Influence of food abundance on aggression in juvenile convict cichlids

Abstract

Optimality models designed to explain the occurrence of feeding territoriality predict that the frequency or intensity of aggression will peak at intermediate levels of food abundance. To test whether this prediction applies to the competition for ephemeral patches of food, we manipulated food abundance over a broad range of values in two separate experiments (24- and 64-fold, respectively) while monitoring the aggressive behaviour of juvenile convict cichlids,Archocentrus nigrofasciatum , competing for the food. In both experiments, the rate of aggression was low when food was scarce, increased as food abundance increased, and decreased when food was provided in excess. This dome-shaped pattern of aggression was caused partly by higher encounter rates between fish and partly by a higher proportion of encounters resulting in aggression, when food was at intermediate levels of abundance. Our results suggest that convict cichlids display behavioural flexibility: in response to changes in food abundance, they appear to change both their likelihood of using aggression when encountering a conspecific and their willingness to enter an occupied patch.

Full text

ANIMAL BEHAVIOUR, 2002, 63, 323–330
doi:10.1006/anbe.2001.1891, available online at http://www.idealibrary.com on
Influence of food abundance on competitive aggression in
juvenile convict cichlids
JAMES W. A. GRANT, ISABELLE L. GIRARD, CINDY BREAU & LAURA K. WEIR
Department of Biology, Concordia University
(Received 20 October 2000; initial acceptance 20 December 2000;
final acceptance 11 June 2001; MS. number: A8913R)
Optimality models designed to explain the occurrence of feeding territoriality predict that the frequency
or intensity of aggression will peak at intermediate levels of food abundance. To test whether this
prediction applies to the competition for ephemeral patches of food, we manipulated food abundance
over a broad range of values in two separate experiments (24- and 64-fold, respectively) while monitoring
the aggressive behaviour of juvenile convict cichlids, Archocentrus nigrofasciatum, competing for the food.
In both experiments, the rate of aggression was low when food was scarce, increased as food abundance
increased, and decreased when food was provided in excess. This dome-shaped pattern of aggression was
caused partly by higher encounter rates between fish and partly by a higher proportion of encounters
resulting in aggression, when food was at intermediate levels of abundance. Our results suggest that
convict cichlids display behavioural flexibility: in response to changes in food abundance, they appear to
change both their likelihood of using aggression when encountering a conspecific and their willingness
to enter an occupied patch.
2002 The Association for the Study of Animal Behaviour
The link between resource abundance and whether or
not animals defend territories has been clear ever since
Brown (1964) introduced the concept of economic
defendability. Optimality models, which are largely ver-
bal (but see Carpenter & MacMillen 1976), predict the
occurrence of feeding territories between a lower and
upper threshold of food abundance (Warner 1980;Myers
et al. 1981;Carpenter 1987). Below the lower threshold,
food is presumably too scarce to meet the energetic costs
of defence, so that a territorial individual would have
lower fitness than an individual competing via scramble
competition. Above the upper threshold, nonterritorial
individuals can acquire the same amount of food as
territorial individuals without paying the costs of
defence. Hence, territorial individuals will have higher
fitness than nonterritorial individuals only between the
lower and upper thresholds of food abundance. Field
observations of nectar-feeding birds (e.g. Gill & Wolf
1975;Carpenter & MacMillen 1976;Carpenter 1987),
pied wagtails, Motacilla alba (Davies & Houston
1984), and water striders, Aquarius remigis (Wilcox &
Ruckdeschel 1982) support the predictions of the
threshold model of territoriality.
The original threshold model has been modified in two
important ways. First, most animals do not defend terri-
tories in an all-or-none fashion (Wolf 1978;Wittenberger
1981;Craig & Douglas 1986). Rather, as the net benefits
of defence increase, defence typically varies continuously
from the infrequent, low-intensity, defence of a non-
exclusive area to the frequent, vigorous, defence of an
exclusive area (Carpenter & MacMillen 1976;Craig &
Douglas 1986). Hence, a continuum model of feeding
territoriality, a more realistic version of the threshold
model, predicts a dome-shaped relationship between the
frequency or intensity of territorial aggression and the
abundance of food (Wolf 1978;Grant & Noakes 1988;
Wyman & Hotaling 1988). Second, territoriality is but
one form of interference competition, which also
includes brief contests over a single unit of resource, the
guarding of ephemeral patches, and dominance hier-
archies (Archer 1988). Optimality models that were orig-
inally designed to explain the occurrence of territoriality
have been remarkably successful at predicting the pat-
terns of aggression during the competition for ephemeral
patches of food (Isbell 1991;Grant 1993). Hence, a
dome-shaped relationship between aggression and food
abundance may also apply to competition in a variety of
social foraging situations.
In contrast to this dome-shaped prediction, a recent
game theory model based on the hawk–dove game
Correspondence: J. W. A. Grant, Department of Biology, Concordia
University, 1455 de Maisonneuve Blvd West, Montre´al, Que´bec
H3G 1M8, Canada (email: grant@vax2.concordia.ca).
0003–3472/02/020323+08 $35.00/0 2002 The Association for the Study of Animal Behaviour323

(Parker 1984) predicts an inverse relationship between
aggression and food abundance, with no decrease in
aggression when food is scarce (Sirot 2000; also see Doyle
& Talbot 1986). Two key assumptions of the game theory
model, neither of which is explicitly made by the verbal
optimality models, may explain its different prediction.
First, the hawk–dove model assumes that interactions
only occur when individuals are contesting a food item.
In spatially predictable environments (sensu Grand &
Grant 1994), however, animals often defend patches
when food is not present, in anticipation of the arrival of
the next food item (e.g. Grant & Kramer 1992;Grand &
Grant 1994). Interactions when food is absent will lower
the fitness of hawks relative to doves by decreasing the
hawk’s energy gain from a food item (i.e. V); a decrease in
Vcauses the evolutionarily stable strategy (ESS) prob-
ability of playing hawk to decrease (Sirot 2000). As food
abundance decreases, the proportion of interactions that
occur in the absence of food will probably increase.
Therefore, the probability of playing hawk may indeed
decline when food is scarce if animals interact in the
absence of food. Second, the hawk–dove model assumes
that all interactions are between two contestants; hawk-
like behaviour is an ESS whenever the value of the
resource is greater than the cost of injury (Parker 1984).
The effectiveness of aggression, however, decreases as
more than two individuals compete for each resource
unit (Grant et al. 2000). When food is scarce, many
individuals may congregate at the best patches, making
aggression uneconomical. Hence, relaxation of these
two assumptions of the hawk–dove model may produce
predictions similar to those of the optimality model.
We are not aware of any study that has documented a
dome-shaped relationship between aggression and food
abundance in a social foraging situation, perhaps because
no single experimental study has provided a sufficiently
broad range of food abundance. While there is much
evidence of a decrease in aggression when food becomes
abundant (e.g. Magnuson 1962;Armstrong 1991;
Kotrschal et al. 1993;Gregory & Wood 1999), there is less
evidence of a decrease in aggression as food density
becomes scarce (but see Ewald 1980;Davis & Olla 1987;
Powers 1987), and no study has shown both results in a
single experiment. Because of both the theoretical and
empirical uncertainty, we tested whether a dome-shaped
relationship exists between the rate of competitive aggres-
sion and food abundance in a social foraging situation.
To test for the generality of this pattern, we manipulated
food abundance over a broad range of availability (24-
and 64-fold, respectively) in two separate experiments
in which food was presented in different ways (i.e.
continuous input versus depleting).
We used convict cichlids, Archocentrus nigrofasciatum,as
our test species because they readily defend ephemeral
concentrations of food in the laboratory (e.g. Grant &
Guha 1993;Grand & Grant 1994;Praw & Grant 1999). In
the wild they defend nesting sites and offspring, but not
food (Wisenden 1995), like many other species of fresh-
water fish (Barlow 1993). Presumably, food is often
economically defendable in the laboratory, where it is
often presented in a spatially clumped and predictable
manner, but is rarely economically defendable in the
wild.
METHODS
General
Approximately 400 juvenile convict cichlids were held
in 15 stock tanks. These fish were descendants of crosses
between laboratory individuals and wild individuals from
Costa Rica. Groups of three or five fish were formed by
haphazardly selecting fish from the stock tanks and
transferring them to an experimental tank (see below).
Each stock and experimental tank contained an under-
gravel filter, a heater, gravel to a depth of 3 cm, and aged
tap water maintained at 27–29 C. The back and sides
of experimental tanks were covered with white paper
to minimize disturbance to the fish and to provide a
background for videotaping.
As in previous experiments with convict cichlids in
social foraging situations (e.g. Grant & Guha 1993;Grand
& Grant 1994;Praw & Grant 1999), almost all aggressive
interactions were chases, short unidirectional bursts of
increased swimming directed at another individual.
Dominance rank is strongly related to body size in con-
vict cichlids (Keeley & Grant 1993), so that the largest fish
typically defends the food patch and initiates over 90% of
the aggressive interactions (Grant & Guha 1993;Praw &
Grant 1999). Smaller fish adopt nonaggressive roles and
attempt to sneak into the patch to forage (Praw & Grant
1999). While the chase rate by the dominant and the
total chase rate will be very similar, we used the latter as
our measure of aggression in a trial because our focus was
on competitive aggression rather than territoriality per se.
Rate of aggression, however, is not always a good measure
of an animal’s aggressiveness because it ignores an indi-
vidual’s opportunity to engage in aggression (see Robb &
Grant 1998). Hence, we used the rate of encounters
between fish, defined as whenever a fish came to within
two body lengths of another fish, as a measure of the
opportunity for fish to engage in aggression, and the
proportion of encounters resulting in chases (hereafter
chases per encounter) as a measure of aggressiveness.
Continuous Input Experiment
We formed groups of three fish so that body weight
differed by less than 30% within a group. We used a
total of 105 fish in the experiment (mean body
weightSD =3.317 1.173 g). Experimental tanks,
measuring 6131.5 41.5 cm (lwh), were covered
with a black Plexiglas sheet, into which three holes had
been drilled, 25 cm apart, along the midline of the tank.
The middle hole was positioned over the centre of the
tank. We inserted a funnel into each hole so that its tip
protruded 2 cm into the water. We fed fish by dropping
food pellets into the funnels; uneaten pellets sank to the
bottom and were collected in small ice cube trays (23
cells) positioned under each funnel. Fish ate pellets from
324 ANIMAL BEHAVIOUR, 63, 2

the trays very infrequently and primarily after they
were satiated and pellets had begun to accumulate in the
trays.
We randomly assigned the groups to one of five levels
of food abundance: we fed fish 15, 30, 60, 120 or 360
pellets during a 30-min period, once per day. We dropped
one pellet of Fry Feed Kyowa (proximate composition:
crude protein not less than 55%, crude fat not less than
10%, crude fibre not more than 4%, and crude ash not
more than 17%) at a time into one of the three funnels,
chosen randomly for each food item, every 2 min, 1 min,
30 s, 15 s or 5 s, respectively. From preliminary work, we
knew that 120 pellets (1000 m in diameter), weighing
0.15 g, just about satiated three, 2.4-g fish in a 30-min
period. Hence, our levels of food abundance ranged from
about 12.5 to 300% of what the fish could eat. When the
mean fish weight was greater than 2.4 g, we adjusted the
total weight of food in direct proportion to fish weight.
For example, a group of fish with a mean weight of 5 g,
would receive 0.31 g of food in the 120-pellet treatment
(i.e. 5/2.40.15 g =0.31 g). To keep the number of pellets
constant, we used a mix of small (1000 m) and large
(2000 m) pellets to achieve the desired weight. The fish
received no other food while being used in experimental
trials. Because the amount of food required to satiate a
fish typically scales as M
0.75
(Elliott 1975), our method of
adjusting the ration may have been slightly biased such
that larger fish may have been more satiated than smaller
fish at a particular treatment. Because no such bias was
evident when we included body mass as a covariate in the
analyses (all Pvalues n0.05), it was deleted from the
analyses described below.
Prior to each feeding period, we turned off the air pump
and placed the funnels in the holes. The fish quickly
learned to associate these procedures with the arrival of
food. Each group of fish experienced the same treatment
once per day for 5 days. On days 6 and 7, we videotaped
the 30-min feeding trial from directly in front of the
tank. On day 8, we weighed the fish and transferred them
to a holding tank for ‘used fish’. We completed seven
replicates for each treatment.
From the videotapes, we counted the number of chases,
encounters with conspecifics, and the food pellets eaten
and rejected. A pellet was considered rejected if the fish
ate then spat out the pellet within 10 s, or if the fish
ignored a pellet that was clearly visible. The time of
satiation, which occurred only in the 60-, 120- and
360-pellet treatments, was defined as the minute of the
trial in which the fish first rejected at least 75% of the
pellets. We summed the behaviour of the three fish for
each recording, and used the mean of the two recordings
as a datum in our analyses. We recorded the position of
each fish during 10 scan samples that were spaced evenly
over the portion of the trial in which fish were actively
feeding
To meet the assumptions of parametric tests, we used
log
10
transformation of the chase and encounter rate data
and used angular transformation for the proportion of
chases per encounter. For graphical purposes, we plotted
the untransformed data for the percentage of chases per
encounter. Because adjacent food levels increased in a
geometric series, we also used log
10
transformation of the
number of pellets before fitting regressions.
Depleting Patch Experiment
We formed groups of five fish so that the smallest
fish was about half the weight of the largest. We used a
total of 120 fish in the experiment (mean body weight
SD= 1.7950.437). We transferred each group to an
experimental tank, measuring 9046 38 cm (lw
h), and randomly assigned each group to one of six
treatments: we fed fish 12, 48, 96, 192, 364 or 768 Kyowa
(1000 m) pellets per day.
We presented the food in a patch composed of two ice
cube trays (62 cells each) attached to form a 6 4 cell
patch, measuring 23.522.6 cm. We spread the required
pellets evenly among the 24 cells. We covered the patch
by an identical 64 cell patch to prevent the food from
floating out of the patch when it was initially submerged
in the tank. To prevent the patch from floating, we
pushed the patch and the cover into an identical 64
cell base that was attached to the undergravel filter. We
removed the cover at the start of the feeding trial.
We trained the fish to feed from the patch for one
30-min period/day for 5 days, after which they eagerly fed
from the patch. On days 6 and 7, we videotaped the
30-min trial from directly in front of the tank. At the end
of day 7, we transferred the fish into a tank for used fish.
In total, we completed four replicates for each level of
food abundance.
From the videotapes, we counted the number of chases
and encounters, as defined above. We judged the time
that all the food was eaten by when fish hovering near
the patch ceased feeding. An observer in front of the tank
confirmed this initial judgement visually. Satiation was
defined as the minute in which fish ceased feeding while
food was still present. We recorded the positions of the
dominant and subordinate fish during 10 scan samples
that were spaced evenly over the portion of the trial in
which fish were actively feeding. We also scored the
behaviour of the dominant fish (feeding on the patch,
chasing a fish off the patch, chasing a fish on the patch,
or on the patch but not otherwise engaged) on all occa-
sions (N=386) when a subordinate fish first entered the
patch, during the portion of the trials in which fish were
actively feeding.
As in the continuous input experiment, we used the
angular transformation for the proportion of chases per
encounter and log
10
transformation of the number of
pellets per patch. Unlike the first experiment, how-
ever, chase rate and encounter rate data did not require
transformation to meet the assumptions of parametric
tests.
RESULTS
Continuous Input Experiment
The dominant fish typically hovered under the central
funnel and kept the subordinate fish to the outside of the
325GRANT ET AL.: FOOD ABUNDANCE AND AGGRESSION

tank. On average, the dominant fish was 6.5 cm from the
midline of the tank (i.e. a line extending down from the
central funnel); its horizontal position did not change
significantly across treatments (analysis of variance,
ANOVA: F
4,30
=1.38, P= 0.26). The subordinate fish were
significantly farther from the midline of the tank
(mean= 19.1 cm) than the dominant fish (paired ttest:
t
34
=12.43, P<0.0001) and, like the dominants, their hori-
zontal position did not differ across treatments (ANOVA:
F
4,30
=1.02, P= 0.41). When not avoiding the dominant
fish, a subordinate fish typically hovered under each of
the side funnels.
Fish fed actively throughout the 15-, 30- and 60-pellet
trials, but were satiated in 10 of 14 (i.e. 7 replicates2
days of recording= 14 trials) 120-pellet trials and in all the
360-pellet trials. On average, the fish ate 15, 29.9, 58.1,
92.2 and 105.7 pellets in the 15-, 30-, 60-, 120- and
360-pellet treatments, respectively. The number of pellets
eaten in the 120- and 360-pellet treatments did not differ
significantly (two-sample ttest: t
12
=0.76, P= 0.46). Not
surprisingly, the average percentage of weight gained by
the three fish differed among treatments (ANOVA:
F
4,30
=13.51, P<0.0001). Fish lost weight in the 15- and
30-pellet treatments. Weight gain was positively corre-
lated with the amount of food eaten per day, expressed
as a percentage of body weight (Pearson correlation:
r
33
=0.90, P<0.0001).
In the 14 groups (2 treatments7 replicates) that
became satiated, we compared behaviour before and
after satiation. The most notable difference was that
the encounter rate among fish decreased significantly
after satiation (paired ttest: t
13
=2.76, P= 0.016). However,
the proportion of encounters resulting in chases (paired t
test: t
13
=1.85, P= 0.087) and chase rate did not differ
significantly before and after satiation (paired ttest:
t
13
=0.49, P= 0.63). Because of the temporal changes
in behaviour, we present data only for the portion of
the trials in which fish were actively feeding (i.e. not
satiated).
The largest fish in each tank initiated most of the
aggressive interactions. The total chase rate by all three
fish differed significantly among the levels of food abun-
dance (ANOVA: F
4,30
=4.84, P= 0.004). Rate of aggression
initially increased with increasing food abundance,
peaked in the 120-pellet treatment, and then decreased
(Fig. 1a). This apparent dome-shaped relationship
between aggression and food abundance was supported
by a significant quadratic regression (polynomial regres-
sion: F
2,32
=9.03, P= 0.0009; r
2
=0.37); the quadratic term
was significant (polynomial regression: partial F
1,32
=
16.28, P=0.0003), even after the linear term was entered
first in the model. In addition, the mean rate of aggres-
sion was higher in the 120-pellet treatment than in either
the 15- or 360-pellet treatments (Scheffe’s multiple
comparison test: Ps<0.05).
The number of encounters between fish, a measure of
their opportunity to engage in aggressive interactions, did
not differ significantly among the levels of food abun-
dance (Fig. 1b; ANOVA: F
4,30
=0.70, P= 0.60). In addition,
there was no evidence of a dome-shaped relationship
between encounter rate and food abundance (polynomial
regression, partial Ftest for quadratic term: F
1,32
=2.15,
P=0.15).
The proportion of encounters that resulted in aggres-
sive chases also did not differ significantly among the
levels of food abundance (Fig. 1c; ANOVA: F
4,30
=1.72,
P=0.17). However, there was weak evidence of a dome-
shaped relationship between chases per encounter and
food abundance (Polynomial regression, partial Ftest for
the quadratic term: F
1,32
=5.65, P= 0.024). The overall
0.2
0.7
Pellets/30 min
(c)
15
0.3
0.4
0.5
0.6
30 60 120
Chases/encounter (proportion)
360
10
(b)
15
5
6
7
8
30 60 120
Encounters/min
360
1
(a)
15
2
4
30 60 120
Chases/min
360
9
4
Figure 1. Effect of food abundance on mean (±SE, N= 7) (a) number
of aggressive interactions/min, (b) number of encounters, times a
fish approached to within two body lengths of another fish, and
(c) proportion of encounters resulting in aggressive interactions.
Note the logarithmic scale on all axes except the ordinate of (c).
The equation of the quadratic regression in (a) is: log
10
chase
rate= 2.47×log
10
pellets−0.69 ×(log
10
pellets)
2
−1.64.
326 ANIMAL BEHAVIOUR, 63, 2

quadratic regression was not significant (polynomial
regression: F
2,32
=2.93, P= 0.068).
Depleting Patch Experiment
On average, the dominant fish hovered directly over
the patch (i.e. within two body lengths) 94% of the time
(N=240 scan samples); this patch fidelity did not differ
across treatments (F
5,18
=2.47, P= 0.07). All fish usually
rushed into the patch when the lid was removed and
scrambled for food for the first 10–30 s, until chased away
by the dominant fish. The subordinate fish then hovered
off the edge of the patch; at least one subordinate fish
was found over the patch 17% of the time (N=240).
Subordinate fish were observed entering the food patch a
total of 386 times during the portion of trials in which
fish were actively feeding. They typically entered the
patch only when the dominant was busy foraging (52%
of occasions) or chasing other fish, either off the patch
(13%) or on the patch (9%). Only 26% of all intrusions
by subordinate fish onto the patch occurred when the
dominant was present and not otherwise engaged.
The fish either quickly ate all the food in the patch or
became satiated long before the end of the 30-min
feeding trial. The portion of the trial in which the fish
were actively eating (i.e. until all the food was eaten or
until satiation) varied significantly among treatments
(ANOVA: F
5,18
=4.99, P= 0.0048); the average time of
active feeding was 1.8, 2.6, 4.9, 8.8, 4.0 and 4.4 min in
the 12-, 48-, 96-, 192-, 384- and 768-pellet treatments,
respectively. All the food was eaten in the 12-, 48- and
96-pellet treatments, whereas all fish were satiated in the
384- and 768-pellet treatments. In the 192-pellet treat-
ment, one group of fish became satiated whereas three
groups ate all the food. Because the behaviour of the fish
changed when the fish were not actively feeding (see
below), we present separate analyses for the portions of
the trials in which the fish were actively versus not
actively feeding.
When fish were actively feeding, the chase rate dif-
fered significantly among the levels of food abun-
dance (ANOVA: F
5,18
=3.27, P= 0.028). Rate of aggression
initially increased with increasing food abundance,
peaked in the 96-pellet treatment and then decreased
(Fig. 2a). This apparent dome-shaped relationship was
supported by a significant difference in aggression
between the 12- and 96-pellet treatments and between
the 96- and 768-pellet treatments (Scheffe’s multiple
comparison test: Ps<0.05). In addition, even though an
overall quadratic regression was not significant (polyno-
mial regression: F
2,21
=2.70, P= 0.09), the quadratic term
in the equation was significant when entered after the
effect of the linear term (polynomial regression: partial
F
1,21
=4.81, P= 0.04). Even though the chase rate
decreased significantly when the fish ceased feeding
(paired ttest: t
23
=3.29, P= 0.003), a similar pattern of
aggression in relation to food abundance was observed
in the nonfeeding portion of the trials (not shown).
While the differences among the treatments were not
significant when the fish were not actively feeding
(ANOVA: F
5,18
=2.33, P= 0.085), both the overall quadratic
regression (polynomial regression: F
2,21
=4.50, P= 0.024)
and the quadratic term in the regression (polynomial
regression: partial F
1,21
=6.44, P= 0.019) were significant.
When fish were actively feeding, the number of
encounters between fish varied with food abundance in a
manner similar to the chase data (Fig. 2b). While the
encounter rate did not differ significantly among treat-
ments (ANOVA: F
5,18
=2.33, P= 0.085), the peak occurred
in the 96-pellet treatment. The overall quadratic regres-
sion was not significant (F
2,21
=2.34, P= 0.12), but the
quadratic term was (F
1,21
=4.51, P= 0.046). As with the
40
100
Pellets/patch
(c)
12
50
60
70
80
48 96 192
Chases/encounter (%)
768
90
384
2
(b)
12
4
6
10
12
48 96 192
Encounters/min
768
14
384
0
14
(a)
12
4
6
8
10
48 96 192
Chases/min
768
12
384
16
8
2
Figure 2. Effect of food abundance on mean (±SE, N= 4) (a) number
of aggressive interactions/min, (b) number of encounters, times a
fish approached to within two body lengths of another fish, and (c)
proportion of encounters resulting in aggressive interactions. Note
the logarithmic scale on the Xaxes.
327GRANT ET AL.: FOOD ABUNDANCE AND AGGRESSION

chase data, the encounter rate decreased significantly
after the fish ceased feeding (paired ttest: t
23
=4.12,
P=0.0001). However, the encounter rate still showed
a dome-shaped relationship with food abundance.
Encounter rate differed among treatments (ANOVA:
F
5,18
=2.75, P= 0.052), except the peak shifted to the
192-pellet treatment, and both the overall quadratic
regression (F
2,21
=5.27, P= 0.014) and the quadratic term
in the regression (partial F
1,21
=7.91, P= 0.01) were
significant.
The percentage of encounters resulting in chases did
not differ significantly among treatments regardless of
whether the fish were actively feeding (Fig. 2c; ANOVA:
F
5,18
=1.05, P= 0.42) or not (ANOVA: F
5,18
=1.58, P= 0.22).
Curiously, chases per encounter increased significantly
when the fish ceased active feeding (paired ttest:
t
23
=4.55, P<0.001).
DISCUSSION
As predicted by the continuum model of territoriality, a
dome-shaped relationship between the frequency of
aggression and food abundance was evident in both
experiments. In the continuous input experiment, the
decrease in aggression when food was in excess could also
potentially be explained by the increase in the temporal
clumping of food arrival; resource monopolization and
aggression are typically inversely related to the temporal
clumping of resource arrival (Blanckenhorn 1991;Grant
& Kramer 1992;Grant et al. 1995;Bryant & Grant 1995).
This alternate explanation could not, however, explain
the decrease in aggression when food was scarce or the
dome-shaped relationship that was also observed in the
depleting patch experiment.
Our data were not consistent with a modified hawk–
dove model that predicts an inverse relationship between
aggression rate and food abundance (Sirot 2000). This
apparent falsification may be due to a mismatch between
our experimental conditions and the assumptions of
Sirot’s (2000) model. We manipulated the value of the
single patch in our experiments. Sirot (2000) predicts that
the level of aggression will increase as the value of the
patch or food item (V) increases (i.e. the left side of the
dome-shaped relationship). When a patch is sufficiently
rich to satiate all competitors, its value will decrease
because Vis the opportunity cost to find an equivalent
resource, not the energetic value of the patch (Parker
1984). Hence, hawk-like behaviour should decrease
when food is in excess because doves can wait to exploit
the patch without paying the cost of injury. Therefore,
the hawk–dove model may also predict a dome-shaped
relationship between the level of aggression and the
amount of food in a patch.
Both optimality and game theory models assume that
individuals can change their behaviour in response to
environmental change; both predict that the probability
of an aggressive response to conspecifics will decrease as
food abundance becomes superabundant. Unfortunately,
documenting this individual flexibility in aggressiveness
is difficult because the same behavioural rule can produce
different amounts of aggression in different environ-
ments. For example, the commonly observed increase in
aggression when food becomes more clumped in space
(Zahavi 1971;Monaghan & Metcalfe 1985;Grant & Guha
1993) could be due entirely to a fixed rule, such as chase
50% of conspecifics, together with an increased encoun-
ter rate with conspecifics attracted to the clumped
resource (Grant & Guha 1993). Hence, chases per encoun-
ter may provide stronger evidence of behavioural flexi-
bility than frequency of aggression (see Jones 1983;Grant
& Kramer 1992;Robb & Grant 1998).
We have some evidence that the dome-shaped relation-
ship in the amount of aggression was the result of
behavioural flexibility. In the continuous input exper-
iment, the quadratic relationship between chases per
encounter and food abundance suggests that fish actually
modified their aggressive behaviour to conspecifics in
response to food abundance. In particular, the percentage
of encounters resulting in aggression decreased from
about 60% in the 120-pellet treatment to 37% when food
was in excess. There was no such evidence in the deplet-
ing patch experiment. In the depleting patch experiment,
however, the dome-shaped pattern of aggression was
partly driven by encounter rate between fish. When food
was scarce and in excess, the encounter rate was low, as if
the fish were ‘time-sharing’ the patch; that is, the subor-
dinates only entered the patch when the dominant fish
was busy foraging or chasing others. Entering an occu-
pied patch and risking an aggressive interaction may not
have been economically feasible when food was scarce or
in excess. When food was at intermediate levels, encoun-
ter rate with conspecifics increased because subordinate
fish were more likely to enter the already occupied patch.
Hence, behavioural flexibility in the depleting patch
experiment may have been primarily driven by the will-
ingness of the subordinate fish to enter the patch rather
than by the percentage of encounters in which large fish
chased small fish. These data imply that the fish learned
to adjust their aggressive behaviour or willingness to
enter a patch in only 5 days.
The best evidence of individual flexibility in aggressive-
ness in response to changes in food abundance occurs
when food is provided in excess; animals typically either
decrease the intensity of their aggressive displays or
sometimes cease territoriality entirely (Wilcox &
Ruckdeschel 1982;Hart 1987;Carpenter 1987;Kotrschal
et al. 1993). Perhaps the best evidence of a lower
threshold of food abundance is Ewald’s (1980) obser-
vation that Anna’s hummingbirds, Calypte anna, only
defend feeders that provide more than 13% of their daily
energy requirement. Anna’s hummingbirds also show a
continuous increase in aggressiveness with increasing
food abundance; the percentage of intruders chased and
the intensity of aggressive responses increased with
increasing food abundance (Ewald & Carpenter 1978;
Powers 1987). In contrast to the paucity of studies docu-
menting an increase in aggressiveness with food abun-
dance, fish and birds have often been reported to increase
their parental defence effort in response to an increase
in brood size (e.g. Carlisle 1985;Montgomerie &
Weatherhead 1988).
328 ANIMAL BEHAVIOUR, 63, 2

Differences in aggressiveness between our two experi-
ments were evident; an encounter resulted in a chase
only about 50% of the time in the continuous input
experiment, but 75% of the time in the depleting patch
experiment. The lower level of aggressiveness in the
continuous input experiment may reflect the conflict in
time that occurs between eating a food pellet, while it was
briefly available, versus chasing a competitor. In the
depleting patch experiment, however, a food item
remained in the patch to be eaten after a chase, unless
eaten first by a conspecific. Hence, the optimal level of
aggressiveness may have been higher in the depleting
patch experiment, because the costs of defence (i.e. lost
feeding opportunities) were lower.
In summary, our study provides evidence of a dome-
shaped relationship between frequency of aggression and
food abundance. At least part of these changes in fre-
quency of aggression seems to be the result of behavioural
flexibility.
Acknowledgments
We thank Miles Keenleyside for providing the initial
stock of fish, Jason Praw, Stacey Robb and Howard Levitt
for help in the laboratory, and Istva´n Imre, Cheryl
Johnson, Stefa´n Steingri´msson, Jeff Lucas and two anony-
mous referes for comments on the manuscript. This
research was financially supported by the Natural
Sciences and Engineering Research Council of Canada via
a Research Grant to J.W.A.G. and an Undergraduate
Student Research Award to C.B.
References
Archer, J. 1988. The Behavioural Biology of Aggression. Cambridge:
Cambridge University Press.
Armstrong, D. P. 1991. Aggressiveness of breeding territorial
honeyeaters corresponds to seasonal changes in nectar
availability. Behavioral Ecology and Sociobiology,29, 103–111.
Barlow, G. W. 1993. The puzzling paucity of feeding territories
among freshwater fishes. In: Behavioural Ecology of Fishes (Ed. by
F. A. Huntingford & P. Torricelli), pp. 155–174. Chur: Harwood.
Blanckenhorn, W. U. 1991. Foraging in groups of water striders:
effects of variability in handling time and prey arrivals. Behavioral
Ecology and Sociobiology,28, 221–226.
Brown, J. L. 1964. The evolution of diversity in avian territorial
systems. Wilson Bulletin,76, 160–169.
Bryant, M. J. & Grant, J. W. A. 1995. Resource defence, monopol-
ization and variation of fitness in groups of female Japanese
medaka depend on the synchrony of food arrival. Animal
Behaviour,49, 1469–1479.
Carlisle, T. 1985. Parental response to brood size in a cichlid fish.
Animal Behaviour,33, 234–238.
Carpenter, F. L. 1987. Food abundance and territoriality: to defend
or not to defend? American Zoologist,27, 387–399.
Carpenter, F. L. & MacMillen, R. E. 1976. Threshold model of
feeding territoriality and test with a Hawaiian honeycreeper.
Science,194, 639–642.
Craig, J. L. & Douglas, M. E. 1986. Resource distribution, aggressive
asymmetries and variable access to resources in the nectar feeding
bellbird. Behavioral Ecology and Sociobiology,18, 231–240.
Davies, N. B. & Houston, A. I. 1984. Territory economics. In:
Behavioural Ecology. 2nd edn (Ed. by J. R. Krebs & N. B. Davies),
pp. 148–169. Sunderland, Massachusetts: Sinauer.
Davis, M. W. & Olla, B. L. 1987. Aggression and variation in growth
of chum salmon (Oncorhynchus keta) juveniles in seawater: effects
of limited rations. Canadian Journal of Fisheries and Aquatic
Sciences,44, 192–197.
Doyle,R.W.&Talbot,A.J.1986. Artificial selection on growth and
correlated selection on competitive behaviour in fish. Canadian
Journal of Fisheries and Aquatic Sciences,43, 1059–1064.
Elliott, J. M. 1975. Weight of food and time required to satiate
brown trout, Salmo trutta L. Journal of Fish Biology,5, 51–64.
Ewald, P. W. 1980. Energetics of resource defense: an experimental
approach. Proceedings of the 17th International Ornithological
Congress,2, 1093–1099.
Ewald, P. W. & Carpenter, F. L. 1978. Territorial responses to
energy manipulations in the Anna hummingbird. Oecologia,31,
277–292.
Gill, F. B. & Wolf, L. L. 1975. Economics of feeding territoriality in
the golden-winged sunbird. Ecology,56, 333–345.
Grand, T. C. & Grant, J. W. A. 1994. Spatial predictability of food
influences its monopolization and defense by juvenile convict
cichlids. Animal Behaviour,47, 91–100.
Grant,J.W.A.1993. Whether or not to defend? The influence of
resource distribution. Marine Behavior and Physiology,23, 137–
153.
Grant, J. W. A. & Guha, R. T. 1993. Spatial clumping of food
increases its monopolization and defense by convict cichlids,
Cichlasoma nigrofasciatum.Behavioral Ecology,4, 293–296.
Grant,J.W.A.&Kramer,D.L.1992. Temporal clumping of food
arrival reduces its monopolization and defense by zebrafish,
Brachydanio rerio.Animal Behavior,44, 101–110.
Grant,J.W.A.&Noakes,D.L.G.1988. Aggressiveness and
foraging mode of young-of-the-year brook charr, Salvelinus fonti-
nalis (Pisces, Salmonidae). Behavioral Ecology and Sociobiology,22,
435–445.
Grant, J. W. A., Bryant, M. J. & Soos, C. E. 1995. Operational sex
ratio, mediated by synchrony of female arrival, alters the variance
of male mating success in Japanese medaka. Animal Behaviour,49,
367–375.
Grant, J. W. A., Gaboury, C. L. & Levitt, H. L. 2000. Competitor-
to-resource ratio, a general formulation of operational sex ratio, as
a predictor of competitive aggression in Japanese medaka (Pisces:
Oryziidae). Behavioral Ecology,11, 670–675.
Gregory, T. R. & Wood, C. M. 1999. Interactions between individ-
ual feeding behaviour, growth, and swimming performance in
juvenile rainbow trout (Oncorhynchus mykiss) fed different rations.
Canadian Journal of Fisheries and Aquatic Sciences,56, 479–486.
Hart, D. D. 1987. The adaptive significance of territoriality in
filter-feeding larval blackflies (Diptera: Simuliidae). Oikos,46,
88–92.
Isbell, L. A. 1991. Contest and scramble competition: patterns of
female aggression and ranging behavior among primates.
Behavioral Ecology,2, 143–155.
Jones, G. P. 1983. Relationship between density and behaviour
in juvenile Pseudolabrus celiodotus (Pisces: Labridae). Animal Behav-
iour,31, 729–735.
Keeley, E. R. & Grant, J. W. A. 1993. Asymmetries in the expected
value of food do not predict the outcome of contests between
convict cichlids. Animal Behaviour,45, 1035–1037.
Kotrschal, K., Hemetsberger, J. & Dittami, J. 1993. Food exploita-
tion by a winter flock of greylag geese: behavioral dynamics,
competition and social status. Behavioral Ecology and Sociobiology,
33, 289–295.
Magnuson, J. J. 1962. An analysis of aggressive behavior, growth,
and competition for food and space in medaka (Oryzias latipes).
Canadian Journal of Zoology,40, 313–363.
329GRANT ET AL.: FOOD ABUNDANCE AND AGGRESSION

Monaghan, P. & Metcalfe, N. B. 1985. Group foraging in wild
brown hares: effects of resource distribution and social status.
Animal Behaviour,33, 993–999.
Montgomerie, R. D. & Weatherhead, P. J. 1988. Risks and rewards
of nest defence by parent birds. Quarterly Review of Biology,63,
167–187.
Myers, J. P., Connors, P. G. & Pitelka, F. A. 1981. Optimal territory
size and the sanderling: compromises in a variable environment.
In: Foraging Behavior (Ed. by A. C. Kamil & T. D. Sargent),
pp. 135–158. New York: Garland.
Parker, G. A. 1984. Evolutionarily stable strategies. In: Behavioural
Ecology (Ed. by J. R. Krebs & N. B. Davies), pp. 30–61. Sunderland,
Massachusetts: Sinauer.
Powers, D. R. 1987. Effects of food quality on the breeding
territoriality of the male Anna’s hummingbird. Condor,89, 103–
111.
Praw, J. C. & Grant, J. W. A. 1999. Optimal territory size in the
convict cichlid. Behaviour,136, 1347–1363.
Robb, S. E. & Grant, J. W. A. 1998. Interactions between the spatial
and temporal clumping of food affect the intensity of aggression
in Japanese medaka. Animal Behaviour,56, 29–34.
Sirot, E. 2000. An evolutionarily stable strategy for aggressiveness in
feeding groups. Behavioral Ecology,11, 351–356.
Warner, R. R. 1980. The coevolution of behavioral and lifehistory
characteristics. In: Sociobiology: Beyond Nature/Nurture? (Ed. by
G. W. Barlow & J. Silverberg), pp. 151–188. Boulder, Colorado:
Westview Press.
Wilcox, R. S. & Ruckdeschel, T. 1982. Food threshold territoriality
in a water strider (Gerris remigis). Behavioral Ecology and
Sociobiology,11, 85–90.
Wisenden, B. D. 1995. Reproductive behaviour of free-ranging
convict cichlids, Cichlasoma nigrofasciatum.Environmental Biology
of Fishes,43, 121–134.
Wittenberger, J. F. 1981. Animal Social Behavior. Boston: Duxbury
Press.
Wolf, L. L. 1978. Aggressive social organization in nectarivorous
birds. American Zoologist,18, 765–778.
Wyman, R. L. & Hotaling, L. 1988. A test of the model of economic
defendability of a resource and territoriality using young Etropus
maculatus and Pelmatochromis subocellatus kribensis.Environmental
Biology of Fishes,21, 69–76.
Zahavi, A. 1971. The social behaviour of the white wagtail Motacilla
alba alba wintering in Israel. Ibis,113, 203–211.
330 ANIMAL BEHAVIOUR, 63, 2