“…Due to the fact that subjects were nested within classes, schools, quarters, municipalities, regions as well as countries, logistic regression using multilevel procedures were used to analyze differences in ever and weekly smoking prevalence rates (De Vries et al, 1994, 2003bUS Department of Health and Human Services, 1994;Dijkstra et al, 1999). Non-significant predictors and interactions were deleted using a backward deletion procedure (a = 0.05), with the restriction that predictors were not removed from the model if they were involved as interaction terms.…”