THE PROFICIENCY TESTING OF LABORATORIES: A FIRST APPROACH TO IMPLEMENT BAYESIAN METHODS IN THE ASSESSMENT OF PERFORMANCE

摘要

Periodical intercomparison exercises have been organised by CIEMAT and CSN amonglaboratories providing data for environmental monitoring programmes in Spain. Since 1.985 different statistical methodologies have been applied for evaluation, with the result that the recommendations of the IUPAC protocol, the z-score, has shown to be the best working method to assess laboratory performance. The last revision of the IUPAC International protocol for proficiency testing of laboratories, recommends the assessment of laboratory performance against established criteria based on fitness for purpose criteria (the z-score). z velence (b-a)/(sigma)_(ffp) where a velence The best estimate of the measurand, b velence The participant's result, (sigma)_(ffp) velence The fitness-for-purpose based "standard deviation for proficiency assessment". This method provides objective means to assess the accuracy of laboratory results and assumes that participants perform in a manner consistent with the scheme's criteria, thus it does not take into account the participants' reported uncertainties. However a more precise assessment of performance should require a better estimation of the quality of the measurement (i.e.: when a participant has a fitness for purpose requirement inconsistent with that of the scheme). On the other hand, the u-score method of performance assessment takes into account the uncertainty of the laboratory result, but unreliable uncertainties reported by participants can disguise the scoring. u velence |b-a|/((sigma)~(2)+(tau)~(2))~(1/2) sigma velence uncertainty on the assigned value tau velence uncertainty on the laboratory result Although experience shows that uncertainty estimates are often incorrectly evaluated by laboratories, the Bayesian statistical methods allow the inclusion of such information by means of probability distribution functions, pdf.