Data on maximum longevity, to the nearest month, were obtained mostly from the Patuxent Wildlife Research Center database, which provides for each species the number of birds banded, the number of recoveries and the maximum longevity recorded for birds captured in North America. In addition, I obtained some data on maximum longevity from the primary bird literature. Maximum longevity usually increases with the number of recoveries (Arnold 1988
). Therefore, number of recoveries was used to adjust estimates of maximum longevity in all models. As a compromise between sample size and model estimate precision, I excluded species with fewer than 15 recoveries.
Biological data for all species were obtained from the literature (appendix A in the electronic supplementary material). Group size data during foraging often include the range but less often the mean or modal values. I therefore used the maximum group size reported during foraging. In previous work, maximum group size has been shown to be responsive to ecological factors (Beauchamp 2004
). I selected data on group sizes obtained during the non-breeding season so as to avoid possible interference from parental duties. I excluded data obtained at night or during migration. Because many species join groups of other species, estimates of maximum group sizes also included the number of heterospecifics. However, given that members of one species may be more responsive to the number of conspecifics than to the total number of companions in the group (Metcalfe 1984
), I also tallied the maximum number of conspecifics in groups. Number of cited references consulted was used as a cofactor because estimates of maximum group size might increase with the number of references investigated. For body mass, I used data from males in the non-breeding season, if available, because body mass in males usually shows less fluctuation throughout the year. I distinguished passerine and non-passerine species using the existing taxonomy. I classified each species as aquatic or terrestrial, depending on the main habitat used for foraging, and as foraging in open or closed habitats depending on the availability of vegetation cover during foraging activities. For migration, I distinguished long-distance migrants, which typically migrated over 30° of latitude, from short-distance or non-migrating species.
All quantitative data were log10
-transformed prior to statistical analysis. I first performed a multiple regression analysis using species as the unit of analysis. The final model was obtained using backward elimination of non-significant variables. A phylogenetic analysis relied on independent contrasts, which were calculated using the PDAP module (Midford et al. 2008
) within Mesquite
(Maddison & Maddison 2009
) assigning all branch lengths to 1. The phylogeny underlying the calculations was based mostly on two recent papers describing phylogenetic relationships among bird families (Jønsson & Fjeldså 2006
; Livezey & Zusi 2007
). In addition, I used several papers describing interspecies relationships within various bird families (appendix B in the electronic supplementary material). Contrasts were obtained for each continuous variable. For categorical variables, I relied on ancestral state reconstruction with Mesquite
to establish the most parsimonious distribution of the trait along the phylogeny lineages. I used the same multiple regression framework to analyse the data with the restriction that the intercept must pass through the origin (Felsenstein 1985
). The phylogenetic analysis was also conducted using mean or modal group sizes and the results were very similar, albeit based on a smaller sample size, and thus ignored. With the present sample size, the power was sufficient to detect small correlations (<0.1).