Cognitive, Perceptual and
Brain Sciences research department
University College London,
UK
Herding is a form of convergent social
behaviour that can be broadly defined as the
alignment of the thoughts or behaviours of
individuals in a group (herd) through local
interaction and without centralized
coordination. We suggest that herding has a
broad application, from intellectual fashion to
mob violence; and that understanding herding is
particularly pertinent in an increasingly
interconnected world. An integrated approach to
herding is proposed, describing two key issues:
mechanisms of transmission of thoughts or
behaviour between agents, and patterns of
connections between agents. We show how bringing
together the diverse, often disconnected,
theoretical and methodological approaches
illuminates the applicability of herding to many
domains of cognition and suggest that cognitive
neuroscience offers a novel approach to its
study.
Introduction
Cognitive psychology generally focuses on
the individual as the fundamental unit of
analysis [1]. Nevertheless, we are all
embedded in a complex system of social
structures, which ground and organize much of
our behaviour [2], ranging from national
identity to religious affiliation. Here, we
consider one of the many bridges that link
agents and the social structures in which they
are embedded: a form of convergent social
behaviour termed 'herding'. Herding can be
broadly defined as the alignment of thoughts or
behaviours of individuals in a group (herd)
through local interactions rather than
centralized coordination. In other words, the
apparent central coordination of the herd is an
emergent property of local interactions. Herding
is an influential and well-documented feature of
human behaviour in a number of domains,
particularly economics and finance
[3&endash;5]. Although the current
economic turmoil has revealed the depth of
herding among financial institutions and
individual investors [6,7] (and by
implication the agents responsible for their
decisions), this concept also has much broader
relevance beyond the economic arena. Examples of
phenomena that have been described as involving
herd behaviour are diverse and varied, ranging
from stock market bubbles and financial
speculation to zealotry (e. g. the 2002 Gujarat
mob violence [8]), political choice
[9] and consumer preferences
[10,11]. The concept is well known in
ethology, where for example the biologist
William Hamilton illustrated how herd behaviour
can emerge from the uncoordinated behaviour of
individuals engaged in predator avoidance
[12]. The process has also been
investigated in social psychology and terms such
as Fad, Fashion, Mass Hysteria, Bandwagon
Effect, Groupthink and Herd Instinct have
entered common parlance. Whereas the concepts
behind herd mentality and herd behaviour have a
rich history (Table 1), the methods, techniques
and approaches currently used to elucidate them
are relatively recent. In this article we review
the extensive range of theoretical frameworks
for describing herding. Similar ideas and
explanations have emerged in many fields, albeit
with different emphases, demonstrating the
interdisciplinary nature of the concept.
Wepropose a framework with which to organize
these diverse approaches, which is based on a
distinction between the mechanisms of
transmission of a particular thought between
individuals and the patterns of connections
between individuals. We also distinguish between
two main types of transmission: automatic
contagion and rational deliberation. We suggest
that cognitive neuroscience can reveal the
mechanisms underlying the transmission of
information, which can in turn help elucidate
patterns of herd behaviour.
Models of herding
As indicated above, herding among
individuals has been studied within a number of
diverse domains. As a result, a number of
different mechanisms and approaches have emerged
across these domains in order to explain herding
behaviour. It is therefore important to develop
a conceptual framework within which the
different approaches and models can be
described, one that also permits the
highlighting of common features. We propose that
understanding how members of a group become
aligned by 'local' interactions requires
determining: (i) the mechanism of transmission
that propagates a particular thought or
behaviour from one agent to another and (ii) the
pattern of interactions between agents.
Conceptualizing herding as representing these
two separate, though interconnected,
perspectives addresses these requirements.
Experimental approaches and modelling typically
focus nearly exclusively on one issue or the
other and this division between mechanisms of
transmission vs. pattern of interactions between
agents can be considered as a local vs. global
division. This framework brings two primary
questions into focus: (i) How do various types
of information transmission, conscious vs.
automatic, rational vs. emotional, affect the
emergent architecture? (ii) What are the
emergent effects of the pattern of
connections?
Cognitive psychology and neuroscience
generally focus on the individual, or on parts
of the herd (Figure la), yet herding also arises
from the organizing relations of those parts,
that is, by how the individuals are configured
(Figure lb). The pattern-based approach, adopted
widely in fields such as Traffic Jam and Crowd
analysis, financial markets, and social network
analysis, emphasizes the static structure of the
system and its relationships. It shows the
dependencies among these components and takes
the relationships between individuals as the
primary unit for herding research and for the
development of theory. However, a full
understanding of herding behaviour requires the
ability to 'shift' between the two levels
represented in our framework. This distinction
aids in the identification of contextual
effects, that is, herding behaviour that is not
apparent at the individual level. Moreover, the
proposed framework indicates at which level
different research questions could be
appropriately addressed. For instance, there are
some enquiries, such as whether herding is
linear (that is, whether we can use conventional
'reductionist' approaches that look at the parts
and employ superposition to investigate how they
work together), which can only be addressed at
the pattern level. In contrast, focusing on the
affective and cognitive mental states of
individuals, specifying the common and
differential neural mechanisms of 'self and
'other', or interpersonal face-to-face
interactions [131 may reveal quite different
aspects of herding from the insights arising out
of the pattern-based level. These two levels are
not mutually exclusive; both levels can
cross-fertilize each other. Moreover we consider
that increasing knowledge of the brain areas
involved in social cognition (the capacity to
understand people's behavioural intentions,
social beliefs, and personality traits) [141
can inform patternbased herding analyses and
constrain the vast space of relations that can
potentially exist between actors.
Pattern-based theories/models of herding
behaviour: structure sets the herd?
Pattern-based explanations treat individuals
as units with certain simple, well-defined
properties and modes of interaction (Figure 2,
left branch). The terminology used in this class
of models, such as 'critical mass',
'self-organized criticality' and 'epidemics', is
inspired by models in either particle physics or
epidemiology and shares a similar structure.
Such models often come under the rubric of
econophysics models of herding and are prevalent
in finance [151.
Pattern-based approaches view herding (and
social phenomena in general) in terms of the
patterns of interaction among the agents,
modelled as simple imitators, or as following
basic heuristics. In other words, people are
treated as units or atoms with certain simple,
well-defined properties and modes of interaction
that yield herding: the focus is patterns, not
people [161. The models generally rely on
physical laws, distances and velocities rather
than the emotional states of the herd. These
approaches to multiagent phenomena are applied
to many areas, naturally lending themselves to
the explanation of phenomena, such as queuing,
crowd and traffic interactions [171.
Cellular Automata models [18,191 and Ising
models [201 are the most extreme examples of
this abstract approach. Closely related are
models of 'flocking' in animal behaviour (for
excellent reviews see [21,221), widely
exhibited in biology by living creatures ranging
from bacteria to birds. These models have been
used to simulate human crowd behaviour
[17,231, social behaviour such as language
emergence and evolution [241, as well as in
attempts to elucidate the general concept of
emergent phenomena [251.
The most intuitively understandable
pattern-based perspective is social network
analysis - this perspective employs the simplest
of structures in which each node (a person) is
attached via some tie to others. This approach
has had an important influence in modern
sociology (Box 1). Such studies have shown that
happiness and obesity tend to spread through
social networks in a manner analogous to a
contagious disease [26,271 (Figure Tin Box
1). Thinking in social network terms has
progressed from being an evocative allegory
[28-301 and has been extended to agent-based
modelling approaches [311. Social network
analysis focuses on how the structure of other
relationships affects individuals [321
rather than treating individuals as the
fundamental unit, and allows epidemiological
methods to be employed.
Transmission mechanisms in herding: how
do we broadcast?
In contrast to the focus on patterns of
interaction, the complementary transmission
perspective seeks to unify and identify
mechanisms of transfer of information in herding
(as illustrated in the split within the right
branch of Figure 2) by concentrating on the role
of cognitive and affective components,
particularly the effortless human capacity known
as 'mentalizing' (the ability to explain and
predict the behaviour of others by attributing
to them independent mental states) [331. A
number of important theoretical distinctions can
be used to classify different transmission
mechanisms in human herding, including rational
vs. emotional, automatic vs. controlled and
conscious vs. unconscious. However, herding is a
social tendency; an essential component of this
behaviour is that it can incorporate beliefs
about the herd. Individuals often converge by
modelling behaviours and beliefs of the larger
group within which they are embedded. Indeed
most economic models are based on the assumption
that single agents are able to view another
agent's perspective [341.
People often explicitly attempt to infer
others' beliefs, attitudes or preferences; and
draw on these to help determine their own
perspective [351. As such, our sub-division
reflects this key mentalizing facet. In Figure
2, the left subbranch under transmission-based
approaches includes approaches that do not
postulate mentalizing as a critical aspect of
transmission. These approaches range from
emotional contagion through to social contagion
and priming. The right 'mentalizing' sub-branch
encompasses social conformity and influence,
culminating with rational processes through
which the agent consciously and deliberately
considers the information in the signals of
others. Although the importance of the presence
of others has been a mainstay of social
psychology, since Le Bon's [361 claim that
collective influence can almost mesmerize the
individual, when herding is viewed from a
rational stance it is generally considered in
two flavours: informational and reputational
(that is, peer pressure) [371. Economists,
in particular, have observed how groups of
individually rational agents, each drawing on
information about the choices of others can fall
into collectively irrational 'informational
cascades', which do not properly reflect the
group's preferences. This assumes a
utility-maximizing behaviour on the part of each
agent [381 (Box 2).
There are, of course, other explicit drivers
that may induce the alignment of the thoughts or
behaviours of individuals in a group (herd),
through local interaction and without
centralized coordination. Social psychological
accounts, such as those of Sherif [391, Asch
[401 or Latané [411, indicate
mechanisms that are consistent with herding and
contagion under a variety of conditions. Social
influence, real or imagined, of others can have
a measurable impact, ranging from obedience and
compliance through to conformity (for a useful
review [421), but generally the latter imply
some awareness of the act of alignment. This
boundary between self and other is where social
neuroscience can perhaps make a valuable
contribution to the investigation of how the
individual's perception [431 and the borders
of self are affected by the group (Box 3).
Social psychologists in particular have studied
this as the process of de-individuation, and
have addressed how the loss of a person's sense
of individuality can reduce normal constraints
against deviant behaviour. De-individuation is
generally conceived of as a collective
phenomenon, where anonymity and reduced feelings
of individual responsibility provide a mechanism
for situational forces to collectively drive
behaviour [441 immersing the individual into
the coup or herd.
Although herding and its consequences may
arise from active choice, responses often occur
without awareness, hence the key role of
mentality in herding. Presumably, these and
other Theory of mind factors can be manipulated
in the laboratory.
The most researched example of the
non-mentalizing approach in describing
transmission in human herding is that of
emotional contagion. Emotional contagion
involves an involuntary spread of feeling
without any conscious awareness of where the
feeling initially originated [451 and
without necessarily requiring interpersonal
empathy. For example, a child's emotion, be it
excitement or a tantrum, can rapidly influence
the emotion of others, generating a group of
rowdy and energetic children. Adults too can
experience such contagion, as when one can
automatically pick up the excitement in a crowd
or audience. However, unlike empathy, emotional
contagion does not require understanding
another's emotion and is largely involuntary, a
less conscious and more 'infectious' effect,
relying principally on non-verbal communication
(although the 'online world' of emails and
instant messaging can also be susceptible to
emotional contagion, where emotion can be
transmitted without non-verbal cues [461).
It is not yet clear how this domain is related
to other contagious behaviours in humans such as
laughter [471 and yawning [481 or
to what extent these contagious behaviours arise
through cultural or innate processes
[491.
Given the importance of implicit processes
in social cognition and the possibility that
emotions play an adaptive role in the social
environment [501, it follows that fast,
automatic and perhaps unconscious routines
[351 provide a signalling channel to
transmit messages to all members of the group.
The analogy with animal herding of this ripple
effect [511, where moods can ripple out,
influencing group members' emotions, their group
dynamics and individual cognitions, provides an
evolutionary perspective into these behaviours
and suggests that they are crucial for the
maintenance of societal norms. Thus, the notion
of emotional contagion can be extended to the
broader concept of social contagion: the
tendency to automatically mimic and synchronize
expressions, vocalizations, postures and
movements with those of another person leading
to behavioural convergence [521.
A strong form of contagion is termed
hysterical contagion (also termed mass hysteria,
collective hysteria or even mass psychogenic
illness). A common manifestation of mass
hysteria occurs when a group of people (falsely)
believe that they are suffering from a similar
disease or ailment [53-561. Such
manifestations have been reported as far back as
1374 (with dancing mania reaching such an
extreme that it caused deaths) [571. Other
manifestations of the dark side of social
contagion are displayed in mobs, riots and
hooliganism [581. These latter phenomena
highlight the importance of a framework in
organizing different approaches, as they
indicate examples of behaviour which most likely
draw upon both mentalizing and non-mentalizing
drivers.
There is, of course, another example of an
implicit effect, well researched by
psychologists, which has applications for
herding, namely priming: another example of a
nonmentalizing phenomenon. People can be primed
into certain forms of similar behaviour, with
primes ranging from smells [591, everyday
objects [601, the surroundings [611, to
the performance of activities such as marching
or dancing in unison which increase loyalty to
the group [62,631. More worryingly, on a
larger scale, the news media [64,651
implicitly influence the public, creating mass
movements of the herd that are invisible to the
individual.
What can social neuroscience
say?
A point of interaction between these two
levels concerns the biological mechanisms that
underlie herding. Social neuroscience is ideally
positioned to connect these levels. Social
structures may be emergent organizations beyond
the individual, yet these emergent organizations
require biological systems in the individual to
create them [66,671. Furthermore, there is a
large body of work on imitation [681. Mirror
neurons (nerve cells that fire when we carry out
an action, or watch someone else carry out a
similar action) may also play a role in this
interaction between the individual and the herd.
Neural evidence seems to support this idea as a
promising line of research. There are, as yet,
few studies investigating the neurobiological
correlates of herding mentality, conformity
[43,691 and emotional contagion [70,711.
However, the evidence so far is promising; for
example, it demarcates how emotion circuitry
diverges in the adolescent male and female brain
under peer approval [721 and rejection
[731.
There are many areas in which the cognitive
neuroscience 'toolbox' can be applied to
research in herding. For example, neuroimaging
could provide evidence for reinforcement models
of information cascades [741. There may also
be analogies of quorum sensing (from ethology)
in human herding. The communication of
chemosensory signals between conspecifics has
been well documented in many vertebrates and
invertebrates, and as such provides a
transmission mechanism for the non-mentalizing
approaches to herding. Indeed, in a recent
study, pre-exam sweat had a specific effect on
brain activity, correlating with areas involved
in empathy and those that process social and
emotional signals [751.
Combining the methods of neuroscience and
the diverse approaches presented in Figure 2 can
generate powerful tools for studying the brain
processes behind human herding. Novel
neuroimaging techniques (scanning many
individuals at a time) could be used to capture
brain-based
correlates of herding [761, whereas the
emerging field of neuroeconomics [77,781
offers the possibility of characterizing and
building biological models of herding (such as
the recent imaging study based on a model from
economics the 'beauty contest' game - which
permits the investigation of how a player's
mental processing incorporates the thinking
process of others in strategic reasoning
[791).
Concluding remarks and future
directions
The concept of herding has been evoked in
many different contexts, ranging from mass
hysteria in neurology [561 to the diffusion
of innovations in economics and to the
propagation of ideas [801. These appeals to
collective behaviour all imply that certain
forms of behaviour go beyond the individual, but
different disciplines yield somewhat dissimilar
accounts of the mechanisms of herding. To
discern structure within this array of
approaches requires a broad integrative
viewpoint. The framework presented here, similar
to other integrative approaches in cognitive
psychology [811, has heuristic as well as
integrative potential. It presents scaffolding
for organizing the questions that can be
addressed about herding and the common and
diverse mechanisms that underlie it across
domains. The proposed classification invites
specific interdisciplinary questions to be
addressed (see also Box 3). For instance, a
pertinent case would be the boundary
specification problem of a herd. Dunbar has
suggested that the typical size of a social
network is restricted to around 150 people due
to possible limits in the capacity of the human
communication channel [821. How do pattern
models integrate this with the limits of
cognitive capacity?
In the group environment we are exposed to
ever shifting emotional messages and are
influenced by the social situation and other
agents. The convergence upon a single mood or
emotion can elicit herd behaviour in which the
agents are connected and process stimuli in a
similar manner. How stable this contagion is,
its neural mechanisms and the role of rational
'top-down' factors remain unclear.
Diffusion theory explores social networks
and their role in influencing the spread of new
ideas and practices. How do authority figures or
'hubs' transmit their information in the more
spatial-based transmission models, and can
knowledge of cognitive biases or biological
limitations in processing influence or provide
more accurate parameters for these
descriptions?
Allying the methods of cognitive
neuroscience with the pattern-based and
transmission-based perspectives on herding
creates both interesting hypotheses and
predictions concerning areas of activation
associated with selfand other-processing and how
the individual transmits (and encodes)
information from/to the herd. Interesting
herding patterns can be expected to arise within
biological constraints that neuroimaging can
elucidate, which in turn could provide insight
into factors ranging from the rationality of
herds [831 to the role of mentalizing and
proximity [841.
One wonders, in this internet age with the
increasing ease of sharing information and ideas
[851, and with ever proliferating points of
contact between people, whether we will be more
susceptible to herding. However, with modern
experimental data tracking techniques permitting
the analysis of individuals in a top-down manner
[86] and neuroscientific methods such as
imaging in a bottom-up manner [71], more
than ever the fields of social cognition and
social neuroscience can play an important role
in exploring the ubiquitous yet sometimes
disquieting, interaction between the individual
and the herd.
