Within Laboratory Variance Outlier Detection: An Alternative to Cochran's Test

• Activity Number: 82 - Computer Experiments, Statistical Engineering, and Applications in Physical Sciences
• Type: Contributed
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Michael Morton*
• Companies: Altria Client Services
• Keywords: Outlier detection; Homogeneity of variance; Levene's Test; Cochran's Test; Interlaboratory study; Collaborative study

Abstract

The interpretation of the results from analytical testing is often guided by the inherent variability of the associated test method. A common approach to characterize the method variability is through interlaboratory studies. One of the standard guides for the conduct and analysis of interlaboratory studies is ISO 5725-2: 1994 in which the use of Cochran's test is recommended to determine laboratories that have overly large intra-laboratory variation. However, Cochran's Test is very sensitive to non-normality, and a simulation analysis shows that the probability of falsely eliminating a laboratory as an outlier can easily be 30% instead of the nominal 1%, even for symmetric distributions that are very hard to analytically distinguish from a normal distribution. This false rejection of outliers has the potential to result in misleading underestimation of method variation and can reduce the data available for the estimation of lab-to-lab variation. An alternative approach, derived from Levene's Test, is much less sensitive to the normality assumption. This approach thereby has the potential to more accurately represent the variability within and between analytical laboratories.