Download The effect of social facilitation on vigilance in the eastern gray

Behavioral Ecology
doi:10.1093/beheco/arp019
Advance Access publication 16 February 2009
The effect of social facilitation on vigilance in
the eastern gray kangaroo, Macropus giganteus
Olivier Pays,a,b Michel Goulard,c Simon P. Blomberg,d Anne W. Goldizen,b Etienne Sirot,e and Peter
J. Jarmanf
a
Laboratoire Paysages & Biodiversité, Université d’Angers, Campus Belle Beille, 2 bd Lavoisier, 49045 Angers
Cedex, France, bSchool of Integrative Biology, University of Queensland, St. Lucia, Brisbane, QLD 4072,
Australia, cInstitut National de Recherche Agronomique Toulouse, Unité Mixte de Recherche 1201
Dynafor, Castanet-Tolosan, France, dFaculty of Biological and Chemical Sciences, University of
Queensland, St. Lucia, Brisbane, QLD 4072, Australia, eLaboratoire Ecobio, Unité Mixte de Recherche
Centre National de Recherche Scientifique 6553, Université de Rennes 1, Campus de Beaulieu, Rennes,
France, and fEcosystem Management, University of New England, Armidale, NSW 2351, Australia
The relevance of vigilance activity to predator detection has been demonstrated in numerous studies. However, few studies have
investigated the effect of one group member being vigilant on the probability of others being vigilant in group-forming prey
species. Thus, we studied vigilance activity of eastern gray kangaroos Macropus giganteus that still experience occasional predation.
We video recorded the behavior of all group members simultaneously and investigated the probability of a focal group member
being vigilant (or nonvigilant) in relation to other individuals’ vigilant and nonvigilant behaviors. Our results show that the
decision of an individual to exhibit a vigilant posture depended on what it and other group members had been doing (scanning
or foraging) at the preceding second and on group size. The probability of an individual being vigilant was positively affected by
the proportion of companions that were vigilant at the previous second, confirming the existence in this species of a tendency for
synchronization of individual vigilance. Group size affected individuals’ vigilance in 3 ways. First, individuals were more likely to
be vigilant if the proportion of their group mates that was vigilant was high, and this was strengthened with increasing group size.
Second, the effect of the individual’s own vigilance state (vigilant or not) at the previous second also increased with group size.
Third, the probability of an individual being vigilant decreased with group size. These findings increase our understanding of the
much-studied relationship between vigilance and group size. Key words: allelomimesis, group size, kangaroos, neighbor effect,
probability of being vigilant, social facilitation, vigilance. [Behav Ecol 20:469–477 (2009)]
umerous theoretical and empirical studies have demonstrated the importance of vigilance activity to the detection of predators (Lima 1995a, 1995b; Bednekoff and Lima
1998; Treves 2000). In social prey species, the vigilance behavior that group members adopt in response to the risk of being
preyed on is assumed to have a strong effect on their survival.
However, our understanding of the factors that motivate each
group member to raise its head and scan its surroundings,
and remain vigilant, is still limited for most wild species.
Although vigilance in group-forming species has already received considerable attention, this has been mostly in the context of changes in individuals’ vigilance rates with changing
group size (Elgar 1989; Lima 1995b; Roberts 1996; Childress
and Lung 2003), predation risk (Lima 1987; Hunter and
Skinner 1998; Laundré et al. 2001; Lung and Childress
2006), or strength of intraspecific competition among group
members (Baldellou and Henzi 1992; Robinette and Ha
2001; Tchabovsky et al. 2001; Hirsch 2002). Many studies,
especially of mammal species, have demonstrated that
a group member’s vigilance is affected by the presence of
its neighbors, although the effect may differ between species
forming open-membership and those forming persistent
groups (Blumstein et al. 2001; Treves et al. 2001; Cameron
and Du Toit 2005; Di Blanco and Hirsch 2006, Pays et al.
2008). In birds, some studies also show that the decision of
N
Address correspondence to O. Pays. E-mail: [email protected].
Received 3 October 2008; revised 5 December 2008; accepted 5
December 2008.
The Author 2009. Published by Oxford University Press on behalf of
the International Society for Behavioral Ecology. All rights reserved.
For permissions, please e-mail: [email protected]
an individual within a flock to be vigilant is partly dependent
on the behavior of its close neighbors (Bekoff 1995; Fernández-Juricic and Kacelnick 2004; Fernández-Juricic et al. 2004,
2005). Several theoretical models have also assumed that the
probability of an individual being vigilant will depend on
other group members’ activities (Ruxton 1996; Bahr and Bekoff 1999; Jackson and Ruxton 2006; Sirot 2006; Sirot and
Touzalin 2008). However, little empirical evidence exists in
the literature on how the probability of an individual being
vigilant varies with other group members’ vigilance activity
and how this is affected by group size.
Synchronization of vigilance among group members has
been shown in several avian and mammal species (Bertram
1980; Fernández et al. 2003; Ebensperger et al. 2006; Pays,
Jarman, et al. 2007; Pays, Renaud, et al. 2007). In these species, vigilant and nonvigilant periods are positively correlated
among group members and thus synchronized in phase. This
synchronization of scanning periods among individuals produces a pattern of waves of collective vigilance during each of
which several individuals are vigilant at the same time. Although such waves of vigilance may be caused by simultaneous
detection of an approaching danger by group members, recent models have shown that reciprocal influences among
individuals in the group may be sufficient to generate them.
In particular, individuals surrounded by neighbors displaying
high levels of vigilance may perceive a high predation risk,
through the process of social information transfer, and respond by increasing their own vigilance, even if no predator
is present (Sirot 2006). Furthermore, when predators display
marked preferences for stragglers in an attack, it also becomes
adaptive for an individual to copy the vigilance of its
Behavioral Ecology
470
neighbors, in order not to react too slowly and lag behind
them whenever an attack occurs (Sirot and Touzalin 2008).
In both situations, vigilance in the group may then become
contagious, as a result of an adaptive response to the vigilance
attitude of companions and relatively independently of the
real level of risk in the environment (see also Giraldeau
et al. 2002).
In a previous study of eastern gray kangaroos Macropus giganteus, Pays, Jarman, et al. (2007) demonstrated that both
the onset and the end of individuals’ scanning bouts, as well
as bouts of nonvigilant activity, tended to be synchronized,
producing waves of collective vigilance. However, although
the methods in that study allowed the detection of synchronization of vigilance activity between group members, they did
not allow the determination of the causes of this observed
collective pattern. Waves of collective vigilance were likely
triggered by several causes, including proactive vigilance
(i.e., the focal animal raised its head without any intra- or
extragroup stimulus) and re-active vigilance (i.e., either several group members raised their heads because they had independently detected the same disturbance outside the group
at the same time or 1 individual raised its head and that action
alerted other group members without their having independently seen the disturbance themselves).
Although the idea that individuals monitor their environment while foraging has received experimental support (Quirici
et al. 2008), the proximate mechanisms and rules that trigger
collective patterns of vigilance (such as coordination or synchronization) are still unknown. We investigated the emergence of collective patterns of synchronization in wild eastern
gray kangaroos. Our aims were to test 1) whether the probability of a focal individual being vigilant is positively affected by
the proportion of companions that are vigilant and 2) whether
this effect is affected by group size. We address these aims by
using a new statistical method to quantify, as precisely as possible, the influence of companions’ activity (scanning or foraging) on a focal group member’s vigilance behavior. We studied
10-min behavioral sequences of foragers in groups during
which the observer did not detect any evident extragroup disturbances (i.e., presence of predators or other heterospecifics,
human disturbance, etc.). Thus, we tried to exclude as much as
possible collective synchronized vigilance events during which
several group members raised their heads simultaneously yet
independently of one another because they reacted to the
same extragroup disturbances. As we recorded the behavior
of all group members simultaneously, we analyzed the probability of a focal group member being vigilant in relation to
other group members’ behaviors.
MATERIALS AND METHODS
Study areas and animals
Fieldwork was carried out in Australia, at Newholme (3025#S,
15138#E) near Armidale, New South Wales, in July 2005 and at
Elanda Point (2615#S, 15259E) in southeast Queensland in
March 2007. More extensively described in a previous study
(Pays, Jarman, et al. 2007), the first study site comprised eucalypt forest, woodland, and open pastures of predominantly
native species. The second area was located on the edge of
Lake Cootharaba and completely surrounded by bush within
Great Sandy National Park. In this area, observations of eastern gray kangaroos were conducted within an approximately
25-ha area of cleared and semicleared land within the national
park. Animals were not marked for individual recognition at
either site. We tested for an effect of study area on vigilance
activity of group members, controlling for group size in our
data set, and did not find any significant differences between
the 2 sites (linear models of effect of site on Logit-transformed
proportion of time individual spent vigilant, P . 0.09). Therefore, we pooled the data from the 2 field sites.
Recording data
To study the pattern of collective vigilance arising from the individual behaviors of group members, we collected behavioral
data by video recording all members of focal groups of eastern
gray kangaroos for 10-min periods. During sampling, all group
members were in the camera’s field of view. Data were collected
by 1 observer (OP) who kept a minimum distance of 100 m
between himself and the focal group. When filming animals,
the observer did not move and was hidden in the landscape.
Only groups whose predominant activity was foraging and that
did not move far during recording were considered. Furthermore, only groups whose size and composition did not change
during recording were retained in the analysis. Identifying
groups on the basis of 30-m maximal separation between members and the maintenance of social and spatial cohesion of
members during the whole focal sample (Jarman and Coulson
1989) produced no ambiguous cases. We began the video
sequence when all animals displayed feeding behavior. When
analyzing video sequences, the observer sought to identify
whether he might have triggered group members’ scans by
looking at the orientation of animals’ head-up stares. No
group was filmed more than once a day to ensure that sampled groups were independent of each other; in fact, individuals change their group membership many times each day,
and sometimes even many times in an hour. We considered an
animal as vigilant when it did not move its feet but raised its
head above horizontal, scanning its surroundings. No ambiguities were encountered in distinguishing a vigilant from
a nonvigilant animal. Nonvigilance corresponded predominantly to feeding. The position of each group member during
the 10-min sequences was not recorded because the dynamics
of individual movements led individuals to change their positions constantly within groups.
As we wished to study whether synchronization of vigilance
occurs that is not triggered by individuals independently
responding to an external stimulus, we did not wish to include
in the analyses behavioral sequences of groups within which at
least 2 individuals reacted to the same extragroup source of
disturbance. We therefore reviewed video sequences, rejecting
from analysis any in which several group members, evidently
reacting to an extragroup source of disturbance, raised and oriented their heads in the same direction at the same second.
There was only one such case in 124 groups studied, which
we have not included in our analyses; we are confident that
in all other cases, individuals were not independently responding to the same external stimulus.
For analysis, video sequences were converted to analytic
sequences. For each individual within each group, a binary sequence was constructed reflecting its activity state (nonvigilance activity: 0 and vigilance activity: 1) precisely at each
second for 600 s. All individual binary video sequences were
aligned within groups, and thus, we recorded at precisely the
same time the activity of each group member.
Following this strict procedure, we studied 510 individuals at
the 2 study sites that were in 123 groups ranging from 2 to 15
individuals (see Table 1 for details). No groups of 13 individuals were recorded in our study.
Data Analyses
Our main focus in this paper is to investigate whether the probability of a group member being vigilant was affected by the
proportion of other group members already vigilant (Phu).
Pays et al.
•
Neighbor effect on vigilance
471
Table 1
For each group size, effects of the individual state of the focal animal at the previous second (S) and the proportion of companions that were
vigilant at the previous second (Phu), analyzed using a generalized linear mixed-effects model fitted by penalized quasi-likelihood (GLMM-PQL)
procedure (binomial, link function: Logit) with individual and group as 2 nested random effects (see Materials and Methods)
Group size
Ngp
Nind
Nevent
Factor
Coefficient
SE
numDF
denDF
t
P
2
19
38
22800
3
21
63
37800
4
14
56
33600
5
5
25
15000
6
2
12
7200
7
5
35
21000
8
6
48
28800
9
4
36
21600
10
6
60
36000
11
3
33
19800
12
5
60
36000
14
1
14
8400
15
2
30
18000
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
Intercept
Phu
S
23.274
0.357
5.247
22.706
0.592
5.669
22.921
0.822
5.310
22.599
0.602
5.234
24.371
1.546
6.537
24.060
0.957
5.978
23.978
1.223
5.474
24.276
1.833
5.821
24.542
2.122
5.670
25.027
2.663
6.875
24.419
1.565
5.760
25.378
3.210
6.779
25.161
2.345
6.614
0.087
0.064
0.058
0.149
0.085
0.053
0.124
0.084
0.049
0.365
0.195
0.079
0.132
0.271
0.150
0.130
0.192
0.078
0.092
0.181
0.063
0.069
0.217
0.083
0.095
0.234
0.070
0.110
0.297
0.121
0.100
0.208
0.069
0.170
1.292
0.222
0.136
0.524
0.134
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
22760
22760
22760
37735
37735
37735
33542
33542
33542
14973
14973
14973
7186
7186
7186
20963
20963
20963
28750
28750
28750
21562
21562
21562
35938
35938
35938
19765
19765
19765
35938
35938
35938
8384
8384
8384
17968
17968
17968
237.472
5.567
90.750
218.174
6.931
106.674
223.494
9.753
108.648
27.111
3.089
66.083
232.938
5.707
43.501
231.148
4.986
76.354
243.108
6.187
86.938
261.611
8.660
69.918
247.698
9.062
80.661
245.814
8.958
56.687
244.183
7.518
83.563
231.551
2.485
30.547
237.974
4.478
49.370
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
0.0001
0.002
0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
,0.0001
0.0001
0.013
0.0001
,0.0001
,0.0001
,0.0001
Ngp is the number of sampled groups, Nind is the number of sampled individuals, and Nevent is the total number of sampled seconds, pooling
the binary sequences of all individuals.
We analyzed each group size separately for 2 main reasons.
First, group size and group geometry might be linked, and
some studies have shown that group geometry may affect the
individual vigilance of group members (Hamilton 1971; Elgar
1989; Bekoff 1995; Bahr and Bekoff 1999; Di Blanco and
Hirsch 2006). Second, even if equal proportions of companions were vigilant at the previous second, the effect on a focal
individual’s vigilance activity of 1 vigilant companion in
a group of 2 might be different to that of 2 vigilant individuals
in a group of 4 or 3 vigilant individuals in a group of 6, and so
on.
To test whether the probability of a focal group member being vigilant was affected by the proportion of vigilant individuals among its companions at the previous second (Phu), we
first investigated whether there was autocorrelation in individual sequences of 600 s. Whenever data are collected over time,
there may be correlations between successive observations.
Therefore, we first checked whether autocorrelation existed
in our data set, and then we used a statistical procedure that
took into account time correlation in the data set. We used an
autocorrelation function that provided an estimate of the correlation between observations separated by a lag of m time
units, for m ¼ 0, 1, 2, . . ., M (see Priestley 1982 for more
statistical details). We estimated the value of the autocorrelation function on all of the individual binary sequences of
vigilance and nonvigilance behaviors. Figure 1a demonstrates
the presence of serial correlation. If no autocorrelation estimate (given by the vertical lines for positive lags) had fallen
outside the strip defined by the 2 dotted lines, we could safely
assume that no serial correlation existed. However, Figure 1a
shows values far outside those limits and thus that substantial
serial correlation existed in individual sequences. Therefore,
a vigilant act displayed by a focal individual seemed to be
affected by its behavior at the previous second, that is, with
a lag not only at t 2 1 s but also at t 2 2, t 2 3 s, and so on.
Consequently, to test whether the probability of a focal group
member being vigilant was affected by the proportion of companions that were vigilant at the previous second (Phu), we
needed to use a statistical procedure that controlled for serial
correlations.
Because the response variable, the focal individual’s behavior (vigilant: 1 or not vigilant: 0), had a binomial distribution,
we used generalized linear mixed-effects models (binomial,
link function: logit) fitted by PQL (penalized quasi-likelihood,
472
Behavioral Ecology
Figure 1
(a) Autocorrelation function (ACF) estimated on all individual binary sequences of 600 s of vigilance and nonvigilance behaviors for groups of
4 individuals. The value of the autocorrelation function (given by the vertical lines for positive lags) at lag 0 is always 1. The horizontal
dotted lines provide an approximate 95% confidence interval for the autocorrelation estimate at each lag. Here, Lag 1 is 1 s. (b) ACF estimated
on the residuals of the model including the individual state of the focal animal at the previous second and the proportion of companions that
were vigilant at the previous second using a GLMM-PQL procedure (see Materials and Methods and Table 1). (c) ACF estimated on the residuals of
the same minimal model but without including the individual state of the focal animal at the previous second.
Breslow and Clayton 1993), including Phu (the proportion of
companions that were vigilant at the previous second) as an
independent variable. According to the link function, the
probability of being vigilant (P1) was Logit transformed in
the procedure (i.e., LogitðP1 Þ ¼ LnðP1 =ð12P1 ÞÞ). Thus, the
computed model in this case assumed that the relationship
between the probability of being vigilant (P1) and the proportion of companions vigilant at the previous second (Phu)
was LogitðP1 Þ ¼ a1bPhu 1e; where a was the intercept; b, the
estimated coefficient derived for the effect of Phu; and e, errors. Moreover, we considered the hierarchical nature of our
data set including individuals and groups as 2 nested random
effects.
We also had to control for autocorrelation in the statistical
procedure. As shown before, the behavior of a focal individual
was affected by its behavior in the previous seconds, that is, lag
not only at t 2 1 s but also at t 2 2 s, t 2 3 s, and so on. We
considered Markov’s assumption that the probability of being
vigilant taking into account the past (t 2 1, t 2 2, t 2 3 s, etc.)
can be reduced to a law accounting for only the immediate
past t 2 1 s (see Meyn and Tweedie 1993). Accepting Markov’s
assumption, we assumed that the probability of a focal group
member being vigilant was affected by its own state at the
previous second (state 0: nonvigilant and state 1: vigilant).
Thereafter, we included this independent variable in the
model.
To summarize, the complete model included 2 fixed variables, focal individual state at the previous second (S) and
the proportion of companions that were vigilant at the previous second (Phu), and could be represented by the following
simplified algorithm: LogitðP1 Þ ¼ a1bPhu 1dS1e; where a
was the intercept; b and d, estimated coefficients; and e, errors. To check whether autocorrelation had been controlled
correctly, we used the same procedure as the one applied to
the individual sequences (the previous autocorrelation function modeling procedure) but on the residuals of the model
that included the individual state of the focal animal at the
previous second, generating the results shown in Figure 1b
Pays et al.
•
Neighbor effect on vigilance
473
Figure 2
For each group size (GS), the
probability of a focal group
member being vigilant in relation to the proportion of companions that were vigilant at the
previous second (Phu); (a) for
GS from 2 to 8 individuals, (b)
for GS from 9 to 15, when the
focal animal was already vigilant
at the previous second (P11), (c)
for GS from 2 to 8, and (d) for
GS from 9 to 15, when the focal
group member was nonvigilant
at the previous second (P01).
See Table 1 for the coefficients
attributed to each factor in the
algorithms of each group size.
For example, for a group of
2 individuals (GS 2), P11 ¼
exp(1.973 1 0.357Phu)/(1 1
exp(1.973 1 0.357Phu)) and
P01 ¼ exp(23.274 1 0.357Phu)/
(1 1 exp(23.274 1 0.357Phu)).
(example for group size of 4 individuals). Figure 1c shows the
autocorrelation function estimated on the residuals from
the same model but without including the individual state
of the focal animal at the previous second. Comparison between Figure 1b and c confirms that including as a variable in
the model the individual state of the focal animal at the previous second and accepting Markov’s assumption were appropriate and allowed us to control for serial correlations in our
procedure.
For each group size, we obtained an estimate of each coefficient derived for the effect of Phu (proportion of companions
that were vigilant at the previous second) and S (focal individual state at the previous second). Then we tested whether
the effects of the proportion of companions that were vigilant
and the individual state of the focal animal at the previous
second varied among group sizes using linear model procedures. We compared whether a linear, second-degree polynomial or exponential function best fitted the coefficients b,
derived for Phu, and d, derived for S, using Akaike’s criterion
(AIC) statistics (Akaike 1974). However, as the quotients
between the number of observations and the number of parameters in models were less than 40, we used corrected
Akaike’s criterion AICc statistics (Hurvich and Tsai 1995).
The model with the smallest AICc statistic was considered
the best (Burnham and Anderson 2002).
All of the statistical analyses were computed using R software
(the R Foundation for Statistical Computing, 2007).
RESULTS
First, for each group size, we tested whether the probability of
a focal group member being vigilant was affected by the pro-
portion of companions that were vigilant at the previous second (Phu) as well as by the focal animal’s own state (vigilant
or not) at the previous second. Results are presented in Table 1.
Whatever the group size (from 2 to 15), the probability of
an individual being vigilant was affected by its own individual
state at the previous second. An individual that was vigilant at
the previous second had a higher probability of being vigilant
than if it had been nonvigilant (foraging). Moreover, according
to the positive sign derived from Phu, the probability of an
individual being vigilant was positively affected by the proportion of companions that were vigilant at the previous second
(Table 1).
The statistical procedure allowed us to compute 2 different
algorithms from the results in Table 1 for the probability of
being vigilant for each group size. For example, according to
the coefficient derived for each factor in Table 1, for a group
of 2 individuals, the probability (P11) of being vigilant for
a group member that was already vigilant at the previous second was given by: LogitðP11 Þ ¼ 23:27415:24710:357Phu . As
LogitðP11 Þ ¼ LnðP11 =ð12P11 ÞÞ, P11 ¼ expð1:97310:357Phu Þ=
ð11expð1:97310:357Phu ÞÞ. Furthermore, the probability
(P01) of being vigilant for a group member that was nonvigilant at the previous second was given by Logit P01 ¼
23:27410:357Phu . So, P01 ¼ expð23:27410:357Phu Þ=ð11exp
ð23:27410:357Phu ÞÞ. From these algorithms, we then calculated the probability of a focal group member being vigilant
when the focal animal was or was not already vigilant at the
previous second in relation to the proportion of companions
that were vigilant at the previous second (Phu) within each
group size (Figure 2). Figure 2 shows 2 major points. First,
given that Figure 2a,c corresponds to small groups (2 group
size [GS] 8 individuals) and Figure 2b,d to larger ones (9 GS 15), global group size effects exist for both the
Behavioral Ecology
474
Table 2
Group size effect on 1) b Phu, the effect of the proportion of companions that were vigilant at the previous second, and 2) d S, the effect of the
focal individual’s state at the previous second on the vigilance of focal individuals
Models
(1)
b Phu ¼ c 1 fGS
b Phu ¼ c 1 fGS 1 gGS2
b Phu ¼ c 1 fGS2
b Phu 5 exp(g 1 zGS)
(2)
d S 5 g 1 zGS
d S ¼ c 1 fGS 1 gGS2
d S ¼ c 1 fGS2
d S ¼ exp(c 1 fGS)
c
f
20.008
20.128
0.661
20.932
0.188
0.225
0.011
0.144
5.122
5.268
5.456
1.640
0.098
0.054
0.006
0.016
g
k
AIC
AICc
DAICc
xi
26.187
26.142
27.585
21.937
3
4
3
3
18.375
20.286
21.171
9.875
18.437
20.390
21.233
9.937
8.500
10.453
11.296
0.000
0.095
0.059
0.047
0.799
26.899
26.840
26.922
216.522
3
4
3
3
19.798
21.680
19.844
27.043
19.860
21.784
19.906
27.105
0.000
1.924
0.046
7.245
0.361
0.223
0.357
0.059
LogLik
20.002
0.003
The best models with the lowest AICc are in boldface. c, f, and g are coefficients. LogLik is the Loglikelihood, k is the number of estimated
parameters, AIC and AICc are Akaike’s criterion and the corrected criterion, DAICc is the difference of AICc between the model and the best
one, and xi is the weight of the model.
probability that an individual would be vigilant when it had
already been vigilant and the probability that it would be
vigilant when it had not been vigilant, at the preceding second. When group size increased, both these probabilities
tended to decrease. Second, the highest values of the 2 probabilities of being vigilant did not correspond to groups of 2
individuals but to groups of 3, 4, and 5 individuals.
These calculations gave us, for each group size, the coefficients for the effects of Phu (proportion of companions that
were vigilant at the previous second) and S (focal individual
state at the previous second) on the probability of an individual being vigilant. We then tested for group size effects on
these coefficients by investigating whether linear, seconddegree polynomial or exponential functions best fitted the
coefficients b, derived for Phu, and d, derived for S. Table 2
shows that the effect of Phu was positively affected by group
size. When group size increased, the effect of Phu on the
probability of an individual being vigilant increased. Moreover, as can be seen by the value of AICc, this increase was
characterized by an exponential function (Figure 3a). Table 2
also shows that the effect of the focal individual’s state at the
previous second (S) was positively affected by group size
through a linear relationship (Figure 3b).
DISCUSSION
Our main findings in this article were that an individual’s probability of being vigilant was positively affected by the proportion of group members that had been vigilant at the
previous second and was affected in 3 ways by group size. First,
individuals were more likely to be vigilant if the proportion of
their group mates that was vigilant was high, and this effect was
strengthened with increasing group size. Second, the effect of
the individual’s own vigilance state (vigilant or not) at the previous second also increased with group size. Third, the probability of an individual being vigilant, in contrast, decreased
with group size.
Our results show that the effect of the behavior of an individual on its probability of being vigilant at the following second was positively affected by group size in a linear way. Thus,
the higher the group size, the greater the effect of the focal
animal’s previous activity (foraging or scanning) will be on
its own probability of being vigilant at the following second.
Thus, our analyses suggested that the decision that an individual made about being vigilant depended on both group size
and what it was doing previously (scanning or foraging). However, multiple other factors (Beauchamp 1998) such as the
individual’s energetic state (McNamara and Houston 1986),
predation pressure (Hunter and Skinner 1998; Laundré et al.
2001), or the presence of competitors (Baldellou and Henzi
1992) may also play key roles in a group member’s decision to
be vigilant, and this combination of factors complicates our
understanding of vigilance.
An individual kangaroo that was already vigilant at the previous second had a higher probability of being vigilant than if it
had been nonvigilant (foraging) at the previous second. This
result may be related to the durations of scan and interscan
bouts. Jarman (1987) and Pays, Jarman, et al. (2007) found
that for small groups (2–5 individuals) of eastern gray kangaroos, the mean duration of individual bouts of vigilance (scanning) was about 10 s and the mean duration of bouts of
foraging (interscan) was about 40 s. Thus, analyzing individual activity at each second, there were finite intrinsic probabilities that an individual’s previously recorded activity state
(vigilant or foraging) would persist.
The probability of an individual being vigilant was also positively affected by the proportion of companions that were vigilant at the previous second, which confirms the tendency for
synchronization of individual vigilance that has been previously described in this species (Pays, Jarman, et al. 2007) as
well as some other marsupial species (Pays et al. 2008), rodents (Ebensperger et al. 2006), ungulates (Pays, Renaud,
et al. 2007), large flightless birds (Bertram 1980; Fernández
et al. 2003), and passerines (Lazarus 1979; Bekoff 1995;
Fernández-Juricic et al. 2004). Moreover, when group size increased, the effect of the proportion of companions that were
vigilant at the previous second on the probability of an individual being vigilant increased. This is a novel finding and is
particularly interesting given that we avoided collecting data
in situations in which group members raised their heads because all individuals had simultaneously detected the same extragroup disturbance. Our results thus tend to show that
reciprocal influences among neighbors play an important role
in the determination of individual vigilance, which may be, in
part, independent of external cues.
Considerable methodological challenges occur in the
study of behavioral synchronization (Ward 1985; Engel and
Lamprecht 1997; Pays, Jarman, et al. 2007). Here, we have
used a new method of analysis to show that the probability
of a group member being vigilant was strongly affected by the
proportion of other group members that were vigilant at the
preceding second. This statistical method has many other
applications in behavioral ecology studies in which temporal
autocorrelations exist. Indeed, it can be used for any binary
Pays et al.
•
Neighbor effect on vigilance
Figure 3
(a) Relationship between group size and the influence of
companions’ alertness at the previous second on the probability of
the focal animal being alert and (b) relationship between group size
and the influence of the focal individual’s alertness at the previous
second on the probability of the focal animal being alert. See Table 1
for the values (6standard error [SE]) of the coefficients b attributed
to Phu and d to S for each group size and Table 2 for statistical details
for the relationship.
behavioral sequences between individuals to investigate
whether the probability that an individual will exhibit a behavior
depends on the behavior exhibited by other group members.
Previous studies that investigated patterns of synchronized
vigilance could not differentiate between the proximate mechanisms that triggered the observed synchronization of vigilance
among group members. Our study showed that the probability
of a group member being vigilant was affected by the numbers
of vigilant and nonvigilant companions at the previous second,
which suggests that animals might often react to the behavior
of other group members by considering those individuals’
475
head-up acts as information about potential risks outside the
group. Thus, our results improve our understanding of the
processes that motivate individuals to scan their environment.
Successive head-up acts seem to emerge progressively in the
group because of the contagious effect triggered by individuals
reacting to each others’ behavior. First, 1 individual might raise
its head proactively (i.e., without a stimulus from either inside
or outside the group) or because of an extragroup stimulus.
This individual’s vigilant act may or may not initiate a wave
of collective vigilance (Pays and Jarman 2008). Such single,
proactive vigilant scans might occur independently between
group members and randomly through time, as assumed by
Pulliam’s (1973) model. Second, an allelomimetic copying
phenomenon might emerge, triggering an overlap of vigilance between the initiator and at least one other group member. These other group members might exhibit re-active
vigilance, reacting to the initiator’s head-up act or vigilant
state. In group-forming prey species in which information
passes between group members, a communication signal
(e.g., an alarm call or the sudden departure of that individual,
Pulliam 1973; Davis 1975; Lazarus 1979) might commonly induce such a pattern. Studying bird flocks under attacks, several authors have recorded that if an individual fleeing
because it has detected a predator communicates this information to its group mates, this triggers an immediate departure of other birds in a few seconds, increasing their chance of
escaping (Hilton et al. 1999; Cresswell et al. 2000; Quinn and
Cresswell 2005). It has also been suggested that birds alerted
by the flight of their companions first take some time to assess
the real level of threat before taking off (Quinn and Cresswell
2005). It could thus be that vigilance, as well as flight, induces
antipredatory behavior among members of a group.
The relationship between group size and the influence of
companions’ alertness at the previous second on the probability of an individual being vigilant in the subsequent second followed an exponential function (Figure 3). This provides
evidence of an amplification effect when the proportion of
companions that are vigilant increases. Amplification processes in which collective behaviors emerge because 1 individual’s behavior is amplified by the action of many
other individuals (positive feedback) have been reported for
a very wide range of group-forming organisms (Parrish and
Edelstein-Keshet 1999; Krause and Ruxton 2002; Couzin and
Krause 2003) such as social insects (Deneubourg and Goss
1989; Bonabeau et al. 1997; Dussutour et al. 2005), birds
(Jackson et al. 2008), and mammals (Meunier et al. 2008).
For example, studying the effect of social facilitation on foraging success in Gyps vultures, Jackson et al. (2008) reported
that vultures rely heavily on cues from conspecifics to locate
carcasses via local enhancement. The probability of a vulture
finding a carcass is a function of vulture density in the habitat
and carcass numbers, following a sigmoid relationship. Thus,
these authors highlighted an amplification process with the
probability of foraging success increasing exponentially (at
least for the first stage of the curve) with the density of vultures close to the carcass.
We found that individuals in smaller groups were more vigilant those in larger groups, as has been found in many studies
(e.g., Jarman 1987); however, the increase with group size of
the effect on vigilance of the proportion of group members
that were vigilant suggests that synchronization of vigilance
would be greater at larger group sizes. However, we might
have expected that group members would act more synchronously in smaller groups than in larger ones because of the
geometry of such groups, in particular the clustering of neighbors, as an individual would synchronize better with its proximate neighbors than with more distant group members.
Thus, we need more studies to understand the relationships
476
between group size, group geometry, degree of synchrony of
vigilance and foraging activity, and the amplification process
among group members.
In our study, we used a lag of 1 s to investigate synchronization of vigilance for statistical reasons (particularly to control
temporal autocorrelation) and because we have observed that
individuals often react to each other’s head-up acts with a very
short delay. However, it is clear that individuals will not always
respond to their neighbors’ actions that quickly. Our analysis
used Markov’s assumption that the probability of being vigilant
taking into account the past (t 2 1, t 2 2, t 2 3 s, etc.) can be
reduced to a law accounting for only the immediate past t 2 1 s
(see Meyn and Tweedie 1993); thus, we partially considered the
history of correlations in behaviors during the previous seconds. However, our procedure did not allow us to test whether
the probability of vigilance was affected by the companions’
behavior and the focal individual’s previous state using lag times
of over 1 s (e.g., 2 or 3 s previously) or lag periods greater than 1 s
in duration (e.g., the previous 1–5 s). We suggest that such
analyses should be done. Experimental studies are also needed
to analyze and quantify the delay shown by a group member in
reacting to an alerted neighbor and to determine the efficacy
of the communication signals that exist in large mammalian
herbivores and more generally in other prey species.
FUNDING
Laboratiore Paysages and Biodiversité d’Angers, France.
We acknowledge M. Curkpatrick for permission to use the Newholme
Field Laboratory at the University of New England. We would like to
express our thanks to Brian Mays and the staff who allowed us to study
kangaroos at Elanda Point and provided logistical support. This research was approved by the University of Queensland’s Animal Experimentation Ethics Committee and carried out under a Scientific
Purposes Permit from Queensland’s Environmental Protection
Agency. We are also grateful to Mike Speed and F. Stephen Dobson
for their comments on an earlier version on this article.
REFERENCES
Akaike H. 1974. A new look at statistical model identification. IEEE
Trans Autom Contr. 19:716–723.
Bahr DB, Bekoff M. 1999. Predicting flock vigilance from simple passerine interactions: modelling with cellular automata. Anim Behav.
58:831–839.
Baldellou M, Henzi SP. 1992. Vigilance, predator detection and the
presence of supernumerary males in vervet monkey troops. Anim
Behav. 43:451–461.
Beauchamp G. 1998. The effects of group size on mean food intake
rate in birds. Biol Rev. 73:449–472.
Bednekoff PA, Lima SL. 1998. Randomness, chaos and confusion in
the study of antipredator vigilance. Trends Ecol Evol. 13:284–287.
Bekoff M. 1995. Vigilance, flock size, and flock geometry: information
gathering by western evening grosbeaks (Aves, Fringillidae). Ethology. 99:150–161.
Bertram BCR. 1980. Vigilance and group size in ostriches. Anim Behav. 28:278–286.
Blumstein DT, Daniel JC, McLean IG. 2001. Group size effects in
quokkas. Aust J Zool. 49:641–649.
Bonabeau E, Theraulaz G, Deneubourg JL, Aron S, Camazine S. 1997.
Self-organization in social insects. Trends Ecol Evol. 12:188–193.
Breslow NE, Clayton DG. 1993. Approximate inference in generalized
linear mixed models. J Am Stat Assoc. 88:9–25.
Burnham KP, Anderson DR. 2002. Model selection and multi-model
inference. A practical information-theoretic approach. New York:
Springer.
Cameron E, Du Toit JT. 2005. Social influences on vigilance behaviour
in giraffes, Giraffa camelopardalis. Anim Behav. 69:1337–1344.
Behavioral Ecology
Childress MJ, Lung MA. 2003. Predation risk, gender and the group
size effect: does elk vigilance depend upon the behaviour of conspecifics? Anim Behav. 66:389–398.
Couzin ID, Krause J. 2003. Self-organization and collective behavior in
vertebrates. Adv Study Behav. 32:1–75.
Cresswell W, Hilton GM, Ruxton GD. 2000. Evidence for a rule governing the avoidance of superfluous escape flights. Proc R Soc
Lond B. 267:733–737.
Davis JM. 1975. Socially induced flight reactions in pigeons. Anim
Behav. 23:597–601.
Deneubourg JL, Goss S. 1989. Collective patterns and decisionmaking. Ethol Ecol Evol. 1:295–311.
Di Blanco Y, Hirsch BT. 2006. Determinants of vigilance behavior in
the ring-tailed coati (Nasua nasua): the importance of within-group
spatial position. Behav Ecol Sociobiol. 61:173–182.
Dussutour A, Deneubourg JL, Fourcassié V. 2005. Amplification of
individual preferences in a social context: the case of wall-following
in ants. Proc R Soc Lond B. 272:705–714.
Ebensperger LA, Hurtado MJ, Ramos-Jiliberto R. 2006. Vigilance and
collective detection of predators in degus (Octodon degus). Ethology.
112:879–887.
Elgar MA. 1989. Predator vigilance and group size in mammals
and birds: a critical review of the empirical evidence. Biol Rev.
64:13–33.
Engel J, Lamprecht J. 1997. Doing what everybody does? A procedure
for investigating behavioural synchronization. J Theor Biol. 185:
255–262.
Fernández GJ, Capurro AF, Reboreda JC. 2003. Effect of group size on
individual and collective vigilance in greater rheas. Ethology.
109:413–425.
Fernández-Juricic E, Kacelnick A. 2004. Information transfer and gain
in flocks: the effect of quality and quantity of social information at
different neighbor distances. Behav Ecol Sociobiol. 55:502–511.
Fernández-Juricic E, Siller S, Kacelnick A. 2004. Flock density, social
foraging: an experiment with starlings. Behav Ecol. 15:371–379.
Fernández-Juricic E, Smith R, Kacelnick A. 2005. Group-size and
distance-to-neighbor effects on feeding and vigilance in brownheaded cowbirds. Anim Behav. 73:771–778.
Giraldeau LA, Valone T, Templeton JJ. 2002. Potential disadvantages
of using socially acquired information. Phil Trans R Soc Lond B.
357:1559–1566.
Hamilton WD. 1971. Geometry for the selfish herd. J Theor Biol.
31:295–311.
Hilton GM, Cresswell W, Ruxton GD. 1999. Intraflock variation in the
speed of escape-flight response on attack by an avian predator.
Behav Ecol. 10:391–395.
Hirsch BT. 2002. Social monitoring and vigilance behavior in brown
capuchin monkeys (Cebus apella). Behav Ecol Sociobiol. 52:458–464.
Hunter TB, Skinner JD. 1998. Vigilance behaviour in African ungulates: the role of predation pressure. Behaviour. 135:195–211.
Hurvich CM, Tsai CL. 1995. Model selection for extended quasi-likelihood
models in small samples. Biometrics. 62:1077–1084.
Jackson AL, Ruxton GD. 2006. Toward an individual-level understanding of vigilance: the role of social information. Behav Ecol.
17:532–538.
Jackson AL, Ruxton GD, Houston DC. 2008. The effect of social
facilitation on foraging success in vultures: a modelling study. Biol Lett.
4:311–313.
Jarman PJ. 1987. Group size and activity in eastern grey kangaroos.
Anim Behav. 35:1044–1050.
Jarman PJ, Coulson G. 1989. Dynamics and adaptiveness of grouping
in macropods. In: Grigg G, Jarman PJ, Hume I, editors. Kangaroos,
wallabies and rat-kangaroos. New South Wales (Australia): Surrey
Beatty & Sons Pty Limited. p. 527–548.
Krause J, Ruxton GD. 2002. Living in groups. Oxford: Oxford University Press.
Laundré JW, Hernandez L, Altendorf KA. 2001. Wolves, elk, and bison: reestablishing the ‘‘landscape of fear’’ in Yellowstone National
Park, U.S.A. Can J Zool. 79:1401–1409.
Lazarus J. 1979. The early warning function of flocking in birds: an
experimental study with captive quelea. Anim Behav. 27:855–865.
Lima SL. 1987. Vigilance while feeding and its relation to the risk of
predation. J Theor Biol. 124:303–316.
Lima SL. 1995a. Back to the basics of anti-predatory vigilance: the
group-size effect. Anim Behav. 49:11–20.
Pays et al.
•
Neighbor effect on vigilance
Lima SL. 1995b. Collective detection of predatory attack by social
foragers: fraught with ambiguity? Anim Behav. 50:1097–1108.
Lung MA, Childress MJ. 2006. The influence of conspecifics and predation risk on the vigilance of elk (Cervus elaphus) in Yellowstone
National Park. Behav Ecol. 18:12–20.
McNamara JM, Houston AI. 1986. The common currency for behavioral decisions. Am Nat. 127:358–378.
Meunier H, Petit O, Deneubourg JL. 2008. Social facilitation of fur rubbing behavior in white-faced capuchins. Am J Primatol. 70:161–168.
Meyn SP, Tweedie RL. 1993. Markov chains and stochastic stability.
New York: Springer.
Parrish JK, Edelstein-Keshet L. 1999. Complexity, pattern, and evolutionary trade-offs in animal aggregation. Science. 284:99–101.
Pays O, Dubot AL, Jarman PJ, Loisel P, Goldizen AW. 2009. Vigilance
and its complex synchrony in the red-necked pademelon, Thylogale
thetis. Behav Ecol. 20:22–29.
Pays O, Jarman P. 2008. Does sex affect both individual and collective
vigilance in social mammalian herbivores: the case of the eastern
grey kangaroo? Behav Ecol Sociobiol. 62:757–767.
Pays O, Jarman P, Loisel P, Gerard JF. 2007. Coordination, independence or synchronisation tendency of individual vigilance in the
eastern grey kangaroo? Anim Behav. 73:595–604.
Pays O, Renaud PC, Loisel P, Petit M, Gerard JF, Jarman P. 2007. Prey
synchronize their vigilant behaviour with other group members.
Proc R Soc Lond B. 274:1287–1291.
Priestley MB. 1982. Spectral analysis and time series. London: Academic Press.
Pulliam HR. 1973. On the advantages of flocking. J Theor Biol.
38:419–422.
Quinn JL, Cresswell W. 2005. Escape response delays in wintering
redshank, Tringa totanus, flocks: perceptual limits and economic
decisions. Anim Behav. 69:1285–1292.
477
Quirici V, Castro RA, Oyarzun J, Ebensperger LA. 2008. Female degus
(Octodon degus) monitor their environment while foraging socially.
Anim Cogn. 11:441–448.
R Development Core Team. 2007. R: a language and environment for
statistical computing. Vienna, (Austria): R Foundation for Statistical
Computing. Available from: http://www.r-project.org.
Robinette RL, Ha JC. 2001. Social and ecological factors influencing
vigilance in northwestern crows, Corvus caurinus. Anim Behav. 62:
447–452.
Roberts G. 1996. Why individual vigilance declines as group size increases. Anim Behav. 51:1077–1086.
Robinette RL, Ha JC. 2001. Social and ecological factors influencing
vigilance in northwestern crows, Corvus caurinus. Anim Behav.
62:447–452.
Ruxton GD. 1996. Group size and anti-predator vigilance: a simple
model requiring limited monitoring of other group members.
Anim Behav. 51:478–481.
Sirot E. 2006. Social information, antipredatory vigilance and flight in
bird flocks. Anim Behav. 72:373–382.
Sirot E, Touzalin F. 2009. Coordination and synchronization of
vigilance in groups of prey: the role of collective detection and
predators’ preference for stragglers. Am Nat. 173:47–59.
Tchabovsky AV, Popov SV, Krasnov BR. 2001. Intra- and interspecific
variation in vigilance and foraging of two gerbillid rodents, Rhombomys opimus and Psammomys obesus: the effects of social environment.
Anim Behav. 62:965–972.
Treves A. 2000. Theory and method in studies of vigilance and aggregation. Anim Behav. 60:711–722.
Treves A, Drescher A, Ingrisano N. 2001. Vigilance and aggregation in
black howler monkeys (Alouatta pigra). Behav Ecol Sociobiol.
50:90–95.
Ward P. 1985. Why birds in flocks do not coordinate their vigilance
periods. J Theor Biol. 114:383–385.