Moderators affecting the dispersal-range size relationship
To investigate how differences between dispersal processes, evolutionary history, and study design affect the overall dispersal-range size relationship, we fitted a Generalized Linear Mixed Model (GLMM) with Binomial error distribution (link probit) and study ID as a random effect. To test the effect of the 17 moderators (Table 1) on the presence or absence of a dispersal-range size relationship, we performed a forward model selection procedure. We started by fitting a model with only random effects, and sequentially added significant fixed factors until reaching a final model. To decide which variable to include in the model in every time step, we tested for the significance of all fixed factors that can be added to the base model (i.e., all factors that are not already included in the model at that point) using a log-likelihood ratio test to identify the most significant variable to add (andp < 0.05). Each model was tested for collinearity using the vif function from the ‘car’ R package (Fox & Weisberg 2019). If a new added variable was collinear with one already selected in the model, the new variable was excluded from the selection procedure. We retained variables with a generalized variance inflation factor (GVIF) smaller than 2, which is the square root of the threshold value for the standard VIF (VIF = 5), and indicates limited collinearity. We used generalized linear mixed-effects models in a Bayesian setting (bglmer function from R package ‘blme’; Dorie et al. 2021).
We only included positive and neutral relationships as response variables, negative relationships were excluded as they were too few to perform an analytical test on (10 out of 478 relationships in total). All continuous variables (number of dispersal-related traits, biogeographical region size and number of species) were rescaled from 0 to 1. Also, 13 relationships for clades that were only assessed in a single study (mycorrhizal fungal, diatoms and liverworts) were excluded as they could not be placed in any of the two broad clades (animals, vascular plants) assessed here. 36 relationships were also excluded as did not report number of species and 13 were excluded as they missed information on highest taxonomic level. The final model included in total 410 relationships from 81 studies.
In addition to a global model including all relationships, we ran separate analyses for the two broad clades: animals and vascular plants. In animals, we also included the specific taxon (e.g., birds, mammals, fish, insects) as a factor related to the departure/transfer dispersal process, as these clades may differ in their dispersal ability based on certain biological features (e.g., presence or absence of wings). We then excluded invertebrates and trematodes, the former because it cannot be assigned to a specific taxonomic group and the latter, because it only has two reported relationship from a single study (Thieltges et al. 2011). Because temperature regulation and latitude were correlated, we ran two separated models including either temperature regulation or latitude. We recoded ‘tropical’ and ‘multiple latitude’ as ‘tropical-multiple latitude’, and ‘subtropical’ and ‘temperate’ as ‘subtropical-temperate’ for the model including latitude, due to complete separation. In plants, we excluded factors that lacked variation within plant studies, such as dispersal type (passive vs. active), temperature regulation, realm (only studies on terrestrial plants), species age (only one study included this), taxonomic unit, and the taxonomic breadth level ‘class’. Latitude was recoded similarly as done for animals. Model fitting and selection for both subsets was done as described above for the complete data-set.