Establishment patterns of exotic birds are constrained by non-random patterns in introduction

Aim: Species transported to and introduced into non-native environments, termed 'introductions', constitute a growing component of many regional floras and faunas, yet not all such species successfully establish. Investigations into why some introductions succeed while others fail are often based on analyses of the outcome of historical introductions. Such 'experiments in nature' have the drawback that the effects of interest may be confounded because they have not been randomized with respect to each other, and because the species and locations chosen for introduction may not be a random subset of the available species or locations. Our aim is to quantify this non-randomness using a global data set of bird introductions, and examine the factors associated with introduction probability in two subsets of these data, one taxonomic (order Anseriformes) and one geographical (British birds). Location: Global. Methods: Statistical analyses of the distributions among taxa and locations for 1378 introduction events for 426 bird species across the world, and statistical analyses of the characteristics of Anseriform and British bird species selected for introduction. Results: Global introductions of birds have been highly non-random with respect to taxon, location of origin, and location of introduction. Most introductions involve species in just five families (Phasianidae, Passeridae, Psittacidae, Anatidae and Columbidae), and most introductions have been to temperate and island locations. Within the taxonomic and geographical subsets, the species chosen for introduction tend to be abundant species that would have been relatively easy to obtain. Main conclusions: The characteristics of the species and locations chosen for introduction are not representative of species and locations in general, which limits our ability to draw general conclusions from historical records, and generates problems of confounding and non-independence in statistical analyses of introduction success. We suggest possible solutions for these problems.