We are pleased that Collins et al. (2011) now agree that for Bismarck bird distributions ‘there are excesses of checkerboard pairs within both genera and defined guilds’ compared with random expectations, and that this is consistent with strong influence by interspecific competition. They also discuss the other hypotheses that we considered in a prior publication (Sanderson et al., 2009). Here, we alert readers to problems in their paper that, if cured, would have considerably strengthened those conclusions.
- 1 We had already provided, in downloadable form, the complete database of 41 Bismarck islands and 142 Solomon islands, and 150 and 141 analysable bird species, respectively (Sanderson et al., 2009). Collins et al. (2011) analysed only a small fraction of those data: 31 Bismarck islands and 154 bird species tabulated by Mayr & Diamond (2001). They ignored the Solomon Islands completely, thereby failing to notice the strikingly consistent behaviour of individual bird genera between the two archipelagoes that we discussed. Furthermore, as noted by Mayr & Diamond (2001) and by Sanderson et al. (2009), four of those 154 species (Casuarius bennettii, Pelecanus conspicillatus, Scythrops novaehollandiae, Halcyon sancta) should not have been included in the analysis because of uncertainty about their breeding status or native status.
- 2 Collins et al. (2011) employed only 1000 null matrices. We found (Sanderson et al., 2009, p. 774, paragraph 2) that even 10,000 null matrices are insufficient to converge on consistent null distributions. We used 1,000,000 matrices for all our published analyses.
- 3 Collins et al. (2011) conducted three sets of analyses (community-wide, congeneric, and intraguild checkerboards), of which we had already published the first two. We did not do the third (intraguild analyses), because Connor & Simberloff (1979) had criticized them previously, and we note that Collins et al. (2011) criticized them again in their paper. We performed four additional analyses that Collins et al. (2011) omitted: positive associations (our p. 775), archipelago differences (p. 775), incidence effects (p. 777), and genus comparisons between archipelagoes (p. 778). Thus, Collins et al. (2011) report no new analyses: they merely repeat one-third of our analyses with one-sixth of our dataset and 1–1000th of our number of randomized matrices. Their discussion also adds nothing substantively new: it merely raises again, without the benefit of personal experience, the issues of supertramp distributions and historical biogeography that we had already discussed with knowledge of the species and archipelagoes involved.