How exceptional are the classic adaptive radiations of passerine birds?

Significance

Adaptive radiations—species descended from a common ancestor that are adapted to a wide variety of ecological niches—occur frequently on remote islands where a colonist finds few competitors in an unexploited ecological landscape. We examined recognized cases of island radiations among passerine birds, asking whether the morphological diversity exceeds that in a set of clades including some composed in part or wholly by continental species. Island clades were slightly more diverse than the other clades, but the focal adaptive radiations were not exceptional. Clade variation conformed to a normal distribution, suggesting that special circumstances, such as evolution of key innovations, are unnecessary to explain the evolutionary diversity of diverse clades.

Abstract

We investigated whether celebrated cases of evolutionary radiations of passerine birds on islands have produced exceptional morphological diversity relative to comparable-aged radiations globally. Based on eight external measurements, we calculated the disparity in size and shape within clades, each of which was classified as being tropical or temperate and as having diversified in a continental or an island/archipelagic setting. We found that the distribution of disparity among all clades does not differ substantively from a normal distribution, which would be consistent with a common underlying process of morphological diversification that is largely independent of latitude and occurrence on islands. Disparity is slightly greater in island clades than in those from continents or clades consisting of island and noninsular taxa, revealing a small, but significant, effect of island occurrence on evolutionary divergence. Nonetheless, the number of highly disparate clades overall is no greater than expected from a normal distribution, calling into question the need to invoke key innovations, ecological opportunity, or other factors as stimuli for adaptive radiations in passerine birds.

Adaptive radiations have long fascinated evolutionary biologists. Defined by Futuyma [(1), glossary] as “evolutionary divergence of members of a single phylogenetic lineage into a variety of different adaptive forms,” adaptive radiations embody the evolutionary origins of much of life’s diversity (2, 3). A sizable literature addresses the causes of evolutionary radiations (e.g., refs. 2 and 4–7), seeking to explain why some clades exhibit remarkable evolutionary diversification. A wide variety of hypotheses—including the evolution of key traits, the colonization of depauperate areas, and the extinction of dominant clades (6, 8–10)—ha been introduced to explain the extraordinary phenotypic diversity, as well as species richness, of clades undergoing adaptive radiation. However, one may reasonably ask whether widely studied adaptive radiations, such as the Darwin’s finches of the Galápagos Islands or the Anolis lizards of the West Indies, are in fact extraordinary in phenotypic variation and have diversified to a greater extent than similar-aged clades as a whole (11, 12).

The usual course of evolution is for clades to diversify, one species begetting many, and for the species within these clades to diverge adaptively as they compete among themselves for resources, contend with different predators and parasites, and occupy different ecological niches, thus encountering different selection pressures. As a result, one would expect any clade to exhibit some degree of adaptive variation among its constituent species (we note that members of a clade may evolve differences for nonadaptive reasons, but we restrict our discussion here to radiations in which the evolved differences result primarily from adaptation in response to divergent selection pressures). Because no two clades diversify in the same way, even under a common underlying stochastic process, some clades will become phenotypically more diverse (i.e., exhibit greater phenotypic “disparity”) than others. If the strength of selection—stemming from changes in resources, climate, competitors, predators, and myriad other factors—were to fluctuate in a stochastic manner, adaptive diversification in a clade might conform to a Brownian motion model of morphological change (13). A reasonable expectation for such a process is a normal or lognormal distribution of clade disparities. Accordingly, variation among clades with respect to phenotypic disparity might result from selective pressures drawn from the same underlying distribution.

One implication of a normal distribution of phenotypic disparity among clades would be that particular “adaptive radiations” are not necessarily special. Rather, they might simply occupy the tails of the distribution of phenotypic disparity among clades produced by the same underlying processes. As a result, clades that by chance experienced the greatest diversifying selective pressures might impress us most and be identified as adaptive radiations. However, one would not have to invoke special circumstances—key innovations, ecological opportunity afforded by colonization of empty landscapes, or extinctions of previous niche occupants—to explain the greater ecological or phenotypic disparity of these clades compared to that of other clades.

Ecological opportunity (6, 14) has long been thought to trigger adaptive radiation, an idea supported by the preponderance of recognized adaptive radiations on oceanic islands, where colonizing species encounter few competing, predatory, and parasitic antagonists. Such examples come from across the biological spectrum: e.g., Galápagos weevils (15), Hawaiian lobeliads (16), Bonin Island snails (17), Caribbean lizards (18), and Rift Lake cichlids (lakes being islands in a terrestrial sea [19]). Some of the best-known examples come from passerine birds (Aves: Passeriformes), particularly the Darwin’s finches (Geospizinae) of the Galápagos archipelago and the honeycreepers (Drepanidinae) of the Hawaiian islands (20–24).

Using these and other classic avian examples, we ask here whether renowned adaptive radiations truly are exceptional in the extent of their (presumably adaptive) phenotypic disparity (25, 26) and whether, more generally, the extent of radiation is greater in island clades than in those occurring on continents. We also ask whether the distribution of phenotypic disparity among bird clades departs from a normal distribution, which would suggest heterogeneity in the underlying causes of diversification. Finally, we ask whether this distribution reveals the existence of clades whose exceptional disparity requires special explanation.

Materials and Methods

Our overall approach involves generating a distribution of phenotypic disparity among clades, based on measurements of eight external morphological traits related to locomotion and diet (27–29). We define disparity for each clade in terms of the SD of morphological principal component scores among species within that clade (see below for details).

Selection of Clades.

We use two conventions for selecting clades to generate this distribution. In the first, we subjectively selected 192 phylogenetically independent clades of passerine birds from the Jetz et al. (30) ultrametric phylogeny (MayrAll_Hackett_25_mcc.tre, birdtree.org/subsets). These clades, which we call “taxon-defined” clades, include 2,627 species or about half of all passerine birds. The clades were selected on the basis of taxonomic distinction, i.e., generally referring to genera or larger subfamilial monophyletic taxa, rather than being of approximately similar age. We adopted this approach in part because the intervals between the stem and crown ages do not overlap among several of the insular clades of interest (e.g., Drepanidinae [Hawaiian honeycreepers], 19.34 Ma [stem] to 12.49 Ma [crown], versus Geospizinae [Darwin’s finches], 10.12 to 9.40 Ma). In this sample of taxon-defined clades as a whole, stem ages varied between 7.4 and 60.0 Ma, and crown ages varied between 6.9 and 39.0 Ma. These clades included between 2 and 83 (average, 13.7) species. Our morphological measurements were available (27) for 1,378 of the species in 190 clades, varying between 2 and 44 species (mean = 7.2 species) per clade, representing about half of the 2,627 species within these clades.

The 192 taxon-defined clades are listed in Dataset S1. Each is monophyletic in the Jetz et al. (30) tree. Also indicated are the stem and crown ages of each clade (Ma), the number of extant species, and the number of species for which eight external morphological traits were measured (see ref. 27). This is followed by the size disparity of the measured species within a clade, calculated as the variance across species within each clade of scores on the first principal component of a PCA of all species included in the analysis. The PCA was calculated from the covariance matrix of the eight log-transformed original measurements (28, 29). Finally, we include the shape disparity, which is the sum of the variances of the remaining seven principal component scores. Because we calculated a single PCA for all species in the analysis, and not for each clade individually, size and shape disparity within each clade are measured on the same axes of morphological variation. At least two species were measured in all but two of the clades (Myiopagis and Todirostrum, both Tyrannidae), providing estimates of morphological disparity for 190 clades.

We also indicated whether the distribution of each clade is primarily tropical (167 clades, 87%) or temperate (25 clades, 13%) and whether the species in each clade occur solely on an island (including archipelagoes; 20 clades, 10.4%), as opposed to exclusively within a continental region (142 clades, 73.9%; Dataset S1). We considered New Zealand, Madagascar, and New Guinea (but not Australia) to be islands. An additional 30 clades (15.6%) consisted of species whose predominant distribution occurs on islands or archipelagoes, but also includes noninsular taxa (see SI Appendix for distributional scenarios). We evaluated whether these clades were primarily radiations within an archipelago with limited dispersal to noninsular regions (11 clades, 5.7%), which we designated as Archipelago-Centered clades (AC), or widespread clades consisting of island species intermingled within a continental clade, which we considered to be Mixed Continental/Island clades (MCI, 19 clades, 9.9%, see SI Appendix for details).

Because of the subjectivity involved in delimiting the taxon-defined clades, we also identified a second, larger set of passerine clades, referred to as “age-defined” clades, which were all the passerine clades (two or more species) defined by a stem-crown interval that includes 15 Ma. That is, the crown ages of the age-defined clades are younger than ca. 15 Ma, but divergence between each age-defined clade and its closest sister taxon (stem age) exceeds ca. 15 Ma. Note that, by definition, these clades are phylogenetically independent—they are monophyletic and nonoverlapping. The 748 age-defined clades (Dataset S2) contain 5,598 species (all the passerines except for 348 species with “ages” in the Jetz et al. (30) phylogeny exceeding 15 Ma, i.e., clade size = 1), of which we have measured 1,555 species, including at least two species in each of 288 clades, allowing calculation of disparity. We also distinguished clades as tropical (627 clades, 84%) versus temperate (121 clades, 16%) and island (145 clades, 19%), Mixed Continental/Island (26 clades, 3.5%), or continental (577 clades, 77.1%) clades based on the predominant distributions of their species. Of the 288 age-defined clades with morphological measurements, 30 (10.4%) occurred only on islands, whereas 232 (80.5%) were entirely continental in distribution. As in the taxon-defined clades, we found 26 (9.0%) clades that included a mixture of insular and continental species. Of these, 15 (5.2%) clades were monophyletic radiations within an archipelago with evidence of dispersal to continental regions (AC clades). Another 11 (3.8%) consisted of continental clades with island species embedded within the radiation (MCI clades, see SI Appendix for details).

Several groups of birds stand out as classic cases of island radiations, including clades restricted to single islands as well as to island archipelagoes. Among the most celebrated of these are the Darwin’s finches (Geospizinae) of the Galápagos Islands, which have been the subject of many analyses of evolutionary diversification (e.g., refs. 20, 21, and 31–40). Other celebrated avian clades include the Hawaiian honeycreepers (Drepaninae; refs. 22–24 and 41–43), the birds-of-paradise of New Guinea (44–46), the vangas (Vangidae) of Madagascar (47), and the West Indian tanagers (Coerebinae: Loxigilla clade) (48, 49). We also included the mockingbirds and thrashers (Mimidae) of the Caribbean Basin as a less-known example of a small archipelagic radiation (50, 51).

These clades exhibit slight differences in their composition in the two analyses: for the Paradisaeidae (taxon-defined clade 85; Dataset S1), the age-defined clade (278; Dataset S2) does not include Lycocorax, which occurs in the taxon-defined clade; the taxon-defined Calicalicus clade of the Vangidae (97) is divided into two age-defined clades—Calicalicus/Oriolia (age-defined clade 310) and Leptopterus/Mystacornis (311); the taxon-defined Shetba clade of the Vangidae (98) is divided into two age-defined clades, Vanga/Euryceros (312), and Pseudobias/Shetba (313); the age-defined Mimidae clade (552) lacks Melanotis; the age-defined Geospizinae (clade 675) combines two clades that are distinct in the taxon-defined analysis, the West Indian tanagers in the genera Loxigilla, Tiaris, Coereba, and others (taxon-defined clade 171) and the Galápagos finches (taxon-defined clade 170).

Statistical Analyses.

We calculated values for Skew and Kurtosis using the functions “Skew” and “Kurt” from the R package “DescTools” package (52). Whether the values for skewness and kurtosis for size and shape disparity matched a normal distribution was assessed using the Jarque–Bera test [function “jarque.bera.test” in the package “tseries” (53, 54)]. We tested whether clades observed to exceed the 95th percentile could be considered extreme values (e.g., Taxon-defined clades 53 and 192) using a generalized extreme Studentized deviate (ESD) test (55). We used the “rosnerTest” function in the package “EnvStats” (56).

We compared size and shape disparity with generalized linear models as implemented in the R (version 3.2.2) statistical computing environment (57). We evaluated two models. The first included the distributional status of each clade (Island, Continental, Archipelago Centered, and Mixed Continental/Island) and whether clades were found in tropical or temperate environments as fixed effects and species numbers, stem age, and crown age as covariates. The second model only included the distributional status of each clade. We compared marginal means for size and shape disparity among the distributional categories using the function “emmeans” in the package “emmeans” (58). We present the results for size and shape disparity comparisons between Island and Continental clades in the main text. Those comparisons including Archipelago-Centered and Mixed Continental/Island clades are given in SI Appendix, Tables S2–S5. We used the Tukey method for adjusting P values for multiple comparisons.

Results

Size and shape disparity are significantly, but not strongly, correlated in both the taxon-defined (Table 1, r = 0.32) and age-defined clades (Table 1, r = 0.24). Among the taxon-defined clades, morphological disparity is positively related to stem and crown age, but is independent of the number of species in a clade. Among the age-defined clades, size and shape disparity are positively, but weakly, correlated with crown age, perhaps reflecting the reduced age variation among age-defined clades, and disparity increases significantly with the number of species per clade.

Table 1.

Correlations among the estimated clade variable for the taxon-defined clades (above) and age-defined clades (below) based on the Jetz et al. (30) tree

				Pearson correlation coefficients (r)
Taxon-defined clades	N	Mean	Std dev	Log(ShapeSD)	Stem age	Crown age	Log(Species)
Log(SizeSD)	190	−0.96	0.29	0.32	0.32	0.33	0.07
Log(ShapeSD)	190	−0.98	0.16		0.24	0.30	0.08
Stem age	192	21.87	8.96			0.84	−0.15
Crown age	192	16.51	6.35				0.08
Log (species)	192	2.27	0.82
Age-defined clades
Log(SizeSD)	288	−1.07	0.45	0.24	−0.02	0.12	0.25
Log(ShapeSD)	288	−1.02	0.19		0.04	0.14	0.20
Stem age	748	17.87	4.00			−0.02	−0.26
Crown age	748	11.82	2.41				0.52
Log(species)	1,100*	0.46	0.43

Note: Bold type indicates P < 0.0001; italic type indicates P < 0.05.

^*Including 348 clades with N = 1 species.

Distinctive and Classic Island Radiations.

All of the taxon-defined clades regarded here as distinctive and classic island “adaptive radiations” exhibited size and shape disparity above the median for all clades (Table 2, values in italics; binomial test of the probability that all clades are above the median: size disparity P < 0.008, shape disparity P = 0.0044); the birds-of-paradise (size and shape), the Shetba clade of vangids (size, Clades 97 and 98), and the Hawaiian drepanids (shape) were above the 95th percentile (bold type) (binomial test, P = 0.0004 for both size and shape).

Table 2.

Descriptive statistics pertaining to taxon-defined clades for stem and crown ages (Ma)

Clade		Taxon	Stem age	Crown age	Species	Measured	Size disparity	Shape disparity
	Mean		21.87	16.52	13.68	5.36	0.031	0.014
	Median		19.70	15.61	9	4	0.015	0.010
	90th %		33.30	24.73	30.9	9	0.072	0.027
	95th %		38.44	27.62	40.45	12.6	0.094	0.037
85	Paradisaeidae	Paradisaeidae	28.45	23.75	34	15	0.136	0.061
97	Vangidae	Calicalicus clade	25.00	17.99	5	4	0.071	0.019
98	Vangidae	Schetba clade	22.32	9.44	3	2	0.099	0.023
137	Mimidae	Caribbean endemics	19.48	16.21	9	8	0.028	0.020
159	Fringillidae	Drepanidinae	19.34	12.49	23	22	0.061	0.057
170	Thraupidae	Geospizinae	10.12	9.4	14	13	0.041	0.023
171	Thraupidae	Coerebinae	11.15	10.72	6	6	0.038	0.011

Numbers of species, and size and shape variances for all clades, and for seven clades representing examples of adaptive radiations, for which italicized values exceed the median for all clades, and bold-faced values exceed the 95th percentile for all clades. The frequencies of extreme variances among the island adaptive radiations are the same when compared only to continental clades.

The age-defined clades (Table 3) present a similar picture to the taxon-defined clades in that most of the clades representing classic island radiations exhibit disparity that is above the median for both size (7/8 clades) and shape (8/8 clades) (both P < 0.01), while three of the eight clades are above the 95th percentile with respect to size variation, and two of eight with respect to shape variation (both P < 0.01). Although the clades in this analysis are not identical to those in the taxon-defined analyses, the outlying clades (>95th percentile) are the same: birds-of-paradise (Paradisaeidae, clade 278) are exceptional for both size and shape, honeycreepers (Drepanidinae, clade 647) for shape, and the same two vangid clades (310, 313) for size.

Table 3.

Descriptive statistics for the age-defined clades for stem and crown ages

Clade	Clade name	Statistic	Stem age	Crown age	Species	Measured	Size disparity	Shape disparity
		Mean	17.98	12.81	12.18	4.61	0.024	0.013
		Median	16.63	13.35	7	3	0.011	0.009
		90th %	21.60	14.73	27	9	0.056	0.027
		95th %	24.26	14.88	38	14	0.097	0.036
278	Paradisaeidae		15.46	14.34	28	14	0.139	0.065
310	Vangidae1*		17.99	10.11	3	2	0.198	0.014
311	Vangidae2*		17.99	8.42	2	2	0.001	0.027
312	Vangidae3†		23.30	8.92	2	2	0.024	0.025
313	Vangidae4†		22.32	9.44	3	2	0.099	0.023
552	Mimidae		16.21	11.46	8	8	0.034	0.025
647	Drepanidinae		19.34	12.50	23	22	0.061	0.057
675	Geospizinae‡		15.11	14.51	77 ‡	41	0.031	0.025

^*These clades correspond to the taxon-defined Calicalicus clade (97).

^†These clades correspond to the taxon-defined Shetba clade (98).

^‡The Geospizinae in this table include West Indian tanagers in the genera Loxigilla, Tiaris, Coereba, and others, constituting a clade within which the Galápagos finches are embedded.

Numbers of species, and size and shape variances (disparity) for which we measured two or more species, and for eight clades representing examples of celebrated adaptive radiations, for which italicized values exceed the median for all clades, and bold-faced values exceed the 95th percentile for all clades.

The Distribution of Clade Morphological Variation.

We first tested whether any individual clades are statistical outliers with respect to morphological variation. To do so, we asked whether the distribution of morphological variance (disparity) is consistent with a normal distribution or whether some clades are outliers beyond the expected range of observations for a normal distribution representing a single underlying process.

For the taxon-defined clades (Fig. 1, Left), we examined the distribution of age-corrected size and shape disparity (given the significant correlations reported in Table 1 for both variables with both stem age and crown age). The distributions of the uncorrected, log-transformed size and shape disparities differed significantly from normality in the direction of leptokurtosis, i.e., extended tails, particularly toward smaller size (skew, −0.50 and −0.05; kurtosis, 1.28 and 1.29; Jarque–Bera test (53), JB = 19.5 and 11.9 chi-square distributed with 2° of freedom, both P < 0.01). The average of the log-transformed SD of the clade-level size variables (0.29) is nearly twice that of the clade-level shape variables (0.16).

Fig. 1.

Distribution of the size and shape disparities variables among the 190 taxon-defined clades (Top) and 288 age-defined clades (Below). Means for size and shape disparity are given as solid vertical lines.

The disparity values for one SD above the mean (84th percentile), and for the 95th and 99th percentiles of the distributions, were 0.27, 0.40, and 0.59 for size, and 0.15, 0.27, and 0.36 for shape. If the residuals were normally distributed, we would expect that about 10 of the 190 taxon-defined clades would exceed the 95th percentile and two would exceed the 99th percentile. With respect to the shape residuals, 10 clades were above the 95th percentile and two were above the 99th percentile; for the size residuals, the numbers were eight and two. Thus, the data suggest close adherence to a normal distribution. With respect to extreme values in a boxplot framework (>1.5 times the interquartile range below the first quartile or above the third quartile) (52, 53), only two clades—both continental—could be considered potential extreme values for logshape (#53, Menura/Atrichornis; #192, Promeropidae), and only #53, which is a small, but extremely old clade, would be considered extreme for logsize. We tested whether these clades constituted extreme values using an extreme value analysis based on the Rosner Test (55). None of the clades met the conditions for extreme values, which supports our conclusion that the data follow a normal distribution.

Among the 288 age-defined clades for which we had measured two or more species, the size distribution differed significantly from normality, being skewed toward small values (skew, −1.07; kurtosis, 1.46), whereas the shape distribution was more nearly normal (skew, −0.18; kurtosis, 0.26; Jarque–Bera tests, JB = 53.7 [size] and 1.6 [shape], chi-square distributed with 2° of freedom, P < 0.01 and P > 0.4) (Fig. 1, Right). The SD among the clade-level size variables (0.80) was nearly twice that among the clade-level shape variables (0.37) (for these analyses, we did not use age-corrected values because of the weak relationship between age and size and shape disparity reported above, which, of course, was the rationale for choosing clades of similar age). We would have expected about 15 clades to be above the 95% level if the variation were normally distributed. With respect to size variance, 14 were above the 95% level; with respect to shape variance, 15 were above the 95% level. Moreover, none of the values of the log shape variance or log size variance would be considered extreme in a boxplot context. Thus, the distributions of these two variables among clades appear to be close to log-normal, without evidence of outlier values, that is, of clades that deviate from a common underlying normal distribution of morphological variance.

To summarize, with respect to the generation of morphological diversity within clades, and in spite of the moderate kurtosis of size and shape variation among clades caused by extended tails of the distributions toward lower clade disparity, we cannot reject the hypothesis that most clades of birds have diversified according to a single underlying process.

Distributions of size and shape disparities among island and continental clades in both the taxon-defined and age-defined datasets are compared in Fig. 2 (for simplicity, we present results for clades designated as either Archipelago Centered (AC) or Mixed Continental/Island (MCI) in SI Appendix, Supplementary Material; these clades, which contain both island and continental species, reaffirm the results presented here; see SI Appendix, Supplementary Material). We tested the effects of place of occurrence for the log-transformed variances of both size and shape within clades using general linear models (function “glm” in R). These models included whether the clades were distributed in tropical versus temperate environments and the type of distribution as predictor variables. We also included the logarithm of the number of species per clade, and the stem and crown ages of the clade as covariates.

Fig. 2.

Distribution of the size and shape disparities comparing continental and island taxon-defined and age-defined clades. In terms of disparity of the continental distributions, the means of the island clades exceeded those of the continental clades by 0.67 and 0.30 (taxon-defined clades, size and shape) and by 0.30 and 0.25 (age-defined clades, size and shape) log₁₀ SD units. Means are given by the solid vertical lines.

For the taxon-defined clades, only island/continent (χ²_1,158 = 21.15, P < 0.0001) was associated with significantly higher values of size disparity. In particular, island clades had greater size disparity than continental clades (comparison of marginal means: t₁₅₄ = −4.47, P = 0.001). Tropical clades had a marginally higher value of size disparity than temperate zone species (χ²_1,154 = 3.25, P = 0.07). For shape variance, crown age (χ²_1,154 = 9.48, P = 0.002) and island/continent (χ²_1,154 = 15.29, P < 0.0001) had a significant influence. Shape disparity for island clades exceeded continental clades (t₁₈₂ = 3.91, P = 0.0001).

When the analysis was run comparing only the geographic distribution of clades, we obtained significant differences among the distribution classes for size variance (χ²_1,158 = 21.11, P < 0.0001) with a significant contrast between island and continental clades (t₁₅₈ = −4.59, P < 0.001). The analysis for shape disparity also resulted in a significant effect of clade distribution (χ²₁,₁₅₈ = 13.9, P = 0.0002). Island clades had shape disparity values greater than continental clades (t₁₅₈ = 3.74, P = 0.0003).

We applied the same linear model to understand the variance in the age-defined clades. The model yielded species number (χ²_1,256 = 13.31, P = 0.0002) and island/continent (χ²_1,256 = 9.29, P = 0.003) as significant predictors of increased size variance. For shape variance, the same model produced a parallel result: species numbers (χ²_1,256 = 9.67, P = 0.0018) and island/continent (χ²_3,256 = 13.42, P = 0.0002) were significant effects. Comparisons among the distribution categories revealed higher values of size disparity (t₂₅₆ = −3.05, P = 0.0025) and shape disparity (t₂₈₀ = −3.66, P = 0.003) for islands than continents.

When the analysis was run with only clade distribution as an effect, we obtained a similar result. Island clades exhibited greater morphological size variance than continental clades (χ²_1,260 = 6.30, P = 0.012). The analysis with shape variance also revealed island clades having higher disparity than continental clades (χ²_1,260 = 11.76, P = 0.006).

In summary, these analyses provide evidence for a modest increase in disparity in island clades relative to mainland ones; effects of species number, age, and tropical vs. temperate were more ambiguous.

Discussion

Evolutionary biologists hold two views concerning adaptive radiations. Traditionally, biologists seek particular explanations for the exceptional phenotypic and ecological disparity produced by adaptive radiations. Both intrinsic factors (e.g., the acquisition of a “key innovation”) and extrinsic factors (e.g., colonization of an empty ecological space and extinction of competing species) are believed to promote extraordinary diversification in a clade (9).

Alternatively, adaptive radiations may be seen as the tail end of the distribution of evolutionary diversification (59). Rather than being dichotomous—some clades being adaptive radiations and others not—this view portrays the extent of adaptive disparity among clades as a continuum; adaptive radiations are simply those clades with the greatest ecological and morphological disparity. Perhaps these radiations have resulted from special events affecting only them that promoted exceptional adaptive diversification. But an alternative possibility is that morphological diversification in the most diverse clades resulted from the same underlying process of evolution that is common, at least in a statistical sense, to all clades in the comparison group.

This second view is a reasonable null hypothesis. If selection pressures (from factors like climate, species interactions, and resource availability) fluctuate stochastically, then species in some clades would experience very little difference in selection, whereas species within other clades would subject to very different selective pressures from each other. As a result of this stochastic variation, members of some clades would adaptively differentiate to a much greater extent than members of other clades, producing great differences in clade-level disparity. With such variation in the distribution of selection pressures among clades, we would not need to invoke special evolutionary or historical events (e.g., the evolution of a key innovation and colonization of an isolated landmass) to explain the evolution of greater phenotypic disparity in some clades than in others.

We chose passerine birds to examine whether some clades present evidence of exceptional adaptive diversification. The evolutionary diversity of birds has been extensively studied, and many clades are textbook examples of adaptive radiation. Ecological opportunity on oceanic islands, in particular, is often invoked to explain the diversity of these clades (12), although key innovations are also sometimes postulated (9, 33, 47).

We took three approaches to test these ideas. First, we examined classic avian examples of adaptive radiation. Second, we analyzed the distribution of phenotypic disparity among bird clades. Third, we focused on two factors thought to promote adaptive radiation—diversification on islands and in the tropics—and asked whether clades exhibiting these factors are more diverse.

A number of conclusions follow from these analyses. First, clades identified as adaptive radiations are not qualitatively distinct from other passerine clades in the sense of having been drawn from a different distribution of size and shape disparity. Some classic bird adaptive radiations—e.g., birds-of-paradise and Hawaiian honeycreepers—are truly extreme in their degree of phenotypic disparity (i.e., beyond the 95th percentile of within-clade disparity) and all are above the median value of phenotypic disparity. Others, however, are less exceptional. Notably, Darwin’s finches and their continental tanager relatives rank only 39th out of 190 taxon-defined clades in size disparity and 32nd in shape disparity. Moreover, some unheralded continental clades lie in the upper tail of the phenotypic disparity distribution, among them Menura (taxon-defined clade 53; 1st in size disparity and 2nd in shape disparity), the tropical American cotingas (Cotingidae: taxon-defined clade 5; 5th in size and 18th in shape), the genera Frederickena and Mackenziaena in the antbird family Thamnophilidae (taxon-defined clade 27; 7th in size and 3rd in shape), and woodcreepers of the genus Lepidocolaptes (taxon-defined clade 30; 10th in shape, but only 133rd in size). Overall, the analyses do not make a statistically supported case that the set of clades included here as “classic” adaptive radiations is notable compared to the overall distribution of within-clade disparity, particularly when one recognizes that these examples of adaptive radiation have been singled out in the literature precisely because of their high disparity (i.e., even a stochastic process of diversification will produce some clades with great disparity).

Second, the distribution of phenotypic disparity among clades of birds conforms closely to a normal distribution. This raises the possibility that no special explanation for apparently exceptional diversity is necessary. If the natural selection pressures that drive the evolution of morphology fluctuate in direction and magnitude in a Brownian-motion way, then we might not have to invoke exceptional events—the acquisition of a key innovation, ecological opportunity, and extinction of competing species—to explain the extensive diversification of some clades (or the extreme lack of diversification in others).

Of course, a normal distribution of disparity does not preclude the possibility that the relatively high disparity of particular clades is the result of one (or more) of these factors; key innovations or ecological opportunity could account for the relatively high phenotypic disparity of birds-of-paradise and Hawaiian honeycreepers. Rather, the failure to reject a normal distribution simply indicates that those factors, if they did operate, are not strong enough to distort the overall normal shape of the distribution among all clades, nor to make clades affected by these factors stand out as extreme exceptions in terms of their great disparity. This viewpoint would suggest that our test lacks power and that particular clades may have been affected by different evolutionary processes, but the deviations they produced were not substantial enough to produce a non-normal distribution of clade disparities. In other words, our analyses cannot reject the idea that a spectrum of processes (e.g., ecological opportunity, key innovations, genetic drift, and fluctuating natural selection) combine to produce a normal distribution of clade disparities. Because a normal distribution is not mechanistically connected to particular processes, failure to reject a normal distribution does not preclude the possibility that many different processes have operated across clades.

We recognize this possibility and suggest that further research should explore the adequacy of the normal distribution as a null expectation for clade disparities. Nonetheless, our findings strongly caution against the widespread interpretation that large disparity is prima facie (i.e., “based on the first impression; accepted as correct until proved otherwise”) evidence that a clade has been affected by evolutionary factors not experienced by other bird clades.

Our message is not, however, entirely nihilistic. Our analyses do indicate that occurrence on islands promotes phenotypic diversification, in accord with the proposed positive role of increased ecological opportunity (9, 15). The overall effect of islands on disparity is relatively small—not even substantial enough to produce a bimodal distribution of clade disparity when all clades are combined—but the difference does indicate that something about occurring on islands tends to promote phenotypic diversification. This effect is reinforced when we consider clades that harbor insular and continental taxa (SI Appendix, Supplementary Material). Clades that diversified on archipelagoes and experienced dispersal to continental habitats were more similar to island radiations. In contrast, clades that ancestrally occurred to a great extent on the mainland but also include an island radiation are similar to entirely continental clades in size and shape disparity.

Although our results provide support for an island effect—perhaps the result of ecological opportunity—driving adaptive diversification, they also make clear that simply picking out diverse groups and labeling them as adaptive radiations may be insufficient or even misleading. To complement the approach developed here, we need detailed studies of particular cases, examining patterns of natural selection (31) and, if possible, making ecological opportunity a quantifiable, rather than heuristic, term (6, 14).

The analytical approach employed here represents a potentially important contribution to the study of adaptive radiation. We suggest that statistical approaches should be taken to identify how exceptional particular clades are within a universe of comparable clades. The results presented here may surprise many in that they exhibit little evidence for a set of clades qualitatively distinct from the distribution of all clades. However, our analyses should be considered preliminary, guided by selection of clades using subjective taxonomic criteria (the taxon-defined clades) or an arbitrary clade age (the age-defined clades). Further analysis of additional groups of avian or other clades, using larger morphological datasets and more refined statistical approaches, would be welcome next steps.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Data, Materials, and Software Availability

All study data are included in supporting information.