Fig. 6: Partial effects plot of the relationship between spatial distance and interaction dissimilarity (β_WN).

Here, we show the fit (solid line) of a Generalized Additive Model (GAM) with interaction dissimilarity as the response variable and all our predictor variables included. Thus, this plot shows the effect of spatial distance on interaction dissimilarity, while controlling for the effect of the other predictor variables in the model. Partial residuals remain on the same scale as the original data, but the sign of values indicates how they differ from what would be expected (i.e., from the fitted values) based on the other predictor variables in the model. The gray area represents two standard errors above and below the estimate of the smooth being plotted. The histogram above the plot shows the distribution of data points across the spatial gradient. Note the sharp increase in interaction dissimilarity until a threshold distance of around 2500 km (dotted red line), beyond which few networks shared interactions (a similar pattern can be seen in Supplementary Fig. 14c). Source data are provided as a Source Data file.