Outlier Test Calculator — Grubbs, Dixon Q, and Generalized ESD

🏭

Industry Use

Outlier Detection in Pharmaceutical QC

Outlier tests identify data points that are statistically inconsistent with the remainder of a dataset. They are used in pharmaceutical quality control to evaluate anomalous assay results, stability data, dissolution measurements, and analytical sequences. Per USP <1010>, outlier testing is recommended for validation and analytical method assessment.

Critical GMP principle: Statistical outlier rejection alone is never sufficient. Per 21 CFR Part 11 and GMP guidance, outliers must be excluded only after investigation confirms an assignable cause (e.g., instrument malfunction, analyst error, sample contamination). A result cannot be discarded solely because statistics say it's unlikely—the investigation must document why the outlier occurred.

Common applications: evaluating OOS results in batch release, detecting contaminated stability samples, assessing analytical method ruggedness, and identifying data entry errors. Three tests are offered: Grubbs (n>6), Dixon Q (n=3–25), and Generalized ESD (Rosner, for multiple outliers).

📐

Calculation Explanation

Three Statistical Outlier Tests

1. Grubbs Test (α=0.05, two-tailed): Best for n > 6. Tests whether a single point is an outlier.

Grubbs Test Statistic

G = max|xᵢ − X̄| / s

Compare G to Grubbs critical value table for n and α. If G exceeds critical value, the point is statistically significant outlier at α.

2. Dixon Q Test: Best for small samples (n=3–25). Simpler than Grubbs.

Dixon Q Statistic

Q = (suspect − nearest) / range

where range = max value − min value. Compare Q to Dixon Q critical table. If Q exceeds critical value, reject outlier.

3. Generalized ESD (Rosner): Detects multiple outliers iteratively. Removes one potential outlier at a time and recalculates limits.

USP <1010> Statistical Techniques ICH Q2(R1) Validation of Analytical Procedures 21 CFR Part 11, GMP Regulations

📋

Variable Definitions

Symbols and Descriptions

Symbol	Name	Description
n	Sample Size	Number of observations in dataset
xᵢ	Individual Value	Each data point (e.g., potency result)
X̄	Mean	Average of all observations
s	Standard Deviation	Measure of data spread
G	Grubbs Statistic	Distance of outlier from mean in SD units
Q	Dixon Q Statistic	Ratio of suspect gap to total range
Range	Data Range	max(xᵢ) − min(xᵢ)
α	Significance Level	Risk level for rejection (e.g., α=0.05 = 5%)
Critical Value	Test Threshold	Tabulated value. If test stat > critical, reject outlier.

📝

Step-by-Step Tutorial

Worked Example — Dissolution Data Outlier Detection

Scenario: Tablet dissolution test at 30 minutes shows n=8 results (%dissolved). One value (64%) appears anomalously low. Use Grubbs test (α=0.05) to assess whether it is statistically significant.
Data Collection:

Dissolution Results (%)
83, 85, 87, 86, 84, 85, 88, 64
Step 1: Calculate mean and SD.
X̄ = (83 + 85 + 87 + 86 + 84 + 85 + 88 + 64) / 8 = 662 / 8 = 82.75%

Deviations squared: 0.0625, 5.0625, 17.5625, 10.5625, 0.5625, 5.0625, 26.5625, 349.5625
Sum = 415.0 / 7 = 59.29
s = √59.29 = 7.70%
Step 2: Identify suspect outlier and calculate Grubbs statistic.
Suspect = 64% (lowest value)
|64 − 82.75| = 18.75
G = 18.75 / 7.70 = 2.43
Step 3: Compare to Grubbs critical value.
Critical G (n=8, α=0.05, two-tailed) = 2.126 (from table)

Is G > Critical? 2.43 > 2.126? YES
Step 4: Statistical Conclusion.
Result: The value 64% is statistically significant as an outlier at α=0.05 level.

IMPORTANT GMP NOTE: Statistical significance alone does NOT justify exclusion. An assignable cause investigation is REQUIRED.
Step 5: Required Assignable Cause Investigation (GMP Mandate).
Investigate: Was tablet 64 from same batch/production run? Check equipment logs for malfunction at time of test. Review analyst notes for procedural deviations. Confirm sample identity. If investigation documents a valid reason (e.g., "tablet physically broken during setup"), outlier may be excluded with documented justification. If no cause found, result must be reported as-is; process change may be warranted.
Final GMP Decision:
Statistical Test: OUTLIER DETECTED (G = 2.43 > 2.126)

GMP Requirement: Investigate and document assignable cause. Exclusion permitted only if cause validated. If no cause found, result stands and may trigger batch retest, root cause analysis, or equipment investigation. Never exclude outliers based on statistics alone.

→ Open Outlier Testing Calculator