In the framework of growing regulations regarding tobacco products, increased requirements are observed for reporting of analytical figures (e.g. FDA). This paper deals with the reporting of data variability in this context. The objective was to evaluate the consistency of the short term standard deviation to describe the variability of measurements as well as the effect of the number of replicates. The dataset of the CORESTA 2006 joint experiment included a number of cigarette smoke constituents identified by FDA for reporting. The testing protocol required the analysis of Kentucky reference cigarettes 2R4F and 1R5F performing 5 replicates (run over consecutive days) in 3 independent experiments (run within different time periods).

This data set provided different sources of variability across measurements: short term variability (replicates), medium term variability (periods) and lab-to-lab variability (laboratories). For each reference cigarette, ANOVA with one factor (laboratories) combined with Newman-Keuls’s multiple range tests was performed to compare data generated across laboratories. The distribution of variance between laboratories and the residual error (period and/or replicate) was estimated with different numbers of replicates (from 2 to 15). A hierarchical ANOVA on the 3 factors (laboratories, period and replicates) allowed the complete split of variance and evaluates their respective relative contributions to the full variability. Results show that the expression of variability as individual standard deviation (repeatability) gives false differentiations, whatever the number replicates, due to the major contribution of the lab-to-lab variability.