ESD Test for Outliers
The Extreme Studentized Deviate (ESD) test is a generalized version of the Grubbs test that can detect multiple outliers in a dataset. Unlike the standard Grubbs test which only finds one outlier at a time, ESD iteratively tests and removes outliers until no more are found.
When to Use the ESD Test
- You suspect there may be more than one outlier in your data
- Your data is approximately normally distributed
- You want to identify extreme values systematically
- The standard Grubbs test is insufficient for multiple outliers
How It Works
- Calculate the mean and standard deviation of the data
- Find the value with the maximum absolute deviation from the mean
- Calculate the test statistic:
R = max|x - mean| / SD - Compare to critical value based on sample size and significance level
- If R > critical value, mark as outlier and remove it
- Repeat steps 1-5 up to the maximum number of outliers specified
- Stop when no more outliers are detected
Comparison with Grubbs Test
| Feature | Grubbs Test | ESD Test |
|---|---|---|
| Outliers detected | 1 only | Multiple |
| Method | Single test | Iterative |
| Best for | Single suspected outlier | Multiple suspected outliers |
Important Considerations
- Data should be approximately normally distributed (check with normality tests)
- Don't automatically remove outliers — investigate them first
- Consider whether outliers represent genuine extreme values or data entry errors
- Report how many outliers were removed and why
Null Hypothesis
H₀: There are no outliers in the data (all values come from the same normal distribution)