Environmental Determinants of Island Species Number in the British Isles: A Reconsideration

Abstract

It is well known statistically that outlying values of independent variables contribute disproportionately to the fitting of a regression line. A previous examination of the environmental determinants of numbers of plant species in the British Isles includes one such outlier, Britain itself. Analyses identical to those previously performed are reported here, excluding Britain. Without the outlier, the prior conclusion that environmental heterogeneity contributes to floristic richness is unsupported. Introduction and methods Johnson & Simberloff (1974) used the technique of mixed multiple regression analysis to examine the relative importance of certain environmental variables in the determination of numbers of species of plants (S) present on the British Isles. This technique, first introduced for the analysis of species richness patterns on islands by Hamilton et al. (1963), employs variables in both transformed and untransformed modes. Johnson & Simberloff's (1974) results singled out number of soil types (ST) as the best predictor of S. Area (A), degrees North latitude (L), and distance from Britain (Di) entered their multiple regressions after ST was included in the model (Johnson & Simberloff, 1974, Tables 3 and 4). Although S, ST, and A were all highly intercorrelated, the authors inferred from their regressions that ST was the more proximal cause of the observed distribution of species on the islands since it was a better predictor of S than was A. Johnson & Simberloff (1974) considered ST to be a measure of environmental heterogeneity, and interpreted its higher simple correlation with S as demonstrating the importance of environmental heterogeneity in the determination of S. We note that one of the islands they considered, Britain, is greater in area than Lewis, the next largest, by more than two orders of magnitude (Johnson & Simberloff, 1974, Table 1), and is more than four standard deviations from the mean of the distribution of island areas (log X= 179; s.d.=088; N= 41) calculated with Britain removed. Britain is thus an outlying value of the independent variable A, and is disproportionately weighted in regression analyses. It is not clear whether Britain, the outlier, is a sample from the same population as the other islands, a necessary assumption in linear regression analysis. To determine this, we proceeded as recommended by Kruskal (1960), to reanalyse the data with Britain removed. If Britain is indeed drawn from the same population as the other islands, its removal should not significantly alter Johnson & Simberloff's (1974) findings. Conversely, if their results are markedly changed by the removal of Britain, then conclusions based upon these data are