Minimum viable population size: A meta-analysis
of 30 years of published estimates
Lochran W. Trailla
, Corey J.A. Bradshawb
, Barry W. Brooka,
*
a
Research Institute for Climate Change and Sustainability, School of Earth and Environmental Sciences, University of Adelaide,
South Australia 5005, Australia
b
School for Environmental Research, Institute of Advanced Studies, Charles Darwin University, Darwin, Northern Territory 0909, Australia
A R T I C L E I N F O
Article history:
Received 5 April 2007
Received in revised form
5 June 2007
Accepted 15 June 2007
Available online 6 August 2007
Keywords:
MVP
Extinction correlates
Population viability
Generalized linear mixed-effects
models
Meta-analysis
A B S T R A C T
We present the first meta-analysis of a key measure in conservation biology: minimum viable
population (MVP) size. Our analysis is based on studies published since the early 1970s,
and covers 141 sources and 212 species (after filtering 529 sources and 2202 species). By
implementing a unique standardization procedure to make reported MVPs comparable,
we were able to derive a cross-species frequency distribution of MVP with a median of
4169 individuals (95% CI = 3577–5129). This standardized database provides a reference
set of MVPs from which conservation practitioners can generalize the range expected for
particular species (or surrogate taxa) of concern when demographic information is lacking.
We provide a synthesis of MVP-related research over the past 30 years, and test for ‘rules of
thumb’ relating MVP to extinction vulnerability using well-known threat correlates such as
body mass and range decline. We find little support for any plausible ecological and life history
predictors of MVP, even though correlates explain >50% of the variation in IUCN threat
status. We conclude that a species’ or population’s MVP is context-specific, and there are no
simple short-cuts to its derivation. However, our findings are consistent with biological theory
and MVPs derived from abundance time series in that the MVP for most species will
exceed a few thousand individuals.
Ó 2007 Elsevier Ltd. All rights reserved.
1. Introduction
Conservation practitioners are challenged to make informed
choices about the allocation of finite resources to mitigate
the current extinction crisis (Ceballos and Ehrlich, 2002; Thomas
et al., 2004), while being cognizant of the complex ecological
(Shaffer, 1985) and socio-political (Woodroffe et al., 2005)
systems in which such decisions are embedded. Accelerating
habitat and species losses have mandated consideration of
this problem in terms of the number of individuals required
for persistence within a specified timeframe (Shaffer, 1981;
Shaffer, 1987) because small and range-restricted populations
are highly vulnerable to extinction (Terborgh and Winter, 1980;
Gilpin and Soule´, 1986; Schoener and Spiller, 1987). The
concept of a ‘minimum viable population’ (MVP; Shaffer,
1981; Lacava and Hughes, 1984) has been used extensively in
species recovery and conservation management programs
(Clark et al., 2002), and is relevant to the IUCN’s Red List
(www.iucnredlist.org) criteria concerning small and rangerestricted
populations. However, the biological and utilitarian
value of MVP to species conservation has remained controversial
(Shaffer, 1987; Caughley, 1994; Reed et al., 1998).
0006-3207/$ - see front matter Ó 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biocon.2007.06.011
* Corresponding author: Tel.: +61 8 8303 3745; fax: +61 8 8303 4364.
E-mail addresses: lochran.traill@adelaide.edu.au (L.W. Traill), corey.bradshaw@cdu.edu.au (C.J.A. Bradshaw), barry.brook@adelaide.
edu.au (B.W. Brook).
B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/biocon
Past reviews of the concept (Samson, 1983; Gilpin and
Soule´, 1986; Ewens, 1990) and its application (Sjogren-Gulve
and Ebenhard, 2000; Bulte, 2001; Stinchcombe et al., 2002)
have been theoretical, qualitative or cursory, with the primary
literature tending to focus on inherent problems of estimation
(Reed et al., 1998; Brook et al., 2000) rather than utility
per se. Despite both debate on the real-world applicability of
the concept (Caughley, 1994; Reed et al., 1998) and its sustained
popularity (Bulte, 2001; Reed et al., 2003; Tear et al.,
2005; Brook et al., 2006), there has been no broad-scale quantitative
assessment of the MVP literature. This is perhaps due
in part to the difficulty of standardization (e.g., definition of
risk and timeframe, alternative model structures) across
studies.
Individual case studies of MVP for any given species cannot
reveal: (a) the form and variance of the cross-species
distribution of MVP, and whether these agree with theoretical
predictions, or match with genetic, demographic or
environmental rules of thumb for MVP; (b) the existence
(or absence) of taxonomic or life history patterns in MVP;
and (c) generalizations useful for conservation management.
Here we provide the first quantitative meta-analysis
of published MVP estimates, to determine the ensemble
properties of MVP and whether useful generalizations
emerge.
2. Methods
2.1. Dataset
We conducted an exhaustive meta-analysis of the MVP-relevant
literature. All MVP data were obtained from published
articles, book chapters and scientific reports. Primary
literature was identified through ISI’s Web of Science
(www.isinet.com) and Elsevier’s Science Direct (www.
sciencedirect.com) databases. The online search engines
Google (www.scholar.google.com) and Yahoo (www.yahoo.
com) were used to identify, and where possible source, scientific
reports and other grey literature. Search terms such
as ‘‘minimum & viable’’ and ‘‘extinction’’ were used, among
others. Monographs and book chapters were sourced
through university library databases. A cross-check of the
reference list of each article permitted further collation,
especially for sources published prior to 1992. Each article
was reviewed for MVP estimates, and where population
viability analysis (PVA) methods were used, populations
were considered ‘viable’ only where P80% of the initial
population survived for P20 years (Shaffer, 1981). If the
initial population was considered unviable but a target
MVP estimate provided, the latter was used. Where MVP
was not specified explicitly, we required at least the risk
of extinction for a defined timeframe and initial population
size to be reported. Data from baseline PVA models were
selected and hypothetical scenarios ignored. MVP estimates
derived through genetic analyses or population censuses
were also included. A database was collated and structured
according to taxonomic group. Attributes such as species
IUCN Red listing (IUCN, 2006) were later assigned, and the
completed database is available online as Supplementary
Material (Table S1).
2.2. Controlling for differences in the modelling technique
used to derive MVP
Data were collated for 287 MVP estimates, initially by collecting
all parameters that some or all of the models used to derive
MVP. These were (1) probability of persistence, (2)
duration of persistence in years, (3) duration of persistence
in generations, (4) model type or method used to derive
MVP estimate, (5) sex ratio at birth, (6) adult sex ratio, (7) form
of density dependence, (8) carrying capacity, (9) Allee effect
(present/absent), (10) inbreeding depression considered, (11)
probability of catastrophe, (12) birth to adult survival, (13)
adult survival, (14) per cent of female population breeding,
(15) fecundity, (16) age at parturition, (17) longevity, (18) density
and (19) dispersal ability. In many cases, data for the
above parameters were omitted or not given by the authors.
Using logic and previous hypotheses based on extinction
theory (Akcakaya, 1998; Brook et al., 2006), we reduced the initial
19 model attributes to six predictors which we hypothesized
would be relatively independent and explain much of
the methodological variation in MVP among studies: (1) Model
used [MOD]: a categorical index of method or model used to
derive MVP. This was restricted to: (a) individual-based simulation,
(b) matrix/cohort-based simulation (including time
series methods), (c) empirical census or (d) genetic analysis;
(2) Persistence probability [PER]: a continuous variable of the
probability of population persistence over a given time period.
If not used, and where the population was stated as viable,
the probability was assumed to be 100%; (3) Duration [DUR]:
a continuous variable being the period of time over which a
population was deemed viable, expressed as a continuous
variable in generations (3–1200). When generation length of
the species was not provided, we assumed it to be equal to
the age at primiparity. Where a MVP was estimated from a
census or genetic analysis, or where the time frame of viability
was not stated explicitly (n = 13), viability was assumed to
be 100 years and the number of generations estimated on this
basis; (4) Density dependence [DEN]: a categorical factor classified
as: (a) density-independent, (b) ceiling-type density
dependence or (c) functional-type density dependence. The
differentiation between categories (b) and (c) was necessary
to account for their opposite effect on MVP – ceiling density
dependence increases extinction risk, whereas non-Allee
functional density dependence (negative feedback) decreases
extinction risk, relative to density-independent models (Ginzburg
et al., 1990); (5) Inbreeding depression [INB]: a categorical
factor indicating whether the loss of genetic variation in the
population was modelled or not. This was most commonly,
although not universally, applied as 3.14 diploid lethal equivalents
on juvenile survival; and (6) Catastrophes [CAT]: a categorical
factor indicating whether random catastrophe
outside the normal distribution of environmental stochasticity
was included or not.
2.3. Ecological extinction predictors
Following previous work (Brook et al., 2006), we reduced a set
of postulated ecological, life history and anthropological
extinction correlates to a set of eight composite predictors.
Where these correlates were not given in the sourced litera-
160 B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6
Table 1 – Summary of generalized linear and generalized linear mixed-effect model (GLM and GLMM, respectively)
comparisons using Akaike’s Information Criterion corrected for small sample sizes (AICc) and Bayesian Information
Criterion (BIC): (a) GLMMs of the MVP-generating model correlates used for standardization (PER = persistence probability,
MOD = model type, DUR = duration in generations, DEN = form of density dependence, INB = inbreeding included,
CAT = catastrophes included) against the original MVP estimates; (b) GLMMs of the standardized MVP (MVPst) against
ecological and life history correlates (BWT = body weight, GNL = generation length, FEC = fecundity SOC = social grouping,
HMP = human impact, DSP = dispersal, RAN = range, TRE = population trend); (c) GLM of the ecological and life history
correlates against MVPst; and (d) binomial GLMM relating species IUCN Red-Listing (listed or not listed) to ecological and
life history correlates
Candidate models LL k DAICc wAICc DBIC wBIC %DE
(a) MVP-generating model correlates
MVP $ PER + DUR + INB + CAT À425.935 7 0.056 0.439 0.000 0.800 6.3
MVP $ DEN + PER + DUR À427.324 7 2.834 0.109 2.778 0.199 6.0
MVP $ PER + DUR À438.280 5 20.489 <0.001 13.978 <0.001 3.6
MVP $ MOD + PER + DUR + DEN + INB + CAT À420.397 12 0.000 0.451 15.708 <0.001 7.5
MVP $ MOD + PER + DUR À435.307 8 20.961 <0.001 24.102 <0.001 4.2
MVP $ null À454.568 3 48.889 <0.001 35.841 <0.001 0.0
(b) GLMM of ecological and life history correlates
MVPst $ null À1.151 3 0.000 0.364 0.000 0.757 0.0
MVPst $ BWT À1.116 4 2.025 0.132 5.096 0.059 3.0
MVPst $ TRE À1.127 4 2.046 0.131 5.114 0.059 2.1
MVPst $ HMP À1.146 4 2.084 0.128 5.151 0.058 0.5
MVPst $ GNL À1.147 4 2.087 0.128 5.155 0.058 0.4
MVPst $ BWT + GNL À1.112 5 4.136 0.046 10.247 0.005 3.4
MVPst $ HMP + TRE À1.127 5 4.167 0.045 10.272 0.004 2.1
MVPst $ BWT + GNL + FEC À0.973 6 6.003 0.018 15.142 0.000 15.5
MVPst $ BWT + GNL + DSP + RAN À1.052 7 8.333 0.006 20.451 0.000 8.6
MVPst $ BWT + GNL + SOC À1.016 8 10.456 0.002 25.555 0.000 11.8
MVPst $ BWT + GNL + FEC + SOC + DSP + RAN + HMP + TRE À0.914 13 21.646 0.000 51.184 0.000 20.6
(c) GLM of ecological and life history correlates
MVPst $ BWT + GNL + FEC 196.466 5 0.000 0.746 0.000 1.000 15.5
MVPst $ BWT 181.859 3 25.039 0.000 18.502 0.000 3.0
MVPst $ null 178.585 2 29.528 0.000 19.692 0.000 0.0
MVPst $ BWT + GNL + SOC 191.868 7 13.454 0.001 19.909 0.000 11.8
MVPst $ TRE 180.875 3 27.006 0.000 20.469 0.000 2.1
MVPst $ BWT + GNL + DSP + RAN 188.197 6 18.658 0.000 21.896 0.000 8.7
MVPst $ BWT + GNL 182.293 4 26.248 0.000 22.989 0.000 3.4
MVPst $ HMP 179.103 3 30.551 0.000 24.014 0.000 0.5
MVPst $ GNL 178.968 3 30.821 0.000 24.283 0.000 0.4
MVPst $ BWT + GNL + FEC + SOC + DSP + RAN + HMP + TRE 203.025 12 2.159 0.253 24.379 0.000 20.6
MVPst $ HMP + TRE 180.878 4 29.079 0.000 25.820 0.000 2.1
(d) GLMM of ecological and life history correlates
IUCN $ BWT + GNL + FEC + SOC + DSP + RAN + HMP + TRE À60.328 13 74.329 0.000 0.000 0.867 54.0
IUCN $ HMP + TRE À74.019 5 0.000 0.556 4.571 0.088 43.6
IUCN $ HMP À76.136 4 0.452 0.444 5.912 0.045 42.0
IUCN $ BWT + GNL + DSP + RAN + TRE À84.747 8 32.400 0.000 36.731 0.000 35.4
IUCN $ BWT + GNL + DSP + RAN À89.857 7 37.429 0.000 44.222 0.000 31.5
IUCN $ BWT + GNL + FEC + TRE À100.322 7 56.577 0.000 66.842 0.000 23.6
IUCN $ GNL + TRE À107.006 5 63.603 0.000 73.362 0.000 18.5
IUCN $ BWT + GNL + TRE À105.896 6 64.494 0.000 74.464 0.000 19.3
IUCN $ TRE À112.062 4 70.888 0.000 80.100 0.000 14.6
IUCN $ BWT + GNL + SOC + TRE À104.677 9 73.018 0.000 82.254 0.000 20.2
IUCN $ BWT + TRE À111.778 5 73.039 0.000 83.116 0.000 14.8
IUCN $ BWT + GNL + FEC À116.044 6 84.142 0.000 95.822 0.000 11.6
IUCN $ BWT + GNL À119.315 5 87.930 0.000 98.583 0.000 9.1
IUCN $ GNL À121.283 4 89.159 0.000 99.019 0.000 7.6
IUCN $ BWT + GNL + SOC À115.406 8 89.623 0.000 101.176 0.000 12.1
IUCN $ null À131.254 3 106.551 0.000 115.564 0.000 0.0
IUCN $ BWT À130.474 4 107.425 0.000 117.768 0.000 0.6
All GLMMs include the taxonomic Class (e.g., Mammalia, Aves, etc.) as a random effect. Shown are model log-likelihood (LL), number of
parameters (k) change in AICc (DAICc), AICc weight (wAICc), change in BIC (DBIC), BIC weight (wBIC) and the per cent deviance explained (%DE).
%DE is a measure of the structural goodness-of-fit of the model. Models sequences are ordered by wBIC for all model sets, because we were
primarily interested in main rather than tapering effects.
B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6 161
ture, data were derived from online databases or published
papers on that species (see Appendix S1, Supplementary
Material). Predictors used were: (1) Body weight [BWT]: an
allometric scaling covariate (mass in g). Body mass data for
the mostly herbaceous plants were estimated using a benchmark
wet-weight for a similar-sized species, and forestry timber
data used to estimate mass for large Monocotyledons; (2)
Generation length [GNL]: taken as age at sexual maturity and
estimated in months; (3) Fecundity [FEC]: a continuous variable
representing the mean number of young produced per
female per year. This included the average number of eggs
laid/young born, but did not account for the probability of survival
to adulthood (such as in birds and herptiles). Multiple
broods within a year were taken into consideration to calculate
a total yearly output of offspring; (4) Social grouping
[SOC]: a categorical index of mating systems. These were (a)
colonial (i.e., large breeding colonies and spawning sites), (b)
polygamous or gregarious, (c) monogamous and (d) solitary
(i.e., a brief period of copulation only or asexual/hermaphroditic
breeding, and plants); (5) Dispersal [DSP]: the migratory
or dispersive capability of a species, where dispersal and
migration are used interchangeably, and categorized broadly
as (a) migratory or (b) constrained. A species was considered
constrained if it remained within a 20-km radius of its place
of birth/hatching; (6) Range [RAN]: the geographic distribution
scored as either (a) geographic range spanning more than one
major biome (Smith and Smith, 2003), or (b) the species was
primarily restricted to a single biome; (7) Human impact
[HMP]: a categorical index of the (a) beneficial or (b) generally
adverse influence of people. Species considered to benefit
from humans were domesticated animals, harvested crops
and commensals, for example; and (8) Population trend
[TRE]: a categorical index of (a) stable or increasing population
or (b) a population in general decline. TRE was assumed to account
for deterministic population decline.
2.4. Statistical analyses
For all analyses we reduced the population dataset from 287
populations to 212 unique species to avoid potential problems
of pseudo-replication caused by multiple representations (different
populations) of the same species. Two a priori model
sets (Burnham and Anderson, 2002) were constructed to examine
the amount of variation explained in MVP (Table 1): (a) six
models encompassing a selection of the model characteristics
used to derive MVP, and (b) eleven models encompassing a
selection of ecological, life history and anthropogenic threat
terms.
To gauge the relative importance of each derived variable
for predicted MVP, we fitted a series of generalized linear
mixed-effects models (GLMM) to loge MVP in the R Language
(R Development Core Team, 2004), using the lmer function
(in the lme4 library). MVP was assumed on a priori grounds
to be log-normally distributed (Brook et al., 2006). The random
effects error structure of GLMM was used to correct for nonindependence
of species due to potential shared evolutionary
life history traits (Felsenstein, 1985) by decomposing the variance
across species by hierarchical Linnaean taxonomy
(Class) (following Blackburn and Duncan, 2001). Class was selected
as the taxonomic random term in preference to Order
because of sample size limitations: many Orders were represented
by a single species only. The importance of considering
taxonomy in the GLMM was assessed also by repeating
the analyses using a series of generalized linear models
(GLM) with the same ecological and life history correlates.
Asymptotic indices of information loss were used to assign
relative strengths of evidence to the different competing
models (Burnham and Anderson, 2002), with both Akaike’s
Information Criterion corrected for small sample sizes (AICc)
and Bayesian Information Criterion (BIC) weights used as an
objective means of model comparison (Burnham and Anderson,
2002). AICc identifies tapering effects where n, per term,
exceed approximately 20 data, whereas BIC only identifies
main effects (Link and Barker, 2006). Full model results are
shown in Table 1.
Because MVP estimates taken from the literature vary due
to the particular methods employed in each case, it was necessary
to standardize estimates (MVPst) to a consistent model
structure. To do this we used the best-ranked GLMM based on
BIC (Table 1) for the model characteristics set (the model
including persistence probability, duration of persistence,
inbreeding depression and catastrophes, and a phylogenetic
correction), setting persistence probability (PER) to 99%, the
number of generations (GNL) over which MVP was estimated
to 40, and set the b coefficients for the factors to have inbreeding
depression (INB) and catastrophes (CAT) included. The
standardizing equation was therefore:
logeMVPst ¼ logeMVPorig þ bPER Á loge
0:99
PER
 
þ bGNL
Á loge
40
GNL
 
þ bINB þ bCAT
where bPER = 22.5618, bGNL = 0.4365, bINB = 1.2306, bCAT =
0.4258. The distributions of the original versus standardized
MVP estimates are shown in Fig. 1. For each species, the
Fig. 1 – Comparison of original versus standardized
minimum viable population sizes. Relative frequencies of
the 212 MVP species estimates (log10 scale) for the original,
uncorrected values, taken directly from the literature (solid
line, Supplementary Notes) and the same values after
standardization for differing structure of the
MVP-generating method/model (dotted line).
162 B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6
respective coefficients were set to zero when its original MVPgenerating
model matched the defined standardization criterion.
Although the per cent deviance explained in MVP by the
highest BIC-ranked model was only $6%, standardization was
still required to avoid potentially spurious relationships in the
analysis of MVP and ecological correlates.
We tested the ecological predictors by fitting GLMM to loge
MVPst with Class set as a random effect for phylogenetic
control, and then fitted GLM without random effects to examine
the importance of including phylogenetic control in the
models. To provide an independent check of the biological
authenticity of the derived ecological predictors with respect
to a measure of extinction proneness, we constructed analogous
models using the IUCN Red Listing (IUCN, 2006) of species
(17 models). Of the 212 species represented in the
meta-analysis, 92 were Red-Listed (anything other than ‘Least
Concern’).
3. Results
We sourced 529 relevant articles published between January
1974 and December 2005, describing up to 2202 species and
a minimum of 1444. The exact count of distinct species could
not be determined because one large study (Fagan et al., 2001)
did not report which species were examined. Excluding a recent
study on MVP which fitted a set of simple phenomenological
models to 1198 abundance time series (Brook et al.,
2006), 141 articles met the selection criteria and listed 287
MVP estimates for 212 species. A gradual increase in MVP-related
publications over the past 30 years was matched by a
concomitant rise in the number of species studied (Fig. 2).
The establishment of public-access online databases (e.g.,
IUCN Red list and Global Population Dynamics Database
[GPDD], www.cpbnts1.bio.ic.ac.uk/gpdd/) and subsequent
multi-species analyses (Fagan et al., 2001; Reed et al., 2003;
Brook et al., 2006) in recent years were responsible for large
increases in the number of species evaluated (Fig. 2).
A bias toward large-bodied species in extinction-related
research was evident. Ultimately, we found that the frequency
distribution of species studied was skewed towards
heavier species, with 53.8% of all species and 85.3% of mammals
exceeding 1000 g (Fig. 3). By contrast, only 31% of 4049
extant mammals listed in a large database of body masses
(Smith et al., 2003) are >1000 g. Moreover, vertebrates accounted
for 47% of all species studied, despite this taxon representing
only a few percent of named species (IUCN, 2006),
and of the 92 species in the meta-analysis that were IUCN
Red-Listed, 62.0% were mammals. Surprisingly, the Red Listing
of species included in all MVP-related studies showed
an over-representation of non-threatened species (Fig. S1,
Supplementary Material), likely due to larger studies (Brook
et al., 2006) being based on abundance time series collected
for purposes not directly related to conservation, such as
monitoring and harvesting.
The reported MVP values were not comparable in a quantitative
meta-analysis because of differences in the specified
risk definitions and structure of the generating models. We
therefore collated relevant model type and structure data
for each species and fitted a set of GLMM and used AICc,
and BIC to select the most parsimonious model(s) for standardizing
MVP (see Section 2). The most parsimonious model
relating MVP to ‘generating-model structure’ was, according
to AICc, the one that included all model characteristics; however
only 7.5% of the deviance was explained by the saturated
model after controlling for phylogeny (Table 1). An analysis on
a reduced dataset, using Class/Order as a nested random effect,
yielded an equivalent result. It has been shown that with
sufficient sample sizes, the Kullback–Leibler prior used to justify
AICc weighting favours more complex models (Link and
Fig. 2 – Publication trends for minimum viable population
size (MVP), 1974–2005. The cumulative number of species in
studies related to population viability and extinction (log10
scale, solid line), and a 5-year moving-average of the
number MVP-related peer-reviewed and unpublished
literature sources (dotted line). A large increase in species
studied since 2001 marked the advent of freely-accessible
online population databases.
Fig. 3 – Relative frequency distribution of body weight (log10
scale in g). All species (open bars) and mammals (solid bars)
with estimates of minimum viable population size are
shown, with the relative distribution of body weights for all
extant mammals for which data are available (Smith et al.,
2003) (dotted line) for comparison.
B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6 163
Barker, 2006), so we also considered model ranking according
to the dimension-consistent BIC weights to identify the main
drivers of structural variation in MVP (i.e., ignoring tapering
effects). The latter metric signalled that only four of the six
correlates considered (probability of persistence, duration,
inbreeding and catastrophe – see Section 2) explained an
important component (6.3%) of the deviance in MVP (Table
1). Thus, using the best BIC-supported model’s coefficients,
we standardized MVPs (MVPst) to a 99% persistence probability,
and time frame of 40 generations (a previously used time
frame – Brook et al., 2006).
Median MVPst was 4169 individuals (3577–5129, 95% CI),
compared to the median reported uncorrected MVP of 3299
individuals. This is similar to the recommended effective
population size of 4500 individuals based on genetic data
(Frankham, 1995), and the median MVP of 5816 reported for
vertebrates (Reed et al., 2003). The frequency distribution of
the standardized published MVP estimates (Fig. 4) was more
symmetrical and peaked at a higher MVP than the modelaveraged
distribution of MVPs derived from an independent
time series analysis (Brook et al., 2006). This result contradicts
the view that estimates of population viability derived from
scalar models may be overly precautionary (Dunham et al.,
2006), probably because Brook et al. (2006) considered functional
density dependence, whereas Dunham et al. (2006) only
used a population ceiling function.
4. Discussion
Deciding how much habitat is needed to achieve conservation
goals requires robust rules of thumb because in many situations
there are insufficient data to develop a species-specific
population viability analysis (Shaffer et al., 2002). So, can we
provide any generalities from this meta-analysis of MVP?
Models relating ecological attributes predicted a priori to
Fig. 4 – Relative frequency distribution of minimum viable
population (MVP) estimates (log10 scale). Standardized
MVPs from the meta-analysis of 212 species examined
since 1976 (solid line) are compared to MVP estimates
derived independently from models fitted to 1198 species’
time series of abundance data (dotted line) (Brook et al.,
2006). Median values are represented by vertical lines for
each distribution.
Table 2 – Summary of median (and bootstrapped 95% confidence bounds) minimum viable population sizes from all
available literature (n = number of species; standardized = MVPst; original = MVPorig)
n MVPst MVPst 95% CI MVPorig
Vertebrates
Birds 48 3742 2544–5244 3310
Fish 8 1,239,727 211,171–2,085,032 500,000
Mammals 95 3876 2261–5095 2901
Herptilesa
31 5409 3611–6779 3999
Sum/median 182 4102 3325–5096 3697
Other taxa
Plantsb
22 4824 2512–15,992 2097
Insects 5 10,841 1650–103,625 2000
Marine invertebratesc
3 3611 1984–1,047,547 2500
Sum/median 30 6111 3165–10,768 2100
Body mass
<1 kg 98 5137 3577–6947 2509
P1 kg 114 3956 2575–4961 3697
IUCN
Listed 92 3611 2261–5033 2484
Not listed 120 4824 3867–5878 3435
All species 212 4169 3577–5129 3299
a Reptiles and amphibians.
b Mosses, ferns, dicotyledons, monocotyledons and gymnosperms.
c Molluscs and crustaceans.
164 B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6
correlate with extinction risk failed to explain much of the
variation in MVPst; the saturated correlates model accounted
for 20.6% of the explained deviance after taking phylogeny
into account as a random effect (Table 1). The most parsimonious
GLMM, according to BIC, failed to find evidence for any
main effects. Yet these predictors explained 54% of the deviance
in whether or not a species was IUCN Red-Listed. This
contrast between the predictability of MVP versus IUCN status
has been described in previous work (Brook et al., 2006),
using MVP estimated from an independent source (time series
data), and effectively highlights two different paradigms
(Caughley, 1994). Ecological predictors of threatened status
indicate a species’ sensitivity to the largely systematic drivers
of extinction (Cardillo, 2003), confirmed here by the support
for IUCN listing. MVP represents, on the other hand, the small
population paradigm (Caughley, 1994); that is, a population already
reduced in size and subject to a host of population-specific
threats (many stochastic) which cannot be accounted for
in broad species comparisons such as ours.
MVP is thus an appropriate measure of the viability of populations
that have declined deterministically (or catastrophically)
to a small size, but subsequently ‘stabilized’ (though
they continue to fluctuate stochastically). As such, contextspecific
factors such as variability of the local environment
are more relevant for determining MVP than the broad-scale
extinction drivers that cause endangerment. MVP size and regional
or global extinction risk are thus unrelated (Brook
et al., 2006). Note that the majority of vertebrates considered
threatened by IUCN are listed under Criterion A, which relates
threat to rate and magnitude of population size or range decline
(IUCN, 2006). The assessment of vulnerability of IUCN
is complementary, but essentially unrelated, to that derived
from MVP.
Despite the lack of predictability of MVP based on plausible
(and measurable) correlates of extinction risk, we can draw
some broad generalizations from the meta-analysis. MVPrelated
studies have gradually increased over the past three
decades, with no apparent decline in the concept’s use, and
with a trend toward multi-species analyses (Fig. 2). Depending
on the strength of density dependence, MVP follows either a
weakly right-skewed or symmetrical distribution (Fig. 4), with
the highest probability density in the range of a few thousand,
rather than hundreds, or tens of thousands of individuals,
comparable to the findings of Brook et al. (2006) and Reed
et al. (2003). While there was some broad taxonomic variation,
the true magnitude of any differences is uncertain because
some taxa have been poorly sampled to date (fish and
invertebrates – Table 2).
A major product of this collation and standardization of
published MVPs, especially when coupled with a previous
phenomenological analysis (Brook et al., 2006), is a database
of MVPs and species attributes that span a broad range of biomes,
body sizes, life histories and threat status. This resource
(Table S1, provided as a searchable spreadsheet table in the
Supplementary Material) can be used by conservation practitioners
as a preliminary guide to the MVP range expected for
particular species or surrogate taxa of concern, or indeed to
derive a target MVP for data-deficient species (we recommend
the upper 95% confidence limit of MVP for the taxon in question,
excluding poorly sampled taxa such as insects, fish and
marine invertebrates). Moreover, these results provide important
baseline data for testing future research hypotheses
regarding population size and extinction risk, particularly
with the now-evident shift toward the Bayesian paradigm
within ecology and the concomitant need for robust informative
prior information (Clark and Gelfand, 2006). We also support
a disciplinary shift away from charismatic species (as
highlighted by the lack of data available for fishes, insects
and marine invertebrates) and focus of expertise and resources
on IUCN-listed species and hotspots of latent extinction
risk (Cardillo et al., 2006).
Acknowledgements
We thank Julian O’Grady for his assistance with some of the
correlates data, and Marcel Cardillo and an anonmymous referee
for helpful review comments. Funding came from the
Australian Research Council (DP02851634). B.W.B. conceived
the work, L.W.T. compiled the data, and all authors contributed
to the data analysis and writing of the paper.
Appendix A. Supplementary data
Supplementary data associated with this article can be found,
in the online version, at doi:10.1016/j.biocon.2007.06.011.
R E F E R E N C E S
Akcakaya, H.R., 1998. RAMAS Metapop: viability analysis for
stage-structured metapopulations (version 3.0). Applied
Biomathematics, Setauket, NY.
Blackburn, T.M., Duncan, R.P., 2001. Establishment patterns of
exotic birds are constrained by non-random patterns in
introduction. Journal of Biogeography 28, 927–939.
Brook, B.W., O’Grady, J.J., Chapman, A.P., Burgman, M.A.,
Akc¸akaya, H.R., Frankham, R., 2000. Predictive accuracy of
population viability analysis in conservation biology. Nature
404, 385–387.
Brook, B.W., Traill, L.W., Bradshaw, C.J.A., 2006. Minimum viable
population size and global extinction risk are unrelated.
Ecology Letters 9, 375–382.
Bulte, E.H., 2001. Minimum viable populations and sluggish
management. Environmental Conservation 28, 191–193.
Burnham, K.P., Anderson, D.R., 2002. Model Selection and
Multimodel Inference: A Practical Information Theoretic
Approach, second ed. Springer-Verlag, New York.
Cardillo, M., 2003. Biological determinants of extinction risk: why
are smaller species less vulnerable? Animal Conservation 6,
63–69.
Cardillo, M., Mace, G.M., Gittleman, J.L., Purvis, A., 2006. Latent
extinction risk and the future battlegrounds of mammal
conservation. Proceedings of the National Academy of
Sciences of the United States of America 103, 4157–4161.
Caughley, G., 1994. Directions in Conservation Biology. Journal of
Animal Ecology 63, 215–244.
Ceballos, G., Ehrlich, P.R., 2002. Mammal population losses and
the extinction crisis. Science 296, 904–907.
Clark, J.S., Gelfand, A.E., 2006. A future for models and data in
environmental science. Trends in Ecology & Evolution 21,
375–380.
Clark, J.A., Hoekstra, J.M., Boersma, P.D., Kareiva, P., 2002.
Improving U.S. Endangered Species Act recovery plans: key
findings and recommendations of the SCB recovery plan
project. Conservation Biology 16, 1510–1519.
B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6 165
Dunham, A.E., Akc¸akaya, H.R., Bridges, T.S., 2006. Using scalar
models for precautionary assessments of threatened species.
Conservation Biology 20, 1499–1506.
Ewens, W.J., 1990. The minimum viable population size as a
genetic and a demographic concept. In: Adams, J., Lam, D.A.,
Hermalin, A.I., Smouse, P.E. (Eds.), Convergent Issues in
Genetics and Demography. Oxford University Press, New York,
pp. 307–316.
Fagan, W.F., Meir, E., Prendergast, J., Ayoola, F., Kareiva, P., 2001.
Characterizing population vulnerability for 758 species.
Ecology Letters 4, 132–138.
Felsenstein, J., 1985. Phylogenies and the comparative method.
American Naturalist 70, 115.
Frankham, R., 1995. Effective population-size adult-population
size ratios in wildlife – a review. Genetical Research 66, 95–107.
Gilpin, M.E., Soule´, M.E., 1986. Minimum viable populations:
processes of species extinction. In: Soule, M.E. (Ed.),
Conservation Biology: The Science of Scarcity and Diversity.
Sinauer, Sunderland, MA, pp. 19–34.
Ginzburg, L.R., Ferson, S., Akc¸akaya, H.R., 1990. Reconstructibility
of density dependence and the conservative assessment of
extinction risks. Conservation Biology 4, 63–70.
IUCN, 2006. IUCN Red List of Threatened Species. World
Conservation Union. www.redlist.org.
Lacava, J., Hughes, J., 1984. Determining minimum viable
population levels. Wildlife Society Bulletin 12, 370–376.
Link, W.A., Barker, R.J., 2006. Model weights and the foundations
of multimodel inference. Ecology 87, 2626–2635.
R Development Core Team, 2004. R: A language and environment
for statistical computing. URL http://www.R-project.org.
Reed, J.M., Murphy, D.D., Brussard, P.F., 1998. Efficacy of population
viability analysis. Wildlife Society Bulletin 26, 244–251.
Reed, D.H., O’Grady, J.J., Brook, B.W., Ballou, J.D., Frankham, R.,
2003. Estimates of minimum viable population sizes for
vertebrates and factors influencing those estimates. Biological
Conservation 113, 23–34.
Samson, F.B., 1983. Minimum viable populations – a review.
Nature Areas Journal 3, 15–23.
Schoener, T.W., Spiller, D.A., 1987. High population persistence in
a system with high turnover. Nature 330, 474–477.
Shaffer, M.L., 1981. Minimum population sizes for species
conservation. BioScience 31, 131–134.
Shaffer, W.M., 1985. Order and chaos in ecological systems.
Ecology 66, 93–106.
Shaffer, M., 1987. Minimum viable populations: coping with
uncertainty. In: Soule, M.E. (Ed.), Viable Populations for
Conservation. Cambridge University Press, Cambridge,
pp. 69–86.
Shaffer, M., Watchman, L.H., Snape III, W.J., Latchis, I.K., 2002.
Population viability analysis and conservation policy. In:
Beissinger, S.R., McCullough, D.R. (Eds.), Population Viability
Analysis. University of Chicago Press, Chicago, pp. 123–142.
Sjogren-Gulve, P., Ebenhard, T., 2000. The Use of PVA in
Conservation Planning.
Smith, R., Smith, T., 2003. Elements of Ecology. Benjamin
Cummings, San Francisco, CA.
Smith, F.A., Lyons, S.K., Ernest, S.K.M., Jones, K.E., Kaufman, D.M.,
Dayan, T., Marquet, P.A., Brown, J.H., Haskell, J.P., 2003. Body
mass of late quaternary mammals. Ecology 84, 3403.
Stinchcombe, J., Moyle, L.C., Hudgens, B.R., Bloch, P.L.,
Chinnadurai, S., Morris, W.F., 2002. The influence of the
academic conservation biology literature on endangered
species recovery planning. Conservation Ecology 6,
15–26.
Tear, T.H., Kareiva, P., Angermeier, P.L., Comer, P., Czech, B., Kautz,
R., Landon, L., Mehlman, D., Murphy, K., Ruckelshaus, M.,
Scott, J.M., Wilhere, G., 2005. How much is enough? The
recurrent problem of setting measurable objectives in
conservation. Bioscience 55, 835–849.
Terborgh, J., Winter, B., 1980. Some causes of extinction. In: Soule´,
M.E., Wilcox, B.A. (Eds.), Conservation Biology: An
Ecological-Evolutionary Perspective. Sinauer, Sunderland, MA,
pp. 119–134.
Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont,
L.J., Collingham, Y.C., Erasmus, B.F.N., de Siqueira, M.F.,
Grainger, A., Hannah, L., Hughes, L., Huntley, B., van Jaarsveld,
A.S., Midgley, G.F., Miles, L., Ortega-Huerta, M.A., Peterson,
A.T., Phillips, O.L., Williams, S.E., 2004. Extinction risk from
climate change. Nature 427, 145–148.
Woodroffe, R., Thirgood, S., Rabinowitz, A., 2005. People and
Widlife. Conflict or Co-existence? Cambridge University Press,
Cambridge, UK.
166 B I O L O G I C A L C O N S E R V A T I O N 1 3 9 ( 2 0 0 7 ) 1 5 9 –1 6 6