Rise of QAnon: A Mental Model of Good and Evil Stews in an Echochamber

Abstract

The QAnon conspiracy posits that Satan-worshiping Democrats operate a covert child sex-trafficking operation, which Donald Trump is destined to expose and annihilate. Emblematic of the ease with which political misconceptions can spread through social media, QAnon originated in late 2017 and rapidly grew to shape the political beliefs of millions. To illuminate the process by which a conspiracy theory spreads, we report two computational studies examining the social network structure and semantic content of tweets produced by users central to the early QAnon network on Twitter. Using data mined in the summer of 2018, we examined over 800,000 tweets about QAnon made by about 100,000 users. The majority of users disseminated rather than produced information, serving to create an online echochamber. Users appeared to hold a simplistic mental model in which political events are viewed as a struggle between antithetical forces-both observed and unobserved-of Good and Evil.

Full text

Rise of QAnon:
A Mental Model of Good and Evil Stews in an Echochamber
J. Hunter Priniski
Mason McClay
Keith J. Holyoak
(priniski@ucla.edu)
(masongmcclay@ucla.edu)
(holyoak@psych.ucla.edu)
Department of Psychology
University of California, Los Angeles
Abstract
The QAnon conspiracy posits that Satan-worshiping
Democrats operate a covert child sex-trafficking operation,
which Donald Trump is destined to expose and annihilate.
Emblematic of the ease with which political
misconceptions can spread through social media, QAnon
originated in late 2017 and rapidly grew to shape the
political beliefs of millions. To illuminate the process by
which a conspiracy theory spreads, we report two
computational studies examining the social network
structure and semantic content of tweets produced by users
central to the early QAnon network on Twitter. Using data
mined in the summer of 2018, we examined over 800,000
tweets about QAnon made by about 100,000 users. The
majority of users disseminated rather than produced
information, serving to create an online echochamber.
Users appeared to hold a simplistic mental model in which
political events are viewed as a struggle between
antithetical forces—both observed and unobserved—of
Good and Evil.
Keywords: QAnon, conspiratorial thinking,
hypercoherence, naturalistic data
Introduction
Deception and misinformation have pervaded American
politics throughout its history. During the Civil War,
Southern sentiment was manipulated by press campaigns
propagating unfounded claims that the North was
planning to lead a revolt of freed slaves against the South
(Necklason, 2020). A century and a half later, false
allegations of stockpiled weapons of mass destruction in
Iraq served to justify a U.S. invasion. This pretense
remained a pervasive belief among the American public
for at least a decade (Altheide & Grimes, 2005). More
recently, online forums promoted a conspiracy theory
asserting that Jewish financial elites helped orchestrate
the Covid-19 pandemic (Zipperstein, 2020). The
emergence of social media has enabled unregulated
production and proliferation of misinformation––a
global "infodemic" of misinformation (Salvi et al.,
2020)—making our era a Golden Age of conspiracy
theories.
One of the most widespread fantasies to emerge from
recent internet forums is the QAnon conspiracy
(Zipperstein, 2020; Papasavva et al., 2020; Saltman,
2020). QAnon’s central narrative involves a covert world
of liberal elites ruling by way of a global child-
trafficking ring and by mass manipulation of the
economy and media. As the main protagonist of QAnon,
former U.S. President Donald Trump is purported to be
the sole defense against this elite cabal. While this
narrative is disturbing and radical, the most nefarious
aspects of QAnon are likely its varied peripheral
narratives and claims that sow distrust in scientific and
democratic institutions (Zipperstein, 2020). For instance,
during the beginning of the Covid-19 pandemic, a meme
posted on a QAnon 4Chan board falsely claimed that
96% of reported deaths attributed to Covid-19 were not
actually due to that disease. While this claim was easily
refutable, within days it inspired massive anti-lockdown
protests in Germany, the United Kingdom, and the
United States, which were primarily attended by
protestors holding signs that bore the QAnon slogan,
WWG1WGA (“where we go one, we go all”).
Belief in QAnon's web of narratives is not reserved for
those on the political extremes. A CIVIQS (2020) survey
reported that one in three Republicans believe that
QAnon is mostly true. Given that a substantial
proportion of the electorate is comfortable with a
radically inaccurate depiction of reality, the
pervasiveness of QAnon-supporting sentiment may pose
a threat to the stability of Western democratic institutions
(Saltman, 2020). Understanding the social and cognitive
factors enabling the rise of QAnon is crucial to prevent
future emergence of similar conspiracies.
Hypercoherence as a Prerequisite for
Conspiratorial Thinking
Generalization of previously learned information to
novel situations is a hallmark of adaptive learning
(Mednick & Freedman, 1960). In the context of belief
formation, new beliefs also tend to generalize from, or
cohere with, features of prior beliefs (Lewandowsky,
Gignac, & Oberauer, 2013; Homer-Dixon et al., 2013).
Indeed, a coherence mechanism has been shown to be
central to various cognitive processes, from visual
perception (Yuille & Grzywacz, 1988) to moral
reasoning (Holyoak & Powell, 2016). According to
explanatory coherence theory, beliefs are often formed
on the basis of congruence with prior beliefs, insofar as
Priniski, J. H., McClay, M., & Holyoak, K. J. (2021). Rise of QAnon: A mental model of good and evil stews in an
echochamber. In T. Fitch, C. Lamm, H. Leder, & K. Teßmar-Raible (Eds.), Proceedings of the 43rd Annual
Meeting of the Cognitive Science Society. Cognitive Science Society.

the acceptance of a new belief increases the explanatory
coherence of the belief network (Findlay & Thagard,
2011). Beliefs may therefore be adopted if they fit the
explanatory model generated from one’s prior beliefs.
But what happens when new evidence can be
construed as coherent with one’s prior beliefs, regardless
of its veridicality? Recent work has shown that people
who engage in conspiratorial thinking tend to attribute
more control and structure to the world than is plausible.
Conspiratorial thinkers do not typically believe in just a
single conspiracy theory, but rather clusters of them (van
Harreveld et al., 2014). For example, Lewandowsky,
Oberaurer, and Gignac (2013) found that people's
propensity to believe that NASA faked the moon-landing
predicted their tendency to believe that climate change
was a hoax. This association arises because conspiracy
thinkers are likely to endorse completely novel
conspiracies that share common conspiracy themes, thus
viewing logically disjointed narratives as mutually
coherent (e.g., NASA/government conspiracy → fake
moon landing; climate scientists conspiracy → climate
science is fake). This evidence suggests that conspiracy
thinkers may readily bind new information with their
conspiratorial view of the world through a maladaptive
level of coherence––hypercoherence. Hypercoherence
combines top-down priors based on broad core attitudes
(e.g., distrust of government and scientific elites),
coupled with bottom-up “data” based on the opinions of
fellow believers that echo on social media. Where
conspiracy thinkers go one, they go all by attributing a
vast network of complex narratives to a single causal
source (Saltman, 2020).
Hypercoherence may be especially easy to achieve
when information is consumed within online
communities where both information and social identity
are radically curated. Conspiracies such as QAnon may
be a natural consequence of a social media environment
that: (1) prioritizes false information over verifiable
information, and (2) allows for the easy and rapid
formation of echochambers, or pockets of online
communities that share and consume nearly identical,
belief-confirming information (Sasahara et al., 2020).
Once misinformation is introduced that coheres with the
narrative of a particular echochamber, it may foster the
generation of additional content by simultaneously
adding to the coherence of the community’s narrative
while reducing its standard of plausibility.
Misinformation may therefore gradually reconfigure a
person's belief network toward stronger degrees of
coherence, making it more capable of binding disparate
and implausible beliefs. The result is belief in
conspiracies that cover a wide range of narrative clusters.
These factors make QAnon no longer merely a single
conspiracy, but a web of conspiratorial plots under the
umbrage of a central narrative and shared identity
(Roose, 2021). Understanding the features of social
media networks in which QAnon has spread may help
elucidate the conditions under which conspiratorial
trends metastasize into super-conspiracies. Such
analyses may help to find ways to stymie the growth of
future conspiracies that take root in internet discussions.
To this end, here we report analyses of the structure and
content of the early QAnon Twitter network, aiming to
infer the mental model that binds members of this
community together while understanding the social
processes underlying the narrative's rapid spread online.
Computational Studies of Twitter Data
We conducted two studies of the early QAnon Twitter
network to understand the social and cognitive processes
shaping the dissemination and content of the
conspiracy's narrative. In Study 1 we analyzed a retweet
network to assess the extent to which users shared versus
produced content. In Study 2 we fit a series of topic
models to the tweets to characterize the general form of
the mental model shared by QAnon users.
From June 29 to July 12, 2018, 834787 tweets from
107,777 unique users were collected using the Twitter
Streaming API. Tweets containing at least one of the
following strings were collected: "qanon", "#q", and
"#qanon". In January 2021, after Twitter removed Q
accounts following the insurrection on the U.S. Capitol,
tweets from many of these accounts were permanently
deleted. This early dataset thus provides a rare and
crucial glimpse into the echochamber that grew to
eventually impact American politics for much of 2020.
Study 1: Network Analysis
Social networks––both online and in the physical
world—are central to the function and maintenance of
conspiratorial beliefs. We analyzed a retweet network to
examine how different users contributed to the
production and distribution of information.
Retweet Network Construction
Retweet networks capture how information is shared in
a Twitter network. Previous work has highlighted the
need to identify different user types in order to
understand messaging patterns in Twitter data (Kwon,
Priniski, & Chanda, 2018). This information may not
only help to understand how conspiratorial narratives
grow via adoption on social media (i.e., attracting more
followers), but also may help identify target points that,
if removed, would effectively disrupt the network (e.g.,

removing a certain user from the platform). A retweet
network is represented mathematically as a digraph
(directed graph) G = (V,E), where V contains a set of
vertices (or nodes) representing unique Twitter users and
E contains a set of directed and weighted relations
between two users vi and vj in V, and where ei,j represents
the number of times user vi retweeted vj. Consequently,
the degree of vertex vi––which equals the sum of all
weighted edges connected to vi––is a proxy for the
amount of information vi shared within the network.
Because G is a digraph, edges encode directionality and
therefore ei,j may not necessarily equal ej,i. For instance,
the number of times user vi retweeted vj may not equal
the number of times user vj retweeted vi. Therefore, a
node's degree can be further decomposed into an out-
degree and in-degree component, respectively
representing the number of times vi shared other users'
tweets, and how many times other users shared vi's
tweets.
Analysis of a user's in- and out-degree can reveal their
information-sharing role in the network. Specifically,
high in-degree values indicate the user produced content
that was heavily retweeted by other users in the network,
suggesting that the person was a producer of information
in the ecosystem. Producers are likely to be central to the
production of information in the echochamber and to
development of the central narrative. In contrast,
relatively high out-degree values indicate users who
more frequently shared other users' content rather than
producing their own. Such distributors are central to the
transmission of misinformation emerging from the core
of the echochamber to users outside the network.
Distributors may lure new users into the echochamber by
exposing them to misinformation via newsfeeds.
Both user types are central to the generation and
spreading of political disinformation and synthesis of a
unifying narrative (e.g., Keller, Schoch, Styler, & Tang,
2020). Understanding the proportion of users playing
each role can reveal how the conspiracy grew and spread.
If there is relative balance between the two user types,
this would suggest that many users contribute to the
narrative by producing information, and the conspiracy
is relatively "grass roots" in nature. In contrast, if there
are more users with large out-degree values, this would
suggest that the majority of users in the network
distribute rather than produce information, suggesting
that only a minority of highly influential users are key to
the production of information in the network.
Figure 1. Distribution of out-degree (number of times a user
retweeted another) and in-degree (number of times a user's
original tweet was retweeted by another) values across the
QAnon conspiratorial network.
Table 1. Summary Statistics Describing Network Structure.
Network Statistics
Nodes
Edges
Unique
Weighted Sum
98352
430036
655216
Out/In-Degree Distribution
Min
0/0
Max
12899/1314
Mean
4.37/4.37
Variance
9011.5/172.7
Skew
76.7/23.0
As shown in Figure 1, for this dataset the range of out-
degrees values is in fact far larger than that of in-degree
values, suggesting that most users distributed
information produced by a small set of influential users.
This interpretation is supported by the differences in
statistics describing the distributions of these values
(Table 1). Notably, out-degree values were far more
variable than in-degree values. Strikingly, the extremes
of the degree distributions suggests that some users
retweeted up to 13,000 times in the dataset.
Uncovering Communities and Central Users
Understanding the global organization of a social
network can shed light on its operations. Due to the
inherent complexity of real-world networks, numerical
methods are required to reveal higher-level structures,
such as tightly-connected clusters of nodes, or
communities. Here, we apply the k-clique percolation
algorithm (or k-clique community detection) to uncover
overlapping communities across levels of
"embeddedness" in the network (Palla, Derenyi, Farkas,
& Vicsek, 2005). The goal of this analysis is two-fold:
(1) to assess the prevalence of communities across the

network, which will guide our understanding of how
information is produced and shared across users in the
network; (2) to reveal which users are most central to the
network in order to analyze their tweet content (Study 2).
K-clique community detection finds substructures in
a network by first finding all k-cliques in a network. A k-
clique is a set of k nodes in a network such that all nodes
in the set share an edge between them. In other words, a
k-clique is a fully connected set of k nodes. For example,
if there are three nodes, v1, v2, and v3 such that v1 is
connected to v2, v2 is connected to v3, and v1 is connected
to v3, then the three nodes form a 3-clique. The k-clique
community detection algorithm uses the sets of k-
cliques––and the nodes that span multiple separate k-
cliques––to assess larger substructures in the network.
Specifically, a community is constructed between two
separate k-cliques if at least one node is shared between
them. For instance, if nodes v1, v2, and v3 form a 3-clique,
and v3, v4, v5, and v6 form a 4-clique, then because the
node v3 is common to both cliques, a community of the
seven nodes is constructed. By sweeping across values
for k and extracting communities, we can get a sense of
the number of tightly-connected communities at varying
sizes. Users belonging to k-clique communities with
larger values of k are more embedded within the
network.
The number of communities resulting from k-clique
community detection with k values ranging from 5 to 12
are shown in Figure 2. There are many small
communities distributed throughout the network (above
250 when k =3), and the number goes to 0 quickly (k =
11). This finding indicates that there are not many
communities of tightly connected individuals,
suggesting that the network does not take the form of a
grassroots campaign in which many users produce and
share information with one another (e.g., Bandari, Zhou,
Qian, Tangherlini, & Roychowhur, 2017).
The lack of large, tightly connected clusters further
suggests that the early Q network may have been easily
disrupted had influential accounts been taken down. At a
much later stage in QAnon development (January 2021),
the removal of Trump and other prominent Q accounts
led to a reduction of misinformation by 73% (Dwoskin
& Timberg, 2021), highlighting the integral role of users
central to the Q conspiracy on Twitter.
Study 2: Modeling Tweet Content
We examined the semantic themes shaping discussion in
the echochamber by applying the topic modeling
algorithm latent Dirichlet allocation (LDA; Blei, Ng, &
Jordan, 2014) to the tweets of the network’s most central
users. We chose to analyze the tweets of users who are
highly central to the network, because many of the
accounts in the network spread rather than produced
information. Furthermore, a cursory analysis of the
tweets from the users more peripheral to the network
suggests that many of these users were bots and not
ordinary human users. These users shared rather than
produced original content, which is a hallmark of bot
behavior (e.g., Lazer et al., 2018), and the limited content
that they produced was largely incoherent. We aimed to
limit our analyses to tweets from human users, such as
those central to the production of content revealed in
Study 1.
Specifically, we extracted the semantic content tied
to users in different communities returned by calculating
cliques of size 9. This selection returned a set of five
distinct communities, allowing us to qualitatively assess
whether there are clear differences in the semantic
content in the tweets produced by users of each of the
five communities. We considered this number to be a
“goldilocks” value for qualitatively exploring semantic
content present in the dataset (larger values of k produce
too few communities, whereas smaller values of k
produce too many distinct communities). Thus, we fit
five topic models to the hashtags in tweets of five
clusters of users uncovered in Study 1 to examine the
extent to which semantic themes varied across the
network.
Figure 2. Number of communities uncovered via the k-clique
community detection across values of k.
Building Community-Specific Topic Models
LDA assigns each document (hashtags in a tweet) to
topics by instantiating a set of n topics, where each topic
is defined by a set of words that compose the documents.
The algorithm assigns a probability value for how likely
a document belongs to each topic. This probability value
is determined by how many of the topic’s
“representative” words appear in the tweet. Analysis of

the keywords describing the topics can shed light on the
broad semantic themes being discussed in a corpus.
Table 2 lists a subset of topics and their associated
keywords returned by the five community-specific topic
models. Common clusters of topics are prevalent across
all five communities. The semantic themes of each topic
suggest that people are framing their discussion as a
battle between Forces for Good as well as both
Unobservable and Observable Maleficent Forces. This
interpretation is supported by recent journalism work on
Q believers (e.g., LeFrance, 2020). Topics representative
of Forces for Good include: Guns (suggesting their
militarization); Trump; Rainmakers; Q, QAnon; "Where
we go one, we go all", and Michael Flynn. Unobservable
Maleficent Forces included: DeepState; Human
Trafficking Network; whereas Observable Maleficent
Forces included: Mainstream Media and DemocRats.
The theme of militarization is pervasive, revealed not
only as a topic solely devoted to Guns in cluster 1, but
also in the consistent presence across clusters of terms
such as: LetTheTribunnalsBegin, PeopleAreWeapons,
and PatriotsFight. Early Q believers apparently
envisioned a literal battle between good and evil forces,
foreshadowing the role played by Q supporters in the
violent insurrection that occurred in January 2021 in
Washington, D.C.
Table 2. Community-specific topics and keywords.
Cluster
Topics
Keywords
1
Guns
Guns, GunsAreTools,
PeopleAreWeapons
Deep State
DeepStateCabal, SuperElite,
Illuminati
Gun Control
Socialists, Guns, GunGrabbers,
SuperElite
2
Media
MSM (Mainstream Media),
QRevolution, FakeNews
Veteran
GodBlessOurTroops,
PatriotsFight, MAGAveteran
3
Deep State
LetTheTribunalsBegin,
Rainmakers, DeepState
4
Conspiracy
FakeNews, PerkinsCoie,
PedoGate, PizzaGate
Trafficking
DrainTheSwamp,
SaveTheChildren,
HumanTrafficking
5
Trafficking
HumanTrafficking, PerkinsCoie,
ChildTrafficking
MAGA
MAGA, Trump, POTUS,
TheStorm
Note. Not all topics from topic models reported here.
A general schema of Good versus Evil has long been
exploited to engage public support for political issues
(e.g., Lakoff, 1991). Understanding how this schema
shapes reasoning in QAnon can help better understand
how the narrative developed. Further, extensive research
has found individual differences in susceptibility to
conspiratorial thinking (Swami et al., 2011). Naturalistic
work on factors leading to people's early involvement in
Q can help understand which individual differences
create vulnerability to conspiracy beliefs. The most
common terms in user descriptions of the profiles in this
dataset shed preliminary light on these factors. As shown
in Table 3, users often self-identify as patriots and
Christians. Given this exploratory finding, we propose
that one’s social identification (e.g., Christian, Patriot)
leads to an adoption of a belief heuristic (Good vs Evil)
that is used to select a specific political stance (e.g., gun
rights) (see Table 4).
Table 3. Most frequent terms in user descriptions.
Term
Proportion
MAGA
.27
patriot
.07
Trump
.19
Christian
.07
love
.11
proud
.07
conservative
.11
country
.06
God
.09
NRA
.06
Table 4. Proposed social-cognitive model of QAnon belief
formation.
Social
Identification
à
Good vs Evil
Schema
à
Political
Beliefs
Christian,
Patriot,
Conservative
Trump versus
Cabal
Gun rights to
militarize
Discussion
Analysis of the QAnon Twitter network revealed that the
majority of users shared, rather than produced,
information. This finding suggests that development of
the QAnon narrative, and its effects on shaping the
beliefs of those in the network, were driven by a few key
users. We applied topic modeling to analyze the tweet
content of these users. This analysis revealed that users
may hold a simplistic mental model in which political
events are generated from an antithetical struggle
between Evil (both observable and unobservable) and
Good forces.
Our hypothesis about the mental schema of a QAnon
user is currently based on an exploratory analysis of a
naturalistic dataset. Future work––both naturalistic and
empirical––is therefore required. More detailed topic
modeling (e.g., models fit to the full text bodies of

tweets) can shed further light on the semantic structure
of this tweet corpus. A study that examines how
endorsement of conspiratorial narratives changes as a
function of their framing (e.g., as a struggle of good
versus evil as compared to alternative framings) could
shed light on the causal connection between this mental
schema and propensity to engage in conspiratorial
thinking. Additional demographic data could be used to
assess which forms of social identification (e.g., religion,
political beliefs) correlate most strongly with belief in a
good-versus-evil archetype, and with belief in
conspiracies more generally. It is possible that certain
individual differences could predict which people are
especially prone to fall for conspiracies when they are
framed as a struggle between good versus evil. Such
additional studies could help to elaborate and test the
social-cognitive model we have proposed here.
Acknowledgements
Preparation of this paper was supported by NSF Grant
BCS-1827374 to KH.
References
Altheide, D. L., & Grimes, J. N. (2005). War
programming: The Propaganda Project and the Iraq
War. Sociological Quarterly, 46(4), 617-643.
Bandari, R., Zhou, Z., Qian, H., Tangherlini, TR, &
Roychowdhury, VP. (2017). A resistant strain:
Revealing the online grassroots rise of the
antivaccination movement. Computer, 50(11), 60-67.
Blei, D. M., Ng, A. Y., & Jordan, M. I. (2014). Latent
dirichlet allocation. Journal of Machine Learning
Research, 3, 993-1022.
CIVIQS (2020). Report: Americans pessimistic on time
frame for coronavirus recovery. Retrieved from
www.civiqs.com.
Dwoskin, E., & Timberg, C (2021, January 16).
Misinformation dropped dramatically the week after
Twitter banned Trump and some allies. The
Washington Post. Retrieved from
www.washingtonpost.com.
Findlay, S. D., & Thagard, P. (2014). Emotional change
in international negotiation: Analyzing the Camp
David accords using cognitive-affective maps. Group
Decision and Negotiation, 23(6), 1281–1300.
van Harreveld, F., Rutjens, B. T., Schneider, I. K.,
Nohlen, H. U., & Keskinis, K. (2014). In doubt and
disorderly: Ambivalence promotes compensatory
perceptions of order. Journal of Experimental
Psychology: General, 143(4), 1666–1676.
Holyoak, K. J., & Thagard, P. (1989). Analogical
mapping by constraint satisfaction. Cognitive Science,
13(3), 295–355.
Holyoak, K.J., & Powell, D. (2016). Deontological
coherence: A framework for commonsense moral
reasoning. Psychological Bulletin, 142(11), 1179-
1203.
Homer-Dixon, T., Maynard, J. L., Mildenberger, M.,
Milkoreit, M., Mock, S. J., Quilley, S., Schröder, T.,
& Thagard, P. (2013). A complex systems approach
to the study of ideology: Cognitive-affective
structures and the dynamics of belief systems. Journal
of Social and Political Psychology, 1(1), 337–363.
Jankowicz, N, & Otis, C. (2020, June 17). Facebook
groups are destroying America. Wired. Retrieved
from www.wired.com.
Kwon, K. H., Priniski, J. H., & Chanda, M. (2018).
Disentangling user samples: A supervised machine
learning approach to proxy-population mismatch in
Twitter research. Communication Methods and
Measures, 2, 216-237.
Lakoff, G. (1991). Metaphor and war: The metaphor
system used to justify war in the Gulf. Peace
Research, 23(2), 25-32.
Lazer, D. M. J., Baum, M. A., Benkler, Y., Berinsky, A.
J., Greenhill, K. M., Menczer, F., Metzger, M. J.,
Nyhan, B., Pennycook, G., Rothschild, D., Schudson,
M., Sloman, S. A., Sunstein, C. R., Thorson, E. A.,
Watts, D. J., & Zittrain, J. L. (2018). The science of
fake news. Science, 359(6380), 1094–1096.
Lewandowsky, S., Gignac, G. E., & Oberauer, K. (2013).
The role of conspiracist ideation and worldviews in
predicting rejection of science. PLOS ONE, 8(10),
e75637.
Lewandowsky, S., Oberauer, K., & Gignac, G. E. (2013).
NASA faked the moon landing—therefore, (climate)
science is a hoax: An anatomy of the motivated
rejection of science. Psychological Science, 24(5),
622–633.
Mednick, S. A., & Freedman, J. L. (1960). Facilitation
of concept formation through mediated
generalization. Journal of Experimental Psychology,
60(5), 278–283.
Necklason, J. (2020, May 29). Conspiracy theories and
the Civil War. The Atlantic. Retrieved from
www.theatlantic.com.
Nyce, C. M. (2020, May 14). QAnon is a new American
religion. The Atlantic. Retrieved from
www.theatlantic.com.
Palla, G., Derényi, I., Farkas, I., & Vicsek, T. (2005).
Uncovering the overlapping community structure of
complex networks in nature and society. Nature,
435(7043), 814–818.
Papasavva, A., Blackburn, J., Stringhini, G., Zannettou,
S., & De Cristofaro, E. (2020). “Is it a Qoincidence?”:
A first step towards understanding and characterizing
the QAnon movement on voat.co. ArXiv:2009.04885
Ranney, M., & Thagard, P. (1988). Explanatory
Coherence and Belief Revision in Naive Physics.
Rodriguez, A., Bollen, J., & Ahn, Y. Y. (2016).
Collective dynamics of belief evolution under

cognitive coherence and social conformity. PLOS
ONE, 11(11), e0165910
Roose, K. (2020, August 21). What is QAnon? The pro-
Trump conspiracy theory movement, explained. Vox.
Retrieved from www.vox.com.
Saltman, K. J. (2020). Salvational super-agents and
conspiratorial secret agents: Conspiracy, theory, and
fantasies of control. Symploke, 28(1), 51–63.
Salvi, C., Iannello, P., Cancer, A., McClay, M., Rago,
S., Dunsmoor, J. E., & Antonietti, A. (2021). Going
viral: How fear, socio-cognitive polarization and
problem-solving influence fake news detection
and proliferation during Covid-19 pandemic.
Frontiers in Communication, 5, 127.
Sasahara, K., Chen, W., Peng, H., Ciampaglia, G. L.,
Flammini, A., & Menczer, F. (2020). Social
influence and unfollowing accelerate the emergence
of echo chambers. Journal of Computational Social
Science, 1-22.
Yuille, Y.L., & Grzywacz, N. (1988). A computational
theory for the perception of coherent visual motion.
Nature, 333, 71-74.
Zipperstein, S. (2020, August 25). The conspiracy theory
to rule them all. The Atlantic. Retrieved from
www.theatlantic.com.