 |
 |
 |
|
 Volume 14 Numbers
3 and 4, 2004
A REVIEW OF NETWORKING THEORIES ON THE FORMATION OF PUBLIC
OPINION
Moses A. Boudourides
Abstract. The aim of this study is to review the network concept
and its relevance on theories of public opinion formation. For this purpose,
after discussing social and policy networks, I review certain network theories
of (1) collective action and (2) voting choices and preferred modes of political
participation. Finally, I present a network simulation of public opinion formation
that generalizes Axelrod’s adaptive culture model. This simulation is
based on both convergent and divergent communicative processes. Introduction The
question regarding to what extent networks (social, political or
heterogeneous-hybrid networks) influence political agency and behavior is of
tremendous theoretical and practical interest. I will not discuss this notion
explicitly or assess the role of electronic networks on processes of formation
of political and public opinions. However, nowadays, the pervasiveness and
ubiquity of computer-mediated communication is a generally acknowledged fact.
What, instead, I aim to do here is to examine certain network theories of (1)
collective action and (2) voting choices and preferred modes of political
participation. This paper will start with a brief sketch of the network concept
and will conclude with a concrete network simulation of collective processes of
public opinion formation. Networks Networks can be found everywhere in the present landscape of society,
culture, politics and science and technology. Thus, the word
“network” appears to have become a fashionable catchword. As Kenis
and Schneider already claimed more than a decade ago: “The term network
is on the way to becoming the new paradigm for the ‘architecture of
complexity’ (compared to hierarchy as the old architectural paradigm of
complexity: see Simon)” (Kenis & Schneider, 1991, p.
25). In fact, besides many occurrences in science and engineering
(for instance, neural, “complex,” self-organized, informational,
environmental, transportation and telecommunication networks), networks can
frequently be found in social science (e.g., social networks, scientometric
networks), in political science (“policy networks”) and in
economics and organization theories (e.g., networks of innovation, networks in
between markets and hierarchies, learning networks). Within
such an extended spectrum of uses and occurrences of networks, it may be very
difficult and even meaningless to try to find a common denominator in a formal
definition of this concept. However, for the present purposes, a generic or
“cognitively shared’ (among so many domains) understanding of
networks could be described in terms of two entities: actors and relations.
This implies that in a network a set of nodes or actors (individual, aggregate
or mixed) are related or linked to each other under specific more or less
stable mechanisms, defining a non-hierarchical (horizontal) set of
relationships among the actors. If one tries to conceptualize
“actors” and “relationships” in a concrete context, one
realizes that several selections of these terms exist, each yielding different
notions of networks. For this reason, I would like to restrict our discussion
to two particular cases: (1) a social network (from sociology) and (2) a policy
network (from political science).
Social Networks In social
theory, social networks are related to what is usually identified as the
“relational aspect” of social structure. As José
López and John Scott remark (2000, p. 1),
although there are many definitions and discussions on social agency and
action, very few deal with social structure. By its meaning in everyday life,
social structure refers to patterns or arrangements of whatever elements are
considered to constitute society. According to the aforementioned authors,
social structure points to three independent and complementary aspects of the
social: the institutional, the relational and the embodied. Institutional structure is
“comprising those cultural or normative patterns that define the
expectations that agents hold about each other’s behavior and that
organize their enduring relations with each other.” While, relational structure is
“comprising the social relations themselves, understood as patterns of
… interconnection and interdependence among agents and their actions, as
well as the positions they occupy.” Finally, embodied structure is “found in the habits and skills that
are inscribed in human bodies and minds and that allow them to produce,
reproduce, and transform institutional structures and relational
structures” (López & Scott, 2000, pp.
3-5). Relational structure,
that is, social structure as a patterning of social relationships, is usually
described as a social network, in which actors are whoever and whatever
performs the agencies through which relations among actors develop and hold
them together. Thus, actors (or agents) in a social network can be individual
people, objects or events. However, they can also be aggregate units such as
organizations, institutions, firms, communities, groups, families, etc. The
very idea of the social network approach is that relations or interactions
between actors are the building blocks or the key factors that sustain and
define social structure, despite the “nature” of actors or any
other attributes with which they might be endowed (Wellman,
1988; Wasserman & Faust, 1994; Scott,
2000). Typically,
interactions between actors result from an exchange of resources from the
specific social and cultural contexts they live and communicate in. More
specifically, these contexts include the existing distribution of power or authority
relationships, accepted social norms, habits, dependencies, practices,
expectations and preferences. In these interactions, exchanged resources can be
either material or informational, such as goods, money, information, services,
social or emotional support, trust, influence. Each kind
of a resource exchange is considered to constitute a social network relation
and actors maintaining the relation are said to maintain a tie. The strength of
a tie may range from weak to strong, depending on the quantity, quality and
frequency of the exchanges between actors (Marsden &
Campbell, 1984). Patterns of who is tied to whom reveal the structure of
the underlying network; they show how resources flow among actors and to what
extent actors are interconnected in the network. In a very well known example
of social network analysis, Mark Granovetter (1973,
1974) investigated the exchange of job
information among acquaintances and found that weak ties are in practice quite
strong for the diffusion of such information. Policy Networks Also, in the context of policy studies and political theory, a variety of
definitions of “policy networks” exists, based on the attributed
sense of the basic terms. For instance, typical actors in a policy network
might be either public (e.g., the state, governmental or other public
institutions, citizens) or private (e.g., the market, corporations, interest
groups, consumers) or even mixed-hybridized forms of the former (e.g., modern
science the way envisaged by Callon, 1994). The situation becomes even more complicated when one tries to theorize in
a general or generic setting the relationships that could possibly tie these
actors in a policy network. The dominant view in literature on policy is that
policy actors carry certain interests and possess certain resources. Therefore,
it is argued, the actors are linked together (i.e., they develop their
relationships of interdependence or seclusion) when they proceed to mediate
their interests and exchange their resources. Nevertheless, this mechanism
should not be understood in the sense that the actors’ interests and
resources (i.e., identities and competences) exist before the actors start to
constitute a network; the very idea of the “relational” perspective
is that a network both defines and is defined by its links. However, the
question that could be raised concerns the way in which actors manage their
relationships. In answering this question, two different “schools”
of policy-making can be identified (cf., Börzel, 1997, p. 3).
Still, these “schools” do not exclude each other, as hybrid conceptualizations between these two
theoretical responses can be distinguished (Mayntz, 1993): ·
According to the “interest intermediation
school” (Marsh & Rhodes, 1992), a
generic meso-level mechanism can be identified in developing relationships
within a policy network, independently of who the actors are. Depending on whether
the involved actors have common or competing interests and what exactly their
interests are, they decide with whom, to what extent and for what purpose to
exchange their resources. Thus, the actors take certain political decisions,
formulating, implementing or changing certain policies by their bargains (to
optimize their profits) or their negotiations (to reconfigure their
identities). ·
According to the “governance school” (Kenis & Schneider, 1991), which is restricted to apply
only to public-private interactions of public policy, a policy network between
public and private actors is a specific form of governance. David Marsh (1998,
p. 8) argues: “Hierarchy is a mode of governance characterized by a
very close structural coupling between the public and private level, with
central coordination, and thus control, being exercised by government. In
contrast, markets as a form of governance involve no structural coupling.
Furthermore, outcomes result from the market-driven interplay between a
plurality of autonomous agents, drawn from the public and the private spheres;
there is no central coordination. To the contrary, policy networks involve a
loose structural coupling; interaction within networks between autonomous
actors produces a negotiated consensus which provides the basis for
coordination.” Epistemic
communities provide a concrete example of a policy network, in which specific
constellations of actors (e.g., public and private actors, experts and involved
citizens) are mediating their interests in order to develop a flexible form of
governance for complex techno-scientific issues. Epistemic communities have
arisen in the field of foreign policy and international relations in the
context of international policy coordination of such issues as Collective Action According
to an old tradition of social theorizing, public opinions and individual
judgments are affected by observations of social aggregates and mass behaviors
in relation to these issues of opinions or judgments. Thus, in the context of
theories of collective action, the formation and change of public opinion can
be considered a mechanism through which a certain action affects other future
or remotely occurring actions. However,
traditionally collective action should be defined as the provision of a
collective good, which in turn is defined by its nonexcludability (Olson, 1965; Hardin, 1982). According
to Olson (1965), the logic of collective action is based
on the assumption that individuals motivated by self-interest will, whenever
possible, avoid investing resources in a joint endeavor, thus leaving others to
contribute their share even though all will benefit, a phenomenon known as
“free riding.” Therefore, Olson argued that “rational,
self-interested individuals will not act to achieve their common or group
interests” (1965, p. 2) unless they are given
private or selective individual incentives which will reward cooperators or
punish noncooperators. Although it
is undeniable that Olson’s arguments opened the discussion on collective
action in social theory, they, subsequently, were often considered problematic
in areas where the existing empirical evidence opposed them (e.g., in theories
of social movements). Thus, Marwell and Oliver and their collaborators in a
series of articles and a book published between 1983 and 1993 (Oliver & Marwell, 1988; Marwell, Oliver, & Prahl, 1988; Oliver, Marwell, & Texeira, 1985; Prahl, Marwell, & Oliver, 1991; Marwell
& Oliver, 1993) developed a theory of “critical mass” which
was encompassing Olson’s facts about collective action but at the same
time attempting to give more realistic solutions to its contested parts. For
instance, like most economic theories of markets, Olson assumed independence of
actors making decisions. Oliver (1993) and Marwell
and Oliver (1993) criticized this view and
emphasized the importance of the network of relations in which interdependent
actors are embedded. Computer simulation experiments by Marwell and Oliver (1993) showed that the extent to which people are
interconnected in communication networks increases their willingness to support
the collective good. Using a similar research strategy, Marwell, Oliver and
Prahl (1988) showed that centralization and resource
heterogeneity in the network influenced aggregate contributions to a collective
good. In the
early 1990s, a number of social theorists argued that collective action is
formed and sustained less by the attention given to the collective good rather
than by mechanisms of formation and diffusion of public opinions. An assessment
and review of the works citing critical mass theory has been published recently
by Oliver and Marwell (2001). In the context
of the present study, three of the studies cited by Oliver and Marwell are
relevant. These three studies draw upon network theories in order to explicate
mechanisms of public opinion formation. Kim and
Bearman (1997), for example, have developed an interesting
model of opinion changes occurring in a network. Actors increase their interest
to participate in public processes if connected to people with higher interest
levels who contribute. In contrast, they decrease their interest if connected
to others with lower level that defects. In this sense, collective action
occurs if and only if there is a positive correlation between interest and
power/centrality and, so, interest heterogeneity is found to have positive effects
on “pulling up” the population’s potential for participation. Bahr and
Passerini (1998a, 1998b) have
developed a statistical mechanics model of collective behavior in analogy with
physical systems. They set up a model in which each actor possesses a
“strength” factor of opinion (or of persuasiveness) and the
probability of choosing an opinion is proportional to the number of actors who
hold that opinion, that is, it depends on the group distribution of opinion.
They made an analysis of the conditions in which a group changes opinions and
the extent to which this depends on the size of the group. The volatility of
opinions is, then, understood as “social temperature.” They also considered
cellular automata in which actors restricted their interaction to those people
at their vicinity, based on some well-defined rules. Their aim was to study the
emergent group patterns. For instance, they observed abrupt phase transitions
from consensus to near consensus, from well-ordered pockets of opinion at low
“social temperature” to less-ordered nonconsensus at higher
temperature. In some cases they even observed chaotic transitions. Ohlemacher
(1996) worked on “social relays,” that
is, mobilization-mediating social networks, based on empirical data regarding
citizens’ campaigns in Finally, I
cite Lohmann’s (1994) “signaling”
model, which is only indirectly a network model. However, as its conclusions
are very important for theories of critical mass, this study is of relevance in
this present study. The Lohmann model is based on informational cascades
(during the Monday demonstrations in Leipzig, East Germany, 1989-91) in which
actions of others signaled the extent of dissent from the regime. Remarkably
enough, Lohmann challenges one of the premises of the critical mass theory,
namely that extremists are important for the formation of the critical mass.
Instead, she argues that protest accelerates when moderates are involved early
in the process. Voting Choices and
Political Participation In general,
both citizen involvement in political institutions and individual
decision-making on voting and participation are said to depend on social
psychological perceptions and beliefs. Social forces impinge upon the citizen
and social and interpersonal interactions among citizens. The dimension of
interactions suggests that there should be a relationship between social
connectedness and civic participation or voting preferences. Many scholars in
political sociology have devoted time and research on exploring this relationship
(cf., Knoke, 1990a, chapter 2, pp. 29-56; Huckfeldt & Sprague, 1995; Verba
et al., 1995). In fact, there are
three distinct streams approaching the relationship of actors’ connectedness
with their political behavior: (1) the classical approach of surveys measuring
actors’ attitudes, (2) the macro-level analysis of collective patterns
derived from observations of social aggregates and (3) the exploration of
micro-level interpersonal dynamics developed among interacting individuals. Ever since
Paul Lazarfeld and his colleagues started their early surveys on individual
voting in presidential elections in the late 1940s, the decisive factor in
individual vote decisions has been the information flow through networks of
interpersonal communication (Lazarfeld et al., 1948).
This approach formed the base of the so-called “ Another old
empirical tradition in political science on the analysis of opinion formation
processes is the “contextual analysis” of voting choices. Here,
social and political environments of voters’ communities (e.g.,
residential, occupational) are assumed to affect their political orientations
through various contextual interventions (Sprague, 1982).
The main reason for this is that individuals’ embeddedness within a given
social, cultural or political context structures their social interactions. It
both constrains and enables their communicative practices within particular
patterns of behavior. However, as Knoke remarks (1990a,
p. 47), the two types of data that traditionally contextual analysts have
employed, do not suffice. These two types are individual attributes and social
unit composition. Knoke argues that a third type of data is needed, the
so-called individuals’ egocentric network. This will avoid erroneous
inferences on the impact of contextual influences on voters’ choices. In fact,
Knoke states that the egocentric “micro-nets may serve as filters that
connect individuals to the larger neighborhood and community social
structures” and “through neighboring relations, friendships, and
work groups, local residents are exposed to selective portions of their
immediate political context” (1990a, p. 47). Thus, Knoke
argues that a network theory of voting should consider the “effects of
both form and content of the egocentric relationships upon an actor’s
vote decision” (1990a, p. 48). The form of an
egocentric network is described by the number of alters to whom an ego is
linked and the strength of these ties, both between the ego and the alters and
among the alters. Moreover, “the content of an egocentric network
consists of exchanges of political information that may influence ego’s
perceptions of political choices and their consequences” (1990a, p. 48).
For example, Laumann in his study of Detroit white men’s voting choices
found that the more densely linked the adjacent alters of an ego were, the
greater their political homogeneity of the network (Laumann,
1973, p. 123). On the opposite end, star-like linked egocentric networks
tend to be less politically homogeneous and more tolerant to political
extremism (1973, p. 127).
Similarly, in considering both form and content dimensions, Knoke (1990a, p. 49) formulated the following two propositions
on the egocentric network effects on voting choices: (1) the influence on ego
of strongly tied adjacent alters is higher than the influence of weakly tied
alters and (2) the more political homogeneous the ego’s adjacent alters,
the higher the influence on the ego to adopt her alters choices. In fact, Knoke
(1990b) was able to test hypotheses shaped by the
above two propositions on data of the 1987 General Social Survey. Of course, an
extensive literature on social networks and voting choices exists already. For
instance, Zuckerman, Valentino, and Zuckerman (1994) have developed an
interesting structural theory of vote choice arguing that, as individuals vary
in their membership in mutually reinforcing social and political networks, they
vary in the likelihood of persistently voting for the same political party and
never voting for other parties. However,
the fact is that the relationship between personal or individual attitudes and
network effects or influences is quite complex. As David Lazer (2001)
has been stressing, such processes should be canvassed on the basis of a
co-evolution between network and individual; individuals simultaneously shape
and are shaped by the networks in which they function. Thus, by exploring
certain longitudinal, attitudinal and sociometric data (from a A different
stream of models about the formation of collective public opinion does not
focus on aggregates but rather details of individual interaction. Theories of
this sort often describe public opinion formation during an election or another
kind of political decision. Following some insights of early work by McPhee and
Smith (1962) and later explored by Huckfeldt
and Sprague (1995), individuals are seen as
parts of loosely knit, flexible networks in which information transmission
occurs through political discussions. Individuals adjust their opinions on the
basis of the perceived quality of the information from individual discussants
and other factors. In Huckfeldt’s analogy (2001),
the formation of public opinion is like collecting the conclusions of thousands
of individuals serving on different juries. From the latter perspective, it is
important to study a theory of political communication, which investigates the
extent to which citizen discussion beyond the boundaries of cohesive groups
might influence the dissemination of public opinion. According to Huckfeldt,
Beck, Finally, I
should mention that often theories of network influence on political
participation are understood from the point of view of social capital (cf., Coleman, 1990; Putnam, 1993),
that is, through studies of how social capital is converted into civic
participation. In this way, considering social capital as a determining
dimension of individual behavior (Lake &
Huckfeldt, 1998; Putnam, 2000; Verba
et al., 1995), it means
that social capital should not be treated as an attitude (e.g., norm, social
trust or tolerance) but rather it should be conceptualized as a resource of
social structure (Newton, 1999). As social networks are
structuring the flow of information surrounding an individual (Granovetter, 1973), the existence of large but
loosely coupled networks linking individuals into a wider context increases the
probability of exposure of the linked individuals to appeals for political
action. In practice, this happens either through participation in informal
networks or membership in formal groups (as voluntary associations). Hence, in
this sense, social capital is related to both “social recruitment”
and “political mobilization” (Teorell, 2000). Formalization If I want
to study how the network concept works in the context of the above theories,
getting a handle with empirical work is indispensable. Among many textbooks in
the field, one could mention a few characteristic: David Knoke’s Political Networks: the Structural
Perspective (1990a), Robert Huckfeldt and John
Sprague’s Citizens, Politics, and
Social Communication: Information and Influence in an Election Campaign (1995) and Gerald Marwell and Pamela Oliver’s
The Critical Mass in Collective Action: A
Micro-Social Theory (1993). However, beyond the
empiricist tradition, there is a second – and complementary –
intellectual strategy that I will employ here in this paper. This is the
theoretical strategy of constructing and investigating formal mathematical
representations of the processes enacted in network formations of public
opinion. To follow
such a mathematical formalization in modeling social phenomena takes for
granted that this mathematical modeling is a desirable tool to study social
realities. But this is far from being unanimously accepted: many theorists
typically reject formal modeling in social research because they are contesting
the prominence of positivist mathematical tractability and empiricist
falsifiability. For instance, as Ian Shapiro and Alexander Wendt (1992) claim, such a formalization might
fall into the trap of rationalist “logicism” when it degenerates
into “an exercise in trying to derive an ever widening class of phenomena
from the theory rather than an attempt to validate the theory
empirically” (1992, p. 202). Thus, Shapiro and
Wendt warn: “By confusing what can never be more than devices for
hypothesis generation with the conduct of social science itself, they often
lose sight of the phenomena that their theories purport to explain, and the
disputes about the fine points of analytical models that occupy much of their
attention often reside so deeply in a world of counterfactuals that they would
never be tested empirically” (1992, p. 203). Nevertheless,
many theorists often realize the potential of formal explorations as a
necessary complement to its alternative – verbal analysis. Thus, the
German sociologist Norbert Elias has tried to combine the two possibilities
offered by qualitative theorizing and formal modeling by claiming that
“the social apparatus for thinking and speaking places at our disposal
only either models of a naively egocentric or magico-mythical kind, or else
models from natural science” ([1970] 1978,
p. 17). Similarly, in
their criticisms of mathematical modeling, Shapiro and Wendt do not neglect to
discuss the potential of formal research as a heuristic to test hypotheses and,
thus, to build theories:
“Taking an explanation and running with it, driving an as-if
causal theory to the hilt, may reveal as faulty the assumptions that hitherto
had been taken for granted and may generate research problems and hypotheses
for investigation that otherwise would not have been thought of” (Shapiro & Wendt, 1992, p. 203). In fact, discussing the
counterfactual alternative to empirical generalizations, Max Weber (1949) has already suggested that the former “involves
first the production of – let us say it calmly – “imaginary
constructs” by the disregarding of one or more of those elements of
“reality” which are actually present, and by the mental
construction of a course of events which is altered through modification in one
or more ‘conditions’” (1949, p.173).
Furthermore, it should not escape our attention that through
“experimentation” of formal models, albeit in the context of a
reification inside a formal-artificial world, one can better appreciate the
complexity of the unanticipated consequences of human choices and social
agencies (Giddens, 1979, p. 258). Therefore,
besides its epistemological limitations, formal modeling may lead to theory
development by revealing ambiguities among “raw” and unprocessed
data and by formulating questions which might never be raised by verbal theory
alone. Formalization necessarily involves detailed descriptions and
specifications of the mechanisms or processes involved in the phenomenon to be
modeled. On the one hand, verbal theory can only roughly – sometimes
vaguely – sketch interdependencies or causalities among the various
relationships appearing in the studies phenomena. On the other hand, formal
theory can spell out answers of how these interdependencies, causalities and
relationships work out at least in an idealized or counterfactual context,
which might serve as a heuristic for the very understanding of the phenomena.
In this sense, formal modeling may scrutinize and filter away all the
ambiguities of verbal theorizing by providing the experimental ground on which
more theoretically complete and empirically consistent reasoning may develop. A Simulation In this section, I am going to discuss an
example of a formal modeling – in the form of a simulation – as an
attempt to understand how the mechanisms of networked formation of public
opinion work in an idealized and counterfactual setting of this simulation. In
this context, many theorists have extensively considered simulations as means
to address questions about the crisscrossing between social and political
networks and their relation to consequences of influence, disagreement,
ambivalence, engagement, participation and voting choices (e.g., Johnson, 1999; Huckfeldt, Johnson,
Sprague, & Craw, 2000; Ikeda &
Huckfeldt, 2001). Here I will narrow down the problem to a family
of simulations which are following Axelrod’s (1997)
suggestion that agent-based models can be useful tools for “thought
experiments” and clarification of theory. Of course, these are not the
only simulations used to analyze social emergent phenomena in political
processes: Latané’s theory of social impact (Latané,
1981; Latané et al., 1994) is another very
well known model providing powerful political (and psychological) simulations. Axelrod’s
original simulation is based on an “adaptive culture model” (as it
was named by Kennedy, 1998), that is, a model of
dissemination of culture through social interaction proceeding by local
convergence and resulting emergence of global polarization at a limited degree.
In fact, Axelrod is dealing with the dissemination of culture over a
rectangular grid (regular lattice) the nodes (or cells) of which are supposed
to be the interacting agents. Each agent possesses a list of numerical
attributes, which Axelrod is interpreting as the agents’ cultural
characteristics. The results of Axelrod’s simulation are indicating that
over the long run the assumed processes of local convergent interactions are
globally homogenizing the culture: almost all individuals are adopting the same
culture or in the best case only a very small number of cultures are surviving
throughout the model. In other words, the outcome of Axelrod’s simulation
is either a complete cultural homogenization or a fragmentation into a small
number of heterogeneous cultural zones. Subsequently,
many scholars (Kennedy, 1998, 1999;
Shibanai et al.,
2001; Greig, 2002) have tried to formulate variants of
Axelrod’s simulation in order to attain a higher degree of global
polarization or cultural heterogeneity, which would be a more realistic model
of real cultural processes. The necessity of breaking away from a
“homogenization prediction” is attributed by Johnson and Huckfeldt
(2001) partly to empirical reasons and partly to
normative reasons. On the one side, recent empirical studies have shown that
high diversity and considerable disagreement can be sustained by interpersonal
networks (Huckfeldt, Johnson & Sprague, 2002).
On the normative side, as Johnson and Huckfeldt (2001)
have argued, “this model has the implication that interaction does erase
all differences” and, thus, “one might argue in favor of
segregation, or cultural apartheid, as the only way to preserve diversity on
the aggregate social level,” which of course is a rather discouraging
contention. Here, I
would present a similar simulation (analytically developed in Boudourides, 2003) in an effort to overcome the restrictions of the
produced homogenization in the original Axelrod’s model and to manifest
that significant diversity and heterogeneity can be sustained over general
networks of agents. For this purpose, before showing our simulation in action,
let’s present where our simulation differs from the previous ones. These
differences appear at three spots: (1) the interpretation of agents’
attributes, (2) the positioning of agents and their linkage patterns and (3)
the dynamics of interactions among agents. The Interpretation of Agents’ Attributes Perhaps this
point is of minor importance but I am referring to it in order to make our
simulation meaningful at the context of the present paper (which is a review of
theories of the formation of public opinions, voting choices, preferences of
political participation, etc.). As Axelrod did, agents’ attributes are
represented by F features (typically, F = 5) and its feature takes values,
called traits, in the set of numbers {0, 1, …, T} (typically, T = 9,
i.e., 10 traits). Now, agents’ features were described by Axelrod as
cultural characteristics but I may follow Johnson’s and Huckfeldt’s
(2001) suggestion and think of these
features as opinions, issue stances, political allegiances or any other
judgments the agents may have on the political sphere which would orient their
voting choices or their political participation in one way or another. Agents’ Positioning In the
original Axelrod’s model as in all of the subsequent simulations that we
are aware of, agents are located at the vertices of a regular lattice (a
rectangular grid) in such a way that each of them is directly linked with four
other agents (with the exception of the side agents – unless boundary
periodicity is assumed). Besides the fact that such a social topology is
extremely ideal and regular, it happens to be a rather unrealistic one. In
fact, from modern theories of social networks and complexity studies, we know
that social aggregates – at least above a certain size which makes them
comparable to large social groups, communities or even society itself –
are composing networks of the so-called “small-worlds” (Milgram, 1967; Watts, 1999). These
networks are rather highly clustered but also they possess many shortcuts among
their nodes, which make them somehow in the middle of the hierarchy between
regular lattices and random graphs (Watts & Strogatz,
1998; Watts, 1999). Therefore,
the social topology (for the relative positioning of the agents and the linkage
among them) we are using in this simulation is that of an arbitrary graph
(“small-world” effects considered elsewhere. As mentioned before,
the graph topology is more realistic than the regular lattice (or grid)
especially when the number of nodes (agents) is large. However, even with
networks of a small number of nodes (agents), the graph topology is certainly
more appropriate to the study of network effects because we can test the
simulation over different fundamental kinds of network topologies (rings, stars,
bridges or relays etc.) in order to explore the role played by various
structural parameters on the dynamics of the studied model. Mixed Dynamics of Interactions
In order to
derive a higher degree of diversity and heterogeneity in the emergent patterns
of the simulation, our basic idea is to consider two types of agents and to
define different rules of interaction for each of them. Since Axelrod’s
convergent interactions were only producing a minimal degree of diversity
(almost homogenization), we were wondering whether the addition of a small
amount of divergent interactions would dramatically increase the level of
diversity. In fact, we saw in our simulations that this was the case. To
implement this mixed architecture of co-existing convergent and divergent
interactions among agents, we thought to divide the population of agents into
two groups, homophilic and heterophilic agents: ·
Homophilic
agents are the ones determined to sustain convergent interactions with other
agents (in various ways depending on whether the other agent with whom they are
interacting is homophilic or heterophilic). In other words, homophilic
agents’ disposition is to sustain convergent or amicable interactions
with other agents leading to agreement or adoption of the otherness, which tend
to produce more similarities among the agents’ features (e.g.,
interpreted as opinions, positions). ·
Heterophilic
agents are the ones determined to sustain divergent interactions with other
agents (again in various ways depending on whether the other agent with whom
they are interacting is homophilic or heterophilic). In other words,
heterophilic agents’ disposition is to sustain divergent or contentious
interactions with other agents leading to disagreement or rejection of the otherness,
which tend to proliferate dissimilarities among the agents’ opinions or
positions (their features). At this
point, we need to give some theoretical motivations for the above types of
interactions. On the one hand, “homophily” is based on the social
psychological “law of attraction,” that is, the tendency of human
beings to interact with similar ones (Homans, 1950; Verbrugge, 1977; Kandel, 1978).
Although quite often homophily is attributed to social influence from social
network neighbors, sometimes an alternative explanation is also given.
According to this alternative conceptualization, group homogeneity is not so
much the product of the interactions of those who are similar but it is could
also be the byproduct of repulsion among those who are different (Rosenbaum, 1986) – for instance, xenophobia might
cluster together homogeneous portions of the population. On the
other hand, if we focus on political processes and we interpret
“heterophily” as the expression of political disagreement (Huckfeldt & Sprague, 1995), then the
realization of disagreement among citizens is central to political deliberation
to the extent that political deliberation among citizens is central to
democratic politics. This is due to the following two reasons. First, the
importance of disagreement and, thus, heterophily usually is attributed to the
fact that, if people do not disagree through their interactions, their own
political views are never challenged and consequently there is no social
communication or deliberation in this case (McPhee, 1963).
Second, it is personal experience of political disagreement (heterophily) among
citizens what sustains democratic politics. In fact, otherwise, when citizens
do not disagree with each other at levels which are kept within manageable
bounds, democratic politics is deprived of its vital defining characteristics (Lipset, 1981) and it is rather considered to be
destabilized. Coming back
to our simulation, the way we denote whether an agent is homophilic or
heterophilic is by putting a sign in front of the agent’s features: +[T1T2…TF]
would be a feature of a homophilic agent and -[T1T2…TF]
would be a feature of a heterophilic agent. Now, we are
in the position to state explicitly the rules of our simulation, which is based
on mixed and opposite interactional dynamics: Step 0. Initially we have the following: ·
a
graph of N nodes over which the population of agents is accommodated; ·
a
randomly chosen distribution of features Φk, k = 1, …, N, for each agent
k (where Φk is a signed list of F numerical traits of the
form Φk = ±[T1kT2k…TFk]); ·
a
proportion n+ (0 £ n+ £ 1) of homophilic agents and a proportion n-
= 1 – n+ of heterophilic agents. Step 1. At random, pick an agent i and then
pick one of its neighbors j (adjacent nodes in the graph). Step 2. Compare Φi with Φj and then modify Φi as follows: ·
If
both agents i and j are homophilic, then select randomly a feature on which i and
j have different traits and with probability equal to the similarity of the
traits of their features change the trait on this feature of i to take the
value of the trait of this feature on j (this is exactly Axelrod’s rule
of convergent interactions). ·
If
agent i is homophilic but agent j is heterophilic, then select randomly a
feature on which i and j have different traits and with probability equal to
the half of their similarity of traits change the trait on this feature of i to
take the value of the trait of this feature on j. ·
If
both agents i and j are heterophilic, then select randomly a feature on which i
and j have the same trait and with probability equal to the dissimilarity of
their traits change the trait on this feature of i to take randomly any value
which is different from the common value of i and j. ·
If
agent i is heterophilic but agent j is homophilic, then select randomly a
feature on which i and j have the same trait and with probability equal to the
half of the dissimilarity of their traits change the trait on this feature of i
to take randomly any value which is different from the common value of i and j. Step 3. Repeat Step 1 and so on. Apparently,
our algorithm coincides exactly with that of Axelrod’s adaptive culture
model when n+ = 1, i.e., all agents are homophilic. However, just
adding a small proportion of heterophilic agents is dramatically increasing the
diversity of the equilibrium state towards which the iterations of the
simulation eventually converge. To see this, we are measuring the groups of
agents (independently of the sign of their features) composed of agents with
identical features (better said: identical traits in all their features).
Let’s call them “equifeatured” groups and let G denote the number
of equifeatured groups of agents in their population. Initially, because of the
random assignment of features, G is almost equal to N, i.e., almost all agents
have different features. When the simulation terminates (its iterations
converge), we are interested in knowing how much bigger than 1 G is, because G
= 1 would signify homogenization while G > 1 diversity. Of course, when n-
® 1 (i.e., almost all agents are heterophilic),
we know that finally G ® N. What is more surprising is that
dropping a tiny touch of disagreement in a population of agents who are
predisposed to interact convergently stabilizes the final level of diversity at
a considerably high value, which is by all means much higher than the rather
poor level of diversity attained by Axelrod’s model, i.e., without any
“spirit of disagreement” inside the population of agents. Elsewhere
we intend to give a detailed analysis and statistics supporting this claim of
ours. Here we are only going to give a couple of examples. The first
example is of a one-dimensional lattice of 100 nodes (agents) with degree 4
(which is topologically equivalent to the rectangular grid of 10 x 10 agents
with periodic boundaries). Here we are treating the case F = 5 and T = 10,
i.e., all the agents have five features and the traits of each one of them
takes a numerical value from 0 to 9. We are assuming that n- = 0.02,
i.e., 2 agents are heterophilic and 98 are homophilic. After running our
simulation 10 times, we found that finally the average number of attained equifeatured
groups is 18.6, which is much bigger than 3.2, Axelrod’s estimate of
“stable regions” under the same conditions. Unfortunately,
the graph of the previous example is too large to be appropriately visualized.
For this purpose, we will give a second example with a smaller graph: A
one-dimensional lattice of 25 nodes (agents) with degree 4 (which is
topologically equivalent to the rectangular grid of 5 x 5 agents with periodic
boundaries). In this example, to compare with the corresponding Axelrod’s
measures, we are taking F = 5 and T = 15. Furthermore, we are assuming that n-
= 0.04, that is, 1 agent is heterophilic and 24 are homophilic. After running
our simulation 10 times, we found that finally the average number of attained
equifeatured groups is 11.7, which is bigger than (approximately) 7,
Axelrod’s estimate of “stable regions” under the same
conditions. A visualization of an initial network follows: Figures
Figure 1: The initial (random) distribution of features in the network. Note that
agents possess dissimilar features (black links) at most coinciding in two
traits (the light gray link). Note that
the heterophilic agent is almost at the centre of the graph. There is one light
gray link, that is, two agents with two similar traits, and all the remaining
links are black, that is, all the remaining 24 agents either have one common
trait or they are completely dissimilar (pair wise). Leaving the
simulation to run for one million iterations, the visualization of the final
network is as in the following figure. Notice that now we have eventually ten
different equifeatured groups: one clique of seven agents, one clique of six
agents, one clique of three agents, two cliques of two agents each and five
isolated agents (one of which is the heterophilic agent).
Figure 2: The final
(equilibrium) distribution of features forming ten equifeatured groups of
agents. Light gray links signify complete similarity and black links complete
dissimilarity. Elsewhere (Boudourides, 2003), we are presenting further results
of this simulation and to explore how different network topologies (rings,
stars, bridges or relays, etc.) influence the formation of the final
equifeatured groups. Conclusions After discussing
the network concept (through social and policy networks) and its relevance on
theories of public opinion formation (through mechanisms of collective action
and political participation), we have presented in this paper a concrete
simulation, which exemplifies a number of aspects of how network
interdependences are developed in an idealized context of a mathematical model.
This simulation is an extension of Axelrod’s adaptive culture model,
where it is assumed that culture disseminates through convergent local
interactions, eventually producing a global cultural landscape. However, when
convergent interactions are formalized as mimetic modifications of
actors’ attributes, it is not surprising that the diversity levels in
such an emerging global landscape tend to be rather low. In fact, the theory of
dynamical systems in mathematics suggests that contractive modes of dynamical
transformations are always attracted by simple equilibrium states of low
complexity. But, in this formal (mathematical) sense, a route to produce chaos
and complexity would be through a “hyperbolic” mixing of both
contractive and expansive modes of interaction. This suggestion seems to work
out in our simulation too: Using Lazer’s (2001)
terminology, by increasing the range of possible plastic deformations through
the mixing of consensual with oppositional actors, even if the latter are very
few, the emergent equilibrium patterns are much more complex than without the
presence of the latter. Of course, to explain or interpret what role the
effects of contentious behavior play in cultural dynamics, opinion formation,
political processes, social communication or collective action is beyond the
scope of this paper. The claim we make here in this simulation that an
infinitesimal disturbance to local norms suffices to produce higher global
diversity of the equilibrium outcomes of an agent-based simulation (of
networked actors’ attributes) cannot be considered as bearing any
normative implications. It is simply a theoretical observation of the immanent
structural instability of such a stochastic (and nonergodic) simulation, which
was manifested in the computer experiments we have performed. References Adler, E., & Haas, P. M. (1992). Epistemic communities, world order and the creation of a reflective
research program. International
Organizations, 46(3):
367-390. Axelrod, R. (1997). The dissemination of culture: A
model with local convergence and global polarization. Journal of Conflict Resolution, 41,
203-226. Bahr, D. B., & Passerini, E.
(1998a). Statistical
mechanics of ppinion formation and collective behavior: Micro-sociology. Journal of Mathematical Sociology, 23, 1-27. Bahr, D. B., & Passerini, E. (1998b).
Statistical mechanics of collective behavior: Macro-sociology. Journal of Mathematical Sociology, 23, 29-49. Berelson, B., Lazarfeld, P. F., & McPhee, W. N.
(1954). Voting: A study of opinion
formation in a Presidential campaign. Boudourides,
M. (2003). A simulation of convergent-divergent public opinion formation on
social networks. Paper presented at the Sunbelt
XXIII International Social Network Conference, Börzel,
T. A. (1997). What’s so special about policy networks? An exploration of
the concept and its usefulness in studying European governance. European
Integration online Papers (EIoP),
1, 16. Callon, M.
(1994). Is science a public good? Science,
Technology, & Human Values, 19(4),
395-424. Coleman, J.
(1990). Foundations of social theory.
Elias, N.
[1970] (1978). What is sociology? Giddens, A.
(1979). Critical problems in social
theory. Granovetter,
M. (1973). The strength of weak ties. American
Journal of Sociology, 78,
1360-1380. Granovetter,
M. (1974). Getting a job. Greig, J. M.
(2002). The end of geography? Globalization, communications, and culture in the
international system. Journal of Conflict Resolution, 46(2), 225-243. Haas, P. M.
(1990). When knowledge is power. Haas, P. M.
(1992). Introduction: Epistemic communities and international policy coordination.
International
Organization, 46(1), 1-35. Hardin, R.
(1982). Collective action. Homans, G.C.
(1950). The human group. Huckfeldt, R. (2001). The social communication of political
expertise. American Journal of Political
Science, 45, 425-438. Huckfeldt, R. R., Beck, P. A., Dalton, R J., & Levine, J.
(1995). Political
environments, cohesive social groups, and the communication of public opinion. American Journal of Political Science, 39(4), 1025-1054. Huckfeldt, R., Johnson, P.E., & Sprague, J.D. (2002).
Political environments, political dynamics, and the survival of disagreement. Journal of Politics, 64(1), 1-21. Huckfeldt, R., Johnson, P. E., Sprague, J. D., & Craw,
M. C. (2000). Influence, communication, and the survival of political
disagreement among citizens. Presented at the annual meeting of the American
Political Science Association, Huckfeldt, R.,
& Sprague, J. D. (1995). Citizens, politics, and social communication: Information and influence
in an election campaign. Ikeda, K., &
Huckfeldt, R. (2001). Political communication and disagreement among citizens
in Johnson, P.
E. (1999). Protests, elections, and other forms of political contagion.
Presented at the 1999 annual meeting of the American Political Science
Association, Johnson,
P. E., & Huckfeldt, R. (2001). Persuasion and political heterogeneity
within networks of political communication: Agent-based explanations for the
survival of disagreement. Presented at the annual meeting of the American
Political Science Association, Kandel, D.
B. (1978). Homophily, selection, and socialization in adolescent friendships. American Journal of Sociology, 84, 427-436. Kenis, P.,
& Schneider, V. (1991).
Policy networks and policy analysis: Scrutinizing a new analytical toolbox. In
B. Marin & R. Mayntz (Eds.), Policy networks, empirical evidence and
theoretical considerations (pp. 25-59). Frankfurt
a. M.: Campus Verlag. Kennedy, J.
(1998). Thinking is social; experiments with the adaptive culture model. Journal
of Conflict Resolution, 42(1),
56-76. Kennedy, J. (1999). Minds and cultures: Particle swarm
implications for beings in sociocognitive space. Adapted Behavior, 7(3-4), 269-287. Kim, H.,
& Bearman, P. S. (1997). The structure and dynamics of movement
participation. American Sociological
Review, 62, 70-93. Knoke, D.
(1990a). Political networks: The
structural perspective. Knoke, D. (1990b). Networks of political action: Toward theory
construction. Social Forces, 68(4), 1041-63. Latané,
B. (1981). The psychology of social impact. American Psychologist,
36, 343-356. Latané, B., Nowak, A., & Liu, J. H. (1994). Measuring emergent social phenomena:
dynamism, polarization, and clustering as order parameters of social systems. Behavioral
Science, 39, 1-24. Laumann, E.
O. (1973). Bonds of pluralism: The forms
and substance of urban social networks. Lazarfeld,
P. F., Berelson, B., & Gaudet, H. (1948). The people’s choice: How the voter makes up his mind in
Presidential campaigns. Lazer, D.
(2001). The co-evolution of individual and network. Journal of Mathematical
Sociology, 25(1), 69-108. Lipset,
S. M. (1981). Political man (Expanded edition). Lohmann, S.
(1994). The dynamics of informational cascades: The Monday demonstrations in Leipzig,
East Germany, 1989-91. World Politics, 47, 42-101. López,
J., & Scott, J. (2000). Social
structure. Buckingham: Open University Press. Marsden, P.,
& Campbell, K. E. (1984). Measuring tie strength. Social Forces, 63,
482-501. Marsh, D.
(1998). The development of the policy network approach. In D. Marsh (Ed.), Comparing policy networks (pp. 3-17). Buckingham: Open
University Press. Marsh, D., & Rhodes, R. A. W. (1992). Policy
communities and issue networks. Beyond typology. In D. Marsh & R. A. W.
Rhodes (Eds.), Policy networks in British government (pp. 249-268). Marwell, G.,
& Oliver, P. (1993). The critical
mass in collective action: A micro-social theory. Marwell, G.,
Oliver, P. E., & Prahl, R. (1988). Social networks and collective action: A
theory of the critical mass, Mayntz, R. (1993).
Modernization and the logic of interorganizational networks. In J. Child et al. (Eds.), Societal change between market and
organization (pp.
3-18). McPhee, W., & Smith, R. (1962). A model for
analyzing voting systems. In W. McPhee & W. Glaser (Eds.), Public opinion and Congressional elections
(pp. 123-179). McPhee,
W. N. (Ed.). (1963). Formal theories of
mass behavior. Milgram, S. (1967). The small-world problem. Psychology Today, 1,
60-67. Ohlemacher,
T. (1996). Bridging people and protest: Social relays of protest groups against
low-flying military jets in Oliver, P.
E. (1993). Formal models of collective action. Annual Review of Sociology, 19,
271-300. Oliver, P. E., & Marwell, G. (1988). The paradox
of group size in collective action: A theory of the critical mass, II. American Sociological Review, 53, 1-8. Oliver, P., & Marwell, G. (2001). Whatever
happened to critical mass theory? A retrospective and assessment. Sociological Theory, 19(3), 292-311. Oliver, P. E., Marwell, G., & Texeira, R. (1985).
A theory of the critical mass, I. Interdependence, group heterogeneity, and the
production of collective action. American
Journal of Sociology, 91(3),
522-556. Olson, M.,
Jr. (1965). The logic of collective
action: Public goods and the theory of goods. Prahl, R.,
Marwell, G., & Oliver, P. E. (1991). Reach and selectivity as strategies of
recruitment for collective action: A theory of the critical mass, V. Journal of Mathematical Sociology, 16(2):
137-164. Putnam, R.
(1993). Making democracy work: Civic
traditions in modern Italy. Princeton: Putnam, R.
(2000). Bowling alone: The collapse and
revival of American community. Rosenbaum,
M. E. (1986). The repulsion hypothesis: On the non-development of
relationships. Journal of Personal and
Social Psychology, 51,
1156-1166. Shapiro, Shibanai,
Y., Yasuno, S., & Ishiguro, Scott, J.
(2000). Social network analysis (2nd
ed.). Sprague, J.
(1982). Is there a micro-theory consistent with contextual analysis? In Teorell, J.
(2000). A resource model of social capital: Networks, recruitment and political
participation in Verba, S., Schlozman, K. L., & Brady, H. (1995). Voice and equality: Civic
voluntarism in American politics. Verbrugge, L. M. (1977). The structure of adult friendship choices. Social Forces, 56, 576-597. Wasserman,
S., & Faust, K. (1994). Social
network analysis: Methods and applications. Weber, M.
(1949). Objective possibility and adequate causation in historical explanation.
In E. A. Shils & H. A. Finch (Eds.), The
methodology of the social sciences (pp. 164-188). Wellman, B.
(1988). Structural analysis: From method and metaphor to theory and substance.
In B. Wellman & S. D. Berkowitz (Eds.), Social structures: A network approach (pp. 19-61). Zuckerman, A.
S., Valentino, N. A., & Zuckerman, E. W. (1994). A structural theory of
vote choice: Social and political networks and electoral flows in Copyright 2004 Communication Institute for Online Scholarship, Inc. This file may not be publicly distributed or reproduced without written permission of the Communication Institute for Online Scholarship, P.O. Box 57, Rotterdam Jct., NY 12150 USA (phone: 518-887-2443). |
