Grubbs and Dixon Q Outlier Test Calculator
Paste your replicate measurements to get the mean, standard deviation (SD), coefficient of variation (CV%, also called %RSD) and 95 % confidence interval, and to check for outliers with Grubbs' test (ISO 5725-2) or Dixon's Q test. Separate values with commas, spaces, semicolons or line breaks, and use a period as the decimal point.
Data distribution
Which statistic answers which question?
Mean (x̄): the central value that represents your replicates.
Standard deviation (s): the spread around the mean. The calculator uses the sample SD (divided by n − 1).
CV% = s / x̄ × 100: the relative spread, also reported as %RSD (relative standard deviation), especially in pharmaceutical work. Because it has no units, it lets you compare methods. As a rough guide: below 2 % excellent, below 5 % acceptable, above 10 % worth investigating.
SEM = s / √n: the standard error of the mean. It tells you how well you know the true mean.
95 % CI = x̄ ± t × SEM: the confidence interval for the mean, with Student's t for n − 1 degrees of freedom. If you repeated the whole experiment many times, 95 % of the intervals built this way would contain the true mean.
Outlier detection
An outlier is a value that lies unusually far from the rest of the data set. A statistical test only tells you that a value is suspicious; before removing it, always look for an assignable cause, such as a transcription error, an instrument fault or a contaminated sample.
G = max |xᵢ − x̄| / s
If G > Gcrit (α = 0.05, two-sided), the most extreme value is a statistically significant outlier. Grubbs' test assumes the data are approximately normally distributed and tests one suspect value at a time.
Q = |xsuspect − xnearest| / range
If Q > Qcrit, the extreme value is an outlier. Because it only uses the gap to the nearest value and the range, it suits the small sets of replicates that are typical in the lab.
Critical values used by the calculator
When n falls between two tabulated values, the calculator uses the next larger n in the table.
| n | Qcrit | n | Qcrit |
|---|---|---|---|
| 3 | 0.970 | 7 | 0.568 |
| 4 | 0.829 | 8 | 0.526 |
| 5 | 0.710 | 9 | 0.493 |
| 6 | 0.625 | 10 | 0.466 |
| n | Gcrit | n | Gcrit |
|---|---|---|---|
| 11 | 2.355 | 18 | 2.651 |
| 12 | 2.412 | 19 | 2.681 |
| 13 | 2.462 | 20 | 2.709 |
| 14 | 2.507 | 25 | 2.822 |
| 15 | 2.549 | 30 | 2.908 |
| 16 | 2.585 | 40 | 3.036 |
| 17 | 2.620 | 50 | 3.128 |
Frequently asked questions
Should I use Grubbs' test or Dixon's Q test?
Dixon's Q test only looks at the gap between the suspect value and its nearest neighbor, so it works with very small data sets (3 to 10 values). Grubbs' test uses the mean and the standard deviation of the whole set and is the test recommended by ISO 5725-2; it becomes more reliable as n grows. The calculator chooses automatically: Dixon's Q test for n ≤ 10 and Grubbs' test for n > 10.
Can I remove an outlier and recalculate?
Only if you can justify the cause, such as a recorded pipetting error or a documented instrument failure. Removing a value on statistical grounds alone, with no physical explanation, is not accepted for methods accredited to ISO/IEC 17025. If you remove a value, document it in the report.
Why divide by n − 1 instead of n?
Because your measurements are a sample (a subset) of all the measurements you could make, not the whole population. Dividing by n − 1 instead of n corrects the bias and gives a better estimate of the true variability of the method.
Related tools
LOD and LOQ CalculatorCalculate the limit of detection and limit of quantitation by the blank, calibration curve (ICH Q2) or signal-to-noise method.
This calculator is also available in Spanish.