The comparison between reproducibility standard deviations from collaborative trials and proficiency tests: a preliminary study from food analysis

Reproducibility conditions of the replication of a measurement include several circumstances. In chemical measurement ‘reproducibility’ is mostly taken to refer to an interlaboratory study, either a collaborative trial (that is, with a strictly defined analytical procedure) or a proficiency test (with no prescribed procedure). At first sight, we might expect the reproducibility standard deviation of the proficiency test to be the greater for the same determination: the various procedures used by the participants will each introduce an extra uncertainty related to their specific biases. No comprehensive study of this potential disparity has been undertaken hitherto. The issue is important because reproducibility standard deviation is closely related to standard uncertainty. Here a comparison is made between the trend of collaborative trial outcomes (standard deviation as a function of concentration) and individual values from the FAPAS proficiency testing scheme in the food analysis sector. Contrary to expectations, the general tendency is for proficiency tests to provide slightly smaller standard deviations than do collaborative trials at mass fractions of the analyte greater than 10−7, and slightly higher at lower concentrations. However, there is considerable variation around the median level of the ratio at all mass fractions