geographic location, that confer large species ranges also help make species bulletproof (although geographical variation in species’ range sizes again complicates separation of geographical variability in threat intensity from intrinsic biological vulnerability).
Large-scale analyses can find general predictors of extinction risk but can miss interesting variation among regions or clades, which more narrowly focused models might pick up (Fisher and Owens, 2004). Order-specific models typically have higher explanatory power than the large-scale models. These models have some common features, such as the importance of geographic range size, but also differ considerably (Cardillo et al., 2008). For example, body size is a predictor in bats but not in rodents, whereas different life history traits predict risk in carnivores (gestation length) and ungulates (weaning age, interlitter interval). Likewise, different environmental factors and measures of human impact are implicated in different taxa. The models also vary regionally, with life history mattering less in north temperate regions than elsewhere (Cardillo et al., 2008).
One likely source of variation in models is that different drivers may select against different characteristics and show spatial variation in intensity. Broad-scale analyses may therefore lump competing signals together (Owens and Bennett, 2000). Within artiodactyls, predictors of extinction risk differ between hunted and nonhunted species: Late weaning age was the sole risk factor among the former, whereas low income levels among local people and small range size predicted risk among the latter (Price and Gittleman, 2007). More generally, low reproductive rates and large size are likely risk factors for overexploitation, but a specialized habitat may matter more under habitat loss (Owens and Bennett, 2000). Analyses focused more tightly on driver-specific responses often tend to consider far fewer species, in which case far fewer predictor variables can be considered simultaneously without overfitting, and statistical power may be lower. On the plus side, the tighter focus can reduce the chance of mixed signals [although interaction terms can also do this (Price and Gittleman, 2007)], and more precise measures of driver intensity and extinction risk might be available than can be had globally (Fisher et al., 2003; Isaac and Cowlishaw, 2004). Broad- and narrow-scale analyses each give part of an obviously very complex picture. Furthermore, we have focused on phylogenetic nonindependence, but to fully consider the interaction between biology and geography, the development of methods that also deal with spatial nonindependence in comparative data will be critical.
Analyses modeling risk as a function of intrinsic biology (i.e., not including driver information) can highlight species at lower risk than expected from their geography, ecology, and life history (Cardillo et al., 2006). Such species may be particularly likely to decline rapidly if drivers intensify, because their attributes are repeatedly found in rapidly declining