Closed-form confidence intervals on measures of precision for an interlaboratory study

Area of Science:

• Pharmaceutical analysis
• Statistical modeling

Background:

• Analytical method transfer between laboratories is crucial in the pharmaceutical industry.
• Estimating method precision (repeatability, intermediate precision, reproducibility) relies on variance components from analysis of variance (ANOVA) models.
• Existing confidence interval methods (Welch, Satterthwaite, MLS) are typically compared for simpler ANOVA models.

Purpose of the Study:

• To compare the performance of Welch, Satterthwaite, and modified large-sample (MLS) methods for constructing confidence intervals on sums of variance components.
• To evaluate these methods specifically for a three-way classification ANOVA model commonly used in pharmaceutical method transfer studies.
• To assess method performance under conditions where variance components might be estimated negatively.

Main Methods:

• A simulation study was conducted to evaluate the coverage probability of different confidence interval methods.
• The study focused on a three-way classification ANOVA model relevant to interlaboratory method transfer.
• Comparisons were made using both standard moment estimates and adjusted estimates (set to zero if negative).

Main Results:

• The modified large-sample (MLS) method demonstrated superiority when using standard moment estimates.
• When using adjusted moment estimates, the Satterthwaite method performed comparably to MLS in many scenarios and better in specific settings.
• The Satterthwaite method consistently outperformed the Welch method, regardless of whether standard or adjusted estimates were used.

Conclusions:

• The Satterthwaite and MLS methods are recommended for constructing confidence intervals on precision measures in three-way classification ANOVA models for method transfer studies.
• Careful consideration of variance component estimation (standard vs. adjusted) is important for selecting the most appropriate confidence interval method.
• These findings enhance the statistical rigor of interlaboratory method validation in the pharmaceutical sector.