Content uploaded by Sander van Doorn
Author content
All content in this area was uploaded by Sander van Doorn on Apr 02, 2014
Content may be subject to copyright.
Evolutionary emergence of responsive
and unresponsive personalities
Max Wolf
†
, G. Sander van Doorn
‡
, and Franz J. Weissing
†§
†Theoretical Biology Group, Centre for Ecological and Evolutionary Studies, University of Groningen, Kerklaan 30, 9751 NN Haren, The Netherlands;
and ‡Santa Fe Institute, 1399 Hyde Park Rd., Santa Fe, NM 87501
Edited by Brian Skyrms, University of California, Irvine, CA, and approved August 29, 2008 (received for review June 16, 2008)
In many animal species, individuals differ consistently in suites of
correlated behaviors, comparable with human personalities. In-
creasing evidence suggests that one of the fundamental factors
structuring personality differences is the responsiveness of indi-
viduals to environmental stimuli. Whereas some individuals tend
to be highly responsive to such stimuli, others are unresponsive
and show routine-like behaviors. Much research has focused on the
proximate causes of these differences but little is known about
their evolutionary origin. Here, we provide an evolutionary expla-
nation. We develop a simple but general evolutionary model that
is based on two key ingredients. First, the benefits of responsive-
ness are frequency-dependent; that is, being responsive is advan-
tageous when rare but disadvantageous when common. This
explains why responsive and unresponsive individuals can coexist
within a population. Second, positive-feedback mechanisms re-
duce the costs of responsiveness; that is, responsiveness is less
costly for individuals that have been responsive before. This
explains why individuals differ consistently in their responsive-
ness, across contexts and over time. As a result, natural selection
gives rise to stable individual differences in responsiveness.
Whereas some individuals respond to environmental stimuli in all
kinds of contexts, others consistently neglect such stimuli. Inter-
estingly, such differences induce correlations among all kinds of
other traits (e.g., boldness and aggressiveness), thus providing an
explanation for environment-specific behavioral syndromes.
architecture of behavior 兩behavioral flexibility 兩behavioral syndromes 兩
individual differences 兩reactivity
Empirical findings in ⬎100 species, ranging from insects to
mammals, suggest that personalities are a widespread phe-
nomenon in the animal kingdom (1–9). Individuals differ pro-
foundly from each other in their behavior, and these differences
are often consistent over time and extend to various contexts. In
birds, fish, and rodents, for example, some individuals are
consistently more aggressive than others, and aggressive indi-
viduals differ from nonaggressive individuals in many other
respects like foraging behavior or the exploration of novel
environments (5). From an adaptive point of view, both the
coexistence of behavioral types and the consistency of individ-
uals are poorly understood (10, 11).
Many researchers believe that a fundamental factor structur-
ing personality differences is the degree to which individual
behavior is guided by environmental stimuli (6–8, 12–21).
Whereas some individuals pay attention to environmental stim-
uli and quickly adapt their behavior to the prevailing conditions,
others show more rigid, routine-like behavior. Such differences
in responsiveness (also termed coping style, reactivity, f lexibility,
plasticity) have been documented in many organisms including
birds [e.g., great tits (12), spice finches (13), and zebra finches
(14)] and mammals [e.g., rats and mice (7), pigs (20), and humans
(15, 16)].
In both mice and rats (21), individuals differ substantially in
their responsiveness to environmental changes in a maze task.
Some individuals quickly form a routine, are not inf luenced by
minor environmental changes, and perform relatively badly
when confronted with a changing maze configuration. Others
omit forming a routine, are strongly influenced by minor
changes, and perform relatively well when confronted with
changing maze configurations. Similarly, some great tits readily
adjust their foraging behavior to a change in the feeding
situation, whereas others stick to formerly successful habits (12).
The finding that humans and other primates differ in their
susceptibility to environmental influences (15, 16) might also be
interpreted along these lines.
These observations raise two important questions. First, why
do responsive and unresponsive individuals coexist within a
population? Should we not expect a single ‘‘optimal’’ phenotype?
And second, why are differences in responsiveness consistent
across contexts and over time? Should we not expect that
individuals adjust their responsiveness to the needs of the
prevailing situation? In this article, we develop a simple but
general evolutionary model to address these questions.
First, we address the coexistence problem. Our crucial insight
is that for many realistic scenarios, the benefits of responsiveness
are negatively frequency-dependent. As a consequence, respon-
siveness spreads when rare but is selected against when common.
This explains coexistence. Second, we address consistency. We
show that stable individual differences in responsiveness arise
whenever the costs of responsiveness are lower for those indi-
viduals that have been responsive before. We argue that many
processes like learning or training give rise to such positive
feedbacks thus explaining consistency. Interestingly, our results
illustrate that individual differences at the level of behavioral
organization (here, the responsiveness to environmental stimuli)
can induce correlative associations among all kinds of otherwise
unrelated traits.
Coexistence of Responsive and Unresponsive Individuals
Basic Scenario. We consider a population of individuals that face
environmental uncertainty. By assessing the prevailing environ-
mental state and adequately responding to it, individuals can
typically increase their payoff. Yet, such a responsive strategy
involves costs (22) such as, for example, the time and energy
costs of sampling the environment, the mortality cost induced by
collecting information, or the costs of building and maintaining
the required sensory machinery.
Fig. 1 shows the structure of a simple model that captures the
key ingredients of this scenario. Individuals have a choice
between the two options L(‘‘left’’) and R(‘‘right’’). The payoffs
from these options depend on the environment, which can be in
either of two states that occur with probability s
i
(i⫽0or1).
Accordingly, we denote the payoffs from choosing Land Ras a
i
Author contributions: M.W., G.S.v.D., and F.J.W. designed research; M.W., G.S.v.D., and
F.J.W. performed research; and M.W. and F.J.W. wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
§To whom correspondence should be addressed. E-mail: f.j.weissing@rug.nl.
This article contains supporting information online at www.pnas.org/cgi/content/full/
0805473105/DCSupplemental.
© 2008 by The National Academy of Sciences of the USA
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0805473105 PNAS
兩
October 14, 2008
兩
vol. 105
兩
no. 41
兩
15825–15830
EVOLUTION
and b
i
, respectively. Before choosing between Land R, individ-
uals choose whether or not to adopt a responsive strategy.
Responsive individuals get to know the current state and can
therefore make their behavior dependent on this information;
that is, choose Lwith probability l
0
or l
1
, depending on the state
of the environment. Yet, responsiveness is costly and reduces the
payoff by C. In contrast, unresponsive individuals cannot dis-
tinguish between the two states and have to use the same
probability l
in both states.
Benefits of Responsiveness. In view of the cost of responsiveness
C, the responsive strategy can only spread if the benefits of
responsiveness exceed these costs. The benefits of responsive-
ness are given by the excess payoff Eof a responsive over an
unresponsive individual. What determines this excess payoff? In
state i, a responsive individual plays strategy l
i
and thus obtains
the payoff l
i
a
i
⫹(1 ⫺l
i
)b
i
. This payoff will typically exceed the
payoff of an unresponsive individual, l
a
i
⫹(1 ⫺l
)b
i
, that has to
use the general-purpose strategy l
. The payoff difference in state
iis therefore (l
i
⫺l
)(a
i
⫺b
i
), and the benefits of responsiveness
are thus given by
E⫽
冘
i
s
i
共l
i
⫺l
兲共a
i
⫺b
i
兲.
Hence, the responsive strategy spreads whenever E⬎C, and the
unresponsive strategy spreads whenever E⬍C.
Frequency Dependence. From now on, we make the crucial as-
sumption that the payoffs a
i
and b
i
are negatively frequency-
dependent, that is, the excess payoff of choosing Lover Rin state
i,a
i
⫺b
i
, decreases with the frequency f
i
of individuals that
choose option Lin state i(f
i
⫽p
r
l
i
⫹(1 ⫺p
r
)l
). As we discuss
below, this is a realistic assumption. In the supporting informa-
tion (SI), we demonstrate that frequency dependence at the level
of the choices between Land Rgives rise to benefits of
responsiveness that are negatively frequency dependent, that is
dE
dp
r
⬍0. [1]
The intuition for this result is as follows. Consider a situation
where in state i, it is advantageous to choose option L(a
i
⬎b
i
).
Hence, responsive individuals choose L(l
i
⫽1), whereas unre-
sponsive individuals have to stick to the general-purpose strategy
l
. However, the payoff difference between Land Rdecreases
with the frequency of individuals that choose option L.Asa
consequence, the benefits of responsiveness in state idecrease
with the frequency of responsive individuals.
Coexistence. Because the benefits of responsiveness E(p
r
) are
negatively frequency-dependent, they will be highest in a pop-
ulation of unresponsive individuals (p
r
⫽0) and lowest in a
population of responsive individuals (p
r
⫽1). We have seen that
responsive individuals can invade a population of unresponsive
individuals whenever E(0) ⬎C, whereas unresponsive individ-
uals can invade a population of responsive individuals whenever
C⬎E(1). Accordingly, both strategies can spread when rare
whenever
E共0兲⬎C⬎E共1兲[2]
leading to the coexistence of responsive and unresponsive indi-
viduals. In the SI, we show that E(0) and E(1) can readily be
calculated. E(0) is given by E(0)⫽s
0
s
1
⌬, where ⌬⫽¥
i
兩a
i
⫺b
i
兩
is the total payoff difference in a population of unresponsive
individuals. E(1) is equal to zero whenever, in a population of
responsive individuals, a mixed evolutionarily stable strategy
(ESS) would be played in any of the environmental states.
Example I: Coexistence in a Patch-Choice Game. We now illustrate
this result and its consequences for a situation where the options
Land Rcorrespond to the alternatives in a patch-choice game,
where each individual has the choice between two patches. The
payoff an individual obtains in any of the two patches is given by
a
i
⫽A
i
/f
i
and b
i
⫽B
i
/(1 ⫺f
i
), where A
i
and B
i
are state-dependent
baseline values of the two patches, and f
i
is the frequency of
individuals that choose patch Ain state i.
Fig. 2Aillustrates that negative frequency dependence on the
level of the patch-choice game gives rise to benefits of respon-
siveness that are negatively frequency-dependent. Responsive
individuals (green line) always obtain a payoff that is as least as
high as that of unresponsive individuals (red line) because they
can choose the better patch in each environment. However, as
predicted by our analysis above (Eq. 1), the payoff difference
between responsive and unresponsive individuals (black line),
that is, the benefit of responsiveness, decreases with the fre-
quency of responsive individuals. Whether decreasing benefits of
responsiveness give rise to the coexistence of responsive and
unresponsive individuals depends on the strength of this de-
crease and on the cost of responsiveness (see Eq. 2). For the
chosen parameter values of A
i
and B
i
, we expect coexistence
whenever the cost of responsiveness Cis between E(1) ⫽0 and
E(0) ⫽0.5 (right axis). For any of these equilibria, one can
readily calculate the corresponding ESS behavior of responsive
and unresponsive individuals in the patch-choice game. This is
illustrated in Fig. 2C, which shows how these strategies change
with the cost of responsiveness.
To test these predictions, we implemented our assumptions in
individual-based computer simulations in which trait frequencies
change over time under the influence of natural selection (see
Appendix, below). The simulation results are in perfect agree-
ment with our analytical predictions. For any value of C⬍0.5,
the population converges to the predicted mixture of responsive
and unresponsive individuals; Fig. 2Bshows two simulations for
the scenario depicted in Fig. 2A(C ⫽0.2), one starting from an
a0b0
LR
l1–l
a1b1
LR
l1–l
a0–C b0–C
LR
l01–l0
a1–C b1–C
LR
l11–l1
state 0 state 1
s0s1
state 0 state 1
s0s1
unresponsive
1– pr
responsive
pr
Fig. 1. Setup of the one-stage model. We consider a scenario where indi-
viduals can find themselves in either of two states, where state ioccurs with
probability si. Individuals have the choice between two options Land R. The
payoffs associated with these options, aiand bi, depend on the state of the
environment i⫽0,1 and, in addition, on the strategy established in the
population. An individual follows the responsive strategy with probability pr.
Responsive individuals can distinguish between the two states and make their
behavior dependent on the current state. Accordingly, the probability liwith
which a responsive individual chooses option Ldepends on the state i.In
contrast, unresponsive individuals cannot distinguish between the two states
and have to use the same probability l
in both states. Although responsiveness
allows more flexible behavior, it is costly and reduces the payoff by C.
15826
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0805473105 Wolf et al.
ancestral population of responsive individuals and the other
from an ancestral population of unresponsive individuals. Fig.
2Dillustrates that also the behavior of responsive and unrespon-
sive individuals in the patch-choice game is in perfect agreement
with our analytical predictions. Unresponsive individuals (red
line) evolve an intermediate tendency to choose between the two
patches (l
*⫽0.58), whereas responsive individuals (green lines)
flexibly employ the two extreme strategies ‘‘always choose patch
A’’ (l
0
ⴱ
⫽1) and ‘‘always choose patch B’’ (l
1
ⴱ
⫽0), depending on
the state of the environment.
Consistent Individual Differences in Responsiveness
Positive Feedbacks and Consistency. Empirical evidence suggests
that individuals that are responsive to environmental stimuli at
one point in time and in one context tend to be responsive at later
points in time and in different contexts as well (6, 7). Why should
natural selection give rise to such consistency? Consider first the
extreme case where being responsive once reduces the cost of
further responsiveness to zero. In this case, it is obvious that
previously responsive individuals should be responsive anew,
because they can reap the benefits of responsive behavior
without incurring additional cost. Hence, responsiveness is con-
sistent within and across contexts. This is an extreme scenario,
because early responsiveness has a very strong feedback on the
cost of later responsiveness. However, we now show that even the
tiniest feedback is sufficient to induce consistent individual
differences in responsiveness.
To investigate the effect of such feedbacks, we now consider
a two-stage scenario. In each of the stages, individuals face the
choice between adopting a responsive or an unresponsive strat-
egy. The two stages might either represent the same context at
different points in time (e.g., patch choice early and late in the
season) or different contexts (e.g., a patch choice and aggressive
encounters). In both stages, individuals face a choice between
two options (say Land Rin stage 1 and say L⬘and R⬘in stage
2), where the payoffs are again negatively frequency-dependent
and depend on the state of the environment. For simplicity, we
assume that the environmental states in both stages are uncor-
related. Individuals that are responsive in any of the two stages
get to know the environmental state in that stage and can fine
tune their behavior accordingly. The fitness of an individual is
given by the sum of payoffs obtained in both stages reduced by
the cost of responsiveness. As above, the cost of responsiveness
in the first stage is given by C. We assume that the cost of
responsiveness in the second stage is smaller for individuals that
were responsive in the first stage (C
r
) than for those individuals
that were unresponsive in the first stage (C
ur
). In the SI, we show
that even the smallest cost reduction gives rise to consistency in
responsiveness: At the ESS, individuals that are responsive in the
first stage have a higher tendency to be responsive in the second
stage (p
r
r
) than individuals that are unresponsive in the first
stage (p
r
ur
), that is
p
r
ⴱ
兩
r
⬎p
r
ⴱ
兩
ur
.[3]
In fact, as we presently show, even a very small feedback gives
rise to strong consistency in responsiveness across stages.
Example II: Consistency and Behavioral Syndromes. We now illus-
trate this result and its consequences for a situation where
individuals have to choose a patch in the first stage (as above)
and are involved in aggressive encounters in the second stage.
C
0.0
0.4
0.6
0.8
1.0
Evolutionarily stable strategies
0.0 0.1 0.2 0.3 0.4
Cost of responsiveness (C)
0.5 0.6
0.2
D
0.0
0.4
0.6
0.8
1.0
Strategies patch choice game
0 5 10 15
Time (1,000 generations)
20
0.2
A
0.0
0.4
1.2
0.8
1.6
Cost and benefit of responsiveness
0.0 0.1 0.2 0.3 0.4
Frequency responsive individuals (p
r
)
0.5 0.6
B
0.0
0.4
0.6
0.8
1.0
Frequency responsive individuals (pr)
0.2
patch choice unresponsive
patch choice responsive state 0
patch choice responsive state 1
f
r
e
q
u
e
n
c
y
r
e
s
p
o
n
s
i
v
e
unresponsive
responsive state 0
responsive state 1
cost of responsiveness
p
a
y
o
f
f
d
i
f
f
e
r
e
n
c
e
p
a
y
o
f
f
u
n
r
e
s
p
o
n
s
i
v
e
p
a
y
o
f
f
r
e
s
p
o
n
s
i
v
e
E(1)
C
E(0)
0 5 10 15
Time (1,000 generations)
20
simulation 2
simulation 1
Fig. 2. Coexistence of responsive and unresponsive individuals due to frequency-dependent selection, illustrated for a situation where the options Land R
correspond to the alternatives in a patch-choice game. (A) Dependence of payoffs on the proportion of responsive individuals in the population. Responsive
individuals always obtain a payoff that is at least as high as the payoff to unresponsive individuals. The benefits of responsiveness (i.e., the excess payoff of
responsive individuals, black line) decreases from a value E(0) ⫽0.5 in a population of unresponsive individuals to E(1) ⫽0 in populations with a high proportion
of responsive individuals. The benefits of responsiveness exactly balance the cost of responsiveness at pr⫽0.32. (B) Two individual-based simulations illustrating
that, independent of the initial conditions, natural selection gives rise to the stable mixture of responsive and unresponsive individuals predicted by A.(C)
Dependence of the evolutionarily stable strategies on the cost of responsiveness. The dashed black line indicates the configuration in Aand B.(D)
Individual-based simulation showing the evolution of behavior in the patch-choice game. At equilibrium, responsive individuals exhibit a state-dependent pure
strategy: ‘‘always choose patch Lstate 0’’ and ‘‘always choose patch Rin state 1.’’ Unresponsive individuals employ a mixed strategy.
Wolf et al. PNAS
兩
October 14, 2008
兩
vol. 105
兩
no. 41
兩
15827
EVOLUTION
Aggressive encounters are modeled as a hawk–dove (23) game
(L⬘⫽‘‘hawk’’ and R⬘⫽‘‘dove’’): Individuals fight for a resource
of value V, and aggressive hawks risk injury, reducing their payoff
by D. Now we assume that the resource value is either V
0
or V
1
,
depending on the state of the environment.
Fig. 3A depicts how the ESS level of responsiveness depends
on the strength of the feedback. For any degree of cost reduction,
first-stage responsiveness is represented by the blue line, and
second-stage responsiveness of previously responsive and unre-
sponsive individuals is depicted by the dashed and solid gray
lines, respectively. Note that for strong feedbacks, all individuals
play a pure strategy in the second stage: Previously responsive
individuals are always responsive, whereas previously unrespon-
sive individuals are never responsive. Remarkably a dichotomy
of similar strength already occurs at very weak feedbacks. In
other words, the smallest cost reduction gives rise to consistent
individual differences. Our individual-based simulations (Fig.
3B) are in perfect agreement with these analytical predictions.
Behavioral Syndromes. As in the one-stage game considered
above, at the ESS, unresponsive individuals play a general-
purpose mixed strategy in both stages, whereas responsive
individuals adapt their behavior to the prevailing conditions and
choose a pure strategy (Fig. 3 Cand D). Notice that, for a given
combination of environmental states, all responsive individuals
play the same combination of pure strategies in both stages. At
the population level, this induces a correlation between the
behavioral choices in stage 1 and stage 2. In other words,
consistent individual differences in responsiveness induce be-
havioral correlations that might be interpreted as behavioral
syndromes (1, 5). Note that this cross-context correlation ref lects
consistency in the behavior of responsive individuals rather than
an intrinsic link between the two contexts. This is also reflected
by the fact that the sign and the strength of these correlations
depend on the environment (Fig. 3E).
Discussion
Frequency Dependence. Our explanation for the coexistence of
responsive and unresponsive individuals is based on the insight
that the benefits of responsiveness are negatively frequency-
dependent. Frequency dependence at the level of responsiveness
is caused by our assumption that the payoffs at the level of the
behavioral choices (e.g., patch choice, aggressive encounters) are
frequency-dependent. This assumption is realistic. For example,
behavior in social interactions (e.g., aggressive or cooperative
behavior) has frequency-dependent payoffs almost by definition,
because the outcome depends on the behavior of all participants
(23–25). Other forms of frequency dependence arise whenever
individuals compete for limited resources as, for example, in a
foraging context. In these situations, individual behavior impacts
on the environment, which, in turn, feeds back on the individuals
(26). Next to such ecological mechanisms, a variety of other
mechanisms can also lead to frequency dependence (27).
Emergence of Polymorphism. In our model, frequency-dependent
selection gives rise to polymorphism. This may reflect our
assumption that individuals face a binary choice between adopt-
ing a responsive or an unresponsive tactic. In some situations, it
is indeed reasonable to view responsiveness as an all-or-nothing
decision; in others, however, responsiveness is better viewed as
st
C
0.0
0.4
0.6
0.8
1.0
Strategies patch choice game
010203040
Time (1,000 generations)
50
0.2
D
0.0
0.4
0.6
0.8
1.0
Strategies hawk-dove game
Time (1,000 generations)
0.2
A
Evolutionarily stable strategies
Strength of feedback ((C
ur
– C
r
) / C
ur
)
B
0.0
0.4
0.6
0.8
1.0
Frequency responsive individuals
0.2
Time (1,000 generations)
01020304050
01020304050
0.0
0.4
0.6
0.8
1.0
0.2
0.0 0.2 0.4 0.6 0.8 1.0
unresponsive
responsive state 0
responsive state 1
E
–0.4
0.0
0.2
0.4
Cross context correlations
Environment patch choice game
–0.2
0
1
0
0
1
1
1
0
Environment hawk-dove game
1
st
stage
2
nd
stage: previously
unresponsive
2
nd
stage: previously
responsive
1
st
stage
2
nd
stage: previously
unresponsive
2
nd
stage: previously
responsive
unresponsive
responsive state 0
responsive state 1
Fig. 3. Evolution of consistent individual differences in responsiveness due to positive feedbacks. (A) Evolutionarily stable responsiveness illustrating that,
independent of the strength of feedback, individuals that are responsive in the first stage (here patch-choice game) show high levels of responsiveness in the
second stage (here a hawk–dove game), whereas previously unresponsive individuals show low levels of responsiveness in the second stage. The dashed black
line indicates the configuration in the individual based simulations B–E.(B) Typical simulation illustrating the evolution of consistent individual differences in
responsiveness. (Cand D) In both the patch choice context (C) and in the hawk–dove context (D), unresponsive individuals evolve a mixed strategy, whereas
responsive individuals evolve the pure strategies that are used dependent on the state of the environment. (E) For each combination of environmental states
in the two stages, a correlation results between the behavioral choices (patch choice and hawk–dove game), induced by the fact that individuals differ
consistently in their responsiveness and that responsive individuals play a pure strategy in either state. The sign and the strength of these correlations depend
on the combination of states in both contexts.
15828
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0805473105 Wolf et al.
a continuous trait. For example, individuals may var y in their
degree of sampling on a scale from superficial to thorough.
Alternatively, individuals may var y their rate of sampling as in
situations where individuals differ in their tendency to interrupt
their ‘‘normal’’ behavior to sample.
When responsiveness varies continuously, negative frequency
dependence may result in either a monomorphism with an
intermediate degree of responsiveness or a polymorphism where
individuals differ in their responsiveness. The evolutionary
outcome will ref lect the shape of the tradeoff individuals face.
Intuitively, polymorphism is favored when the costs and benefits
associated with responsiveness give rise to a convex tradeoff,
whereas monomorphism is favored by concave tradeoffs (28).
Interestingly, the coexistence of responsive and unresponsive
phenotypes has been suggested in other contexts as, for example,
the coexistence of plastic and canalized development al strategies
(29) or the coexistence of generalists and specialists (30).
Positive Feedbacks. Our model explains consistency in respon-
siveness by a positive-feedback mechanism. Previously respon-
sive individuals have a higher tendency to be responsive again
because they face lower costs (or higher benefits) than previously
unresponsive individuals. Remarkably, the smallest such asym-
metry translates into a strong positive association of responsive-
ness across stages.
It is highly plausible that positive feedbacks act within contexts
as, for example, in the case when responsive individuals get
better at being responsive (e.g., assessing cues) with repeated
experience (31). Cross-context feedbacks might seem less likely,
but they can be caused by various mechanisms. We give three
examples. First, the cost of responsiveness may consist of a
context-independent part (e.g., screening the environment) and
a context-specific part (e.g., screening for specific cues). With
respect to a second context, the context-independent part rep-
resents fixed costs that do not have to be paid again. Second,
individuals that are responsive in one context may build up
knowledge and skills that can be used in a different context. If,
for example, individuals get better in interpreting environmental
cues, the costs are lower for experienced than for inexperienced
individuals. Third, information gathered in one context may
prove useful for assessing the state of the environment in a
different context, that is, information acquired in one context
may spill over to a different context.
Responsiveness and Behavioral Flexibility. In the empirical literature,
differences in responsiveness are also referred to as differences in
flexibility, plasticity, and reactivity. These categories are often used
synonymously (e.g., refs. 7 and 8). However, this is not always
adequate. Whereas responsiveness refers to the propensity of an
individual to adjust its behavior to the prevailing environmental
conditions, behavioral f lexibility refers to the tendency of an
individual to show varying behavior when confronted with the same
context repeatedly. One might think that responsive individuals are
flexible (i.e., show varying behavior) and unresponsive individuals
are rigid (i.e., show the same behavior). Our analysis shows,
however, that this relation is more ambiguous.
Consider a situation where individuals are repeatedly con-
fronted with the same context under uncertainty. Both respon-
sive and unresponsive individuals will appear f lexible to an
observer. Responsive individual are f lexible because they play a
state-dependent pure strategy and thus change their behavior
with the environmental state. Yet, unresponsive individuals are
also flexible because they play a mixed strategy and hence
change their behavior due to randomization. There is, however,
a crucial difference between the two strategies: Only responsive
individuals vary their behavior systematically in response to the
environmental conditions.
The relation between responsiveness and flex ibility is not
always as ambiguous. We give two examples. First, consider the
above scenario where individuals are repeatedly confronted with
the same context but now assume that there is a cost associated
with changing behavior (e.g., switching patches might be costly).
Such a cost has a differential effect on responsive and unre-
sponsive individuals. Whereas responsive individuals change
their behavior only when it pays to, unresponsive individuals do
not improve their payoff by changing behavior. Consequently,
whenever there is a cost associated with changing behavior,
unresponsive individuals should rigidly stick to the behavior once
chosen, whereas responsive individuals should keep changing
their behavior f lexibly whenever the environmental state
changes. Note that in such a case, unresponsive individuals still
mix between both alternatives on a population level: Some
consistently choose option L, whereas others choose option R.
Second, consider a situation where individuals, instead of
choosing between two discrete alternative Land R, face a choice
between a continuum of alternatives. For example, instead of
choosing between an aggressive hawk and a nonaggressive dove
strategy, individuals might choose an intensity of aggression that
varies continuously between a minimum level Land a maximum
level R. In this case, the mixed strategy of unresponsive individ-
uals does not correspond to a randomization but to an interme-
diate intensity of aggression. Thus, when confronted with such
a context repeatedly, unresponsive individuals rigidly show the
same intermediate level of aggression, whereas responsive indi-
viduals f lexibly exhibit max imal and minimal levels of aggres-
sion, depending on the state of the environment.
Implications for Understanding Animal Personalities. The defining
feature of animal personalities is that individual behavior is corre-
lated over time and across contexts. Such correlations, or behav ioral
syndromes (1, 5), seem puzzling because a more flexible structure
of behavior should be advantageous. Current explanations fall into
two classes. According to the ‘‘constraints view,’’ trait correlations
result from constraints on the architecture of behavior (5). This
view emphasizes seemingly nonadaptive aspects of behavior and
limited plasticity. However, it remains unclear why the underlying
constraints are not removed by natural selection. Interestingly, our
model exemplifies that a flexible architecture may invade the
constrained one, without necessarily going to fixation.
According to the ‘‘adaptive view,’’ trait correlations are the
result of natural selection. Particular combinations of traits
appear together because they work well together (32–37). For
example, the boldness–aggressiveness syndrome has been ex-
plained in terms of differences in energy reserves (33), differ-
ences in future fitness expectations (36), and differences in
growth rates (37). Although being in the realm of the adaptive
view, our results provide a different type of explanation. Indi-
vidual differences at the level of the behavioral organization can
give rise to correlative associations of all kinds of otherwise
unrelated behaviors.
Consider the above scenario where evolution gives rise to a
correlation between the patch choice behavior of individuals and
their aggressiveness (Fig. 3E). Suppose, for the sake of the
argument, that the patches differ in their riskiness (e.g., presence
of a predator) such that patch choice might be interpreted as a
choice between being bold and being shy. In this case, the
correlation pattern in Fig. 3Eresembles an environment-specific
boldness–aggressiveness syndrome that has been found in nat-
ural populations of sticklebacks (38, 39). Notice, however, that
this correlation is not caused by an intrinsic link between
boldness and aggressiveness. Rather it is caused by the fact that
the coexisting responsive and unresponsive individuals employ
different decision rules to choose between the behavioral alter-
natives. Whereas responsive individuals use a fine-tuned rule
that conditions the behavior on the prevailing conditions, un-
responsive individuals employ a general-purpose rule that does
Wolf et al. PNAS
兩
October 14, 2008
兩
vol. 105
兩
no. 41
兩
15829
EVOLUTION
not distinguish between these conditions. Trait correlations do
not, therefore, necessarily reflect an inherent connection be-
tween the associated traits but can be a byproduct of stable
individual differences at the level of behavioral organization.
Appendix
Here, we give the setup of the individual-based simulations. The
analytical results are derived in the SI. We simulated a spatially
heterogeneous metapopulation with many ‘‘islands’’ (i.e., local
patches). The islands differ in their environmental conditions for
the choice situations (Lvs. R): For the patch choice (hawk–dove)
context in stage 1 (stage 2) a fraction s
0
(s⬘
0
) of the islands is in
state 0, and the remaining islands are in state 1. For the two-stage
scenario, the environmental conditions for each of the stages are
drawn independently from each other. In all our simulations, we
studied populations of 5,000 individuals that are distributed
among 50 islands.
After birth, individuals disperse to a random island. Individ-
uals are haploid and characterized by a suite of heritable traits
corresponding to (i) the tendency to choose the responsive
strategy in each of the stages and (ii) the tendency to choose
option Lin each of the stages. Each of these traits is encoded by
a separate gene locus; for the two-stage scenario, separate loci
encode the behavior in the first and in the second stage. After
obtaining the payoff (see below), individuals mate at random
within each island, reproduce, and then die. Each island con-
tributes equally to the total offspring generation; within islands,
the relative contribution of an individual to the offspring gen-
eration is proportional to its net payoff. During reproduction,
mutations occur with probability
⫽1⫻10
⫺3
. Whenever a
mutation occurs, it has a small effect on a random locus: It
changes the strategy of an individual by a value that is drawn
from a normal distribution with mean 0 and standard deviation 0.1,
with the constraint that it remains in the inter val from 0 to 1.
For the one-stage scenario (Fig. 2), individuals face a situation
as depicted in Fig. 1. We assume that the options Land R
represent the behavioral alternatives in a patch-choice game. For
an island with environment al state i, the payoff that an individual
obtains in any of the two patches is given by a
i
⫽A
i
/fand b
i
⫽
B
i
/(1 ⫺f), where A
i
and B
i
are state-dependent baseline values,
and fis the frequency of individuals that choose patch Lin this
island. All simulations are based on A
0
⫽0.8, B
0
⫽0.2, A
1
⫽0.3,
B
1
⫽0.7, C⫽0.2, and s
0
⫽0.5.
For the two-stage scenario (Fig. 3), individuals face a situation
as depicted in Fig. 1 twice. We assume that the options represent
the behavioral alternatives in a patch-choice game in the first
stage and the alternatives in a hawk–dove game in the second
stage. In the hawk–dove game, individuals within an island are
paired at random and fight for a resource with a value V
i
that
depends on the island’s state of the environment for the hawk–
dove game. Payoffs are obtained as in the standard hawk–dove
game (23): When two hawks meet, one gets V
i
, and the other gets
⫺D; when two doves meet, both get V
i
/2; when a hawk meets a
dove, the hawk gets V
i
, and the dove gets 0. Fig. 3 is based on
V
0
⫽2, V
1
⫽8, D⫽10 for the hawk–dove game, C⫽0.2,
C
ur
⫽0.2, C
r
⫽0.1 for the cost of responsiveness, and s
0
⫽0.5,
s⬘
0
⫽0.5.
ACKNOWLEDGMENTS. We thank T. W. Fawcett, O. Leimar, D. S. Wilson, and
an anonymous referee for numerous helpful suggestions and D. Visser for
preparing the figures. G.S.v.D. was supported by a Rubicon grant from the
Netherlands Organisation for Scientific Research.
1. Clark AB, Ehlinger TJ (1987) in Perspectives in Ethology, eds Bateson PPG, Klopfer PH
(Plenum, New York), pp 403–420.
2. Wilson DS, Clark AB, Coleman K, Dearstyne T (1994) Shyness and boldness in humans
and other animals. Trends Ecol Evol 9:442–446.
3. Gosling SD, John OP (1999) Personality dimensions in nonhuman animals: A cross-
species review. Curr Dir Psychol Sci 8:69–75.
4. Gosling SD (2001) From mice to men: What can we learn about personality from animal
research? Psychol Bull 127:45–86.
5. Sih A, Bell AM, Johnson JC, Ziemba RE (2004) Behavioral syndromes: An integrative
overview. Q Rev Biol 79:241–277.
6. Groothuis TGG, Carere C (2005) Avian personalities: Characterization and epigenesis.
Neurosci Biobehav Rev 29:137–150.
7. Koolhaas JM, et al. (1999) Coping styles in animals: Current status in behavior and
stress-physiology. Neurosci Biobehav Rev 23:925–935.
8. Korte SM, Koolhaas JM, Wingfield JC, McEwen BS (2005) The Darwinian concept of
stress. Neurosci Biobehav Rev 29:3–38.
9. Nettle D (2006) The evolution of personality variation in humans and other animals.
Am Psychol 61:622–631.
10. Bell AM (2007) Future directions in behavioural syndromes research. Proc R Soc London
Ser B 82:755–761.
11. Reale D, Reader SM, Sol D, McDougall PT, Dingemanse NJ (2007) Integrating animal
temperament within ecology and evolution. Biol Rev 82:291–318.
12. Verbeek MEM, Drent PJ, Wiepkema PR (1994) Consistent individual differences in early
exploratory behavior of male great tits. Anim Behav 48:1113–1121.
13. Mottley K, Giraldeau LA (2000) Experimental evidence that group foragers can con-
verge on predicted producer–scrounger equilibria. Anim Behav 60:341–350.
14. Beauchamp G (2001) Consistency and flexibility in the scrounging behaviour of zebra
finches. Can J Zool 79:540–544.
15. Aron EN, Aron A (1997) Sensory-processing sensitivity and its relation to introversion
and emotionality. J Pers Soc Psych 73:345–368.
16. Belsky J, Bakermans-Kranenburg MJ, van Ijzendoorn MH (2007) For better and for worse:
Differential susceptibility to environmental influences. Curr Dir Psychol Sci 16:300–304.
17. Frost AJ, Winrow-Giffen A, Ashley PJ, Sneddon LU (2007) Plasticity in animal personality
traits: Does prior experience alter the degree of boldness? Proc R Soc London Ser B
274:333–339.
18. Schjolden J, Winberg S (2007) Genetically determined variation in stress responsiveness
in rainbow trout: Behavior and neurobiology. Brain Behav Evol 70:227–238.
19. Sinn DL, Gosling SD, Moltschaniwskyj NA (2008) Development of shy/bold behaviour in
squid: Context-specific phenotypes associated with developmental plasticity. Anim
Behav 75:433–442.
20. Hessing MJC, Hagelso AM, Schouten WGP, Wiepkema PR, van Beek JAM (1994)
Individual behavioral and physiological strategies in pigs. Physiol Behav 55:39– 46.
21. Benus RF, Koolhaas JM, van Oortmerssen GA (1987) Individual differences in behav-
ioural reaction to a changing environment in mice and rats. Behaviour 100:105–122.
22. DeWitt TJ, Sih A, Wilson DS (1998) Costs and limits of phenotypic plasticity. Trends Ecol
Evol 13:77–81.
23. Maynard Smith, J (1982) Evolution and the Theory of Games (Cambridge Univ Press,
Cambridge, UK).
24. Dugatkin LA, Reeve HK (2000) Game Theory and Animal Behavior (Oxford Univ Press,
Oxford).
25. Svensson E, Sheldon BC (1998) The social context of life-history evolution. Oikos
83:466– 477.
26. Kokko H, Lopez-Sepulcre A (2007) The ecogenetic link between demography and
evolution: Can we bridge the gap between theory and data? Ecol Lett 10:773–782.
27. Sinervo B, Calsbeek R (2006) The developmental, physiological, neural and genetical
causes and consequences of frequency-dependent selection in the wild. Annu Rev Ecol
Evol Syst 37:581–610.
28. Rueffler C, Van Dooren TJM, Metz JAJ (2004) Adaptive walks on changing landscapes:
Levins’ approach extended. Theor Popul Biol 65:165–178.
29. Plaistow SJ, Johnstone RA, Colegrave N, Spencer M (2004) Evolution of alternative
mating tactics: Conditional versus mixed strategies. Behav Ecol 15:534–542.
30. Wilson DS, Yoshimura J (1994) On the coexistence of specialists and generalists. Am Nat
144:692–707.
31. Rosenzweig MR, Bennett EL (1996) Psychobiology of plasticity: Effects of training and
experience on brain and behavior. Behav Brain Res 78:57–65.
32. Wilson DS (1998) Adaptive individual differences within single populations. Philos
Trans R Soc London Ser B 353:199–205.
33. Rands SA, Cowlishaw G, Pettifor RA, Rowcliffe JM, Johnstone RA (2003) Spontaneous
emergence of leaders and followers in foraging pairs. Nature 423:432–434.
34. Dall SRX, Houston AI, McNamara JM (2004) The behavioural ecology of personality:
Consistent individual differences from an adaptive perspective. Ecol Lett 7:734–739.
35. McElreath R, Strimling P (2006) How noisy information and individual asymmetries can
make ‘personality’ an adaptation: A simple model. Anim Behav 72:1135–1139.
36. Wolf M, van Doorn GS, Leimar O, Weissing FJ (2007) Life-history trade-offs favour the
evolution of animal personalities. Nature 447:581–584.
37. Stamps JA (2007) Growth-mortality tradeoffs and ‘personality traits’ in animals. Ecol
Lett 10:355–363.
38. Bell AM, Stamps JA (2004) Development of behavioural differences between individ-
uals and populations of sticklebacks, Gasterosteus aculeatus. Anim Behav 68:1339–
1348.
39. Dingemanse NJ, et al. (2007) Behavioural syndromes differ predictably between 12
populations of three-spined stickleback. J Anim Ecol 76:1128–1138.
15830
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0805473105 Wolf et al.
