2.4 | Statistical analyses
All statistical analyses and calculations were conducted in R (version
3.5.1, R Core Team, 2016). Tests for trait-environment relationships
were performed using single and multiple ordinary least-squares
regression models, with the average colour lightness and body size of
Odonata assemblages as dependent variables and climatic variables as
independent variables. Differences in the slopes of the relationships of
colour lightness and body size with climatic variables between lentic
and lotic habitats were determined by fitting interaction terms between
the independent variables and habitat type. In all models, independent
variables were scaled and centred (z-standardised) to facilitate their
comparison. To ensure very low mulitcollinearity among predictors, the
variance-inflation factors (vif) for the predictor variables in
regression models were checked using the vif function of the
R-package car (Table S2; Fox et al., 2016).
Since spatial autocorrelation in the survey data could violate the
assumptions of our statistical models, i.e., that all data points are
independent of each other, spatial correlograms of the model residuals
were calculated using functions of the R-package ncf (Bjornstad,
2016). These correlograms indicated significant spatial autocorrelation
in our data. Therefore, all analyses were repeated using spatial
autoregressive error models (Dormann, 2007) that included a spatial
distance weight according to the model-specific point of spatial
independence (extracted from spatial correlograms shown in Figs S1-S2).