as a single summary for a multimodal distribution; like the mean and SD it summarises, CV assumes unimodal symmetric data. The fifth is forgetting the percentage convention — CV is conventionally multiplied by 100, but some software returns the raw ratio, and mixing the two produces 100× errors. The sixth is treating CV as a measure of skewness or shape; it’s not — it’s only a relative spread measure and tells you nothing about distribution shape.
When should I not use this calculator?
Skip it when the mean is close to zero or can be negative — CV is undefined or misleading in both cases. Avoid it for ordinal or categorical data where mean and SD have no clear meaning. It is the wrong tool for log-transformed data where the geometric coefficient of variation (GCV) is more appropriate. Do not use it for time-series with autocorrelation, where the basic SD underestimates true variability and CV inherits the bias. Skip it for highly skewed distributions where the mean is not a representative central value; use median absolute deviation (MAD) divided by the median instead. And do not use it as the sole summary statistic — pair it with the raw mean, SD, and a histogram or boxplot so readers can see the underlying distribution shape, not just the relative-variability ratio.