TSRC, Tob. Sci. Res. Conf., 2018, 72, abstr. 045

A streamlined means of outlier detection in collaborative and proficiency studies

SLONE S.; SHELTON B.; SHEARER A.; CANETE S.
University of Kentucky, Lexington, KY, USA

Regardless of the industry, one of the first steps in analyzing data from a collaborative or a proficiency study is to detect any outliers in the data. Guidelines published by the International Organization for Standardization, ISO, enumerate several tests to consider. To identify outlying means in the laboratories participating in proficiency study, the guidelines suggest the single Grubb’s test and the Mandel’s h test. Similarly, the guidelines propose using the Cochran’s test and the Mandel’s k test to identify outlying standard deviations. While the guidelines present a partial table of significant values for these four outlier tests, a programmatic solution to find the scores and critical values leads to savings in time and effort. In his 2013 paper in AStA Advances in Statistical Analysis, Wilrich derived the formulas for the critical values for each of the aforementioned tests. A SAS® macro was written to perform these tests on outliers for data from the Center for Tobacco Reference Products (CTRP) at the University of Kentucky. The CTRP utilizes this macro for proficiency testing and for characterization/certification of constituents in tobacco reference products as part of an FDA funded Cooperative Agreement. Since the macro is written for flexibility, a user can choose the tests to be performed as well as various options for printing which include significant results for the Grubbs or Cochran test, graphs for the Mandel’s h and k, with reference lines at the 95% and 99% critical values, and a summary table of results. Details on and sample runs of the macro will be presented. The macro has been updated to handle missing data.