Long-term shifts in the functional composition of communities
After the taxa had obtained coordinates in the PCoA, the functional composition was described by the position of the community in the PCoA plot. The position of the community was calculated as a mean position of the taxa in the sample with respect to A1 and A2, weighted by their relative abundances. Long-term trends (explanatory variable year) were then visualized and quantified by Generalized Additive Model (GAM) smooth functions (family Gaussian, link identity), using the community coordinates A1 or A2 in May or August as response variables. Models were fitted to A1 and A2 separately, since it is more difficult to visualize and interpret the temporal trends on a 2D surface. To analyse the trends in A1 and A2 independently was also meaningful, because the distribution of trait values along A1 and A2 is straightforward (feeding type varies along A1, complexity and body size vary along A2).