Content uploaded by Stuart L Pimm
Author content
All content in this area was uploaded by Stuart L Pimm on Feb 09, 2018
Content may be subject to copyright.
Contributed Paper
Estimating the normal background rate of species
extinction
Jurriaan M. De Vos,∗† Lucas N. Joppa,‡ John L. Gittleman,§ Patrick R. Stephens,§
and Stuart L. Pimm∗∗
∗Institute of Systematic Botany, University of Zurich, Zollikerstrasse 107, 8008, Z¨
urich, Switzerland
†Department of Ecology and Evolutionary Biology, Brown University, Box G-W, Providence, RI, 02912, U.S.A.
‡Microsoft Research, 21 Station Road, Cambridge CB1 2FB, United Kingdom
§Odum School of Ecology, University of Georgia, Athens, GA, 30602, U.S.A.
∗∗Nicholas School of the Environment, Duke University, Durham, NC, 27708, U.S.A., email stuartpimm@me.com
Abstract: A key measure of humanity’s global impact is by how much it has increased species extinction rates.
Familiar statements are that these are 100–1000 times pre-human or background extinction levels. Estimating
recent rates is straightforward, but establishing a background rate for comparison is not. Previous researchers
chose an approximate benchmark of 1 extinction per million species per year (E/MSY). We explored disparate
lines of evidence that suggest a substantially lower estimate. Fossil data yield direct estimates of extinction
rates, but they are temporally coarse, mostly limited to marine hard-bodied taxa, and generally involve
genera not species. Based on these data, typical background loss is 0.01 genera per million genera per year.
Molecular phylogenies are available for more taxa and ecosystems, but it is debated whether they can be used
to estimate separately speciation and extinction rates. We selected data to address known concerns and used
them to determine median extinction estimates from statistical distributions of probable values for terrestrial
plants and animals. We then created simulations to explore effects of violating model assumptions. Finally,
we compiled estimates of diversification—the difference between speciation and extinction rates for different
taxa. Median estimates of extinction rates ranged from 0.023 to 0.135 E/MSY. Simulation results suggested
over- and under-estimation of extinction from individual phylogenies partially canceled each other out when
large sets of phylogenies were analyzed. There was no evidence for recent and widespread pre-human overall
declines in diversity. This implies that average extinction rates are less than average diversification rates.
Median diversification rates were 0.05–0.2 new species per million species per year. On the basis of these
results, we concluded that typical rates of background extinction may be closer to 0.1 E/MSY. Thus, current
extinction rates are 1,000 times higher than natural background rates of extinction and future rates are likely
to be 10,000 times higher.
Keywords: diversification rates, extinction rate, fossil record, lineages through time, molecular phylogenies
Estimaci´
on de la Tasa Normal de Extinci´
on de Especies
Resumen: Una medida clave del impacto global de la humanidad es cu´
anto han incrementado las tasas de
extinci´
on de las especies. Las declaraciones conocidas establecen que estas son 100 – 1,000 veces los niveles
de extinci´
on pre-humanos o de fondo. Estimar las tasas recientes es un proceso directo, pero establecer una
tasa de fondo para comparar no lo es. Investigadores previos han elegido un punto de referencia aproximado
de una extinci´
on por mill´
on de especies por a˜
no (E/MEA). Exploramos l´
ıneas dispares de evidencia que
sugieren un estimado sustancialmente m´
as bajo. Los datos f´
osiles producen estimados directos de las tasas de
extinci´
on, pero son temporalmente burdos, en su mayor´
ıa limitados a los taxones marinos de cuerpos duros,
Paper submitted February 11, 2014; revised manuscript accepted June 22, 2014.
452
Conservation Biology, Volume 29, No. 2, 452–462
C
2014 Society for Conservation Biology
DOI: 10.1111/cobi.12380
de Vos et al. 453
y generalmente involucran a los g´
eneros y no a las especies. Bas´
andonos en estos datos, la p´
erdida de fondo
t´
ıpica es de 0.01 g´
eneros por mill´
on de g´
eneros por a˜
no. Las filogenias moleculares est´
an disponibles para
m´
as taxones y ecosistemas, pero se debate si pueden usarse para estimar por separado las tasas de extinci´
on
y especiaci´
on. Seleccionamos datos para dirigirnos a asuntos conocidos y los usamos para determinar los
estimados de extinci´
on medios a partir de distribuciones estad´
ısticas de valores probables para plantas y
animales terrestres. Despu´
es creamos simulaciones para explorar los efectos de las suposiciones del modelo
de violaci´
on. Finalmente, recopilamos los estimados de diversificaci´
on – la diferencia entre las tasas de
especiaci´
on y extinci´
on para taxones diferentes. Los estimados medios de las tasas de extinci´
on variaron
desde 0.023 hasta 0.135 E/MEA. Los resultados de la simulaci´
on sugirieron una sobre- y subestimaci´
on de
la extinci´
on a partir filogenias individuales que se cancelaron unas a otras cuando se analizaron conjuntos
grandes de filogenias. No hubo evidencia de declinaciones generales pre-humanas, recientes y extensas en
la diversidad. Esto implica que las tasas de extinci´
on promedio son menores a las tasas de diversificaci´
on
promedio. Las tasas medias de diversificaci´
on fueron 0.05 – 0.2 especies nuevas por mill´
on de especies por
a˜
no. Con base en estos resultados, concluimos que las t´
ıpicas tasas de extinci´
on de fondo pueden ser m´
as
cercanas a 0.1 E/MEA. As´
ı, las tasas de extinci´
on actuales son mil veces m´
as altas que las tasas naturales de
extinci´
on de fondo y que las tasas futuras probablemente sean 10, 000 veces m´
as altas.
Palabras Clave: filogenias moleculares, linajes a trav´
es del tiempo, registro f´
osil, tasa de diversificaci´
on, tasa de
extinci´
on
Introduction
. . . a mass extinction crisis, with a rate of extinction now
1,000 times higher than the normal background rate.
Al Gore (2006)
Some quantitative statement about the current extinc-
tion crisis is needed. As Gore’s remark exemplifies, it
has become standard to quantify present extinctions as
rates and then to compare them to some background that
typifies geological history. Mass extinction events are un-
derstood to be exceptional events when extinction rates
are episodically much higher. Gore follows Pimm et al.
(1995) who estimated “recent extinction rates are 100 to
1000 times their pre-human levels.” Comparing current
and background extinction rates raises a series of issues
that we address in this article. Estimating the current
rates of extinctions is straightforward (Pimm et al. 1995,
2006, 2014). Here, we explored the more difficult task
of estimating the background rate of extinction. By this,
we mean the geologically recent rate of extinction before
human actions inflated them.
Three data sources bear on the background rate: the
fossil record, the overall diversification rates of species,
and the detailed patterns of how many species accumu-
late in a lineage over time elucidated from molecular
phylogenies. One can use fossil data to estimate extinc-
tion rates directly and contribute essential information,
but their use has limitations. Most data are for marine
animals, most studies assess genera, not species, and the
temporal resolution of the data is poor. Importantly, they
can be used to identify periods when extinction rates
exceed speciation events, that is, when lineages shrink
in their numbers of species.
We concentrated on insights that may be drawn from
molecular phylogenies. Richly detailed phylogenies are
readily available for many taxa in many ecosystems.
Moreover, it may be possible to estimate speciation and
extinction rates separately from a lineage’s diversification
from one to many species. Problematically, the patterns
of branching times in the phylogenies of many lineages
result in a zero estimate of extinction rate for the
simplest model of constant speciation and extinction.
That motivated us to ask two questions. First, what are
the upper bounds of extinction rates of lineages derived
from this model? Second, how severe are the effects of
violating the simple model’s assumptions? We created
an extensive set of simulations under a wide variety of
conditions of changing speciation and extinct rates.
Overall net rates of diversification—the difference be-
tween speciation and extinction—are very widely avail-
able and simple to calculate, but they seem unpromising.
Any given net diversification rate might result from a very
high rate of extinction and an only slightly higher rate
of speciation. Moreover, there might be many lineages
for which extinctions exceed speciation. Of course, if
this occurred in the recent (but pre-human) past, then
the analyses of detailed phylogenies and the fossil record
would demonstrate it. That this is not happening implies
that average extinction rates are less than average diver-
sification rates.
Expressing the Magnitude of Present Day Extinctions
Before 1995, the literature often quoted statistics on
current extinctions in terms of species per day. Estimates
ranged from a minimum of three (Myers 1989) to “a
hundred species per day” (Stork 2010). More than the
uncertainties about the extinctions themselves, the
Conservation Biology
Volume 29, No. 2, 2015
454 Background Rate of Extinction
numbers reflected the wide range of estimates for how
many eukaryote species there are (Pimm et al. 2014).
The uncertainty in the species-per-day estimates also
posed problems when dealing with critics of environ-
mental concerns who demanded the scientific names
of recently extinct species (Stork 2010). Of course, tax-
onomists have described only a small fraction of species,
while the IUCN’s Red List (www.iucnredlist.org) has
assessed an even smaller fraction of those (approxi-
mately 72,000 species). To avoid the necessarily complex
caveats for extinctions per day estimates, Pimm et al.
(1995) replaced this metric with a proportional rate that
they could calculate for a given taxon.
This approach measures extinction rates by follow-
ing cohorts of species and then tallying how many suc-
cumbed subsequently. For example, by 1845 taxonomists
had described 5000 bird species. Sixty-one of these
were extinct by 2012. Dividing the number of extinct
species (61) by the product of 167 years (2012–1845)
and total species (5000), one derives the extinction rate
as 61/(167∗5000) =0.000073. Multiplying that rate by
1,000,000 gives an extinction rate of approximately 73
extinctions per million species per year (E/MSY), or
0.073% per decade (Pimm et al. 2006). In contrast to
an uncertain estimate of any day’s extinctions, most of
which will be unknown to taxonomists, we know these
extinctions in detail (Pimm et al. 1995).
Calculating the Background Rate from the Fossil Record
The fossil record suggests that recognized taxa live from 1
to 10 million years, with some obvious exceptions. Pimm
et al. (1995) chose the shorter duration, which gives a
background rate of 1 E/MSY, and called it a “benchmark.”
Barnosky et al. (2011), Harnik et al. (2012), and Alroy
(1996) greatly expand on this superficial estimate.
Alroy (1996) estimated 0.165 extinctions of genera
per million genera years for Cenozoic mammals. Most
paleontological studies assess genera, not species (Flessa
& Jablonski 1985). Harnik et al. (2012) examined marine
taxa. Their extinction rate is a dimensionless extinction
fraction, the natural logarithm of the fractional survival
of genera measured over an average stage length of
7 million years. Converting these fractions to their
corresponding rates yields values for the last few million
years of 0.06 genera extinctions per million genera for
cetaceans, 0.04 for marine carnivores, and from 0.001
(brachiopods) to 0.01 (echinoids) for a variety of marine
invertebrates. Put another way, 1% of echinoid genera
are lost per million years.
Species Diversification Rates from Molecular Phylogenies
Molecular phylogenies provide an appealing alternative
to the fossil record’s shortcomings because they cover
a large range of taxa, periods, and environments. In the
1
10
100
-15 -10 -5
Number of lineages
Time back to basal divergence (My)
Disa (Orchidaceae)
Bursera (Burseraceae)
Figure 1. The logarithms of the number of lineages
over time for two plant genera and the relationships
expected from a simple birth–death model Eq. (1) that
assumes speciation and extinction rates are constant
through time. The simple model expects an
exponential increase in the numbers of lineages over
time, or a linear increase on this logarithmic scale
(see text). The exponential fit of Disa lineages versus
time from 10 to 2 million years before present is
shown with the dashed line, thereafter, the observed
data are above this line. For Bursera, the dashed black
line shows exponential fit of lineages versus time from
10 to 4 million years before present, thereafter, the
observed data are below this line.
simplest case, a single lineage grows to a clade size of N
species, E(N), during time, t, according to
E(N)=exp(r∗t),(1)
where ris the net diversification rate (i.e., the difference
between speciation, λ, and extinction, μ), which can be
estimated easily (Kendall 1948; Moran 1953):
(λ−μ)=r=ln(N)/t.(2)
Can one separately estimate the speciation and
extinction rates? With Eq. 1, the plot of the logarithm
of the number of lineages through time (LTT) is linear
(slope λ–μ), but with an important qualification. In
the limit of the present day, there are no extinctions
of the most recent taxa—they have not yet happened.
Thus, near the present, the LTT slope should increase
and approach λ, the speciation rate (Nee et al. 1994;
Nee 2006). Figure 1 shows an example for South African
orchids of the genus Disa.
Practice is considerably more complicated than simple
theory (Morlon et al. 2011; Etienne et al. 2012). Lambda
and μmay be time dependent, functionally related, or
Conservation Biology
Volume 29, No. 2, 2015
de Vos et al. 455
depend on the number of species already present, and
they will likely vary from place to place and among taxa.
We considered the compilations of McPeek (2008) in
which 80% of the studies have LTT graphs that curve
downward on the log-linear scale. Figure 1 shows an
example of this for Central American species in the genus
Bursera (De-Nova et al. 2012).
Such studies yield a maximum likelihood (ML) esti-
mate of zero extinction, but for all statistical estimates it
is standard to determine their probability distributions.
Furthermore, we used simulations to explore whether
such methods appropriately reconstruct parameter value
ranges when the diversification process is more complex
than the simple model. Moreover, some phylogenies have
features that make them unsuitable for separating speci-
ation and extinction rates. That motivated our selection
of phylogenies to analyze.
Methods
Diversification Rates
Our choices of data reflected our need to compare es-
timates of diversification. Phillimore and Price (2008)
provide 40 dated phylogenies of bird groups. Ferrer and
Good (2012b) provide net diversification rates of all plant
families. Valente et al. (2010) searched the literature to
find exceptionally high rates of diversification. We use
two measures of diversification: Kendall-Moran (Kendall
1948; Moran 1953) and Magallon-Sanderson (Magallon &
Sanderson 2001).
The Kendall-Moran estimator measures the average net
diversification Eq. 2; tis stem age, the time since a lineage
diverged from its sister; in contrast, crown age is the
time a lineage has been known from its earliest species.
This estimator (see below) is corrected by the Magallon-
Sanderson estimator. (In their original publication, Ferrer
and Good [2012b] used base 10 for their logarithms, not
base e, and later posted a corrigendum [Ferrer & Good
2012a].)
The Magallon-Sanderson estimator corrects net
diversification for non-zero extinction. If extinction is
not zero, some extant species will go extinct in the
near future, which makes the estimates larger than they
should be. One does not know the extinction rate a
priori. So, following Magallon and Sanderson (2001), we
used arbitrarily low and high extinction fractions (μ/λ)
of 0 (i.e., the Kendall-Moran estimate) and of 0.9 and
compared the estimates.
Diversification and its Component Speciation and Extinction
Rates
In a perfect world, phylogenies would be complete for
all species and evolutionary relationships fully resolved.
Practice falls short. Thus, we included two kinds of
relevant data: data from many different groups and from
taxonomically diverse lineages and data specifically
selected for some taxa.
To cover considerable taxonomic breadth, we used
McPeek’s (2008) data set of 182 dated phylogenies
of Chordata, Mollusca, Magnoliophyta, and Arthropoda
from which we excluded small phylogenies. This data
set places no particular emphasis on the quality of the
phylogenetic trees in terms of number of taxa or dating
methodology employed. The fraction of extant species
included in each clade is known and high (>50%), how-
ever. That property is essential because the signal of ex-
tinctioninphylogenies—anupturninanLTTtowardthe
present—is only detectable when one samples a high
proportion of extant species. That is also the reason for
using a clade-by-clade approach, rather than one based on
information available from a single very large supertree
that covers considerable taxonomic breadth with few
sampled species (e.g., Rabosky et al. 2012). We com-
pared results of the plant phylogenies of McPeek with
carefully selected plant phylogenies to confirm that the
concerns associated with use of McPeek’s data set would
not invalidate the results (Supporting Information). We
included mammals because the data are typically for a
genus or some other subset of species at a taxonomic
rank below family.
Mammals and Plants
We compiled a list of all mammal families with at least
five species, based on Wilson and Reeder (2005), and
estimated their ages from a published chronogram of
all mammals (Fritz et al. 2009). We also identified mam-
malian orders and families that were at least 90% resolved
in the supertree and extracted these subtrees for use
in additional calculations. The latter process yielded 15
subtrees. We used the R package Ape v 2.7 for tree ma-
nipulations (Paradis et al. 2004).
We carefully selected 37 dated plant phylogenies
based on the inference and taxon sampling methods
used so as to ensure branch lengths and divergence
times were derived with the best available methods
(Supporting Information).
Analyses
Given the large number of uncertainties in the data on
observed species and the origination data we had for
them, our overall aim was to generate probability distri-
butions for the parameters given these uncertainties. We
did so by employing the constant-rate birth–death model
(Nee et al. 1994) with a correction for un-sampled taxa
(Bokma 2008), as implemented in a Bayesian framework
in the software SubT (Ryberg et al. 2011).
Conservation Biology
Volume 29, No. 2, 2015
456 Background Rate of Extinction
We created one joint frequency histogram by adding
the Markov Chain Monte Carlo samples that the algorithm
generated from the analyses of all phylogenies. Summary
histograms represent an estimate of the parameter for
the entire taxonomic group, while fully accounting for
the uncertainty associated with speciation and extinc-
tion estimates from any of the individual phylogenies.
We also calculated the interval containing 95% of the
observations.
Simulated Data
We simulated the possible effects of violations of the
assumptions of the constant rate birth–death model
on estimates of extinction rates from phylogenies in
subT. We also assessed the consequences of missing
species. We first generated phylogenies under models
of constant, lineage-specific, time-slice-specific, and
diversity-dependent speciation and extinction rates, and
combinations thereof, for a total of 36 diversification
scenarios. Then, we pruned one-third of extant tips to
reflect incomplete taxon sampling. Subsequently, we
estimated the rates of speciation and extinction from
the pruned and un-pruned simulated data with two
methods based on the assumption of constant rates: the
Bayesian implementation (BI) of the constant-rates birth–
death model (which accounts for missing species by
treating them as random variables and assumes positive
diversification rates), as employed for the empirical
phylogenies (subT; Bokma 2008; Ryberg et al. 2011), and
a ML estimator (which allows for negative diversification
rates and accounts for missing species via its likelihood
expression; R-package diversitree [FitzJohn 2012]).
For each simulated phylogeny, we determined
whether the estimated 95% interval of highest posterior
density of the subT result included the true, generating
value or values of extinction and determined the differ-
ence between the estimated rate and these true values.
We then combined all results to determine whether there
was a systematic bias in over- or underestimation extinc-
tion rates compared with their true values. For phyloge-
nies with missing taxa, we compared the performance of
the Bayesian and ML methods for accounting for them.
Technical details of analyses are discussed in Supporting
Information, where we also address further criticisms of
these models and how we addressed them.
Results
Rates of Diversification
Median rates of diversification were approximately 0.1
species per million species years (Table 1). All but one
of the intervals that spanned 95% of the estimated diver-
sification rates were below 1 species per million species
years (Table 1).
Rates of Speciation and Extinction
These results are complex in their detail and generate
numerous caveats that are discussed in Supporting Infor-
mation. Nonetheless, their key features are simply stated
and best viewed in comparison to the summaries of data
already published.
The most frequent extinction rate estimates (μ;Fig.2
middle panels) were near zero. Median and upper 95%
extinction rates, respectively, were for chordates 0.064
and 0.586, for plants 0.053 and 0.352, for arthropods
0.09 and 0.934, for mammals 0.023 and 0.102, and for
Molluscs 0.135 and 1.672. Median and upper 95% esti-
mates, respectively, were for chordates 0.241 and 0.751,
for plants 0.352 and 0.877, for arthropods 0.320 and
0.923, for mammals 0.257 and 0.778, and for molluscs
0.357 and 0.825. For plants, the data set of McPeek
yielded fully congruent results to our carefully selected
phylogenies, though the former were associated with
slightly wider confidence intervals (Supporting
Information).
The assumptions underlying the model may not have
been met. So, we explored various violations of the as-
sumptions. The results of our simulated data were numer-
ous and complex and are discussed in detail in Supporting
Information. We only briefly summarize them here.
Both Bayesian and ML methods often incorrectly in-
ferred extinction rates from individual phylogenies (in
10–49% of phylogenies, depending on the scenario; Sup-
porting Information). This entailed both over- and un-
derestimations. Overall, without missing species, these
effects cancelled each other out to a large extent so
that the overall mean or median inferred rates approx-
imated the true, generating values (Supporting Informa-
tion). When there were missing species, the Bayesian
method outperformed the ML method. It only slightly un-
derestimated extinction, whereas the ML method greatly
overestimated extinction (Supporting Information). Gen-
erally across scenarios, high extinction rates were un-
derestimated, low rates were overestimated, and when
a phylogeny contained multiple rates, an intermediate,
average rate was found. Although complete phyloge-
nies yielded more accurate results than phylogenies with
many missing species, the correct order of magnitude
was inferred. Complete phylogenies and phylogenies
with missing species yielded congruent results under all
36 combinations of simulation conditions (Supporting
Information).
While estimates for individual phylogenies were some-
times in substantial error, estimates averaged across phy-
logenies were accurate in order of magnitude.
Conservation Biology
Volume 29, No. 2, 2015
de Vos et al. 457
Figure 2. Diversification and extinction rates (expressed as rates per lineage per million years) and
extinctionfraction (dimensionless) in descending order by sample size for five taxonomic groups. For each
taxonomic group,the top graph shows the frequency distribution of 7500 sample parameter values from each of
the phylogenies inthe group. Thus, the data underlying the summary histogram of, for example, mollusks has
8∗7500 samples. Thebold horizontal line shows the shortest interval containing 95% of the values. The bottom
graphs show descendingorder of extinction rate. We list the individual studies in the order shown in this figure in
the SupportingInformation. The median values are the vertical lines within the bars, the 95% confidence intervals
(i.e., the 95% highest posterior density) are the open boxes, and points are outliers.
Conservation Biology
Volume 29, No. 2, 2015
458 Background Rate of Extinction
Figure 2. (continued)
Discussion
Extinction Rates from the Fossil Record
Fossils provide essential information, but are limited
in temporal resolution and taxonomic breadth (Purvis
2008). Estimates based on the fossil record had the
obvious bias that taxa identified to species may have
been present somewhere before they were first recorded
and after they were last recorded. Unless corrected,
longevities are potentially underestimated and extinction
rates are overestimated (Foote & Raup 1996).
Interpolating fractional survival based on stages aver-
aging several million years to shorter intervals of a million
years is problematical. P. G. Harnik (personal communi-
cation) suggests interpolation imputes homogenous rates
of extinction across geological stages, while evidence
suggests extinction rates are pulsed toward their end
(Alroy 2008; Foote 2005). Indeed, changes in floras and
faunas often define when one stage ends and the other
starts. If so, for millions of years, the rates would have
been even lower than values we have presented above
and would have been followed by episodes when they
were higher. Thus, our earlier estimate of 1% of echinoid
genera lost per million years at an implied constant rate
may be incorrect. We can say only that 7% of echinoid
genera are lost over 7 million years.
Comparing rates of generic extinctions with species
extinction is complex (Russell et al. 1998). Mammal
genera contain an average of 4.4 species. Were all
mammal genera to have four species in them, then
individual species extinction rates of 0.63 E/MSY would
give the observed generic extinction rate of 0.165 (Alroy
1996). For five species, the species extinction rate would
be 0.69 E/MSY. There are two problems with this. Most
generic extinctions are likely those in monotypic genera.
Second, these calculations are based on an assumption of
Conservation Biology
Volume 29, No. 2, 2015
de Vos et al. 459
statistical independence. As Russell et al. (1998) show,
species extinctions within a genus are highly contingent.
Some genera and some places are much more vulnerable
to extinction than others. Combined, these considera-
tions would lead to the species extinction rate being
close to the generic rate of extinction (i.e., 0.165 E/MSY).
In short, when one is comparing the extinction rates
of fossils with present biota, one is contrasting genera
with species and, of course, typically very different taxa
(often marine taxa with terrestrial taxa). Nonetheless,
to the extent these comparisons are reasonable, the data
strongly suggest that a benchmark of 1 E/MSY is too high.
Extinction Rates from Phylogenies
Estimating extinction rates from phylogenies is contro-
versial. First, our selection of phylogenies did not include
groups that are both young and small because we would
not have had the statistical power to detect anything.
Nor did it include groups that are very old and small. For
these, we additionally were concerned that strong, yet
undetectable fluctuations in rates over time would bias
our inference. The groups we included showed strong
similarities in the rates of extinction inferred (Fig. 2). We
do not know if groups we could not study would differ
greatly. The advantage of including larger groups is that
our results are informative for many species, hence, we
hope representative.
Second, as Rabosky (2009b) showed in his uncompro-
misingly titled paper “Extinction Rates Should Not Be
Estimated from Molecular Phylogenies,” ML estimates can
produce an overestimation of the real extinction rate if
among-lineage variation in diversification rate increases.
We addressed this concern in two ways. We excluded
large phylogenies (>500 species) for which this issue
may be particularly problematic. Then, we provided a
range of possible values for each phylogeny by means of
Bayesian inference. The expectation is that when there is
conflict within the subsets of data on the relative extinc-
tion rate (e.g., high or low), a Bayesian analysis would
provide the range of possible solutions as output. Alter-
native approaches to accommodate such among-lineage
rate variation by splitting a phylogeny into multiple rate
partitions (Alfaro et al. 2009; Rabosky et al. 2013) pose
other statistical challenges (Supporting Information). Im-
portantly, if Rabosky’s (2009a) concerns apply, it means
our estimates of extinction rates are too high and the
comparisons to present day rates too low. In short, his
concerns make our results conservative.
Third and conversely, many phylogenies showed a
slower than exponential increase in lineages over time
near the present. Ecological factors may limit the maxi-
mum size a clade may achieve (Rabosky & Lovette 2008;
Morlon et al. 2010; Morlon et al. 2011; Etienne & Rosin-
dell 2012; Etienne et al. 2012). Price et al. (2014) provide
a detailed example for Himalayan birds that argues niche
saturation forces such slower than exponential patterns.
In addition, speciation takes time to become complete
(Etienne & Rosindell 2012).
Whatever the causes, the consequence of this slowing
is that it obscures on-going extinctions and so makes
our estimates of their rate too low. Indeed, for Bursera
(Fig. 1), it seems improbable a priori that their extinction
rate has been exactly zero for 7 million years.
To address these concerns, we undertook numerous
simulations of known diversifications, with varying frac-
tions of species removed from these modeled clades and,
in some cases, different rates of diversification (Support-
ing Information). At best, our simulations demonstrated
that the underestimation of extinction due to complex
diversification processes may be slight. We recovered
the correct order of magnitude of absolute extinction
across replicate phylogenies, even though individual es-
timates are associated with large uncertainties. Our find-
ings underline the recent conclusion that phylogenies of
extant taxa contain some information on extinction rates
(Pyron & Burbrink 2013), even when assumptions of sim-
ple models are not fully met.
Finally, taxonomists may fall short of recognizing
all lineages that will give rise to new species in the
future (Phillimore & Price 2008). The entirely arbitrary
taxonomic decision of whether to group geographically
isolated populations as one species or split them into
several recently derived species affects estimates of
extinction rates.
Despite those difficulties, an important conclusion
emerges. Of the 140 phylogenies we analyzed (Fig. 2),
all but four had median estimated extinction rates
of <0.4 E/MSY and only two (one arthropod and one
mollusk) had rates >1, and those were <1.5. These
estimates match those of Weir and Schluter (2007), who
estimated bird and mammal extinctions at 0.08 E/MSY
at the equator—where there are the most species—and
at 0.4–0.6 E/MSY at 50°latitude—where there are
fewer species. (They based their estimates on the
ages of a large number of sister species divergences
of New World birds and mammals.) Thus, despite the
methodological hurdles and the potentially confounding
whims of taxonomists, there is consistent evidence
that background extinction rates are <1 extinction per
million species-years and likely much less than this.
Diversification Rates
Median diversification rates were from 0.05 to 0.2
new species per species per million years for a dis-
parate group of animals and plants (Table 1). Diversifi-
cation rates >1.0 were exceptional. Valente et al. (2010)
explicitly addressed the issue of how fast taxa can di-
versify. They analyzed the genus Dianthus (carnations,
Caryophyllaceae) and found net diversification rates of
up to 16 new species per species per million years. This
Conservation Biology
Volume 29, No. 2, 2015
460 Background Rate of Extinction
Table 1. Diversification rates for selected taxa derived from different methods of analysis.
Net diversification rate
Study Method an Median 95%
Plants [22]bK-M 204 0.096 0.011–0.271
M-S 204 0.060 0.002–0.189
Birds [21] K-M 45 0.216 0.092–0.543
M-S 45 0.147 0.061–0.393
Mammals (see text)bK-M 106 0.066 0.000–0.225
M-S 121 0.047 0.013–0.161
McPeek Chordatabthis study 45 0.204 0.007–0.787
Plants (see text)bthis study 37 0.088 0.002–0.324
McPeek Arthropodsbthis study 34 0.173 0.003–0.698
McPeek Mammalsbthis study 16 0.066 0.009–0.13
McPeek Molluscabthis study 8 0.135 0.018–1.395
aAbbreviations: K-M, Kendall-Moran estimator; M-S, Magallon-Sanderson estimator (see text). The K-M and M-S estimators are applied to data for
plants from Ferrer and Good (2012b), for birds from Phillimore and Price (2008), and for mammals from this study. The estimation approach
applied to the remaining data was implemented in subT as described in the text for data on 37 plant taxa (selected for this study), arthropods,
mollusks, and chordates (other than birds and mammals) from McPeek (2008) and for a subset of the mammal data.
bCrown ages (see text) were used.
puts them well above 11 other plant groups, the highest
rate of which was for Andean Lupinus (lupins, Fabaceae)
at approximately 2 (Hughes & Eastwood 2006; Koenen
et al. 2013). For birds, the record holders are the South-
east Asian Zosterops (White-eyes, Zosteropidae), at 2.6
new species per species per million years. Others classify
some of these species as subspecies, which would reduce
that rate.
Valente et al. (2010) also discuss the cichlids of east
African lakes and estimate stem diversification of up to six
new species per species per million years for Lake Malawi
cichlids. They show that rates roughly ten times these
for Lake Victoria cichlids are possible if the 500 species
now present are all descended from just one ancestor
after the lake dried out 14,700 years ago. To achieve
such rates, however, one must completely exclude the
possibility of several species surviving that desiccation in
refugia. An additional example is the rapid divergence of
Enallagma damselflies that have added 23 new species
from 7 lineages in the last 250,000 years (Turgeon et al.
2005). In contrast are lineages such as Ginkgo biloba that
appear to have changed little since the Jurassic (Zhou &
Zheng 2003).
Synthesis
We reviewed three lines of evidence toward obtaining
an order of magnitude estimate of the background rate
of extinction. The fossil data set a broad expectation of
how fast such species go extinct. Separating extinction
and speciation rates from phylogenies is methodologi-
cally difficult. We selected phylogenies carefully so as
to find credible estimates and used simulations that sug-
gested that the correct order of magnitude is recovered,
even when the diversification process is complex. In
general, we estimated extinction rates that were typically
much smaller than Pimm et al.’s (1995) benchmark of 1
E/MSY. There are statistical uncertainties, but generally
high background extinction rates would surely be notice-
able either in some phylogenies or within the range of
uncertainties in all of them.
Rates of diversification are less direct, but there are
many compilations of them across different taxa and
ecosystems. Diversification rates were 0.05 to 0.2
new species per species per million years, with excep-
tional rates of >1. The question is what does these tell us
about extinction rates.
The fossil record shows that overall species richness
increases over time (Rosenzweig 1995). Certainly, some
clades are shrinking (examples in Quental and Marshall
[2011]), but fewer are shrinking than increasing. The
direct estimates of extinction rates from phylogenies esti-
mated above are also low. Our simulation models showed
that when averaged across a set of phylogenies, the mod-
els placed these rates within an order of magnitude. In
short, what we saw in observed phylogenies also pre-
cluded high extinction rates.
Simply, there is no widespread evidence for high
recent extinction rates from either the fossil record or
from molecular phylogenies; thus, in general extinction
rates cannot exceed diversification rates. Combining
the evidence from fossils, the separation of speciation
and extinction rates from molecular phylogenies, and
from overall diversification rates, we conclude that the
benchmark estimate of the natural background rate of
1 E/MSY extinctions per million species per years is too
high. A more defensible estimate should be closer to 0.1
E/MSY, whereupon current extinction rates are 1,000
times higher than the natural background and future
rates of extinction are likely to be 10,000 times higher.
Conservation Biology
Volume 29, No. 2, 2015
de Vos et al. 461
Acknowledgments
We thank A. Antonelli, A. D. Barnosky, P. G. Harnik, M.
McPeek, G. J. Russell, M. Ryberg, T. Price, D. Silvestro,
and D. Tittensor for comments and discussion.
Supporting Information
Additional numerical details of data in Figure 2 (Appendix
S1), data sources (Appendix S2), list of selected mammal
studies (Appendix S3), list of plant studies from McPeek
(2008) (Appendix S4), details of procedure used to se-
lect plant phylogenies (Appendix S5), details on use of
program SubT (Appendix S6), model criticisms and how
we addressed them (Appendix S7), criteria used to se-
lect phylogenies (Appendix S8), details of the simula-
tion approach (Appendix S9), absolute extinction rates
estimated from complete trees and from trees with one-
third missing species (Appendix S10), detailed results of
complete phylogenies (Appendix S11), and additional ref-
erences (Appendix S12). The authors are solely respon-
sible for the content and functionality of these materials.
Queries (other than absence of the material) should be
directed to the corresponding author.
Literature Cited
Alfaro, M. E., F. Santini, C. Brock, H. Alamillo, A. Dornburg, D. L.
Rabosky, G. Carnevale, and L. J. Harmon. 2009. Nine exceptional ra-
diations plus high turnover explain species diversity in jawed verte-
brates. Proceedings of the National Academy of Sciences 106:13410-
13414.
Alroy, J. 1996. Constant extinction, constrained diversification, and un-
coordinated stasis in North American mammals. Palaeogeography,
Palaeoclimatology, Palaeoecology 127:285–311.
Alroy, J. 2008. Dynamics of origination and extinction in the marine
fossil record. Proceedings of the National Academy of Sciences
105:11536–11542.
Barnosky, A. D., N. Matzke, S. Tomiya, G. O. U. Wogan, B. Swartz, T. B.
Quental, C. Marshall, J. L. McGuire, E. L. Lindsey, and K. C. Maguire.
2011. Has the earth’s sixth mass extinction already arrived? Nature
471:51–57.
Bokma, F. 2008. Bayesian estimation of speciation and extinction prob-
abilities from (in) complete phylogenies. Evolution 62:2441–2445.
De-Nova, J. A., R. Medina, J. C. Montero, A. Weeks, J. A. Rosell, M. E.
Olson, L. E. Eguiarte, and S. Magall´
on. 2012. Insights into the his-
torical construction of species-rich Mesoamerican seasonally dry
tropical forests: the diversification of Bursera (Burseraceae, Sapin-
dales). New Phytologist 193:276–287.
Etienne, R. S., B. Haegeman, T. Stadler, T. Aze, P. N. Pearson, A. Purvis,
and A. B. Phillimore. 2012. Diversity-dependence brings molecular
phylogenies closer to agreement with the fossil record. Proceedings
of the Royal Society B: Biological Sciences 279:1300–1309.
Etienne, R. S., and J. Rosindell. 2012. Prolonging the past counteracts
the pull of the present: protracted speciation can explain observed
slowdowns in diversification. Systematic Biology 61:204–213.
Ferrer, M. M., and S. V. Good. 2012a. Correction. Annals of Botany
110:1079–1081.
Ferrer, M. M., and S. V. Good. 2012b. Self-sterility in flowering plants:
preventing self-fertilization increases family diversification rates. An-
nals of Botany 110:535–553.
FitzJohn, R. G. 2012. Diversitree: comparative phylogenetic analyses of
diversification in R. Methods in Ecology and Evolution 3:1084–1092.
Flessa, K. W., and D. Jablonski. 1985. Declining Phanerozoic back-
ground extinction rates: Effect of taxonomic structure? Nature
313:216–218.
Foote, M. 2005. Pulsed origination and extinction in the marine realm.
Paleobiology 31:6–20.
Foote, M., and D. M. Raup. 1996. Fossil preservation and the strati-
graphic ranges of taxa. Paleobiology 22:121–140.
Fritz, S. A., O. R. P. Bininda Emonds, and A. Purvis. 2009. Geographical
variation in predictors of mammalian extinction risk: big is bad, but
only in the tropics. Ecology Letters 12:538–549.
Gore, A. 2006. An inconvenient truth: The planetary emergency of
global warming and what we can do about it. Rodale Books, New
York.
Harnik, P. G., H. K. Lotze, S. C. Anderson, Z. V. Finkel, S. Finnegan,
D. R. Lindberg, L. H. Liow, R. Lockwood, C. R. McClain, and J. L.
McGuire. 2012. Extinctions in ancient and modern seas. Trends in
Ecology & Evolution 27:608–617.
Hughes, C., and R. Eastwood. 2006. Island radiation on a continental
scale: exceptional rates of plant diversification after uplift of the
Andes. Proceedings of the National Academy of Sciences
103:10334–10339.
Kendall, D. G. 1948. On the generalized “birth-and-death” process. The
Annals of Mathematical Statistics 19:1–15.
Koenen,E.,J.deVos,G.Atchison,M.Simon,B.Schrire,E.deSouza,L.
de Queiroz, and C. Hughes. 2013. Exploring the tempo of species
diversification in legumes. South African Journal of Botany 89:19-30.
Magallon, S., and M. J. Sanderson. 2001. Absolute diversification rates
in angiosperm clades. Evolution 55:1762–1780.
McPeek, M. A. 2008. The ecological dynamics of clade diversification
and community assembly. The American Naturalist 172:E270–E284.
Moran, P. 1953. The estimation of the parameters of a birth and death
process. Journal of the Royal Statistical Society. Series B (Method-
ological) 15:241–245.
Morlon, H., T. L. Parsons, and J. B. Plotkin. 2011. Reconciling molecu-
lar phylogenies with the fossil record. Proceedings of the National
Academy of Sciences 108:16327–16332.
Morlon, H., M. D. Potts, and J. B. Plotkin. 2010. Inferring the dynam-
ics of diversification: a coalescent approach. PLoS Biology 8(9).
DOI:10.1371/ journal.pbio.1000493.
Myers, N. 1989. Extinction rates past and present. BioScience 39:39–41.
Nee, S. 2006. Birth-death models in macroevolution. Annual Review of
Ecology, Evolution, and Systematics 37:1–17.
Nee, S., E. C. Holmes, R. M. May, P. H. Harvey, S. Nee, E. C. Holmes,
R. M. May, and P. H. Harvey. 1994. Extinction rates can be esti-
mated from molecular phylogenies. Philosophical Transactions of
the Royal Society of London. Series B: Biological Sciences 344:
77–82.
Paradis, E., J. Claude, and K. Strimmer. 2004. APE: analyses of
phylogenetics and evolution in R language. Bioinformatics 20:
289–290.
Phillimore, A. B., and T. D. Price. 2008. Density-dependent cladogenesis
in birds. PLoS biology 6. DOI: 10.1371/journal.pbio.0060071.
Pimm,S.,P.Raven,A.Peterson,C¸. H. S¸ekercio˘
glu, and P. R. Ehrlich.
2006. Human impacts on the rates of recent, present, and future
bird extinctions. Proceedings of the National Academy of Sciences
103:10941–10946.
Conservation Biology
Volume 29, No. 2, 2015
462 Background Rate of Extinction
Pimm, S., G. J. Russell, J. Gittleman, and T. M. Brooks. 1995. The future
of biodiversity. Science 269:347–350.
Pimm, S. L., C. N. Jenkins, R. Abell, T. M. Brooks, J. L. Gittleman, L.
Joppa, P. H. Raven, C. M. Roberts, and J. O. Sexton. 2014. The
biodiversity of species, their rates of extinction, distribution, and
protection. Science 344: 987. DOI: 10.1126/science.1246752
Price, T. D., et al. 2014. Niche filling slows the diversification of Hi-
malayan songbirds. Nature 509:222–225.
Purvis, A. 2008. Phylogenetic approaches to the study of extinction.
Annual Review of Ecology, Evolution, and Systematics 39:301–319.
Pyron, R. A., and F. T. Burbrink. 2013. Phylogenetic estimates of spe-
ciation and extinction rates for testing ecological and evolutionary
hypotheses. Trends in Ecology & Evolution 28:729–736.
Quental, T. B., and C. R. Marshall. 2011. The molecular phylogenetic
signature of clades in decline. PloS one 6. DOI: 10.1371/journal.
pone.0025780
Rabosky, D. L. 2009a. Ecological limits and diversification rate: alterna-
tive paradigms to explain the variation in species richness among
clades and regions. Ecology Letters 12:735–743.
Rabosky, D. L. 2009b. Extinction rates should not be estimated from
molecular phylogenies. Evolution 64:1816–1824.
Rabosky, D. L., and I. J. Lovette. 2008. Density-dependent diversification
in North American wood warblers. Proceedings of the Royal Society
B: Biological Sciences 275:2363–2371.
Rabosky,D.L.,F.Santini,J.Eastman,S.A.Smith,B.Sidlauskas,J.Chang,
and M. E. Alfaro. 2013. Rates of speciation and morphological evo-
lution are correlated across the largest vertebrate radiation. Nature
Communications 4:1958. DOI: 10.1038/ncomms2958.
Rabosky, D. L., G. J. Slater, and M. E. Alfaro. 2012. Clade age and
species richness are decoupled across the Eukaryotic tree of life.
PLoS Biology 10(8). DOI:10.1371/journal.pbio.1001381
Rosenzweig, M. L. 1995. Species diversity in space and time. Cambridge
University Press, Cambridge.
Russell, G. J., T. M. Brooks, M. M. McKinney, and C. G. Anderson.
1998. Present and future taxonomic selectivity in bird and mammal
extinctions. Conservation Biology 12:1365–1376.
Ryberg, M., R. H. Nilsson, and P. B. Matheny. 2011. DivBayes and SubT:
exploring species diversification using Bayesian statistics. Bioinfor-
matics 27:2439–2440.
Stork, N. E. 2010. Re-assessing current extinction rates. Biodiversity and
Conservation 19:357–371.
Turgeon,J.,R.Stoks,R.A.Thum,J.M.Brown,andM.
A. McPeek. 2005. Simultaneous quaternary radiations of
three damselfly clades across the Holarctic. The American
Naturalist 165:E78–E107.
Valente, L. M., V. Savolainen, and P. Vargas. 2010. Unparalleled rates of
species diversification in Europe. Proceedings of the Royal Society
B: Biological Sciences 277:1489–1496.
Weir, J. T., and D. Schluter. 2007. The latitudinal gradient in recent
speciation and extinction rates of birds and mammals. Science
315:1574–1576.
Wilson, D. E., and D. A. M. Reeder. 2005. Mammal species of the world:
a taxonomic and geographic reference. Johns Hopkins University
Press, Baltimore, Maryland.
Zhou, Z., and S. Zheng. 2003. Palaeobiology: the missing link in Ginkgo
evolution. Nature 423:821–822.
Conservation Biology
Volume 29, No. 2, 2015
