Calibration model averaging in chemical analysis: a case study for the method of standard additions

In this study we demonstrate that the method of standard additions can provide unreliable results in chemical analysis when the linear calibration model is forced to fit the data. A well designed standard addition experiment can still yield results biased by 10% when the analyst relies only on the linear model. Recently, the Joint Committee for Guides in Metrology (GUM-6) has emphasized the need to address the uncertainty inherent to the choice of measurement models and here we show how model averaging can provide a practical way to account for model uncertainty in the method of standard additions.

Automated summary

In this study, it is demonstrated that the method of standard additions can lead to unreliable results in chemical analysis when the linear calibration model is forced to fit the data. Even with a well-designed standard addition experiment, results can still be biased by 10% when relying solely on the linear model. The Joint Committee for Guides in Metrology (GUM-6) has recently stressed the importance of addressing uncertainty in measurement models, and this study shows how model averaging can offer a practical solution to account for model uncertainty in the method of standard additions.