406 Social Influence and Group Dynamics
and each group has the same relative proportions of strong
and weak members (high vs. low boxes). Figure 16.2 shows
the equilibrium reached after six rounds of simulated discus-
sion. Now the majority is 90% and the minority is 10%. Note
that the minority opinion has survived by forming clusters of
like-minded people and that these clusters are largely formed
around strong individuals.
These two group-level outcomes—polarization and
clustering—are commonly observed in computer simulations
(cf. Nowak et al., 1996; Latané, Nowak, & Liu, 1994) and are
reminiscent of well-documented social processes. As noted
earlier in this chapter, the average attitude in a group be-
comes polarized in the direction of the prevailing attitude as
a result of group discussion (e.g., Moscovici & Zavalloni,
1969; Myers & Lamm, 1976). In the simulations, polariza-
tion reflects the greater influence of the majority opinion. In
the initial random configuration (Figure 16.1), the average
proportion of neighbors holding a given opinion corresponds
to the proportion of this opinion in the total group. The aver-
age group member, then, is surrounded by more majority
than minority members, a difference that results in more mi-
nority members’ being converted to the majority position
than vice versa. Some majority members are converted to
the minority position, however, because they happen to be
located close to an especially influential minority member, or
because by pure accident, more minority members happen to
be at this location.
Clustering is also pervasive in social life. Attitudes, for ex-
ample, have been shown to cluster in residential neighbor-
hoods (Festinger, Schachter, & Back, 1950). Pronounced
clustering also characterizes political beliefs, religions, cloth-
ing fashions, and farming techniques. Clustering reflects the
relatively strong influence exerted by an individual’s neigh-
bors. When opinions are distributed randomly, the sampling
of opinions through social interaction provides a reasonably
accurate portrait of the distribution of opinions in the larger
society. When opinions are clustered, however, the same sam-
pling process will yield a highly biased result. Because the
opinions of those in the nearby vicinity are weighted the most
heavily, the prevalence of one’s own opinion is likely to be
overestimated. Hence, opinions that are in the minority in
global terms can form a local majority. Individuals who hold
a minority opinion are therefore likely to maintain this opin-
ion in the belief that it represents a majority position.
Control Factors for Social Influence
The results concerning polarization and clustering have been
confirmed analytically (Lewenstein et al., 1993) and have
received empirical support as well (Latané, Liu, Nowak,
Bonavento, & Zheng, 1995; Latané & Nowak, 1997). This
research has also identified several control factors that are re-
sponsible for the emergence of these macroscopic properties
(Latané & Nowak, 1997; Lewenstein et al., 1993; Nowak
et al., 1996). Individual differences in strength, first of all, are
indispensable to the survival of minority clusters. This con-
clusion is consistent with evidence demonstrating the impor-
tance of leaders for maintaining the viability of minority
opinions. The literature on brainwashing, for example, docu-
ments that natural leaders were commonly removed from the
group before attempts were made to brainwash prisoners of
war (cf. Schein, 1956). By counteracting the sheer number of
majority opinions, the strength of leaders stops minority clus-
ters from decaying. It is worth noting that as a result of social
influence, individual differences in strength tend to become
correlated with opinions. This is because the weakest minor-
ity members are most likely to adopt the majority position, so
that over time the average strength of the remaining minority
members will grow at the expense of the majority. This sce-
nario is consistent with the observation that individuals advo-
cating minority positions are often more influential than
those advocating majority positions.
A second critical control factor is nonlinearity in attitude
change. Abelson (1979) demonstrated that when individuals
move incrementally toward the opinions of interaction part-
ners as a result of social influence, the invariable outcome of
simulations is uniformity and the complete loss of minority
clusters. In the model depicted here, however, attitudes are
assumed to be categorical in nature (Latané & Nowak, 1994).
This means that individuals hold a fixed position and actively
resist influence attempts until a critical threshold of influence
is reached, at which point they switch dramatically from one
category to another rather than incrementally on a dimension
of judgment. There is empirical evidence in support of the
nonlinearity assumption for attitude topics that are personally
important (cf. Latané & Nowak, 1994). Such attitudes dis-
play a bimodal distribution, with almost no individuals occu-
pying the intermediate points on the attitude dimension. This
suggests, incidentally, that one way to achieve consensus in a
group is to decrease the subjective importance of the topic in
question.
A third critical feature concerns the geometry of the social
space (Nowak, Latané, & Lewenstein, 1994). People do not
communicate equally with everyone in a group, nor are their
interactions random. Specific communication patterns can
be approximated with different geometries of social space.
In most of the simulations, social space is portrayed as a
two-dimensional matrix ofnrows andncolumns. This geom-
etry reflects the assumption that interactions typically occur
in two-dimensional spaces, such as neighborhoods, town