Estimation of measurement uncertainty by using regression techniques and spiked samples

Abstract

In this paper, we describe how to assess the trueness of analytical procedures using spiked samples and regression techniques. We show then how to calculate the uncertainty of future samples using the information generated in this process. Two types of bias are calculated in the assessment of trueness: a proportional bias (normally expressed in terms of recovery) and a constant bias. Since in many cases blank samples are not available, we propose a method for assessing trueness when samples that already contain the native analyte are spiked at several levels of concentration. Only proportional bias can be estimated from spiked samples, while constant bias must be estimated by the Youden method, i.e. by taking different amounts of sample. We propose two ways of assessing trueness. The first expresses results as instrumental responses and the second expresses results as concentration. We then present expressions for calculating uncertainty that cover these two situations. Finally, we use the expressions to calculate uncertainty from validation data of the analysis of esters in wine by SPME-GC.