Aspects of the design protocol and the statistical methods for analysis of tar, nicotine and carbon monoxide yields in cigarette smoke that can affect the measurement variability within collaborative studies.

Statistical principles described in ISO 5725-1 (1994) are a robust basis for evaluating cigarette smoke data from collaborative studies under the ISO 3308 machine smoking and for specifying the criteria for the removal of outlier data and determination of mean yields and their variability. However, the standard only provides recommendations on outlier removal that should be taken into account by experts who undertake data interpretation. The potential for over-interpretation of data from small numbers of laboratories is highlighted and recommendations made to deal with this possibility. Key variables to the statistical analysis, the number of cigarettes per replicate and replicates performed in each laboratory, the number of participating laboratories and the use of linear and rotary smoking machines in smoke collection, are identified and their relevance to obtaining robust data are considered. The statistical methods routinely used for data analysis from the ISO regime, are re-assessed for their suitability to analyse data obtained under the Canadian intense (CI) regime, where yield differences between linear and rotary smoking machines are found. This machine effect can lead to more outliers being detected and difficulties in outlier detection which may affect the provision of robust estimates of mean yields, repeatability and reproducibility.