518 R. P. Duncan and others Prehistoric bird extinctions Table 1. Significance tests for the fixed-effect predictors included in two GLMMs that predict prehistoric extinction at Marfells Beach, with the taxonomic variable ���family��� included as a random effect. (In the unadjusted model, the observed values of the minimum number of individuals (MNI) in the dune and midden deposits are used in calculating the selection ratios. In the adjusted model, the MNI of non-extinct species in the dune deposits are discounted by a proportion equal to 0.61 to account for differential accumulation times (see �� 3).) unadjusted model adjusted model fixed effect type III 2 p type III 2 p selection ratio 12.3 0.001 3.9 0.049 habitat 16.6 0.001 6.3 0.012 loge (body mass) 6.8 0.009 7.9 0.005 selection ratio �� loge (body mass) 12.6 0.001 4.3 0.039 We excluded from our analyses taxa that were not in the pre- human breeding avifauna of New Zealand, because these may have colonized the area only after it was abandoned by the Maori (see also �� 3). We also excluded birds that forage prim- arily at sea (petrels, gannets, penguins and shags), because these species may be over-represented in the dune deposits due to storm wrecks (Powlesland & Powlesland 1993). Out of the remaining 59 taxa (615 MNI) included in our study, 18 had become extinct in the northeast of South Island by the time Europeans arrived in New Zealand (Holdaway 1999). Four of these extinct taxa were species of moa, but moa were relatively uncommon at Marfells Beach, comprising only 16 out of 615 MNI in our analysis. To estimate relative hunting pressure, we calculated a selec- tion ratio (Manly et al. 1993), wi, using the MNI of each taxon collected in the dune and midden deposits at Marfells Beach: wi = oi/pi , (where oi is the MNI of taxon i recovered from the middens divided by total MNI recovered from the middens, and pi is the MNI of taxon i recovered from dunes divided by total MNI recovered from dunes). A value of wi 1 indicates positive selection of a taxon by Maori hunters: the taxon was dispro- portionately abundant in the middens relative to its abundance in the local area, as measured by its abundance in the natural dune deposits. A value of wi 1 indicates avoidance: a taxon was less abundant in the middens relative to its abundance in the natural dune deposits. Overall, larger values of wi indicate greater hunting pressure. For prehistorically extinct birds, we cannot estimate directly life-history traits relating to population growth rate, such as clutch size, age at sexual maturity and inter-clutch interval. However, among birds in general, such traits are correlated with body mass (Bennett 1986 Saether 1987). These life-history traits also tend to vary relatively little among species classified in the same family, but show significant variation across families (Bennett 1986 Owens & Bennett 1995). To account for differ- ences in life history, we therefore included body mass and a vari- able coding for family in our analyses. Classification to family followed Sibley & Monroe (1990) (or Turbott (1990) for the extinct species that were not classified in Sibley & Monroe (1990)). Body mass values were taken from Holdaway (1999). A further confounding factor is that the taxa recovered from Marfells Beach occurred in different habitats: 26 out of the 59 taxa occupied coastal or freshwater habitats, and the remaining 33 taxa occupied terrestrial, mostly forest, habitats. Extensive forest clearance by fire occurred in the northeast of South Island shortly after Maori arrival (McGlone 1983 McGlone & Basher 1995) and this habitat loss may have selectively reduced popu- lations of terrestrial birds, increasing their susceptibility to Proc. R. Soc. Lond. B (2002) extinction. We therefore included an additional variable coding species by habitat (either coastal and freshwater, or terrestrial) in our analyses. We tested if greater hunting pressure (larger values of the selection ratio, wi ) was associated with a higher probability of extinction, having controlled for differences among taxa in body mass, habitat use and family membership by fitting a generalized linear mixed model (GLMM), using the SAS macro Glimmix (Littell et al. 1996). GLMMs allow the modelling of variances and covariances through the specification of random effects, accommodating situations in which observations at one level are correlated because they are clustered into higher level groups and therefore may not be independent (Goldstein 1995). In this case, we included a variable coding for family as a random effect in the model. This acts to control for unmeasured life-history traits that vary among taxa at the family level, and ensures that the significance tests for the fixed-effect predictors (see below) are not biased by non-independence due to unmeasured traits shared at the family level (e.g. any avian life-history traits that vary mainly among families, as noted above). Our response vari- able was binary (whether taxa became prehistorically extinct or not in the Marfells Beach region), and we specified a binomial error distribution and logit link function. We included the vari- ables selection ratio (wi ), body mass and habitat as fixed-effect predictors in the model. In addition, we tested for a significant interaction between wi and body mass, which tested for a dispro- portionately higher extinction risk among larger-bodied (lower population growth rate) taxa for a given level of hunting pressure. 3. RESULTS All three fixed effects along with the interaction between selection ratio and body mass were highly significant pre- dictors of prehistoric extinction in a GLMM with family as a random effect (table 1). The form of these relation- ships is evident from table 2 and figure 1. Among fresh- water and coastal species, the extinction rate was higher in taxa that were positively selected by Maori hunters (5 out of 13) compared with those that were not (0 out of 13). Furthermore, among those freshwater and coastal species that were positively selected by hunters, extinc- tions were concentrated in the larger-bodied taxa (figure 1). Hence, greater hunting pressure in combination with larger body mass (lower population growth rate) led to the extinction of a predictable subset of freshwater and coastal species. The same pattern holds for species from terrestrial habitats. Of the five terrestrial taxa that were positively on April 22, 2011 rspb.royalsocietypublishing.org Downloaded from

Prehistoric bird extinctions R. P. Duncan and others 519 Table 2. The number of fossil taxa in the prehuman avifauna (excluding seabirds) collected from Marfells Beach by habitat, selection ratio and extinction response categories. (A selection ratio of more than 1 indicates positive selection by Maori hunters (see �� 2).) extinction response habitat selection ratio surviving extinct freshwater and coastal 1 13 0 1 8 5 terrestrial 1 19 9 1 1 4 selected by hunters, it was the four largest-bodied taxa that became extinct. However, nine large-bodied terres- trial taxa became extinct despite relatively low hunting pressure at Marfells Beach, elevating the extinction rate among terrestrial species (13 out of 33) compared with freshwater and coastal species (5 out of 26). Our finding that greater hunting pressure is associated with a higher extinction rate rests on the assumption that the relative abundance of a taxon in the dune deposits provides an unbiased estimate of its relative abundance in the prehistoric avifauna. With the exception of seabirds (which we excluded from the analysis), it is unlikely that a significant bias would arise as a result of species being selectively deposited on the dunes. Unlike many natural preservation sites, which are biased towards trapping cer- tain kinds of bird, species ought to be deposited on the dunes in direct proportion to their natural abundance in the local avifauna. We would expect, however, a signifi- cant size bias in bone preservation and collection, with the bones of larger-bodied species surviving for longer and being more readily noticed and collected (sieves are not recorded as being used at Marfells Beach). This would result in individuals of larger-bodied taxa being over- represented in the dunes relative to their abundance in the local avifauna. A similar size bias in bone recovery should also occur in the middens. The crucial point is whether the strength of this bias differs between the dune and mid- den deposits, thus biasing the selection ratios. To test this, we estimated the rate of bone recovery for each taxon in the two sources (dune or midden deposit) as loge (total number of bones collected/MNI). As expected, taxa of larger body mass had a significantly higher rate of bone recovery and, for a given body mass, the bone recovery rate was also higher in the dune relative to the midden deposits (table 3), presumably because bones of hunted individuals would have been destroyed by cooking and eating. More importantly, however, the body mass �� source interaction term was non-significant when added to this model (t = 1.30, p = 0.20), showing that any difference in the size-related recovery rate for the two sources does not seriously bias the selection ratios. The Marfells Beach site was probably abandoned by the Maori before European arrival (abandonment in the per- iod ca. AD 1300���1769). There must, therefore, have been a period of at least ca. 200 years during which only the bones of birds that survived prehistoric Maori occupation were accumulating in the dunes. As a result, we would expect these taxa to be over-represented in the dune Proc. R. Soc. Lond. B (2002) Figure 1. The mean loge (body mass) (�� s.e.m.) of prehistorically extinct and surviving taxa collected at Marfells Beach, by habitat and selection ratio categories. A selection ratio of more than 1 indicates positive selection by Maori hunters (see �� 2). Statistics show the results of t-tests comparing mean loge (body mass) between surviving and extinct taxa in each category ((a) no statistical comparison possible (b) t11 = 2.5, p = 0.030 (c) t26 = 4.7, p 0.001 (d) t3 = 2.7, p = 0.077). deposits relative to their availability to prehistoric hunters, biasing the selection ratios in favour of the outcomes we observed. However, our conclusions are robust to this effect. For those species that survived prehistoric Maori occupation, we reduced the MNI in the dunes to a proportion equal to 0.61. Assuming that bones have accumulated in the dunes at a constant rate, this has the effect of ���resetting��� the MNI of non-extinct species in the dune deposits to a point 700 years ago (the approximate time of Maori arrival, given that the dunes are 1800 years old Ota et al. 1995). We then recalculated the selection ratios using these adjusted MNI values. This procedure on April 22, 2011 rspb.royalsocietypublishing.org Downloaded from