About the Grubbs' Test Calculator
The Grubbs' Outlier Test Calculator checks whether the single most extreme value in a dataset is likely a genuine statistical outlier, assuming the rest of the data is approximately normally distributed. It is a formal, single-value alternative to looser rules of thumb like the 1.5xIQR method, producing a precise statistic you can compare against published critical value tables.
How It Works
Enter a list of at least three numbers. The calculator computes the mean and sample standard deviation of the full dataset, then finds whichever value deviates furthest from the mean and expresses that deviation in standard-deviation units as Grubbs' statistic, G. It also reports which specific value was flagged as the most extreme.
Formula & Methodology
To compute by hand: find the mean of all values, compute the sample standard deviation (dividing the sum of squared deviations by n-1), then find the absolute distance of each value from the mean. The largest of those distances, divided by the standard deviation, is G. Compare G to a published Grubbs' critical value for your sample size and chosen significance level - if G exceeds that critical value, the flagged point is a statistically significant outlier.
Examples
One clearly high reading
With the dataset 12, 14, 13, 15, 12, 14, 32, 13 (mean 15.625, sample standard deviation about 6.696), the value 32 has the largest deviation from the mean, 16.375, giving Grubbs' statistic G = 16.375 / 6.696, about 2.445.
Five equal readings and one spike
With the dataset 10, 10, 10, 10, 10, 20 (mean 11.667, sample standard deviation about 4.083), the value 20 deviates furthest from the mean, 8.333, giving G = 8.333 / 4.083, about 2.041.
Advantages
- Produces a precise, comparable statistic instead of a subjective visual judgment about which point looks unusual.
- Uses the dataset's own sample standard deviation, reflecting its actual spread rather than an assumed threshold.
- Identifies both the numeric value of G and the specific data point flagged, so you know exactly which observation to investigate further.
Common Mistakes
- Treating G on its own as proof of an outlier without comparing it to the appropriate Grubbs' critical value for the dataset's sample size and significance level.
- Running the test repeatedly to strip out multiple suspected outliers one at a time without adjusting for the fact that Grubbs' test is designed to detect one outlier at a time, which inflates the false-positive rate.
- Applying the test to data that is clearly skewed or non-normal, since Grubbs' test assumes the non-outlying data is approximately normally distributed.
Edge Cases to Watch For
- Requires at least 3 numbers, since the test needs enough data points to establish a meaningful mean and spread.
- Undefined when every value in the dataset is identical, since the standard deviation would be zero and G would require dividing by zero; the calculator returns an error in this case.
- The calculator reports G itself but does not look up a critical value automatically; you still need to compare G against a table for your sample size and significance level to reach a conclusion.
- Designed to flag only the single most extreme value; datasets suspected of having two or more outliers need a different, multi-outlier procedure.
Common Use Cases
- Laboratory scientists checking whether a single unusual measurement in a batch of repeated readings should be flagged before averaging results.
- Quality control analysts screening a small sample of measurements for one suspicious value before it is included in a report.
- Students and researchers learning formal outlier-detection methods as an alternative to eyeballing a boxplot or applying the IQR rule.