Subspecies as a Meaningful Taxonomic Rank in Avian Classification

Abstract

Dissatisfaction with the subspecies unit of classification is, in part, a consequence of the failure of many of those who have described subspecies to follow the conceptual definition of the subspecies, namely that it should represent diagnosable units. The antiquity of the descriptions of most subspecies (median year of description of currently recognized subspecies estimated to be 1908–1909) means that the majority predated any statistical tools for assessing diagnosability. The traditional subspecies concept, as originally construed, identifies minimum diagnosable units as terminal taxa, and I suggest that it is thus essentially synonymous with the phylogenetic species concept. Therefore, both must deal with the fundamental difficulties inherent in using diagnosability as a criterion. Application of monophyly as a criterion for taxon rank at the population level has inherent difficulties. An advantage of the biological species concept is that it incorporates, in its classification of taxa, ass...

Full text

Ornithological Monographs
Volume (2010), No. 67, 62–78
© The American Ornithologists’ Union, 2010.
Printed in USA.
62
CHAPTER 6
SUBSPECIES AS A MEANINGFUL TAXONOMIC RANK
IN AVIAN CLASSIFICATION
Abstract.—Dissatisfaction with the subspecies unit of classification is, in part, a consequence of
the failure of many of those who have described subspecies to follow the conceptual definition of
the subspecies, namely that it should represent diagnosable units. The antiquity of the descriptions
of most subspecies (median year of description of currently recognized subspecies estimated to be
1908–1909) means that the majority predated any statistical tools for assessing diagnosability. The
traditional subspecies concept, as originally construed, identifies minimum diagnosable units as ter-
minal taxa, and I suggest that it is thus essentially synonymous with the phylogenetic species con-
cept. Therefore, both must deal with the fundamental difficulties inherent in using diagnosability as a
criterion. Application of monophyly as a criterion for taxon rank at the population level has inherent
difficulties. An advantage of the biological species concept is that it incorporates, in its classification
of taxa, assessments of gene flow and reproductive isolation, which are critical components of the
evolutionary process. Critics of the biological species concept persistently overlook the fact that it
includes the subspecies rank as a necessary component of that concept for distinct populations within
biological species. Analyses that require terminal taxa can, with care, be conducted under the biologi-
cal species concept using subspecies plus monotypic species. Critics of the biological species concept
with respect to its application have missed the biological and political disadvantages of treating mini-
mum diagnosable units as the primary unit of conservation concern. Human perception is in accord
with ranking such minimum diagnosable units below the species rank; socially and scientifically,
humans consider diagnosable units of other humans as distinct groups but not separate species.
J. V. Remsen, Jr.1
Museum of Natural Science, Louisiana State University, Baton Rouge, Louisiana 70803, USA
1E-mail: najames@LSU.edu
Ornithological Monographs, Number 67, pages 62–78. ISBN: 978-0-943610-86-3. © 2010 by The American Ornithologists’ Union.
All rights reserved. Please direct all requests for permission to photocopy or reproduce article content through the University of
California Press’s Rights and Permissions website, http://www.ucpressjournals.com/reprintInfo.asp. DOI: 10.1525/om.2010.67.1.62.
Las Subespecies como un Rango Taxonómico Significativo en la
Clasificación de las Aves
Resumen.—En parte, la insatisfacción con la unidad de clasificación de subespecie es consecuen-
cia de que muchos de aquellos que han descrito subespecies no han seguido la definición concep-
tual de la subespecie como una unidad diagnosticable. La antigüedad de las descripciones de la
mayoría de las subespecies (la mediana del año de descripción de las subespecies actualmente
reconocidas se estima en 1908–1909) significa que la mayoría precedió a las herramientas estadís-
ticas para evaluar la diagnosticabilidad. El concepto tradicional de subespecie, como se concibió
originalmente, identifica unidades diagnosticables mínimas como taxones terminales, por lo que
sugiero que esencialmente es sinónimo del concepto filogenético de especie. Por lo tanto, ambos
deben lidiar con las dificultades fundamentales inherentes vinculadas con el uso del criterio de
diagnosis. La aplicación de la monofilia como un criterio para la clasificación de los taxones al
nivel poblacional tiene dificultades inherentes. Una ventaja del concepto biológico de especie es
que incorpora, en su clasificación de los taxones, evaluaciones del flujo génico y del aislamiento
reproductivo, que son componentes fundamentales del proceso evolutivo. Las críticas al concepto
Key words: species concepts, species definitions, subspecies definitions.
OM67_06.indd 62 4/6/10 6:30:50 PM

SUBSPECIES IN AVIAN CLASSIFICATION 63
biológico de especie persistentemente pasan por alto el hecho de que éste incluye el rango de sub-
especie como un componente necesario para poblaciones diferentes dentro de la especie biológica.
Los análisis que requieren taxones terminales pueden ser conducidos, con cuidado, bajo el concepto
biológico de especie usando subespecies y especies monotípicas. Las críticas del concepto biológico
de especie con respecto a su aplicación han pasado por alto las desventajas biológicas y políticas de
tratar a las unidades diagnosticables mínimas como las unidades principales de preocupación con-
servacionista. La percepción humana coincide en clasificar estas unidades diagnosticables mínimas
por debajo del rango de especie; social y científicamente, los humanos consideran unidades diag-
nosticables de otros humanos como grupos distintivos pero no como especies separadas.
Whether the subspecies rank in classification
is considered useful depends on whether one’s
concept of species includes room for geographi-
cally non-overlapping, diagnosable units within a
species. Can a species be subdivided into distinct,
biologically meaningful units? Should such units
be formally named? As noted by all who have
written about the classification of organisms, im-
posing a categorical scheme, such as the Linnaean
system of classification, on the pattern of continu-
ous variation produced by evolutionary processes
is doomed to be unsatisfactory. As noted by Strese-
mann (1936:157), “Whoever wants to hold firm
rules, should give up taxonomic work. Nature is
too disorderly for such a man.” Empirical exam-
ples can be mustered that defy the tidy either/or
demands of any species concept. Yet human per-
ception, dominated by categorical thinking, uses
such schemes to produce the vocabulary of labels
needed for communication. In short, biological
classification attempts to inflict an unrealistic cat-
egorical scheme on the patterns produced by a dis-
orderly, fundamentally noncategorical process.
Controversy over the utility and definition of
the subspecies rank in such a categorical classifica-
tion has a long history, with episodic reappraisals
(e.g., Wiens 1982), yet the category survives in al-
most all modern classifications of birds. This sur-
vival, since the mid-1800s, is presumably driven
by a perception among most humans that the cat-
egory that we term “species” can include within it
named subpopulations to identify nonclinal geo-
graphic variation. This, in turn, may follow from
our own widespread, long-standing perception of
the nature of the species Homo sapiens, in which
pronounced, nonclinal geographic variation is
included within that species rather than each dis-
tinct group being considered a separate species.
What Is a Subspecies?
Conceptual Definitions
To recognize nonclinal intraspecific geographic
variation in animals, some taxonomists have ap-
plied trinomials as subspecies names since at least
1844 (fide Simpson 1961). The concept behind
subspecies definitions centers on the existence of
separate units or geographic units within the rank
of species. Historically, dissatisfaction with rank-
ing every distinctive geographic population as a
species was the catalyst for the use of trinomials,
which were regarded a century ago as a radical
and progressive step in classification (Knox 2007).
Definitions of subspecies extracted from standard
references and textbooks (Table 1) are founded on
the theme that a “species” may consist of subunits
that differ from each other in diagnosable ways
yet share the characters attributed to the species
itself. I combine these ideas into the following
definition: “geographic populations diagnosable
by one or more phenotypic traits.”
The theme that unifies these definitions of
subspecies is that subspecific names identify dis-
tinct population units: they are phenotypic pre-
dictors of past or current genetic continuity, the
phenotypic analogue of genetic markers. Hennig
(1966:102) stated that the goal of species-level
taxonomy was to relegate to subspecies rank “all
vicarying reproductive communities.”
That the subspecies category has biological
meaning is reinforced by the observation that
populations known to be reproductively isolated,
and thus considered species by any definition,
typically differ from close relatives in the same
kinds of phenotypic characters and patterns, but
to a greater degree (e.g., as seen among subspe-
cies that intergrade where in contact). Whether
such characters and patterns represent causation
or correlation is an open question.
A definition of subspecies as “geographic pop-
ulations diagnosable by one or more phenotypic
traits” is a simple statement concerning the cur-
rent geographic distribution of distinct pheno-
typic traits. This definition makes no assumptions
about whether the traits are adaptive or whether
they represent populations that are incipient spe-
cies, and thus makes no predictions concerning
the future. Assuming that a phenotypic trait has
a genetic rather than environmental basis, sub-
species boundaries imply that all individuals of
OM67_06.indd 63 4/6/10 6:30:50 PM

64 ORNITHOLOGICAL MONOGRAPHS NO. 67
Table 1. Subspecies definitions from textbooks and reference works.
Source Definition
Mayr et al. 1953, Mayr 1963 Geographically defined aggregates of local populations which differ
taxonomically from other such subdivisions of a species
Mayr and Ashlock 1991:43, 430 An aggregate of local populations of a species inhabiting a geographic
subdivision of the range of the species and differing taxonomically
[differing by sufficient diagnostic characters] from other populations
of the species
Futuyma 1979 A set of populations of a species that share one or more distinctive features
and occupy a different geographic area from other subspecies
Futuyma 2005:213, 356 A recognizably distinct population, or group of populations, that occupies
a different geographic area from other populations of the same species;
populations of a species that are distinguishable by one or more
characteristics and are given subspecific names
Strickberger 2000 A taxonomic division of a species often distinguished by special phenotypic
characters and by its origin or localization in a given geographic region
the subspecies share the genes responsible for the
diagnostic trait, which arose in a common ances-
tor, and, thus, form a monophyletic group with
respect to those genes. Whether they also form a
monophyletic group with respect to other gene
trees is an open question. As I discuss below,
however, application of the term “monophyly” at
the population level is problematic.
What Is a Subspecies?
Operational Definitions
Mayr et al. (1953) provided objective, quanti-
tative definitions of subspecies based on degree
of overlap that can be applied across taxa. They
outlined why using simple linear overlap in mea-
surements, for example, overemphasizes extreme
individuals in a population and overestimates
true population overlap. They also discussed
various interpretations of the “75% rule” as the
threshold for naming subspecies. Although one
interpretation is that only 75% of the individu-
als of each sample have to be correctly classified,
the rule as defined by Amadon (1949), Mayr et al.
(1953), and Patten and Unitt (2002) is based on
standard deviations from the mean of normally
distributed data. Depending on which metric is
applied, in essence these definitions mean that
90–97% of the individuals of one population
must be distinguishable from the equivalent per-
centage of the other population to be considered
subspecies under the somewhat misleadingly
named 75% rule.
As emphasized by Mayr et al. (1953) and Pat-
ten and Unitt (2002), defining subspecies solely
on the basis of statistically significant differences
in population means is an unfortunate misinter-
pretation of the conceptual definition. Given
large enough sample sizes, the means of any two
populations likely differ significantly (>95%),
even though actual overlap can be nearly com-
plete, and so statistically significant differences
in the means alone provide almost no informa-
tion on how distinctive two populations are in
terms of diagnosability, the key theme of the
conceptual definitions of subspecies. The prob-
lem is that the conceptual definitions emphasize
the population as a whole, not the individuals
that constitute it, and so statistically significant
differences between means can be interpreted
as diagnosability if the population is the unit of
analysis.
Although the 75% rule has a long history in or-
nithology, its application has been erratic at best.
For example, it is generally not mentioned as a
criterion for recognizing subspecies in classifica-
tions (e.g., American Ornithologists’ Union 1957,
Dickinson 2003) or in any of the Handbook of the
World series (del Hoyo et al. 1992–2008). It is not
possible to tell how many of the subspecies cur-
rently recognized in such sources would qualify
as subspecies under the 75% rule, but it is certain
that many subspecies, especially in North America,
would not qualify as valid taxa under this rule,
particularly those defined by mensural differ-
ences. From personal experience in attempting
to use subspecies diagnoses, such as the keys in
the Birds of North and Middle America series (Ridg-
way and Friedmann 1901–1950), I predict that
more than 75% of North American subspecies
OM67_06.indd 64 4/6/10 6:30:51 PM

SUBSPECIES IN AVIAN CLASSIFICATION 65
taxa delimited by mensural data would not sur-
vive application of the 75% rule.
Although Patten and Unitt (2002) used 75% as
their target level of degree of diagnosability in
deference to tradition, they advocated a higher
level, 95%, as a standard for diagnosability. I
also propose that this level of diagnosability be-
come the operational definition of diagnosable.
McKitrick and Zink (1988) gave reasons why
aiming for 100% diagnosability for phylogenetic
species is conceptually and methodologically
unreasonable. Also, geographic sampling to de-
termine diagnosability in the case of parapatric
populations must exclude any zones of intergra-
dation in statistical treatments because heavy
sampling from that zone would eliminate any
potential diagnosability of two populations. To
avoid circularity, delimitation of zones of inter-
gradation must be objective (e.g., Harrison 1993),
because, hypothetically, one could expand the
area considered the zone of intergradation until
diagnosability reaches 95% for populations on ei-
ther side of it. Of course, the amount of intergra-
dation occurring at parapatric zones of contact is
of considerable biological and taxonomic interest,
especially in determining whether taxa are sub-
species or full species (Mayr and Ashlock 1991).
By incorporating a quantitative operational
definition into the conceptual historical defini-
tion, I produced the following definition, modi-
fied from Futuyma (2005), including an explicit
statement that diagnosability refers to individu-
als that comprise the population: A subspecies is
a distinct population, or group of populations,
that occupies a different breeding range from
other populations of the same species; individu-
als are distinguishable from those other popula-
tions by one or more phenotypic traits at the 95%
level of diagnosability.
Application of this rigorous definition would
result in the synonymization of many subspe-
cies names in North America and elsewhere
where broad geographic patterns of smoothly
clinal differences in coloration and, especially,
morphometrics have been artificially categorized
as subspecies. Quantitative analyses of this geo-
graphic variation typically found that much or
most of this variation, at least in terms of morpho-
metrics, cannot be partitioned into diagnosable
units (e.g., Power 1969, Behle 1973, Tacha et al.
1985, Aldrich and James 1991, Wood 1992, Rising
et al. 2001, Rising et al. 2009). The implied agenda
of much of the work in the first half of the 20th
century was that all geographic variation had to
be described in a categorical way, namely by use
of subspecies names (see Knox 2007). However,
patterns of geographic variation in phenotype
provide valuable insights into population struc-
ture and the process of evolution, regardless of
whether the variation can be apportioned into
diagnosable units (James 1970, Zink and Remsen
1986). Note also that this definition is based on
phenotypic traits in plumage and morphology
with an assumed genetic basis, not on other phe-
notypic traits such as behavior or physiology, nor
on genetic markers not expressed in the pheno-
type; one could make a case for recognizing any
diagnosable, geographically distinct genetic unit
with a subspecies label. Note also that this defini-
tion is incomplete with respect to distinguishing
subspecies from species, which I address below.
To refer to populations or individuals that repre-
sent extremes of clinal variation or populations
that do not meet statistical thresholds of diagnos-
ability, for convenience one could use an infor-
mal vocabulary using the formerly recognized
subspecies names in quotes, followed by “grade”
(e.g., “the ‘nigrideus’ grade” for the darker north-
easternmost populations of the American Robin
(Turdus migratorius). Note that use of a 95% di-
agnosability criterion applies only to two-way
comparisons, as in the 75% rule, not to multiple
simultaneous comparisons. It also applies only to
populations in the breeding ranges; an empirical
consequence of such a rigorous standard (if in-
dividuals from intergrade zones were excluded)
would also allow assignment of individuals from
the nonbreeding range to breeding population
with a high level of statistical certainty.
A persistent criticism of the subspecies con-
cept is that analysis of different characters may
produce different subdivisions (e.g., Wilson and
Brown 1953). In other words, the characters are
not distributed in a concordant geographic man-
ner; for example, three characters might show
geographic variation, but each character could
show three different patterns that would delimit
subspecies boundaries in three conflicting ways.
The existence of such conflict is inevitable, and
application of the subspecies concept in such
cases is unwarranted. However, my impression,
based on examining many hundreds of primar-
ily Neotropical bird species over the past 30 years
with respect to distinct plumage characters, is that
such conflicts are greatly outnumbered by exam-
ples of concordance. For example, if three traits
OM67_06.indd 65 4/6/10 6:30:51 PM

66 ORNITHOLOGICAL MONOGRAPHS NO. 67
show nonclinal geographic variation, all three
may not show breaks at the same point, but ac-
tual conflict in where the breaks are is infrequent.
For exceptionally thorough quantifications of pat-
terns of geographic variation in morphometrics
and plumage that showed strong concordance,
see Johnson (1980) and Cicero (1996). By contrast,
smoothly clinal variation regularly shows con-
flicting patterns, with trends in, for example, tail
length showing a different pattern from that of,
for example, back color.
Ridley (2004) used lack of concordance in geo-
graphic variation as grounds for essentially dis-
missing the importance of the entire subspecies
unit of classification. He used the example of cli-
nal size variation in North American House Spar-
row (Passer domesticus) to illustrate clinal variation
and then noted that there was no reason to expect
clines in other characters to match the body-size
cline. However, he did not point out that no sub-
species have been described or recognized in
North American House Sparrows because the
variation is smoothly clinal rather than discrete,
so his example is inappropriate, and his premise
is flawed. If the geographic pattern of variation in
distinct characters produces conflicting patterns,
then this implies complex underlying population
structure that may not be amenable to diagnoses
and, therefore, subspecies names. However, John-
ston and Selander (1964:549) found that “color
differences between samples are both marked
and consistent, permitting 100 percent separa-
tion of specimens from two localities” but did not
name any subspecies because they did not think
that the situation was temporally stable.
Subspecies versus Phylogenetic Species
Given that conceptual definitions of subspe-
cies have always emphasized diagnosable units,
how do they differ from phylogenetic species
other than that, under the biological species
concept, many diagnosable units are ranked as
species? Cracraft (1983:170) defined phylogenetic
species as “the smallest diagnosable cluster of
individual organisms within which there is a pa-
rental pattern of ancestry and descent.” Cracraft
did not define how these differ from subspecies
but emphasized the heterogeneous nature of the
results of applying the subspecies concept. He
urged abandoning the subspecies rank in classifi-
cation, without detailing how it differs from phy-
logenetic species. The methodological difficulties
that produced the heterogeneity in units called
subspecies are assumed to disappear if Cracraft’s
phylogenetic species concept is adopted—when,
in fact, delimitation of “diagnosable clusters” en-
tails all the methodological problems that com-
plicate subspecies delimitation. Any renaming of
all minimum diagnosable units as species would
require determining what units are actually diag-
nosable and at what statistical thresholds of di-
agnosability.
Cracraft (1983) pointed out the biological spe-
cies concept lacks equivalency among the units
called species. However, the same problem per-
vades species defined under the phylogenetic
species concept, in which, for example, species
reproductively isolated from all other lineages,
including syntopic sister taxa, are treated as the
same taxonomic unit as populations that dif-
fer only in the possession of a single diagnostic
character and cannot coexist syntopically with
sister taxa. By contrast, use of the subspecies rank
within biological species as the unit of analysis
reduces the problems of heterogeneity because
population units diagnosed only by minor plum-
age differences are not treated as the same unit
as lineages known or inferred to be sealed from
other lineages by reproductive isolation.
Another criticism of the biological species
concept is that biological species are not the ap-
propriate unit for biogeographic and speciation
analyses (Cracraft 1983). I agree. The appropriate
units are indeed minimum diagnosable units—
that is, subspecies under the biological species
concept. That subspecies can be used productively
for such analyses is shown inadvertently by none
other than Cracraft (1983), who used subspecies
names in outlining his methods for determining
areas of endemism. Cracraft (1985) later also used
lengthy lists of trinomials to demarcate and name
areas of endemism in the Neotropics. Those ar-
eas, defined by the terminal taxonomic unit of the
biological species, namely subspecies, are still the
standard nomenclature for Neotropical biogeo-
graphic analyses, thereby demonstrating the util-
ity of the subspecies unit of classification.
Other definitions of phylogenetic species re-
peat the essence of Cracraft’s phylogenetic species
concept, with the emphasis on diagnosability and
common ancestry, and they do not address how
this definition differs from that of the subspecies.
Futuyma (2005) and Freeman and Herron (2007)
also reported the definition of the phylogenetic
species concept without explaining how it differs
OM67_06.indd 66 4/6/10 6:30:51 PM

SUBSPECIES IN AVIAN CLASSIFICATION 67
from the rank of subspecies within the biological
species concept. The explicit conceptual defini-
tion of phylogenetic species is that they represent
monophyletic units, whereas subspecies are not
defined explicitly with respect to monophyly. In
practice, however, when phenotypic characters
define phylogenetic species, the issue of mono-
phyly is often ignored. In fact, the phylogenetic
species concept’s method of the minimum diag-
nosable unit, when applied to phenotypes, can be
applied to inanimate objects and is not inherently
phylogenetic (Johnson et al. 1999). Further, when
genetic criteria are used to define monophyly,
these criteria are typically just one or two loci,
typically non-recombining mitochondrial DNA
(mtDNA) genes, and monophyly with respect to
other loci is not addressed (see below).
Furthermore, using unique or even multiple
characters to identify minimum diagnosable units
does not guarantee monophyly of the taxa in ques-
tion if three or more populations are involved. If
rates of character evolution are unequal, then
some populations will become diagnosable be-
fore others, leading to paraphyletic groupings of
populations that have not become diagnosable,
similar to the “metaspecies” and “plesiospecies”
problems (Donoghue 1985, Olmstead 1995, Will-
mann and Meier 2000). If three populations have a
known history C+(A+B), if B is the first population
to acquire a diagnostic character, leaving C and A
with nothing but ancestral character states, then
even if the true history were known, there would
be no way to avoid a paraphyletic taxon A+C
if the strict rules of diagnosability are followed.
If their geographic ranges are linear, the taxa are
sedentary, and the central taxon acquires an apo-
morphy first, then at least we would be suspicious
that the character distribution represents unequal
rates of character evolution. In fact, such a linear
array provided a clue suggesting that many such
cases in Andean birds represented cases of un-
equal character acquisition that would potentially
mislead phylogeny (Remsen 1984). In many cases,
however, the populations’ ranges are not linear,
and in such cases, geography cannot provide hints
that the populations without diagnostic characters
form a paraphyletic taxon.
Advocates of the phylogenetic species con-
cept often promote its adoption because it makes
the fundamental unit of classification “histori-
cal taxa” (Zink and McKitrick 1995), whereas
in the biological species concept non-sisters can
be treated as a single species. As noted above,
morphology-based applications of the phyloge-
netic species concept do not necessarily produce
historical taxa. History is a continuum, and the
exercise of recognizing which historical units
within this continuum are named taxa is inher-
ently arbitrary. Worse, at the population level,
defining historical units depends on which char-
acters or which loci are thought to represent
the true history. As acknowledged by Zink and
McKitrick (1995), it is well known that use of any
one set of markers can lead to misrepresentations
of history (Tateno et al. 1982, Neigel and Avise
1986, Pamilo and Nei 1988). Only by knowing the
gene trees of a large number of polymorphic loci
can the true population history be reconstructed,
and even then, incomplete lineage-sorting may
complicate resolving a single history even if en-
tire genomes are sequenced (Pollard et al. 2006).
Further, for all populations with topographically
and climatologically heterogeneous ranges, this
history likely dates no farther back than the most
recent pulse in the cycle of fragmentation and
secondary contact.
Using diagnosability as a criterion for nam-
ing taxa has inherent methodological problems
that affect phylogenetic species and subspecies
(under the biological species concept) equally,
for four reasons. (1) Any diagnosability level is
arbitrary. Because diagnosability is a continuum,
from 0 to 100%, any cutoff is inherently arbitrary
and cannot be defended conceptually (Johnson et
al. 1999). Setting the threshold at 95% is a reason-
able level because of the widespread use of that
arbitrary level for statistical “significance.” None-
theless, the consequence is that two populations
that are, for example, 95% diagnosable are given
taxon status, whereas those at 94% are not and are
included in the same unnamed category as those
population samples diagnosable at 0%. (2) A corol-
lary of arbitrary diagnosability is that the outcome
is driven in part by sampling. The closer the di-
agnosability approaches the threshold, the higher
the chance that an increase of one additional indi-
vidual in the sample will determine the outcome;
thus, such an addition to the sample could change
the ranking from unnamed taxon to phylogenetic
species or subspecies without any true change in
the biology and history of the populations. (3) The
geography of sampling is critical to the outcome
if the character assessed shows any geographic
variation (Zink and Remsen 1986). Past gene
flow or residual geographic variation in the once-
continuous populations makes it essential that
OM67_06.indd 67 4/6/10 6:30:52 PM

68 ORNITHOLOGICAL MONOGRAPHS NO. 67
sampling be focused on geographically proximate
populations. And (4) diagnosability is driven by
the resolution of the technique used (see Collar
1997, Avise 2004).
Zink (2006) objected to my definition of sub-
species and phylogenetic species as synonyms
(Remsen 2005) because, in essence, some mini-
mum diagnosable units under the biological
species concept would be ranked as species if
reproductively isolated from other such units.
Zink’s (2006) argument is largely semantic, be-
cause a biological species that is monotypic (con-
tains no units ranked as subspecies) would still
be treated as equivalent to the subspecies unit in
those analyses for which minimum diagnosable
units are the appropriate unit of analysis. In other
words, an analysis using minimum diagnosable
units under the biological species concept would
include all taxa ranked as subspecies plus all
monotypic species. The difference between the
biological species concept and the phylogenetic
species concept is not in defining minimum diag-
nosable units but in the ranking of some of those
units as species. Under the biological species con-
cept, 4,677 (48%) of the 9,722 species in Dickinson
are monotypic (D. Lepage pers. comm.)
Monophyly at the Population Level?
The original conceptual theme of the phy-
logenetic species concept is that minimum di-
agnosable units are not only diagnosable, but
monophyletic (Cracraft 1983). As is now well
known, the problem is that at the population level,
monophyly is difficult to define and determine
(e.g., de Queiroz and Donoghue 1990, Wheeler
and Nixon 1990, Davis and Nixon 1992). Only if
all gene trees within a series of populations that
share a common ancestor have topologies that
do not conflict can a single population be labeled
unambiguously monophyletic. Genetic data (e.g.,
Avise 1989) confirm what common sense pre-
dicts: the turbulent history and complex popu-
lation genetics of real-world situations are often
unlikely to produce true monophyly because of
incomplete lineage-sorting and gene flow among
populations that are not reproductively isolated.
Gene tree topologies, superimposed, probably
look more like a tangled net than a tree (Degnan
and Rosenberg 2009). Further, it is now well un-
derstood that under some circumstances the gene
trees of independently segregating loci are not ex-
pected to recover the true species tree (Rosenberg
and Tao 2008) and that postdivergence gene flow
may make reconstructing species trees from gene
trees particularly problematic (Takahata and
Slatkin 1990, Eckert and Carstens 2008). Add to
this the historical likelihood of repeated phases
of expansion, range fragmentation, and second-
ary contact, and the use of the term “monophyly”
becomes problematic. For a particularly well doc-
umented example of how a single gene tree can
misrepresent species trees of buntings in the ge-
nus Passerina, see Carling and Brumfield (2008).
In part because of this, Hennig (1966: fig. 4) rec-
ognized and illustrated this problem graphically,
did not apply the term “monophyly” below the
species level, and used the reasoning of the bio-
logical species concept in his definition of species.
Although Hennig used characters to label species
in his diagrams illustrating cladistic methodol-
ogy and is thus widely cited as an advocate of
the phylogenetic species concept, Hennig clearly
considered reproductive isolation the essential
component of speciation (e.g., Hennig 1966:54).
Reproductive isolation is the necessary first step
toward true monophyly.
Even with respect to a single gene, monophyly
at the population level differs fundamentally
from monophyly at higher levels because it can
be ephemeral, perhaps typically persisting only
during the refugial phase of range expansion and
contraction cycles, and even then being vulner-
able to dispersal-generated gene flow. There-
fore, the objection to subspecies or biological
species because they are not monophyletic (e.g.,
McKitrick and Zink 1988) is not condemning.
Paraphyly and polyphyly at the population level
are predicted, and empirically demonstrated,
to be widespread (Funk and Omland 2003). For
example, Hull et al. (2008) showed that Swain-
son’s Hawk (Buteo swainsoni) is paraphyletic with
respect to Galapagos Hawk (B. galapagoensis)
in terms of mtDNA; however, there is no other
biological support for merging B. galapagoensis
into B. swainsoni or for recognizing two or more
species within traditionally defined B. swainsoni.
Further, possession of a diagnostic character, the
criterion needed for phylogenetic species rank, is
no guarantee of monophyly with respect to other
genes. For example, Swainson’s Hawk has a suite
of diagnostic phenotypic characters despite its be-
ing a paraphyletic unit with respect to Galapagos
Hawk. Labeling clusters of populations as species
on the basis of monophyly with respect to single
gene trees indicates monophyly only with respect
OM67_06.indd 68 4/6/10 6:30:52 PM

SUBSPECIES IN AVIAN CLASSIFICATION 69
to that gene tree, and not necessarily with respect
to the the population or species tree (Edwards et
al. 2005, 2007). Additionally, in practice, far too
few individuals are typically sampled to deter-
mine whether two populations are monophyletic
with respect to the loci surveyed (for an exam-
ple of how differences in sample size can affect
conclusions concerning population monophyly,
see Brumfield 2005). In fact, an earlier study of
Swainson’s and Galapagos hawks (Riesing et al.
2003) had reported that the two were reciprocally
monophyletic only because too few individuals
had been sampled (Hull et al. 2008). In summary,
the putative advantage of the phylogenetic spe-
cies concept in establishing monophyletic units
as the fundamental unit of taxonomy is appeal-
ing rhetoric but elusive reality. Hennig’s (1966)
restriction of the term “monophyly” to levels of
classification above that of species under the bio-
logical species concept reflects remarkable wis-
dom given the state of knowledge of population
genetics at that time.
What Is a Species?
One cannot discuss subspecies without also
defining species. The controversy over species
concepts is obviously too large and complex to
treat here; see Coyne and Orr (2004) for a com-
prehensive review. De Queiroz (2005a, b) pointed
out that all species concepts share the property,
explicit or implicit, that the unit called “species”
represents the uniquely biological property of a
separately evolving metapopulation lineage. The
problem is how to apply that concept and which
criteria are used to delimit species. Although de
Queiroz (2005a, b) tried to present his broadly
defined species concept as a solution, he offered
no real operational definition with respect to
explicitly defining such a unit; in fact, at one ex-
treme, a pair of individuals colonizing an island
and successfully reproducing could fit the defi-
nition of “separately evolving metapopulation
lineage” after a single generation. Nonetheless,
de Queiroz’s (2005b) simple diagram of the split-
ting and subsequent divergence of populations
crisply illustrates the underlying problem of set-
ting criteria to demarcate species boundaries. His
use of continuous shading aptly emphasizes the
continuum of degrees of divergence and the in-
herently arbitrary decisions necessary. The island
example above would represent the first point
past divergence on his time axis. Therefore, some
level of subjectivity inevitably influences one’s
choice of criteria.
As expressed more fully elsewhere (Johnson
et al. 1999, Remsen 2005), I favor definitions of
species based on a fundamental process of evo-
lution at the population level, namely gene flow
or lack of it; that is the essence of the biological
species concept. My support for process-based
definitions—rather than being “blind allegiance”
to the biological species concept, the accusation
leveled by Peterson et al. (2006)—is based on rec-
ognition that severe diminishment or cessation
of gene flow is clearly critical to diversification.
Personally, I regard the biological species concept
as an imperfect attempt at inflicting a typology
on a continuum; however, I dislike even more
any other categorical scheme proposed so far
(e.g., various versions of the phylogenetic species
concept). Rather than become disillusioned at the
failures, I recommend rejoicing in the underlying
complexity that the failures reveal.
The primary operational problem of the bio-
logical species concept, as emphasized by Ernst
Mayr from the outset (e.g., Mayr 1942b), is in deal-
ing with ranking allopatric differentiated popu-
lations. Here, I note that human cognition deals
directly with this problem in recognizing differ-
entiated but reproductively fully compatible units
within Homo sapiens as conspecifics. This predates
science, much less the Modern Synthesis, in that
even the earliest historians treated allopatric dif-
ferentiated populations of humans as “people,”
rather than as some other type of species. Ter-
ritoriality and combat, typical manifestations of
intraspecific competition but relatively rare in
interspecific competition, were expressions of
that cognitive framework. Therefore, in treating
distinct interpopulational differences as part of
the same species, the biological species concept
has a subjective appeal that the phylogenetic spe-
cies concept lacks. The phylogenetic species con-
cept could also produce some unknown number
of species within Homo sapiens, a result refuted
by human behavior long before modern societal
influences.
The problem of assigning rank to differenti-
ated allopatric populations is not as intractable
as is often portrayed. By placing the degree of
differentiation in a comparative phylogenetic
framework, namely comparing degree of differ-
entiation in the allopatric form to that seen in
closely related sympatric or parapatric popula-
tions, a reasonable and testable hypothesis can be
OM67_06.indd 69 4/6/10 6:30:52 PM

70 ORNITHOLOGICAL MONOGRAPHS NO. 67
made concerning whether the allopatric form has
or has not differentiated to the degree shown by
related forms that do or do not freely interbreed
(Miller 1955; Mayr 1969, 1996; Mayr and Ashlock
1991; Helbig et al. 2002; Futuyma 2005). Any ar-
bitrariness involved in assignment of taxon rank
through this process is no greater than that inher-
ent in assessing minimum diagnosable units un-
der the phylogenetic species concept.
Reproductive Isolation
The importance of reproductive isolation in
guaranteeing independent evolutionary lineages
has been emphasized by many authors, including
Hennig (1966) and Cracraft (1983). Proponents
of the phylogenetic species concept can seem
schizophrenic toward reproductive isolation,
first acknowledging its importance, then dismiss-
ing its importance. For example, McKitrick and
Zink (1988:6) stated that “the ‘closure’ or sealing
of a gene pool is therefore an important evolu-
tionary event.” Yet they explicitly denied a role
to reproductive isolation in ranking taxa because
interbreeding is a “primitive trait” or “ancestral
character” (e.g., Zink 2006). The ability to inter-
breed could perhaps be construed as an ancestral
character, but empirical evidence in birds sug-
gests a severe limit to interbreeding in terms of
time since divergence: Price and Bouvier’s (2002)
survey indicated that postzygotic incompatibili-
ties begin to originate by ~2 million years after
divergence. Moreover, free interbreeding (i.e.,
nonassortative mating with hybrids having equal
fitness to pure parental) provides a highly reliable
indicator of a close relationship. Empirically, it is
limited in birds to populations that have diverged
to a limited degree; if not sisters, such popula-
tions are members of a lineage that abruptly re-
place each other geographically. In other words,
the ancestral component of free interbreeding is
highly restricted to parapatric representatives of
a single lineage. Zink and McKitrick (1995) reiter-
ated the importance of reproductive isolation and
considered studies of it valuable, but they also ar-
gued that it should not have a role in delimiting
species. Missed altogether is that reproductive
isolation or its absence governs the distribution
of characters that delimit the phylogenetic spe-
cies concept’s minimum diagnosable units in
sympatric and parapatric taxa; therefore, the pat-
tern of diagnosability is a product of the process
dismissed as an “ancestral character.” As noted
previously (Avise and Wollenberg 1997, Remsen
2005), denying a role in classification to the most
important threshold in the history of a lineage
seems incongruous if that classification is sup-
posed to be based on the history of a lineage.
Zink and McKitrick (1995) implied that some
proponents of the biological species concept
place theoretical emphasis on reproductive isola-
tion because the lack of it, namely hybridization,
means that the two populations may eventually
homogenize. Similarly, Zink (2006) portrayed the
biological species concept as placing importance
on the “potential future outcome of current inter-
breeding.” Rather than making such predictions,
the classification of two differentiated, freely in-
terbreeding populations as one biological species
represents only a statement concerning the cur-
rent interaction of the two populations, namely
that in terms of mate selection and recognition,
individuals of both populations treat each other
as equivalents, regardless of any previous his-
tory of differentiation. It does not necessarily
predict the future (although considered by some
a hallmark of a mature research field, not specu-
lation), nor does it necessarily group historical
taxa. However, it represents important informa-
tion concerning the current situation in terms
of individual behavior and its consequences for
population genetics. In summary, such popula-
tion interactions provide taxonomist-free data
on whether (or to what degree) two populations
consider themselves “the same” or “different.”
At least some of the controversy over the im-
portance of reproductive isolation is caused by
disagreement over, or misrepresentation of, the
definition of reproductive isolation. Mayr ’s defi-
nition of the biological species concept empha-
sizes free interbreeding, widely interpreted as
nonassortative mating in contact with no reduc-
tion in hybrid fitness. In contrast to any criteria
based on diagnosability, the advantages of these
criteria are (1) that ranking depends on the bio-
logical behavior of the individuals involved and
(2) that any change in that behavior has conse-
quences for gene flow. That patterns of mate
choice may change temporally or geographically
is inevitable, and these differences will generate
problems for anyone who expects a typological
categorization scheme to nimbly handle all real-
world variation. Populations that interbreed but
still mate assortatively (e.g., no hybrid swarm in
contact zone) are treated as separate species un-
der most interpretations of the biological species
OM67_06.indd 70 4/6/10 6:30:53 PM

SUBSPECIES IN AVIAN CLASSIFICATION 71
concept. If new data reveal that existing taxa
classified by the biological species concept are
actually freely interbreeding, that classification
should be changed.
Subspecies as Straw Men
and Phylogenetic Species
Finding that existing classifications of subspe-
cies are defective at some level is not an indict-
ment of the subspecies concept itself, no more
than a reanalysis of a phylogenetically based
classification that found problems with a previ-
ous classification would be an indictment of the
phylogenetic species concept. Those who attack
subspecies as a taxon rank consistently miss the
distinction between a concept and the correct ap-
plication of that concept. An everyday analogy
would be to blame the car, not the mechanic, for
a botched repair job. For example, McKitrick and
Zink (1988:11) advocated abandonment of the
subspecies rank largely because of the histori-
cal inconsistency in its application and admitted
that properly characterized subspecies—namely,
in their words, those “distinct from other popu-
lations in one or more characters”—”would be
called [phylogenetic] species by our criteria” (as
echoed by Zink 2006).
Using existing subspecies classifications as an
indictment of anything is disingenuous. The vast
majority of such classifications has not been sub-
jected to a modern, quantitative analysis since
their original presentation, often the Check-list
of the Birds of the World series (Peters 1934–1987),
many dating as far back as the 1930s. More re-
cent synopses, such as Dickinson (2003) and the
Handbook of the Birds of the World series (del Hoyo
et al. 1992–2008), largely repeat the initial classi-
fications in Peters’s Check-list unless subsequent
studies have altered them. Although the 1960s and
early 1970s saw a wave of quantitative studies,
particularly in North America, few such studies
have been published since then. Thus, the vast
majority of subspecies-level classifications remain
mostly unchanged from those of Peters’s Check-list
and are maintained largely by historical inertia, a
diminishment in this type of biodiversity science,
and a lack of adequate material to readdress his-
torical hypotheses.
However, many critiques of the subspecies
concept seem to assume that these classifications
undergo some sort of constant, modern, quantita-
tive scrutiny. As pointed out previously (Remsen
2005), the majority of subspecies were described
in a prestatistical era. In fact, the term “statistics”
and even the simplest statistical analyses, such
as the t-test, postdate the majority of subspecies
descriptions. The percentage of subspecific clas-
sifications in the Peters’s Check-list that have ever
been subjected to statistical evaluation is minute,
perhaps <1%. Therefore, the chances that any of
these classifications would not require modifica-
tion after a modern reanalysis are also minute.
I am unaware of any quantitative reanalyses of
existing subspecies designations that have not
produced modifications of existing subspecies
classifications. For example, see Cicero (1996),
who found that 4 of the 10 subspecies in the Bae-
olophus inornatus complex were not diagnosable,
and Patten and Pruett (2009), who found that
only 25 of 51 subspecies of Melospiza melodia rep-
resented diagnosable units.
To illustrate these points, I plotted (Fig. 1) the
date of the type descriptions of all subspecies
currently recognized by Dickinson (2003) for two
bird families, Parulidae and Pycnonotidae, of
similar size but contrasting features. The family
Parulidae is restricted to the New World, much of
its diversity is at temperate latitudes, and many
species are highly migratory. The family Pycnon-
otidae is restricted to the Old World tropics and
includes no highly migratory species. Despite
the differences, the chronology and pattern of
subspecies descriptions are remarkably similar.
Fifty percent of all descriptions predate the first
publication of Student’s t-test (1908), much less
its widespread use in ornithology, 70% predate
Fisher’s (1930) seminal work on population ge-
netics, and 79% predate Huxley’s (1942) book
on the Modern Synthesis. Therefore, to use such
classifications as ammunition to attack subspe-
cies as a concept is a classic straw-man approach
that is counterproductive to elucidating the pat-
terns of diversification and the processes that
produce them. Any critique of the subspecies
unit as a concept using empirical results should
start by determining which named subspecies fit
the conceptual definition. Failure to apply such a
conceptual definition to subspecies designations
over the past century has, in my opinion, directly
catalyzed the origin of the phylogenetic species
concept.
In contrast to subspecies designations, the
phylogenetic species concept benefits from hav-
ing few empirical applications to examine on any
large scale. A reasonable prediction is that if all
OM67_06.indd 71 4/6/10 6:30:53 PM

72 ORNITHOLOGICAL MONOGRAPHS NO. 67
classifications started using a phylogenetic spe-
cies classification as of today, then 100 years from
now those results would be viewed with the same
disdain directed at current subspecies classifi-
cations. Published analyses using phylogenetic
species as units already provide ample fodder
for criticism, with a near absence of quantitative
rigor in determining whether their units actually
represent minimum diagnosable units. Given the
importance in many analyses of using minimum
diagnosable units, whether called subspecies or
phylogenetic species, the first step requires a rig-
orous determination of what those units are.
However, Navarro-Sigüenza and Peterson’s
(2004) listing of bird species for Mexico based on
diagnosable units is quantitatively inferior to that
of Robert Ridgway’s volumes from the early 1900s
that cover the same area (Remsen 2005), although
Peterson and Navarro-Sigüenza (2006) assured
us that unpublished analyses supported their
designations. Peterson and Navarro-Sigüenza
(2006:886) also assured us that their 2004 clas-
sification was a “consistent taxonomy” that was
“based on the same criteria as in all other clades,”
yet their methodology and criteria remain un-
specified. Likewise, Peterson’s (2006) synopsis
of diagnosable units in Philippine birds rests on
unspecified sample sizes (noted as “woefully
small” for many populations) and qualitative as-
sessments. See Collar (2007a) for a full critique of
Peterson’s (2006) approach. Simply dismissing all
trinomial nomenclature and then labeling as spe-
cies all populations that by qualitative inspection
appear diagnosable is not an acceptable program
for assessing biodiversity; see Collar (1997) for
similar comments on fundamental problems with
Cracraft’s (1992) revision of a single family, the
Paradisaeidae, based on the phylogenetic species
concept. In fairness to all these attempts, reevalu-
ation of the diagnosability of currently described
subspecies, especially for a rich avifauna such as
that of Mexico, is a daunting, monumental task
that will require detailed research, and Navarro-
Sigüenza and Peterson have made a noble start.
Unfortunately for biodiversity assessment, these
kinds of baseline analyses of geographic variation
Fig. 1. Historical pattern of dates of descriptions of subspecies in the Parulidae and Pycnonotidae. The data
plotted are the publication years of the type descriptions for all subspecies currently recognized by Dickinson
(2003), not including, of course, the type description of the species.
OM67_06.indd 72 4/6/10 6:31:00 PM

SUBSPECIES IN AVIAN CLASSIFICATION 73
are not considered groundbreaking research.
Nonetheless, casual, qualitative inspection of
study skins is no longer an acceptable practice for
taxonomic revisions, whether the taxa are labeled
“subspecies” or “phylogenetic species.”
Some biogeographic analyses using phyloge-
netic species as units start with the assumption
that certain described subspecies accurately rep-
resent minimum diagnosable units, declare them
to be species, and then proceed with the analysis.
These analyses typically do not report sample
sizes or the geography of their sampling distribu-
tion, much less character analyses, diagnosability
indices, or anything else that would permit repli-
cation. Notable recent exceptions are the analyses
of McKay (2008) and D’Horta et al. (2008). Prior
to any phylogeographic analysis, they began with
a quantitative analysis of geographic variation in
plumage characters to define minimum diagnos-
able units. As noted previously, the antiquity of
most subspecies names makes it inevitable that
many will fail diagnosability tests. However,
analyses that do not include the characters used
to diagnose the taxa are unlikely to address di-
agnosability adequately. For example, Drovetski
et al. (2009) quantitatively analyzed geographic
variation in breast plumage in the currently rec-
ognized species of North American rosy-finches
in the genus Leucosticte but omitted those plum-
age characters (face pattern) formally (e.g., Ridg-
way 1901, MacDougall-Shackleton et al. 2000,
Johnson 2002, Johnson et al. 2002) used to diag-
nose the taxa. Snow (1997) pointed out that in-
complete geographic sampling and small sample
sizes for many taxa make it necessary to study
geographic variation and taxonomy in detail be-
fore determining what constitutes minimum di-
agnosable units.
Environmental Induction
In his widely used textbook, Gill (2007:575)
stated that “geographical differences in size or
color may be due directly to environmental dif-
ferences rather than evolved genetic differences
among populations” but provided no further
details or citations. One likely source of such
statements is a tiny number of studies that have
documented minor environmental effects on body
size and shape in relation to the genetic compo-
nent (James 1983, Larsson and Forslund 1992,
Leafloor et al. 1998), although most such stud-
ies have not found an environmental component
(see Merilä and Fry 1998). Many subspecies have
been described on the basis of measurements that
reflect overall body size. Regardless of whether
such differences have an environmental compo-
nent, I suspect that many or most of these subspe-
cies will be shown to fail diagnosability tests. The
vast majority of such subspecies have not been
analyzed using any test of degree of overlap, and
their validity often rests on differences between
means and various qualitative assessments of
the ranges.
The other potential source of Gill’s (2007)
statements concerning environmental effects is
the relationship between diet, or other measures
of condition, and feather pigmentation or struc-
ture. Environmental effects on the ability to ex-
press appropriate coloration are widely known,
in that poor condition or disease may effect the
coloration of individuals within a bird popula-
tion. Coloration based on carotenoids can be
affected strongly by diet because carotenoid pig-
ments must be acquired from food (reviewed by
McGraw 2006a), and the expression of carotenoid-
based coloration can be altered by environmental
conditions, such as parasite load (reviewed by
Hill 2006). As for melanin-based coloration, the
most widespread source of coloration in birds,
documentation of environmental effects is not
as clear-cut, and some experiments have failed
to find an effect of diet on melanin production
or expression (e.g., Gonzalez et al. 1999, Buch-
anan et al. 2001, McGraw et al. 2002). Nonethe-
less, because melanin is synthesized from amino
acid precursors using metabolic energy and their
deposition is influenced by at least four classes of
hormones (McGraw 2006b), the potential remains
for an effect on its production owing to general
health and nutrition. As for structural colors, lim-
ited experimental data suggest that nutrition dur-
ing molt may affect their expression (Hill 2006),
but such experiments are limited mainly to glossy
black species, are largely correlational, and have
not addressed potential confounding influences
of age (Prum 2006). Nonetheless, given the com-
plex pathways involved and the extraordinary
structural precision required to produce normal
coloration (reviewed by Prum 2006), the poten-
tial for environmental effects would seem large.
Whether coloration is based on nanostructure or
pigments, environmental effects on individuals
within a population are highly likely. (And this is
discounting the ways in which the environment
can alter plumage through time, such as through
OM67_06.indd 73 4/6/10 6:31:00 PM

74 ORNITHOLOGICAL MONOGRAPHS NO. 67
fading and wear.) That environmental differences
might also produce region-wide effects on certain
populations of a species remains a potential ex-
planation for geographic variation in coloration
and body size in birds. However, data that docu-
ment a link between environmental effects and
among-population differences in coloration are
lacking.
Evidence for natural selection on color shades
and patterns is reasonably strong (e.g., Burtt
1981, Rohwer and Ewald 1981, Prum 1997, Ne-
gro et al. 1998, Dumbacher and Fleischer 2001,
Mumme 2002, Tickell 2003). Certainly, the strong
associations between patterns of coloration and
various ecological and social factors (for reviews,
see Bortolotti 2006, Dale 2006) imply that natu-
ral selection on the underlying genetic basis of
these patterns is widespread. Dramatic seasonal
changes in plumage coloration in some species,
typically associated with changes in social sys-
tem, also imply strong selection (although envi-
ronmental effects caused by a seasonal shift in
food supply are not necessarily ruled out by such
correlations).
Subspecies and Conflicts
with Gene-based Phylogenies
Several recent papers have attacked the utility
of subspecies by comparing current subspecies-
level classification with patterns of diversification
shown by mtDNA (e.g., Zink et al. 2001). A mis-
match between units defined by mtDNA versus
subspecies is then proclaimed as evidence that
subspecies mask patterns of diversity or obscure
analyses of the process of historical diversifica-
tion (Zink 2004).
Conflicts between mtDNA trees and subspe-
cies units may result from faulty delineation of
subspecies boundaries, because most have not
been critically or quantitatively examined (see
above). However, if the subspecies boundaries
represent diagnosable units, then I am unaware
of any model of evolution that predicts perfect
concordance between diagnosable phenotypic
units and any single gene tree, particularly those
of presumably neutral loci. Lost in the discussion
of such conflicts is that (except for populations
without any history of fragmentation and sec-
ondary contact) gene trees and population trees
not only differ, but also are expected to do so
because of the influences of incomplete lineage-
sorting and gene flow (for review, see Coyne and
Orr 2004). This is especially true for the most fre-
quently analyzed genes, those of mtDNA, which
are matrilinearly inherited as a single linkage
unit. Empirically, mtDNA markers may do as
well as any in tracking population history (Zink
and Barrowclough 2008), but to uncritically treat
an mtDNA gene tree as equivalent to the true
population history should be termed “mtDNA
myopia.” For example, Zink (2004:563) stated that
“subspecies should be judged to fail as meaning-
ful units if they do not predict the evolutionary
history of the populations they represent,” but
in Zink’s view mtDNA phylo-groupings repre-
sent the only history worth recognizing taxo-
nomically, without recognizing that an mtDNA
phylogeny is merely a gene tree. A population
marked by a phenotypically diagnosable char-
acter, provided that character has a genetic basis,
also shares a common history but on a different
time-scale. For subspecies units to show perfect
concordance with an mtDNA gene tree, each
subspecies would also have to have a unique
haplotype (or haplotype lineage), an unrealis-
tic expectation. Even so, Phillimore and Owens
(2006) showed that Zink’s estimates were an order
of magnitude too low because of sampling bias
and that broader sampling indicated that more
than a third of the taxa ranked as subspecies were
monophyletic even by the highly restrictive and
unrealistic criterion of mtDNA haplotypes. The
title of Zink’s (2004) paper proclaimed that sub-
species obscured biological diversity; however,
one could also make a case that using mtDNA
phylogroups as taxonomic units obscures biodi-
versity because it ignores biologically important,
phenotypic markers of recent population history.
Under Zink’s extreme view, some diversity even
at the species level would be erased, with most
Galápagos finches merged into a few monotypic
species because their mtDNA gene trees are not
reciprocally monophyletic (Zink 2002). Described
as “an unfortunate reliance on a single, poten-
tially misleading molecule” by Grant and Grant
(2006), such a treatment as single species would
ignore the reproductive isolation and divergence
of multiple lineages within this radiation.
Researchers who do not find concordance
between genetic data and subspecies boundar-
ies often proclaim that such subspecies are not
genetically distinct. Two fundamental problems
beset such statements. First, such studies typi-
cally analyze one or two genes, often mitochon-
drial—that is, a tiny fraction of the genome. The
OM67_06.indd 74 4/6/10 6:31:00 PM

SUBSPECIES IN AVIAN CLASSIFICATION 75
appropriate qualifier for such statements would
be that a subspecies is not genetically distinct
with respect to whatever number of genes was
analyzed. Second, if a subspecies is diagnosable
by phenotypic characters (external manifesta-
tions of genetic characters), then indeed it is also
likely genetically distinct, but the gene(s) that
control those characters have not been located or
analyzed. If two or more populations share the
same phenotypic characters that have arisen by
common selection pressure, then their grouping
into a single taxon would mislead phylogenetic
classification; this is where mtDNA or other ge-
netic markers can elucidate the true population
history that phylogenetic classification requires.
Subspecies as Impediments
to Conservation
Some (e.g., Hazevoet 1996, Sangster 2000,
Peterson 2006) have claimed that ranking diag-
nosable units as species under the phylogenetic
species concept or a similar concept rather than
as subspecies under the biological species con-
cept benefits conservation. See Collar (1996, 1997,
2007a), Garnett and Christidis (2007), and Winker
et al. (2007) for opposing views. A benefit of the
biological species concept to conservation is that
it provides a degree of triage in terms of prioritiz-
ing resources at the global level. Restricting the
species rank to populations known to be repro-
ductively isolated or to have diverged to a level
comparable to that shown by reproductively
isolated populations (i.e., species by anyone’s
definition) allows limited conservation resources
to be concentrated on those populations. For ex-
ample, if one had a limited amount of funding
to be divided evenly among Caribbean parrot
species in the genus Amazona, using the classi-
fication based on the biological species concept
would divide those funds among species that all
differ strongly from one another and are species
by any reasonable criterion. By contrast, elevat-
ing all diagnosable subspecies to species rank
under the phylogenetic species concept would
give equivalent taxonomic rank and funding, for
example, to Amazona leucocephala hesterna (en-
demic to Cayman Brac and differing from nearby
A. l. caymanensis of Gran Cayman only in hav-
ing a larger patch of red in the belly plumage) as
to the bizarrely plumaged, highly distinctive A.
guildingii of St. Vincent. Advocates for conserva-
tion on Cayman Brac naturally would be pleased
with such an outcome, and so it is no surprise
that among the most vocal advocates for the phy-
logenetic species concept are those devoted to
the conservation of small areas or islands (e.g.,
Hazevoet 1996), whose cause benefits from rais-
ing every endemic subspecies to species rank. A
more global view, however, would be that a pri-
oritization scheme based in part on taxon rank is
beneficial in that populations diagnosable only
by characters that do not impede on gene flow,
i.e., taxa ranked as subspecies under the biologi-
cal species concept, do not receive the resources
allocated to taxa ranked as species under the that
concept.
The other criticism of the use of subspecies in de-
fining conservation units is that many do not cor-
respond to “historically significant groups” (Zink
2004). However, these groups are typically delim-
ited only by patterns of shared mtDNA haplo-
types (e.g., Zink et al. 2001). Whether such groups
are the only historically significant groups, how-
ever, is open to discussion. Because these genetic
markers are assumed to be neutral, by definition
they have no biologically meaningful manifesta-
tion. Further, because of their matrilineal pattern
of descent and because of the widely recognized
problem of incomplete lineage sorting, these hap-
logroups represent only the history of perhaps
one or two non-recombining genes (Edwards and
Bensch 2009). Although such markers are useful
tools for tracking aspects of population history,
phenotypic markers also have the potential to do
the same. Moreover, in contrast to haplotype dif-
ferences, phenotypic markers have the potential
to be biologically meaningful and should thus be
of greater conservation concern (Crandall et al.
2000). Differences in pattern and coloration, for
example, frequently correspond to abrupt discon-
tinuities in gene flow in birds, a taxonomic class
in which sexual selection has played a key role
in diversification; their more subtle manifestation
as diagnosable characters that mark subspecies
boundaries gave rise to the phrase “incipient spe-
cies” for some subspecies. To ignore this aspect
of geographic variation and population biology
only because of lack of correspondence to neu-
tral mtDNA markers in vogue today should be
regarded as myopic by those interested in pat-
terns of biodiversity or the identification of units
of conservation concern—or, indeed, the process
of evolution.
A particularly disingenuous criticism of the
biological species concept as an impediment to
OM67_06.indd 75 4/6/10 6:31:00 PM

76 ORNITHOLOGICAL MONOGRAPHS NO. 67
conservation is the claim that it masks biodiver-
sity. For example, Peterson (2006) denounced the
biological species concept for overlooking nu-
merous distinct populations but did not mention
that under this concept all of those populations
are named, as subspecies, and overlooked only
if one restricts an analysis to the species rank.
Thus, Peterson (2006) found much higher levels
of species richness and unrecognized or under-
appreciated patterns of endemism by application
of a diagnosability-based species concept; how-
ever, he did not point out that an analysis that in-
cluded subspecies would have revealed the same
patterns that he “discovered.”
Application of the phylogenetic species con-
cept produces two potentially severe problems
for conservation. First, opponents of conserva-
tion would quickly discover that the definition of
species had been changed to elevate more taxa to
higher threat levels, with accusations of manipu-
lation of the rules. Changing the definition would
only fuel the suspicions of conservation oppo-
nents that scientists have abandoned objectivity
in favor of a pro-conservation agenda. Second,
elevating to species rank many taxa diagnosable
only by characters that conservation opponents,
the general public, and most biologists would
justifiably label as trivial could diminish confi-
dence in conservation science, undermine the
credibility of taxonomists, and erode support for
programs to protect threatened species.
Subspecies Are Overlooked
as a Component of Biological Species
De Queiroz and Donoghue (1988:334) con-
cluded that “no one species concept can meet
the needs of all comparative biologists.” I sug-
gest that use of a biological species concept that
identifies minimum diagnosable units as subspe-
cies spans more of those needs than is appreci-
ated. Debates over the merits of species concepts
based on whether they emphasize reproductive
isolation or minimum diagnosable units overlook
that subspecies, an integral part of the biological
species concept, are its minimum diagnosable
units. Criticizing the biological species concept
for not allowing analyses of basal evolutionary
units overlooks that the subspecies rank is an in-
tegral part of the concept. The biological species
concept encompasses units that fit the conceptual
definition of phylogenetic species but calls these
minimum diagnosable units subspecies rather
than species (if they are not ranked as biological
species). Proponents of the phylogenetic species
concept would point out that regardless of con-
ceptual definitions, in practice many subspecies
are not diagnosable units. As discussed above,
this (1) is largely the consequence of incorrect ap-
plication of the definition and (2) has to be dealt
with regardless of whether these units are called
subspecies or species. In fact, Phillimore et al.
(2007) showed that analyses of subspecies as an
index of intraspecific geographic differentiation
within a species yield sensible results with re-
spect to biogeographic influences on intraspecific
variation. Under the biological species concept,
classification with diagnosable units provides
two levels of information: one that emphasizes
genetic discontinuities (species) and another that
emphasizes geographic units within the species
identified by diagnostic characters (subspecies).
Analyses that require terminal taxa can use popu-
lations ranked as subspecies (e.g., Cracraft 1985),
whereas analyses based on active or potential
barriers to gene flow can use the species rank.
Geographic variation not partitioned into diag-
nosable units may occur within taxa ranked ei-
ther as species or subspecies under the biological
species concept.
A recurring misconception in some recently
published papers is that the phylogenetic species
concept reveals diversity and the biological spe-
cies concept obscures it. For example, Reddy’s
(2008) application of the phylogenetic species
concept to Pteruthius, currently considered to
consist of 5 species under the biological species
concept, first required determining which of the
23 recognized subspecies were diagnosable units;
that is the same procedure that would be neces-
sary under a modern reevaluation of the genus.
Although the geography of sampling and sample
sizes were not reported, Reddy found that 19 taxa
were diagnosably distinct. She then claimed that
this was “almost a four-fold increase in recog-
nized diversity”; in fact, all of that diversity was
recognized under the biological species concept, 5
as species and the other 14 as subspecies of those
species. In terms of overall taxonomic diversity,
this application of the phylogenetic species con-
cept actually reduced the number of recognized
taxa by some 15%. Once nondiagnosable taxa are
identified and eliminated (a problem shared by
all species concepts), the differences are not in di-
versity per se but in the ranks assigned to those
units of diversity.
OM67_06.indd 76 4/6/10 6:31:01 PM

SUBSPECIES IN AVIAN CLASSIFICATION 77
In summary, the biological species concept
provides two levels of information, whereas the
phylogenetic species concept provides one. The
biological species concept incorporates the acqui-
sition of diagnostic characters into its classification
by ranking diagnosable populations minimally as
subspecies. The biological species concept also in-
corporates reproductive isolation, acknowledged
even by many proponents of the phylogenetic
species concept as an important evolutionary step
in the history of any lineage, by ranking such pop-
ulations as species.
Subspecies and Human Perception
What would happen if the phylogenetic spe-
cies concept’s minimum diagnosable units were
applied to Homo sapiens? Certainly, until recent
decades, humans classified one another into ra-
cial groups thought to have diagnostic charac-
ters (e.g., Hall and Kelson 1959), and even today,
one’s race is a data field in many nonscientific
categorization schemes. Research has shown that
such schemes fail to classify individuals reliably
and that, at the genetic level, ≤95% of all genetic
variation is among-individual, not among-group.
Nonetheless, despite rampant ongoing gene flow
and the relatively recent origin of Homo sapiens,
the residual variation may accurately predict
region of origin and show strong geographic
structuring. For example, even different groups
of Native Americans differ strongly in haplotype
frequencies (Malhi et al. 2003). Research on the
genetic basis of human diseases has spawned an
interest in ancestry-informative markers that pre-
dict the geographic origin of individual humans.
Although complex computations are required to
identify unique combinations of alleles, the geo-
graphic structure of this variation can identify
individuals with respect to continent of origin
(Rosenberg et al. 2002, Collins-Schramm et al.
2004, Mao et al. 2007, Li et al. 2008) and subregion
(Tian et al. 2008a). Recently, Tian et al. (2008b), us-
ing a sample of European Americans categorized
according to “self-reported” region of European
descent, showed that principal component analy-
sis of single nucleotide polymorphisms allowed
accurate discrimination of individuals as either
northern vs. southern European ancestry and
found further evidence of structure within the
northern European sample.
If geographic variation in Homo sapiens were
sampled in the same limited way that it is in most
birds, then application of the phylogenetic species
concept to Homo sapiens would certainly produce
“minimum diagnosable units” that are neither
biologically nor socially acceptable as “species.”
However, the detailed structure of this variation,
both phenotypic and genotypic, is sufficiently
well studied that we can be sure that few if any
character states analogous to those used in bird
taxonomy would unambiguously diagnose any
subpopulations of humans. Even today, after
much global movement and genetic mixing, our
own genetic and morphological (e.g., Shriver et
al. 2003) diversity could be partitioned into an
unknown number of diagnosable units by use of
unique combinations of characters and allele fre-
quency differences. By contrast, although cultural
barriers prevent full application of the biological
species concept to humans, this concept would
consider all humans conspecific (Homo sapiens).
In terms of perception and the absence of biologi-
cally based reproductive isolation, humans clearly
think of themselves as belonging to one species,
as defined by the biological species concept, de-
spite marked geographic variation within Homo
sapiens. Given that species definitions are scientifi-
cally untestable matters of taste (Brookfield 2002),
human perception has spoken with resounding
clarity that “species” are not minimum diagnos-
able units.
Common Ground
The debate over species and subspecies con-
cepts is healthy, particularly in forcing a reevalu-
ation of currently recognized subspecies names.
I strongly agree with critics of the biological spe-
cies concept that terminal taxa should be used in
analyses of, for example, biogeography and bio-
diversity. The uncertainty of the diagnosability of
many subspecies, especially in temperate North
America, requires that anyone undertaking an
analysis using terminal taxa must carefully scru-
tinize their diagnosability. Empirically, however,
using named subspecies from Peters’s Check-list
series, even without critical evaluation (e.g., Cra-
craft 1985), successfully demarcates areas of en-
demism. So, if the sample is large enough and the
error rate (nondiagnosable taxa) small enough,
real patterns should emerge even if current sub-
species names are taken as is.
I also strongly concur with McKitrick and
Zink (1988) and others that subjective notions of
whether a character is too trivial to use to diagnose
OM67_06.indd 77 4/6/10 6:31:01 PM

78 ORNITHOLOGICAL MONOGRAPHS NO. 67
a taxon are unscientific. What matters is whether
that character is a marker for a cohesive evolu-
tionary unit, regardless of any known functional
significance. If that character is “one extra hooklet
on a barb of the seventh primary” (McKitrick and
Zink 1988:9), and it passes the 95% diagnosability
test, then it defines an entity worthy of a name, in
my opinion.
Some defenders of the biological species con-
cept worry that adoption of the phylogenetic
species concept would lead to too many species
(e.g., Mayr [1993] as cited by Zink and McKitrick
[1995]). Preconceived notions of how many spe-
cies there ought to be are scientifically indefen-
sible. I echo McKitrick and Zink (1988) and Zink
and McKitrick (1995) on the importance of letting
the data determine the number of populations
ranked as species. Even under the biological spe-
cies concept, the number of species is increasing
dramatically, particularly in the tropics, where
many taxa formerly ranked as subspecies are el-
evated to species rank through careful study of
vocalizations and population interactions at con-
tact zones. For example, field studies of polytypic
species of antbirds (Thamnophilidae), many using
the comparative framework of Isler et al. (1998),
have already elevated 31 taxa previously treated
as subspecies to species rank. These 31 species,
ranked as subspecies either by Peters (1951) or
by Meyer de Schauensee (1970), were subsumed
under 16 species names, including one, Myrme-
ciza castanea, considered a synonym of an exist-
ing subspecies. They include Frederickena fulva,
Cymbilaimus sanctaemariae, Thamnophilus zarumae,
T. tenuepunctatus, T. cryptoleucus, T. atrinucha, T.
stictocephalus, T. sticturus, T. pelzelni, T. ambiguus,
Thamnomanes schistogynus, Dysithamnus leucostic-
tus, Epinecrophylla spodionota, Myrmotherula ignota,
M. multostriata, M. pacifica, Herpsilochmus atricap-
illus, H. motacilloides, H. dugandi, Drymophila ru-
bricollis, Cercomacra laeta, Hypocnemis flavescens,
H. peruviana, H. subflava, Hypocnemis ochrogyna, H.
striata, Schistocichla humaythae, S. brunneiceps, S.
rufifacies, S. saturata, and Myrmeciza castanea (for
references, see Zimmer and Isler 2003, Remsen
et al. 2009). Species richness has thus increased
by 88% in the 18 cases studied so far. Ongoing
studies of other groups of thamnophilids will
undoubtedly increase this tally, perhaps by as
many as 50 species (M. L. Isler pers. comm.). If
the thamnophilid results can be extrapolated to
tropical avifaunas as a whole, many hundreds of
subspecies will be elevated to species rank under
the guidelines of the biological species concept
when critical data become available. This does not
represent a shift toward the phylogenetic species
concept but, rather, an increase in data on repro-
ductive isolation.
McKitrick and Zink (1998) provided a protocol,
based in part on Zink and Remsen (1986), for apply-
ing the phylogenetic species concept to real-world
situations. I suspect that they would also share my
concern that few studies undertake the necessary
steps to determine diagnosability. I disagree only
semantically. I call the diagnosable units revealed
by such analyses subspecies, not species.
Acknowledgments
I thank M. Batzer, R. Brumfield, N. Collar, E.
Dickinson, K. McGraw, M. Patten, F. Sheldon, and K.
Winker for comments on sections of the manuscript.
OM67_06.indd 78 4/6/10 6:31:02 PM