The potential of multivariate quality control as a diagnostic tool in geoanalysis

Abstract

Univariate analytical quality control (UVQC) is now a mature discipline. By contrast, multivariate QC (MVQC) techniques have been relatively little used to date in industrial quality control and this is particularly true of analytical quality control. MVQC provides a new research area based on understanding the behaviour of the errors occurring in many analytes simultaneously. Such an approach has considerable potential for improving the criteria that are used to assess data quality in routine production of analyses by ICP-AES, ICP-MS, XRF and INAA. The results of an investigation of the potential of MVQC to control the characteristics of a 25 element dataset for an in-house reference material analysed by ICP-AES over a period of 6 years are described. It is suggested that a combination of Hotelling’s T² statistic and principal component analysis can provide powerful tools for both error recognition and subsequent diagnosis of probable cause. In this case, critical values of T² defining the conventional 95, 99 and 99.9% control limits were established (using computer simulation) to suit the fitness-for-purpose requirements of the analytical regime, which required that both element recovery during sample preparation and the relative importance attached by users to determinations of different elements had to be taken into account.