Data analysis
All data manipulations and statistical analyses were performed in R version 4.1.3 (R Core Team, 2021). Every analysis was carried out for two groups: for the total ruderals and for the non-native ruderals. Using the historical Rallarvägen surveys, we analyzed 1) region-wide species richness as a function of the observational year, and 2) the mean EIV-T as a function of the observational year and first year of observation. Using the 2021 Rallarvägen survey, we analyzed 3) species richness as a function of distance to the railroad, distance to the E10, and soil temperature variables and 4) the Z -score abundances as a function of the ruderal species richness. Finally, using the MIREN trail survey, we modeled 5) species richness as a function of elevation and 6) the elevational maximum as a function of the first year of observation.
Models consisting of a dependent variable with count data (number of species) were analyzed using generalized linear models (functionglm , poisson or quasipoisson distribution), otherwise linear models were used (function lm ). We identified the best fitting models using the Akaike Information Criterion with a correction for smaller sample sizes (AICc) from the AICcmodavg package (Mazerolle, 2020). For significant interactions consisting of two continuous variables, we centered one independent variable at its sample mean to make interpretation easier (Schielzeth, 2010). In multiple regression analyses, we checked for possible multicollinearity of independent variables by calculating the variance inflation factor (vif) using the vif function from the car package (Fox & Weisberg, 2011). We considered results to be significant when p ≤ .05 and marginally significant when p < .10.
To visualize the dissimilarities in vegetation composition between subregions and observational years, we conducted a Principal Coordinates Analysis (PCoA, = Multidimensional scaling, MDS). PCoA is an ordination technique to explore and visualize dissimilarities in species composition data by focusing on distances. The more similar the compositions are, the closer together they occur in the plot. Distances were calculated with the function vegdist from the Vegan package (Oksanen, 2022), and from this distance matrix the principal coordinate scaling was computed with the pcoa function from the ape package (Paradis, 2022). We used the Jaccard distance which is defined as:Jaccard distance = 2B/(1+B) , where B is the Bray-Curtis dissimilarity. Bray-Curtis dissimilarity usually focuses on the dissimilarity of abundance, but by specifying binary = TRUE in the function it calculates distances based on presence-absence data. The obtained dissimilarity is a number between 0 and 1 – this value is 0 when two communities share all the same species, and 1 when they do not share any species.