A multiple regression analysis was performed to explain variation in the log number of naturalized plant species on islands. Predictor variables included history of occupation, time of European settlement/trade, latitude, log of island elevation, log of island area, and log of human population size (Table 5A.1). Several variables (as indicated above) were log-transformed to meet the assumption of normally distributed data. Stepwise analyses were performed by using the backward and forward procedures for adding and removing variables. The probability for a variable to enter the model was set at 0.250, and the probability for a variable to leave was set at 0.100. The best model constructed (judged by lowest AIC value and statistically significant predictor variables) has three predictors: history of occupation, log of elevation, and log of human population size; the model has an adjusted R2 value of 0.91. All statistical analyses were performed in JMP software, Version 5.0.1.
Data were compiled from the literature, using the criteria described above in Island Characteristics. Because sampling efforts varied among historic accounts of the flora, the number of exotics recorded in the Appendix are “range-through” data—such that species believed to be established at two points in time are recorded here as being established at all points of time between these. Therefore, if a survey in the 1800s and the modern flora both list a plant as established, then it was assumed to occur on the dates between these, even if it was not recorded on one of the interim dates. In most cases, this resulted in relatively minor alterations from the number of species recorded at any one point in time. This standardization allowed islands where only range-through data were available to be compared with those where all counts were independent. This procedure is particularly appropriate for large, topographically complex islands where individual species are easily missed on any one survey of an island. The one island considered here where this procedure may not have been necessary is Heron Island—a small (19-ha) island in the Great Barrier Reef. However, the differences on Heron Island between range-through and point-time data are relatively small, as indicated in the Appendix. Range-through data were not calculated for New Zealand; consequently, data from 1940 (when a range-through calculation is anticipated to make a substantial difference in recorded values) were not used in analyses of change in naturalized richness through time (see Appendix). Note that range-through and point-time data are always equivalent for the first and last time steps of any given island. The most recent (modern) publications used as data sources for islands are listed in Sax et al. (2002); older records