Correspondence Paul W. Sherman, Department of Neurobiology and Behavior, Mudd Hall, Cornell University, Ithaca, NY 14853, USA E-mail: firstname.lastname@example.org
We comprehensively reviewed information on maximum life spans of wild birds (based on banding recoveries) and nine ecological, physiological and behavioral variables that have been hypothesized to affect the evolution of avian life spans. Data on maximum longevities and body masses were available for 936 species, and data on all variables were available for 470 species in 40 families from 15 orders. The Phoenicopteriformes (flamingos), Psittaciformes (parrots) and Procellariiformes (petrels and shearwaters) had the longest mean maximum life spans (>30 years), and the Passeriformes (perching birds), Podicipediformes (grebes) and Piciformes (woodpeckers) had the shortest mean maximum life spans (<10 years). Other orders were intermediate, with the Gruiformes (cranes and rails), Anseriformes (waterfowl), Ciconiiformes (herons and egrets) and Pelecaniformes (pelicans) living a mean maximum of 20–30 years, and the Columbiformes (pigeons), Strigiformes (owls), Falconiformes (hawks), Sphenisciformes (penguins) and Charadriiformes (shorebirds) living a mean maximum of 10–20 years. Within the speciose order Passeriformes, the Corvidae (crows) had longest mean maximum life spans (>17 years), and the Tyrannidae (flycatchers) and Parulidae (wood warblers) had the shortest mean maximum life spans (6 years). Multivariate regression analyses revealed that the independent variables together explained 80.3% of the variation in maximum longevities among 40 avian families, and 69.6% of the variation among 17 families of Passeriformes. In the comprehensive analysis four variables significantly affected maximum longevities, namely body mass, diet, sociality and breeding insularity (mainland vs. island), whereas breeding latitude, breeding habitat, nest-site location and migratory behavior did not have significant effects. These results are consistent with evolutionary theories of senescence, which predict that morphological and behavioral attributes that reduce extrinsic mortality should select for mechanisms that postpone physical deterioration, resulting in longer life spans and extended breeding opportunities.
Senescence is ‘a persistent decline in age-specific fitness components of an organism due to internal physiological deterioration’ (Rose, 1991). Senescence is progressive, irreversible, endogenous, and ubiquitous (Strehler, 1962). The occurrence of senescence poses an important puzzle for evolutionary biology (Williams, 1957; Hamilton, 1966; Austad, 1997) because, all else being equal, longer-lived individuals have more opportunities to reproduce than shorter-lived conspecifics, so natural selection should consistently favor greater longevities. Surprisingly, therefore, in all major taxonomic groups of plants and animals life lengths exhibit negative binomial distributions, with far more short-lived than long-lived species (e.g. Finch, 1990; Hulbert et al., 2007; de Magalhaes, Costa & Church, 2007; Ricklefs, 2008).
There are three, closely related evolutionary explanations for senescence (Medawar, 1952; Williams, 1957; Kirkwood, 1977, 2002). All of them propose that senescence is an outcome of population demography that is affected by natural selection only indirectly, rather than something that natural selection on individuals and their genes has favored directly. The core idea is that when rates of extrinsic mortality are high enough that most individuals in any population do not survive very long, natural selection will be relatively ineffective in promoting physiological mechanisms that repair damage and defects among the few surviving elderly, resulting inevitably in the creeping in of senescent decline. Genes that have beneficial effects early in life will be favored even if they promote or fail to prevent late-life deterioration (reviewed by Ricklefs, 1998, 2008; Kirkwood & Austad, 2000; Monaghan et al., 2008). As expected, rates of senescence are positively correlated with rates of extrinsic mortality in many vertebrates (Ricklefs, 1998, 2000, 2008; Ricklefs & Scheuerlein, 2001). Of course, selection consistently favors traits that reduce susceptibility to extrinsic mortality, but because it can be stochastic (e.g. food shortage, bad weather) or co-evolutionary (competition, parasitism, predation), extrinsic mortality can never be eliminated.
However, not all birds have long life spans. Indeed even among flighted birds longevities are known to differ considerably, but the data have not been synthesized or analyzed. We therefore collected all available information on maximum life lengths of free-living and captive birds, and then used multivariate statistical techniques to investigate the effects of nine extrinsic variables that have been hypothesized to affect avian senescence and longevity (e.g. Brawn, Karr & Nichols, 1995; Finch & Ricklefs, 1995; Martin, 1995; Böhning-Gaese et al., 2000; Martin et al., 2006; Møller, 2006, 2007; Fontaine et al., 2007). Our results indicate that maximum life spans are affected by multiple attributes, all of which tend to reduce extrinsic mortality, as predicted by evolutionary theories of senescence.
Maximum longevity data were available for many birds in the wild (based on banding recoveries) and for some in captivity (based on zoo records). To see whether these two types of data yielded similar results, we compared them for 98 species with both types of information. There was a significant positive relationship between longevities in the wild and in captivity (F=88.4, d.f.=1,97, P<0.0001), and there were no significant differences between maximum longevities in captivity and in the wild for the six avian families that included at least three species with both data types (Wilcoxon's signed rank tests: Table 1). Similarly, Ricklefs (2000) reported that rates of actuarial senescence (quantified using the Weibull aging function) did not differ between captive and wild bird populations. In light of the much larger sample size and greater ecological validity of field data we focused our analyses on banding recoveries with one exception, the order Psittaciformes (parrots). In these long-lived birds, only data on maximum longevities in captivity are available for 45 of 47 species (in the families Cacatuidae and Psittacidae). In view of the parallel between maximum longevities in the wild and in captivity, it seemed unlikely that the mix of banding recoveries (for 425 species) and zoo records (45 species) that we analyzed would yield misleading results.
Table 1. Tests of the equivalence of maximum longevities in the wild (banding recoveries) and in captivity among six avian families with species for which both types of information were available (Wilcoxon's signed rank tests)
Our data base contains a single mean mass and maximum longevity for each species, regardless of sex. The quality of these data varied considerably among species due to differences in sample sizes (often <20 individuals) and lengths of studies relative to maximum life spans. For some families, especially those that are regularly hunted (e.g. the Anatidae), accurate information on maximum longevities of both sexes exists for many species, but for most families such information exists for only a few species. Longevity records for species in poorly sampled families will undoubtedly be superseded by results of ongoing and future long-term studies. To partially address these problems, Møller (2006, 2007) recommended controlling for sampling effort statistically. Unfortunately, we were unable to do so because for most species in our data base (Appendix 1) the information was not reported. There also is considerable intra-specific variability in life spans within species of birds (e.g. Fox et al., 2006; Nussey et al., 2008), and some are sexually dimorphic in size (reviewed by Shine, 1989) and also exhibit sexual dimorphisms in extrinsic mortality, senescence rates and maximum longevities (Promislow, 2003; Carranza et al., 2004; Christe, Keller & Roulin, 2006; Clutton-Brock & Isvaran, 2007; Bonduriansky et al., 2008). Moreover, both captive and wild data probably underestimate actual maximum life spans because in the field the recovery of very old, banded birds requires considerable luck, and in captivity mortality can result from accidents, animal care practices and inappropriate living conditions rather than senescence (e.g. Sherman & Jarvis, 2002). We were forced to use a single mass and longevity datum per species by lack of other information: intra-specific variation in life spans has been quantified for only a few birds (e.g. Fox et al., 2006; Jones et al., 2008; Keller et al., 2008), and in the 11 data bases we consulted (Appendices 1 and 2) body masses of males and females typically were not separated and the sex of the longest-lived individual usually was not specified. All our analyses assume that, like noise in a signal, deficiencies in the quality and quantity of maximum longevity data for individual species would increase variance and mask associations with ecological, physiological and behavioral variables that actually exist, but they would not generate associations that do not in fact occur.
(A)Diet– Each species was categorized based on its typical diet as being a: (1) Carnivore; (2) Herbivore; (3) Omnivore.
(B)Breeding latitude– Each species was categorized based on its typical breeding latitude: (1) 0° to ±30°; (2) ±31° to ±60°; (3) ±61° to ±90°. No distinction was made between the northern and southern hemispheres. Large increments were used (30° which equals 3333.69 km) because many species breed over broad latitudinal gradients.
(C)Breeding insularity– Each species was categorized based on whether it typically nests on: (1) Islands; (2) the Mainland.
(D)Age at first reproduction– Each species was categorized based on the number of years after hatching when individuals typically first mate.
(E)Breeding habitat– Each species was categorized based on whether its typical breeding habitat is (1) Grassland; (2) Scrubland; (3) Wetland; (4) Woodland.
(F)Nest location– Each species was categorized based on whether its typical nest-site location within its breeding habitat is (1) Aquatic (floating on water); (2) Cavity; (3) Ground; (4) Shrub; (5) Tree Branch.
(G)Migratory behavior– Each species was categorized based on whether it is typically: (1) Migratory; (2) Sedentary Year-Round. Migration was defined as the seasonal movement of a complete population. Species were not ranked based on their average migration distance because there often is so much variability among populations of the same species.
(H)Sociality– Each species was categorized based on its breeding social behavior: (1) Social; (2) Non-Social. Social breeding was defined broadly to include colonial nesting and cooperative breeding.
Ultimately, information on both continuous variables (maximum longevity and mean mass) and all eight categorical variables was available for 470 species (Appendix 2). This represents almost 5% of the world's avifauna, and we believe it is the largest data base of its kind available. The Passeriformes was the most speciose order in our data base, containing complete information on 179 species in 17 families. We included this order in our comprehensive analysis and also analyzed the Passeriformes separately, to see if intra- and inter-order results corresponded; no other orders could be analyzed separately due to insufficient sample sizes.
Initially we had planned to include ‘age at first reproduction’ in our multivariate analyses. However, we decided not to do so for two reasons. First, preliminary exploration of our data base revealed that age at first reproduction was so tightly correlated with mean mass (F=11.1727, d.f.=1314, P<1.29E−24) that the two variables could not be treated as independent. Second, age at first reproduction is more appropriately considered an effect rather than a primary cause of the environmental factors generally hypothesized to underlie variations in life spans and senescence rates. Age at first reproduction is a life-history trade-off between advantages of early sexual maturation (e.g. shortening the vulnerable juvenile period and increasing the likelihood of realizing reproduction) versus advantages of continued growth and social maturation (e.g. larger body size, reduced predation and enhanced reproductive output) (Charnov, 1993; Roff, 2002). Although Ricklefs & Cadena (2007) reported that age at first reproduction did not strongly influence avian life spans (in captivity), de Magalhaes et al. (2007) found that time to reproductive maturity was correlated with adult life span in mammals and birds.
We also considered including ‘chemical protection’ as an additional independent variable in our analyses (see Blanco & Sherman, 2005). Edibility scores, based mostly on responses of ‘unnatural’ predators (e.g. insects, humans), have been published for 105 species of birds from southern Africa (Cott & Benson, 1970; Götmark, 1994); in addition, nine species in the New Guinean family Pachycephalidae (especially the genus Pitohui) have been found to contain defensive neurotoxins (batrachotoxins) in their skin and feathers (Dumbacher et al., 2008; Jønsson et al., 2008). Unfortunately, however, information on maximum life spans in nature is not available for most of these 114 species.
In light of these uncertainties about avian phylogenies and analytical techniques, we chose an alternative approach to minimize possible effects of non-independence of species: testing for hypothesized relationships at higher taxonomic levels (families), as suggested by Reeve & Pfennig (2003). Thus we computed mean values for each continuous and discrete variable for all the species in each avian family, and entered these mean family values in our multivariate models. To try to ensure that families had been sampled adequately to yield meaningful results, we included only those for which data on body masses and maximum longevities were available for >5 species. To reveal the details of the variables that were significant predictors in the multivariate analyses, we conducted a posteriori univariate analyses using all species that were included in each continuous and discrete variable category.
Before analysis, data on maximum longevities and mean masses were log transformed to adjust for unequal variances among families. The composite data base was then entered into a multivariate regression model using jmp® 7.0 statistical software (SAS Institute Inc., 2007). Mean maximum longevities of 40 avian families and, separately, 17 passerine families was the dependent variable, Y, and mean masses and means of the eight categorical variables were the independent variables, Xi, i=1, …, p, with ɛ defined as the error term representing the unpredicted variation in the response variable. The data were modeled with the following equation:
Longevities of free-living birds
Maximum longevities in nature differed markedly among 15 avian orders (Fig. 2a). The Phoenicopteriformes (flamingos), Psittaciformes (parrots) and Procellariiformes (petrels and shearwaters) had the longest mean maximum life spans (>30 years), whereas the Passeriformes (perching birds), Podicipediformes (grebes) and Piciformes (woodpeckers) had the shortest mean maximum life spans (<10 years). Other orders were intermediate, with the Gruiformes (cranes and rails), Anseriformes (waterfowl), Ciconiiformes (herons and egrets) and Pelecaniformes (pelicans) living a mean maximum of 20–30 years, and the Columbiformes (pigeons), Strigiformes (owls), Falconiformes (hawks), Sphenisciformes (penguins) and Charadriiformes (shorebirds) living a mean maximum of 10–20 years.
Sample sizes of families of Passeriformes were large enough to enable a separate analysis of 17 families in this order (Fig. 1b). The longest-lived Passeriformes were the Corvidae (crows: mean maximum of >17 years) and the shortest-lived were the Tyrannidae (flycatchers) and Parulidae (wood warblers: both c. 6 years). Among other families, the Hirundinidae (swallows), Paridae (titmice), Passeridae (sparrows) and Turdidae (thrushes) lived a mean maximum of 10–13 years, the Sittidae (nuthatches), Motacillidae (motmots), Fringillidae (finches), Sylviidae (old-world warblers), Cardinalidae (grosbeaks) and Emberizidae (buntings) lived a mean maximum of 8–9 years, and the Laniidae (shrikes), Mimidae (mockingbirds), Vereonidae (vereos) and Icteridae (blackbirds) lived a mean maximum of 7–8 years.
Our comprehensive multivariate regression model, which included mean maximum longevities of 40 families as the dependent variable and family-level means for the eight independent variables, explained 80.3% of the variance in maximum life spans (Table 2). Details of the results are presented in Appendix 3. Among the independent variables, mean mass, diet and sociality were highly significant predictors, and breeding insularity was marginally significant (P=0.053). The multivariate regression model of the Passeriformes dataset, which included mean maximum longevities of 17 families as the dependent variable and family-level means for the eight independent variables, explained 69.6% of the variance in maximum life spans (Appendix 3, F=2.2898, d.f.=8, 16, r2=69.6%, P=0.13). However, none of the independent variables significantly predicted mean maximum longevities of passeriform families, probably due to large inter-family variability and small sample sizes (i.e. ≤10 species for nine of the 17 passerine families).
Table 2. Results of the multivariate analysis of the comprehensive dataset, to explore sources of variation in mean maximum longevities of 40 avian families (F=15.8028, d.f.=8, 39, r2=80.3%, P<.0001)
Sum of squares
Details of parameter estimates and effects tests are presented in Appendix 3.
Maximum longevities of several species in our data base (Appendix 2) were anomalously low relative to congeners, for example, Puffinus auricularis, Puffinus gravis, Anas diazi, Anas fulvigula and Anas laysanensis. Longevity records for these species at least likely are artifacts of inadequate sampling. We therefore removed them and re-ran the model. The significance of the overall results was unchanged and, as before, mass, diet and sociality were significant predictors and breeding insularity was marginally significant. Apparently longevities of these ‘questionable’ species did not have a major influence on the results, especially since we analyzed family-level mean values. Of course data deficiencies and biases may have existed for other species as well, but we were unable to identify them a priori. Our ‘total evidence’ approach (Sherman, Holland & Shellman Sherman, 2008) assumes that any unknown data shortcomings were more likely to obfuscate associations with extrinsic variables that actually exist than to falsely create relationships that do not occur.
To explore the details of the relationships revealed by the multivariate model we conducted a posteriori analyses of each significant variable individually, using species-level data. Maximum longevities were significantly related to body masses in the comprehensive dataset (Fig. 2a) and in the family Passeriformes (Fig. 2b). Regarding diets, in the comprehensive dataset (Fig. 3a) mean maximum longevities of herbivorous species (c. 27 years) were significantly longer than those of carnivores (17 years) and omnivores (14 years). Among the Passeriformes (Fig. 3b), herbivores and omnivores had similar mean maximum life spans (c 10 years), which were longer than maximum life spans of carnivores (7 years). Regarding sociality, in the comprehensive dataset (Fig. 4a) mean maximum longevities of social species (24 years) were considerably longer than non-social species (13 years). Among Passeriformes (Fig. 4b) social species also had greater mean maximum longevities than non-social species (13 vs. 9 years). Regarding breeding insularity, in the comprehensive dataset (Fig. 5) mean maximum longevities of island-breeders (26 years) were considerably longer than those of mainland breeders (15 years). It was not possible to perform the parallel analysis of the effects of insularity within the Passeriformes because there were only three island-breeding species.
Our review and analysis of maximum life spans of free-living birds revealed considerable variability among 40 families of birds from 15 orders (Fig. 1a) and among 17 families in the order Passeriformes (Fig. 1b; Appendices 1 and 2). Multivariate analyses of the comprehensive dataset indicated that mean maximum longevities were significantly influenced by body mass, diet, sociality, and breeding insularity (marginally) (Figs 2–5, Table 2, Appendix 3), but not by breeding latitude, breeding habitat, nest location or migratory behavior. Separate analyses of families of Passeriformes yielded quantitatively similar, but generally non-significant results, likely due to variability associated with the smaller number of families and small sample sizes for many families.
There is an interesting reversal of the body mass–longevity correlation in a mammal that further clarifies the evolutionary forces molding senescence patterns in general. Usually extrinsic mortality is inversely related to body size, but in domestic dogs Canis familiaris the small-bodied breeds live longer than large-bodied breeds (Li et al., 1996; Speakman, Acker & Harper, 2003; Galis et al., 2007). This anomaly is illuminating because larger breeds of dogs were artificially selected for participation in dangerous activities such as hunting large game, fighting and protecting their owners, all of which carry high mortality risks, and for rapid growth and early maturation (rather than somatic maintenance and repair) to facilitate these activities. By contrast, smaller dog breeds were selected to serve as companion animals and lap dogs or to capture vermin (rats and mice), so they lived in more protected environments, suffered lower extrinsic mortality and matured more slowly. As a result, the onset of senescence occurs later in small-bodied breeds than large-bodied breeds.
Among families of birds, diet significantly affected maximum longevities (Table 2, Appendix 3). Follow-up analyses indicated that among all birds, herbivores lived significantly longer than carnivores or omnivores (Fig. 3a), and that among passerine families herbivores and omnivores lived longer than carnivores (Fig. 3). There are several reasons to hypothesize that herbivores generally experience lower rates of extrinsic mortality than carnivores, all else being equal. First, carnivorous (and some omnivorous) species can be injured or killed during chases and attacks on prey, whereas herbivores experience no direct danger from their food. Second, herbivores are less likely to contract parasites or pathogens from their food than carnivorous or omnivorous species. Third, the food supply of herbivores is more stable, consistently available and evenly distributed than the prey of carnivores. To further examine these possibilities, we tried to separate herbivores into grass/leaf eaters and frugivores, and to separate carnivores into meat, fish and insect eaters. However, small sample sizes and high intra-category variances thwarted statistical analyses of these sub-categories.
Overall, our dietary results parallel those of Munshi-South & Wilkinson (2006), who found that diet explained a significant amount of the variance in maximum life spans of parrots, and that granivorous species lived longer than omnivorous and fruit-eating/insectivorous species. Granivory combines the effects of a low risk, low parasite/pathogen diet (Brittingham, Temple & Duncan, 1988; Lombardo et al., 1996) with a readily available, evenly distributed and relatively stable food source (Dostine & Franklin, 2002), all of which could presumably reduce extrinsic mortality.
In our comprehensive multivariate analysis, breeding sociality significantly affected mean maximum longevities of avian families (Table 2; Appendix 3). A posteriori analyses revealed that social species lived longer than non-social species (Fig. 4). These results agree with those of Arnold & Owens (1998), who reported that cooperative breeding was correlated with low annual mortality and long life spans, as predicted by kin selection theory and life-history theory (Bourke, 2007). However, subsequent analyses by Møller (2006) and Blumstein & Møller (2008) called into question the role of sociality in the evolution of avian longevities and senescence patterns. The reasons for the difference between our results and theirs probably lie in differences in both sample sizes and definitions of sociality. Whereas Møller (2006) defined sociality as ‘colonial nesting’ and Blumstein & Møller (2008) defined it as ‘cooperative breeding,’ our definition included both. We took the broader approach because both colonial nesting and cooperative breeding have often been linked to reduced predation rates on adult birds, chicks and eggs, due to shared vigilance, sentinels, alarm calling, cooperative group defense, safety in numbers and selfish herd effects (e.g. Hoogland & Sherman, 1976; Hoogland, 1981; Hailman, McGowan & Woolfenden, 1994; Clutton-Brock et al., 1999; Hatchwell & Komdeur, 2000; reviewed by Safran et al., 2007).
The link between sociality and longevity is illustrated by the characteristics of the four longest and shortest-lived avian orders (Fig. 1). All four species of Phoenicopteriformes (flamingos) in our data base (Appendix 2) breed in colonies and crèche their chicks, all 25 Procellariiformes (petrels and shearwaters) and all 16 Pelecaniformes (pelicans) nest colonially, and 25 of 47 species (54%) of Psittaciformes (parrots) nest colonially or breed cooperatively. By contrast, only 38 of 179 (21%) Passeriformes (perching birds) and only two of nine (22%) Columbiformes (pigeons) in our data base nest colonially or breed cooperatively, only one of four (25%) Podicipediformes (grebes) breeds colonially, and only three of 15 (20%) Piciformes (woodpeckers) breed cooperatively. There are no consistent differences in diets, nest locations or breeding insularity between the four longest- and shortest-lived orders: the Phoenicopteriformes are omnivorous, mainland, aquatic nesters, the Psittaciformes are herbivorous, mainland, cavity nesters and the Procellariiformes and the Pelecaniformes are carnivorous, insular, ground nesters; the Podicipediformes are carnivorous, mainland, aquatic nesters, the Piciformes are carnivorous, mainland, hole-nesters, the Columbiformes are herbivorous and nest in various locations on mainlands, and the Passeriformes are highly variable in diets and nest-site locations (Appendix 1). Of course, it is possible that the effects of sociality are confounded by other factors. For example species in the family Corvidae are especially long-lived (Fig. 1b). Although this probably relates to their sociality, an intriguing additional possibility is that the defensive neurotoxins and aposematism that occur in the basal family Pachycephalidae actually are more widespread in corvids (Jønsson et al., 2008).
Colonial breeding represents a dynamic evolutionary balance between anti-predator benefits and costs of predator attraction due to increased conspicuousness of groups. To evaluate the relative importance of these antagonistic effects in the evolutionary history of the Ciconiiformes, Varela, Danchin & Wagner (2007) used a character-mapping approach. Their analyses led them to infer that predator avoidance was not the primary selective force that originally favored coloniality in ciconiiform history. However, as Varela et al. (2007) were careful to point out, their results do not mean that breeding in colonies provides no anti-predator benefits. Rather, once breeding colonies formed for any ecological reason (e.g. limited breeding habitat, locally concentrated food sources, or advantages of social foraging on ephemeral food patches), individuals living in those colonies also could have benefited from reductions in per capita predation (i.e. reductions in rates of extrinsic mortality).
In our comprehensive multivariate model, breeding insularity had a marginally significant effect on mean maximum longevities of avian families (Table 2; Appendix 3). Follow-up analyses indicated that birds which breed on islands can live almost twice as long as mainland breeders (Fig. 5). Presumably this is because island-breeding species experience lower mortality due to the relative lack of predators, parasites and pathogens on islands compared with the mainland (Blondel, 2000; Goüy de Bellocq et al., 2003; Zoellick et al., 2004). For example, Austad (1993) found that island-dwelling opossums Didelphis virginiana senesced more slowly and lived longer than conspecifics on the mainland, and he attributed the difference to lower rates of extrinsic mortality due to absence of predators.
Given the magnitude of the longevity differences between island and mainland breeding birds, it is surprising that breeding insularity did not more strongly affect mean maximum life spans in our comprehensive analysis (Table 2; Appendix 3). This was probably due to sample-sizes because our dataset was heavily skewed toward mainland-breeding species (n=318 of 470 species), with a limited number of island-breeding ducks and petrels. Moreover, there was considerable variability in maximum longevities (and body masses) among the island-breeding species (Fig. 5), perhaps due to differences in degrees of contact these birds have with mainland predators, parasites, and pathogens. Some islands are close to a mainland, and birds that breed there regularly forage on or migrate to the mainland and so are exposed to continental biotic enemies; other islands are far out in the ocean, and birds that breed there and spend the rest of their lives at sea are rarely exposed to mainland-dwelling biotic enemies.
Four variables that did not significantly affect maximum longevities in our multivariate analyses were nest location, breeding habitat, breeding latitude and migratory behavior (Table 2; Appendix 3). We originally included these variables because previous investigators had called attention to their possible effects on rates of extrinsic mortality and thus senescence. For example, predation on eggs and nestlings varies with nest location in many bird species (Schaub, Mumme & Woolfenden, 1992; Martin, 1995; Owens & Bennett, 1995; Martin & Ghalambor, 1999; Doerr, Doerr & Jenkins, 2006; Fontaine et al., 2007). However, which nesting locations are most and least susceptible to predation varies across species and habitats, and nest location has less impact on survival of adults and post-fledging juveniles than on eggs and nestlings in most species (Martin & Li, 1992). This is important because, theoretically, the onset of senescence is not expected to occur until reproduction commences (Williams, 1957; Hamilton, 1966), a prediction that has been supported empirically for birds and mammals (Charmantier et al., 2006; Møller, 2006; Jones et al., 2008). In addition, in many avian families nesting locations are variable among species, resulting in intermediate mean values in our family-level analyses that may have obscured any effects of nest location on mean maximum longevities.
Breeding habitat type also can affect the likelihood of predation, especially on eggs and nestlings (Martin, 1995; Doerr et al., 2006; Fontaine et al., 2007). However, within breeding habitats rates of extrinsic adult mortality due to predation often depend on breeding density. Breeding density also can increase reproductive costs (e.g. competition for food, mates and nest sites, parasitism, etc.), and thus affect life-history characteristics including senescence (Mysterud et al., 2001; Wilkin et al., 2006; Williams et al., 2006). Unfortunately, data on breeding densities and adult survival rates within and among nesting habitats were not available for the populations of the species whose maximum longevities appear our data base, so we were unable to investigate whether breeding habitat type affects maximum longevity while controlling for breeding densities.
We also did not find significant effects of breeding latitude or migratory behavior on maximum longevities (Appendix 3). By contrast, Møller (2007) reported that breeding latitude and migration distance explained, respectively, 3.7 and 2.3% of the variation in avian maximum longevities. He hypothesized that longevities decreased with increasing latitude due to ‘slow life histories’ at low latitudes (Jones et al., 2008), and that longevities increased with increasing migratory distance because migrants spend more time per year in benign abiotic environments and also possess special physiological adaptations for reducing oxidative stress (which were selected for in the context of enabling long-distance flights; see also Hulbert et al., 2007; Costantini, 2008).
Our failure to corroborate Møller's (2007) results may have been due to: (1) differences in sample sizes and analytical techniques [he conducted independent contrast analyses on 169 species of European birds, whereas we analyzed means of 40 avian families (470 species) world-wide]; (2) differences in the quantification of breeding latitude (he analyzed breeding latitude as a continuous variable, calculated as the mean of the northernmost and southernmost breeding season latitude for each species, whereas we analyzed breeding latitude more conservatively as a categorical variable, with three ±30° increments, due to considerable intra-specific variability in breeding locales); (3) differences in quantification of migration (Møller analyzed mean migratory distances of entire species as a continuous variable, whereas we considered migration as a categorical variable, again due to intra-specific variability in migration distances); (4) the relatively small amount of variation in longevities (<4%) that Møller explained by considering either breeding latitudes or migration distances.
The puzzle of senescence is being actively investigated at multiple levels of analysis (Sherman, 1988; Jenkins, 2004; Monaghan et al., 2008; Ricklefs, 2008), especially in birds (reviewed by Holmes & Martin, 2009). Our results indicate that much of the variation in avian longevities can be explained by differences in body mass, diet, breeding sociality and breeding insularity. The longest-lived species were large, herbivorous, social, island-dwellers, which is consistent with evolutionary theories of senescence (Medawar, 1952; Williams, 1957; Kirkwood, 1977, 2002) because all four traits can contribute to reducing rates of extrinsic mortality. In general, birds live longer than similar-sized mammals because flight facilitates escape from predators. Among avian families, behavioral and life-history characteristics that further reduce extrinsic mortality underlie much of the variability in maximum longevities and, probably, rates of senescence.
For helpful commentaries on preliminary versions of the manuscript we thank Ronald Booker, Walter D. Koenig, John W. Fitzpatrick, H. Kern Reeve, Kaitlin Stanmyer, the anonymous reviewers and, especially, Robert E. Ricklefs and Janet Shellman Sherman. Francoise Vermeylen and Sherry Weitzen provided statistical advice; Richard Wrangham suggested the aposematism hypothesis. For financial support, we thank the US Fish and Wildlife Service, and the College of Arts and Sciences, the Agricultural Experiment Station (Hatch Grant Program), and the S.H. Weiss Presidential Fellowship Fund at Cornell University.
Table Appendix 1. Data base containing information on mean masses and maximum longevities of 936 species of birds, along with all available information on seven ecological, behavioral, and physiological variables that have been hypothesized to affect avian senescence and longevity (Wasser and Sherman, Avian Longevity and Senescence)
Max longevity in wild
Max longevity in captivity
Decimals in the longevity data refer to months (e.g. 27.083=27 years, 1 month).
Abbreviations are as follows: breeding habitat: grass (grassland), scrub (scrubland), wet (wetland), wood (woodland); nest location: 1 (ground), 2 (shrub), 3 (tree branch), 4 (cavity), 5 (aquatic – floating on water); diet: omni (omnivore), herbi (herbivore), carni (carnivore); breeding insularity: M (mainland), I (island); breeding latitude: 1 (0° to ± 30°), 2 (30° to ± 60°), 3 (60° to ± 90°); migratory behavior: Y (yes), N (no); sociality: S (colonial or cooperative breeder), N (non-social breeder).
Table Appendix 2. Data base containing information on mean masses and maximum longevities of 470 species of birds, along with complete information on seven ecological, behavioral, and physiological variables that were entered in the multivariate analyses (Wasser and Sherman, Avian Longevity and Senescence)
Max longevity in wild
Max longevity in captivity
Decimals in the longevity data refer to months (e.g. 27.083=27 years, 1 month).
Abbreviations are as follows: breeding habitat: Grass (Grassland), Scrub (Scrubland), Wet (Wetland), Wood (Woodland); nest location: 1 (Ground), 2 (Shrub), 3 (Tree Branch), 4 (Cavity), 5 (Aquatic – floating on water); Diet: Omni (Omnivore), Herbi (Herbivore), Carni (Carnivore); Breeding insularity: M (Mainland), I (Island); Breeding latitude: 1 (0° to ± 30°), 2 (30° to ± 60°), 3 (60° to ± 90°); Migratory behavior: Y (Yes), N (No); Sociality: S (Colonial or Cooperative Breeder), N (Non-social Breeder).