Forward selection with Monte Carlo permutation testing was used to select the minimal set of bacteriological and environmental variables that contributed significantly and independently to explaining the variation in both FLP datasets (ter Braak and Smilauer, 1998).