Pesticide analysis in ground water. Statistical evaluation of certification data of a multicomponent reference material

Abstract

21 laboratories participated in the certification of a multicomponent ampouled reference material containing 24 pesticides. This paper describes the statistical analysis of the unbalanced data set returned by the participants in the certification. The simplest method of certification is achieved when laboratories report all the required data. In practice, however, with complex analytical methods such as multicomponent analyses of organic compounds, laboratories are not able to report all the required data. The response for each determinand in the material has been modelled in a three-factor nested model. For computational comfort the model has been rephrased into a general linear model. The statistical problems with unbalanced data are discussed in relation to ISO 5725 and ISO Guide 35. Tests for homogeneity are exact and uninfluenced by the unbalance in data. Unbalanced data cause dependencies between mean squares above the second lowest level of the model, thus exact hypothesis testing is not possible at higher levels. Approximate tests for variation between rounds and laboratories has been constructed by means of Satterthwaite’s formula [1]. A graphical plot of deviances from certified values has been applied to illustrate laboratory performance.