Coefficient of Variation

Coefficient of variation (CV) is another name for Relative Standard Deviation (RSD). It is a dimensionless measure that describes relative dispersion. Coefficient of Variation was introduced by Karl Pearson and is expressed as the ratio of standard deviation to mean times 100.CV = (σ / μ) × 100. In case of low CV values, it means greater uniformity, while high values correspond to greater diversity compared to the mean. While standard deviation cannot be compared between two variables with different units, CV makes it possible.

What Is the Coefficient of Variation?

But there is one issue with standard deviation. Standard deviation measures the spread of your data, but it cannot say if this spread is big or small compared to your average value. And it makes no sense once you start comparing two datasets that use different measurement units.

Here’s an example. Two manufacturing plants make identical components. The first plant uses grams: average value = 500g, standard deviation = 25g. The second uses ounces: average = 17.6 oz, standard deviation = 1.1 oz. Which of them operates more consistently? Just by looking at those standard deviations, you won’t get any conclusions since their measurement units are different. However, CV helps.

Plant A CV = (25 / 500) × 100 = 5% Plant B CV = (1.1 / 17.6) × 100 = 6.25%

The process operated by the first plant is more consistent. This is the type of insight that can be achieved only by using coefficients of variation.

This is why people needed such measures, which made CV necessary. It was proposed by Karl Pearson in 1896 paper Mathematical Contributions to the Theory of Evolution, Philosophical Transactions of the Royal Society.In analytical chemistry and laboratory science, the same metric goes by “relative standard deviation” (RSD). Different field, same formula, different name.

The Coefficient of Variation Formula

CV = (Standard Deviation ÷ Mean) × 100

Where:

• σ = standard deviation of the dataset
• μ = arithmetic mean of the dataset
• Result is expressed as a percentage

Population CV vs Sample CV

The formula above looks identical in both cases. What changes underneath it is how standard deviation gets calculated.

Population SD — use this when you have every single data point in the group:

σ = √[Σ(xᵢ − μ)² / N]

Sample SD — use this when you’re working with a subset of a larger group:

s = √[Σ(xᵢ − x̄)² / (n − 1)]

The (n − 1) denominator in the sample version is called Bessel’s correction. It compensates for the fact that a sample tends to systematically underestimate the true population variance. In real-world data analysis, you’re almost always working with samples — so (n − 1) is the version you reach for by default.

Step-by-Step Coefficient of Variation Examples

Example 1 — Basic Single Dataset

Dataset: {4, 8, 15, 16, 23, 42}

Step 1 — Find the mean: (4 + 8 + 15 + 16 + 23 + 42) / 6 = 108 / 6 = 18

Step 2 — Find the standard deviation (population): Deviations from mean: (−14, −10, −3, −2, 5, 24) Squared deviations: (196, 100, 9, 4, 25, 576) → Sum = 910 SD = √(910 / 6) ≈ 12.32

Step 3 — Calculate CV: CV = (12.32 / 18) × 100 = 68.4%

A CV of 68.4% indicates high variability relative to the mean. Whether that’s acceptable depends entirely on what this data represents — which is why benchmarks matter.

Example 2 — Comparing Two Datasets (The Actual Use Case)

To be honest, most statistical classes breeze through this topic without even explaining the real power behind the coefficient of variation. All they do is give you a formula and maybe go through one example before moving on to the next topic.

Now imagine a case from medical research. Take two biomarkers that can be measured in adult males within a certain range according to Endocrine Society guidelines:

Biomarker	Units	Mean	SD	CV
Testosterone	ng/dL	600	90	15.0%
Cortisol	μg/dL	18	2.1	11.7%

Different units. Different magnitudes entirely. But now you can make a meaningful comparison: cortisol levels are measured more consistently across this population than testosterone levels. That’s a real clinical statement — made possible only because CV removes the units from the comparison.

Example 3 — Investment Risk Comparison

In finance, CV works as a risk-efficiency ratio. An investor evaluating three assets:

Asset	Expected Return	Volatility (SD)	CV
Stock ABC	14%	10%	71.4%
S&P 500 ETF	13%	7%	53.8%
Government Bond	3%	2%	66.7%

The SD seems to be small, thus making the bond appear to be safe. However, its CV indicates otherwise. In effect, the risk taken for each $100 gain expected on investment is 66.7. The ETF offers high gains relative to the risks involved – which is exactly what CV highlights.

How to Calculate Coefficient of Variation in Excel

No Python required. If your data sits in cells A1 through A10, type this into any empty cell:

=STDEV(A1:A10)/AVERAGE(A1:A10)*100

That’s it. Excel handles the rest and gives you CV as a percentage.

For population CV — meaning you have the entire population, not a sample — swap STDEV for STDEVP:

=STDEVP(A1:A10)/AVERAGE(A1:A10)*100

Google Sheets uses the same syntax. Identical formula, identical output.

One thing to check before you run it: make sure your AVERAGE isn’t zero or negative. Excel won’t throw an error — it’ll just return a number that means nothing, or infinity. That’s not a bug; it’s a signal your data isn’t appropriate for CV in the first place. See the limitations section below for exactly when this becomes a problem.

Coefficient of Variation in Python

Two approaches, depending on what you need.

Option 1: NumPy — most control over population vs sample

python

import numpy as np

data = [4, 8, 15, 16, 23, 42]

# Population CV (default)

cv = (np.std(data) / np.mean(data)) * 100

print(f"CV: {cv:.2f}%")

# Output: CV: 68.39%

# Sample CV — add ddof=1 for Bessel's correction

cv_sample = (np.std(data, ddof=1) / np.mean(data)) * 100

print(f"Sample CV: {cv_sample:.2f}%")

# Output: Sample CV: 75.05%

Option 2: SciPy — cleaner one-liner

python

from scipy import stats

data = [4, 8, 15, 16, 23, 42]

cv = stats.variation(data) * 100  # Returns ratio by default; multiply by 100 for %

print(f"CV: {cv:.2f}%")

# Output: CV: 68.39%

scipy.stats.variation uses population SD internally. For sample SD, stick with the NumPy version and set ddof=1. Use SciPy when you’re building analysis pipelines and want cleaner code; use NumPy when you need to control exactly which SD formula runs underneath.

Interpreting the Coefficient of Variation — What Does Your Number Mean?

General CV Benchmarks

The number on its own means nothing without context. That said, these ranges work as first-pass guidance across most fields:

CV Range	What It Signals
Below 10%	Low variability — data is highly consistent
10–20%	Moderate variability — acceptable in most applications
Above 20%	High variability — significant spread relative to the mean
Above 100%	Extreme variability — SD exceeds the mean itself

Treat these as starting points, not hard rules. Because in practice, “good” CV means something completely different depending on where you work.

Industry-Specific CV Benchmarks

This is where generic thresholds break down. A CV of 15% is outstanding in behavioral research and alarming in clinical diagnostics. Here’s what the standards actually say, field by field.

Clinical Laboratories: Under 42 CFR Part 493 (CLIA – Clinical Laboratory Improvement Amendments) rules, there are defined maximum allowable error limits for each analyte, with ±7% of the assigned value or ±1 percentage point specified for HbA1c. In practical terms, it will be necessary to have a coefficient of variation well below 5%. If the coefficient of variation exceeds 15%, in repeatability tests, this is indicative that the procedure must be revised.

Manufacturing / Quality Control: Capability measures in Six Sigma assess the consistency of a process, using Cpk capability indexes; a frequently aimed-for Cpk of 1.33 reflects a process operating very narrowly, which translates into a CV less than 5% in precision-based production, although there is no universal value of CV set for that purpose. A process maintaining its consistency at 3% but shifting towards 8% reveals early indications of deterioration that usually occur many weeks before defective products come to light.

Finance: There is no universal cut-off point here. The lower the coefficient of variation, the better if the return expectation is positive. The S&P 500 index typically yields about 10%-11% per year and experiences a standard deviation of 15%-20% each year, meaning that its CV is about 150%-200% in annual terms, yet it drops to under 100% in rolling annualized 10-year terms due to reduced volatility. Individual stocks can experience CV levels of 200%-300%+ during times of high volatility.

Behavioral and Social Science Research: If the coefficient of variation is less than 20-25%, it can be taken to be an acceptable range for surveys and behavior-related data, as variations between humans is quite large in these cases. If it exceeds 30%, it is advisable to ensure that your sample population is indeed homogeneous.

Analytical Chemistry / Spectroscopy: Repeat measurements on the same sample under repeatability conditions should show CV below 2–5% for validated analytical methods.

When to Use Coefficient of Variation (and When Not To)

When CV Works Well

• Comparing variability across datasets with different units — the core use case
• Evaluating investment risk relative to expected return
• Quality benchmarking across production lines or lab methods
• Assessing consistency of measurement instruments across different scales

When CV Is Misleading or Invalid

Three situations where CV produces a wrong answer… and won’t warn you that it’s wrong.

1. Interval scales (Celsius, Fahrenheit, IQ scores)

Coefficients of variation can be calculated using only ratio-scaled measures, whereby the value of zero indicates an absolute absence of the object being measured. For example, temperature measured in Celsius and Fahrenheit degrees is a case of an interval measure, whereby the value of zero does not necessarily mean absolute absence.

The problem with the above approach: If we have a set of temperatures measured in °C, whose mean is 20 and standard deviation is 5, then CV = 25 percent. However, if these temperatures are expressed in °F with mean at 68 and standard deviation at 9, then CV = 13.2 percent. Therefore, the coefficient of variation is invalid when applied to interval measures.

2. Zero or near-zero mean

In case of a mean of zero, the denominator becomes zero, rendering the result undefined. In the event of a very small mean value, for instance 0.001, CV produces a huge number that makes no sense at all. The scenario can be found in cases like a net financial position, error term residual from an experiment, or data whose center is around zero.

3. Heavily skewed distributions

CV relies on the assumption that the mean value is a good measure for describing the central tendency of the data set. If the distribution is significantly skewed to the right, the mean is dragged towards the longer tail of the distribution, and therefore, makes the CV greater than what the average data point is going through.

Coefficient of Variation vs Standard Deviation — Which Should You Use?

Situation	Use SD	Use CV
Understanding absolute spread in one dataset	✅	Works but adds nothing
Comparing variability within same units	✅	Also works
Comparing variability across different units or scales	❌	✅
Data has zero or negative mean	✅	❌
Communicating to a non-technical audience	✅ Simpler	⚠️ Needs explanation
Evaluating investment risk-to-return ratio	❌	✅
Heavily skewed data	Consider IQR	Consider CQV

The practical rule: if you’re working within a single dataset and want to understand absolute variability, SD is clearer and more intuitive. The moment you need to compare across different scales — different units, different populations, different experiments — CV is what you use.

Applications of Coefficient of Variation

Finance and Investment: CV represents an efficiency ratio between risks. Two projects having the same risk and different rates of return will have different amounts of effective risk, and this difference can be shown using CV. Investment professionals can compare assets regardless of class through CV.

Manufacturing and Quality Control: An increase in CV is usually the first measureable sign that a trend has begun to occur in a stable manufacturing process. An increase in the CV is easier and less costly to detect than an increase in defects; this is economically advantageous.

Clinical Laboratories: Reliability in diagnostics hinges on reproducibility of the test. If the test lacks reproducibility (i.e., high CV in terms of repeatability), then regardless of the accuracy of the mean results obtained, it is unreliable in the clinical sense. The CV plays an important role in laboratory diagnostics.

Ecology and Environmental Science: Comparison of variability in animal populations between regions, rainfall during different seasons, or variation in pollutant concentrations between sample locations. Different species and locations have widely varying magnitudes — the coefficient of variation accounts for that.

Education Research:  A school whose average exam score is 90 and whose standard deviation is 9 (coefficient of variation = 10%) is relatively more stable in terms of performance than a school whose average score is 50 and whose standard deviation is 4 (coefficient of variation = 8%).

Frequently Asked Questions