ArticlePDF Available

Echo chambers and viral misinformation: Modeling fake news as complex contagion


Abstract and Figures

The viral spread of digital misinformation has become so severe that the World Economic Forum considers it among the main threats to human society. This spread have been suggested to be related to the similarly problematized phenomenon of “echo chambers”, but the causal nature of this relationship has proven difficult to disentangle due to the connected nature of social media, whose causality is characterized by complexity, non-linearity and emergence. This paper uses a network simulation model to study a possible relationship between echo chambers and the viral spread of misinformation. It finds an “echo chamber effect”: the presence of an opinion and network polarized cluster of nodes in a network contributes to the diffusion of complex contagions, and there is a synergetic effect between opinion and network polarization on the virality of misinformation. The echo chambers effect likely comes from that they form the initial bandwagon for diffusion. These findings have implication for the study of the media logic of new social media.
Content may be subject to copyright.
Echo chambers and viral misinformation:
Modeling fake news as complex contagion
Petter To
Sociology Department, University of Amsterdam, Amsterdam, The Netherlands
The viral spread of digital misinformation has become so severe that the World Economic
Forum considers it among the main threats to human society. This spread have been sug-
gested to be related to the similarly problematized phenomenon of “echo chambers”, but the
causal nature of this relationship has proven difficult to disentangle due to the connected
nature of social media, whose causality is characterized by complexity, non-linearity and
emergence. This paper uses a network simulation model to study a possible relationship
between echo chambers and the viral spread of misinformation. It finds an “echo chamber
effect”: the presence of an opinion and network polarized cluster of nodes in a network con-
tributes to the diffusion of complex contagions, and there is a synergetic effect between
opinion and network polarization on the virality of misinformation. The echo chambers effect
likely comes from that they form the initial bandwagon for diffusion. These findings have
implication for the study of the media logic of new social media.
The way we become informed, debate, and form our opinions have changed profoundly with
the advent of online media [15]. Today’s media is less organized through centralized deci-
sion-making, and more through complex cascade processes, where news items spread like
wild-fire over social networks through direct connections between news producers and con-
sumers—categories between which it is becoming increasingly hard to distinguish. Disinter-
mediation has changed the way we as a society form narratives about our common world with
promises of more egalitarian ways of meeting and discussing.
But despite early optimism about this ostensibly decentralized and democratic meeting-
place, the online world seems less and less like a common “table” that “gathers us together” [6]
(p.52) to freely discuss and identify societal problems in a new type of “public sphere” [7].
Instead, it seems to bring forth the worst of human instincts: we cluster together into tribes
that comfort us with reaffirmation and protect us from disagreement; “echo chambers” that
reinforce existing perspective and foster confirmation biases [811]. Technology that was pur-
ported to help weave tighter the bonds between humans instead seem to has led to a fraying of
our social fabric, as we are torn into social groups with separate world-views.
PLOS ONE | September 20, 2018 1 / 21
Citation: To¨rnberg P (2018) Echo chambers and
viral misinformation: Modeling fake news as
complex contagion. PLoS ONE 13(9): e0203958.
Editor: Chris T. Bauch, University of Waterloo,
Received: March 28, 2018
Accepted: August 30, 2018
Published: September 20, 2018
Copyright: ©2018 Petter To¨rnberg. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: Data for empirical
simulations are available from: To¨rnberg, Petter
(2018): Empirical simulation graphs. figshare.
Funding: This project has received funding from
the European Union’s Horizon 2020 research and
innovation programme project ODYCCEUS (grant
agreement n˚ 732942). The funders had no role in
study design, data collection and analysis, decision
to publish, or preparation of the manuscript.
Competing interests: The author has declared that
no competing interests exist.
Simultaneous with this development is an on-going reduction in the quality and credibility
of available information: while the Internet was initially hailed as an unprecedented source of
easily accessible knowledge, it increasingly appears to instead have brought an information cli-
mate characterized by biased narratives, “fake news”, conspiracy theories, mistrust and para-
noia. The digital world seems to provide fertile soil for the growth of misinformation, as
studies show that false news diffuse faster, farther and deeper than true news in social networks
[12]. This changing online climate is relevant not merely within the realm of social media, but
may influence opinions and behavior also in other areas of human life [1318]. This rapid cul-
tural shift has quickly become an onerous threat, with viral misinformation now being seen as
a major risk to human society [19].
There are certain signs that point to a link between these two phenomena—echo chambers
and the spread of misinformation—since homogeneous clusters of users with a preference for
self-confirmation seem to provide capable green-houses for the seedling of rumors and misin-
formation. A polarized digital space where users tend to promote their favorite narratives,
form polarized groups and resist information that does not conform to their beliefs may be the
fertilizer that makes the Internet so fertile for the growth of misinformation [20]. Armies of
supporters for products, brands or presidents are quickly rallied, spreading their views in an
environment where trust in traditional knowledge authorities is increasingly frail [2125].
But the potential link between echo chambers and misinformation has proven difficult to
evaluate due to the complexity underlying social media dynamics. The transition in informa-
tion dynamics that digital media has brought can be usefully seen through the notion of media
logic, i.e. the “particular institutionally structured features of a medium [. . .] that will tend to
structure particular perceptual and cognitive biases” [26] (p.63-64). This notion thus points to
structural factors of a medium, which in the case of social media implies an entwined causal of
distributed actors [27,28]. Contemporary media logic has taken a distinctly postmodern turn
with digital technology, as networks and identities now carry more weight than the truths of
traditional authorities [29]. It has gone from the structured top-down logic of the machine—
with centralized institutions and trained journalists making decisions from editorial chairs—
to the decentralized and dynamic complexity of what resembles a swarm or herd—millions of
dopamine-driven smartphone-users swiping through click-bait news on porcelain thrones [30,
31]. Such swarm logic often departs from the terrain accessible by human intuition, and travels
into the realm of emergence and complexity where neither un-aided cognition or traditional
scientific methods are able to disentangle the complex chains of causality [32].
This paper suggests an emergent causal link that could serve to connect the spread of misin-
formation and online echo chambers. The nature of this link is perhaps most clear through the
lens of a metaphor: if we think of the viral spread of misinformation in a social network as
akin to a wildfire [19], an echo chamber has the same effect as a dry pile of tinder in the forest;
it provides the fuel for an initial small flame, that can spread to larger sticks, branches, trees, to
finally engulf the forest (it should be noted that this metaphor has a specific and technical
meaning: forest fires received significant early interest within complexity theory, see e.g. [33,
34], the metaphor also recalls Schelling’s [35] threshold model). Just as how segregation has
been, famously, shown to involve emergent feedback processes [36], it is in this paper sug-
gested that an emergent network effect results in that misinformation and rumors spread easier
in networks where there is a presence of an echo chamber. The model implies that news originat-
ing in a segregated cluster of users tends to spread further than in a network without clusters.
It furthermore shows that the mere coming together of users with similar views may be enough
to increase the prevalence of misinformation, as virality increases with network homophily. It
finally suggests that the combination between these two factors, arguably what defines an echo
chambers, is synergetically more prone to induce global cascades than either taken in isolation.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 2 / 21
This could mean that echo chambers make fake news more viral, since information that reso-
nates with biased clusters of users has a higher likelihood to spread through a network.
This paper thus uses a simple network model to study whether the network structure of
echo chambers can in itself have an effect for the spread of information. Online echo chambers
are modelled as set of users characterized by opinion and network polarization, i.e. they are
clusters of like-minded users, that are (i) more separated from the rest of the network, and (ii)
have a lower threshold for being convinced by a given narrative. This approach may provide
direction for further study in online media dynamics in general, and the spread of misinforma-
tion in particular.
1 Modeling spread of misinformation on social media
The model presented here aims to be simple as this both increases generality and makes the
results of the model easier to interpret and bring into a sociological narrative [37,38]. Observ-
ing that there is a tendency for echo chambers to form around conspiracies and mistrust of tra-
ditional authorities, the question that this model asks is: if a diffusant is associated to an echo
chamber, will this affect its virality in a social network? The model thus does not deal with the
content of the diffusant, meaning that its dynamics applies to any type of information spread-
ing from an echo chamber.
The model describes the spread of a news item on social media as a complex diffusion in a
social network. This means that we understand a social network as a number of connected
nodes—representing users—that each are assigned a threshold that describes how difficult
they are to convince about a given narrative. If a large-enough fraction of their neighbors
spread a given narrative, the node will be convinced, and will continue the spread of the diffu-
sant. Hence, the threshold can be seen as representing to what extent a given news item or idea
resonates with a certain user.
In this network, we understand an echo-chamber as a set of users characterized by two prop-
erties: opinion and network polarization. Opinion polarization means that they, in relation to a
given question, are more inclined to share similar views. Network polarization means that they
are more densely connected with each other than with the outside network. In other words, an
echo chamber is a tightly connected set of nodes more inclined to share a common view on a
given narrative. For the sake of simplicity, this model focuses on the existence of a single echo-
chamber in a larger network.
This type of clusters in networks have generally been understood to constitute impediments
to the spread of diffusion, as they reduce the number of the weak ties that have in turn been
found to be central enablers of diffusion (see e.g. [3943]). Modeling results have shown that
things like information, diseases, and so on, have a harder time of spreading into and out of
clusters, since these are less connected to the surrounding network, which means that they will
also reduce the probability for network-wide cascades.
However, these results specifically concern diffusion of entities that require only single
contact to spread. This makes sense in many cases, for example disease, but it has been
argued to be a problematic assumption when studying behavioral phenomena (see e.g.
[44,45]). In these cases, complex contagion, is often a more plausible view: that multiple con-
tact increases the likelihood for a contagion to spread. In these cases, weak ties have been
shown to be less relevant, or even have a negative effect on diffusion, and there are situations
where clusters may in fact be beneficial for the diffusion. But not always: even for complex
contagions, clusters may have negative impact on diffusion. For example, Easley and Klein-
berg [46] show using mathematical reasoning that given that cascades start outside of the
cluster, that they are seen as successful only if every node in the network is activated, and
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 3 / 21
that every node has the same number of neighbors, clusters will have negative effects for
We argue that online news and rumors are better characterized as a complex rather than
simple contagions, since a news item or idea becomes more convincing with the number of
people that argue for it [44]. For example, a single user in your network claiming that vaccines
cause autism or that Russia is an important ally to the US may easily be disregarded as a nut-
job. But if a large fraction of your social network argues for the same thing, it not only makes
the argument seem more authoritative, but you may also feel the need to conform to your
social group [47]. This majority effect becomes especially clear if we think of political discus-
sions and online interaction not as much as a spread of information, but as having elements
of group identity and belonging, an understanding that is becoming increasingly prevalent
throughout the social sciences [48,49]. Echo chambers are generally understood as also play-
ing a role in the formation of interpretive frames and collective identities, rather than simply
constituting a hub for information diffusion [50].
The second aspect of the echo chambers, opinion polarization, implying that their thresh-
olds are lower than that of the surrounding network, is related to what in network terms is
referred to as homophily: the probability that neighboring nodes have similar thresholds for
activation. Moderate levels of homophily is generally understood to increase the network’s
capacity for diffusion [5154]. This can also be connected to the suggestion in critical mass
theory that an initial group can solve the large group problem by creating a “bandwagon effect”
[5558]. This model differs from studies looking only at homophily (e.g. [51]) as it combines
homophily with network clustering, to thereby capture the essence of a social network echo
The model here presented primarily differs from previous approaches in the literature by:
(i) looking at the dynamics of complex rather than simple contagions, which may have impor-
tant effects for the results; (ii) loosening the assumptions that every node in the network needs
to be activated for a cascade to be understood as complete, and that each node has the same
number of neighbors; (iii) exploring the systemic interaction between network and opinion
polarization, i.e. between lower activation thresholds and level of clustering, as these are both
characteristic of echo chambers. In other aspects, the model generally follows the established
design of existing network cascade models that focus on how different factors affect the proba-
bility for activation to diffuse in a network [44,51,59].
1.1 Model design
We refer to the probability that activation will spread to a majority of the network’s nodes,
given a certain distribution of network structures, as the “virality” of the diffusion. Since the
connections between the nodes represent the existence of mutual social contact, the model
uses undirected ties. The distribution of these ties follows the so-called Erdős-Rényi structure,
where each tie is assigned with uniform probability, which is commonly used in this type of
models [43]. The distribution of ties in an Erdős-Re
´nyi network follows a Poisson distribution.
To form the clusters, which here represent free social spaces, a fraction of the ties that span
from inside the cluster to outside the cluster are removed and replaced with ties inside the clus-
ters, resulting in more internal cluster ties than external. The fraction between internal and
external ties represents the previously introduced notion of network polarization. For example,
with a polarization value of 0.85, 85% of the cluster’s external connections are relocated, while
a polarization value of 0 represents a standard Erdős-Re
´nyi structure. This definition of net-
work polarization captures the idea that echo chambers are clusters of more densely connected
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 4 / 21
A cascade is initiated by the activation of a randomly selected node and its neighbors. This
could for example be in the form of an actor posting a blog post or tweet with misinformation.
In each following time step, nodes that have more than a certain fraction—called their thresh-
old—of their neighborhood activated themselves become activated—what has been referred to
as a complex contested contagion [44]. This continues until a steady state is reached. If at this
point, a majority of the network has become activated, the cascade is classified as successful.
Since the focus is on echo chambers, we assume that the cascade is initiated inside the clus-
ter. Furthermore, we implement the previously introduced notion of opinion polarization as a
parameter lowering the activation threshold for the cluster nodes relative to the average thresh-
old, meaning that the nodes inside the cluster are more easily activated. This of course trivially
has an effect of increasing the virality, and we hence compare to a control case, in which the
reduction of threshold is assigned to random nodes in the entire network.
The primary question that we aim to investigate with this model is whether echo chambers
have any effect on the virality of cascades, and to furthermore look at the interaction between
opinion and network polarization in terms of virality.
1.2 Model implementation
We define: P
as the network polarization parameter, P
as opinion polarization parameter, kas
the average degree of the nodes, θis the activation threshold, cis the fraction of nodes belonging
to the echo chamber. Eis the total number of edges, Nare number of nodes. For each combina-
tion of steps of parameter values, the model is run 1,000 times, with a new network structure
generated for each run to compensate for network structure heterogeneity and allow for higher
generality and robustness. Parameters P
, and θare systematically varied over an interval, to
find how these parameters affect the model (additionally, other parameters where tested for sen-
sitivity in separate runs.) With 100 steps in P
and 200 steps in θand 12 steps in P
, the model is
run a total of 2.4 10
times, which is arguably a reasonably thorough exploration of parameter
space, thereby allowing an in-depth assessment of the model dynamics. Furthermore, these runs
were evaluated for a number of network sizes, to further validate their robustness and relevance.
In each such run, an Erdős-Rényi network structure is constructed in the following way. N
nodes are created, cN of which are specified as belonging to the echo chamber. The specified
mean degree kis divided by two (since the network is undirected and any edge has two sides)
and multiplied with the number of node edges, i.e. E¼Nk
2. From this set, P
kE edges are
selected where exactly one of the connected nodes belong to the cluster. These are removed,
and replaced by edges where both nodes belong to the cluster. Following this, nodes outside
the cluster are set to have activation threshold θ, and nodes in the cluster are set to have thresh-
old θP
. A random node in the cluster is then selected, and set as activated, to act as activa-
tion seed. All nodes connected to this node are also set as activated. After this, the following is
repeated until no more changes occur: for each node, if the fraction of activated connected
nodes is larger than the node threshold, the node is set as active in the next step. If the fraction
of active nodes at the end of such a run is larger than 0.5, the cascade is considered global (this
definition prevents the mere activation of the echo chamber to affect the measured level of vir-
ality, and follows previous research, see e.g. [45]). This procedure is repeated 1,000 times with
new random network structure for each parameter step. Virality,V(θ,P
), is defined as the
fraction of times in these runs that a majority of the nodes are activated.
2 Results and analysis
We will now analyze the results of the model through a step-by-step process illustrating and
describing the model output through different graphs.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 5 / 21
2.1 Relation between network polarization (P
) and virality (V)
We start by looking at how the presence of a network polarized cluster affects the virality—i.e.
the likelihood that the complex contagion diffuses to a majority of the network nodes—with-
out taking opinion polarization into account.
Looking at Fig 1, it is clear that the presence of a cluster has a positive effect on virality.
For example, at θ= 0.270, a completely random network without cluster has an around 65%
chance of a global cascade, while for Pn¼Pc
n(i.e. at the optimal level of network polarization
with regard to virality) it is more than 85%. This is intuitively rather unexpected, as it implies
that the network polarization itself—rather than only together with opinion polarization—
impacts through emergent structural effects. This implies that it is not necessary for the indi-
viduals associated to echo chambers to be unusually inclined to believe a narrative for them to
have disproportionate effect on its diffusion: the simple fact that the diffusion originates in an
isolated cluster is enough to increase virality. It should also be noted that this is yet another dif-
ference in the cascade dynamic between complex and simple contagions [44].
We can see in Fig 1, that the effects of having a cluster increases gradually from around
= 0.4, and peaks at P
= 0.6. At P
>0.7, virality falls quickly, likely as it becomes increas-
ingly difficult for the cascades to spread from the cluster as the number of external connections
are reduced.
As can be seen in the figure, by comparing the slopes of the different lines, the effect of
having a cluster depends strongly on the threshold level: for threshold levels where no cas-
cade is possible, or where cascades are almost certain to occur, the presence of clusters of
course has little effect. The threshold levels that are most interesting are thus the ones where
the effect of the presence of the echo chamber is as large as possible, i.e. the θ
for which
n;PoÞ  Vðyc;0;PoÞ(henceforth denoted DVðyc;Pc
n;PoÞ) is maximized, for some given
and where Pc
nis, similarly, the network polarization level for which virality peaks. In the
following analysis, we will thus focus on these threshold levels, which we refer to as the “criti-
cal threshold”, or θ
. As can be seen in Fig 1, the effect of varying θis far from linear and,
unsurprisingly, has a large impact on virality. To further investigate the effects of changing
threshold levels, we begin by looking at how varying the average threshold affects virality for
different parameter settings.
2.2 Relation between virality (V) and threshold (θ)
We now also introduce opinion polarization, P
, in the echo chambers, i.e. the threshold is
reduced inside the cluster compared to the outside. We begin by looking at how the average
threshold level, θ, affects virality. This sheds light on the question of how the quality or reso-
nance of a certain news item or idea affects its possibility to spread in the network, which in
turn constitutes a central aspect of the media logic of social media. Is the spread of an idea
directly proportional to the resonance of that idea in the network, as would be naïvely
As can be seen in Fig 2, the relationship between threshold (θ) and virality (V) is far from
linear: the transition in virality is fairly rapid. The slope seems to be equally steep for the differ-
ent settings, the difference between which seem to be mainly expressed in an offsetting of the
transition to lower threshold values. The graph thus shows how relatively small changes in
threshold can result in increased probability for rapid global cascades. In other words, the vir-
ality of a narrative is not proportionally related to its quality or resonance, but small differences
can have big effects. This illustrates an important aspect of the logic of new social media.
It is no surprise that opinion polarized nodes result in higher virality, in both the case and
the control, since the definition of opinion polarization here is that some nodes have a lower
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 6 / 21
Fig 1. Virality as a function of network polarization. For parameters, see Table 1. This figure shows the effects of the echo chamber with P
= 0, i.e.
without opinion polarization (implying no difference in activation threshold between the cluster and the overall network.) The results are shown for
different average threshold levels. As can be seen, the cluster increases the virality until the network polarization passes 0.6, from which it starts having a
negative impact on virality. The lower graph shows the effects of varying the number of nodes in the network. The lower graph shows the variance for
network polarization with θ= 0.27 for varying node counts, to show that the results are robust for varying network sizes. As can be seen, the virality falls
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 7 / 21
threshold, which naturally increases the probability for transitions. More unexpected and
interesting are the differences between the case and the control. For example, the surface
between the dashes and filled line are larger with opinion polarized echo chambers than with
the control, in particular for higher levels of opinion polarization (P
). This implies that the
effects of having an echo chamber is larger when the echo chamber is also opinion polarized. This
in turn indicates that there is a synergetic effect between opinion polarization and network
To study this closer, we return to looking at virality as a function of the level of network
polarization (as in Fig 1), but this time, we include also opinion polarization, P
2.3 Relation between virality (V), and network (P
) and opinion
polarization (P
Fig 3 shows the relationship between network polarization, opinion polarization and virality.
Comparing the case and the control, there are multiple striking differences. First, the effect of
having an echo chamber (i.e. DVðyc;Pc
n;PoÞ) is significantly larger for the case than the control.
Second, in the case, Pc
n(i.e. the level of network polarization for which virality peaks) depends
on P
(opinion polarization), but in the control, it remains constant. Looking at how Pc
changes as a function of P
, we see that the peak level gradually shifts to the left with increasing
opinion polarization, i.e. toward lower levels of network polarization (note that the lines can-
not be compared with regard to the level of virality, since θ
varies with P
.) This suggests that
with higher levels of opinion polarization, virality peaks at lower levels of network polarization.
In other words, this may imply that if an echo chamber is highly coherent in terms of views on
a specific news item or idea, its effects on virality becomes larger if it is more well-connected
with the surrounding network.
We can look more closely at this relationship by plotting the network and opinion polariza-
tion at each point which virality peaks, giving optimal network polarization as a function of
the level of opinion polarization. This is shown in Fig 4.
Fig 4 indicates that echo chambers with higher levels of opinion polarization have the most
effect on virality when they have more external connections (i.e. lower P
). A plausible expla-
nation for this is that it follows from the levelling between having enough internal ties to
enable cohesion of the cluster, but having as many external connections as possible in order to
spread the cascade globally. Since opinion polarized echo chambers need fewer internal ties to
achieve internal cohesion, this implies that these ties have more impact when external.
This suggests that what is playing out in the dynamics of diffusion from echo chambers fol-
lows along the lines of Ghasemiesfeh et al. [60], who observe that one can distinguish two
phases in the diffusion process: first, the cascade spreads locally via strong, short-range ties,
and gathers the critical momentum it needs to transition into the second phase. In this second
phase, the diffusion starts to spread also through long-range ties, to the rest of the network.
Hence, in the first phase, strong local connections are beneficial, while the second phase is
more similar to simple contagions, spreading quickly through weak ties.
This suggests the hypothesis that echo chambers are beneficial for virality as they provide a
boost to the first phase, by growing an initial momentum: echo chambers provide a foundation
from which the global cascade can then follow. If this is indeed the way the role that echo
chambers play in diffusion processes, successful viral spread should take the form of an initial
with larger network sizes, however, the effects of having a cluster present seems to possibly increase with network size. The lower graph was averaged over
300 iterations, with degree 8, and 20% of nodes in cluster. (Runs performed with opinion polarization showed the same result).
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 8 / 21
Fig 2. Virality as a function of threshold. Shows virality (V) as a function of the average threshold value (θ), for different levels of
opinion polarization (P
). The upper, filled line for each P
shows virality at the critical level of network polarization (i.e the P
for which
the effects of the presence of a cluster, V(θ
, 0) = ΔV(θ
), peaks for some given P
, denoted Pc
n) and the lower,
dashed line for an Erdős-Re
´nyi network. The upper figure shows the case (the nodes in the echo chamber are opinion polarized), while
the lower shows the control (opinion polarization is assigned to random nodes in the network.) Two things should be noted: (i) The
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 9 / 21
wave of activation inside the echo chamber, followed by a second longer wave of activation in
the larger network. To explore whether this is the case, we can look at the order in which
nodes in the network are activated in successful spread. This is shown in Fig 5. As can be seen,
there is indeed two phases of activation: first, the activated nodes almost exclusively belong to
the echo chambers, second, it spreads to the larger network. This implies that a such a two-
phase process is indeed playing out, and that the echo chambers may contribute to it.
So, we have thus far noted that there is an interaction effect between network and opinion
polarization in echo chambers, and that both have a positive effect on virality. However, the
perhaps most pertinent question remains: are there any synergic effects between these two fac-
tors with relation to virality? In other words, is the effect on virality of an echo chamber greater
than sum of its parts? To explore this, we look at the effect of the echo chamber on virality as a
function of opinion polarization.
2.4 Virality (V) as a function of opinion polarization (P
Fig 6 shows the effect of an echo chamber with an optimal level of network polarization as a
function of opinion polarization. In other words, the graph shows the impact of the presence
of an echo chamber depends on how opinion polarized the echo chamber is. What we can see
from the graph is that the effect of having an echo chamber increases with higher levels of
opinion polarization, up until a level where the effect pans out, and then starts to decrease.
Another way to put this is that up until a certain level of opinion polarization, the more opin-
ion polarized the echo chamber is, the more effect does its presence have on virality. In the
control case, there is almost no additional positive effect on virality from increasing opinion
polarization. Indeed, when the opinion polarization is high, the presence of a cluster makes
virtually no difference.
This implies that we indeed do have a synergistic relationship between opinion and net-
work polarization for network virality. That this effect becomes negative at high levels of opin-
ion polarization may be explained by Fig 4; as opinion polarization increases, lower level of
network polarization is needed to maintain internal coherence in the echo chamber. So, with
high opinion polarization, the optimal network polarization becomes lower. This also implies
relationship between threshold and virality is far from linear, and (ii) as can be seen, the distance between the dashed and filled lines
becomes significantly wider in the case than in the control: this tentatively suggests the presence of a synergic interaction between network
and opinion polarization.
Table 1. Model parameters.
Parameter Value
N(network size) 100
k(average degree) 8
c(cluster fraction) 0.2
(step size for opinion polarization) 0.03
(step size for network polarization) 0.075
Δθ(step size for threshold) 0.0015
Iterations for each parameter set 1,000
These parameter values were used to generate all graphs, unless otherwise specified in the figure text. The step sizes
constitute the “resolution” of the parameter values. The iteration count describes how many times the model was run
for each parameter step, with new random network structures for each run to compensate for network heterogeneity.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 10 / 21
Fig 3. Virality and opinion polarization. In these graphs, we look V(θ
) (i.e. at the critical threshold level) for each level of P
(opinion polarization) (note: since θ
varies with P
, the lines in the graphs cannot be compared with regard to the absolute level of
virality.) The different lines denote different level of opinion polarization (P
). The upper graph shows the case (when the echo chamber is
polarized), and the lower the control (when opinion polarization is assigned to random nodes in the network.) As can be seen, in the case,
(the level of opinion polarization) affects Pc
n, i.e. at what level of network polarization that virality peaks. In the control case, there is no
such effect.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 11 / 21
that the difference between having a cluster and not becomes smaller, which in turn means
that not having a network polarized cluster becomes more efficient (this can most clearly be
seen in Fig 3.)
Hence, we have two ostensibly contradictory results, as the effect of the cluster increases
with higher opinion polarization, but the more opinion polarized the cluster is, the less net-
work polarized should it be to maximize virality. The latter effect counteracts the former, and
as the opinion polarization increases, this counteraction becomes stronger until the difference
between having cluster and not becomes nominal.
In summary, we can hence conclude that the model implies that the existence of an echo
chamber increases the virality in networks, and that the reason for this is that the echo cham-
ber constitutes a “protected space” in which an idea or narrative can find a firm footing for fur-
ther diffusion through the network. This suggests a theoretical connection to the “innovation
niches” of socio-technical transitions, which are argued to act as “safe havens” in which new
technologies can develop, free from market pressures, before spreading to the general market
(see e.g. [61]). It also relates to the observation that while long ties are highly beneficial for the
spread of simple contagions, complex contagions first have to spread locally before they can
take advantage of such long-range ties. This also accounts for the synergetic effect between
Fig 4. Critical network polarization as a function of opinion polarization. This graph shows how Pc
n(the network polarization for which the highest virality is
reached) depends on P
(the opinion polarization of the cluster). Comparing to Fig 3, this shows the network and opinion polarization at each point at which virality
peaks. As the graph shows, the optimal echo chamber has more external connections and fewerinternal connections as P
increases. In the control (dashed line), there
is no interaction effect between network and opinion polarization.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 12 / 21
opinion and network polarization, as, if the echo chambers nodes have a very low activation
threshold, they are likely to be activated even without being strongly separated from the rest of
the network, meaning that highly opinion polarized groups have increased opportunity to
focus on weak external ties.
2.5 Echo chambers in empirical networks
While these dynamics play out on the generated Erdős-Re
´nyi networks, real world social net-
works differ in many ways from the type of simple networks that are generally used in this type
of models. For instance, empirical social networks, in particular online variants, are well-
known to display highly uneven degree distributions, high local clustering coefficient, and hav-
ing multiple clusters or echo chambers. Thus, to validate that these observed dynamics are
indeed playing out also in real-world networks structures, simulations were run also on empir-
ical data. While using empirical data strengthens the claim that these emergent dynamics are
in fact playing out also in real world social media, it does however also make it more difficult
to separate the causal mechanism at play, as the networks structures will differ in a multitude
of ways. Because of this, empirical and generated networks have somewhat different purposes,
both being invaluable to investigate the phenomenon at hand.
To acquire empirical social networks for this analysis, the data presented in [62] were used.
This dataset is gathered from Twitter, containing all retweet between politicians in a number
of countries during 6 months (for details and definitions, see [62].) As the authors of the data-
set argue, retweets are used by politicians to show their allegiances, therefore providing a link
to an underlying network structure. These networks have central nodes—often party leaders—
and are clustered along party lines, as party members to varying degrees tend to retweet within
their party. The data are especially appropriate for this simulation as they are of relatively man-
ageable size, thus allowing a thorough exploration of parameter space, they have a ground
truth in the form of the political parties of the nodes, and they have a natural boundary follow-
ing from this ground truth, meaning that it is not necessary to select the nodes to include on
the basis of some arbitrary criteria.
Fig 5. Order of node activation. Illustrates the order in which the node activation occurs in successful cascades, by showing the fraction of echo chamber and non-
echo chamber nodes to be activated at each time step. The left graph has no difference in activation threshold between inside and outside the cluster, while the right has
= 0.2. As can be seen, the cluster is activated the first few steps, while the bulk of the non-cluster nodes are activated later, peaking around step 6 or 7. The runs are
averaged over 1000 iterations, P
= {0.7, 0.75, 0.85}, with θ= 0.27.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 13 / 21
Fig 6. Effect of echo chamber. The graph shows the effect of the presence of the echo chamber for the network polarization level with
highest effect, as a function of opinion polarization (i.e. DVðyc;Pc
n;PoÞas a function of P
). In other words, looking at Fig 3, for each
level of opinion polarization, it substracts the virality without an echo chamber from peak virality, Vðyc;Pc
n;PoÞ  Vðyc;0;PoÞ. The
dashed line represents the control case where the opinion polarization has been randomly assigned. As can be seen, the presence of an
opinion polarized echo chamber has more impact than the presence of one that is not opinion polarized. In the control case, there is
little positive interaction between network and opinion polarization. The lower graph shows the same curve for different network sizes,
to show that the results are robust for varying network sizes. As can be seen, the effects of having a cluster present actually increases
with the network size. Lower graph runs were averaged over 300 iterations, with average degree 8, and 20% of nodes were assigned to
be part of echo chamber.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 14 / 21
To identify the effects of network polarization, a large number of networks structures are
needed. Networks are generated from this data by seeing retweeting as a stochastic process
revealing an underlying social network. The number of tweets between nodes is thus used to
set the probability for a tie to exist between two nodes, which is then used as a set of probabili-
ties from which a given number of edges are drawn. This allows us to generate networks with
varied average degree by, in a sense, simulating varying the time interval during which the
data were collected. Since data is collected over a finite time interval, meaning that retweets
between all nodes will not yet have been collected, a low base probability is added to each net-
work edge (here, = 0.1) to allow us to generate a network with higher average degree. This
stochastic approach allows creating a large number of networks that describe the network
properties of the data, while reducing the influence of the average network degree on the
model outcome.
To additionally reduce the impact of networks differences, networks and parties were fil-
tered on their size. The countries with between 80 and 150 nodes, and their parties of between
5 and 40 nodes were selected, generating 200 random networks for each of the 37 combina-
tions. This resulted in a total of 7,400 generated networks. For each such network, opinion
polarization (P
) and threshold (Θ) were varied, running 5 iterations for each combination,
resulting in a total of 18.5 million model runs. An average degree of 8 was used in these runs.
The level of network polarization of the clusters were calculated on these networks, using the
same procedure as in previous simulations, and rounded off to closest 0.05.
Fig 7 show the result of these simulations. We see that the familiar dynamics are at play also
in these empirical networks: virality increases with network polarization, and as opinion polar-
ization increases, the effect of network polarization on virality becomes significantly stronger.
Looking at the critical activation threshold for each opinion polarization level in Fig 8, the
pattern observed in Fig 7 becomes even clearer: the cluster has a strong effect on virality, in
particular for higher values of P
While these results overall match with the results on the Erdős-Re
´nyi networks, some inter-
esting differences can be noted. One such difference to the model dynamics on generated net-
works is that the virality peak does not vary as clearly over P
as P
changes (compare to Fig 4).
This difference is likely the result of that while the generated networks had to, in a sense, prior-
itize between internal or external ties, these empirical networks are not restricted in this way,
as they can vary their total number of connections. In other words, different clusters can vary
in their internal and external connectivity, and it is quite possible for a party to have plenty of
both internal and external ties. In fact, these are rather likely to correlate, as parties with highly
active users will tend to tweet more both internally and externally.
Fig 7. Virality as a function of network polarization in empirical networks. These figures correspond to Fig 1, and show the effects of network
polarization on virality for different activation thresholds. As can be seen, virality (V) is affected by network and opinion polarization. As opinion
polarization (P
) increases, the effect of network polarization (P
) becomes stronger, and we see a familiar peak at around P
= 0.55 to 0.6. For higher
values of network polarization, virality starts to fall.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 15 / 21
The right figure in Fig 8 shows the difference in activation threshold at the critical level
between the case and the control run. As the figure illustrates, higher levels of opinion polari-
zation result in the echo chamber effect becoming stronger. However, there is little to no effect
when political polarization is 0—despite that the left figure shows a significant effect of net-
work polarization also without political polarization. This ostensibly unexpected result is in
fact not very surprising: since these networks will tend to consist of multiple clusters, randomly
activating a node—as in the control case—will still tend to activate a node within a cluster.
When political polarization increases, however, the echo chamber effect comes into effect, and
the difference between case and control grows significantly.
In summary, these empirical runs show that the observed dynamics are indeed at play also
in empirical networks, with multiple clusters and uneven degree distribution. Despite that the
level of network polarization co-varies with various related structural differences, there is a
clear correlation between virality and network polarization, influenced by political polariza-
tion. This strengthens the claim that the model dynamics are at play also in real-world social
3 Conclusion
The model presented in this paper indicates that there may be general structural effects of echo
chambers that contribute to the viral spread of misinformation. The model adds to the existing
literature on network diffusion (see e.g. [44,51,59]) by showing that a combination between
clustering and homophily has disproportionate effects on the capacity for complex diffusion,
as it contributes to initiating a band-wagon effect. This stands in sharp contrast to both intui-
tion and existing research on the spread of simple contagions, where clusters have instead
been found to reduce virality [43]. This link between the presence of a polarized cluster and
the spread of a complex contagion indicates a possible connection between echo chambers
and the spread of misinformation. In other words, the result of this model suggests that echo
chambers may be linked to the spread of misinformation through an emergent network effect.
When misinformation resonates with the views of an echo chamber, the chamber can function
as an initial platform from which the diffusion can occur globally through weak ties.
The model furthermore suggested that the combination between opinion and network
polarization, quintessential of echo chambers, results in a synergetic effect that increases
the virality of narratives that resonate with the echo chamber. This means that the simple
Fig 8. Critical virality on empirical networks. The left figure corresponds to Fig 4, showing the data from Fig 7 at critical threshold, i.e. the threshold
where the impact of the cluster is the highest. (As in Fig 4, comparison between the lines should be done with care, as they represent different values of
Θ.) The right figure shows the difference between the case and the control for these critical threshold levels. The figure shows that there is indeed a
strong interaction effect between political and network polarization, also when the model is run on empirical networks.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 16 / 21
clustering together of users with a deviant world-view is enough to affect the virality of infor-
mation items that resonate with their perspective.
More broadly, this suggests that not only algorithmic “filter bubbles” [63] affect what news
and perspectives we are exposed to online, but that the mere fact of social media permitting a
dynamic of social clustering can change the dynamics of online virality. The possibility of self-
segregation [64] can therefore affect not only what the segregated users see, but also what per-
spectives non-segregated users are exposed to. This can occur as subtle and complex network
dynamics of the interaction structure of social media can play into the diffusion dynamics, in
ways that are not necessarily even understood by the developers of the media platforms.
While simulations offer an unparalleled possibility to study specific causal mechanisms, it
should however be noted that the result of a computational model is not enough to draw defin-
itive conclusions about real world dynamics, since the observed mechanism may be overshad-
owed by other factors. The simplicity of the model described in this paper, while having the
benefit of providing generality and permitting a thorough exploration of parameter space, also
means that there are many factors that are not taken into account in the model. Additional
research is therefore necessary to investigate whether the described mechanism is in fact in
play in real world social media.
By engaging in this exploration, this paper constitutes a step toward the study of the “media
logic” [27] of social media, which, as has been noted by multiple scholars, differ in important
ways from the more institutionalized logic of traditional media [26,28]. Since social media
logic is the result of complex cascades and emergent effects, they require the use of computa-
tional models to study and discern their often-surprising dynamics [32]. Although clearly of
central relevance for the understanding of the contemporary media landscape, these questions
remain under-researched and poorly understood among both scholars and the lay public.
A fascinating self-similarity can be noted here, as the type of unintuitive complex dynamics
characteristic of social media has been observed to be precisely what misinformation and con-
spiracy theories tend to seize upon: scholars have noted that conspiracy theorists often con-
struct narratives that attempt to “restore a sense of agency, causality and responsibility” [65]
(p.1). This seems related to a type of modernist thinking of society—what Andersson and
To¨rnberg [38] refer to as the “complicated” system perspective—which would follow from
approaching social media through the lens of traditional media. In this way, conspiracism is
not only a symptom of the breakdown of knowledge authority, but of the breakdown of intui-
tive causality itself, as society becomes an increasingly complex system [66,67]. In line with
this, Gambetta and Hertog [68] show that engineers are over-represented among extremists,
and argue that this comes from seeking the order and hierarchy of machines also in the social
world. Conspiracy theories hence tend to describe a society under the conscious control of a
global elite, seeking order and agency in a world which is becoming more and more like a leaf
in the wind of social processes that are emergent and intangible, contingent and plural.
This is part of the larger question of how social media has affected our social and political
processes. While online platforms have been thought of as equalizers and social levelers, which
through disintermediation could provide equal opportunities for all, they in practice seem
rather to result in the perpetuation and reinforcement of processes of stratification and
inequality, as well as in self-segregation and tribalization [69]. The disappearance of media
intermediation seems not to have, as was believed, fostered a space for direct meetings in a sort
of online Habermasian public sphere, but rather to have implied that the “world between them
has lost its power to gather them together, to relate and to separate them” [6] (p.52). As we
find ourselves in this new condition of “worldlessness”, we go in search of new communities,
new worlds in which we may again feel a sense of togetherness. These communities become
places “where we see all people suddenly behave as though they were members of one family,
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 17 / 21
each multiplying and prolonging the perspective of his neighbor. [. . .] The end of the common
world has come when it is seen only under one aspect and is permitted to present itself in only
one perspective.” [6] (p.57-58).
The question of how to resolve the issues of misinformation and polarization, now consid-
ered among the main threats to human society, hence concerns how to design social media
technology as to again constitute the foundation for a common world; how to shape our digital
spaces as to help weave, rather than fray, our social fabric. For this, studies of the emergent
externalities of the interaction structures of social networks will likely need to play a central
This project has received funding from the European Union’s Horizon 2020 research and
innovation programme project ODYCCEUS (grant agreement n˚ 732942).
Author Contributions
Conceptualization: Petter To¨rnberg.
Formal analysis: Petter To¨rnberg.
Methodology: Petter To¨rnberg.
Project administration: Petter To¨rnberg.
Resources: Petter To¨rnberg.
Software: Petter To¨rnberg.
Visualization: Petter To¨rnberg.
Writing – original draft: Petter To¨rnberg.
Writing – review & editing: Petter To¨rnberg.
1. Brown J, Broderick AJ, Lee N. Word of mouth communication within online communities: Conceptualiz-
ing the online social network. Journal of interactive marketing. 2007; 21(3):2–20.
2. Kahn R, Kellner D. New media and internet activism: from the ‘Battle of Seattle’to blogging. New media
& society. 2004; 6(1):87–95.
3. Quattrociocchi W, Conte R, Lodi E. Opinions manipulation: Media, power and gossip. Advances in
Complex Systems. 2011; 14(04):567–586.
4. Quattrociocchi W, Caldarelli G, Scala A. Opinion dynamics on interacting networks: media competition
and social influence. Scientific reports. 2014; 4:4938. PMID:
5. Kumar R, Mahdian M, McGlohon M. Dynamics of conversations. In: Proceedingsof the 16th ACM
SIGKDD international conference on Knowledge discovery and data mining. ACM; 2010. p. 553–562.
6. Arendt H. The human condition. University of Chicago Press; 1958.
7. Habermas J. The structural transformation of the public sphere: An inquiry into a category of bourgeois
society. MIT press; 1991.
8. Bessi A, Zollo F, Del Vicario M, Scala A, Caldarelli G, Quattrociocchi W. Trend of Narratives in the Age
of Misinformation. PloS one. 2015; 10(8):e0134641.
PMID: 26275043
9. Mocanu D, Rossi L, Zhang Q, Karsai M, Quattrociocchi W. Collective attention in the age of (mis) infor-
mation. Computers in Human Behavior. 2015; 51:1198–1204.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 18 / 21
10. Sunstein CR. Echo chambers: Bush v. Gore, impeachment, and beyond. Princeton University Press
Princeton, NJ; 2001.
11. Garrett RK. Echo chambers online?: Politically motivated selective exposure among Internet news
users. Journal of Computer-Mediated Communication. 2009; 14(2):265–285.
12. Vosoughi S, Roy D, Aral S. The spread of true and false news online. Science. 2018; 359(6380):1146–
1151. PMID: 29590045
13. Centola D. The spread of behavior in an online social network experiment. science. 2010; 329
(5996):1194–1197. PMID: 20813952
14. Christakis NA, Fowler JH. The spread of obesity in a large social network over 32 years. n engl j med.
2007; 2007(357):370–379.
15. Christakis NA, Fowler JH. The collective dynamics of smoking in a large social network. New England
journal of medicine. 2008; 358(21):2249–2258. PMID:
16. Fowler JH, Christakis NA. Dynamic spread of happiness in a large social network: longitudinal analysis
over 20 years in the Framingham Heart Study. Bmj. 2008; 337:a2338.
a2338 PMID: 19056788
17. Salganik MJ, Dodds PS, Watts DJ. Experimental study of inequality and unpredictability in an artificial
cultural market. science. 2006; 311(5762):854–856. PMID:
18. Bond RM, Fariss CJ, Jones JJ, Kramer AD, Marlow C, Settle JE, et al. A 61-million-person experiment
in social influence and political mobilization. Nature. 2012; 489(7415):295–298.
nature11421 PMID: 22972300
19. Howell L. Digital wildfires in a hyperconnected world. WEF Report. 2013; 3:15–94.
20. Quattrociocchi W, Scala A, Sunstein CR. Echo chambers on facebook. 2016;.
21. Sunstein CR, Vermeule A. Conspiracy theories: Causes and cures. Journal of Political Philosophy.
2009; 17(2):202–227.
22. Furedi F. Culture of fear revisited. A&C Black; 2006.
23. Gu B, Konana P, Raghunathan R, Chen HM. Research Note—The Allure of Homophily in Social Media:
Evidence from Investor Responses on Virtual Communities. Information Systems Research.2014; 25
24. Bessi A, Coletto M, Davidescu GA, Scala A, Caldarelli G, Quattrociocchi W. Science vs conspiracy: Col-
lective narratives in the age of misinformation. PloS one. 2015; 10(2):e0118093.
1371/journal.pone.0118093 PMID: 25706981
25. Zollo F, Novak PK, Del Vicario M, Bessi A, MozetičI, Scala A, et al. Emotional dynamics in the age of
misinformation. PloS one. 2015; 10(9):e0138740. PMID:
26. Dahlgren P. Media logic in cyberspace: Repositioning journalism and its publics. Javnost-the public.
1996; 3(3):59–72.
27. Altheide DL. Media logic. Wiley Online Library; 1979.
28. Van Dijck J, Poell T. Understanding social media logic. 2013;.
29. Jameson F. Postmodernism, or, the cultural logic of late capitalism. Duke University Press; 1991.
30. Carr N. The shallows: What the Internet is doing to our brains. WW Norton & Company; 2011.
31. E
´rdi P. Complexity explained. Springer; 2007.
32. Mitchell M. Complexity: A guided tour. Oxford University Press; 2009.
33. Bak P, Chen K, Tang C. A forest-fire model and some thoughts on turbulence. Physics letters A. 1990;
34. Karafyllidis I, Thanailakis A. A model for predicting forest fire spreading using cellular automata. Eco-
logical Modelling. 1997; 99(1):87–97.
35. Schelling TC. Micromotives and macrobehavior. WW Norton & Company; 2006.
36. Schelling TC. Dynamic models of segregation. Journal of mathematical sociology. 1971; 1(2):143–186.
37. Macy MW, Willer R. From factors to actors: Computational sociology and agent-based modeling.
Annual review of sociology. 2002; p. 143–166.
38. Andersson C, To
¨rnberg P. Wickedness and the anatomy of complexity. Futures. 2017;.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 19 / 21
39. Granovetter M. The strength of weak ties. American journal of sociology. 1973; 78(6):l.
40. Chwe MSY. Structure and strategy in collective action. American Journal of Sociology. 1999; 105
41. Gould RV. Collective action and network structure. American Sociological Review. 1993; p. 182–196.
42. Macy MW. Learning theory and the logic of critical mass. American Sociological Review. 1990; p. 809–
43. Watts DJ, Strogatz SH. Collective dynamics of ‘small-world’networks. Nature. 1998; 393(6684):440–
442. PMID: 9623998
44. Centola D, Macy M. Complex contagions and the weakness of long ties. American Journal of Sociology.
2007; 113(3):702–734.
45. Watts DJ. A simple model of global cascades on random networks. Proceedings of the National Acad-
emy of Sciences. 2002; 99(9):5766–5771.
46. Easley D, Kleinberg J. Networks, crowds, and markets: Reasoning about a highly connected world.
Cambridge University Press; 2010.
47. Cialdini RB, Goldstein NJ. Social influence: Compliance and conformity. Annu Rev Psychol. 2004;
55:591–621. PMID: 14744228
48. Achen CH, Bartels LM. Democracy for realists: Why elections do not produce responsive government.
Princeton University Press; 2016.
49. Mason L. Uncivil agreement: How politics became our identity. University of Chicago Press; 2018.
50. Boutyline A, Willer R. The social structure of political echo chambers: Variation in ideological homophily
in online networks. Political Psychology. 2017; 38(3):551–569.
51. Chiang YS. Birds of moderately different feathers: Bandwagon dynamics and the threshold heterogene-
ity of network neighbors. Journal of Mathematical Sociology. 2007; 31(1):47–69.
52. Krassa MA. Social groups, selective perception, and behavioral contagion in public opinion. Social Net-
works. 1988; 10(2):109–136.
53. Valente TW. Social network thresholds in the diffusion of innovations. Social networks. 1996; 18(1):69–
54. Del Vicario M, Bessi A, Zollo F, Petroni F, Scala A, Caldarelli G, et al. The spreading of misinformation
online. Proceedings of the National Academy of Sciences. 2016; 113(3):554–559.
55. Oliver PE, Marwell G. The Paradox of Group Size in Collective Action: A Theory of the Critical Mass. II.
American Sociological Review. 1988; p. 1–8.
56. Nadeau R, Cloutier E, Guay JH. New evidence about the existence of a bandwagon effect in the opinion
formation process. International Political Science Review. 1993; 14(2):203–213.
57. Marwell G, Oliver PE, Prahl R. Social Networks and Collective Action: A Theory of the Critical Mass, Ill.
American Journal of Sociology. 1988; 94(3):502–534.
58. Oliver P, Marwell G, Teixeira R. A theory of the critical mass. I. Interdependence, group heterogeneity,
and the production of collective action. American journal of Sociology. 1985; p. 522–556.
59. Centola DM. Homophily, networks, and critical mass: Solving the start-up problem in large group collec-
tive action. Rationality and society. 2013; 25(1):3–40.
60. Ghasemiesfeh G, Ebrahimi R, Gao J. Complex contagion and the weakness of long ties in social net-
works: revisited. In: Proceedings of the fourteenth ACM conference on Electronic Commerce. ACM;
2013. p. 507–524.
61. Geels FW. Ontologies, socio-technical transitions (to sustainability), and the multi-level perspective.
Research Policy. 2010; 39(4):495–510.
62. Teernstra L, Uitermark P Justus To
¨rnberg. Politicians’ retweet networks: Political coalitions and divi-
sions on Twitter in 23 countries. In: NetSci; 2018.
63. Pariser E. The filter bubble: What the Internet is hiding from you. Penguin UK; 2011.
64. Lawrence E, Sides J, Farrell H. Self-segregation or deliberation? Blog readership, participation, and
polarization in American politics. Perspectives on Politics. 2010; 8(1):141–157.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 20 / 21
65. Knight P. ILOVEYOU: Viruses, paranoia, and the environment of risk. The Sociological Review. 2000;
66. Castells M. The rise of the network society: The information age: Economy, society, and culture. vol. 1.
John Wiley & Sons; 2011.
67. Urry J. Global complexity. Polity Cambridge; 2003.
68. Gambetta D, Hertog S. Engineers of jihad: The curious connection between violent extremism and edu-
cation. Princeton University Press; 2016.
69. Boy JD, Uitermark J. Reassembling the city through Instagram. Transactions of the Institute of British
Geographers. 2017;.
Echo chambers and viral misinformation
PLOS ONE | September 20, 2018 21 / 21
... Notably, even accurate information could be misleading when quoted out of context and lead to false inferencing (Treen et al., 2020). Adopting the qualifiers of information as above or related concepts such as rumors, propaganda, or 'fake news,' a host of works have examined the role of SM therein and the tenacious association of such fallacies with polarization (e.g., Del Vicario et al., 2016;Törnberg, 2018). Kaligotla et al. (2020) argue that SM accentuates falsehoods on multiple counts: social linkages aid believability, verifiability of information is neither immediately possible nor of value in the ecosystem, and crowdsourced content is presumed by many to be more accurate than it often is. ...
... Malicious accounts employing 'sock puppets' (i.e., fake identities) and bots aid in the process (Mohammad et al., 2015;Yan et al., 2021). Systemic falsity is reciprocally constitutive of polarization as propaganda solidifies positions and emotions, which in turn makes one credulous (Humprecht et al., 2020a;Törnberg, 2018). It can thus be argued that while human falsity has always existed in perpetuity, SM has not just extended its reach but fundamentally redefined it. ...
The world of today presents a duality of phenomenal progress and persistent ills. Amid such existential contradictions, public deliberation forms one of the central pillars of a functional and progressive society. Though its relevance remains undoubted, interactions in the public sphere may often give way to misinformation, affect-driven predisposition, and homophily-based interactions, all reminiscent of polarization. While polarization remains a concern worldwide, structural changes, most notably, social media's advent and remarkable progress, have further redefined the meaning, scale, and diffusion of information. Accordingly, a tireless debate rages regarding the valence and strength of social media's influence on polarization. As an incremental means of resolving the complexity, we perform a systematic review of the extant scholarship and identify contingencies and mechanisms of social media's relationship with polarization. Further, we provide a conceptual framework, incorporating these intricacies while emphasizing the need to place this association in a broader frame. Our work contributes to theory by being one of the few reviews linking social media to polarization and providing a synthesis of contingent factors and underlying processes. We guide policy and practice by suggesting a future research framework.
... Future research on this topic could consider: i) the temporal analysis of segregated communities and their relation with gaining more or less citations over time, ii) the analysis of diversity of the scientific publications inside the communities using opinion distance [31] and their demographic diversity to understand if the segregated and isolated communities are not diverse and echoing research to the point of becoming polarised, iii) the definition of lead researchers (using the hub/spoke core or author position in the publications) and the understanding of their relationship to segregated communities [12], iv) the measurement of the impact of the segregated communities on the topology of the network formation and the spreading processes of scientific theories [36]. ...
Full-text available
Collaboration networks, where nodes represent authors and edges coauthorships among them, are key to understanding the consumption, production, and diffusion of knowledge. Due to social mechanisms, biases, and constraints at play, these networks are organized in tight communities with different levels of segregation. Here, we aim to quantify the extent and impact of segregation in collaboration networks. We study the field of Computer Science via the Semantic Scholar Open Research Corpus. We measure the segregation of communities using the Spectral Segregation Index (SSI) and find three categories: non-segregated, moderately segregated, and highly segregated communities. We focus our attention on non-segregated and highly segregated communities, quantifying and comparing their structural topology and core location. When we consider communities of both categories in the same size range, our results show no differences in density and clustering, but evident variability in their core position. As community size increases, communities are more likely to occupy a core closer to the network nucleus. However, controlling for size, highly segregated communities tend to be located closer to the network periphery than non-segregated communities. Finally, we analyse differences in citations gained by researchers depending on their community segregation level. Interestingly, researchers in highly segregated communities gain more citations per publication when located in the periphery. They have a higher chance of receiving citations from members of their same community in all cores. Researchers in non-segregated communities accrue more citations per publication in intermediary and central cores. To our knowledge, our work is the first to characterise segregated communities in scientific collaboration networks and to investigate their relationship with the impact measured in terms of citations.
... However, the exercise of simulation is more oriented towards imitation, since it triggers a supplanting of the real by the signs of the real (Baudrillard, 1978). Therefore, unlike representation, it could be said that simulation is closer to imitation than to interpretation; meaning that in this context, political discussions on Twitter merely imitate a real-life debate, though it is very close to the real thing from a discursive and participatory perspective (Törnberg, 2018). ...
Full-text available
This paper argues in favor of the manifestation of a liquid democracy on Twitter during the Colombian presidential election campaigns (2018-2022) through a qualitative analysis of over 2484 responses to the candidates' messages. Users appropriated the platform through the use of memes, fake news and hashtags (#) to attack or defend the politicians, ignoring any of their campaign proposals. A flowchart was created to illustrate the emergence of subjectivities that emerge during political discussions on the platform, such as saboteur/troll, or cyberactivists. Likewise, the profiles of the candidates were constructed under a marketing strategy appropriate to this new cyber-citizenship. The findings support a lack of political debate or discussion and a simulated political participation that does not guarantee the exercise of democracy
... 16,17 Innerhalb dieser Echokammern bleiben Falschmeldungen meistens unwidersprochen und verbreiten sich sogar schneller als wahre Informationen. 18,19 Wie kann man mit Falschmeldungen umgehen? ...
Full-text available
Immer wieder tauchen Falschmeldungen über Geflüchtete in den sozialen Medien auf, die sich teilweise hartnäckig halten. Vor allem im Zuge der sogenannten EU-Flüchtlingskrise im Sommer 2015 war ein starker Anstieg von Falschmeldungen über Geflüchtete (oder als Geflüchtete wahrgenommene) Personen zu beobachten. Woher kommen diese Falschmeldungen und wieso sind sie so verbreitet? Welche Konsequenzen haben sie und wie kann man ihnen entgegentreten? Dieser Beitrag stellt die Hintergründe von Falschmeldungen über Geflüchtete dar und bietet Hilfestellung im Umgang mit solchen Informationen.
Full-text available
Eine demokratische Gesellschaft ist auf Bürger*innen angewiesen, die auf verlässliche und fak-tentreue Informationen zurückgreifen können – nicht nur, aber gerade in Krisenzeiten wie wäh-rend der aktuellen Corona-Pandemie. Desinformation, also absichtlich gestreute Falschnach-richten, stellt für die Demokratie deshalb ein Problem dar. Bislang ging man davon aus, dass Desinformation in der Schweiz nicht sehr stark ausgeprägt ist. Die Covid-19-Pandemie hat aber das Thema stärker auf die öffentliche Agenda gebracht. Der vorliegende Bericht untersucht, wie Schweizer*innen mit Desinformation umgehen und welche Rolle Fehlinformationen, Al-ternativmedien und Verschwörungstheorien in der Öffentlichkeit und in der Wahrnehmung der Bürger*innen spielen. Der Bericht zeigt Ergebnisse einer repräsentativen Bevölkerungsbefra-gung zur Problematik der Desinformation in der Schweiz, bietet ein Inventar von Websites sowie Social-Media-Angeboten von Alternativmedien und zeigt den Stellenwert von Alterna-tivmedien anhand einer computer-unterstützten Analyse der Aktivitäten sämtlicher aktiven Nutzer*innen der Schweizer Twitter-Sphäre.
Full-text available
Purpose In the age of social media, when publishers are vying for consumer attention, click-baits have become very common. Not only viral websites but also mainstream publishers, such as news channels, use click-baits for generating traffic. Therefore, click-bait detection and prediction of click-bait virality have become important challenges for social media platforms to keep the platform click-bait free and give a better user experience. The purpose of this study is to try exploring how the contents of the social media posts and the article can be used to explain and predict social media posts and the virality of a click-bait. Design/methodology/approach This study has used 17,745 tweets from Twitter with 4,370 click-baits from top 27 publishers and applied econometric along with machine learning methods to explain and predict click-baitiness and click-bait virality. Findings This study finds that language formality, readability, sentiment scores and proper noun usage of social media posts and various parts of the target article plays differential and important roles in click-baitiness and click-bait virality. Research limitations/implications The paper contributes toward the literature of dark behavior in social media at large and click-bait prediction and explanation in particular. It focuses on the differential roles of the social media post, the article shared and the source in explaining click-baitiness and click-bait virality via psycho-linguistic framework. The paper also provides explanability to the econometric and machine learning predictive models, thus performing methodological contribution too. Practical implications The paper helps social media managers create a mechanism to detect click-baits and also predict which ones of them can become viral so that corrective measures can be taken. Originality/value To the best of the authors’ knowledge, this is one of the first papers which focus on both explaining and predicting click-baitiness and click-bait virality.
Fake news reference to misleading or false articles that is been spread on social network. The diffusion of misleading information on social network causes an impact on civil life of the people in terms of psychological, financial etc., which makes them difficult to lead their life in the society. Therefore, developing a system to predict the diffusion of fake news is essential. This paper proposes a fake news diffusion prediction model involving several psycho-sociological facets of social media users. The underlying presumptions are two-fold: (a) it is essential to understand the probable impact of a posted fake news message, i.e., how many persons will be influenced by the information spread; and (b) although information propagation is a well-explored topic, most of the work has been restricted to network topological analysis. To this end, we present psycho-sociological analysis as a better alternative to understand fake propagation in a social network, with the aim to answer a few fundamental questions: (i) Who initiates fake posts on social media? (ii) Who consumes (replies to, shares, or likes) such comments? (iii) Can we model fake news diffusion better if we know the psycho-sociological traits of individuals towards fakeful content? To present psycho-sociological analysis, we bring in three different dimensions of human behaviour: personality (Big-5: Openness, Agreeableness, Conscientiousness, Extraversion, and nNeuroticism), values & ethics (Schwartz model: Self-Direction, Stimulation, Hedonism, Achievement, Power, Security, Conformity, Tradition, Benevolence and Universalism), and the dark side of the personality (dark triad: Narcissism, Machiavellianism, and Psychopathy). The Big5 model describes five essential person-level traits, the values & ethics model depicts societal behaviour, and the dark triad model encapsulates the dark side of human personality. We develop a suite of classifiers to detect behavioral traits. We also propose a model to predict the diffusion of fake news and classifier to classify user as fake or real. Various empirical studies are reported in this paper in order to comprehend how human behavioral traits correlate with online fake spreader and fake posting user behavior. From our empirical analysis some of the key observations found are: (i) fake posts on social media are commonly started by people with neurotic personality. (ii) People who spread political fake posts are traditional and power oriented. (iii) Fake user is extrovert and neurotic in nature, and those who post fake and real news on social networks tend to have balanced dark side behaviour. The proposed models also outperform state-of-the-art with a significant margin.
Full-text available
The inner dynamics of the multiple actors of the informations systems - i.e, T.V., newspapers, blogs, social network platforms, - play a fundamental role on the evolution of the public opinion. Coherently with the recent history of the information system (from few main stream media to the massive diffusion of socio-technical system), in this work we investigate how main stream media signed interaction might shape the opinion space. In particular we focus on how different size (in the number of media) and interaction patterns of the information system may affect collective debates and thus the opinions' distribution. We introduce a sophisticated computational model of opinion dynamics which accounts for the coexistence of media and gossip as separated mechanisms and for their feedback loops. The model accounts also for the effect of the media communication patterns by considering both the simple case where each medium mimics the behavior of the most successful one (to maximize the audience) and the case where there is polarization and thus competition among media memes. We show that plurality and competition within information sources lead to stable configurations where several and distant cultures coexist.
Full-text available
Traditional scientific policy approaches and tools are increasingly seen as inadequate, or even counter-productive, for many purposes. In response to these shortcomings, a new wave of approaches has emerged based on the idea that societal systems are irreducibly complex. The new categories that are thereby introduced - like "complex" or "wicked" - suffer, however, by a lack of shared understanding. We here aim to reduce this confusion by developing a meta-ontological map of types of systems that have the potential to "overwhelm us": characteristic types of problems, attributions of function, manners of design and governance, and generating and maintaining processes and phenomena. This permits us, in a new way, to outline an inner anatomy of the motley collection of system types that we tend to call "complex". Wicked problems here emerge as the product of an ontologically distinct and describable type of system that blends dynamical and organizational complexity. The framework is intended to provide systematic meta-theoretical support for approaching complexity and wickedness in policy and design. We also points to a potential causal connection between innovation and wickedness as a basis for further theoretical improvement.
Full-text available
How do people represent the city on social media? And how do these representations feed back into people's uses of the city? To answer these questions, we develop a relational approach that relies on a combination of qualitative methods and network analysis. Based on in-depth interviews and a dataset of over 400 000 geotagged Instagram posts from Amsterdam, we analyse how the city is reassembled on and through the platform. By selectively drawing on the city, users of the platform elevate exclusive and avant-garde establishments and events, which come to stand out as hot spots, while rendering mundane and low-status places invisible. We find that Instagram provides a space for the segmentation of users into subcultural groups that mobilise the city in varied ways. Social media practices, our findings suggest, feed on as well as perpetuate socio-spatial inequalities.
Full-text available
The focus is on the nature and significance of media logic and an ecology of communication for a theory of mediation and for understanding and investigating political communication. Media logic is defined as a form of communication, and the process through which media transmit and communicate information. The logic and guidelines become taken for granted, often institutionalized, and inform social interaction. A basic principle is that media, information technologies, and communication formats can affect events and social activities. An overview of media logic, including correcting some unfortunate misinterpretations, is followed by discussion of attempts at clarification and revision (e.g., mediatization, mediality, mediatric, etc.), along with uses, relevance, and applications for political communication with various media (e.g., television, Internet, social media).
We investigated the differential diffusion of all of the verified true and false news stories distributed on Twitter from 2006 to 2017. The data comprise ~126,000 stories tweeted by ~3 million people more than 4.5 million times. We classified news as true or false using information from six independent fact-checking organizations that exhibited 95 to 98% agreement on the classifications. Falsehood diffused significantly farther, faster, deeper, and more broadly than the truth in all categories of information, and the effects were more pronounced for false political news than for false news about terrorism, natural disasters, science, urban legends, or financial information. We found that false news was more novel than true news, which suggests that people were more likely to share novel information. Whereas false stories inspired fear, disgust, and surprise in replies, true stories inspired anticipation, sadness, joy, and trust. Contrary to conventional wisdom, robots accelerated the spread of true and false news at the same rate, implying that false news spreads more than the truth because humans, not robots, are more likely to spread it.