INTRODUCTION
Populations of large carnivores are declining around the globe,
often with dramatic effects on lower trophic levels (Estes et al.
2011). These species typically range over such wide areas that it
can be difficult to maintain viable populations without some indi-
viduals coming into close proximity to humans, posing serious
threats to human safety and domestic livestock. Conservationists
have therefore sought methods to promote human –carnivore co-
existence outside the confines of national parks and wilderness
areas (Woodroffe et al. 2005; Dickman et al. 2011). Given the
potential conflicts with humans, however, separation of large carni-
vores from human communities may ultimately be preferable to a
landscape-level conservation approach as has been demonstrated
for forestry (Boscolo & Vincent 2003) and agriculture (Phalan et al.
2011).
Few species encapsulate these problems more dramatically than
the African lion. Lion densities are directly dependent on prey bio-
mass (Van Orsdol et al. 1985; Hayward et al. 2007), and annual
range requirements for a single lion pride can exceed 1000 km
2
(Funston 2011). Habitat loss in the past 100 years has reduced the
lion’s range by 75% (Riggio et al. 2012), and human–lion conflicts
have intensified because lions kill livestock (Woodroffe & Frank
2005; Kissui 2008) and people (Packer et al. 2005a, 2011a). In addi-
tion, poorly regulated sport hunting has resulted in over-harvesting
in several countries (Packer et al. 2009, 2011b), the effects of which
can extend into unhunted National Parks (Loveridge et al. 2007;
Caro 2008; Kiffner et al. 2009). Finally, numerous lion populations
are genetically isolated (Slotow & Hunter 2009), and inbreeding has
caused measureable reductions in reproductive rates and disease
resistance in several small populations (Kissui & Packer 2004; Trin-
kel et al. 2008, 2011; also see Johnson et al. 2010).
Yet, not all lion populations have declined. The Serengeti lions, for
example, have steadily increased over the past half-century (Packer
et al. 2005b), populations have remained stable in several large South
African national parks (Ferreira & Funston 2010; Funston 2011), and
numerous private reserves in South Africa and Zimbabwe have
successfully restored lions to areas where they had previously been
extirpated (Hunter et al. 2007; Lindsey et al. 2009a,b; Slotow &
Hunter 2009). However, lions are considered so dangerous in South
Africa that they can only be re-introduced after management authori-
ties erect lion-proof fencing and agree to recapture or destroy any
escaping lions (Hunter et al. 2007; Slotow & Hunter 2009).
Wildlife-proof fences effectively prevent most potential conflicts
between lions and humans in southern Africa (Ferguson & Hanks
2010), yet this strategy runs counter to a long-standing conservation
ethic of keeping protected areas unfenced and contrasts with the
wildlife policies of many range states (Hayward & Kerley 2009;
Licht et al. 2010; Slotow 2012). Depending on the size of the
enclosed population, fencing often also necessitates routine genetic
and demographic management of smaller populations via transloca-
tions of breeding-aged individuals (Trinkel et al. 2008; Johnson et al.
2010). Thus, many conservationists have instead sought to encour-
age human–wildlife co-existence through conflict-mitigation pro-
grammes, compensation schemes, insurance plans or payments for
tolerance (e.g. Dickman et al. 2011). However, the costs of manag-
ing dangerous wildlife are formidable. For example, effective ele-
phant and tiger conservation has been estimated to cost $365–930
per km
2
per year (Leader-Williams & Albon 1988; Walston et al.
2010), and the overall costs of anti-poaching and compensation will
only increase in range states with growing human populations
(Wittemyer et al. 2008; Pfeifer et al. 2012), declining purchasing
power of external funds (Garnett et al. 2011) or worsening
corruption (Garnett et al. 2011).
African lions are among the most extensively studied carnivores
in the world with population data available from a wide variety of
protected areas in nearly a dozen different countries with divergent
conservation practices. Several recently developed ecological models
can accurately estimate lion carrying capacities across a wide range
of ecological conditions (Hayward et al. 2007; Loveridge & Canney
2009), making it possible to estimate the effectiveness of lion con-
servation in a given reserve by measuring how closely the observed
population density matches the expected density. The large number
of long-term studies also provides measures of population trends
across a wide variety of circumstances. Here, we explicitly test the
effectiveness of fencing and management budgets on lion popula-
tion size and growth rates, while including the impacts of human
population density, governance, sport hunting, private management
and protected area size.
MATERIALS AND METHODS
Data come from repeated surveys in 38 sites (median
span = 12 years; range: 3–46 years) and single surveys in an addi-
tional four sites. Population growth rates were estimated from the
exponents of exponential regressions of population size over the
most recent 10 years for each time series, using nonlinear models in
Program R (R Development Core Team 2011), function nls. Because
many long-term study sites were surveyed irregularly, data were
sometimes only available up to 1995–2004, and the median time span
was 9 years (range: 3–14 years) (Table S1); Figure S1 shows time ser-
ies as densities (lions/100 km
2
) except for Mole Park, Ghana, where
data were collected as number of ‘contacts per 100 ranger patrols’.
In an analysis of historical data from 49 undisturbed sites, Love-
ridge & Canney (2009) found a tight correlation (r
2
= 0.9271)
between contemporaneous population sizes of lions and large- to
33
Silole Sanctuary, PO Box 938, Karen, Kenya
34
Bubye Valley Conservancy, Zimbabwe
35
Mammal Research Institute, Department of Zoology & Entomology,
University of Pretoria, Pretoria, South Africa
36
Botswana Predator Conservation Trust, Maun, Botswana
37
69 Comeragh Rd, London, W14 9HT, UK
38
Bioinformatics Unit, University of Hohenheim, Stuttgart, 70599, Germany
39
Wildlife Conservation Society, PO Box 7487, Kampala, Uganda
40
Department of Zoology, Field Museum of Natural History, Chicago, IL,
60605, USA
41
Graduate Group in Ecology, University of California, Davis, CA, 95616,
42
Selous Lion Project, PO Box 34514, Dar es Salaam, Tanzania
43
Laboratory of Applied Ecology, University of Abomey-Calavi, Cotonou,
Benin
44
Ongava Research Centre, PO Box 58, Okaukeujo, Namibia
45
Centre for Wildlife Management, University of Pretoria, Pretoria, South
Africa
46
Department of Applied Economics, University of Minnesota, St. Paul, MN,
55408, USA
*Correspondence: E-mail: packer@umn.edu
© 2013 Blackwell Publishing Ltd/CNRS
2 C. Packer et al. Letter
medium-sized ungulates; the resultant equation between lion and prey
biomass was Y = 0.0109x
0.8782
. Where ungulate surveys were not
available, Loveridge & Canney found a close fit for ungulate biomass
by modeling habitats according to NOAA’s Africa Data Dissemina-
tion Service Rainfall Estimate (ADDS-RFE) and cation exchange
capacities taken from the ISRIC-WISE soil profile data set (www.is-
ric.org/data/isric-wise-international-soil-profile-dataset) separated into
high-, medium- and low-nutrient levels. In the current analysis,
‘expected’ lion densities were calculated from known prey biomass
where possible (34 sites); otherwise, herbivore densities were pre-
dicted from rainfall and soils (8 sites); the method used for estimating
‘lion carrying capacity’ did not significantly affect any of our results.
Each site is classified as managed by public or private agencies,
subjected to sport hunting, separated from surrounding communi-
ties by wildlife-proof fencing, country/geographical region, and
method of estimating carrying capacity (prey biomass vs. rainfall/
soils); we also tested effects of reserve size. Human population data
were taken from the AfriPop Project (www.afripop.org) (Linard
et al. 2012; measuring human densities within one kilometre of
protected area boundaries extracted from the World Database of
Protected Areas (IUCN & UNEP 2009)(see Pfeifer et al. 2012).
Governance was based on UNDP’s six indicators (Voice/Account-
ability, Political Stability, Government Effectiveness, Regulatory
Quality, Rule of Law and Control of Corruption) (UNDP 2010).
Principal Components Analysis showed that 87% of variation
between indicators was captured by a single component (‘Gover-
nance’) (Table S2). In the statistical analyses, management budgets
are US$ per km
2
per year while controlling for purchasing power
and likely losses to corruption (Garnett et al. 2011). Budgets could
not be partitioned according to anti-poaching, outreach, fence
repairs, road maintenance, etc.
For 14 of 42 sites, wildlife surveys were restricted to the best-pro-
tected portion of each reserve, whereas budgets were only available
for the entire reserve. Expenditures per km
2
were based on two
alternative measures: first, total budget divided by the size of the
overall protected area (a lower bound which assumes that manage-
ment expenditures are spread evenly over the entire reserve);
second, total budget divided by the size of the survey area (an
upper bound which assumes that management expenditures are
spent exclusively within the survey area). These alternative measures
produced virtually identical results; statistical tests are based on the
geometric mean of the two extremes.
Human population densities, protected area sizes, annual manage-
ment budgets and the ratios of current-to-expected population size
were all lognormal, so statistics on the two response variables
(population growth rate and current-to-expected population density)
were run on the log-transformed data. We used an information-
theoretic approach (Burnham & Anderson 2002), with Akaike’s
Information Criterion (AIC) to calculate statistical models, using
simple linear models in Program R, function lm. We determined the
magnitude and direction of the coefficients for each independent
variable using multi-model averaging across all models with ∆AIC
less than 4.0 (Grueber et al. 2011). These outputs were examined to
determine which predictors were statistically significant and to mea-
sure the relative importance of each variable (Tables 1–3). ‘Relative
importance’ refers to the sum of the Akaike weights over all of the
models containing the parameter of interest.
Given the nested nature of the geographical data, we evaluated a
mixed-effects model with nested random intercepts for Region and
Country. Log-likelihood ratio tests provided no support for including
random effects: the fixed-effects model outperformed all random-
effects models (testing Region only, as well as Country nested within
Table 1 Multi-model averages across all reserves for A. ratio of current-to-expected population densities (n = 40) and B. exponential growth rates over the past 10 years
(n = 33). See Table S3 for the full list of models with ∆AIC less than 4.0
Variable Estimate SE Adj. SE z value P-value
Relative
importance
A. Multi-model averages for Current vs. Expected in all reserves:
(Intercept) !0.990 0.177 0.182 5.435 0.000*** 1.00
Fence 0.478 0.112 0.115 4.153 0.000*** 1.00
Management Budget 0.102 0.029 0.030 3.427 0.001*** 1.00
Namibia + South Africa 0.212 0.138 0.142 1.493 0.136 0.50
Human Pop. Density !0.109 0.068 0.071 1.548 0.122 0.46
Governance 0.003 0.040 0.041 0.077 0.939 0.16
Method 0.089 0.121 0.126 0.706 0.480 0.15
Size of PA 0.044 0.073 0.076 0.578 0.563 0.12
Hunted 0.040 0.117 0.121 0.328 0.743 0.08
State run 0.013 0.091 0.094 0.141 0.888 0.07
B. Multi-model averages for exponential growth rates in all reserves:
(Intercept) 0.040 0.070 0.072 0.565 0.572 1.00
Fence 0.094 0.043 0.045 2.098 0.036* 0.78
State Run !0.096 0.044 0.045 2.113 0.035* 0.69
Initial Pop. Size !0.096 0.051 0.053 1.830 0.067 0.52
Namibia + South Africa 0.079 0.055 0.057 1.386 0.166 0.44
Size of PA 0.026 0.026 0.027 0.965 0.335 0.17
Method 0.058 0.061 0.064 0.901 0.368 0.15
Governance 0.006 0.014 0.015 0.385 0.700 0.14
Human Pop. Density 0.006 0.030 0.031 0.198 0.843 0.08
Hunted 0.010 0.048 0.050 0.201 0.841 0.07
Management Budget 0.001 0.012 0.013 0.086 0.932 0.07
*P < 0.05, **P < 0.01, ***P < 0.001
© 2013 Blackwell Publishing Ltd/CNRS
Letter Conserving large carnivores 3
Region). However, South Africa and Namibia deviated most strik-
ingly from other countries and geographical configurations, so we
ran all AIC models using ‘Namibia + South Africa vs. Other’ as a
fixed effect to minimise the number of coefficients. Note that
because many of the fenced reserves were smaller than the overall
average, ‘fenced/non-fenced’ showed a moderate degree of co-line-
arity with protected area size (Spearman rank-order correlation,
r
s
= !0.516); however, protected area size was not strongly corre-
lated with either of the dependent outcome variables in a univariate
analysis, and the effects of fencing remained robust in all AIC mod-
els that included protected area size. Finally, we extrapolated popula-
tion sizes at 5-year intervals for 100 years into the future by
combining current population size with the exponential growth rate
over the past 10 years. Populations were considered likely to persist
if their extrapolated population sizes exceed 10% of their potential
carrying capacities at particular time points in the future.
RESULTS
Table 1 summarises the variables with the strongest effects on lion
population status and population growth rates across Africa. Current
population densities are highest compared to their expected values in
Table 2 Multi-model averages of the fenced reserves for A. ratio of current-to-expected population densities (n = 17) and B. exponential growth rates over the past
10 years (n = 16). See Table S4 for the full list of models with ∆AIC less than 4.0
Variable Estimate SE Adj. SE z value P-value
Relative
importance
A. Multi-model averages for Current vs. Expected in fenced reserves:
(Intercept) 0.297 0.411 0.421 0.706 0.480 1.00
Size of PA !0.169 0.095 0.100 1.691 0.091 0.60
Namibia + South Africa 0.238 0.137 0.148 1.604 0.109 0.45
State Run 0.233 0.133 0.142 1.634 0.102 0.38
Governance !0.036 0.030 0.032 1.132 0.258 0.38
Human Pop. Density !0.008 0.106 0.109 0.073 0.942 0.15
Hunted !0.089 0.314 0.325 0.274 0.784 0.14
Management Budget !0.063 0.073 0.076 0.827 0.408 0.13
Method 0.005 0.145 0.159 0.034 0.973 0.02
B. Multi-model averages for exponential growth rates in fenced reserves:
(Intercept) 0.225 0.081 0.084 2.688 0.007** 1.00
Initial Pop. Size !0.108 0.037 0.040 2.706 0.007** 0.83
State Run !0.091 0.041 0.044 2.063 0.039* 0.37
Size of PA !0.039 0.018 0.020 1.924 0.054 0.37
Human Pop. Density 0.025 0.019 0.022 1.165 0.244 0.08
Management Budget !0.013 0.012 0.014 0.985 0.325 0.06
*P < 0.05, **P < 0.01, ***P < 0.001
Table 3 Multi-model averages of the unfenced reserves for A. ratio of current-to-expected population densities (n = 22) and B. exponential growth rates over the past
10 years (n = 17). See Table S4 for the full list of models with ∆AIC less than 4.0
Variable Estimate SE Adj. SE z value P-value
Relative
importance
A. Multi-model averages for Current vs. Expected in unfenced reserves:
(Intercept) !1.186 0.332 0.344 3.443 0.001*** 1.00
Management Budget 0.159 0.034 0.036 4.365 0.000*** 1.00
Human Pop. Density !0.326 0.127 0.136 2.405 0.016* 0.93
Hunted !0.420 0.282 0.295 1.423 0.155 0.35
Namibia + South Africa 0.517 0.388 0.405 1.278 0.201 0.25
Size of PA 0.149 0.124 0.131 1.141 0.254 0.18
State Run 0.169 0.157 0.167 1.011 0.312 0.14
Method 0.078 0.150 0.161 0.486 0.627 0.06
Governance !0.012 0.044 0.047 0.265 0.791 0.05
B. Multi-model averages for exponential growth rates in unfenced reserves:
(Intercept) !0.046 0.073 0.077 0.592 0.554 1.00
Namibia + South Africa 0.422 0.100 0.109 3.865 0.000*** 1.00
Hunted !0.258 0.085 0.094 2.752 0.006** 1.00
Method 0.113 0.082 0.091 1.239 0.215 0.16
State Run 0.069 0.062 0.069 1.006 0.314 0.11
Initial Pop. Size !0.060 0.061 0.068 0.886 0.376 0.09
Governance !0.015 0.016 0.017 0.836 0.403 0.09
Size of PA 0.026 0.033 0.036 0.717 0.474 0.08
Management Budget 0.004 0.012 0.013 0.313 0.755 0.06
*P < 0.05, **P < 0.01, ***P < 0.001
© 2013 Blackwell Publishing Ltd/CNRS
4 C. Packer et al. Letter
reserves that (1) are fenced and (2) have the highest management
budgets per km
2
(Fig. 1, Tables 1a and S3a). Over the past 10 yrs,
population growth rates have been highest in (1) fenced reserves
(Fig. 2) and (2) privately managed reserves (Tables 1b and S3b).
Because fences have such a profound impact on lion management,
we performed separate analyses for fenced and unfenced reserves.
For fenced reserves, none of the tested variables had a significant
effect on current population status (Tables 2a and S4a), whereas
recent population growth has been highest in populations that had
been farthest below their potential densities 10 years earlier (Fig. 2)
with additional positive effects from private management (Tables 2b
and S4b). For unfenced populations, current status is highest in
reserves with the largest management budgets (Fig. 1) and lowest
when surrounded by high human population densities (Tables 3a
and S5a); growth rates were highest in Namibia + South Africa and
in populations that were not subjected to trophy hunting (Tables 3b
and S5b). Given current population sizes and recent trends, all of
the fenced populations are expected to remain at or above their full
potential for the next 100 years, whereas less than half of the
unfenced reserves are likely to persist above 10% of their carrying
capacities for the next 20–40 years (Fig. 3), including unfenced sites
in Botswana, Kenya, Cameroon, Ghana, Tanzania and Uganda.
DISCUSSI ON
Negative conservation impacts of human land use can often be
minimised by restricting conflicting activities to separate areas rather
than by encouraging their co-existence. For example, concentrating
crop production in areas of intensive agriculture and sparing land as
nature reserves can improve species conservation and crop produc-
tion more effectively than land-sharing strategies that integrate con-
servation and low-intensity agricultural production (Phalan et al.
2011). Establishing separate areas of intensive timber production
while maintaining well-defined forest reserves is also preferable to
low-intensity harvests over a greater proportion of forest (Boscolo
& Vincent 2003). Similarly, physical separation is highly effective
for conserving African lions: all of the fenced lion populations were
close to their estimated carrying capacities (Fig. 1), growth rates of
the fenced populations were density dependent (Fig. 2), and every
fenced population is expected to remain close to its carrying capac-
ity for the next century. Indeed, managers in many of the smaller
fenced reserves currently remove ‘excess’ lions in attempts to stabi-
lise ungulate numbers (see Fig. S1). Fenced lion populations were
less sensitive to human densities in adjacent areas than were
unfenced populations, presumably because fences reduce poaching,
minimise habitat loss, curtail illegal grazing and prevent direct
human–lion conflict (Kiffner et al. 2012). Such density-independent
‘edge effects’ likely prevented recovery of numerous unfenced lion
populations that had fallen substantially below their respective carry-
ing capacities 10 years earlier.
Conservationists have long recognised that large carnivores should
be kept apart from humans. However, fencing has so far only been
widely employed in a few African countries because of aesthetic
objections, financial costs and the impracticality of enclosing large-
scale migratory ungulate populations. Thus, recent conservation
efforts have increasingly promoted human–wildlife co-existence,
either by initiating conflict-mitigation projects in buffer zones or by
providing economic incentives for local people to tolerate the costs
R
2
= 0.59125
–0.3
–0.2
–0.1
0
0.1
0.2
0.3
100%10%1%
Exponential growth rate
Initial population status
Figure 2 Effect of population density on population growth rate over the
following 10 years for fenced (black) and unfenced (red) reserves. ‘Initial
population status’ refers to the observed population density at the start of each
time series compared to the expected density. The black regression line is for
fenced reserves.
0%
25%
50%
75%
100%
2020 2030 2040 2050 2060 2070 2080 2090 2100 2110
Percentage of populations
Year
Figure 3 Percentage of populations expected to persist at densities > 10% of their
potential in the future. Differences between fenced (black squares, n = 16) and
unfenced (red circles, n = 21) reserves each year are all significant by Fisher test.
R
2
= 0.43952
1%
10%
100%
$1 $10 $100 $1000 $10 000
Current/predicted density
Management budget (US$ km
–2
year
–1
)
Figure 1 Percentage ratio of current population density to predicted carrying
capacity of African lions in fenced (black squares) and unfenced (red circles)
reserves according to management budget per square kilometre of lion survey
area. The red regression line is for unfenced reserves; the effect of management
budget in the fenced reserves is not statistically significant.
© 2013 Blackwell Publishing Ltd/CNRS
Letter Conserving large carnivores 5
