vol. 155, no.

4 the american naturalist april 2000

Detecting the Geographical Pattern of Speciation from

Species-Level Phylogenies

Timothy G. Barraclough1,* and Alfried P. Vogler1,2,†

1. Department of Biology and Natural Environment Research species (Nee et al. 1996; Purvis 1996; Sanderson and Don-
Council Centre for Population Biology, Imperial College at oghue 1996; Barraclough et al. 1998, 1999a). Hence, pub-
Silwood Park, Ascot, Berkshire SL5 7PY, United Kingdom;
lication of large numbers of species-level phylogenies will
2. Department of Entomology, Natural History Museum,
Cromwell Road, London SW7 5BD, United Kingdom
provide the opportunity for a comparative biology of spe-
ciation across a range of groups. However, this approach
Submitted September 9, 1998; Accepted October 24, 1999 involves reconstructing past events, and so it is important
to consider how changes occurring since speciation events
influence our ability to infer the pattern of speciation
(Mayr 1963). In this article, we investigate one particular
abstract: We introduce a general approach for investigating the
issue, namely the use of species-level phylogenies for in-
role of geography in speciation, based on analyzing the geography vestigating the role of geography in speciation.
of sister clades across all nodes in a species-level phylogeny. We Geography is widely recognized as the key factor in the
examine the predictions of allopatric, sympatric, and peripatric mod- study of speciation. The primary classification of specia-
els of speciation in several animal groups, using patterns of range tion, into so-called geographic modes, is based on the
overlap and range size symmetry between sister clades. A simple pattern of geographic ranges seen among daughter species
model of cladogenesis incorporating random movements of species’
(Mayr 1963; Bush 1975; Templeton 1981). However, the
ranges is used to illustrate the effects of range changes on expected
patterns. We find evidence for a predominantly allopatric mode of theoretical and empirical basis for different modes remains
speciation in our study groups, with sympatry arising through post- controversial, with particular emphasis on two issues. First,
speciational range changes. In addition, we find that relatively recent there is the question of whether geographical isolation is
speciation events are characterized by greater asymmetry in range necessary for the formation of new species, resulting in
size between sister clades than expected under our null models, pro- debate over the relative frequencies of allopatric and sym-
viding potential support for the peripatric model of speciation. We patric speciation (Mayr 1963; Maynard Smith 1966; Ro-
discuss the possible confounding effects of postspeciational range
changes on our conclusions.
senzweig 1978; Bush 1994; Doebeli 1996; Duffy 1996;
Johnson et al. 1996; Shoemaker and Ross 1996). If allo-
Keywords: speciation, diversity, null models, biogeography, modes.
patric speciation is the predominant mode, then we need
to explain the subsequent origin of sympatry within clades.
Second, within the context of allopatric speciation, many
The molecular revolution in phylogenetics has provided authors advocate the importance of small, founder pop-
new sources of information for the study of evolution and ulations in the process of speciation (Mayr 1963; Carson
macroecology, answering questions that were difficult to and Templeton 1984). This view is manifest in the pre-
address until recently (Lutzoni and Pagel 1997; Omland diction of the peripatric theory of speciation that new
1997; Bleiweiss 1998; Farrell 1998; Losos et al. 1998). One species tend to form as small range fragments around a
area gaining new vitality from this information is the study widely distributed ancestral species. An alternative view is
of speciation. Phylogenies represent the pattern of clad- that speciation is caused by geographic features that split
ogenetic splits leading to present-day species and may pro- a species’ range and that these may arise at any point
vide a trace of processes involved in the origin of those within the ancestral species’ range. Active debate over these
two alternatives persists to the present day (Barton and
* E-mail: [email protected]. Charlesworth 1984; Rice and Hostert 1993; Coyne 1994;
†
E-mail: [email protected]. Moya et al. 1995; Slatkin 1996, 1997; Templeton 1996;
Am. Nat. 2000. Vol. 155, pp. 419–434. q 2000 by The University of Chicago. Charlesworth 1997).
0003-0147/2000/15504-0001$03.00. All rights reserved. The original evidence for these discussions was based
420 The American Naturalist

on analyses of species’ ranges, in the light of systematic Material and Methods

revisions of genera (reviewed in Mayr 1963). Recent au-
We performed a literature search for molecular phyloge-
thors have performed more strictly statistical tests using nies including all, or a large majority of, described extant
phylogenetic information on monophyletic groups of spe- species within clades for which species’ range data are
cies. For example, Chesser and Zink (1994) assessed the available. The chosen groups are not an exhaustive list of
relative frequencies of different modes of speciation among all such clades but were chosen to illustrate the method.
a sample of American bird groups, following criteria used Details are given in the appendix. They cover a range of
by Lynch (1989). However, the authors recognized two animal groups and molecular markers. In many cases, the
problems with the approach. First, it relies on an arbitrary authors present several phylogenies obtained using differ-
classification of modes of speciation, depending on thresh- ent methods of phylogenetic reconstruction. We chose par-
old values in the proportion of overlap between sister spe- simony trees for consistency: nearly all papers presented
cies or their relative range sizes. Second, species’ ranges parsimony trees, whereas distance methods varied among
may have changed subsequent to speciation, in which case studies. The one exception was the phylogeny of Sylvia
the distribution of newly formed species at the time of the warblers, which was reconstructed by the distance-based
split may no longer be recognizable from current ranges. method of DNA-DNA hybridization. We used the Fitch-
In addition, previous accounts have not compared ob- Margoliash tree presented for this group. The relative ages
of nodes were calculated by applying a local molecular
served patterns to those predicted by explicitly stated mod-
clock approach to distance data for the chosen topology.
els of the geographic mode of speciation. For example,
This involves progressively averaging branch lengths for
in relation to the peripatric mode of speciation, the ob-
sister taxa, starting from the tips through to the base, and
served frequency of peripheral species with small range then using distances to nodes as estimates of relative age.
sizes may not be unusual compared to that expected simply Phylograms of each group suggest at least rough rate con-
by splitting the ancestral range into two randomly sized stancy among lineages within these groups. The use of
fragments. branch lengths to estimate relative ages of nodes will be
We propose an approach based on analysis of the geo- discussed below.
graphic ranges of species within a clade in the light of their Species’ ranges were obtained from sources listed in the
phylogeny. Species-level phylogenies reconstruct the pat- appendix. Ranges were redrawn from source material as
tern of cladogenesis leading to extant species, and so the shaded regions onto equal area maps in the computer.
geographic ranges of sister clades identified from the phy- Some sources provide shaded regions of species occur-
logeny can be used as an estimate of the geographic mode rence, others provided dot maps of localities. We drew
at the time of speciation. In addition, subsequent range shaded regions around dot maps by eye, following text
changes may be accounted for in part by considering the descriptions of the ranges of each species. Computational
pattern of geographic ranges for sister clades across all methods of representing species’ ranges for these groups
nodes in the phylogeny (Brown and Gibson 1983; Lynch could be employed, but for present purposes, these would
be unlikely to lead to major differences in final ranges and
1989; Chesser and Zink 1994). Deeper nodes in the phy-
in our conclusions. The ranges of higher clades were ob-
logeny represent relatively older speciation events; there-
tained by shading the area where at least one constituent
fore, changes in pattern with relative node age may reflect
species is found.
the extent of changes in geographic ranges over time.
Hence, our approach is to calculate measures of geographic
mode of speciation, such as range overlap between sister Allopatry or Sympatry
clades, for all nodes in the phylogeny and to consider the Methods
pattern of those measures in relation to the relative age
Using the shaded range maps, we calculate the degree of
of nodes. We introduced this approach briefly elsewhere
sympatry between sister clades for each node in the phy-
(Barraclough et al. 1998, 1999a) but develop it further in
logeny of each group. The degree of sympatry is defined
this article. To illustrate the effects of range movements, as the percentage of the more restricted clade’s range over-
we introduce a simple simulation model of cladogenesis, lapped by its more widespread sister (Chesser and Zink
incorporating random movements of species’ ranges 1994):
within a continuous area. We use our findings to interpret
real examples from bird, fish, and insect groups and discuss area of overlap
.
the likely utility of this approach for distinguishing modes range size of clade with smaller range
of speciation.
 Geographic Pattern of Speciation 421

This ranges from 0.0, signifying no range overlap, to 1.0, combines a standard stochastic, constant birth rate model
signifying that the range of one clade is entirely overlapped of cladogenesis (Raup et al. 1973; Harvey et al. 1994), with
by its sister. The index differs from one we used previously species’ ranges coded for simplicity as rectangular shapes
where our aims included testing for character divergence within a continental area. At speciation, the ranges of de-
between sympatric lineages (Barraclough et al. 1998, scendent species are determined by the chosen mode of
1999a, 1999b). Areas and overlaps were calculated by im- speciation, either allopatric or sympatric. Ranges are free
age analysis of the range maps. As a measure of time since to move by random independent changes to their edges,
speciation, we use the relative distance of each node from leading to changes in size, shape, and location of species’
the tips of the phylogeny, based on branch length data ranges over time. The exact form of range movements
obtained from source phylogenies. This assumes that may influence expected patterns, so we consider two al-
branch lengths provide at least rough information on the ternatives differing in the relative frequency and magni-
relative ages of each node. We have repeated our analyses tude of changes. The first is that ranges move by frequent
using an alternative measure, node level, which is the level but small-scale changes to their boundaries, such that
of each node above the tips of the phylogeny: sister species boundaries undergo a random walk over time. The second
are level 1, their nesting clade is level 2, and so on. The is that range movements occur by occasional large-scale
results are statistically indistinguishable, and so for brevity shifts of entire species’ ranges, such as those observed in
we present only those using relative branch length data. many temperate groups during periods of climate change
Potential limitations of this measure are discussed below. (Coope and Wilkins 1994; Graham et al. 1996). Details
We plot the degree of sympatry against relative node age. are shown in figure 1. We do not include extinction in
our model but discuss its likely impact on observed pat-
terns in the discussion. The model is not intended to cover
Predictions and Simulations
all possible scenarios; it is merely the simplest model in-
Elsewhere, we made verbal predictions for the expected corporating mode of speciation and random changes to
patterns of these plots under alternate scenarios (Barra- two-dimensional species’ ranges.
clough et al. 1998, 1999a), which can be summarized as Sample plots obtained by simulation are shown in fig-
follows: If speciation is predominantly allopatric, recently ures 2 and 3. Under an entirely allopatric model of spe-
diverged sister species are expected to display little or no ciation (figs. 2A, 3A), range movements lead to increased
overlap in geographic ranges. Alternatively, if speciation sympatry between more ancient sister clades, which is
is predominantly sympatric, recently diverged sister species largely associated with an increase in variance. The same
are expected to display sympatries of around 1.0 because pattern is found whether range changes occur by gradual
one species’ range must enclose the other entirely. Hence, accumulation of small-scale changes to range boundaries
the intercept of the plot between range overlap and node (fig. 2) or by occasional large-scale shifts of species’ ranges
age provides a summary of the predominant geographic (fig. 3). A high frequency of sympatric speciation increases
mode of speciation within a clade. The level of range over- the number of recent splits with sympatries around 1.0,
lap between more anciently diverged sister clades depends leading to an increase in intercept (figs. 2C, 3C). However,
on the extent of range movements occurring after speci- as the level of range movements increases, plots generated
ation events. If speciation is predominantly allopatric, under alternate modes of speciation become harder to
newly formed species have nonoverlapping ranges, and so distinguish: intercepts converge on an intermediate value,
subsequent range changes are likely to increase overlap as shown by repeated trials summarized in figure 4. Note
between more anciently split species (at least for clades that, because our measure of sympatry is bounded between
occupying a finite area). No increase in the observed range 0.0 and 1.0, the degree of sympatry was arcsine trans-
overlap between sister clades at older nodes is expected if formed before fitting intercepts to the plots by linear re-
there are no range movements subsequent to speciation. gression (Sokal and Rohlf 1981).
Hence, if speciation is predominantly allopatric, the slope The effects of range movements can be treated further
of the plot between range overlap and node age provides by considering the expected pattern when geographic
a summary of the extent or rate of postspeciational range ranges have changed so much that present-day ranges re-
movements. Range changes may include changes in range tain no information on the geographic mode of speciation.
size and shape, as well as drifting of ranges across the Even if species’ ranges are randomly distributed shapes
landscape. If speciation is predominantly sympatric, then within a continental area, range overlap is likely to be
range overlap is likely to decrease for relatively older nodes. greater between sister clades originating from older nodes.
The predictions can be illustrated using a simple sim- Older clades tend to occupy a larger area than younger
ulation model for the origin and movements of species’ clades because they contain more species, and thus, there
ranges during cladogenesis, outlined in figure 1. The model is greater chance of range overlap with their sister clades
422 The American Naturalist

Figure 1: Simulation model for the evolution of species’ ranges during cladogenesis. A, Model considers species with rectangular ranges occupying
a square continental area. Diversification starts at time 0 with a single lineage. Lineages have a constant probability of splitting per unit time, leading
to an exponential build-up of species numbers through time. B, At each speciation event, the ranges of descendent species are allocated according
to the chosen mode of speciation. Allopatric speciation involves a dichotomous split at a uniform random point along either the horizontal or
vertical edge of the species’ range. Sympatric speciation involves the appearance of a new species within the range of an existing species (the
maximum and minimum (x, y) coordinates of the new range are drawn from a uniform distribution of possible values). The frequency of these
two modes is defined by the probability per speciation event that speciation is sympatric. C, Range movements are specified by random independent
changes to the minimum and maximum (x, y) coordinates of each species’ range. We consider two alternatives. First, we allow ranges to move by
continuous random walk of their minimum and maximum (x, y) coordinates through time. The number of steps per unit time is drawn from a
Poisson distribution so that small movements are more frequent than large ones. Step size is a small constant that can be scaled to vary the rate
of movement. An example of the movement of a single range over time is shown (C, i). Second, we assume that species have a constant probability
per unit time of undergoing a major shift to their ranges, upon which their minimum and maximum (x, y) coordinates are set to random points
from a uniform distribution within the continental area (C, ii). The frequency of range shifts can be varied. The continent presents a boundary to
range movements: parts of ranges that step outside the continent are truncated before the next iteration. For simplicity, we do not include extinction
in this model. Instead, species are prevented from reaching 0 range size: if the maximum coordinate drops below the minimum coordinate within
a time step, the maximum and minimum coordinates are simply switched to ensure positive range size. Hence, species close to 0 range size maintain
small, but positive, range sizes until any such time that the maximum coordinate moves away from the minimum coordinate again. D, Endpoint
of each simulation is a series of species’ ranges and a phylogeny showing the sequence and timing of splits leading to those species.

(Barraclough et al. 1998). Sister species may tend to display on the mode of speciation in this case. The mean intercept
low sympatry simply by chance placement of their ranges obtained over repeated trials is shown in figure 4.
within a large continental area. For example, one possible The above predictions assume that species’ ranges move
formulation of this situation is to allocate the present-day at random and independently of one another. However,
range of each species in our simulation model as randomly additional processes affecting species’ ranges may also in-
chosen points within the continental area (i.e., the max- fluence our ability to detect the mode of speciation. For
imum and minimum [x, y] coordinates are drawn from example, if species compete for similar ecological re-
a uniform distribution of possible values). This model sources, then allopatric distributions may be maintained
produces a roughly log-normal distribution of range sizes by competitive exclusion (Letcher et al. 1994). In principle,
(see “Peripatry” below). Samples of sympatry versus node this might obscure the pattern of speciation, most seriously
age plots under this model are shown in the third column if species formed in sympatry were subsequently forced
of figure 3. The plots display little overlap between recently into allopatry by competition. Although this cannot be
split species, apparently consistent with allopatric specia- ruled out, most if not all models of sympatric speciation
tion, but in fact species’ ranges provide no information require mechanisms permitting the coexistence of sym-
 Geographic Pattern of Speciation 423

Real Examples
Plots from the sample groups are shown in figure 5. Cor-
relation coefficients and intercepts are shown in table 1.
Correlation coefficients were determined using Spearman’s
rank tests. So that the intercept of each plot could be
estimated, regression lines were fitted following arcsine
transformation of the degree of sympatry. Model criticism
revealed that residuals from the lines were normally dis-
tributed but showed slight increase in variance at inter-
mediate fitted values in a few groups. Nonetheless, the fit
is good enough to provide a reasonable measure of the
intercept.
The results can be summarized as follows. First, in all
cases, intercepts are distributed around 0, and in most
cases, the most recent nodes display 0 sympatry. There are
two exceptions to the latter generalization: the fruitfly ge-
nus Rhagoletis, which has one very recent node with sym-
patry of 1.0 (between Rhagoletis chionanthi and Rhagoletis
osmanthi), and the tiger beetle subgenus Ellipsoptera,
Figure 2: Samples of sympatry versus node age plots obtained when
which has one recent node with sympatry around 0.47.
range movements occur by the random walk model. The frequency of The genus Rhagoletis has provided the classic example of
allopatric and sympatric speciation varies among rows. A, Entirely al- sympatric race formation by host shifts (Bush 1969). Note
lopatric. B, 50% sympatric. C, Entirely sympatric speciation. The Y-axis
is the degree of sympatry, and the X-axis is the age of nodes in units
time. Simulations started with an ancestral range of 200 # 200 units in
a continental area of 220 # 220 units. The per-lineage probability of
speciation was 0.01 per unit time. They were stopped once 20 species
have evolved. The number of range steps occurring per unit time was
drawn from a Poisson distribution with X = 0.1 ; that is, steps occurred
at 10 times the rate of speciation events. The overall level of range move-
ment was then varied by scaling the step size from small (step size = 0.2
units), to medium (=2.0 units), to large (=20 units).

patrically originating forms, so subsequent competition

seems unlikely. However, the sympatric origin of species
could be obscured if competition were too weak to prevent
the initial establishment of sympatric forms but strong
enough to favor populations that escape from sympatry
and if subsequent movement into allopatry occurred over
shorter timescales than those elapsed since the most recent
speciation events. It has been suggested to us by a referee
that sympatric speciation by chromosomal rearrangements
could fulfil these requirements. Information on chromo-
somal structure and ploidy levels could be used to test this
possibility, but in the absence of additional evidence, we Figure 3: Sympatry versus node age plots obtained when range move-
ments occur by large-scale shifts to entire species’ ranges. The frequency
believe it remains an unlikely bias. For species originating of allopatric and sympatric speciation varies among rows. A, Entirely
by allopatric speciation, competitive interactions may de- allopatric. B, 50% sympatric. C, Entirely sympatric speciation. The
lay the origin of sympatry until species have diverged in Y-axis is the degree of sympatry, and the X-axis is the age of nodes. The
phenotype sufficiently that competition no longer occurs, frequency of range shifts varies from low (=0.01 per unit time, the same
rate as speciation), to medium (=0.05 per unit time), to high (the extreme
thereby accentuating the pattern obtained under neutral case in which all species’ ranges have shifted to random locations since
range movements. The effects of competition and how we the most recent speciation event). Other details are as described in fig-
might detect them are discussed further below. ure 2.
424 The American Naturalist

so much that they provide unreliable information on

modes of speciation.
In combination, the observation of low intercepts and
increasing sympatry with relative node age suggests a gen-
eral pattern of allopatric speciation with range changes
leading to a build-up of sympatry over time. However, as
outlined in the previous section, similar patterns may be
obtained when there are large-scale range changes, irre-
spective of mode of speciation. Possible ways to determine
whether current ranges retain information on mode of
speciation will be discussed below.

Peripatry
Figure 4: Effect of increasing range movements on the intercept of sym- Methods
patry versus node age plots. Each point is the mean intercept from 15
trials of the random walk model outlined in figure 2. Standard errors Using the shaded range maps, we calculated the degree of
are shown. Intercepts were fitted following arcsine transformation of range size symmetry at each node, defined as the range
sympatry, for reasons discussed in the text. Under higher levels of range size of the clade with the smaller range divided by the sum
movements, the intercepts obtained under entirely allopatric (diamonds)
of the range sizes of each clade:
and entirely sympatric (triangles) speciation converge toward an inter-
mediate value. A similar pattern is found using the range shift model
(not shown). The bar to the right of the graph shows the mean intercept range of clade 1
,
obtained when all species’ ranges have shifted to random points within range of clade 1 1 range of clade 2
the continent; that is, ranges retain no information on the mode of
speciation.
where clade 1 has the smaller range size. The measure
takes values between limits of 0.0 and 0.5, the latter rep-
that two species within the group, Rhagoletis zephyria and resenting sister clades with equal-sized ranges. Lynch
Rhagoletis mendax, are embedded within the species Rha- (1989) and Chesser and Zink (1994) used a similar mea-
goletis pomonella in the source phylogeny. Our method sure in their assignments of peripheral-isolates speciation.
cannot address patterns within a paraphyletic clade of this We plot the degree of symmetry against relative node age.
kind because the exact sister groups of R. zephyria and R.
mendax within R. pomonella are unknown (see appendix).
Predictions and Simulations
Rhagoletis mendax is entirely overlapped by part of the
range of R. pomonella, and so this may represent an ad- We extend the approach outlined in the previous section
ditional recent split with high sympatry, whereas R. ze- to provide a statistical framework for testing the theory
phyria displays only slight overlap with other R. pomonella of peripatric speciation. Under a model of predominantly
populations. However, despite one and possibly two recent peripatric speciation, the geographic ranges of recently
sympatric splits, the pattern in the remainder of the group split sister species should tend to display asymmetry of
appears not to be suggestive of predominant sympatric range size. Under the alternative model that small pop-
speciation since other recent nodes display sympatries well ulation size, and hence range size, does not increase the
below 1.0. likelihood of speciation, we expect no particular tendency
Second, in all groups, sympatry tends to increase with toward range size asymmetry. One simple null model for
relative node age suggesting that range changes have oc- this alternative is that, at speciation, the range of the an-
curred in these groups. There is variation in the pattern cestral species is split into two randomly sized fragments,
of this increase among groups, ranging from Malurus, the in other words that a split occurs at some random point
cranes, and Xiphophorus, with a majority of allopatric splits within the ancestor’s range. This corresponds to a broken
even between quite ancient splits, through to Sylvia, the stick model, similar to those commonly used in other areas
auks, Rhagoletis, Flexamia, and Ellipsoptera, with a high of evolutionary ecology (MacArthur 1960; Naeem and
degree of sympatry even among quite recent splits and a Hawkins 1994; Nee et al. 1996); hence, we refer to it as
large scatter of values. The plots in the latter clades re- the “phylogenetic broken stick.” It produces an even dis-
semble closely those obtained under a null model of ran- tribution of range size symmetry immediately after spe-
dom ranges (third column, fig. 3). Hence, we cannot rule ciation that ranges from 0.0 to 0.5 with a mean value of
out the possibility that present ranges may have changed 0.25. Lower values would suggest a tendency toward range
 Geographic Pattern of Speciation 425

Figure 5: Plots of the degree of sympatry against relative node age in representative clades. The Y-axis is the degree of sympatry, and the X-axis is
relative node age. Units vary among clades depending on the molecular markers used in the source phylogenies.

size asymmetry and the possible importance of small asymmetric model in which ranges are allocated at spe-
ranges in speciation. The degree of symmetry between ciation with a bias toward asymmetric ranges (details in
older sister clades will depend on the nature of subsequent legend of fig. 6). Our ability to distinguish the two modes
range changes. If range movements occur within a finite of speciation declines as the level of range movements
continental area, then older clades will tend to occupy a increases (fig. 7). Note that, because our measure of range
greater proportion of the total area and range size sym- size symmetry is bounded between 0.0 and 0.5, the value
metry may increase with node age. of range size symmetry was doubled and arcsine trans-
Figure 6 shows sample plots obtained using the simu- formed before fitting intercepts to plots by linear regres-
lation model outlined above. At low levels of range move- sion (Sokal and Rohlf 1981).
ment, the phylogenetic broken stick model produces the As before, we also consider the expected pattern of sym-
expected even distribution of symmetries. However, in- metry when all species’ ranges have shifted to random
creasing the level of range movements increases the slope locations since the most recent speciation event. In this
of plots and reduces the intercept slightly below 0.25 (fig. case, symmetry between sister species will depend on the
7). For comparison, we show results from an arbitrary statistical distribution of range sizes among species, since
426 The American Naturalist

Table 1: Summary statistics for plots of the degree of sym- Real Examples
patry against relative node age
Spear-
Plots of symmetry against relative node age are shown in
man’s figure 8. Intercepts are shown in table 2. They were fitted
rank by linear regression of arcsine transformed values of sym-
Standard metry, as described above. We used simulations to test for
test
error of
significant departures from the null expectations outlined
Taxon n rs P Intercept intercept
above. First, we performed 1,000 simulations under the
Warblers 14 .66 .02 .054 1.120 phylogenetic broken stick model, splitting an ancestral
2.052 range successively according to the phylogeny for each
Fairy wrens 11 .76 .02 .049 1.091 group (assuming no subsequent range movements). From
2.045
each simulation, we calculated the intercept of symmetry
Cranes 14 .54 .07 .012 1.046
2.012
against node age to obtain the distribution of intercepts
Auks 22 .63 .004 .011 1.081 expected under this model. Second, we performed one
2.011 1,000 simulations allocating relative range sizes from a
Swordtail fish 18 .47 .14 .024 1.082 simultaneous broken stick distribution, corresponding to
2.024 our model that species’ ranges have shifted to random
Fruitflies 19 .48 .001 .174 1.185 locations within a continental area. As noted above, the
2.127 range size of higher clades depends in part on overlaps
Leafhoppers 35 .57 .06 .079 1.080
2.054
Tiger beetles 12 .39 .21 .208 1.147
2.116
Note: Intercepts were fitted by linear regression of arcsine transformed
sympatry, hence the standard errors of the untransformed intercept are
asymmetric; n = number of nodes.

the ranges of sister species are effectively drawn at random

from the available distribution. Although several forms of
range size distribution can be imagined, we consider a
single model in which range sizes are drawn from a si-
multaneous broken stick distribution; that is, we break a
stick of given length at n 2 1 uniform random points to
produce n randomly sized pieces. Note that this model
differs from the phylogenetic broken stick model where
ranges were subdivided successively according to phylog-
eny. Our reasons for choosing the model are threefold.
First, it corresponds to the range size distribution pro-
duced by the model of random range locations outlined
in the “Allopatry or Sympatry” section. Second, it appears
to match real distributions, leading to a left-skewed log
normal distribution of range sizes similar to those com-
monly observed in real clades (Gaston 1998). Third, it
allows straightforward simulation of expected patterns for Figure 6: Samples of range size symmetry plots obtained under three
real clades (see below) because the distribution of relative models of range size allocation at speciation. A, Phylogenetic broken stick
model, in which the ancestral range is subdivided at a uniform random
range sizes depends solely on the number of species within point along its perimeter. B, Arbitrary asymmetric model, in which the
a clade. The pattern of symmetry across nodes will also probability of a split occurring increases in proportion to the square of
depend on levels of range overlap, because the range size its distance away from a 50 : 50 split. C, Simultaneous broken stick model
of higher clades depends on overlaps between their con- in which the minimum and maximum (x, y) coordinates of each species’
stituent species. A sample plot obtained under this model range are set to random points from a uniform distribution within the
continental area. The Y-axis is the degree of symmetry, and the X-axis
is shown in figure 6. In our simulations with 20 species is the age of nodes. A, B, We show two levels of range movements: no
the model produces an expected mean intercept of around range movements and a random walk model with a step size of 5.0 units.
0.18 (fig. 7). Other details are as described in figure 2.
 Geographic Pattern of Speciation 427

In conclusion, we find evidence for greater range size

asymmetry between recently split sister clades than ex-
pected either by assuming random subdivision of ancestral
ranges at speciation or under a simple null model for the
random distributions of species’ ranges within a conti-
nental area. The auks follow an opposite pattern with re-
cent sister clades having more symmetric range sizes than
older sister clades. Overall, the results appear consistent
with a peripatric mode of speciation. However, alternative
explanations will be discussed below.

Discussion and Conclusions

Our analyses utilize phylogenetic information and geo-
Figure 7: Effect of increasing range movements on the intercept of range graphic ranges of present-day species to investigate the
size symmetry versus node age plots. Each point is the mean intercept
geographic pattern of speciation within clades. We use a
from 15 trials using the random walk model. Standard errors are shown.
Intercepts were fitted following doubling and arcsine transformation of simple model of cladogenesis and random range move-
range size symmetry, for reasons discussed in the text. Under high levels ments to help interpret observed patterns. Using this gen-
of range movements, the intercepts obtained under the arbitrary asym- eral approach, we looked at three issues.
metric model of range subdivision (diamonds) converge on those ob-
tained under the phylogenetic broken stick model (triangles). A similar
pattern is obtained using the range shift model (not shown). The bar to Allopatric or Sympatric Speciation
the right of the graph shows the mean intercept obtained in 15 trials of
the simultaneous broken stick model of range allocation. We used the pattern of sympatry across nodes to estimate
the general mode of speciation within each group, allo-
patric or sympatric. In nearly all cases, the most recent
between their constituent species. Therefore, we used the nodes are characterized by 0 range overlap between sister
observed overlaps for each node to calculate the range clades, suggesting a predominantly allopatric mode of spe-
sizes of higher clades from the simulated range sizes of ciation in these groups. This conclusion is unsurprising
each species. Probabilities under the two models are shown given the weight of evidence from other studies that sym-
in table 2. Correlation coefficients of the relationship be- patric speciation is likely to be relatively rare. However,
tween symmetry and node age were determined by Spear- the simulations show that similar patterns may arise if
man’s rank tests and are shown in table 2. species’ ranges are random with respect to cladogenetic
The results can be summarized as follows. Intercepts history, simply because there is little chance of overlap
are lower than expected under the phylogenetic broken between randomly placed ranges of sister species. For
stick model in six groups. This difference is significant in clades with large numbers of allopatric splits, such as Mal-
the warblers, leafhoppers, and fruitflies, marginally non- urus and Xiphophorus, the observed low levels of overlap
significant in the fairy wrens (P ! .1), and nonsignificant would seem unlikely to arise by chance placement of spe-
in the swordtail fish and tiger beetles. The intercept is cies’ ranges. Hence, there appears to be positive evidence
higher than expected under the phylogenetic broken stick for allopatric speciation in these groups. However, plots
model in the cranes and the auks: only 5% of simulated for Sylvia, the auks, and Flexamia closely resemble those
trials produce a higher intercept in the latter group. When produced in simulations of the null model that species’
the eight groups are treated as independent tests, there is ranges are distributed at random, suggesting that these
a significant deviation toward low intercepts across the plots may not provide reliable information on the mode
study groups (P ! .001, Fisher’s combined probability; So- of speciation.
kal and Rohlf 1981). Similar probabilities are obtained In principle, we could test the null explanation by com-
under the simultaneous broken stick model. Although this paring the observed patterns of overlap to those obtained
is only one possible model for the distribution of range by randomly dropping species’ ranges onto the continental
sizes following range shifts, it suggests that our results may area. However, formulating a realistic null model for a
be robust to the occurrence of random range movements. randomization of this kind is fraught with difficulties, in-
Symmetry increases with node age in the warblers, leaf- cluding problems of defining the area available to the clade,
hoppers, and fruitflies (significant in the latter two groups) treatment of the geometries of species’ ranges and con-
but decreases with node age in the remaining groups, most tinents, and in choosing additional constraints to include
strongly in the auks (P ! .1). in the model, for example, habitat restrictions (Strong et
428 The American Naturalist

Figure 8: Plots of range size symmetry against relative node age in representative clades. The Y-axis is range size symmetry, and the X-axis is relative
node age.

al. 1984; Gotelli and Graves 1996). On a case-specific basis, ring after speciation. Our simulation results suggest that
it may be possible to address these problems (Gotelli et sympatric speciation may be detectable from biogeo-
al. 1997), but in the absence of such tests, we conclude graphic information alone only if it occurs at high fre-
simply that recent nodes tend to be allopatric in these quencies, since then we observe a clump of recent splits
groups. However, this may not reflect positive evidence of distributed around 1.0, which is unlikely to arise through
allopatric speciation in groups displaying rapid increases chance range movements alone (figs. 2, 3). Low levels of
in sympatry with relative node age. sympatric speciation will probably not be distinguishable
The recent sympatric splits observed in Rhagoletis and from the effects of range changes. For both groups, the
Ellipsoptera raise the possibility of a low incidence of sym- observed levels of sympatry between recent sister clades
patric speciation in those groups. In Rhagoletis, sympatric seem well within those obtainable under simulations of
speciation seems plausible given the well-documented ex- allopatric speciation with high rates of range change. How-
ample of host shifts among races of Rhagoletis pomonella ever, we cannot rule out a low incidence of sympatric
(Bush 1969, 1994). However, sympatry between recently speciation as an alternative explanation.
split species may also arise through range changes occur- An alternative approach to the problem of range changes
 Geographic Pattern of Speciation 429

Table 2: Summary statistics for plots of range size symmetry against relative
node age
Spearman’s
Phylogenetic Simultaneous
rank test
broken stick broken stick
Taxon rs P Intercept (P) (P)
Warblers .39 .16 .08 .03 !.001
Fairy wrens 2.04 .91 .13 .07 .03
Cranes 2.16 .57 .3 .65 .56
Auks 2.36 .10 .4 .95 .98
Swordtail fish 2.09 .72 .14 .14 .30
Fruitflies .50 .05 .12 .04 .04
Leafhoppers .60 !.01 .12 .002 .006
Tiger beetles 2.11 .71 .22 .36 .44
Note: Sample sizes are the same as in table 1. Symmetry was doubled and arcsine
transformed before fitting intercepts by linear regression. Probabilities of obtaining the
observed intercept or less under the two null models are shown (calculations described
in the text).

is to assess the likely extent of range changes during the tioning sympatry among nodes, we may be able to distin-
history of the clade based on additional information guish these alternatives with further tests (Barraclough et
(Chesser and Zink 1994). For example, paleoecological al. 1998, 1999a). For example, in the Ellipsoptera tiger
data regarding the effects of glacial cycles on the ranges beetles, we find no evidence that levels of ecological di-
of north temperate insect species (Elias 1992; Coope and vergence correlate with the degree of sympatry between
Wilkins 1994) suggest that the insect groups we studied sister clades (Barraclough et al. 1999b), thus failing to
are likely to have undergone major changes since many reject the possibility that sympatry arises through random
of the speciation events represented on the phylogeny.
range movements in this group. Similar tests may allow
Similarly, Blondel et al. (1996) discuss paleoecological ev-
future evaluation of the role of competition in character
idence for likely range changes among species of Sylvia
divergence, and the build-up of sympatry, across a wider
warbler. However, use of this method requires that nodes
in the phylogeny are dated relative to the most recent range of clades.
known climatic events, which may prove difficult in some The pattern of increase in sympatry with node age varies
cases (Klicka and Zink 1997; Avise and Walker 1998). In among study groups. The fairy wrens, cranes, and sword-
addition, for many groups, fossil evidence on the likely tail fish display a pattern of predominantly vicariant dis-
extent of range movements is absent. Therefore, simula- tributions, with most nodes characterized by only low lev-
tion analyses of expected patterns under different scenarios els of overlap between sister clades. In contrast, the Sylvia
of range changes offer a useful additional tool for tackling warblers, auks, Flexamia leafhoppers, Ellipsoptera tiger bee-
this issue. tles, and Rhagoletis fruitflies display much greater levels of
sympatry with considerable range overlap between some
The Pattern of Sympatry within Clades even fairly recently split clades. Interestingly, the three in-
sect groups display the highest levels of sympatry between
In addition to the mode of speciation, plots of the degree recently split clades. Several possible factors may explain
of sympatry versus relative node age provide information the differences among clades, including differences in their
on the origin of sympatry in each group. In all cases, older environmental histories, differences in dispersal abilities
sister clades tend to display greater levels of sympatry than
(Gutierrez and Menendez 1997; McCall et al. 1998), and/
younger sister clades. Irrespective of speciation mode, this
or differences in the strength of any interspecific inter-
pattern is only possible if species’ ranges have changed
since speciation events. Furthermore, it shows that sym- actions influencing geographic exclusion (Letcher et al.
patric species richness within these groups tends to be 1994; Brown et al. 1996; Gotelli et al. 1997). Our study
associated with range overlap between relatively distantly groups compose an insufficient sample to look for cor-
related species. This may occur as a result of random shifts relates of sympatric versus vicariant distributions, but fu-
in species’ ranges over time, but alternatively there may ture comparisons among comprehensive samples of clades
be constraints to sympatry such that only species with will allow tests for general correlates of sympatric species
accumulated ecological differences can coexist. By parti- richness.
430 The American Naturalist

Peripatric Speciation assessed by observing phylograms: if the tips roughly align,

then branch lengths are likely to provide a reasonable es-
We used the pattern of range size symmetry across nodes timate of relative ages.
to investigate the predictions of the peripatric model of Errors in estimates of species’ ranges are likely to add
speciation. Overall, we find that recently split sister clades noise to observed patterns. Although species’ ranges may
display significant asymmetry in range size compared to be undersampled in some groups, particularly rare or trop-
two null models of range size allocation. The auks are an ical ones, there is likely to be reasonable data for any group
exception to this pattern, with greater symmetry between of sufficient interest to warrant a complete phylogenetic
recently split sister clades than between older sister clades. analysis. Information is typically presented in one of two
The theory of peripatric speciation predicts that asym- forms: dot maps showing locality records or shaded dis-
metry arises because the genetic changes leading to new tribution maps. The latter are usually calculated from the
species are more likely to occur in small, narrowly dis- former, combined with the author’s intuition of the entire
tributed populations. However, there are at least two pos- range of each species. Shaded maps inevitably involve a
sible alternative explanations for the observed levels of simplification of the true distribution of species, but they
asymmetry. First, in particular circumstances, asymmetry provide a summary that allows calculation of areas and
may result from the geometry of the landscape, for ex- overlaps at the scale traditionally used for discussion of
ample, if a barrier promoting geographic isolation such as modes of speciation and factors influencing species’
a mountain range happens to be located near the edge of ranges. However, sympatry at the regional scale may not
a continental area (e.g., the Andes). Second, asymmetry correspond to fine-scale co-occurrence. Interpretation of
may arise as a result of range changes occurring since the implications of such sympatry, including mechanisms
speciation events. We considered one model of random of sympatric speciation or levels of species interactions,
range movements, but other types of movement may lead should consider fine-scale species distributions as well.
to higher levels of range size asymmetry between sister
clades, for example, if major range contractions and/or
expansions occur. Alternatively, some range movements Simulation of Species’ Ranges
may lead to greater symmetry between the ranges of sister We introduced a simple model of cladogenesis, incorpo-
clades, such as situations in which range size is heritable rating random movements of species’ ranges, and used it
(Jablonski 1987; Ricklefs and Latham 1992). We cannot to compare observed patterns of geographic range. Pre-
distinguish these alternatives here, but our approach pro- vious work has taken a related approach to island bio-
vides a null model framework for future investigations of geography (Collwell and Winkler 1984; Hayden et al. 1993)
the predictions of the peripatric model and, more gen- and used similar models to investigate spatial patterns of
erally, of the geometry of species’ ranges arising during genotypes within continuous populations (Neigel and Av-
cladogenesis. ise 1993). Our model differs in considering contiguous
species’ ranges occupying continental systems and in our
Methodological Issues express aim of generating broad predictions to guide phy-
logenetic analyses. The simulations use two specific models
Our approach requires a phylogeny and species’ range data of range movements and rely on a rectangular geometry,
for all species within the study clade. Although species- but we believe our general conclusions are robust to other
level phylogenies of entire clades are relatively scarce at formulations of random changes to species’ ranges. Future
present, current levels of molecular automation mean they models could test this assertion and include additional
will soon be available for a wide range of representative features. For example, we have not considered the effects
clades, and they will be based on datasets of multiple mark- of extinction on observed patterns. Random extinction of
ers. In addition to reconstructing the topology of clado- species may add noise to observed patterns of sympatry
genetic splits, molecular phylogenies provide branch and asymmetry by removing parts of the range of higher
lengths that may be useful for estimating the relative ages clades. However, this may not greatly reduce our ability
of nodes. However, these estimates may sometimes in- to discriminate modes because random extinction may
corporate substantial error because of rate variation among tend to prune relatively older branches from a phylogeny,
lineages and the intrinsic stochasticity of base substitutions leaving recent splits relatively unaffected (Nee et al. 1994).
(Mindell and Thacker 1996; Rambaut and Bromham Nonrandom extinction of species with relatively small
1998). Node level may be a useful surrogate measure for ranges may tend to reduce asymmetry between extant sis-
time since speciation in some situations but will be un- ter clades but would perhaps have less effect on patterns
reliable if nodes of equivalent level differ substantially in of sympatry. More sophisticated spatial models of clado-
their ages. The reliability of branch lengths can be roughly genesis may prove useful in investigating this and related
 Geographic Pattern of Speciation 431

questions into the origin and movements of species’ size symmetry between recently split species than expected
ranges. under two null models of range allocation. This may reflect
In conclusion, our results show how phylogenetic anal- intrinsic influences of population size on the probability
ysis of species’ range data can help illuminate the role of of speciation, as predicted by the peripatric model of spe-
geography in speciation and the pattern of subsequent ciation. However, other factors may also lead to asym-
changes in species’ ranges. We find a general pattern of metric ranges between sister clades. We believe that de-
increasing sympatry with relative node age across the velopment of this approach will help to unravel the general
groups examined. This is consistent with a predominantly processes behind the geographic pattern of species
allopatric mode of speciation, and with postspeciational diversification.
range changes leading to increased sympatry among spe-
Acknowledgments
cies. The alternative explanation, that sufficient changes
in species’ ranges have occurred to obscure the mode of We acknowledge grant support from the Natural Envi-
speciation, may be equally consistent in some of the ronment Research Council (GR3/10632) and thank J. Al-
groups. We discussed ways to assess this problem. In ad- roy, J. Coyne, P. Harvey, A. Purvis, I. Ribera, and D. Sim-
dition, we find a general pattern of lower levels of range berloff for comments on the manuscript.

Appendix

Table A1: List of clades used as examples

Number
of species Molecular Source of Sources of Distribution
Clade sampled marker phylogeny range data of group
Fairy wrens (genus: Malurus) 12/12 Allozyme Christidis Rowley and Australia
and Russell 1997 and New
Schodde Guinea
1997
Sylvid warblers (genus Sylvia)a 15/16 DNA-DNA Blondel et Baker 1997 Palearctic
al. 1996
Auks (family: Alcidae) 23/23 Sequence Friesen et Gaston and Holarctic
al. 1996 Jones 1998
Cranes (family: Gruidae) 15/15 Sequence Krajewski Johnsgard Worldwide
and Fetz- 1983
ner 1994
Swordtail fish (genus: Xiphophorus) 19/22 Sequence Meyer et al. Rauchenberger Central
1994 et al. 1990; America
Rosen 1960,
1979
Fruitflies (genus: Rhagoletis)b 20/21 Sequence Smith and Foote et al. Nearctic
Bush 1993
1997
Leafhoppers (genus: Flexamia) 36/44 Sequence Dietrich et Whitcomb and Nearctic
al. 1997 Hicks 1988
Tiger beetles (subgenus:
Ellipsoptera) 13/13 Sequence Barraclough Pearson et al. Nearctic
et al. 1997
1999b
a
In the genus Sylvia, we consider the monophyletic group of 16 species identified by the source phylogeny, which does not
include two species (Sylvia atricapilla and Sylvia borin) traditionally placed in the genus.
b
In the genus Rhagoletis, we consider the monophyletic group containing 21 North American species identified in the source
phylogeny. One species, Rhagoletis pomonella, is paraphyletic with respect to Rhagoletis mendax and Rhagoletis zephyria. Since
our methods compare monophyletic sister clades, we compare the clade R. pomonella, R. mendax, R. zephyria to its sister clade
but cannot consider finer patterns within this group.
432 The American Naturalist

Literature Cited in birds: a test of Lynch’s method. Evolution 48:

490–497.
Christidis, L., and R. Schodde. 1997. Relationships within
Avise, J. C., and D. Walker 1998. Pleistocene phylogeo-
the Australo-Papuan fairy-wrens (Aves: Malurinae): an
graphic effects on avian populations and the speciation
evaluation of the utility of allozyme data. Australian
process. Proceedings of the Royal Society of London B,
Journal of Zoology 45:113–129.
Biological Sciences 265:457–463.
Collwell, R. K., and D. W. Winkler. 1984. A null model
Baker, K. 1997. Warblers of Europe, Asia, and North Af-
rica. Christopher Helm, London. for null models in biogeography. Pages 344–359 in D.
Barraclough, T. G., A. P. Vogler, and P. H. Harvey. 1998. R. Strong, D. S. Simberloff, L. G. Abele, and A. B. This-
Revealing the factors that promote speciation. Philo- tle, eds. Ecological communities: conceptual issues and
sophical Transactions of the Royal Society of London the evidence. Princeton University Press, Princeton, N.J.
B, Biological Sciences 353:241–249. Coope, G. R., and A. S. Wilkins. 1994. The response of
———. 1999a. Revealing the factors that promote spe- insect faunas to glacial-interglacial climatic fluctuations.
ciation. Pages 202–219 in A. E. Magurran and R. M. Philosophical Transactions of the Royal Society of Lon-
May, eds. The evolution of biological diversity. Oxford don B, Biological Sciences 344:19–26.
University Press, Oxford. Coyne, J. A. 1994. Ernst Mayr and the origin of species.
Barraclough, T. G., J. E. Hogan, and A. P. Vogler. 1999b. Evolution 48:19–30.
Testing whether ecological factors promote cladogenesis Dietrich, C. H., R. F. Whitcomb, and W. C. Black. 1997.
in a group of tiger beetles (Coleoptera: Cicindelidae). Phylogeny of the grassland leafhopper genus Flexamia
Proceedings of the Royal Society of London B, Biological (Homoptera: Cicadellidae) based on mitochondrial
Sciences 266:1061–1067. DNA sequences. Molecular Phylogenetics and Evolution
Barton, N. H., and B. Charlesworth. 1984. Genetic revo- 8:139–149.
lutions, founder effects, and speciation. Annual Review Doebeli, M. 1996. A quantitative genetic competition
of Ecology and Systematics 15:133–164. model for sympatric speciation. Journal of Evolutionary
Bleiweiss, R. 1998. Relative-rate tests aid biological causes Biology 9:893–909.
of molecular evolution in hummingbirds. Molecular Bi- Duffy, J. E. 1996. Resource-associated population subdi-
ology and Evolution 15:481–491. vision in a symbiotic coral-reef shrimp. Evolution 50:
Blondel, J., F. Catzeflis, and P. Perret. 1996. Molecular 360–373.
phylogeny and the historical biogeography of the war- Elias, S. A. 1992. Late quaternary zoogeography of the
blers of the genus Sylvia (Aves). Journal of Evolutionary chihuahuan desert insect fauna, based on fossil records
Biology 9:871–891. from pack-rat middens. Journal of Biogeography 19:
Brown, J. H., and A. C. Gibson. 1983. Biogeography. 285–297.
Mosby, St. Louis. Farrell, B. D. 1998. “Inordinate fondness” explained: why
Brown, J. H., G. C. Stevens, and D. M. Kaufman. 1996. are there so many beetles? Science (Washington, D.C.)
The geographic range: size, shape, boundaries, and in- 281:555–559.
ternal structure. Annual Review of Ecology and Syste- Foote, R. H., F. L. Blanc, and A. L. Norrbom. 1993. Hand-
matics 27:597–623. book of the fruitflies (Diptera; Tephritidae) of America
Bush, G. L. 1969. Sympatric host race formation and spe- north of Mexico. Comstock, Ithaca, N.Y.
ciation in frugivorous flies in the genus Rhagoletis. Evo- Friesen, V. L., A. J. Baker, and J. F. Piatt. 1996. Phylogenetic
lution 23:237–251. relationships within the Alcidae (Charadriiformes: Aves)
———. 1975. Modes of animal speciation. Annual Review inferred from total molecular evidence. Molecular Bi-
of Ecology and Systematics 6:339–364. ology and Evolution 13:359–367.
———. 1994. Sympatric speciation in animals—new wine Gaston, A. J., and I. L. Jones. 1998. The auks. Oxford
in old bottles. Trends in Ecology & Evolution 9:285–288. University Press, Oxford.
Carson, H. L., and A. R. Templeton. 1984. Genetic revo- Gaston, K. J. 1998. Species-range distributions: products
lutions in relation to speciation phenomena: the found- of speciation, extinction and transformation. Philo-
ing of new populations. Annual Review of Ecology and sophical Transactions of the Royal Society of London
Systematics 15:97–131. B, Biological Sciences 353:219–230.
Charlesworth, B. 1997. Is founder-flush speciation defen- Gotelli, N. J., and G. R. Graves. 1996. Null models in
sible? American Naturalist 149:600–603. ecology. Smithsonian Institution, Washington, D.C.
Chesser, R. T., and R. M. Zink. 1994. Modes of speciation Gotelli, N. J., N. J. Buckley, and J. A. Wiens. 1997. Co-
 Geographic Pattern of Speciation 433

occurrence of Australian land birds: Diamond’s assem- MacArthur, R. H. 1960. On the relative abundance of spe-
bly rules revisted. Oikos 80:311–324. cies. American Naturalist 94:25–36.
Graham, R. W., E. L. Lundelius, M. A. Graham, E. K. Mayr, E. 1963. Animal species and evolution. Harvard Uni-
Schroeder, R. S. Toomey, E. Anderson, A. D. Barnosky, versity Press, Cambridge, Mass.
et al. 1996. Spatial response of mammals to late qua- Maynard Smith, J. 1966. Sympatric speciation. American
ternary environmental fluctuations. Science (Washing- Naturalist 100:637–50.
ton, D.C.) 272:1601–1606. McCall, R. A., P. H. Harvey, and S. Nee. In press. Avian
Gutierrez, D., and R. Menendez. 1997. Patterns in the dispersal ability and the tendency to form island-en-
distribution, abundance and body size of carabid beetles demic species on continental and oceanic islands. Bio-
(Coleoptera: Caraboidea) in relation to dispersal ability. diversity Letters.
Journal of Biogeography 24:903–914. Meyer, A., J. M. Morrisey, and M. Schartl. 1994. Recurrent
Harvey, P. H., R. M. May, and S. Nee. 1994. Phylogenies origin of a sexually selected trait in Xiphophorus fishes
without fossils. Evolution 48:523–529. inferred from a molecular phylogeny. Nature (London)
Hayden, D., R. R. Radtkey, and E. R. Pianka. 1993. Ex- 368:539–542.
perimental biogeography: Interactions between sto- Mindell, D. P., and C. E. Thacker. 1996. Rates of molecular
chastic, historical and ecological processes in a model evolution—phylogenetic issues and applications. An-
archipelago. Pages 117–130 in R. E. Ricklefs and D. nual Review of Ecology and Systematics 27:279–303.
Schluter, eds. Species diversity in ecological commu- Moya, A., A. Galiana, and F. J. Ayala. 1995. Founder-effect
nities: historical and geographical perspectives. Univer- speciation theory—failure of experimental corrobora-
sity of Chicago Press, Chicago. tion. Proceedings of the National Academy of Sciences
Jablonski, D. 1987. Heritability at the species level: analysis of the USA 92:3983–3986.
of geographic ranges of Cretaceous mollusks. Science Naeem, S., and B. A. Hawkins. 1994. Minimal community
(Washington, D.C.) 238:129–133. structure—how parasitoids divide resources. Ecology
Johnsgard, P. A. 1983. Cranes of the world. Croom Helm, 75:79–85.
Beckenham. Nee, S., E. C. Holmes, R. M. May, and P. H. Harvey. 1994.
Johnson, P. A., F. C. Hoppensteadt, J. J. Smith, and G. L. Extinction rates can be estimated from molecular phy-
Bush. 1996. Conditions for sympatric speciation—a dip- logenies. Philosophical Transactions of the Royal Society
loid model incorporating habitat fidelity and non-hab- of London B, Biological Sciences 349:25–31.
itat assortative mating. Evolutionary Ecology 10: Nee, S., T. G. Barraclough, and P. H. Harvey. 1996. Tem-
187–205. poral changes in biodiversity: detecting patterns and
Klicka, J., and R. M. Zink. 1997. The importance of recent identifying causes. Pages 230–252 in K. J. Gaston, ed.
ice ages in speciation: a failed paradigm. Science (Wash- Biodiversity: a biology of numbers and difference.
ington, D.C.) 277:1666–1669. Blackwell Science, Oxford.
Krajewski, C., and J. W. Fetzner, Jr. 1994. Phylogeny of Neigel, J. E., and J. C. Avise. 1993. Application of a random
cranes (Gruiformes: Gruidae) based on cytochrome-b walk model to geographic distributions of animal mi-
DNA sequences. Auk 111:351–365. tochondrial DNA variation. Genetics 135:1209–1220.
Letcher, A. J., A. Purvis, S. Nee, and P. H. Harvey. 1994. Omland, K. E. 1997. Correlated rates of molecular and
Patterns of overlap in the geographic ranges of Palearctic morphological evolution. Evolution 5:1381–1393.
and British mammals. Journal of Animal Ecology 63: Pearson, D. L., T. G. Barraclough, and A. P. Vogler. 1997.
871–879. Distributional maps for North American species of tiger
Losos, J. B., T. R. Jackman, A. Larson, K. deQueiroz, and beetles (Coleoptera: Cicindelidae). Cicindela 29:33–84.
L. Rodriguez Schettino. 1998. Contingency and deter- Purvis, A. 1996. Using interspecies phylogenies to test
minism in replicated adaptive radiations of island liz- macroevolutionary hypotheses. Pages 153–168 in P. H.
ards. Science (Washington, D.C.) 279:2115–2118. Harvey, A. J. Leigh Brown, J. Maynard Smith, and S.
Lutzoni, F., and M. Pagel. 1997. Accelerated evolution as Nee, eds. New uses for new phylogenies. Oxford Uni-
a consequence of transitions to mutualism. Proceedings versity Press, Oxford.
of the National Academy of Sciences of the USA 94: Rambaut, A., and L. Bromham. 1998. Estimating diver-
11422–11427. gence dates from molecular sequences. Molecular Bio-
Lynch, J. D. 1989. The gauge of speciation: on the fre- logy and Evolution 15:442–448.
quency of modes of speciation. Pages 527–553 in D. Rauchenberger, M., K. D. Kallman, and D. C. Morizot.
Otte and J. A. Endler, eds. Speciation and its conse- 1990. Monophyly and geography of the Rio Panuco
quences. Sinauer, Sunderland, Mass. basin swordtails (genus: Xiphophorus) with description
434 The American Naturalist

of four new species. American Museum of Natural His- organization on gene flow in the fire ant Solenopsis in-
tory Novitates 2975:1–41. victa. Nature (London) 383:613–616.
Raup, D. M., S. J. Gould, T. J. M. Schopf, and D. S. Sim- Slatkin, M. 1996. Defense of founder-flush theories of spe-
berloff. 1973. Stochastic models of phylogeny and the ciation. American Naturalist 147:493–505.
evolution of diversity. Journal of Geology 81:525–542. ———. 1997. Is founder-flush speciation defensible? Re-
Rice, W. R., and E. E. Hostert. 1993. Laboratory experi- ply. American Naturalist 149:604–605.
ments on speciation: what have we learned in the last Smith, J. J., and Bush, G. L. 1997. Phylogeny of the genus
forty years? Evolution 47:1637–1653. Rhagoletis (Diptera: Tephritidae) inferred from DNA se-
Ricklefs, R. E., and R. E. Latham. 1992. Intercontinental quences of mitochondrial cytochrome oxidase II. Mo-
correlation of geographical ranges suggests stasis in eco- lecular Phylogenetics and Evolution 7:33–43.
logical traits of relict genera of temperate perennial Sokal, R. R., and F. J. Rohlf. 1981. Biometry. W. H. Free-
herbs. American Naturalist 139:1305–1321. man, New York.
Rosen, D. E. 1960. Middle American poeciliid fishes of the Strong, D. R., D. S. Simberloff, L. G. Abele, and A. B.
genus Xiphophorus. Bulletin of the Florida State Mu- Thistle. 1984. Ecological communities: conceptual issues
seum, Biological Sciences 5:57–242. and the evidence. Princeton University Press, Princeton,
———. 1979. Fishes from the uplands and intermontane N.J.
basins of Guatemala: revisionary studies and compar- Templeton, A. 1981. Mechanisms of speciation—a pop-
ative geography. Bulletin of the American Museum of ulation genetic approach. Annual Review of Ecology and
Natural History 162:267–376. Systematics 12:23–41.
Rosenzweig, M. L. 1978. Competitive speciation. Biological Templeton, A. R. 1996. Experimental evidence for the ge-
Journal of the Linnean Society 10:275–289. netic transilience model of speciation. Evolution 50:
Rowley, I., and E. M. Russell. 1997. Fairy-wrens and grass- 909–915.
wrens: Maluridae. Oxford University Press, Oxford. Whitcomb, R. F., and A. L. Hicks. 1988. Genus Flexamia:
Sanderson, M. J., and M. J. Donoghue. 1996. Recon- new species, phylogeny, and ecology. Great Basin Nat-
structing shifts in diversification rates on phylogenetic uralist Memoirs 12:224–323.
trees. Trends in Ecology & Evolution 11:15–20.
Shoemaker, D. D., and K. G. Ross. 1996. Effects of social Associate Editor: Daniel Simberloff