Location: Food Quality LaboratoryTitle: Evaluation of a collaborative study through outlier detection and calculation of precision figures of merit: SAS code
Submitted to: AACC International
Publication Type: Other
Publication Acceptance Date: 10/1/2018
Publication Date: 10/1/2018
Citation: Delwiche, S.R. 2018. Evaluation of a collaborative study through outlier detection and calculation of precision figures of merit: SAS code. AACC International. https://methods.aaccnet.org.
Technical Abstract: A computer program was written in SAS to evaluate the results of a collaborative study, also known as a ring test. In such studies, blindly duplicated samples are quantitatively evaluated for one or more analytes (e.g., concentration, physical property) by several participated laboratories using an agreed upon protocol. Data from the laboratories must first be compiled into one spreadsheet. Upon reading in the data, the program proceeds with a Cochran outlier test (for identifying large discrepancies between blind duplicates), a ranks test (for identifying laboratories that are habitually low or high compared to the rest of the laboratories across all samples), a single-Grubbs test (for identifying a laboratory's single sample that is low or high compared to the other laboratories), and a paired-Grubbs test (for identifying a laboratory's pair of samples that are unusually both low, both high, or split between being low for one member of the pair and high for the other member). The program then calculates, by ANOVA, the repeatability (s_r, 2.8 x s_r and RSD_r) and reproducibility (s_R, 2.8 x s_R and RSD_R) precision statistics as defined by IUPAC guidelines and further refined by AOAC International. The program, while submitted in batch, is structured to be run iteratively in order to flag outliers, whereupon after each run the user must decide to keep or remove the identified outliers.