Content uploaded by Craig A. Tovey
Author content
All content in this area was uploaded by Craig A. Tovey
Content may be subject to copyright.
Individual differences versus social dynamics in the
formation of animal dominance hierarchies
Ivan D. Chase*
†‡
, Craig Tovey
§
, Debra Spangler-Martin
¶
, and Michael Manfredonia
¶
*Department of Sociology, State University of New York, Stony Brook, NY 11794-4356;
†
Department of Ecology and Evolution, State University of New York,
Stony Brook, NY 11794-5245;
§
School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0205; and
¶
Division of
Biological Sciences, State University of New York, Stony Brook, NY 11794-5110
Communicated by A. Kimball Romney, University of California, Irvine, CA, February 21, 2002 (received for review February 28, 2001)
Linear hierarchies, the classical pecking-order structures, are
formed readily in both nature and the laboratory in a great range
of species including humans. However, the probability of getting
linear structures by chance alone is quite low. In this paper we
investigate the two hypotheses that are proposed most often to
explain linear hierarchies: they are predetermined by differences in
the attributes of animals, or they are produced by the dynamics of
social interaction, i.e., they are self-organizing. We evaluate these
hypotheses using cichlid fish as model animals, and although
differences in attributes play a significant part, we find that social
interaction is necessary for high proportions of groups with linear
hierarchies. Our results suggest that dominance hierarchy forma-
tion is a much richer and more complex phenomenon than previ-
ously thought, and we explore the implications of these results for
evolutionary biology, the social sciences, and the use of animal
models in understanding human social organization.
L
inear hierarchies, classic pecking-order structures, are
formed readily in nature and the laboratory by many species:
some insects and crustaceans and various fish, birds, and mam-
mals including human children and adolescents (1–10). How-
ever, the probability of generating linear hierarchies by chance
alone is low. We do not know how these social structures develop
their linear form, and even the types of mechanisms that might
produce linearity are controversial. In this paper we evaluate
hypotheses concerning the two most commonly proposed factors
for explaining the formation of linear hierarchies through a
series of experimental studies using cichlid fish.
Two individuals have a dominance relationship if one chases,
threatens, or bites, but receives little or no aggression, from the
other. Dominance hierarchies, known in the mathematical lit-
erature as tournaments, are social structures consisting of dom-
inance relationships between all pairs of individuals in a group.
In a linear hierarchy one individual dominates all the other
individuals in a group, a second dominates all but the first, and
so on down to the last individual who is dominated by all the
others. Dominance relationships in a linear hierarchy are always
transitive. For any three individuals (triad) in the group, if A
dominates B and B dominates C, then A also dominates C. If a
hierarchy is not linear, it contains at least one intransitive triad
(A dominates B, B dominates C, but C dominates A), and the
more intransitive triads there are, the further the hierarchy is
from linearity (by many measures of linearity). Perfectly linear
hierarchies are most common in groups under 10 members, and
as groups grow larger, irregularities may appear (11). Rank in
hierarchies influences such important things as behavior, phys-
iology, health, and ability to produce offspring (12–16).
The first and most often suggested hypothesis concerning the
mechanisms accounting for linearity is that individuals’ positions
in hierarchies are predetermined by differences in dominance
ability. We term this the ‘‘prior attributes’’ hypothesis. It pro-
poses that the ladder-like structure of linear hierarchies can be
explained by another, preexisting ladder-like structure, one on
which individuals about to form a hierarchy are ranked by
attributes indicative of their dominance ability. According to this
hypothesis, the animal highest in dominance attributes takes the
top position in the hierarchy, the animal second-highest takes the
next position, and so on.
General support for this hypothesis comes from the many
studies that demonstrate the association, sometimes quite high,
between various attributes of individuals and their positions in
hierarchies (2, 17, 18). The attributes that are correlated with
rank are varied sorts, depending on study and species, but age,
sex, physical size and strength, physiology, and level of aggres-
siveness are among the most common (12–16, 18). More spe-
cifically, some researchers have shown that in groups of three
animals with great discrepancies in prior attributes (e.g., A, a
recent winner and 30–40% larger than the others; B and C of
similar size, but B a recent winner; and C a recent loser),
individuals more often form hierarchies according to their rank
in attributes than expected by chance alone (19, 20). Other
researchers have argued that attribute differences ultimately
determine the rank order of individuals in hierarchies by dic-
tating the behavioral strategies used during hierarchy formation
(21, 22).
The second hypothesis is that processes of social interaction
among group members are the mechanisms that generate linear
hierarchies, and these processes are not predetermined by
differences in individuals’ attributes (23, 24). We term this the
‘‘social dynamics’’ hypothesis. Although researchers have not yet
demonstrated experimentally which specific dynamics actually
generate linear hierarchies, possibilities include (i) winner ef-
fects, with individuals winning earlier contests increasing their
probability of winning later ones (25, 26), (ii) loser effects, with
individuals losing earlier contests, increasing their probability of
losing later ones (25, 26), and (iii) bystander effects, with
individuals observing others’ encounters and adjusting their
behavior accordingly (27–29). In this hypothesis, if social inter-
action in a group context were prohibited, hierarchies should not
develop their usual linear structures. Thus the behavior that
occurs when groups are assembled would be central to explain-
ing the structure of hierarchies rather than being derivative. In
this case, dominance hierarchies would be ‘‘self-organizing’’ or
‘‘self-structuring’’ systems, the overall structures of which are
determined by interaction among the elements comprising the
systems (30–32).
Landau (33) and Chase (34) provided some initial support for
this hypothesis by demonstrating that stringent mathematical
conditions were required to generate highly linear hierarchies on
the basis of prior differences among individuals: extremely high
correlations between ranking on prior attributes and rank in
dominance hierarchies and highly skewed distributions for the
probabilities of winning encounters among the members of
groups. Examination of the relevant data indicated that such
conditions were rarely fulfilled.
In his ‘‘jigsaw puzzle’’ model, Chase (23) classified various
sequences by which dominance relationships could form in triads
of animals. Some of these sequences ensured the development of
‡
To whom reprint requests should be addressed. E-mail: Ichase@notes.cc.sunysb.edu.
5744–5749
兩
PNAS
兩
April 16, 2002
兩
vol. 99
兩
no. 8 www.pnas.org兾cgi兾doi兾10.1073兾pnas.082104199
transitive dominance relationships and thus the efficient pro-
duction of linear hierarchies, whereas others led to either
transitive or intransitive relationships and thus, possibly, non-
linear hierarchies. Researchers applying the model have found
high proportions of the sequences ensuring transitivity in a range
of species: chickens, rhesus monkeys, Japanese macaques, cichlid
fish, and crayfish (4, 5, 23, 35–38). Winner, loser, and bystander
effects may account for the high proportion of sequences
ensuring transitive relationships, and researchers using mathe-
matical models and computer simulation have demonstrated
that if these effects occur during hierarchy formation, they can
enhance the production of linear structures even when all
individuals are identical initially in prior attributes (39–42).
Methods
Study Species. We used female Metriaclima zebra (formerly
Pseudotropheus zebra), OB morph, as the species for our exper-
iments. This cichlid fish is native to east Africa, aggressive in
nature, and readily forms dominance hierarchies in the labora-
tory. We obtained the fish as surplus stock from the New York
Aquarium for Wildlife Conservation and from a commercial
breeder in Florida.
Experimental Design. In designing our experiments, we faced a
vexing problem. Previous research suggested that differences in
some relatively obvious attributes, chiefly weight, might be
important for predicting hierarchy ranks in fish (43). But what
about differences in undiscovered attributes, or if discovered,
attributes that could not be measured easily or could be mea-
sured only in invasive and upsetting ways that might change the
fish’s scores on the attributes measured? Thus, in our experi-
ments we decided to control for weight and to use an experi-
mental design that would allow us to assess the effects of all these
other potentially important attributes simultaneously: the un-
known and hard-to-measure ones. In the second experiment’s
Discussion, we consider the potential importance of weight
differences in the production of linear hierarchies.
To evaluate these undiscovered and difficult-to-measure at-
tributes, we designed two experiments. The first experiment
directly tested the prior attributes hypothesis. In it we assembled
groups of fish to form hierarchies, separated them long enough
to forget one another (44, 45), and then brought them back
together to form a second hierarchy. In this experiment, we
wanted to ‘‘rewind the tape’’ of each fish in a group to the
greatest extent possible and let the individuals form a second
hierarchy starting from scratch again. If rank on prior attributes
determined hierarchy rank in the first hierarchy, then rank on the
attributes should do so in the second hierarchy, because we
would avoid all possible things that might alter the fish’s rankings
on attributes. In other words, if rank on prior attributes ac-
counted for hierarchy ranks, we would expect identical rankings
for individuals in both hierarchies. If, however, rank on the prior
attributes did not determine hierarchical rank, the first and
second hierarchies should not be identical, and a considerable
proportion of fish should change ranks between them.
The second experiment directly contrasted whether the prior
attributes or social dynamics hypotheses could account better for
the high proportion of linear hierarchies commonly observed. In
this experiment we formed hierarchies by two methods: round-
robin competition and group assembly. In round-robin compe-
tition all the fish in a group met one another, but they did so only
in separate, pairwise encounters, out of sight of other fish, and
with 1 or 2 days between successive encounters. As a result,
interactional processes that might occur normally in a group
context were prohibited (e.g., observing others’ contests, attack-
ing others after a win, or being attacked oneself after a loss), but
the ability of individual qualities (‘‘biological,’’ ‘‘psychological,’’
or ‘‘sociological’’) to control the outcomes of these contests was
not inhibited. In group assembly, we placed all the fish in a group
in a tank simultaneously to establish their relationships in
whichever way they chose without interfering with their inter-
actional processes.
Experiment One
We took fish from their stock tanks, weighed them, and placed
them in separate isolation chambers, 21 liters in volume, for 2
weeks before they took part in a trial to remove any possible
effects of relationships that were established in their stock tanks
(44, 45). We made up groups of four fish allowing a 7%
maximum difference between the heaviest and lightest fish using
their weights at isolation. We fed all fish in a group identical
rations over the course of the experiment (⬇1.5% of body weight
per day). When the 2-week isolation period was complete, we
simultaneously transferred the fish in a group to a 76-liter
observation tank in which the individuals were separated by
partitions. We then removed the partitions and returned 24 h
later to observe from behind one-way mirrors. We recorded all
instances of nips, chases, and mouth fighting (46) and considered
that two fish had a stable relationship if one of the fish (the
dominant) delivered six aggressive acts, in any combination of
nips and chases, to the other without retaliation. Mouth fighting
was considered as a mutually aggressive act, and we began
recounting consecutive aggressive acts by either fish after an
incidence of this behavior. If after ⬇30 min of observation a
group had a stable hierarchy, i.e., all of the pairs had stable
relationships, we terminated observations. If all relationships
were not stable, we performed two or three more observations
that day and, if needed, the next day until relationships were
stable. After a stable hierarchy was achieved, we transferred the
fish back to their original, separate chambers for 2 more weeks
of isolation so that they would forget one another (44, 45).
Finally, we reassembled them to form a second hierarchy with
the same conditions and procedures as used for the first
hierarchy.
Results. Fig. 1 shows the first and second hierarchies for the 22
groups of fish we observed. Nearly all the groups formed linear
hierarchies both times they met (90.9 and 95.5%, respectively).
Fig. 1 also shows the transition patterns between the ranks of
individual fish in the first and second hierarchies, with all fish
keeping the same ranks in both hierarchies, the top two fish in
the first hierarchy swapping ranks in the second hierarchy, the
bottom two fish in the first hierarchy swapping ranks in the
second hierarchy, etc. Inspection indicates the great diversity in
the relationships between the two hierarchies. Only four tran-
sition patterns occurred in more than one group of fish, and 12
different patterns occurred, or 54.5%, out of the maximum of 22
possible if each group of fish had formed a different transition
pattern. This figure shows that it is difficult to achieve the same
hierarchy twice.
Table 1 shows the proportion of groups in which the two
hierarchies were identical, i.e., in which all fish had the same
ranks in both, as well as the proportions of groups in which two
(half) or more fish changed ranks. The percentage of identical
hierarchies (27.3) is significantly higher than would be expected
if second hierarchies were random linear orders, that is, if the
hierarchies were linear and the fish could take any rank (one-
sided binomial test: n ⫽ 22, P ⬍ 0.001). This is strong evidence
that prior attributes do affect hierarchy formation.
On the other hand, in order for the prior attributes hypothesis
in and of itself to be a robust explanation of linear hierarchies,
a high proportion of identical first and second hierarchies would
be required. If we set the standard for prior attributes to account
for linear structures, for example, at a moderate level of 75%
identical hierarchies or even at a low level of 50%, the 27% result
would be significantly smaller than either of these standards
Chase et al. PNAS
兩
April 16, 2002
兩
vol. 99
兩
no. 8
兩
5745
SOCIAL SCIENCES
(one-sided binomial test: n ⫽ 22, P ⬍ 0.001 and P ⬍ 0.03,
respectively). In this light, 27% of the groups with identical
hierarchies is very small.
Discussion. When we rewound the tape of the fish to form new
hierarchies, we usually did not get the same hierarchy twice. The
linearity of the structures persisted and the individuals stayed the
same, but their ranks did not. Thus our results differ considerably
from those predicted by the prior attributes hypothesis. The fact
that more identical hierarchies occurred than expected by chance
alone supports the hypothesis that rank on prior attributes
influences rank within hierarchies but not the hypothesis that
rank on prior attributes of itself creates the linear structure of the
hierarchies. Although 50% of the fish changed ranks from one
hierarchy to the other, almost all the hierarchies were linear in
structure. Some factor other than differences in attributes seems
to have ensured high rates of linearity. In the next experiment,
we tested to determine whether that factor might be social
dynamics.
It might seem possible that ‘‘noise,’’ random fluctuations in
individuals’ attributes or behaviors, could account for the ob-
served differences between the first and second hierarchies.
However, a careful consideration of the ways in which fluctua-
tions might occur shows that this explanation is unlikely. For
example, what if the differences were assumed to have occurred
because some of the fish changed their ranks on attributes from
the first to the second hierarchies? To account for our results,
this assumption would require a mixture of stability and insta-
bility in attribute ranks at just the right times and in just the right
proportion of groups. The rankings would have had to have been
stable for all the fish in all the groups for the day or two it took
them to form their first hierarchies (or we would not have seen
stable dominance relationships by our criterion). Then, in three-
quarters of the groups (but not in the remaining one-quarter)
various numbers of fish would have had to have swapped ranks
on attributes in the 2-week period of separation so as to have
produced different second hierarchies. And finally, the rankings
on attributes for all the fish in all the groups would have had to
have become stable once more for the day or two it took them
to form their second hierarchies.
Alternatively, instead of attribute rank determining domi-
nance rank as in the prior attribute model, dominance in pairs
of fish might be considered to have been probabilistic, such that
at one meeting one might dominate, but at a second meeting
there was some chance that the other might dominate. The
problem with this model is that earlier mathematical analysis
demonstrates that in situations in which one of each pair in a
group has even a small chance of dominating the other, the
probability of getting linear hierarchies is quite low (34). And
even in a more restrictive model in which only pairs of fish that
are close in rank in the first hierarchies have modest probabilities
of reversing their relationships, such as the level (0.25) we
observed in this experiment, the probability of getting as many
linear hierarchies as we observed is still very low (details are
available from the authors).
We know of only one other study (47) in which researchers
assembled groups to form initial hierarchies, separated the
individuals for a period, and then reassembled them to form a
second hierarchy (but see Guhl, ref. 48, for results in which
groups had pairwise encounters between assembly and reassem-
bly). Unfortunately, their techniques of analysis make it impos-
sible to compare results, because they examined correlations
between the frequency of aggressive acts directed by individuals
in pairs toward one another in the two hierarchies rather than
comparing the ranks of individuals. With these techniques it is
possible to get a positive correlation and thus a ‘‘replication’’ of
an original hierarchy in situations in which several animals
actually change ranks from the first to the second hierarchies.
Table 1. Percentage of groups with different numbers of fish
changing ranks between first and second hierarchies (n ⴝ 22)
No. of fish changing ranks Percentage of groups
0 27.3
2 36.4
3 18.2
4 18.2
Fig. 1. Transition patterns between ranks of fish in the first and second
hierarchies. Frequencies of experimental groups showing each pattern are
indicated in parentheses. Open-headed arrows indicate transitions of rank.
Solid-headed arrows show dominance relationships in intransitive triads; all
the fish in an intransitive triad share the same rank.
5746
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.082104199 Chase et al.
Experiment Two
We took fish from their stock tanks, weighed them, and made up
groups of four and five fish using weights at isolation; in a set of
four the largest was no more than 7% heavier than the smallest,
and in a set of five, no more than 9% heavier. After 2 weeks of
isolation, as in Experiment One, we formed hierarchies by
round-robin competition and group assembly.
In round-robin competition we randomly selected two pairs of
fish from a set for the first round of encounters. Each pair was
transferred to a 21-liter observation tank separated by a partition
and given2htoacclimate to the tank. We then removed the
partition and observed them through a one-way mirror. Again,
all instances of nips, chases, and mouth fighting were recorded
until one fish reached dominance over the other (a total of 15
consecutive, aggressive acts against the other without retalia-
tion). Of these 15 we scored only nips for the first seven acts and
any combination of nips and chases for the remaining eight. As
before, we considered mouth fighting as a mutually aggressive
act and began recounting consecutive acts by either fish after
such an incidence. When one fish reached the dominance
criterion, we separated them and returned them to their isolation
tanks.
We continued the rounds of encounters until all fish in a set
had met one another. Each fish in sets of four had 2 days between
encounters, and in sets of five each had 1 day except for the odd
fish out, which had 2 days. Where possible, we matched winners
to winners and losers to losers.
In group assembly we simultaneously transferred all fish in a
set from their isolation tanks to a 76-liter aquarium; observations
began 24 h later. We observed the fish and determined stable
dominance relationships and hierarchies using the same proce-
dures as described for experiment one.
We used six rather than 15 consecutive aggressive acts in group
assembly to determine stable dominance relationships, because
upon first meeting, as in round-robin competition, fish often
exchange aggressive acts before one clearly establishes domi-
nance and initiates all aggressive activity; such contests require
a fairly large number of consecutive acts to ensure a stable
relationship. In contrast, after some time together, relationships
are often in place, fish do not trade acts back and forth, thus
fewer acts suffice to determine which fish in a pair is dominant.
Results. As indicated earlier in the discussion of our experimental
design, both hypotheses predicted that the hierarchies formed
through group assembly should be linear, but they disagreed
about the extent of linear hierarchies formed via round-robin
competition. Although the prior attributes hypothesis antici-
pated linear structures, the social dynamics hypothesis fore-
casted nonlinear ones, because round-robin competition did not
allow interaction in a group context.
Fig. 2 shows the various hierarchy structures and frequencies
of sets of fish forming them in round-robin competition and
group assembly. Most of the hierarchies formed under group
assembly were linear, and the few others tended to show
relatively simple structural deviations from linearity. In contrast,
many hierarchies formed with round-robin competition were not
linear, including several with quite complicated structures.
In Table 2, the probabilities of linear and nonlinear hierarchies
in sets of four and five fish expected by chance alone (if each fish
had an independent 0.5 probability of dominating each other)
are compared with the proportions observed in round-robin
competition and group assembly. Although round-robin com-
petition only produced significantly higher proportions of linear
hierarchies than expected by chance in sets of five fish, group
assembly did so in sets of both sizes (one-sided binomial tests:
round robin competition, n ⫽ 16, P ⫽ 0.10 in sets of four fish,
and n ⫽ 12, P ⬍ 0.005 in sets of five; group assembly, n ⫽ 25,
P ⬍ 0.001 in sets of four and n ⫽ 11, P ⬍ 0.001 in sets of five).
Fig. 2. The structure of hierarchies forming in group assembly and round-robin competition for sets of four and five fish. An animal dominates all those listed
below it except as indicated by heavy arrows; three fish in an intransitive triad sharing the same rank are placed on the same level in a hierarchy. Frequencies
of experimental groups showing each structure are indicated in parentheses.
Table 2. Percentage of linear structures expected in random
hierarchies and observed in round-robin competition and group
assembly in sets of four and five fish
Size of set
Method of forming hierarchy
Random, % Round robin, % Group assembly, %
4 37.5 56.2 (n ⫽ 16) 92.0 (n ⫽ 25)
5 11.7 50.0 (n ⫽ 12) 90.9 (n ⫽ 11)
Chase et al. PNAS
兩
April 16, 2002
兩
vol. 99
兩
no. 8
兩
5747
SOCIAL SCIENCES
Concerning support for the hypotheses, further inspection of
Table 2 indicates that the proportions of linear hierarchies
formed with group assembly were much higher than those
established with round-robin competition. With group assembly
over 90% of the hierarchies in both sizes were linear, but with
round-robin competition, only about half, 54 and 50%, were
linear in sets of four and five, respectively. Statistical tests
indicate that these differences are highly significant (Fisher’s
exact test: P ⬍ 0.002 for sets of four, and P ⬍ 0.05 for sets of five).
Discussion. Group assembly experiments allowed fish to interact
in a group context, and nearly all the hierarchies were linear. But
when we prohibited this form of interaction in round-robin
competition, only about half the hierarchies were linear. Social
interaction greatly enhanced the formation of linear hierarchies.
Consequently, there is strong support for the social dynamics
hypothesis. The prior attributes hypothesis also finds support in
that sets of five fish meeting in round-robin competition formed
linear hierarchies at a rate significantly higher than chance.
It might be argued that differences in a certain class of prior
attributes (social attributes) could only be expressed in group
contexts but not in round-robins, e.g., the ability to react
strategically to others’ contests. However, if such attributes
existed and were vital to forming linear hierarchies, they should
have manifested in the first experiment in which fish met in
groups. We should have seen a large proportion of identical first
and second hierarchies, but we did not.
Our results do not demonstrate that social dynamics by
themselves can produce linear hierarchies. In the group assembly
experiments, in which group interaction proved so effective in
producing linear hierarchies, the individuals still varied in prior
attributes. For this reason we say that our results demonstrate
that social interaction greatly enhances the production of linear
hierarchies rather than that social dynamics by themselves
generate linear hierarchies. To determine whether social dy-
namics on their own can generate high proportions of linear
hierarchies, we propose investigation of hierarchy formation in
genetically identical fish or other animals with equal physical
characteristics. We conjecture that even without attribute dif-
ferences such animals would form linear hierarchies as readily as
those with variation in attributes.
Would some variation in weight, the one prior attribute we
controlled, be sufficient to produce high rates of linear hierar-
chies even without interaction in a group context? We suggest
that the differences required would be high, probably greater
than that found in many groups forming linear hierarchies either
in nature or the laboratory. For example, based on our exper-
iments with pairs of fish, we estimate that even when each
individual was ⬇25% heavier than the next (C 25% heavier than
D, B 25% heavier than C, and A 25% heavier than B), only
⬇34% of the hierarchies formed would be linear and identical to
the prior ranking on weight. And at this variation, A would be
about twice the weight of D! (also see Francis, ref. 43, for an
insightful discussion of the social mediation of weight differences
in established hierarchies).
General Discussion
Our results support both the prior attributes and social dynamics
hypotheses. More identical hierarchies appeared in the first
experiment and more linear hierarchies in the second experi-
ment (in sets of five fish) than expected by chance alone. But the
rates of identical hierarchies still were relatively low, and the
rates of linear hierarchies from round-robin competition were
significantly less than those from group assembly. To ensure that
most hierarchies were linear, as they are in many nature- and
laboratory-based studies, the fish had to interact in a group
context. Linear hierarchies did not consistently preexist on some
attribute ranking, ready to be revealed when individuals met.
Instead, social dynamics were crucial to the production of these
social structures. Although variation in attributes significantly
affected the respective placement of individuals on hierarchy
ladders, social dynamics were necessary for the dependable
manufacture of the ladders themselves.
These findings require a reconceptualization of the phenom-
enon of dominance hierarchies. Linear structures should not be
assumed to result simply from variation among individuals or
from cumulative conflicts among pairs of individuals. Instead, to
account for the common occurrence of linear hierarchies and
provide more accurate accounts of how individuals acquire their
ranks, we must look at patterns of interaction across whole
groups and understand how these patterns produce hierarchy
ladders. Even in dominance hierarchies among simple creatures,
in which individual differences in raw physical power, aggression,
and strategy might seem crucial, interaction processes are vital
in providing the typical forms of social organization that we
observe. Simply put, the formation of dominance hierarchies is
a richer and more complex phenomenon than has been thought
previously.
The importance of interaction among individuals for produc-
ing the patterns of organization in dominance hierarchies reveals
these structures as self-organizing or self-structuring systems.
These experiments are an empirical demonstration that domi-
nance hierarchies are indeed self-organizing, and they confirm
previous theoretical work (40–42).
But what particular social processes actually promote the
formation of linear hierarchies, and how do they do so? Our
results here show that these processes are at the core of the
as-yet-unanswered riddle of hierarchy formation. Empirical and
theoretical work on winner, loser, and bystander effects (25–29,
39–42) and applications of Chase’s jigsaw puzzle model (5, 23,
35–38) seem to be very promising starting points. Both of these
lines of research suggest that linear structure is promoted by
positive feedback to initial wins and losses during hierarchy
formation, i.e., when an animal dominating in one contest goes
on to dominate in others and when an animal becoming subor-
dinate in one contest goes on to be subordinate in others. We
suggest that further investigation of these and other dynamical
patterns hold the key to understanding how hierarchies come to
develop linear structures so often.
One implication of our results is that current models in
sociobiology are either too simple or too concerned with indi-
vidual differences to account adequately for the evolution of
behaviors leading to dominance hierarchies. Many of these
models use game theoretic techniques to consider conflicts
between pairs. Although these models are helpful, we concur
with Oliveira et al. (27) in that the evolution of behavior in
dominance hierarchy formation must be seen as contextual to
networks of individuals rather than independent dyads. Forming
dominance hierarchies and being a social animal in general may
require the evolution of considerable cognitive power in indi-
viduals to meet the contingencies of interaction in groups.
Another line of evolutionary thinking notes that individuals
ranking higher in hierarchies often produce more offspring than
those ranking lower. Consequently, if there are genetic linkages
to any differences in attributes that help influence higher rank,
they would be favored by natural selection (but see Francis, ref.
49, for careful consideration of the requirements for demon-
strating selection for attributes associated with dominance). But
even if these differences in dominance potential are selected for,
our results suggest that they are not doing the heavy lifting in the
production of linear hierarchies.
In their seminal work on friendship in human groups, Holland
and Leinhardt (50) argue that any network of relationships in
which higher-level properties can be modeled adequately using
only the properties of actors or pairs of actors has no true social
structure. In their words, there is nothing inherently social about
5748
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.082104199 Chase et al.
the structure of such a network of relationships. By this defini-
tion, our fish are genuine social creatures and their dominance
hierarchies are true social structures. Fish apparently can be
active and aware social actors, not just taking places in domi-
nance hierarchies that reflect their intrinsic, biological charac-
teristics. And if fish can form true social structures, why not other
animals? Thus our results suggest that there may be no funda-
mental discontinuities between social structure in humans and
animals.
Finally, our results suggest that dominance hierarchies in fish
and perhaps other social structures in ‘‘simple’’ animals might
serve as models to help us understand the development of
hierarchies and other forms of social organization in humans.
Ordinarily, the absence of higher-level cognitive ability, behav-
ioral complexity, language skills, and elaborate cultural forms
argue against applying lessons learned from studying social
organization in simple creatures to the investigation of social
systems in humans. However, finding that social interaction is so
important in producing organized structures in fish strengthens
the argument for investigating the importance of social dynamics
in producing dominance hierarchies and other social structures
in humans. Just as animals have served as invaluable models for
understanding genetics, health and disease processes, and cog-
nition and perception in humans, animal models may enable us
to better understand how some of our social systems develop as
they do.
We thank Ritu Bapat, Michelle Cornog, Peter Murch, Kristine Seitz, and
Nam Thai for help in data collection; Ginny and Charlie Eckstein and
Tom Keegan for advice on fish care; Paul Loiselle and the New York
Aquarium for Wildlife Conservation for providing fish; Peter Bearman,
Al Carlson, Stephen Cole, Andrea Tyree, and Everett Waters for
comments on earlier drafts; and Eugene Danner, Inc., Penn Plax
Corporation, Python Products, Rena Corporation, and Tetra USA for
equipment donations. Support was provided by the Harry Frank
Guggenheim Foundation, National Science Foundation Grant SES-
9424006, the Guy Jordan Endowment Fund of the American Cichlid
Association (to I.D.C.), and grants from the Institute Fellows program
at Georgia Institute of Technology and the Georgia Tech Foundation
(donation by John Grigsby, to C.T.).
1. Heinze, J. (1990) Naturwissenschaften 77, 41–43.
2. Wilson, E. O. (1975) Sociobiology (Harvard Univ. Press, Cambridge, MA).
3. Vannini, M. & Sardini, A. (1971) Monitore Zool. Ital. (N.S.) 5, 173–213.
4. Goessmann, C., Hemelrijk, C. K. & Huber, R. (2000) Behav. Ecol. Sociobiol.
48, 418–428.
5. Nelissen, M. H. J. (1985) Behavior 94, 85–107.
6. Post, W. (1992) Anim. Behav. 44, 917–929.
7. Barkan, C. P. L., Craig, J. L., Strahl, S. D., Stewart, A. M. & Brown, J. L. (1986)
Anim. Behav. 34, 175–187.
8. Addison, W. E. & Simmel, E. C. (1970) Bull. Psychon. Soc. 15, 303–305.
9. Hausfater, G., Altmann, J. & Altmann, S. (1982) Science 217, 752–755.
10. Savin-Williams, R. C. (1980) J. Youth Adolesc. 9, 75–85.
11. Jameson, K. A., Appleby, M. C. & Freeman, L. C. (1999) Anim. Behav. 57,
991–998.
12. Raleigh, M. J., McGuire, M. T., Brammer, G. L., Pollack, D. B. & Yuwiler, A.
(1991) Brain Res. 559, 181–190.
13. Sapolsky, R. M. & Share, L. J. (1994) Am. J. Primatol. 32, 261–275.
14. Clutton-Brock, T. H., Albon, S. D. & Guiness, F. E. (1984) Nature (London)
308, 358–360.
15. Beacham, J. L. (1988) Anim. Behav. 36, 621–623.
16. Holenkamp, K. E. & Smale, L. (1993) Anim. Behav. 46, 451– 466.
17. Drews, C. (1993) Behavior 125, 283–313.
18. Jackson, W. M. & Winnegrad, R. L. (1988) Anim. Behav. 36, 1237–1240.
19. Cloutier, S., Beaugrand, J. P. & Lague, P. C. (1996) Behav. Processes 38,
227–239.
20. Beaugrand, J. P. & Cotnoir, P.-A. (1996) Behav. Processes 38, 287–296.
21. Slater, P. J. B. (1986) Anim. Behav. 34, 1264–1265.
22. Jackson, W. M. (1988) Ethology 79, 71–77.
23. Chase, I. D. (1982) Behavior 80, 218–240.
24. Francis, R. C. (1988) Ethology 78, 223–237.
25. Chase, I. D., Bartolomeo, C. & Dugatkin, L. A. (1994) Anim. Behav. 48,
393–400.
26. Hsu, Y. Y. & Wolf, L. L. (1999) Anim. Behav. 57, 903–910.
27. Oliveira, R. F., McGregor, P. K. & Latruffe, C. (1998) Proc. R. Soc. London
Ser. B 265, 1045–1049.
28. Johnsson, J. I. & Akerman, A. (1998) Anim. Behav. 56, 771–776.
29. Silk, J. B. (1999) Anim. Behav. 58, 45–51.
30. Bak, P. (1996) How Nature Works (Springer, New York).
31. Resnick, M. (1994) Turtles, Termites, and Traffic Jams (MIT Press, Cambridge,
MA).
32. Kauffman, S. A. (1993) The Origins of Order (Oxford Univ. Press, New York).
33. Landau, H. G. (1951) Bull. Math. Biophys. 13, 1–19.
34. Chase, I. D. (1974) Behav. Sci. 19, 374–382.
35. Mendoza, S. D. & Barchas, P. R. (1983) J. Hum. Evol. 12, 185–192.
36. Chase, I. D. (1985) Anim. Behav. 33, 86–100.
37. Barchas, P. R. & Mendoza, S. D. (1984) in Social Hierarchies: Essays Toward
a Sociophysiological Perspective, ed. Barchas, P. R. (Greenwood, Westport, CT),
pp. 23–44.
38. Eaton, G. G. (1984) Int. J. Primatol. 5, 145–160.
39. Skvoretz, J., Faust, K. & Fararo. T. J. (1996) J. Math. Soc. 21, 57–76.
40. Theraulaz, G., Bonabeau, E. & Deneubourg, J. L. (1995) J. Theor. Biol. 174,
312–323.
41. Hogeweg, P. (1989) in Artificail Life I, ed. Langton, C. (Addison–Wesley,
Redwood City, CA), pp. 297–316.
42. Hemelrijk, C. K. (2000) Anim. Behav. 59, 1035–1048.
43. Francis, R. C. (1988) Anim. Behav. 36, 1844–1845.
44. Johnsson, J. L. (1997) Ethology 103, 267–282.
45. Miklosi, A., Haller, J. & Csanyi, V. (1997) Behav. Processes 40, 97–105.
46. Barends, G. P. & Barends-Van Roon, J. M. (1950) Behavior Suppl. 1, 1–242.
47. Dugatkin, L. A., Alfieri, M. S. & Moore, A. J. (1994) Ethology 97, 94–102.
48. Guhl, A. M. (1975) in Social Hierarchy and Dominance, ed. Schein, M. W.
(Dowden, Hutchinson, and Ross, Stroudsburg, PA), pp. 156–201.
49. Francis, R. C. (1984) Behavior 90, 25–45.
50. Holland, P. W. & Leinhardt, S. (1979) in Perspectives on Social Network
Research, eds. Holland, P. W. & Leinhardt, S. (Academic, New York), pp.
63–83.
Chase et al. PNAS
兩
April 16, 2002
兩
vol. 99
兩
no. 8
兩
5749
SOCIAL SCIENCES