Explanation of outlier analysis - remarks based on the evaluation of the WG10 study
In 2010 a collaborative study was undertaken to investigate the intra- and inter-laboratory variability of the determination of mainstream smoke NFDPM, nicotine and carbon monoxide yields from a range of commercial cigarettes, a reference cigarette and the CORESTA CM6 Monitor Test Piece. These test articles were smoked in accordance with ISO and Health Canada Intense (HCI) methods (the study with its main results is described in the presentation of D. Mariner et al.). In the statistical analysis of this study we proceeded according to ISO 5725 in general. Concerning the outlier analyses different strategies can be applied for this large and structured data set. These strategies will be explained and described in their effects on final estimates like the intra- and inter-laboratory variability. In addition to the "standard" analysis of a data set statistical simulation experiments will underline possible effects of outlier analysis when certain data structures are given (e.g. bimodal distributions).