AI Glossary / Time series and signal processing / Beginner
Coefficient of variation
Definition
The coefficient of variation, or CV, is the standard deviation divided by the mean. It describes variability relative to the average level and is often expressed as a percentage. Its usual interpretation is most useful for positive measurements on a ratio scale with a meaningful zero and a mean sufficiently far from zero.
What it means
A standard deviation of ten units may be large for a process averaging twenty units and small for one averaging a thousand. The CV places the spread in relation to the average, making some comparisons across scales easier to interpret.
It is dimensionless when numerator and denominator use the same units. This does not mean it is appropriate for every numerical dataset. Changing units by multiplication preserves the ratio, but shifting the zero changes it. It is also a measure of observed relative dispersion, not automatically a measure of forecast accuracy, reliability, risk or uncertainty about the mean.
How it works
Calculate the mean and the chosen standard deviation, then divide the latter by the former. For a sample CV, a common choice uses the sample standard deviation computed with an n - 1 variance denominator. A population-style descriptive calculation uses n instead, so small-sample results can differ between software defaults.
Multiply the ratio by 100 to report a percentage. State the convention and data window. A zero mean makes the usual ratio undefined, while a near-zero mean can make it unstable. Negative means can produce negative numerical ratios in some implementations, but that output often lacks the usual relative-variability interpretation. Some reports use an absolute mean and call the result relative standard deviation; definitions should not be silently mixed.
Three practical examples
1. Process comparison: A quality team compares relative dispersion in two positive measurements with very different average magnitudes, while also retaining their absolute standard deviations.
2. Demand description: A planner summarizes variability across comparable product-demand histories. The planner checks for trends, seasonal effects and many zero periods before treating CV as a useful classification measure.
3. Data validation: An analyst reproduces a reported CV and checks whether the original calculation used sample or population standard deviation. The distinction can explain a discrepancy without any data error.
Think of it this way
Think of comparing the size of typical fluctuations with the size of the usual measurement. A two-unit fluctuation means something different around an average of ten than around an average of one hundred. The CV expresses that relative scale, while leaving the actual units and consequences for the decision to be considered separately.
Worked example
Take an original sample of three positive measurements: 8, 10 and 12. Their mean is 10. Squared deviations from the mean sum to 4 + 0 + 4 = 8. Using the sample variance denominator n - 1 = 2 gives variance 4 and standard deviation 2.
The sample CV is 2 / 10 = 0.20, or 20%. Now multiply every measurement by ten, producing 80, 100 and 120. The mean becomes 100 and sample standard deviation becomes 20, so the CV remains 20 / 100 = 20%.
Adding ten instead gives 18, 20 and 22. The standard deviation remains 2, but the mean becomes 20, so CV falls to 10%. This demonstrates why a meaningful zero matters. Using the population-style denominator on the original sample would give standard deviation sqrt(8/3), approximately 1.633, and CV approximately 16.33%. These are transparent descriptive calculations, not precise estimates from a large study.
Illustrative case study
Fictional case: Harbor Components compares two production measurements using CV and concludes that their relative variability is similar. A reviewer asks for the raw units and tolerance limits as well. One measurement operates much closer to an engineering limit, so the same relative dispersion has different operational implications.
The team keeps CV in the descriptive dashboard but adds specification-aware measures and time plots. It also standardizes the standard-deviation convention across reports. Managers now see both proportional variation and the consequences of that variation, rather than using one dimensionless number as a universal process-quality score.
Common misconceptions and mistakes
A CV of 20% does not mean that 20% of observations are wrong or outside limits. It is not a percentage forecast error. A smaller CV is not inherently better when a process's required range or purpose differs.
The ratio is inappropriate for arbitrary-zero scales such as Celsius temperature when interpreted as relative variability. Near-zero means can create extreme values even without extreme absolute spread. Outliers affect both the mean and standard deviation. In a time series, a high CV can reflect trend or seasonality rather than unpredictable noise, and it says nothing directly about serial dependence.
Business use
CV can help compare relative dispersion when measurements and context make the ratio meaningful. It is useful as one descriptive statistic alongside mean, standard deviation, sample size and a view of the data distribution.
Before using it to rank products or processes, check that the quantities share a comparable interpretation and that the observation windows are appropriate. Inspect zeros, outliers and changing levels. Do not use universal low, medium and high CV labels without justification for the application. A business decision should connect variability with costs, tolerances or service requirements rather than rely on a ratio alone.
People also ask
Is CV measured in the original units?
No. The units cancel when standard deviation and mean use the same scale.
Does changing meters to centimeters change CV?
Multiplying both mean and standard deviation by the same positive factor leaves it unchanged.
Does adding a constant leave CV unchanged?
No. It changes the mean but not the standard deviation.
Is a negative CV necessarily a software bug?
Not if the implementation divides by a negative mean, though the usual interpretation may be unsuitable.
Check your understanding
What is the sample CV of 8, 10 and 12?
Answer: It is 20% using the stated sample-standard-deviation convention.
Why does scaling all values by ten preserve CV?
Answer: Both numerator and denominator scale by ten.
Why is a near-zero mean problematic?
Answer: Small changes in the denominator can produce very large changes in the ratio.
Related terms
- Standard deviation
- Mean
- Ratio scale
- Dispersion
- Sample variance
- relative standard deviation
- forecast error
- seasonality
Learn next
Learn next how sample and population standard deviations differ, why measurement scales matter, and when robust dispersion measures are useful. For time-series comparisons, examine trend and seasonal structure before interpreting raw variability.
More in Time series and signal processing
- Box-Jenkins methodology - The Box-Jenkins methodology is an iterative approach to building time-series models, especially ARMA and ARIMA models, through identification, parameter...
- Box-Pierce test - The Box-Pierce test is a portmanteau test that assesses whether a sequence has autocorrelation across a chosen set of initial lags.
- Classical decomposition method - The classical decomposition method separates a time series into estimated trend-cycle, seasonal and remainder components using moving averages and seasonal...
- Collaborative filtering - Collaborative filtering is a recommendation approach that learns from patterns of interaction between users and items.
- Cumulative forecasting - Cumulative forecasting predicts the total amount of a variable over several future periods.
- Curve fitting - Curve fitting is the process of choosing parameters of a mathematical function so that the function approximates observed data according to a stated criterion.
Sources
NIST Dataplot, Coefficient of Variation: https://www.itl.nist.gov/div898/software/dataplot/refman2/auxillar/coefvari.htm . SciPy developers, variation: https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.variation.html . These primary technical sources explain the ratio, scale limitations and implementation conventions. The scaling and shifting examples are original and independently calculated.
Previous: Coefficient of multiple correlation Next: Cognition enhanced Natural language Information Analysis Method