Globally, elevated extinction risk in mammals is strongly associated with large body size. However, in regions where introduced predators exert strong top-down pressure on mammal populations, the selectivity of extinctions may be skewed towards species of intermediate body size, leading to a hump-shaped relationship between size and extinction risk. The existence of this kind of extinction pattern, and its link to predation, has been contentious and difficult to demonstrate. Here, we test the hypothesis of a hump-shaped body size–extinction relationship, using a database of 927 island mammal populations. We show that the size-selectivity of extinctions on many islands has exceeded that expected under null models. On islands with introduced predators, extinctions are biased towards intermediate body sizes, but this bias does not occur on islands without predators. Hence, on islands with a large-bodied mammal fauna, predators are selectively culling species from the lower end of the size distribution, and on islands with a small-bodied fauna they are culling species from the upper end. These findings suggest that it will be difficult to use predictable generalizations about extinction patterns, such as a positive body size–extinction risk association, to anticipate future species declines and plan conservation strategies accordingly.
The biological selectivity of species decline and extinction offers clues to the relative importance of different threatening processes. Recognizing predictable patterns of extinction, or particular traits that make species especially vulnerable to different threats, presents the possibility of identifying species at potential future risk and planning conservation strategies pre-emptively . In mammals, one of the strongest patterns of biological selectivity involves body size: in most comparative analyses of mammal extinction risk, larger size is associated with higher risk [2–5]. An obvious explanation for this association is the negative scaling of life-history speed with size, leading to lower rates of population growth and recovery , and lower population densities [6,7], in larger species. Furthermore, there appear to be synergistic effects whereby larger species are more sensitive to a given threatening factor than small species .
In this context, any exceptions to the expected positive size–risk association are worthy of investigation, for two main reasons. First, such exceptions would undermine the generality of the positive size–risk association and hence its usefulness for pre-emptive conservation planning. Second, such exceptions may point to agents of species decline that exert an unusually strong pressure on mammal populations in the lower or intermediate parts of the body size distribution. It has been suggested that Australian mammals represent a major exception to the positive size–risk relationship. Most extinctions and severe declines in Australian mammals have been within an intermediate body mass range of 35–5500 g, known as the ‘critical weight range’ . This corresponds broadly to the preferred prey-size range of feral cats and red foxes, which are probably responsible for catastrophic declines and extinctions among Australian mammals . An energetic explanation has also been offered, which argues that medium-sized species have the combined disadvantage of higher energy needs than small species and lower mobility than large species [8–10].
However, the existence of a hump-shaped association between size and extinction risk in Australian mammals has been contentious. Cardillo & Bromham  showed that the numbers of extinct and threatened species of 35–5500 g are no greater than expected under null models and supported instead a positive size–risk association. Other studies have supported the existence of a hump-shaped pattern within restricted geographical and ecological subsets of Australian mammals [12–14]. Whether introduced predators drive such patterns, however, is difficult to determine. While local eradications of cats and foxes have been linked to increases in populations of medium-sized mammals, the role of predators in driving a hump-shaped size–risk association across whole assemblages has only been inferred from the overlap (or lack thereof) between the distributions of mammal species and introduced predators across broad geographical regions [12,13].
We use a large database of island mammal populations to test the hypothesis that introduced predators bias extinctions towards intermediate body sizes. Islands provide an excellent system for testing this hypothesis, because they represent, in effect, a large set of independent replicate experiments in the exposure of an indigenous mammal fauna to exotic predators. If predators exert sufficient pressure on mammal populations to generate a hump-shaped size–risk pattern, then on islands with predators we should expect the size distribution of extinct species to cluster around an intermediate body size within the typical prey-size range of the predators. On islands lacking predators, a plausible alternative pattern is that extinctions cluster around body sizes towards the upper end of the body size distribution, as expected under the typical positive association between size and risk. On all islands, the null expectation is that the body size distributions of extinct species are indistinguishable from those of the same number of species sampled randomly from the original fauna of each island.
2. Material and methods
Our database includes records of the presence and extinction of 927 populations of 106 native non-volant mammal species, on 321 Australian islands, together with data on the presence of introduced exotic species, species-average body masses and island geographical data. A population is defined as a given species on a given island. Where island-specific body mass data were available, these were used in preference to species values . The introduced predators we consider are cats, foxes and dingoes. It is difficult to distinguish the ecological role of dingoes, established in Australia for several thousand years, from feral domestic dogs, so we consider both as ‘dingoes’. Full details of database construction are provided in , and the database itself is provided in the electronic supplementary material, table S1.
(b) Testing size-selectivity and size-bias of island extinctions
For each island with at least one extinct and one extant native mammal species, we calculated a test statistic de that quantifies the degree of dispersion of extinct species body masses (w) around a given body mass value (m) where n is the total number of extinct species. For each island, we found the value of m that minimized de, using the ‘optimize’ function in R to search m values across the interval 10–20 000 g. We generated a null model for m by repeating this procedure for 1000 sets of n species sampled from the original mammal fauna of the island (i.e. extant + extinct indigenous species). Extinctions were considered size-selective if the observed m deviated significantly from the null distribution under a two-tailed test with α = 0.05.
We then examined the direction of the extinction size bias with respect to the body size distribution of the original fauna on each island. We calculated a standardized effect size for the extinction size bias as (observed m − median null m)/s.d. of null m. Plotting the size-bias effect size against the mean body size of island faunas reveals patterns of size-selective extinction (figure 1). If extinctions are biased towards intermediate sizes, there should be a negative relationship in which the slope crosses zero on the y-axis (figure 1b).
Of the 321 islands in the database, 59 have suffered mammal extinctions, of which 43 have at least one extant indigenous species. Of these 43 islands, we found evidence for size-selective extinction (observed m ≠ null) on 20 islands (electronic supplementary material, table S2). The direction of size-bias in extinctions was divided almost equally, with observed m > null on 10 islands and observed m < null on 11 islands. The plot of extinction size-bias effect size against the mean mass of original island faunas has a negative slope, although the association is not significant across all islands (figure 2a; slope = −7.85, p = 0.14, d.f. = 41). Across islands with size-selective extinctions, however, there is a significant negative association (figure 2b; slope = −2.85, p = 0.006, d.f. = 18). Across islands with at least one introduced large predator (cats, foxes or dingoes), the association is significantly negative (figure 2c; slope = −1.89, p = 0.01, d.f. = 24), but non-significant across islands without any of these predators (figure 2d; slope = 0.3, p = 0.51, d.f. = 14). In all models, body mass was log-transformed, and we removed one island with a studentized residual more than 60. As this island had a mean body mass of original fauna at the lower end of the x-axis (103.5 g), removal of this outlier had a conservative effect on the models. We tested for spatial autocorrelation in the residuals of the model presented in figure 2a, but found no significant spatial effect on size-bias residuals (Moran's I = −0.038, p = 0.68), justifying the use of non-spatial regressions.
The immediate causes of extinction are no great mystery—usually, one or more of the ‘evil quartet’ (hunting, habitat loss, introduced species or coextinctions; ). It is less obvious why different species respond differently to similar threat types. Large-scale comparative analyses have begun to answer this question by showing that numerous biological traits mediate the influence of external threats on species’ extinction risk. Evidence for some particularly consistent and common relationships, including a positive association between body mass and extinction risk, presents the possibility of assigning data-deficient species to provisional threat categories or planning pre-emptive conservation measures . While the largest mammal species have notable disadvantages in the face of human impacts, our results show that medium-sized species are likely to be even more vulnerable than large species where exotic predators are the primary threat, although this will depend on the size of the key predators in any given system. This casts doubt on the universality of a positive association between size and extinction risk.
The major novelty of our analysis is twofold. First, data on mammal extinctions for a large number of islands, with and without introduced predators, provide a degree of replication not found in previous continental-scale analyses [11,12,14]. Second, our tests for extinction size bias are performed with reference to the body size distribution of each island's original fauna, which provides an appropriate null model. Although we found a clear difference in extinction size bias between islands with and without predators, it is possible that the presence of predators covaries with some environmental feature of islands that drives the size bias. However, this seems unlikely. We tested the size-bias effect size against three island environmental features (area, distance from mainland and mean annual rainfall), but none showed significant univariate associations, or improved the fit of the models presented in the Results. Our results also reject the energetic hypothesis for higher extinction risk in medium-sized species [8–10], because under this hypothesis the hump-shaped pattern should also be seen on islands without predators. The presence of at least one species of large introduced predator therefore appears to be the best explanation for the pattern of extinction size bias towards intermediate body sizes.
While the findings of our study support previous claims that medium-sized mammal species in Australia have been unusually vulnerable to predator-driven extinction [8,12], the suggested ‘critical weight range’ of 35–5500 g appears to have little biological reality. If this range did reflect the zone of elevated extinction vulnerability, we would expect the extinction size bias to be close to zero for islands with a mean faunal body mass within this range. Instead, these islands tend to have an extinction size bias that is highly positive (figure 1c), suggesting extinctions are more frequent for species above 5500 g. This pattern is actually more consistent with evidence for an upper prey size for foxes of around 10 kg . Dingoes also take prey larger than 5500 g , but only three islands have dingoes as the sole large predator species, making it difficult to distinguish their role from that of foxes.
The data and full analysis results are provided in the electronic supplementary material.
E.H. was supported by an Australian Postgraduate Scholarship.
We thank Xia Hua for advice on the analysis.
- Received December 16, 2013.
- Accepted March 11, 2014.
- © 2014 The Author(s) Published by the Royal Society. All rights reserved.