The coefficient of variation (CV) is the standard deviation divided by the mean, expressed as a percentage. It measures how big the spread is relative to the size of the thing being measured.

Because it has no units, the CV lets you compare variability between quantities measured on completely different scales, such as the consistency of a balance weighing in grams against a pipette dispensing in microlitres. A standard deviation of 5 means very different things on a mean of 10 and a mean of 1000; the CV tells them apart.

Rule of thumb: under 10% is low variability, 10 to 20% is moderate, 20 to 30% is high, and above 30% the data are very variable. Laboratory assays usually demand a CV under 15%.

The coefficient of variation answers a question the standard deviation cannot: is this spread large or small for numbers of this size? A standard deviation of 3 kg is enormous if you are weighing mice and trivial if you are weighing cattle. Dividing the standard deviation by the mean cancels the units, leaving a pure ratio that you can multiply by 100 and read as a percentage. That is the entire idea, and it is why the CV appears under different names in almost every applied field.

What this calculator reports for each group:

• CV (%), the coefficient of variation as a percentage, the headline number.
• CV (ratio), the same value before multiplying by 100, which some software reports instead.
• Mean and standard deviation, in your original units, so you can see what the CV was built from.
• Interpretation band, low, moderate, high or very high, based on the conventional thresholds.
• Confidence interval for the CV, an approximate range for the true population value.
• Signal-to-noise ratio, the inverse of the CV, used in engineering and imaging.
• Validity checks, flagging negative values, a mean near zero, or mixed signs, all of which make the CV unusable.

A worked one-liner: if a set of leaf measurements averages 52.44 mm with a standard deviation of 4.19 mm, the coefficient of variation is 4.19 / 52.44 = 0.08, or 8.0%, which counts as low variability.

Who uses it, and what they call it: analytical chemists and clinical laboratories call it the relative standard deviation and use it as the core precision metric for an assay. Haematologists meet it as red cell distribution width, which is literally the CV of red blood cell volume. Cardiologists compute the CV of RR intervals as a heart-rate-variability index. Finance uses it as risk per unit of return, where a lower CV means a better risk-adjusted investment. Biologists use it to compare morphological variability across species of very different body size.

4.1 What the CV percentage actually means

Read the coefficient of variation as "the typical deviation from the average is this percentage of the average". A CV of 8% says that a typical observation sits about 8% away from the mean. That reading works because both the standard deviation and the mean are in the same units, so dividing one by the other cancels them and leaves a pure proportion.

This is what makes the CV so useful for comparison. A standard deviation of 4.2 mm and a standard deviation of 4.2 grams cannot be compared at all. Their coefficients of variation can, because both are just percentages. The moment your question is "which of these is more consistent" rather than "how big is the spread", the CV is the statistic you want.

4.2 Reading the interpretation bands honestly

The conventional bands are under 10% low, 10 to 20% moderate, 20 to 30% high, and above 30% very high. These are rules of thumb, not laws, and they came from general-purpose applied statistics rather than from any particular theory. Every field has its own expectations. An analytical laboratory typically demands under 5% for a well-controlled assay and will reject anything above 15%. Biological measurements between individuals routinely sit at 20 to 40% and nobody is alarmed. Financial returns can produce CVs above 100% because the mean return is small relative to the volatility.

So use the band as a starting point and then apply the standard for your own field. The band selector in section 2 lets you switch between general, laboratory and finance thresholds so the verdict matches your context rather than a generic table.

4.3 Why the CV needs a true zero

This is the constraint that invalidates more CV calculations than anything else. The coefficient of variation is only meaningful on a ratio scale, meaning a scale where zero represents a genuine absence of the quantity and where doubling the number means twice as much of the thing. Length, mass, concentration, count, time duration and money all qualify.

Temperature in Celsius does not. Zero Celsius is the freezing point of water, an arbitrary reference, not an absence of heat. The same data expressed in Kelvin or Fahrenheit would give a completely different CV, which proves the number is not describing anything real about the data. The same problem applies to calendar years, pH, IQ scores, Likert ratings and any index with an arbitrary origin. If shifting your scale by a constant changes the answer, the CV is not valid for that variable.

4.4 The mean-near-zero problem

Because the mean sits in the denominator, the CV becomes unstable as the mean approaches zero and undefined when the mean is exactly zero. A tiny change in the mean then produces a wild change in the CV, so the statistic stops carrying information and starts amplifying noise. As a practical guide, if the mean is smaller than its own standard error, or if the confidence interval for the mean includes zero, do not report a CV at all. Report the standard deviation instead and say why.

Negative values create a related problem. If some observations are negative and some positive, the mean can be near zero while the data are genuinely spread out, and the CV will be meaningless or negative. If every value is negative, the CV comes out negative purely as an artefact of sign. This calculator flags all three situations rather than quietly returning a number.

4.5 CV, RSD and relative standard deviation are the same thing

Analytical chemistry, pharmacology and clinical laboratory science almost always say relative standard deviation or RSD, while statistics textbooks say coefficient of variation. They are the same calculation: standard deviation divided by the mean, times 100. If a method validation document specifies "RSD below 2%", it is specifying a CV below 2%. Some older chemistry literature also uses the term percent coefficient of variation, written %CV, which again is identical.

One genuine variant does exist. Some laboratories distinguish intra-assay CV, computed from replicate measurements within a single run, from inter-assay CV, computed across runs on different days. Both are ordinary coefficients of variation; what differs is which set of replicates you feed in.

4.6 What the confidence interval for the CV adds

The CV you compute is an estimate from a sample, so it carries uncertainty just like any other statistic. This calculator reports an approximate confidence interval built from the normal-theory variance of the CV. It tells you the range of population values consistent with your data.

Two cautions. The approximation assumes the underlying data are roughly normal and works best when the CV is below about 33%; beyond that the interval becomes unreliable and you should use a bootstrap instead. It is also sensitive to sample size: with fewer than about 20 observations the interval is wide enough that comparing two CVs is rarely conclusive. If two groups have overlapping intervals, resist the temptation to declare one more consistent than the other.

4.7 Comparing CVs across groups

Comparing coefficients of variation is the single most common reason people use this statistic, and it is where most of the mistakes happen. A lower CV means more consistent relative to its own mean, which is not the same as more consistent in absolute terms. A group with a CV of 5% around a mean of 1000 has a standard deviation of 50, while a group with a CV of 20% around a mean of 10 has a standard deviation of 2. The second group is far more variable relative to its size but far less variable in raw units. Both statements are true; be explicit about which one you mean.

Also remember that a difference in CV can come from a difference in the mean rather than a difference in the spread. If one group has a higher mean and identical scatter, its CV will be lower without anything about its consistency having improved. Always report the mean and standard deviation alongside the CV so the reader can see which component moved.

4.8 The inverse: signal-to-noise ratio

Turning the CV upside down gives the signal-to-noise ratio, mean divided by standard deviation. Engineering, imaging and instrumentation prefer this form because larger is better, which is more intuitive when you are tuning a system. A CV of 10% is a signal-to-noise ratio of 10; a CV of 50% is a ratio of 2. The two carry exactly the same information, so pick whichever direction your audience expects and stay consistent within a document.

4.9 What the CV cannot tell you

The coefficient of variation says nothing about the shape of the distribution. A symmetric dataset and a heavily skewed one can share the same CV. It says nothing about outliers, and because both the mean and the standard deviation are sensitive to extremes, a single wild value can inflate the CV substantially without the number revealing that anything unusual happened. It says nothing about sample size on its own, and it offers no test, so two CVs that look different are not thereby significantly different.

It is also not appropriate for data that have been standardised, log transformed or otherwise rescaled, because those transformations destroy the ratio-scale property that makes the CV meaningful in the first place.

4.10 Practical judgement

Use the CV when the question is genuinely about relative consistency and the variable has a true zero and a mean comfortably away from it. Report it alongside the mean, the standard deviation and the sample size, never on its own. State which field convention you used for the interpretation bands. And when the validity checks in section 3 flag a problem, believe them: a CV computed on Celsius temperatures or on a near-zero mean is not a slightly imperfect statistic, it is a number with no meaning attached.

▶ Run the analysis above to auto-fill all five examples with your results.

1. Enter your data. The default tab takes comma-separated numbers, exactly as the placeholder shows: 52, 48, 55, 61, 47, .... Newlines, tabs, semicolons and spaces all work, so a column pasted straight from Excel is fine.
2. Name each cluster. The group name field above every textarea is editable. Type something meaningful such as Assay run 1 and that name flows into the results table, all four charts and the exported report.
3. Add or remove clusters. Press Add cluster for a second, third or fourth group. Comparing groups is the main reason to use a CV, so this is where the tool earns its keep. Every column has its own Clear and Remove button, and Remove is disabled on the last remaining column.
4. Or upload a file. On the Upload tab, choose a CSV or Excel file and click the column names you want. Every column you click becomes its own cluster with its own CV, so a spreadsheet of six assay runs gives six coefficients of variation in one pass.
5. Try a sample dataset. Ten named datasets are built in, covering laboratory replicates, clinical measurements, finance, ecology and three deliberately invalid cases so you can see exactly what the validity warnings look like.
6. Choose sample or population standard deviation. Sample (n − 1) is correct unless you genuinely measured every member of the population. This is the setting that makes Excel's STDEV.S and STDEV.P disagree.
7. Pick your interpretation bands. General purpose thresholds suit most work, laboratory bands are stricter, and finance bands are far wider because returns are volatile relative to their mean.
8. Press Calculate Coefficient of Variation. Nothing is computed until you do, and changing a dataset or setting clears the results so you never read stale numbers.
9. Read the headline CV, then the validity panel. If your data contain negative values, a mean near zero, or mixed signs, the calculator will warn you before you build an argument on a meaningless number.
10. Check the four charts and export. Chart 1 places every group against the interpretation bands, which is usually the figure you want. Section 5 auto-fills five reporting styles, and the buttons under the charts export a text report or a print-ready PDF.

Excel has no CV() function, which is the reason this is one of the most searched spreadsheet questions in statistics. You build it from two functions Excel does have, STDEV.S and AVERAGE, and then format the result as a percentage. This section walks through it with a picture of the spreadsheet at each stage, then covers the percentage-formatting trap, doing many columns at once, and the equivalents in Google Sheets, R, Python and SPSS.

8.1 The functions you need

In Excel 2007 and earlier, STDEV.S is called STDEV and STDEV.P is called STDEVP. Everything else works identically.

8.2 Step by step, with the spreadsheet at every stage

8.3 The one-cell shortcut

Once the logic is clear, nest it and forget the intermediate rows.

To make it survive people adding rows later, point at the whole column. Both functions ignore the text header and blanks, so this is safe:

8.4 Several groups at once

The main reason to compute a CV is to compare groups, so this is the layout that matters. Write the formula once with relative references, then drag the fill handle sideways.

Type the formula in B8 only, then drag right. Run 2 is the most consistent at 5.96%, despite not having the largest mean.

8.5 Why your Excel answer might differ from R or Python

8.6 Charting the coefficient of variation in Excel

1. Lay out one row of group names and one row of CV values, as in section 8.4.
2. Select both rows and go to Insert → Charts → Clustered Column.
3. Right-click the vertical axis, choose Format Axis, and set the number format to 0.0"%" so the units are unmistakable.
4. To show the interpretation bands, add three more series holding the constant values 10, 20 and 30, then change each to a Line chart type via Change Chart Type → Combo.
5. Label the chart axis "Coefficient of variation (%)" and state in the caption which SD version you used.

Chart 1 in section 3 of this page does all of that automatically, including the shaded bands, so it is usually quicker to screenshot that than to rebuild it in Excel.

8.7 Excel errors you will probably hit

9.1 Interpretation bands by field

The same CV means very different things depending on what you are measuring. Pick the column that matches your work.

Conclusion: never quote a CV verdict without naming the standard you applied. A 12% CV is a good result in field ecology and a failing one in a validated bioanalytical assay.

9.2 Typical coefficients of variation in practice

Conclusion: use this table to sanity-check your own result. A CV of 40% on repeated balance weighings signals a broken instrument, while the same 40% on species counts is unremarkable.

9.3 CV and signal-to-noise ratio conversion

Conclusion: signal-to-noise is simply 100 divided by the CV percentage. Engineering audiences usually prefer it because larger is better.

9.4 When the coefficient of variation is not valid

Conclusion: the ratio-scale requirement is the one most often broken. If adding a constant to every value changes your CV, and it always does, then the zero point has to be real for the statistic to mean anything.

9.5 How sample size affects the reliability of a CV

Approximate 95% confidence interval width for a true CV of 20%, using the normal-theory approximation.

Conclusion: with fewer than about 30 observations the interval is too wide to distinguish a 15% CV from a 25% one, so do not rank groups by CV on small samples.

Study design: the coefficient of variation assumes a set of independent measurements of a single quantity on a ratio scale, where zero means a genuine absence of the thing measured.

1. Confirm the scale has a true zero before you collect anything. Length, mass, concentration, count, duration and money qualify. Celsius, pH, calendar year, IQ and Likert ratings do not. If your variable fails this test, plan to report the standard deviation instead and save yourself an invalid analysis.
2. Check the expected mean is well away from zero. If the quantity can plausibly average near zero, or take both positive and negative values, the CV will be unstable. Decide in advance what you will report if that happens.
3. Define the population and the replicate unit precisely. Be explicit about whether a replicate is a separate specimen, a separate aliquot of the same specimen, or a repeat reading of the same aliquot. These give very different CVs and they answer different questions.
4. Decide whether you are measuring within-run or between-run precision. Intra-assay replicates go in the same run; inter-assay replicates span days, operators or reagent lots. Label them clearly, because a reader cannot tell them apart from the number alone.
5. Fix the instrument, units and resolution. Record them in the protocol and never change them mid-study. Changing instruments inflates the CV for reasons that have nothing to do with the specimen.
6. Plan the sample size against the precision you need. Ten replicates give a very wide confidence interval for the CV; thirty is a realistic minimum if you intend to compare groups. See table 9.5 for the interval width at each n.
7. Write the acceptance criterion before collection. State the threshold, for example "intra-assay RSD must be below 15%", so the verdict is not chosen after seeing the result.
8. Set the outlier rule in advance. Both the mean and the standard deviation are sensitive to extremes, so a single bad replicate can move the CV substantially. Decide how you will handle it before you see the data.
9. Record raw data in one column per group, exactly as the layout table below shows, ready to paste or upload here.

One column per run or group, one row per replicate, numbers only below the header row. Blank cells are ignored, so unequal replicate counts are fine.

Pre-registration and ethics: record the acceptance threshold, the SD version (sample or population), the replicate definition and the outlier rule before you look at the data. Choosing the interpretation band after seeing the CV is one of the easiest ways to turn a failing result into a passing one without noticing.

Common collection mistakes:

• Mixing within-run and between-run replicates into one column, which produces a CV that describes neither.
• Computing a CV on a variable whose zero point is arbitrary, most often temperature.
• Reporting a CV for a group whose mean is near zero without flagging the instability.
• Changing operator, reagent lot or instrument mid-study, which inflates the CV for procedural reasons.
• Comparing CVs across groups measured at very different concentrations, where the mean rather than the precision drives the difference.

Use the coefficient of variation when the question is about consistency relative to size, rather than about spread in absolute units.

Use it when:

• ✓ Your variable is on a ratio scale with a genuine zero point.
• ✓ The mean is comfortably positive and well away from zero.
• ✓ You are comparing variability between groups measured in different units or at very different magnitudes.
• ✓ You are reporting assay precision, instrument repeatability or manufacturing consistency.
• ✓ You need a unit-free score that can be checked against an acceptance threshold.
• ✓ You are assessing risk per unit of return in finance.

Do not use it when:

• ✗ The scale has an arbitrary zero, such as Celsius, Fahrenheit, pH, calendar years or IQ. Report the standard deviation instead.
• ✗ The mean is at or near zero, where the CV becomes unstable or undefined.
• ✗ The data contain both positive and negative values, which makes the mean unreliable as a scaling factor.
• ✗ Your data are ordinal, such as Likert ratings. Use the median and interquartile range.
• ✗ The data have been log transformed or standardised, which destroys the ratio-scale property.
• ✗ You need a formal test of whether two variabilities differ. Use a modified signed-likelihood ratio test or a bootstrap, not a visual comparison of CVs.

Real-world examples:

1. Clinical laboratory, intra-assay and inter-assay RSD for method validation, benchmarked against a 15% limit.
2. Haematology, red cell distribution width, which is the CV of red blood cell volume reported on every full blood count.
3. Cardiology, the CV of RR intervals as a heart-rate-variability index that falls with age and illness.
4. Ecology and biology, comparing morphological variability between species of very different body size.
5. Agronomy, comparing yield consistency across plots and seasons.
6. Finance, risk per unit of return when comparing funds with different average returns.
7. Manufacturing, process capability monitoring where a rising CV signals drift before any single part fails specification.

Decision tree: ratio-scale variable → mean well away from zero → question is about relative consistency → use the coefficient of variation. If the scale has an arbitrary zero or the mean is near zero, use the standard deviation. If you want the precision of the mean rather than the spread of the data, use the standard error. If the data are skewed or ordinal, use the interquartile range.

Assumptions

1. The variable is measured on a ratio scale. Check whether zero means a genuine absence of the quantity. If it fails, the CV changes when you change units and is not a property of the data at all.
2. All values are positive. Check the minimum. Mixed signs make the mean unreliable as a scaling factor, and an all-negative dataset returns a negative CV that has no interpretation.
3. The mean is well away from zero. Check the mean against its own standard error. As the mean approaches zero the CV inflates without limit and stops carrying information.
4. Observations are independent. Check the sampling design. Repeated readings of the same aliquot are not independent replicates and will understate the true variability.
5. The sample is large enough. Check n. Below about 10 the CV is very unstable, and below about 30 the confidence interval is too wide for group comparison.
6. The data are roughly normal, for the confidence interval only. The point estimate needs no distributional assumption, but the interval reported here does. Check with a histogram or a normality test.

Limitations

• The CV is not robust. Both the mean and the standard deviation are sensitive to outliers, so one extreme value can move it substantially. For a robust alternative use the quartile coefficient of dispersion, (Q3 minus Q1) divided by (Q3 plus Q1).
• It says nothing about distribution shape. Symmetric and heavily skewed datasets can share the same CV.
• It provides no significance test. Two CVs that look different are not thereby statistically different; use a modified signed-likelihood ratio test or a bootstrap.
• The confidence interval is approximate. It assumes normality and degrades badly above a CV of roughly 33%. Use a bootstrap in that range.
• It is not comparable across transformed data. A CV computed on log-transformed values is not the CV of the original variable.
• For strongly log-normal data, the CV is better estimated as the square root of exp(s squared on the log scale) minus one, which this calculator does not compute.
• This tool assumes a simple unweighted sample. Weighted, stratified or clustered survey designs need a design-consistent variance estimator.

A coefficient of variation calculator answers a question the standard deviation is structurally incapable of answering: is this amount of spread large or small for numbers of this size? Dividing the standard deviation by the mean cancels the units and leaves a pure percentage, and that single move is what lets you put a balance measuring in milligrams and a pipette measuring in microlitres on the same axis and say honestly which is more precise.

The habit worth building is to always report the CV with its ingredients. A CV of 12% means nothing on its own, because it could come from a standard deviation of 1.2 on a mean of 10 or from 120 on a mean of 1000. Give the mean, the standard deviation, the sample size and the CV together, and your reader can see immediately which component drove the result. This also protects you from the most common misreading, where a group appears more consistent simply because its mean is larger.

The constraint to respect above all others is the ratio-scale requirement. If adding a constant to every value changes your answer, and for the CV it always does, then the zero point of your scale has to be real for the statistic to mean anything. Temperature in Celsius is the classic trap: the same weather gives you three different coefficients of variation in Celsius, Fahrenheit and Kelvin, which is proof enough that the number is not describing the weather. The validity panel in this tool checks for that, for a mean near zero, and for mixed signs, and when it raises a flag the right response is to report the standard deviation instead rather than to publish a number with no meaning attached.

Finally, treat the interpretation bands as conventions borrowed from a field rather than as facts. Under 10% is low in general applied work, under 5% is the expectation in a validated assay, and over 100% is unremarkable for financial returns. Name the standard you applied whenever you state a verdict, because the same 12% is a pass in one discipline and a failure in another.

Paste your numbers into the calculator above, check the validity panel before anything else, compare your groups on chart 1 against the bands for your field, and copy the reporting sentence that matches your journal. Then come back to this coefficient of variation calculator whenever you need to compare consistency across things that are not measured in the same units, because that is the one job it does better than any other statistic.

Cards marked Coming soon are not yet published, so they are shown without a link. Wire them up once each page goes live.

The following peer-reviewed references support the methods used in this coefficient of variation calculator, covering relative variability, assay precision, confidence intervals for the CV and best practice in reporting.

1. Reed, G. F., Lynn, F., & Meade, B. D. (2002). Use of coefficient of variation in assessing variability of quantitative assays. Clinical and Diagnostic Laboratory Immunology, 9(6), 1235–1239. https://doi.org/10.1128/cdli.9.6.1235-1239.2002
2. Vangel, M. G. (1996). Confidence intervals for a normal coefficient of variation. The American Statistician, 50(1), 21–26. https://doi.org/10.1080/00031305.1996.10473537
3. McKay, A. T. (1932). Distribution of the coefficient of variation and the extended t distribution. Journal of the Royal Statistical Society, 95(4), 695–698. https://doi.org/10.2307/2342041
4. Sokal, R. R., & Braumann, C. A. (1980). Significance tests for coefficients of variation and variability profiles. Systematic Zoology, 29(1), 50–66. https://doi.org/10.2307/2412626
5. Krishnamoorthy, K., & Lee, M. (2014). Improved tests for the equality of normal coefficients of variation. Computational Statistics, 29, 215–232. https://doi.org/10.1007/s00180-013-0445-2
6. Forkman, J. (2009). Estimator and tests for common coefficients of variation in normal distributions. Communications in Statistics, Theory and Methods, 38(2), 233–251. https://doi.org/10.1080/03610920802187448
7. Shechtman, O. (2013). The coefficient of variation as an index of measurement reliability. In Methods of Clinical Epidemiology (pp. 39–49). Springer. https://doi.org/10.1007/978-3-642-37131-8_4
8. Bland, J. M., & Altman, D. G. (1996). Statistics notes: Measurement error proportional to the mean. BMJ, 313(7049), 106. https://doi.org/10.1136/bmj.313.7049.106
9. Salkind, N. J. (2010). Coefficient of variation. In Encyclopedia of Research Design. SAGE. https://doi.org/10.4135/9781412961288
10. Abdi, H. (2010). Coefficient of variation. In N. J. Salkind (Ed.), Encyclopedia of Research Design (pp. 169–171). SAGE. https://personal.utdallas.edu/~herve/abdi-cv2010-pretty.pdf
11. Evans, S. J. W., Lachin, J. M., et al. (2011). Assay validation and the coefficient of variation in bioanalysis. Bioanalysis, 3(14), 1567–1575. https://doi.org/10.4155/bio.11.132
12. Salvagno, G. L., Sanchis-Gomar, F., Picanza, A., & Lippi, G. (2015). Red blood cell distribution width: A simple parameter with multiple clinical applications. Critical Reviews in Clinical Laboratory Sciences, 52(2), 86–105. https://doi.org/10.3109/10408363.2014.992064
13. Shaffer, F., & Ginsberg, J. P. (2017). An overview of heart rate variability metrics and norms. Frontiers in Public Health, 5, 258. https://doi.org/10.3389/fpubh.2017.00258
14. Pelaez-Coca, M. D., Hernando, A., et al. (2021). Statistical methods for heart rate variability: The role of normalised indices. Sensors, 21(6), 2131. https://doi.org/10.3390/s21062131
15. Albrecht, G. H., Gelvin, B. R., & Hartman, S. E. (1993). Ratios as a size adjustment in morphometrics. American Journal of Physical Anthropology, 91(4), 441–468. https://doi.org/10.1002/ajpa.1330910404
16. Hopkins, W. G. (2000). Measures of reliability in sports medicine and science. Sports Medicine, 30(1), 1–15. https://doi.org/10.2165/00007256-200030010-00001
17. Atkinson, G., & Nevill, A. M. (1998). Statistical methods for assessing measurement error (reliability) in variables relevant to sports medicine. Sports Medicine, 26(4), 217–238. https://doi.org/10.2165/00007256-199826040-00002
18. American Psychological Association. (2020). Publication manual of the American Psychological Association (7th ed.). https://doi.org/10.1037/0000165-000
19. NIST/SEMATECH. (2013). e-Handbook of statistical methods. National Institute of Standards and Technology. https://www.itl.nist.gov/div898/handbook/
20. R Core Team. (2024). R: A language and environment for statistical computing. R Foundation for Statistical Computing. https://www.R-project.org/