Coefficient of Variance Calculator
Calculate the coefficient of variance (CV) to understand relative variability in your data. Enter your data points below to get instant results with visual analysis.
Introduction & Importance of Coefficient of Variance
Understanding relative variability in your data through coefficient of variance (CV) calculations
The coefficient of variance (CV), also known as relative standard deviation (RSD), is a standardized measure of dispersion of a probability distribution or frequency distribution. Unlike the standard deviation which measures absolute variability, the CV expresses the standard deviation as a percentage of the mean, making it particularly useful for comparing the degree of variation from one data series to another, even if the means are drastically different.
In statistical analysis, the CV is invaluable because it:
- Normalizes variability to account for differences in measurement units or scales
- Allows comparison of data sets with different means or units
- Provides insight into the precision of measurements relative to the mean
- Is commonly used in quality control, manufacturing processes, and scientific research
The formula for coefficient of variance is:
CV = (σ / μ) × 100%
Where:
σ = standard deviation
μ = mean (average) of the data set
Industries that frequently rely on CV calculations include:
- Pharmaceutical manufacturing (assessing consistency in drug formulations)
- Environmental science (comparing pollution levels across different locations)
- Finance (evaluating investment risk relative to expected returns)
- Biological research (analyzing variability in experimental results)
- Quality assurance (monitoring production consistency)
How to Use This Calculator
Step-by-step guide to calculating coefficient of variance with our interactive tool
Our coefficient of variance calculator is designed for both statistical professionals and beginners. Follow these steps for accurate results:
-
Enter Your Data:
- Input your numerical data points in the text area, separated by commas
- Example format: 12.5, 14.7, 16.2, 13.9, 15.4
- Minimum 2 data points required for calculation
- Maximum 1000 data points supported
-
Select Decimal Precision:
- Choose how many decimal places you want in your results (2-5)
- Higher precision (4-5 decimals) recommended for scientific applications
- 2-3 decimals typically sufficient for business applications
-
Calculate Results:
- Click the “Calculate CV” button
- Results appear instantly below the calculator
- Visual chart updates automatically to show data distribution
-
Interpret Your Results:
- CV values below 10% generally indicate low variability
- CV between 10-20% suggests moderate variability
- CV above 20% indicates high variability relative to the mean
- Compare your CV to industry standards for context
-
Advanced Features:
- Hover over chart elements for detailed values
- Use the “Copy Results” button to export your calculation
- Clear all data with the “Reset” button to start fresh
- Mobile-responsive design works on all devices
For large data sets, consider using our CSV import tool to upload your data directly from Excel or Google Sheets.
Formula & Methodology
Understanding the mathematical foundation behind coefficient of variance calculations
The coefficient of variance calculation involves several statistical steps. Let’s break down each component:
1. Calculating the Mean (μ)
The arithmetic mean represents the central tendency of your data set:
μ = (Σxᵢ) / n
Where:
Σxᵢ = sum of all data points
n = number of data points
2. Calculating the Standard Deviation (σ)
Standard deviation measures the absolute variability of your data:
σ = √[Σ(xᵢ – μ)² / (n – 1)]
For population standard deviation (when your data represents the entire population):
σ = √[Σ(xᵢ – μ)² / n]
Our calculator uses the sample standard deviation formula (with n-1 in the denominator) as this is more commonly appropriate for real-world data analysis where your sample represents a larger population.
3. Calculating the Coefficient of Variation
The final CV calculation combines these elements:
CV = (σ / μ) × 100%
Key mathematical properties of CV:
- CV is dimensionless (no units), allowing comparison across different measurements
- CV is always non-negative (CV ≥ 0)
- When μ = 0, CV is undefined (our calculator handles this edge case)
- CV is sensitive to small changes when the mean is close to zero
4. Interpretation Guidelines
| CV Range | Variability Level | Typical Interpretation | Common Applications |
|---|---|---|---|
| CV < 5% | Very Low | Exceptionally consistent data | Pharmaceutical manufacturing, precision engineering |
| 5% ≤ CV < 10% | Low | High consistency, minimal variation | Quality control, standardized testing |
| 10% ≤ CV < 20% | Moderate | Acceptable variation for many applications | Biological measurements, market research |
| 20% ≤ CV < 30% | High | Significant variation relative to mean | Early-stage research, exploratory data |
| CV ≥ 30% | Very High | Extreme variation, may indicate issues | Pilot studies, highly variable phenomena |
5. Mathematical Considerations
When working with CV calculations, be aware of these mathematical nuances:
- Mean Proximity to Zero: As the mean approaches zero, CV becomes increasingly sensitive to small changes in either the mean or standard deviation
- Negative Values: CV is always calculated using absolute values, even if some data points are negative
- Outliers: Extreme values can disproportionately affect CV calculations
- Data Distribution: CV assumes your data is approximately normally distributed for most interpretations
Real-World Examples
Practical applications of coefficient of variance across different industries
Let’s examine three detailed case studies demonstrating CV calculations in professional settings:
Example 1: Pharmaceutical Quality Control
Scenario: A pharmaceutical company tests the active ingredient concentration in 10 tablets from a production batch.
Data: 98.5 mg, 101.2 mg, 99.7 mg, 100.1 mg, 98.9 mg, 102.3 mg, 99.4 mg, 100.8 mg, 99.1 mg, 101.0 mg
Calculation:
- Mean (μ) = 100.1 mg
- Standard Deviation (σ) = 1.21 mg
- CV = (1.21 / 100.1) × 100% = 1.21%
Interpretation: The extremely low CV (1.21%) indicates excellent consistency in tablet production, well within the FDA’s typical requirement of CV < 6% for drug uniformity.
Example 2: Environmental Pollution Monitoring
Scenario: An environmental agency measures PM2.5 concentrations (μg/m³) at 8 monitoring stations across a city.
Data: 32, 45, 28, 51, 37, 42, 29, 48
Calculation:
- Mean (μ) = 39 μg/m³
- Standard Deviation (σ) = 8.9 μg/m³
- CV = (8.9 / 39) × 100% = 22.8%
Interpretation: The moderate-high CV (22.8%) suggests significant variation in pollution levels across the city, indicating potential “hot spots” that may require targeted intervention. This level of variation is common in environmental monitoring where geographic factors play a major role.
Example 3: Financial Investment Analysis
Scenario: An investor compares the annual returns of two mutual funds over 5 years.
| Year | Fund A Returns (%) | Fund B Returns (%) |
|---|---|---|
| 2018 | 8.2 | 12.5 |
| 2019 | 6.7 | 18.3 |
| 2020 | 5.1 | -2.7 |
| 2021 | 9.4 | 25.1 |
| 2022 | 7.8 | 8.9 |
Calculations:
- Fund A:
- Mean = 7.44%
- σ = 1.62%
- CV = 21.8%
- Fund B:
- Mean = 12.42%
- σ = 11.21%
- CV = 90.3%
Interpretation: Despite Fund B having higher average returns (12.42% vs 7.44%), its extremely high CV (90.3%) indicates much greater volatility. Fund A’s lower CV (21.8%) suggests more consistent performance, which may be preferable for risk-averse investors. This demonstrates how CV helps compare investments with different return profiles.
These examples show how CV provides actionable insights across completely different domains by normalizing variability relative to the mean value of each specific data set.
Data & Statistics
Comparative analysis of coefficient of variance across industries and applications
The following tables present comprehensive data on typical coefficient of variance ranges across various fields, based on published research and industry standards:
Table 1: Industry-Specific CV Benchmarks
| Industry/Application | Typical CV Range | Acceptable CV Threshold | Regulatory Standard (if applicable) | Key Influencing Factors |
|---|---|---|---|---|
| Pharmaceutical Tablet Weight | 1-3% | <5% | FDA, EMA <6% | Manufacturing precision, excipient uniformity |
| Clinical Laboratory Tests | 2-8% | <10% | CLIA ’88 standards | Instrument calibration, technician skill |
| Environmental Water Sampling | 10-25% | <30% | EPA Method Detection Limits | Sample heterogeneity, field conditions |
| Manufacturing Process Control | 3-12% | <15% | ISO 9001 quality standards | Equipment maintenance, raw material quality |
| Agricultural Field Trials | 15-35% | <40% | USDA guidelines | Soil variability, weather conditions |
| Financial Market Returns | 20-100%+ | Varies by asset class | SEC reporting requirements | Market volatility, economic conditions |
| Biological Assays | 5-20% | <25% | NIH guidelines | Sample preparation, assay sensitivity |
Table 2: CV Comparison by Sample Size
How coefficient of variance typically varies with different sample sizes (assuming normal distribution):
| Sample Size (n) | Expected CV Stability | Minimum Detectable Difference | Confidence in CV Estimate | Recommended Applications |
|---|---|---|---|---|
| n < 10 | High variability | Large (>20% of mean) | Low | Pilot studies, preliminary analysis |
| 10 ≤ n < 30 | Moderate variability | Medium (10-20% of mean) | Moderate | Exploratory research, process monitoring |
| 30 ≤ n < 100 | Good stability | Small (5-10% of mean) | High | Confirmatory studies, quality control |
| 100 ≤ n < 500 | Excellent stability | Very small (<5% of mean) | Very High | Regulatory submissions, large-scale analysis |
| n ≥ 500 | Optimal stability | Minimal (<2% of mean) | Extremely High | Population studies, big data analytics |
Key observations from these tables:
- Industries with strict quality requirements (pharmaceuticals, clinical labs) maintain very low CV thresholds
- Natural variability in environmental and agricultural data leads to higher acceptable CV ranges
- Financial data shows the widest CV ranges due to inherent market volatility
- CV stability improves significantly with larger sample sizes
- Sample sizes above 100 provide reliable CV estimates for most applications
When comparing CVs between groups, use NIST’s Engineering Statistics Handbook guidelines for determining statistical significance, especially for sample sizes under 30.
Expert Tips
Advanced techniques and best practices for coefficient of variance analysis
Master these professional techniques to maximize the value of your CV calculations:
1. Data Preparation Best Practices
-
Outlier Handling:
- Use the Tukey’s fence method (1.5×IQR) to identify potential outliers
- Consider winsorizing (capping outliers) rather than complete removal
- Always document outlier treatment in your analysis
-
Data Transformation:
- For right-skewed data, apply log transformation before CV calculation
- For percentage data, consider logit transformation
- Always back-transform results for interpretation
-
Sample Size Considerations:
- For CV comparisons, aim for equal sample sizes
- Use FDA guidance on sample sizes for bioanalytical methods
- Pilot studies should have at least 12 samples for meaningful CV estimation
2. Advanced Calculation Techniques
- Weighted CV: For stratified data, calculate CV within each stratum then combine using weighted averages based on stratum size
- Bootstrap CV: Use bootstrapping (1000+ resamples) to estimate confidence intervals for your CV when sample sizes are small
- Robust CV: Replace mean with median and standard deviation with MAD (Median Absolute Deviation) for outlier-resistant estimation
- Bayesian CV: Incorporate prior distributions for small sample sizes using Bayesian estimation techniques
3. Interpretation Nuances
-
Context Matters:
- Compare your CV to published benchmarks in your specific field
- Consider biological/technical variability inherent to your measurement system
- Account for measurement error (subtract from total CV if known)
-
CV vs. Standard Deviation:
- Use CV when comparing groups with different means
- Use standard deviation when absolute variability is more important
- Report both metrics for comprehensive data characterization
-
Temporal Analysis:
- Track CV over time to monitor process stability
- Use control charts with CV as the metric for quality monitoring
- Investigate sudden CV changes as potential process shifts
4. Visualization Techniques
- CV Heatmaps: Create heatmaps showing CV across multiple groups/samples for quick visual comparison
- CV vs. Mean Plots: Plot CV against mean values to identify patterns (e.g., CV often decreases with larger means)
- Forest Plots: Useful for displaying CV with confidence intervals across multiple studies
- Boxplots with CV: Overlay CV values on boxplots to show both absolute and relative variability
5. Software Implementation Tips
- Excel: Use =STDEV.S()/AVERAGE() for sample CV calculations
-
R:
cv <- sd(x)/mean(x)(considercv <- sd(x, na.rm=TRUE)/mean(x, na.rm=TRUE)for missing data) -
Python:
import numpy as np; cv = np.std(data)/np.mean(data) -
SAS: Use PROC MEANS with CV option:
proc means data=yourdata cv;
Never compare CVs across groups where some means are negative or close to zero. In such cases, consider using:
- Modified CV (using absolute values)
- Log-transformed CV
- Alternative metrics like IQR/median
Interactive FAQ
Common questions about coefficient of variance calculations answered by our experts
What’s the difference between coefficient of variance and standard deviation?
The key difference lies in how they express variability:
- Standard Deviation (σ):
- Measures absolute variability in the same units as your data
- Shows how much individual data points deviate from the mean
- Dependent on the scale of measurement
- Example: If measuring height in cm, σ would be in cm
- Coefficient of Variation (CV):
- Measures relative variability as a percentage of the mean
- Dimensionless (no units), allowing comparison across different scales
- Normalizes variability relative to the mean value
- Example: CV of 5% means the standard deviation is 5% of the mean
When to use each:
- Use standard deviation when you care about absolute variability in original units
- Use CV when comparing variability between groups with different means or units
- Use both together for complete data characterization
Can CV be greater than 100%? What does that mean?
Yes, CV can absolutely exceed 100%, and this conveys important information:
- Mathematical Interpretation:
- CV > 100% means the standard deviation is larger than the mean
- Occurs when σ > μ in the CV formula
- Mathematically valid but requires careful interpretation
- Practical Implications:
- Indicates extremely high variability relative to the mean
- Often seen when the mean is very small (close to zero)
- May suggest measurement issues or fundamental data problems
- Common Scenarios:
- Early-stage research with high measurement noise
- Data with many negative values or values near zero
- Phenomena with inherently high natural variability
- Small sample sizes that don’t represent the population
- What to Do:
- Verify your data for errors or outliers
- Consider data transformations (e.g., log transformation)
- Examine if the high CV is expected for your field
- Report the exact CV value rather than capping at 100%
Example: If measuring rare events where most values are 0 but some are large (e.g., 0, 0, 0, 0, 100), the CV would be very high, reflecting that the non-zero values are extremely variable relative to the small mean.
How does sample size affect coefficient of variance calculations?
Sample size has several important effects on CV calculations and interpretation:
1. CV Stability
- Small samples (n < 30) produce less stable CV estimates
- CV becomes more reliable as sample size increases
- For n ≥ 100, CV estimates are generally very stable
2. Mathematical Relationships
- The standard deviation in the CV formula becomes more accurate with larger n
- Sample mean converges to population mean as n increases (Law of Large Numbers)
- Confidence intervals around CV narrow with larger samples
3. Practical Considerations
| Sample Size | CV Stability | Minimum Detectable Difference | Recommended Use |
|---|---|---|---|
| n < 10 | Very unstable | >30% of CV | Pilot studies only |
| 10 ≤ n < 30 | Moderately stable | 15-30% of CV | Exploratory analysis |
| 30 ≤ n < 100 | Good stability | 10-15% of CV | Most applications |
| n ≥ 100 | Excellent stability | <5% of CV | Definitive analysis |
4. Special Cases
- Very Small Samples (n < 5):
- CV calculations are highly sensitive to individual data points
- Consider using range/mean ratio instead of CV
- Report individual data points alongside CV
- Very Large Samples (n > 1000):
- CV approaches the population parameter
- Small CV differences may be statistically significant
- Consider computational efficiency for calculations
5. Sample Size Determination
To determine appropriate sample size for CV estimation:
- Estimate expected CV from pilot data or literature
- Determine acceptable margin of error for your CV estimate
- Use power analysis to calculate required n
- For most applications, n ≥ 30 provides reasonable CV stability
What are the limitations of using coefficient of variance?
While CV is extremely useful, it has several important limitations to consider:
1. Mathematical Limitations
- Undefined for μ = 0: CV cannot be calculated when the mean is zero
- Sensitive to small means: CV becomes artificially large as the mean approaches zero
- Assumes ratio scale: Requires data where zero represents true absence of the quantity
2. Statistical Limitations
- Assumes normal distribution: CV interpretation relies on approximately normal data
- Affected by outliers: Extreme values can disproportionately influence CV
- Sample dependence: Small samples produce unstable CV estimates
3. Interpretational Challenges
- Context-dependent: “Good” vs “bad” CV depends entirely on the field
- Can be misleading: Low CV doesn’t always mean “good” (could indicate insufficient variability)
- Directional ambiguity: Doesn’t indicate whether variation is systematic or random
4. Practical Considerations
- Data requirements: Needs both mean and standard deviation calculations
- Computational sensitivity: Floating-point precision can affect very small/large CVs
- Communication challenges: Non-statisticians may misinterpret percentage values
5. When NOT to Use CV
- When your data includes negative values that aren’t just measurement noise
- When comparing groups with means very close to zero
- When the standard deviation and mean have different interpretations
- For ordinal data or data without a true zero point
6. Alternative Metrics
Consider these alternatives when CV isn’t appropriate:
| Scenario | Alternative Metric | When to Use |
|---|---|---|
| Data with negative values | Modified CV (using absolute values) | When negative values are meaningful |
| Mean near zero | Log-transformed CV | For ratio data with values near zero |
| Non-normal distributions | IQR/median ratio | For skewed or heavy-tailed distributions |
| Ordinal data | Coefficient of quartile variation | For ranked or categorical data |
| Small sample sizes | Range/mean ratio | When n < 10 and distribution unknown |
How can I reduce the coefficient of variance in my measurements?
Reducing CV requires addressing both technical and procedural sources of variability:
1. Technical Improvements
- Instrument Calibration:
- Follow manufacturer’s calibration schedule
- Use NIST-traceable standards when available
- Document all calibration activities
- Equipment Maintenance:
- Implement preventive maintenance schedules
- Replace worn components that may introduce variability
- Monitor equipment performance with control charts
- Measurement Protocol:
- Standardize all measurement procedures
- Use the same operator for critical measurements
- Implement automated measurements where possible
2. Procedural Enhancements
- Training:
- Provide comprehensive training for all personnel
- Implement certification for critical measurements
- Conduct regular competency assessments
- Standard Operating Procedures:
- Document every step of the measurement process
- Include visual aids and decision trees
- Update SOPs when processes change
- Environmental Controls:
- Control temperature, humidity, and other environmental factors
- Minimize vibrations and electrical interference
- Standardize lighting conditions for visual measurements
3. Statistical Approaches
- Increase Sample Size:
- More measurements reduce random variation
- Follow power analysis to determine optimal n
- Consider composite sampling for destructive testing
- Stratified Sampling:
- Divide population into homogeneous subgroups
- Calculate CV within each stratum
- Combine using weighted averages
- Outlier Management:
- Identify outliers using robust statistical methods
- Investigate root causes of outliers
- Consider winsorization for extreme values
4. Process Optimization
- Design of Experiments (DOE):
- Identify key factors affecting variability
- Optimize process parameters
- Implement statistical process control
- Material Standardization:
- Use consistent raw materials
- Implement supplier qualification programs
- Test incoming materials for consistency
- Automation:
- Automate repetitive measurement steps
- Implement robotic sampling where feasible
- Use LIMS (Laboratory Information Management Systems)
5. Continuous Improvement
- Implement Six Sigma or Lean methodologies
- Track CV over time using control charts
- Set progressive CV reduction targets
- Celebrate and share improvements across the organization
Not all variability is bad – some processes require inherent variability (e.g., biological systems). Focus on reducing unwanted variability that affects your specific goals.
Is there a relationship between coefficient of variance and confidence intervals?
Yes, coefficient of variance and confidence intervals are mathematically related through the standard error of the mean:
1. Fundamental Relationships
- Standard Error (SE):
- SE = σ/√n
- Where σ is the standard deviation from CV calculation
- n is the sample size
- Confidence Interval for Mean:
- CI = μ ± (t-critical × SE)
- t-critical depends on sample size and desired confidence level
- For large n, t-critical ≈ 1.96 for 95% CI
- CV Connection:
- Since CV = σ/μ, we can express SE in terms of CV
- SE = (CV × μ)/√n
- This shows how CV directly affects the width of confidence intervals
2. Practical Implications
- CV and CI Width:
- Higher CV leads to wider confidence intervals
- For a given n, doubling CV roughly doubles CI width
- Reducing CV by 50% can cut CI width by ~30%
- Sample Size Planning:
- Use pilot CV data to calculate required n for desired CI precision
- Formula: n = (t-critical × CV / margin of error)²
- Example: For CV=20%, 95% CI, margin=5%, need n≈62
- CV Confidence Intervals:
- CV itself has a confidence interval (not just the mean)
- Use Fieller’s theorem or bootstrap methods for CV CIs
- Typically wider than mean CIs due to ratio distribution
3. Interpretation Guidelines
| CV Range | Typical CI Width (relative to mean) | Implications | Recommended Action |
|---|---|---|---|
| CV < 10% | <5% of mean | Very precise estimates | Confident decision-making possible |
| 10% ≤ CV < 20% | 5-10% of mean | Moderate precision | May need larger samples for critical decisions |
| 20% ≤ CV < 30% | 10-15% of mean | Low precision | Investigate variability sources |
| CV ≥ 30% | >15% of mean | Very low precision | Significant process improvement needed |
4. Advanced Considerations
- Ratio Distribution:
- CV follows a ratio distribution (standard deviation/mean)
- Confidence intervals may be asymmetric
- Consider log transformation for symmetric CIs
- Bayesian Approaches:
- Incorporate prior information about CV
- Can provide more precise estimates with small n
- Useful when historical CV data is available
- Multivariate CV:
- Extend CV concept to multiple variables
- Use generalized variance or multivariate CV
- Requires advanced statistical software
When presenting results, always report:
- The point estimate of CV
- The confidence interval for CV
- The sample size used
- Any data transformations applied
This provides complete information for proper interpretation.
Can coefficient of variance be used for non-normal distributions?
While CV is most appropriate for approximately normal distributions, it can be used with non-normal data under certain conditions:
1. When CV Can Be Used
- Mild Skewness:
- CV remains reasonably valid for mildly skewed distributions
- Rule of thumb: |skewness| < 1 is generally acceptable
- Check with normality tests (Shapiro-Wilk, Anderson-Darling)
- Log-Normal Data:
- Common in biological, environmental, and financial data
- Apply log transformation, calculate CV, then back-transform
- Result is called the “geometric CV”
- Large Samples:
- Central Limit Theorem makes CV more robust with n > 30
- Sample CV approaches population CV as n increases
- Confidence intervals become more reliable
2. When to Avoid CV
- Severe Skewness:
- |skewness| > 2 makes CV interpretation problematic
- Mean may not represent central tendency well
- Consider median-based alternatives
- Bimodal/Multimodal:
- CV assumes unimodal distribution
- Multiple modes violate this assumption
- Consider mixture models or stratified analysis
- Heavy-Tailed Distributions:
- Extreme outliers disproportionately affect CV
- Standard deviation becomes less meaningful
- Use robust alternatives like MAD/median
- Discrete Data:
- CV assumes continuous data
- For count data, consider Poisson-based metrics
- For binary data, CV is not appropriate
3. Alternatives for Non-Normal Data
| Distribution Type | Alternative Metric | Formula | When to Use |
|---|---|---|---|
| Skewed (right) | Geometric CV | exp(σ_ln) – 1 (where σ_ln is SD of log data) | Log-normal or right-skewed data |
| Skewed (left) | Reciprocal CV | σ_(1/x)/mean(1/x) | Left-skewed or inverse Gaussian data |
| Heavy-tailed | Robust CV | MAD/median | Data with extreme outliers |
| Discrete counts | Dispersion Index | variance/mean | Poisson or negative binomial data |
| Ordinal | Coefficient of Quartile Variation | (Q3-Q1)/(Q3+Q1) | Ranked or Likert-scale data |
| Circular | Circular Variance | 1 – |mean resultant vector| | Angular or directional data |
4. Assessment Workflow
- Check Distribution:
- Create histogram and Q-Q plot
- Calculate skewness and kurtosis
- Perform normality tests (for n > 50)
- Evaluate CV Appropriateness:
- Is the distribution unimodal?
- Is |skewness| < 2?
- Are there extreme outliers?
- Consider Alternatives:
- If CV is inappropriate, select alternative metric
- Document rationale for metric choice
- Consider data transformation options
- Validate Results:
- Compare with non-parametric methods
- Check sensitivity to outliers
- Consult field-specific guidelines
5. Field-Specific Guidelines
- Clinical Chemistry:
- CLSI C28-A3 provides CV acceptability criteria
- Typically requires CV < 10% for most assays
- Higher CV may be acceptable for low-abundance analytes
- Environmental Science:
- EPA methods often specify maximum allowable CV
- Typically 20-30% CV acceptable for field measurements
- Lower CV (10-15%) expected for lab analyses
- Manufacturing:
- Six Sigma targets CV < 5% for critical processes
- Process capability indices (Cp, Cpk) incorporate CV concepts
- Control charts monitor CV over time
The NIST Engineering Statistics Handbook provides excellent guidance on when CV is appropriate and alternatives for non-normal data.