Appendix A. Determination of minimal adequate models.
In the maximal model, containing all factors, interactions and covariates that might be of interest, the interactions were fitted by regressing the response variable against each covariate within each level of each factor. In the first step of model simplification, the different slopes of response variable on each covariate within each factor level were in turn replaced by a common slope of the response variable on each covariate. The common slopes were regressed on the factors one after another, and the changes in residual deviance caused by removal of the different slopes for each covariate were assessed. After all covariates with a common slope were assessed, all non-significantly different slopes were deleted, and the reduced model was assessed. The analysis then continued on the reduced model. In this model, all remaining terms were deleted in turn from the reduced deviance table, and only those leading to a significant (P < 0.05) increase in residual deviance were retained. The deletion tests were repeated on the reduced models until, after removal from the last deviance table, the minimal adequate model that contained nothing but significant terms was determined.
When evaluating the proportional representations in all three categories of weeds, a care was taken when interpreting proportional representation of native species. An apparent reverse trend for native species was observed just as a consequence of the opposite trends in the proportions of archaeophytes and neophytes, that were analyzed before native species.