Chapter 1 Introduction Frances Haugen revealed

Frances Haugen revealed to the world in September 2021 that Facebook had

designed its algorithm to incite its users into highly polarized discussions because it led

to increased user time spent on its platform, expanding its ad revenues (Zubrow et al.,

2021). Following this leak, it soon followed that Twitter was also leveraging its algorithm

to incite further polarization by ranking political figures with more polarizing views and

massive followings (Hong & Kim, 2016). This study generally defines polarization as the

general population's increasingly divided opinions and beliefs such that consensus

becomes increasingly more improbable. Some argue that the two-party system has

always created a polarized culture in America dating back to the political discourse

between urban and rural communities in America.

This dissertation looks at the rapid adoption of social media and its impact on

widening the political divide in the United States. Some research concludes that echo

chambers created on social media are to blame for the rise in affective and pernicious

polarization (Bessi, 2016; Bessi et al., 2016; Cinelli et al., 2021). Pernicious polarization

can be defined as the breakdown of democracy such that two distinctly different groups

distance themselves based on increasingly unmergeable ideology (McCoy & Somer,

2019). Contrarian research concludes that partisan sorting mediated by social media

platforms like Facebook and Twitter has dramatically increased opinion dynamics,

intensifying conflict, and polarization (Dubois & Blank, 2018; Flaxman et al., 2016;

Törnberg, 2022). This dissertation uses Epistemic Network Analysis, an instrument

created by Dr. Shaffer (2018) and the team at the University of Madison Wisconsin

(Marquart et al., 2019), to show how both a combination of echo chambers and partisan

sorting generated by algorithm used by Twitter are increasing the political divide in

America to increase profits at the expense of American democratic values.

Many of the problems with social media surrounding echo chambers and

increased sharing of misinformation were recognized in during my tenure as a digital

marketing consultant (2014-2021). While the study relies on other academic research,

my experience working in social media platforms observed firsthand the rapid rise of

misinformation online to sell products and sway voters. My bias is in the belief that these

platforms enable and encourage harmful content-sharing behavior, as supported by

Francis Haugen in her statements to the SEC and media publications (Whistleblower's

SEC Complaint, 2021). While this study's objective is to remain neutral, some of my bias

led to specific design choices and approaches to the methodology and research

questions. In particular, Research Question Two looks at the algorithm's impact on

Twitter and develops a method for measuring a tweet’s level of engagement, i.e.,

counting the number of likes and shares.

America’s Polarization Problem

Polarization has existed in the United States for quite some time. According to

Duverger's Law, the lack of a proportional representation system in America that favors

two parties, coupled with the Constitution's separation of House and Senate,

inadvertently created a vehicle for polarization in its attempt to preserve individual

states' identities among the nation (Huder, n.d.). A 2014 Pew Research report reported

that the American ideological divide was at the center of political polarization; Figure 1

below shows how in just twenty years, polarization has increased significantly between

the Democrat and Republican parties (Geiger, 2014).

Figure 1   
Democrats and Republicans More Ideologically Divided than in the Past  
  
Note. Reprinted from 2 from “Political Polarization in the American Public.” Pew 
Research Center, Washington D.C. (12 June 2014) 
https://www.pewresearch.org/politics/2014/06/12/political-polarization-in-the-
americanpublic/. With permission from https://www.pewresearch.org/about/terms-and-
conditions/  
  
Since social media's rapid adoption and inclusion in the daily lives of Americans 
and politicians, there has been a more significant divergence from the centrist 
ideologies of the 20th century. This dissertation seeks to determine if social media and 
the political elite drive the polarization America is experiencing today. This study will 
define the political elite as current United States 118th Congress members and former 
political leaders Donald Trump and Jair Bolsonaro. The former Brazilian President is 
included in this study to show that this problem of elite polarization isn’t limited to the 
United States but a worldwide problem on social media. Additionally, the January 8th 
uprising in Brazil was equally significant as the January 6th uprising, as there is sufficient 
evidence linking the attacks on government to the the leaders uses of Twitter (Dwoskin, 
2023; Harton, 2022). Political analyst and commentator Ezra Klein believes that the 
recent shift towards more extraordinary polarization results from a systemic imbalance 

that has encouraged parties to distance themselves as they have become ideologically

driven (Klein, 2020). It becomes nearly impossible for voters to feel comfortable shifting

from one party to another (Klein, 2020).

Historically the term conservative and liberal was not associated with political

ideology rather it referred to people's values on individual policies (Levendusky, 2009, p.

3). Conservative, previously defined in American political circles as individuals with

fiscally conservative parties evolved into an ideologically bound definition created by the

political elite (Levendusky, 2009, p. 25). Inevitably Republicans, identifying as

conservatives, started actively promoting socially conservative values in addition to their

fiscal values during the former President Regan’s tenure (Levendusky, 2009, p. 25).

This same process of tying ideology to parties occurred among Democrats, who started

identifying and sorting themselves as liberals be means of promoting social welfare, and

corporate taxes (Levendusky, 2009, pp. 3, 126). The political elite's use partisan sorting

inevitably led to much of the pernicious polarization experienced around the world today

as individuals of differing political ideologies become increasing divided (McCoy &

Somer, 2019).

Pernicious and affective rhetoric has become commonplace in modern social

media as platforms expand and monitoring and regulation become increasingly more

complicated (Levy, 2021). The vitriol dialogue among politicians makes it harder for

individuals to compromise, as negotiating with an alternative viewpoint can bring

backlash from within their community (McCoyet al., 2018). While there are many

theories and beliefs behind the causation of polarization in America, more important,

however, is the ability of the government to continue to function even when it is

polarized. This level of polarization intensity seen in America today is a significant threat

to the future democracy in the United States (Carothers & O'Donohue, 2019). It is not

without a warrant to be concerned with Twitter's potential impact on democracy

worldwide as it has been at the center of political uprisings worldwide and in the

storming of the United States Capital on January 6, 2021 (Budenz et al., 2019;

Dreisbach, 2022; Theocharis et al., 2015; Valenzuela et al., 2018).

McCoy and Somer (2019) dive deeper into the roots of pernicious polarization

due to "us vs. them" with a strong mentality distrusting the other party; this is further

ingrained because of the rhetoric of politicians like Senator Hawley or Congresswomen

Greene of Louisiana. These extremist polarized views are subsequently ending

conversations between moderate conservatives and liberals as each one might fear

their constituents seeing them as some traitor for reaching across the aisle (Klein, 2020;

McCoy & Somer, 2019).

How does a democracy avoid becoming deadlocked into endless "us vs. them"

debates? The challenges of pernicious polarization presented by McCoy et al. (2018)

suggest that the future of American democracy is at stake due to the inability for rational

debate to occur as individuals like Congresswomen Greene make highly incendiary

policy positions targeted at reasonable discourse in Congress. To further validate this, a

Pew Research poll conducted in 2016 before the Presidential elections found that

antipathy of both democrats and republicans towards the opposing party grew

significantly in the previous 22 years, 91% of Republican respondents found the

Democratic Party position unfavorable (up from 74% in 1994), and 86% of Democratic

respondents found the Republican party position unfavorable (up from 59% in 1994;

Geiger, 2016).

From a historical perspective, the growth of the internet, the expansion of

globalization, and the increase of general welfare have all potentially contributed

elements to the increase in polarization worldwide. This dissertation aims to focus on

social media's impact, specifically Twitter, on polarization and determine if it is a part of

the cause or just another effect of the rapidly dividing world population.

Purpose of Research

The following paragraph utilizes Creswell's formula for developing a purpose

statement to determine the aim of the dissertation and identify critical elements in

determining the author's goals in their work (Creswell & Poth, 2016). The purpose of this

quantitative ethnographic study will be to understand the impact of social media on the

users' political views as the world seemingly becomes increasingly more polarized. The

importance of this is immeasurable in determining if Twitter is contributing to events like

January 6 and whether there is cause for further concern regarding political divisiveness

being resonated by United States politicians on Twitter. By examining members of the

House and Senate's Twitter accounts, this study will reveal and compare the sentiments

inside members of Congress's (MoC) tweets and determine if Twitter's algorithm

rewards polarizing behavior. The study seeks to determine if polarization among political

elites is the driving force behind polarization in America, or conversely, polarization is a

pre-existing social construct and identity of the American system. Determining how

polarization occurs will impact the study of social media, especially Twitter. This study

assumes that current polarization is beyond partisan sectioning/sorting as populations

become increasingly disaffiliated from one another, and populist leaders become more

prevalent around the world (Klein, 2020).

While there is much debate between the theories of partisan sorting and echo

chambers, this study will assume that both are relevant to current polarization in the

United States and that political elites are leveraging partisan sorting to create their echo

chambers (Bail et al., 2018; Brown & Enos, 2021; Törnberg, 2022). These echo

chambers, powered by the political elite, create dangerous narratives reinforced by

unsubstantiated claims online that are wildly repeated by Twitter's algorithms (Simon et

al., 2020). One study found that while Facebook has tried to combat the spread of

misinformation since 2016, Twitter continues to fail to do so as misinformation is rapidly

increasing (Allcott et al., 2019). Misinformation is defined as the unintentional spread of

inaccurate information. At the same time, disinformation is the pervasive use of

spreading malicious or salacious rhetoric to dissuade an individual from further research

(Misinformation and Disinformation, 2023). This study aims to determine if civil unrest,

like the January 6 Uprising and the subsequent uprising following the former President

Bolsonaro in the 2022 Brazilian Presidential Elections, are instigated by the

aforementioned politicians' Tweets (Allcott et al., 2019; Bugs et al., 2023; Cinelli et al.,

2021).

The general intention of the study is to gain insight and determine if Twitter's

algorithm favors polarizing content such that it increases their overall advertising

revenue. Many publications have revealed that since Elon Musk's takeover, advertisers

have been leaving Twitter (Ghaffary, 2023; Saeedy et al., 2023). The Facebook

whistleblower, Frances Haugen, identified that the financial motivation for favoring

polarizing content led to higher levels of user engagement due to human's propensity to

share salacious content (Akinwotu, 2021; Rathje et al., 2021). Many academic

researchers have shown correlations between polarized populations engaging actively

on social media in specific echo chambers; however, the causal formation of these echo

chambers is still widely debated (Cinelli et al., 2021; Hong & Kim, 2016; Rathje et al.,

2021). Using ENA to compare politicians' rhetoric on Twitter will help identify if Twitter's

algorithm favors highly inflammatory posts.

The primary question this study seeks to answer is who is liable for the rapid

polarization of Americans. Is it Twitter that is actively contributing by not monitoring the

algorithms? Alternatively, is it the political elite, i.e., former President Donald Trump,

Senator Elizabeth Warren D-MA, and Congresswomen Greene R-GA, who propagate

provocative agendas on Twitter, which are furthering the political divide in America? By

comparing users' responses to the Twitter threads created and posted by the former

Presidents, Quantitative Ethnography and ENA will quantify the level of impact these

tweets had on the target population.

Follow-up research questions will look at users and their average levels of

engagement with Twitter based on the post type. Rubin defines the uses and

gratification approach (U&G) as a central component to understanding how people

engage with social media from a psychological perspective and will also evaluate how

social media can satisfy their needs to communicate and engage (Nabi & Oliver, 2009).

A study conducted in 2010 concluded that Twitter users were using tweets and retweets

to satisfy their communication needs with other individuals (Chen, 2011). Extending

these studies to examine how polarized users engage within their echo chambers will

be crucial in understanding if users are coming to Twitter with pre-existing bias (partisan

segregation) or if echo chambers are furthering new biases.

Summary of Research Questions and Hypotheses

The following research proposed above will consider the following questions

followed by their subsequent hypotheses:

• RQ1: Do the political elite (Members of Congress) leverage Twitter to promote

identity politics furthering political polarization in America?

• RQ2: Is Twitter’s algorithm giving preferential status to the political elite who use

polarizing tweets to generate higher user engagement resulting in higher ad

revenue?

• RQ3: Were President Donald Trump's (@realDonaldTrump) and President Jair

Bolsonaro of Brazil (@jairbolsonaro) tweets leading up to the uprising of January

6, 2021, and January 8th, 2023, respectively, responsible for the unfortunate and

subsequent events?

• H10: There is no correlation between political elite tweets and affective

polarization in America.

• H1a: ENA reveals a correlation between the tweets of the political elite and the

rise in affective polarization in America.

• H20: Twitter’s algorithm does not treat polarizing tweets with preferential

treatment.

• H2a: Twitter’s algorithm favors highly polarizing content and tweets by giving it

more significant viewership.

• H30: Former President Donald Trump’s and former Brazilian President Jair

Bolsonaro’s tweets did not significantly impact the January 6, 2021, uprising in

Washington D.C. or the January 8, 2023, uprising in Brazil.

• H3a: Using ENA, former President Donald Trump and former President Jair

Bolsonaro's tweets used rhetoric that may have significantly impacted the events

that took place on January 6 and January 8.

Methodological Approach

While there is continued debate about the existence and causes of polarization,

the research conducted in this dissertation assumes that polarization exists both in

political spheres (elite polarization) and in the public of the United States (social

polarization; Banda & Cluverius, 2018; Hare & Poole, 2014; Iyengar et al., 2019). What

is also to be assumed and measured is that societal-level grievances are being

exploited and exacerbated by the elite political rhetoric (McCoy & Somer, 2019). This

dissertation seeks to determine if politicians leverage Twitter's algorithm to maintain

leadership positions by creating non-navigable divisions using us-or-them rhetoric,

furthering pernicious polarization (McCoy & Somer, 2019). This study reviewed select

Members of Congress (MOC) from both the House of Representatives and Senate; the

account of the selected for the study will be based on their official Congressional

account, while some of these accounts may be managed by their staffers, for this study

it will be assumed that these accounts are a wholistic representation of their political

ideology and values. Additionally, since these MOC are using their official title on their

Twitter title page, the study will assume that even staff-managed accounts are

representations of their values.

Other assumptions included are that Twitter's algorithm is intended to increase

user time on the platform, such as advertisers generating more views and increasing ad

spending to maximize Twitter's profits (Hines, 2023). Buzzword or not, the algorithm is

one of the core foundations of 21st-century technology companies. Even as artificial

intelligence is making its way into the world, AI was only possible with massive

algorithmic and machine learning research developments. This study's challenge is

determining the algorithmic bias; inherently, algorithms are intentionally designed to

create a bias (Boddington, 2017, p. 17). For example, the European Court of Justice

2011 required insurance companies to eliminate gender bias in their algorithm because

it set lower premiums for females seeking pensioners insurance due to average more

extended lifespans (Boddington, 2017, p. 17). The ethics of using AI and algorithms will

continue to challenge the future of social media platforms as their rapid user uptake and

increase in world presence continue. One of the most robust assumptions made in this

study is determining the intention of the user of Twitter.

As society becomes increasingly divided, morality and ethics become less about

a shared system of justification and more about the reasoning of specific groups or echo

chambers (Boddington, 2017, p. 20). For example, followers of congresswomen

Marjorie Taylor Green (R-GA; on Twitter as @mtgreenee) will most likely have a widely

different code of ethics than those who are following Senator Elizabeth Warren (D-MA;

on Twitter as @SenWarren). Former Rep. Frank D-MA stated that "ethics were not

political until Gingrich" was elected in 1978 and while serving his first term in Congress

(Tolchin, 2019, pp. 1-2). Since the Republican Revolution of 1994, the Republican party

has latched onto its perceived view of ethics linked to Christian values, while

Democrats, like Sen. Warren, have attributed their ethics to prioritizing citizens over

corporate profits (Tolchin, 2019). Prior to this, the ethical bounds of party lines were

rarely drawn. The framers of the Constitution designed a system similar to the approach

in The Republic by Plato, in which Philosopher Kings determine the moral high ground

through careful thought (Boddington, 2017, p. 20). The Philosopher Kings are for the

sake of this dissertation, not @mtgreenee and @SenWarren; this paper assumes that

the framers of the Constitution determined a framework for moral high ground that

values Freedom of Speech (the First Amendment) and the sanctity of democracy in the

form of a constitutional republic (Boddington, 2017, p. 20).

Much of this study will use Epistemic Network Analysis and other tools to bring

quantitative results to qualitatively observed actions on a social platform. Quantifying

emotional responses on Twitter while using various existing tools like the Vader

Sentiment Analysis tool and nCoder, each has potential flaws and limitations (Hutto,

2023; Marquart et al., 2019). The tool for verifying measures of polarization created by

this study will be the DW-NOMINATE tool (Lewis et al., 2023). DW-NOMINATE was

created by Poole & Rosenthall (1985) to use congressional roll call voting to measure

each MoC voting behavior on a liberal versus conservative scale (Lewis et al., 2023). A

website maintained by UCLA named Voteview.com updates DW-NOMINATE scores

after each Congressional roll call vote and maintains up-to-date, accurate scores for

each MoC (Lewis et al., 2023).

Hemphill et al. (2016) created the #Polar scores tool using the coding language

of Python to determine how polarized a particular tweet is from a member of the United

States Congress. Additionally, #Polar scores can differentiate between users with similar

scores based on the tags used by tweeting politicians and their frequency of tweeting

(Hemphill et al., 2016). Because of the new design of Twitter and the shift towards quote

tweeting, hashtags are no longer an effective tool for gaining views and shares. This

change makes #Polar scores less effective for measuring modern politicians' Twitter

feed polarities. However, the study conducted by Hemphill et al. (2016) reviewed some

of the same MoC examined in this study, so legacy scores can be compared to the

results from the current study's ENA results to determine if members of Congress's

tweets are more of less polarized in correlation to their previous #Polarscore. This study

proposes introducing a new tool that combines the VADER sentiment analysis tool with

nCoder (software for matching large data sets with codes) to create a polarization score

based on the discourse used by each MoC (Hutto, 2023; Marquart et al., 2019). Using

DW-NOMINATE, this study can validate the polarization scores of the combined SA and

nCoder scores generated in this dissertation (Hutto, 2023; Lewis et al., 2023).

To determine the impact on users, the number of likes and retweets of members

of Congress will be categorized and scored based on the level of engagement, then

mapped on ENA to determine if the tweet has any other polarizing effect (Hemphill et al.,

2016; D. W. Shaffer, 2018). This has become increasingly more important as identity

politics continues to dominate the media's election narratives in social media (Herrera &

Sethi, 2022).

The outcome of the 2022 mid-term elections left the Republicans failing to

capture the Senate but gaining an advantage in the House. More academic research

needs to be revealed regarding the 2022 elections. However, many speculate that the

Republicans lost the Senate due to their hyper-polarized candidates pushing identity

politics rather than pushing for more centrist candidates (Everett et al., 2022). Using

tweets generated by the political elite and DW-NOMINATE will allow the study to

determine if the House is, in fact, more polarized than the Senate both in their roll call

votes and use of social Twitter (Lewis et al., 2023).

VADER sentiment analysis will again be used to help gain a broad and quick

understanding of user behavior on Twitter (Lyu & Kim, 2016). Lyu and Kim (2016)

created a sentiment dictionary to measure the strength of specific user responses in

social media. Similarly, the author of this dissertation and Dr. Eric Hamilton of

Pepperdine University used ENA to model the political discourse of user interactions on

Facebook (Hamilton & Hobbs, 2021). From these sentiments, categorizing users by

political ideology is often less challenging, as many can interact in specific echo

chambers (Baumann et al., 2020). However, independent voters/moderates are often

more challenging to identify as they might engage with media sources of all types and

interact with politicians on Twitter from both parties.

Lastly, understanding how one's engagement on social media within specific

echo chambers translates into potentially anti-democratic behavior is more challenging

to research, apart from January 6 and, in Brazil, the insurrection on January 8, 2023.

Studies have shown that algorithms on Facebook have tended to skew to show users

posts that specifically align with their political views (Van Bavel et al., 2021). Tying

incidents like January 6 to specific individuals' behaviors outside of Twitter is more

complicated; however, understanding how this type of user engaged in Twitter before

January 6 can better understand how Twitter might have impacted their decision to enter

the capitol on January 6. Similarly, specific Tweets and the former President of Brazil

Jair Bolsonaro's Telegram page have been attributed to instigating the insurrection on

January 8, 2023 (Dwoskin, 2023). In this portion of the study, the tweets of former

President Donald Trump and former President Jair Bolsonaro will be coded using

nCoder, VADER, and the ENA web tool to determine if the rhetoric used by both

presidents elicited the subsequent insurrections (Hutto, 2023; Marquart et al., 2019,

2021).

Definition of Terms

This dissertation considers the many forms of polarization to determine how and

if Twitter's elites drive any or all forms. Elite polarization is created by political and social

elites (Politicians, business & thought leaders, and celebrities) whose highly radicalized

political interests disseminate to the general population's polarization (Banda &

Cluverius, 2018). Affective polarization is when partisans become so polarized that they

dislike and distrust members of the opposite party based on party identity alone

(Druckman et al., 2021).

Van Bavel et al. (2021) defines echo chambers as a group of like-minded

individuals found on communities on platforms, i.e., Facebook or Twitter, where they can

share and confirm strongly biased opinions without considering opposing viewpoints. At

the same time, not all researchers agree on whether social media polarizes individuals;

they often agree that the echo chambers found in social media provide a catalyst for

further polarization and confirmation bias of certain groundless beliefs (Van Bavel et al.,

2021). Echo chambers, also referred to by some researchers as "selective exposure,"

are perhaps the prominent hypothesis for current theories on social media-driven

affective polarization in America (Baumann et al., 2020; Bessi,

2016; Bessi et al., 2016; McPherson et al., 2001; Tokita et al., 2021; Van Bavel et al.,

2021) However, others have argued that social media does not enable echo chambers

as users are more exposed to contrarian viewpoints in social media and its pre-existing

segregation knows as partisan sorting is driving affective polarization in America (Bail et

al., 2018; Brown & Enos, 2021; Törnberg, 2022).

Partisan sorting is the belief that pre-existing forms of segregation of voters in

America are causally linked to the increase of polarization in the United States (Brown &

Enos, 2021; Mason, 2016; Törnberg, 2022). Brown & Enos (2021) conducted a study

using election data compared with geographic data and found that much of partisan

sorting is an extension of racial/ethnic sorting, as seen in the urban and rural divide.

Perhaps the most surprising part of the study found that the most isolated ten percent of

Democratic voters had 93% or more encounters with other Democrats in their

respective urban areas (Brown & Enos, 2021). These isolated environments have

increased the "us vs. them" mentalities leading to the most problematic form of

polarization, pernicious polarization (McCoy & Somer, 2019).

As defined by McCoy and Somer (2019), Pernicious polarization is the idea that

polarization eventually becomes so unreconcilable that political actions taken by

individuals are dictated solely by their status within their subgroup or echo chamber

rather than free thought. The greatest challenge in combating pernicious polarization is

that individuals struggle to cross the political divide as their constituents see their

actions reprehensible (McCoy & Somer, 2021). This way, democracy is threatened by

pernicious polarization's rigidness in limiting free thought and speech.

Sentiment analysis is a key term used in social media studies to determine users'

attitudes, emotions, and the strength of their emotional responses when engaging in

discourse online. Specific sentiment dictionaries have been created to map these

sentiments into measurable responses such that they can be compared to other

responses (Lyu & Kim, 2016). Other sentiment analysis uses come from studying

political discourse in an analytic framework that can be translated into models used in

various quantitative and qualitative studies (Hamilton & Hobbs, 2021). Epistemic

network analysis is a mixed-method approach designed by Dr. Shaffer of the University

of Wisconsin Madison in his book, Quantitative Ethnography, which allows researchers

to use "big data" to capture quantitative and qualitative results into meaningful networks

that can highlight correlations that might not be quantifiable without a visual

representation of their network connections (D. W. Shaffer, 2018).

On average, today's social media users view their political expression or

selfpresentation online as politically active and knowledgeable (Lane et al., 2019).

These concepts of political self-awareness are central in approaching social media

political discourse as one can avail superior knowledge while feeling more

knowledgeable under the guise and protection of a digital platform that allows for

expression with managed consequences (Lane et al., 2019). For example, users can

correlate ideas and relay inaccurate historical references in echo chambers without

being corrected. Increases in political self-efficacy, the level at which one believes one

can influence politics, have given rise to political movements like QANON and the Proud

Boys (Abramson & Aldrich,

1982).

One of the more challenging terms to define is misinformation, as political

commentators regularly use the term "fake news" to describe it; however, the latter has

no actual relevance to any academic study (Ng et al., 2022). Misinformation is widely

understood to mean information not verifiable by either science or common historical

knowledge spread unintentionally (Misinformation and Disinformation, 2023). It is

essential to study how politicians disseminate misinformation on social media,

reaffirming potentially harmful ideologies leading to identity politics (Ng et al., 2022).

Social identity theory can help classify users on Twitter and determine engagement on

these platforms about events like the January 6 uprising (Ng et al., 2022).

Disinformation, as described earlier, is also imperative in understanding, although

harder to prove, as the intent of spreading false information requires insight into the

creator's intentions. Social identity theory is the belief that an individual's identity

becomes increasingly more attached to their social networks. This alludes to the

potential increases in polarization worldwide (Wakefield & Wakefield, 2023).

Significance of the Study

The study of Twitter's aims and uses of its algorithms is essential to

understanding how social media is changing political behavior amongst active users.

More specifically, gaining insight into Twitter's algorithms might help anticipate/predict

potential uprisings using machine learning algorithms (Bahrami et al., 2018). While

many might believe January 6 to be a one-off incident, and Twitter is not entirely to

blame, there are many examples of even foreign governments leveraging Twitter to try

and influence the public, for example, the Hong Kong protests in 2019 and the

insurrection on January 8, 2023, in Brazil (Dwoskin, 2023; D. Wood et al., 2019). Much

of the challenge the US government faces in regulation is determining who the

responsible actor is; Section 230 of The Communication Decency Act of 1996 suggests

that the third party, the one posting the troublesome content, is the only liable actor

(D.O.J., n.d.). However, more research suggests that the platforms could play a more

significant part in prompting potentially harmful content through its algorithms design.

One study found that when a user shares a tweet with their readers, retweeting

misinformation leads to more significant viewership and distribution of misinformation

(Pang & Ng, 2017). Perhaps more importantly, the study conducted by Pang and Ng

(2017) found that primary users were not more likely to spread misinformation in all

cases. However, followers retweeting their posts had a more significant impact (Pang &

Ng, 2017). This could indicate that further research regarding Twitter's algorithm needs

further study to determine whether Twitter users spreading potential misinformation or

Twitter's positioning of that tweet could create more harm.

Following the Mueller and Cambridge Analytica Scandal, allegations of using

misinformation on social media to interfere with elections became apparent to the

government and the United States Public (Mueller, 2019). The FTC levied the most

significant fine against a company for Facebook's negligence in protecting its user data

from Cambridge Analytica in 2019 (Fair, 2019). However, since then, there has been

little to no legislation passed in Congress to prevent such further attacks on American

data. The rapid spread of misinformation on Twitter, YouTube, Facebook, WhatsApp,

and TikTok revealed the continued lack of controls by tech companies in managing

harmful content during the early months of the COVID-19 epidemic in 2020 (Gisondi et

al., 2022; F. Simon et al., 2020).

While some believe that polarization is an irreparable situation in American

politics, the level of pernicious polarization seen both in the United States government

and the American public is cause for alarm (J. Campbell, 2018, p. 57). The rate at which

Americans are drifting away from centrist politics, as seen in Figure 1, indicates that

January 6, 2021, will not be the last attack on American democracy (Geiger, 2016).

Summary of the Proposed Study

This dissertation seeks to determine if the political elite, for this study, select

Members of Congress (MOC), are leveraging Twitter to further incite political discord in

America by leveraging identity politics through pernicious rhetoric. The second question

is to determine if Twitter's algorithms are giving preferential treatment to the MOC

studied to generate. The final question will look specifically at former President Donald

Trump's and former Brazilian President Jair Bolsonaro's tweets leading up to January 6

and January 8 uprisings to determine if the rhetoric used by the former presidents could

be correlated to the actions taken by those who led their respective insurrections.

Previous studies on elite polarization done by Lewis et al. (2023) and Pew

Research have generally shown increasing polarization in the United States over the

past 50 years, with an even more dramatic rise in the past ten years (Abramson &

Aldrich, 1982; Geiger, 2016; Voteview, 2022). Social media's impact on said polarization

has been debated on whether it is to blame. Theories on previous partisan segregation

put for by Bail et al. (2018), Brown & Enos (2021), and Törnberg (2022) suggest that

social media is not to blame as users were already polarized and social media merely

has amplified the awareness of affective polarization. While there is significant evidence

to support prior partisan segregation in America, the competing echo chamber theories

by Bessi (2016), Tokita (2021), Cinelli et al. (2021), and Baumann et al. (2020) coupled

with this study's focus on elite polarization driving mass polarization helps determine

that social media is being utilized as a tool to polarize.

The studies conducted on COVID-19 misinformation helped create the

hypotheses that the political elite is driving political polarization (Gisondi et al., 2022; F.

Simon et al., 2020). Before these studies, academics were less likely to point the finger

at the political elite. Views of political moderatism dominating voting behavior are still

pushed by Fiorina et al. (2021) versus the counter view that political elites are driving

affective polarization (A. Abramowitz, 2008; Banda & Cluverius, 2018; Hetherington,

2001; Zingher & Flynn, 2018).

Most researchers have encountered challenges studying polarization online

Twitter varies significantly as the platform frequently changes both the algorithm and the

users engaging on Twitter. One study on the amplification of politics on Twitter showed

that randomizing control groups in studying the effects of the interaction on social media

was impossible due to the nature of the content being shared by the user on the

platform (Huszár et al., 2022). VADER Sentiment Analysis (VADER) tool also presented

challenges to other researchers, as the tweets are character-limited, and users might

employ the use of sarcasm or jargon not yet classified by the VADER tool (Hutto, 2023;

Lyu & Kim, 2016; Misiejuk et al., 2021).

This study is needed now more than ever; misinformation and disinformation

campaigns are driven by social media platforms' algorithms and political elites who fuel

messages of affective polarization (Akinwotu, 2021; Atad et al., 2023; Tønnesson et al.,

2022). Social commentator and comedian Sascha Baron Cohen revealed in his

acceptance speech to the Anti-Defamation League in 2019 that he was able to convince

a slightly radicalized Trump follower to use fake explosives (which the subject believed

were real) to attack and theoretically kill members of Antifa based on misinformation he

was feeding him (Cohen, 2019). Sascha Baron Cohen, in his speech, referred to

Voltaire's famous quote in Questions sur les Miracles, "Those who can make you

believe in absurdities can make you commit atrocities" (Cohen, 2019). While

experiments conducted by Sascha Baron Cohen have zero academic credibility, it

provides an allegory to a potential situation that has been witnessed in the case of many

tragedies, including the Capitol Breach on January 6, where five people died. This study

ultimately seeks to show how extremist language used in social media, specifically on

Twitter, leads disenfranchised public members to commit violent uprisings (Capitol

Breach Cases, 2021; Harton et al., 2022). These members of society are distorted by

pernicious polarization. They are told by political elites that there is no compromise with

the other side and that, in some cases, violence and uprising are the only solutions

(McCoyet al., 2018). The threats to democracy are genuine with the rise in populism

worldwide and the constant reaffirmation of such radical beliefs.

The elite movement catalyzes the affective and pernicious polarization society is

encountering across the globe (McCoy & Somer, 2019). This dissertation aims to show

that using ENA coupled with the VADER tool and nCoder will exemplify how Twitter's

rapid and mass delivery capability spreads uncontrollable pernicious discourse, leading

to attempted and potential democracy in America.

Chapter 2: Review of Literature

Background and Research

While the origins of polarization in American society are debated widely by

academics and politicians, the effects of social media on polarization were not seriously

considered until the COVID-19 epidemic and Frances Haugen's interview with 60

Minutes (Akinwotu, 2021; Hart et al., 2020). Some argue that polarization was

preexisting in social media and that, most likely, platforms like Facebook and Twitter

gave greater awareness of the existing polarization (Banks et al., 2021). This sparked

the debate between partisan sorting (existing segregation of voters into polarized

groups) and echo chambers (homophilic groups that create platforms to extend

confirmation bias to users everywhere). Törnberg (2022) is perhaps one of the most

prominent critics of echo chambers' causal effects on polarization and believes that

existing segregation is amplified on social media to create the perception that social

media is creating echo chambers. While substantial evidence supports the pre-existing

segregation, the subsequent echo chambers have become too large to ignore. The

creation of platforms like Truth Social, Parler, and even Elon Musk's purchasing of

Twitter signaled the dominance of the homophilic interest of a few to create and buy

social media platforms such that their voices can be heard (M. Otala et al., 2021).

This literature review will seek to explain how the political elite has leveraged the

flaws of human nature and social media to influence increased partisan segregation into

seeming uncontrollable polarization. The first portion of this literature review will focus

on the historical perspectives of segregation and polarization in America. With a general

review of the theories and types of polarization. Leading to definitions of the harsher,

more extreme cases of pernicious polarization followed by a general review of social

media regulation. Lastly, this literature review will follow the methodical approach of

reviewing previous studies measuring discourse and polarization on social media and

how they impact this study.

History of Polarization

While Americans have had ideological differences since its foundation, mainly

seen in urban and rural divides, divisive political polarization is something that has only

varied in intensity throughout its history. The principles of the Constitution called for

politicians to compromise and put aside differences as a means to an end (J. Campbell,

2018). However, the electoral college and development of the House of

Representatives, made up of congress members from area-specific districts, as

opposed to a whole state or country, lends itself to a two-party system according to

Duverger's Law (Hare & Poole, 2014, p. 414). Duverger's Law is the understanding that

in proportional electoral systems, the United States included participants tend to vote for

the party/person with the best chance of expressing and matching their ideologies

(Schlesinger & Schlesinger, 2006). Hare & Poole (2014) reference the historical

polarization and depolarization of the United States as a cycle. This cycle remained in

balance; however, in the past ten years, ideological framing and identity have become

more critical to party identity than ever, further accelerating what seems like

unremovable polarization.

The first mass political polarization in the United States occurred in the lead-up to

the Civil War (Hare & Poole, 2014). This mass polarization occurred because the

Southern Democrats and Whigs felt affronted by the North with their high tariffs on

exporting commodities and the North's push to end slavery (Hare & Poole, 2014). Mass

polarization, in this case, was brought on because it impacted Southerners economically

and culturally, as well as a fear of overreach from the Federal government (Hare &

Poole, 2014). The impact of the end of the Civil War should have narrowed much of the

preexisting ideological and cultural gaps; however, a disgruntled President, Andrew

Johnson, reinstated racial animosity and inequality when he returned power to white

supremacists (Klein, 2020, pp. 34-35).

The dominance of the Republican Party following the Civil War leading to the Great

Depression led to some deep-seated resentment among Southern Democrats (B. D.

Wood & Jordan, 2017, p. 49). However, from the Great Depression until the Civil Rights

movement, there were significantly low levels of polarization (B. D. Wood & Jordan,

2017, p. 128). The slow return to institutionalized racism in America, especially in the

South, continued for the next hundred years following the Civil War. These cultural

differences between parties in America remained at ease, and mass polarization did not

occur again until the Civil rights movement in the middle of the 1960s (Hare & Poole,

2014). The Civil Rights Movement of the 1960s was monumental because it

transcended both parties and brought a resurgence of mass polarization in America

(Campbell, 2018, pp. 54). As it had after the Civil War, Southern Democrats felt

affronted by the Federal Government's overreach with the passage of the Civil Rights

Act of 1964 and the Voting Rights Act of 1965 (J. Campbell, 2018; Hare & Poole, 2014).

These affronts not only led to the inevitable end of the Dixiecrat as Southern Democrats

became Republicans and Democrats in urban communities brought on minorities to its

party (Carmines & Stimson, 1989, pp. 62). Campbell's Revealed Polarization Theory

(2018) credits the Civil Rights movement, and the challenges faced in the 1960s were at

the heart of the foundation of polarization today.

In the late 1960s through 1970s, the continued liberalization of the Democratic

party, coupled with the Republican Party's shift towards strict conservativism, led to the

modern forms of polarization seen in American politics today (Hare & Poole, 2014; K.

Poole, 2008). During this time, a divergence started forming among political elites as

Democrats and Republicans started to become more isolated from one another. Pauline

Kael was famously quoted for saying in 1972 that she "did not know how Nixon could

have won because I did not know anyone that would have voted for him" (Brandt &

Spälti, 2018). This belief in perceived social norms by social sampling is possibly the

beginning of the future's echo chambers and filter bubbles (Brandt & Spälti, 2018;

Flaxman et al., 2016). In the 1980s and early 90s, there was an increase in elite political

polarization as income inequality, and immigration rose (K. Poole, 2008). Former

President Regan rolled back civil rights reforms and began dialogues introducing

religious ideology (Levendusky, 2009, p. 25). While at the time, these issues weren't

vital enough concerns to polarize the masses, politicians and political elites became

increasingly divided among these ideological concerns (Hare & Poole, 2014).

Gradually members of Congress became increasingly more divisive over

concerns of gun control, abortion, and social welfare to the point where ideological

values started becoming platforms for politicians(D. Green et al., 2004, pp. 210–211).

The use of sorting individuals' ideological values by politicians inevitably leads to the

general polarization seen in society today (Iyengar et al., 2019; Klein, 2020). Like the

1960s, many social issues centered around race (D. P. Green et al., 2004, p. 3). The

Democratic party has become increasingly more diverse as the Republican party finds

itself losing favor with minorities (Klein, 2020).

Former President Obama's win in the presidential election in 2008 was not met

without racism as birtherism, touted primarily by Donald Trump, that Obama was not

born in the United States but instead an African Muslim (Klein, 2020). This racism only

continued into President Donald Trump's electoral victory as he campaigned on

ideological concerns over concerns of illegal immigration (Harton et al., 2022). President

Trump's tenure, coupled with the advent of the COVID-19 epidemic, accelerated

polarization to new heights as party ideology became coupled with public health

strategy (Morris, 2021). A study conducted using 2016 voter data coupled with COVID19

cases and morality data in the Spring and Summer of 2020 found that while counties

that did not vote for Trump had higher death rates in the earlier months later were

surpassed by Republican counties as the summer passed on (Morris, 2021). Morris

(2021) noted in the study that political ideology led to wildly divergent health strategies

for managing COVID-19. This draws much concern as COVID-19 revealed how deep

partisan ideology became more important in the United States in manners of public

health (life or death) than the well-being of the people (Morris, 2021). Sadly, both parties

were equally to blame for politicizing COVID-19 responses (Morris, 2021). The media

essentially aided this political elite manipulation by giving more airtime to politicians than

scientists and medical professionals during the epidemic's early stages (Hart et al.,

2020).

From COVID-19 to January 6, the United States has been besieged with the

realization that social media was primarily to blame for the massive influx of

misinformation online (Gisondi et al., 2022). Further research will reveal that COVID-19,

coupled with the alternative narratives pushed forward by former President Donald

Trump and the company, led to a greater distrust of mainstream media, driving

polarization to a new height.

Theories & Types of Polarization Explained

Campbell (2018, pp. 40-41) attributes modern polarization to three different

theories; the first is the Emerging Polarization Theory, which argues that polarization

happens in political party leadership, and their electorate follows. The Emerging

Polarization Theory identifies polarization as a more modern concept and that extreme

polarization did not occur until the 2000s (Campbell, 2018, p. 40). According to a Pew

Research poll (Duggan e al., 2016), which showed voters the departure from centrist

ideology in the late 2010s, historically, as identified by Hare and Poole (2014), the cycle

of polarization has been around since the Civil War.

Campbell's (2018, p. 40) second theory of polarization is the No Polarization Theory

which suggests that the public is still largely centrist and that polarization is, in fact, a

myth. The midterm elections in 2022 hypothetically indicated that America is growing

tired of extremist politics, as decidedly polarized candidates did poorly while moderate

candidates thrived. On the contrary, a Harvard University Poll (Harvard IOP Youth

Survey, 2022) revealed that 59% of Generation Z voters polled were planning on voting

in the midterms in record numbers because they felt their rights were under attack by

extremist politicians. Campbell's (2018, p. 40) No Polarization Theory is problematic and

can only be defended as a form of ideological sorting (Klein, 2020).

The third theory of polarization, defined by Campbell (2018, p. 41), is the

Revealed Polarization Theory which claims that Americans have been highly polarized

since the 1960s and that politically homogenous parties have masked polarization. This

only became apparent to the population as the Republican and Democratic parties in

the 1990s and 2000s started to take on polarized platforms (Campbell, 2018, p. 41).

Ezra Klein, author of the book Why We Are Polarized (2020), argued that the core

component of current polarization is an extension of Campbell's Revealed Polarization

theory (2018), which looks at identity politics as the core component of current

polarization.

According to several polls, over 40 percent of Americans believed a second civil

war was imminent (Orth, 2022; Zogby, 2021). While there is some debate about whether

America is on the precipice of a second civil war, the division of Americans during the

1960s and early 70s by ethnicity and socioeconomics led to what is known as greater

mass polarization (J. Campbell, 2018). Mass polarization, or group polarization, resulted

from this as communities again, like the first industrial revolution leading to the Civil War.

Political parties divided communities as they embraced the ideological concerns of their

constituents and pushed away from bipartisanship (J. Campbell, 2018;

Sunstein, 1999;).

The continued focus on identity politics by the Democratic and Republican

parties further revealed the partisan sorting in America (Fiorina et al., 2008; Klein,

2020). Issues like abortion, healthcare, social security, continued racial inequality, police

reform, and immigration are transitioning from sorting to party mainstays as they ingrain

themselves into party identity and further deepen polarization (D. P. Green et al., 2004;

Klein, 2020). Opportunistic political actors, like Marjorie Taylor Greene (R-GA) and Matt

Gaetz (R-FL), leverage existing societal cleavages (created by partisan sorting) to build

their agendas and dominate political discourse among their radicalized subgroups

(McCoy & Somer, 2019). Political cleavages do not account for polarization; however,

once political actors embrace these cleavages as part of their platforms and ideologies,

they create formative rifts, resulting in pernicious polarization (McCoy & Somer, 2019).

There are many examples of how identity politics continue to interfere with the success

of the United States and the rest of the world. The Greater Idaho Movement seeks to

expand Idaho's borders into Eastern Oregon to create a more significant conservative

state because it does not match the "cultural divide" (The Greater Idaho Movement,

2023). This is quite terrifying in theory as it suggests that states should be bound by

cultural identity and nothing else. The principles set forth by the founding fathers of the

United States of America relied on compromise and civil debate to frame a constitution

that created platforms that allowed reasonable discourse and checks and balances.

When the political elite push identity politics to determine a state's borders, this is a

prime example of albeit reprehensible and pernicious behavior.

To outsiders, one of the most fascinating aspects of American politics is the

focus on national politics over regional politics, which typically have a far more

significant impact on their daily lives (Hopkins, 2018). The internet and social media

have stimulated national political identity more than ever. In the 2000 presidential

election, Tim Russert, who at the time was at NBC, was credited for popularizing the

red-state vs. blue-state terminology while discussing the elections (Crouch & Rozell,

2014). These "red vs. blue" narratives exploded in the media as increasingly everything

seemed to become some form of "us vs. them" (Crouch & Rozell, 2014). It has gone so

far that most Americans no longer perceive media as non-partisan (Crouch & Rozell,

2014). Social media has seemingly fanned the flames of the already highly charged

political rivalry among political elites in the past decade.

Affective to Pernicious Polarization

Affective polarization is when partisans become so ideologically bound to their

party's values that they disaffect themselves from the contrary opinions (Iyengar &

Westwood, 2015). Ultimately the level of disaffection from other members of society

leads the affectively polarized to believe the other party is uncompromising (Iyengar &

Westwood, 2015). For example, fundamentalist Christians support the Republican party

based on ideological values and avow support to any of its leaders regardless of their

status as a Christian, i.e., former President Donald Trump. Affective polarization

increases as ideological values are amplified and increased political turmoil (Iyengar et

al., 2019). The events following the January 6 insurrection and division in mainstream

media and later isolated disclosure of evidence to a partisan media source, Fox News,

by House Speaker McCarthy enabled both affective and pernicious polarization

(Grisales & Swartz, 2023).

Pernicious polarization, coined by McCoy and Somer (2019), is when individuals

cannot cross party lines or ideological boundaries without fear of reprisal from their

respective cohorts. With pernicious polarization, one's political identity becomes

tantamount to each of their actions. Any action out of touch with that particular ideology

can further that individual's social relationships and positions even in their respective

neighborhood (McCoy & Somer, 2019). One catalyst for this behavior has been social

media which has led to the self-segregation of individuals into echo chambers (Törnberg

et al., 2021). Group polarization is the belief that when groups segregate themselves

into ideological enclaves and expel contrary discourse, they become more radicalized

(Sunstein, 1999). Democracy struggles to thrive when pernicious polarization occurs

because civil discourse is limited and often blocked.

Partisan media has exploded since the advent of the internet and exacerbated

pernicious polarization. Normative views of bipartisanship are frequently under attack by

mainstream media. As a result, viewers find themselves far more uncompromising and

furthering themselves into affective modes of polarization (Levendusky, 2013). One

study found that partisan polarization was a key physiological driver in spreading

misinformation on Twitter (Osmundsen et al., 2020).

Divisive Alienation

Divisive alienation has become one of the unfortunate outcomes for many as

pernicious polarization continues to plague America on social media. One study found

that as the shareability of news on social media increased, polarization increased as

users reflected less tolerance to contrarian viewpoints and media sources (Coscia &

Rossi, 2022). While shareability was previously viewed as a positive for social media,

the speed at which false information can be easily shared and repeated has potentially

harmful circumstances (Coscia & Rossi, 2022). One Twitter engineer who worked on the

"Retweet" button later admitted that he regretted this decision as harmful content

sharing exploded in use (Haidt, 2022). Since 2020, Twitter has tried to limit frivolous

sharing by asking users if they want to share a link they have not opened yet (Hern,

2020). The idea was that if the user has to read the link first rather than just the

prepopped headline, the user might second-guess sharing the article (Hern, 2020).

Twitter and Facebook have made other attempts at reducing misinformation by

using misinformation labels on posts (Papakyriakopoulos & Goodman, 2022). One study

found that warning labels on Donald Trump's tweets about election fraud in the 2020 US

Presidential election did not change the magnitude of users' interaction with the tweets

(Papakyriakopoulos & Goodman, 2022). However, the study found that labels did

reduce users' propensity to create harmful content and retweet information labeled as

misinformation (Papakyriakopoulos & Goodman, 2022). Sharing content in the hopes of

becoming viral became a part of the social media game, and users tend to contribute

maliciously as the platforms seem to favor content that incites mob dynamics (Haidt,

2022).

The COVID-19 epidemic in 2020 gave significant rise to a partisan divide across

the globe as governments attempted to manage an unmanageable virus with various

strategies, each currying or losing favor with its respective base. The general approach

across right-leaning states in the United States was to remain open and wait for herd

immunity; Governor Ron DeSantis was praised for keeping Florida open and gained

significant status as a Republican national leader. On the Opposite side, left-leaning

states, like California and New York, experienced lengthy and big pushes for mass

vaccination. Governor Gavin Newsome of California was praised for his response by

Democratic politicians across America. Neither leader had any noticeable similarities in

their COVID-19 mitigation strategies, and both emerged as successful leaders during

the epidemic, each praised for their approaches. Much of this resulted from the highly

politicized viewpoints of traditional newspaper and network news coverage (Hart et al.,

2020). Hart et al. (2020) found that politicians were more regularly featured than

scientists in newspaper coverage.

Two studies found that conservative respondents showed greater trust in

government authorities to manage COVID-19 than the World Health Organization

(WHO) and scientists (Kerr et al., 2021). Among Congressmen, the typical tweets

among Democratic members were promoting COVID-19 safety and threats to public

health, whereas Republican members blamed China and spoke of damages to

American industry (Kerr et al., 2021). The decrease in cross-party relationships and

cooperation was further exacerbated by COVID-19 and the January 6 uprising, as

politicians showing any amount of party distancing were immediately discredited (Haidt,

2022). Liz Cheney suffered the consequences of Pernicious polarization when she

voted to impeach Donald Trump following his role in the January 6 uprising. She was

subsequently voted by the Republican Party of Wyoming to no longer recognize her as

a Republican (Associated Press, 2021). This type of pernicious polarization has

occurred in political circles and communities around America, both on and offline.

Politically sorted social networks have emerged more robust as users tend to

favor homogenous viewpoints and begin to isolate themselves from contradictory

ideologies (Tokita et al., 2021). A study on Twitter users found that individuals were

becoming increasingly more likely to unfollow users and create homogenous social

environments where they were less likely to continue to follow new sources with

crossideology (Tokita et al., 2021).

Social Media Regulation Effects on Polarization

Social media regulation has been virtually nonexistent on a Federal level as

social media companies claim immunity from misleading or harmful content posted on

their platform using Section 230 of the Communications Decency Act of 1996 (Cramer,

2020). Section 230 was created before the existence of social media as it is seen today

while enforcing it is legally valid, morally unethical, and dubious as it has enabled the

proliferation of misbehavior in social media (Cramer, 2020). Section 230 has also been

misused in assuming that it allows social media to censor specific individuals; for

example, politicians Ted Cruz (R-TX) wrongfully criticized the law saying Facebook

leveraged it to censor right pundits like Alex Jones (Cramer, 2020).

The challenge with Section 230 is that it does protect free speech, but at what

cost? The reality of Internet companies policing and moderating content on their

platforms has led to disastrous consequences, like the rallying of users to commit

genocide in Myanmar (Cramer, 2020). While Facebook certainly did not intend to enable

this behavior, its platform created the vessel to allow for such actions to occur;

ultimately, legal expert Cramer argues that corporate social responsibility (CSR) that will

leverage Facebook to be its police for the good of its users as well as profits (Cramer,

2020). The moderation challenge is inevitably cost-driven; Twitter and Facebook could

not be profitable if humans prescreened all tweets and posts (Goldman, 2018). While

some want to create more significant restrictions around Section 230, the reality is that

doing so would create greater protections for existing internet giants like Google and

Facebook, who can take on any new regulatory costs (Goldman, 2018). Start-ups can

compete against Facebook and Google with the ability not to worry so much about

moderation and focus more on growth (Goldman, 2018).

While the Federal Government has taken it upon itself to remain committed to

Section 230, other states have enacted stricter laws enforcing greater scrutiny of social

media platforms. In March 2023, the state of Utah enacted a law that requires explicit

parental permission for anyone in households under the age of eighteen to use

Facebook, TikTok, Instagram, and Facebook (Singh, 2023). The ban also calls for

platforms to moderate and adapt their platforms such that they are non-addicting for

underage users (Singh, 2023). While this ban is explicitly targeted at youth, there is also

a more significant cause for concern as now Civil Liberties groups point out that parents

have control over their children's accounts and might be able to single out LGBTQ+

(Singh, 2023). While it is difficult to answer precisely, children are subject to first

amendment rights (Garvey, 1979; Tinker v. Des Moines Independent Community School

District, 393 U.S. 503 (1969)). Whether or not this applies to children's content being

monitored by their parents is still debatable, and it seems there will be potential legal

cases against Utah's bill.

Regulating content on social media platforms and the debate of free speech has

been a challenging subject, one that now Twitter CEO Elon Musk has faced in the past

with an SEC violation regarding a tweet he had made online (Krisher, 2022). The subject

of hate speech, insider trading, child pornography, and misinformation have all

presented legal challenges in the United States with little or no avail to setting up

concrete laws to protect users from what is still a wildly complicated platform to police.

Misinformation, commonly referred to as "fake news on Twitter, is perhaps the most

relevant subject regarding determining if social media is in effect, leading to any form of

affective or pernicious polarization. A complex study looking at "fake news" tweets in

which articles promoting misinformation were shared on line garnered a significant

amount of polarized debated in the comment threads (Ribeiro et al., 2017). Ironically,

the study found that "fake news" was generally used not as a tool for identifying

misinformation but for some users, particularly right-leaning, to express disagreement

(Ribeiro et al., 2017).

The Mueller Report presented evidence of Russian election interference with

misinformation campaigns on social media (Polyakova, 2019). Subsequently, the CEOs

of Facebook, Mark Zuckerberg, Twitter, Jack Dorsey, and Google, Sundar Pichai, were

called to testify in a Senate hearing multiple times over the allegations of potential

election interference on their successive platforms (Guynn, 2020). Unsurprisingly, even

with the evidence that Robert Mueller had presented, no laws were passed in Congress

to protect Americans from any future harm caused by misinformation campaigns (Kim,

2020). Even less surprising is that election interference through social media campaigns

continued even into the 2020 presidential elections, except this time, rather than

individuals being surprised, it seemed par for the course (Kim, 2020).

In 2020 the COVID-19 epidemic led to a monumental rise in misinformation being

distributed on social media (Gisondi et al., 2022). The ease of sharing on social

platforms allowed users to leverage algorithms and rapidly disseminate antivaccine

information, questionable cures, and wild conspiracy theories as to the origins of the

virus (Gisondi et al., 2022). Even before COVID-19, misinformation on human wellbeing

was being exploited by users on social media; one study found that over seven years

(2012-2018), the number of eligible articles addressing health misinformation online

increased from 7 to 41 a year with a substantial rise in 2017 (Wang et al., 2019). Some

of the sharp rises in vaccine hesitancy, even before the COVID-19 epidemic, was

identified as some community's objection to a corrupt elite (McKee & Diethelm, 2010;

Wang et al., 2019). The massive uptake in 2021 is primarily attributed to the increase in

celebrity and high-level authority figures' misinformation (Gisondi et al., 2022; F. Simon

et al., 2020). Simon et al. (2020) found that these authority figures and celebrities only

accounted for 20% of the misinformation on social and traditional media but 69% of the

total share of engagements on social media. These massive uptakes in misinformation

led to some of the first significant examples of censorship led by Facebook and Twitter;

Facebook alone removed over 7 million posts and added warning labels to another 98

million between April and June 2020 (Lerman, 2020).

The sheer volume of posts flagged by Facebook in just three months should offer

insight into their capabilities in managing misinformation. Gisondi et al. (2022) believe

that part of the problem is in the scientific and medical communities' failure to use social

media correctly and convey messages that are accessible to users at all levels. These

massive rifts and echo chambers created among users on social media by health

misinformation are not unlike the rifts seen in political discourse online. Monitoring and

holding politicians accountable for their social posts is tantamount to the reduction of

polarization across the world. The unprecedented blocking of former President Donald

Trump's Facebook and Twitter accounts on January 7, 2021, was monumental in

establishing a clear line in First Amendment rights online (Conger et al., 2021).

Ethical Implications and Future of Media Regulation

The challenge of social media and media regulation is that the government is an

elected body that changes every two years in the case of the House, six years in the

Senate, four years in the Presidency, and a lifetime in the Supreme Court. On the other

hand, companies are reactive. New tech companies are creating technologies faster

than lawmakers can respond to past technologies. By the time bills are signed into law,

they are often no longer effective at managing the new technologies. Facebook and

Google have resisted becoming arbiters of political discourse, yet they actively monitor

paid content on their platforms (Kreiss & Mcgregor, 2019). Industry self-regulation is

needed now more than ever as artificial intelligence becomes a large part of modern

corporations' futures. Social media never intended to become a source of American

news (Bell, 2016). Even more surprising is how many Americans consume their news

on social media; a Pew Research Poll found that, even though social media news

consumption has decreased since 2020, nearly 50% of Americans Sometimes reported

or Often get their news on media (See Figure 3 below; Pew Research Center, 2022). Of

those, 31% got their news from Facebook, 14% from Twitter, and 25% from YouTube

(Pew Research Center, 2022). These are massive numbers with powers to generate

substantial effects with little government oversight.

Figure 2  
News Consumption on Social Media Sites from Pew Research Center  
  
Note. Reprinted from “Social Media and News Fact Sheet.” Pew Research Center, 
Washington, D.C. (20 September 2022) 
https://www.pewresearch.org/journalism/factsheet/social-media-and-news-fact-sheet/. 
Reprinted with permission.  
While there are arguments for and against protecting Section 230 of the 
Communication Decency Act, the challenge of regulating platforms extends beyond the 
free speech protections as Facebook, Twitter, and other platforms allow paid 
promotional content (Cramer, 2020, p. 135). Kreiss & Mcgregor (2019) conducted a 
study on Facebook and Google’s paid political advertising and found that while paid 
media was being monitored rigorously, public content was seldom reviewed. Advertisers 
were a primary force in pressuring social media companies to hire more content 
moderators after brands became concerned with their advertisements being placed next 
to conflictual content (Cramer, 2020, p. 135). The fear of reprisal or backlash from brand 
loyalists has been a concern of many companies on social media as social media 

content has seemingly become more and more disdainful. However, the platforms have

argued that social media promotes free speech. Before his takeover of Twitter, Elon

Musk was embroiled in legal controversy regarding his tweet suggesting he had the

offer to take Tesla private at a higher valuation, which subsequently drove the stock

price up (SEC.GOV, 2018). The SEC subsequently charged Elon Musk with securities

fraud and required him to have all of his tweets preapproved by a lawyer (Hawkins,

2023; SEC.GOV, 2018).

Twitter vs. The World: Who is Responsible?

The foundation of modern corporations can be traced back to the East India

Company of the 17th century (Roy, 2012, p. xi). The significant difference between the

formation of the East India Corporation in the 17th century and corporations today is

that in the 17th century, they relied on private interest from other lords and ruling elite

members (Roy, 2012, p. xiii). In contrast, today, the general public has the power to be

owners of corporations (Roy, 2012, p. xiii). By the 18th century, the industrial revolution

pushed away from monopolistic structures managed by the ruling elite as individual

members of the public became enabled to become owners of corporations (Roy, 2012,

p. ix). Today significant corporations like Google, Facebook, Twitter, and Amazon can

create changes that impact society in instrumental ways, often with more significant

impact and speed than the government. This presents an ethical dilemma faced by the

United States government and governments worldwide.

When Elon Musk acquired Twitter in October 2022, many conservative elites and

pundits rejoiced. Senator Ted Cruz (R-TX) hailed it as a significant win for free speech

(Rohlinger et al., 2023). The immediate aftermath of Musk’s takeover was significant as

he cleared out his auditing team, and hate speech dramatically increased (Conger &

Frenkle, 2022). Shortly following Musk’s takeover, previously banned accounts, i.e.,

@realDonaldTrump and several others, were unbanned, giving access to users deemed

by the previous leadership as harmful (Rohlinger et al., 2023). After this unbanning, a

study looked explicitly at rhetoric before and after the banning in regards to the Arizona

election audit, the study found that even with the ban, the type of information regarding

the audits was no different, and in fact, conspiracy theories regarding the ban seemed to

increase (Rohlinger et al., 2023). This marks a challenge for social media companies

and not just Twitter. Does banning certain accounts prevent the dissemination of fake

news? Rohlinger et al. (2023) study suggests that banning Twitter accounts did not

mitigate any misinformation spread or amplification on Twitter. This presents a challenge

for not only Elon Musk but other technology leaders like him to determine what their

responsibilities are in the spreading of misinformation.

January 6 and the genocide in Myanmar were violent and tragic events strongly

correlated to the misuse of social media platforms (Harton et al., 2022; Tønnesson et

al., 2022). Harton et al. (2022) use the dynamic social impact theory (Latané, 1981,

1996) to determine that the ease of communication on social media between likeminded

and troubled individuals led to the January 6 uprising. DSIT is an extension of Latané’s

social impact theory (1981) which suggest the emergence of cultural elements are

connected and form group-oriented values and dynamics based on clustering, once

thought to be a regional now extending into social media as users can cluster online

based on their values (Latané, 1981, 1996). A literature review found that individuals

were persuaded by one another, and affirmations from then-President Donald Trump led

to the rise of the mob that unlawfully entered the capitol on January 6 (Harton et al.,

2022). In Myanmar, military groups made Facebook pages to convey their status as

legitimate states (Tønnesson et al., 2022). While Facebook attempted to ban these

groups’ pages as they were created, the forces would turn to hashtags to continue

sharing misinformation (Tønnesson et al., 2022). The Myanmar government was forced

to shut down the internet in seven Rahkine townships from June 2019 to February 2021

because paramilitary groups used Facebook for military operations and commands

(Tønnesson et al., 2022). As these social media platforms grow, managing content

becomes extremely difficult and costly. However, these social media companies profit

from sharing user-generated content by selling advertiser space. They have a moral

obligation to protect both their users and their advertisers.

Methodologies and History of Polarization Studies

Methods for Measuring Political Divide

To determine how affective polarization impacts society can be done using the

Törnberg et al. (2021) model, which looks at identity-centered politics as a driver for

social or affective polarization. The model looks at how the internet, specifically social

media, allows individuals to isolate themselves and engage in spaces where other

likeminded individuals can avoid conflictual ideas (Törnberg et al., 2021).

Social media has the power to both enable access to free speech and

disseminate democratizing values, but using the same rapid information release can

also harm democratic values by propagating propaganda aimed at harming particular

groups or instilling doubt among citizens (Lorenz-Spreen et al., 2022). To study the

effects of social media on democratization, Lorenz-Spreen et al. (2022) used two

approaches focused on observational data that provided correlational evidence; the first

was looking at articles that examined social media and democracy. The second

approach was a deep analysis of the articles reporting causal evidence of these

breakdowns in democratic values (Lorenz-Spreen et al., 2022). The results indicated

that there were, in fact, a significant number of negative correlations of polarization

found on Twitter, "Political Parties," and "Social Media" (Lorenz-Spreen et al., 2022).

These correlational studies also found that Twitter users were consistently embracing

homophily and engaging in their echo chambers at higher rates than other social media

sources (Lorenz-Spreen et al., 2022).

Some studies suggest that while polarization does exist, it exists in the absence

of social media; these preexisting echo chambers have dominated negative discourses

online, leading to a perceived increase in polarization (Buder et al., 2021) (Banks et al.,

2021). It is suggested that negative social media frames are more likely to increase

perceptions of polarization even if they lack policy content (Banks et al., 2021).

However, another study looking especially at Twitter users and their engagement with

contrarian political views through a guided study found that previously conservative

users became more conservative following engaging with the study (Bail et al., 2018).

These studies, however, are limited in that it is hard to isolate independent voters in the

findings as they are less likely to engage in a predictable or repeatable manner such as

a highly liberal or conservative Twitter user would (Bail et al., 2018; Banks et al., 2021;

Buder et al., 2021).

One of the most used tools to determine political polarity on members in both

houses of Congress is NOMINATE (now referred to as DW-NOMINATE, Dynamic

Weighted NOMINA Three-step Estimation; K. T. Poole, 2005, 2007; Voteview, 2022).

DW-NOMINATE, created by Keith T. Poole and Howard Rosenthal in the 1980s,

evaluates every congressional vote and spatially places each member on a

conservative to liberal scale (Examples seen in Chapter 4; K. T. Poole & Rosenthal,

1985; Voteview, 2022)

Hare and Poole found that the study of polarization in contemporary politics

measuring political actors' ideologies was contingent on each other (2014). The

DWNominate procedure is an effective estimation tool for the ideological scoring of

Senators and Representatives (Hare & Poole, 2014). Bringing the DW-Nominate tool

and referencing said Senators' and Representatives' Twitter accounts would help

determine if the polarization expressed in said Congressman's' tweets are comparable

to their voting behaviors (Voteview, 2022).

In the past, many politicians have often used hashtags on Twitter to indicate

keywords or topics associated with their tweets (Hemphill et al., 2016). Hemphill et al.

found that these hashtags are used in that they provide metadata and are practical tools

for organizing political discussions (2016). Hemphill et al. created the #Polarscores tool

to determine the political spectrum identity of Members of Congress by enumerating

their hashtag positioning, the act of using specific hashtags to identify a political position

in a tweet, and scaling it into a political identity score (Hemphill et al., 2016). #Polar

Scores found a strong correlation between Members of Congress's DW-Nominate score

and their #Polar Score (Hemphill et al., 2016).

Other tools include the Valence Aware Dictionary for Sentiment Reasoner

(VADER) tool, which uses "grammatical and syntactical rules" to detect language and

generate a positive or negative sentiment score (Hutto, 2023). The VADER sentiment

tool can measure responses and tweets using the coding language Python in real-time

(Elbagir & Yang, 2019; Hutto, 2023). Another tool for measuring sentiment is the Twitter

sentiment analysis tool; the tool can detect various human sentiments. Unfortunately,

there are some sentiments it struggles to detect, particularly humor (18% error rate) and

neutral, mistaken for sentiment (16% error rate) (Zimbra et al., 2018). Some of these

mistakes are also seen in the VADER tool; however, sentiment analysis often fails to

improve even with a human review (Elbagir & Yang, 2019; Hutto, 2023). Epistemic

Network Analysis can unpack extensive SA data sets into more meaningful groups by

splitting them into single-subject focus groups (Misiejuk et al., 2021). This study can test

SA's effectiveness in comparing the results of SA with nCoder results and DW-

NOMINATE scores (Lewis et al., 2023; Marquart et al., 2019).

nCoder, not unlike Twitter's SA tool, relies on artificial intelligence to help unpack

and code extensive datasets (Marquart et al., 2019). Unlike SA, nCoder requires

advanced coding by the researcher; for example, when creating a set of codes to

identify potentially harmful sentiments, the researcher would list as many possible

variables (a minimum of five keywords or phrases) reflecting the users' sentiments

(Marquart et al., 2019). Following this, the researcher can train the AI by manually

coding a training set instance for as many variables of their choosing (minimum of ten

and 80 to get validation; Marquart et al., 2019). The results of nCoder, while again rarely

perfect, will allow this study to analyze massive data sets with the hopes that, in

combination with ENA and VADER, the results will be reliable.

Significance of Social Media Data for Polarization Study

Facebook whistleblower Francis Haugen first made the world aware of using

algorithms to amplify contemptuous discourse on the platform (Whistleblower's SEC

Complaint, 2021). Since then, studies have been conducted to determine if Twitter's

algorithm has amplified specific politicians' and commentators' voices (Huszár et al.,

2022). Huszár et al. developed an algorithmic amplification model to measure tweets

from politicians from left and right-leaning groups and measured Twitter users'

responses (2022). The study found that the right across seven countries, including the

United States House and Senate, saw benefits from its increases in the amplification

versus the left, which had minor amounts in comparison (Huszár et al., 2022). The study

concluded that additional research was needed to complete if media sources and

politicians' use of Twitter were responsible for increased political extremism (Huszár et

al., 2022). This dissertation aims to continue this research and determine if the

responses generated by users on the tweets made by politicians and media sources

validate the rise of political extremism.

Understanding moral sentiment in social media and its connection to the rise of

political extremism is a rigorous task and has been modeled in various ways. The MAD

model (motivation, attention, and design) was developed to explain mortal contagion

online (Brady et al., 2022). The primary premise of the MAD model is that people "have

group-identity-base motivations to share moral content," triggering and capturing

audiences such that they continue to share and repeat (Brady et al., 2022). Social

networks are commonplace for over 3 billion users worldwide; disseminating political

and moral discourse by its users has created a demand for politicians, even at the local

level, to embrace it to win elections (Brady et al., 2022). A Pew Research Poll conducted

in 2016 found that nearly 84% of respondents believed that people post things on social

media that they would never say in person (Duggan & Smith, 2016). This is troubling as

research has indicated that social media is structured such that it is designed to amplify

attention to moral content, and in particular, algorithms favor highly engaged content

(Brady et al., 2022).

Human nature has played a part in social media's rise in moral and emotional

content, as humans are biologically and psychologically driven to engage with specific

content (Brady et al., 2022). This human behavior, along with the algorithms designed

by Twitter and Facebook to keep users engaged (spending more time means more ad

revenues), has led to an increased focus on group identity or echo chambers (Barberá

et al., 2015; Brady et al., 2022). One study found that while Twitter has a vast array of

interpersonal networks that are not always bound by ideological configurations

regarding political ideologies, users tend to be more apprehensive about engaging with

content from dissimilar sources (Barberá et al., 2015). The study and Bail et al. found

that liberal users tend to engage slightly more with opposing viewpoints, while

conservative users tend to fall into increasingly polarized viewpoints (Bail et al., 2018;

Barberá et al., 2015).

Estimating political viewpoints on social media while gauging ideological

placement was done using three steps. Using the Twitter REST API, Barberá et al.

found followers of Congress and political elites such as the President and Vice

President, then correlated their followers with other non-political figures but users who

were favored among liberal and conservative voters (Barberá et al., 2015). For example,

liberals matched users with following accounts such as The Huffington Post and

Stephen Colbert; and then matched users following Tea Party and Tucker Carlson with

having conservative ideology (Barberá et al., 2015). Once this was completed, Barberá

et al. (2015) validated the demographics by comparing the 113th U.S. Congress roll call

votes, which saw a correlation of r=.95. While estimating political ideology was far more

complex in years past, social media has seemingly made political preferences easier to

determine.

Much of this preferencing and ideological authentication has been made more

accessible by Facebook's algorithm, which tends to circulate content that validates and

reassures their confirmation bias (Roee, 2021). The field study by Levy (2016)

determined that Facebook's algorithm was less likely to deliver posts from

counterattitudinal outlets. A case study in Israel compared interactions between

Facebook, Twitter, and WhatsApp users and found that of all platforms, Twitter users

interacted more homophilic than the others and supported the case of enabling echo

chambers (Yarchi et al., 2021). A systematic literature review also concluded that social

media favored the emergence of polarizing echo chambers, which limited said users to

diverse information (Iandoli et al., 2021).

Echo chambers have tremendously shaped social media across all platforms as

users find homophily and refuge amongst their constituents (Cinelli et al., 2021). Cinelli

et al. (2021) compared how users consumed news on Facebook, Gab, Reddit, and

Twitter and gave scores to the users and the media sources to determine their

ideological preferences. The study found that Facebook and Twitter users were

particularly homophilic, and the algorithms tended to mimic their behaviors as users

were significantly more segregated than Gab and Reddit (Cinelli et al., 2021). When

comparing how specific news articles or posts were shared on Facebook and Reddit,

the study found that on Facebook news articles, final recipients tended to be exclusively

those of the seed users' leanings, indicating the presence of echo chambers, whereas

on it did not occur this way on Reddit (Cinelli et al., 2021).

One of the limitations of this study is in the review of just Twitter. Since the

January 6 uprising, new social media platforms have been released and created

specifically to enable greater free speech targeted at conservative communities (this is

according to their stated function; Fischer, 2022). Theoretically, these platforms will have

an increased level of pernicious and affective dialogue that is not currently allowed on

Twitter or was not allowed in the past, as the new owner of Twitter, Elon Musk, has

theoretically opened the platform to users like the former president (Elon Musk

[@elonmusk], 2022; Fischer, 2022).

Modern Theories on Political Polarization

Polarization: Echo Chamber vs. Partisan Sorting

One of the challenges many researchers have faced when scrutinizing social

media and polarization is whether social media is the cause of polarization. One study

using Dutch Panel data on social media and affective polarization found little correlation

that one caused the other (Nordbrandt, 2021). The challenge with the study is that the

data is from the Netherlands, which the author admits exhibits a lower level of

polarization than other countries (Nordbrandt, 2021; Reiljan, 2020). This theory lends

itself to Campbell's third theory of revealed polarization, such that polarization only now

seems to increase the masses' ability to articulate with the significant presence of social

media (J. Campbell, 2018). Nordbrandt (2021) and Campbell (2018) fail to account for

social media's ability to project outside users' homogenous groups and bring outside

their local bubbles, fueling polarization (Törnberg, 2022). Multiple studies concluded that

partisan sorting is causally linked to increased affective polarization, especially among

conservative Americans (Bail et al., 2018; Törnberg, 2022).

A survey conducted in an Ipsos panel posted in 2022 found that nearly 50% of

Americans believed there was a likelihood of a civil war in the next few years (Garen J.

Wintemute et al., 2022). This is cause for significant alarm following the uprising on

January 6, 2021. Americans are more willing to believe that the use of force will be

required to end what is likely the most significant level of polarization seen since the

Civil War (J. Campbell, 2018). Selective exposure, the assumption that users are

isolating themselves in echo chambers, has long been touted as the central hypothesis

for affective polarization; however, Törnberg (2022) reveals in his recent study that echo

chambers are merely an "intellectual cul-de-sac.” Additional research conducted by Bail

et al. (2018), Flaxman et al. (2016), and Dubois & Blank (2018) have also fostered this

result when looking at social media platforms, as it believes that studies on echo

chambers have neglected to identify social media's increasing exposure to crosspolitical

content. While these are all sound peer-reviewed studies with evidence, it is hard to

determine a proper methodology for approaching social media as there are many

complexities. Most studies on polarization find it hard to filter individuals with centrist

ideologies engaging with politicized content online (Flaxman et al., 2016).

Most of the centrist or moderate voter identification challenge stems from the

understanding that a massive spectrum of moderate voters exists (Drutman, 2019). A

FiveThirtyEight study examined YouGov America and Democracy Fund Voter Survey

data to break down moderate, undecided, and independent voters (Drutman, 2019).

The study revealed that even those who are ideologically moderate and independent

(only 2.4% of the electorate) are still wildly different in ideological values (Drutman,

2019).

The disappearing centrist voice in all media forms correlates to the general

population's rapid polarization of the center (Drutman, 2019). Opinion dynamics

formulated by social media algorithms and attitude polarization push media consumers

into wildly extreme positions (Jones et al., 2022). Jones et al. model found that rationally

behaving major-party candidates will gain from supporting highly polarized platforms

more so than pivoting to the center (2022). Upon examining highly polarized Congress

members like Margorie Taylor Green R-GA, and Josh Hawley R-IL, they have leveraged

extremist positions to furth their base and maintain a significant presence in Congress

with little centrist positioning.

Multiple forms of segregation have existed and continue to exist since the

foundation of the United States in the 18th century. The continued segregation of human

social groups has led to numerous negative consequences that are mainly responsible

for the massive levels of 180 million sorted votes in the United States (Brown & Enos,

2021). Political party affiliation has become a critical social identity despite little impact

on one's day-to-day life (Green et al., 2004). Brown & Enos attribute these increases in

partisan sorting to isolated partisan environments that lack interpersonal contact (2021).

Exposure to cross-political viewpoints is reasonably low in the United States, with the

average Democrat seeing Republicans at just .30 and converse at .36 (Brown & Enos,

2021). Many voters live in extreme isolation, with 10% of Democrats having virtually no

exposure to Republicans (Brown & Enos, 2021). These low exposure levels are vital in

understanding how subsequent affective polarization occurs in the American public

(Luttig, 2018).

Partisan sorting and affective polarization in America are seemingly occurring

together, and Americans' need for closure drives part of this partisan divide (Luttig,

2018). The closure theory drives the "us" versus "them" mentality as individuals find it

increasingly more challenging to connect with partisan outgroups (Luttig, 2018). Brown

& Enos identify the part of the problem as the lack of exposure and segregation that

leads to prejudiced personalities and inevitably pernicious polarization (Brown & Enos,

2021; Luttig, 2018; McCoy & Somer, 2019).

Some have argued that mainstream media are not responsible for the forces

driving the current polarization (Garen J. Wintemute et al., 2022). However, the rapid

rise of news and media consumption via social media has driven primary mainstream

news sources like Fox News, CNN, and MSNBC to take more partisan positions.

Psychology of Polarization

There are many beliefs and theories about the psychological causes of

polarization; one particular social psychologist, Dr. Bibb Latané of Florida Atlantic

University, pioneered group dynamics research in the 1980s and 90s (Latané, 1981,

1996). In the social impact theory, Latané (1981) suggested that the influence of one

person is directly related and multiplied by the strength, immediacy, and number of

individuals engaging in the target (person). By 1996, social impact theory expanded to

dynamic social impact theory (DSIT), which added the tendency for individuals to be

influenced by those nearby, with immediacy and high redundancy, thus accelerating the

social impact (Latané, 1996). Social media created the perfect platform for the explosion

of highly influential subcultures with similar values and ideologies, as they could quickly

form echo chambers and like-minded users willing to share similar ideologies (Latané,

1996). Social media has been the catalyst for many subcultures and groups like

QANON, Proud Boys, and Oath Keepers; these communities formed online as

individuals continued to feed their needs for validation and securing social standing in

like-minded communities with strong leaders (Harton et al., 2022).

While DSIT is a relatively new theory in which social media and the creation of

echo chambers sufficiently provide evidential examples to support it, human nature and

behaviors have always seen this type of radicalization, enabling it to occur much faster.

The allegory of Hitler's rise to power in Germany is a story told by many but understood

by few regarding the consequences of media and access to information. The inclusion

of political programming on the radio in Germany didn't start until 1929; subsequently, it

became increasingly politicalized (Adena et al., 2015). While the Weimar Republic was

effective in slowing Nazi growth by controlling the radio news programs in 1929, by

1933, Nazis took control over radio programming and were able to convince many

Germans to engage in antisemitic behavior and join the Nazi party (Adena et al., 2015).

The speed in which pro-Nazi messages were distributed on the radio was unlike any

other forms of media seen in previous human history; it allowed for less scrutiny and

opened the doors for newspapers to follow suit furthering the DSIT model of social

impact (Adena et al., 2015; Latané, 1996).

The rapid adoption of social media platforms has allowed for little to no regulation

and oversight leaving the doors wide open for bad actors to influence those with

misinformation at incredible speeds. This is referred to as a "post-truth" world by many

media members and political pundits. Twitter enabled former President Donald Trump to

leverage DSIT because of Twitter's simplicity, impulsivity, and incivility (Ott, 2017). The

original 140-character limit of Twitter (it doubled during Trump's tenure in office and now

is 10,000) encouraged simplicity which meant that context is often lost, and users are

left with the ability to misinterpret information however they see fit (Ott, 2017;

Weatherbed, 2023). Former President Trump leaned hard and successfully into his

ability to create ambiguity and stir up the media and interest with his tweets as people

craved the potential for being viral.

Twitter's ability to create impulsivity follows this desire to become a viral social

media influencer (Ott, 2017). The ease of posting on Twitter or sharing discourages

selfreflection and encourages users to act quickly for the best quote-tweet or tweet

regarding an incident. A study on disseminating fake and real news on Twitter concluded

that fake news spread faster and more effectively because humans were sharing the

news, not bots (Vosoughi et al., 2018). The human desire to engage in selfpromotion

inadvertently relies on humans' choices to act negatively toward those they see as

threatening (Ott, 2017). This use of incivility for self-promotion is one that former

President Donald Trump and other polarizing politicians have leveraged frequently on

Twitter (Ott, 2017).

Former President Donald Trump mastered Twitter in many ways by capitalizing

on Twitter's strengths and flaws. He was able to create communities to spread his

messages through DSIT and create easily repeatable messages like "Stop the Steal,"

Build the Wall," and "Make America Great Again" (Harton et al., 2022). According to the

Washington Post Fact Checker team, on the day before the election on Nov. 2, 2020,

then-President Donald Trump made 503 false or misleading claims regarding election

interference; over his four years, this number is 30,573 about 21 false claims a day

(Kessler et al., 2021). Perhaps more disturbing is that significant media publisher Fox

News validated these beliefs, even though they knew these claims were false (Levine &

Lerner, 2023). To the devotees of some Trump followers, the only truth is the one he

created (Ott, 2017).

The Trump base, however, is not necessarily to blame for their blind faith in his

seemingly endless stream of false claims. Before the 2016 election, affective

polarization and ideological contempt had been engrained in both the Republican and

Democratic parties (Geiger, 2016). Former President Trump offered a completely

different narrative than the existing aristocracy and gave a small-disenfranchised group

hope following the economic struggle and long recovery of the 2008 crash (S. Simon,

2021). Using simple language coupled with emotionally and morally charged tweets,

Trump quickly captured the attention of the previously disenfranchised (Ott, 2017).

The challenge media faces today is that often the news isn’t as emotionally triggering

as the fake news, and as a result, it gets shared less (Brady et al., 2017). This

phenomenon is called "moral contagion," in which the spread of moral ideals is shared

more frequently and with higher engagement (Brady et al., 2017). Brady et al. 2022

explain moral contagion with their psychological model, MAD (motivation, attention, and

design). The MAD model exemplifies how individuals are more likely to share

“moralemotional content” which is fueled by group identities because of their likelihood

to garner responses (Brady et al, 2022).

Former President Trump’s tweets exemplify the MAD model at the core; even

when he deleted the infamous "covfefe" tweet, the media responded with questions over

his mental acuity, and rather than getting defensive, he leaned in with the subsequent

tweet, "Despite the negative press covfefe" (Estepa, 2018). Trump continued to engage

his base and feed his existing narratives without hesitation and did so successfully by

constantly invalidating his opponents, the out-groups (Harton et al., 2022). Trump’s

consistent use of inflammatory and frankly racist language when talking about the

border issues between the United States and Mexico was also an effective example of

the MAD model (Brady et al. 2022)

This polarizing strength of former President Trump waylays into social media,

specifically Facebook and Twitter, favoritism towards posts promoting "out-group

animosity" (Rathje et al., 2021). Rathje et al. (2021) found that the likelihood of a Tweet

or a Post going viral increased by 4.8 times when using out-group terminology over

negative affect language and 6.7 times more than moral-emotional language. Out-group

language is the number one predicting political and media account-sharing behavior

(Rathje et al., 2021a. It can be assumed that the success of highly polarized politicians

on Twitter, like Rep. Greene (R-GA), Sen. Hawley (R-IL), and Senator Warren (D-WA),

is from their leveraging of this out-group animosity phenomenon in their tweets.

Elite Polarization in the 21st Century

While there is still much room for debate, ultimately, this dissertation seeks to

demonstrate how elite polarization continues to drive mass/group polarization in

America. Social media, not unlike mass media in the 1930s in Germany, is acting as a

catalyst for segregated groups to form partisan echo chambers fueled by pernicious

polarization. While there have undoubtedly been minority groups pushing for highly

ideological party formation, the super-minority's magnification due to Twitter's nature is

giving these outliers much more power (Banda & Cluverius, 2018; Rathje et al., 2021).

In 2001 it was common for political scholars to state that American political parties were

weak and in decline (Hetherington, 2001). However, in the 1980s, there was a shift in

the political elite (members of Congress specifically) as partisan voting became

increasingly more common (Hetherington, 2001). By the mid-1990s, voters began voting

for a party based on personal ideology more than ever (Hetherington, 2001). Another

Pew Research Center Analysis study using DW-NOMINATE found that Congressional

polarization was at an all-time high as the divide between Democrats and Republicans

had increased significantly (Desilver, 2022). Figure 3 shows average political ideology

scores for members of the Democratic and Republican parties using

DW-NOMINATE in ten-year intervals from the 92nd to 117th Congress (Desilver, 2022).

Much of this divergence from centrist politics is due to the loss of moderate-to-liberal

Republicans and moderate-Conservative-Democrats as both have vanished from

Congress (Desilver, 2022).

Figure 3

Republicans Have Moved Further to the Right than Democrats to the Left

Note. Reprinted from “Republicans have moved further to the right than Democrats

have to the left.” Pew Research Center, Washington, D.C. (9 March 2022)

https://www.pewresearch.org/ft_22-02-22_congresspolarization_featured_new/.

Reprinted with permission

While there is much debate over the causes of elite polarization in Congress,

there is certainly no doubt in its existence. Some political theorists speculate that the

affective polarization in echo chambers drives elite polarization (Diermeier & Li, 2019).

These theorize that voters are becoming increasingly more responsive to ingroup

deviations than out-group deviations, i.e., politicians fear partisan reprisal for going

against party ideology (Diermeier & Li, 2019). Former Rep. Liz Cheney (R-WY) and

former Rep. Adam Kinzinger (R-IL) experienced this partisan reprisal firsthand as both

were chastised by other Republicans for their work in the January 6 committee and lost

re-election in their districts (Harknik, 2022). There is little research, however,

determining if affective polarization is driving elite polarization or to the contrary, it is

hard to create a model in which the choices of politicians can be measured such that it

reflects decision-making on their values, party values, or the values of the affected

masses (Banda & Cluverius, 2018).

Banda & Cluverius (2018) argue that elite polarization drives affective polarization

such that the political elite leverage out-group leadership and takeover fear to rally their

base. Social identity threat is ingrained in American psychology, and when an individual

is categorized as a group member, this can be perceived as a threatening experience

(Branscombe et al., 1999). The fear of miscategorization to an out-group based on an

ideological value can be highly deterministic in creating voting behavior (Banda &

Cluverius, 2018; Branscombe et al., 1999). This experience is leveraged by pernicious

polarization as cross-cutting ties between elites become systematic conduct of the

political elite (McCoy & Somer, 2019). The challenge is that there is no constructive

element in this type of us vs. them political bargaining (McCoy & Somer,

2019). The Republican party experienced this self-destructive behavior as

Congressmen Kevin McCarthy (R-CA) battled with the far-right conservatives to secure

his position as House Speaker (Greve & Gambino, 2023). Representative McCarthy

(RCA) failed fifteen times before making several concessions which inevitably

weakened his status as the leader of his party (Greve & Gambino, 2023).

While the Republican party has shifted further right than the Democrats have to

the left, the reality is that both parties have put tremendous focus on identity politics in

their campaigns (Desilver, 2022). One study found that as elite-level polarization

increased among the Democratic party, those further on the left became increasingly

polarized (Zingher & Flynn, 2018). Republican elites did not experience the same

increase in polarization with their more conservative base as the Democrats saw with

their extreme liberal base (Zingher & Flynn, 2018). However, regardless of ideology,

both parties saw increases in mass polarization as their elite became increasingly more

ideological (Zingher & Flynn, 2018). Crossover voting is becoming extremely rare, as

noted by shifts in presidential campaigns focused on core supporters rather than

independent voters (J. E. Campbell, 2005; Zingher & Flynn, 2018). The 2004

Presidential Election marked a significant election in that NES election data revealed

that 56% of votes had determined whom they would vote for before the national

conventions (J. E. Campbell, 2005, p. 227). Democrats stood solidly behind John Kerry

because he was not former President G.W. Bush; however, this was not enough as

states who had previously voted for Bush in 2000 saw a significant uptake in voter

turnout (J. E. Campbell, 2005, p. 237). This behavioral shift signified the start of the

extreme polarization that followed in subsequent elections as ideological voting

behaviors of the masses began to take place and the fight for the disappearing

independent voters diminished (J. E. Campbell, 2005; Zingher & Flynn, 2018).

The general shift towards affective and pernicious polarization has sufficiently

eliminated much of the moderate voter category in America (A. I. Abramowitz, 2010;

Drutman, 2019). In the 2008 presidential election, presidential nominees in both the

Democratic and Republican parties took hardline positions that reflected the party views

of their majority (A. Abramowitz, 2008). This partisan-ideological polarization among the

political elite, especially in the case of presidential candidates, brought these partisan

beliefs front and center to the American voter (A. I. Abramowitz, 2010, p. 36). Contrary

to this argument, a few have argued that these divisions have existed since the 1950s

and that the center is alive and well hence the existence of mixed blue and red states

(Republican governors voting blue in a presidential election; Fiorina et al., 2008).

Fiorina et al. (2008) frequently met with a rebuttal as their views are highly

idealist and grant significant credit to American voters' knowledge of the government

system (A. I. Abramowitz, 2010, p. 36). A survey poll conducted with Penn State

University after the 2018 midterm congressional elections suggested that Republican

voters had reasonably strong views regarding the intentions of the average Democratic

voter (Plutzer & Berkman, 2018). Only one in four Republican voters surveyed believed

that Democratic voters had the country's best interest in their mind when voting (Plutzer

& Berkman, 2018). The other parties' lack of social trust is furthering affective

polarization among American votes (Lee, 2022). Social trust and perception of others

following suit in the belief that other Americans will follow through for the public's good

was tested and failed drastically during the COVID-19 epidemic (Lee, 2022). Lee's

(2022) survey found that the higher levels of perceived polarization led to a significant

reduction in social trust cross-party lines. The amplification of trust issues among the

political elite only leads to greater affective polarization in America (Theiss-Morse et al.,

2015).

McCoy et al. (2018) found that elite polarization significantly created and

intensified divisive pernicious rhetoric in Hungary, Turkey, the United States, and

Venezuela. The study found that the elites effectively targeted societal cleavages to

push their base to a position of distrust and animosity (McCoyet al., 2018). Elites used

messages instilling fear of cross-political ideologies in their base to garner political

support and funding (Iyengar & Westwood, 2015). Former President Obama signified

the begging of partisan cues entering the daily lives of Americans as individuals became

increasingly more determined to express their political identities (Iyengar & Westwood,

2015). This increasing behavior of expressing partisanship, coupled with the ability to

disseminate information rapidly, accelerates the polarization created by social media.

Using Epistemic Network Analysis to Evaluate Twitter

Several studies cited in this paper have successfully used epistemic network

analysis (ENA) to provide quantitative and qualitative insights into political rhetoric used

on Twitter (Hamilton & Hobbs, 2021; Misiejuk et al., 2021). ENA identifies and quantifies

connections of critical themes and sentiments in coded tweets by creating dynamic

network models that generate and illustrate qualitative connections through quantitative

summary statistics (D. W. Shaffer et al., 2016). The challenges of traditional multivariate

statistical models for social media analysis lie in the inability to indefinity networks and

patterns in massive data sets like Twitter feeds (D. W. Shaffer et al., 2016). ENA allows

this study to identify key phrases to model the rhetoric used by those engaging in highly

charged political feeds. As mentioned early in the chapter, the study conducted by

Misiejuk et al. (2021) incorporated SA data into ENA by coding sentiments of Twitter

feeds of both Democratic and Republican users regarding the COVID-19 epidemic.

Figure 4 below shows the networks created by adding SA data into their coding scheme

and developing a second network; they could identify sentiments expressed by

individuals in a particular party and a positive and negative correlation (Misiejuk et al.,

2021). The thickness of the line determines the strength of the connection in ENA; in

Figure 4 below, the correlation between stimulus action and favorable among

Democrats was a strong correlation which is not surprising considering the media's

representation of Democratic support for COVID-19 stimulus relief packages (Misiejuk

et al., 2021).

Figure 4:

Incorporation Sentiment Analysis into ENA

Note. Obtained from “Incorporating Sentiment Analysis with Epistemic Network

Analysis” by K. Misiejuk, J. Scianna, R. Kaliisa, K Vachuska, D.W. Schaffer, A. Ruis, and

Reprinted with Permission.

A systematic literature review of social media sentiment analysis studies found

that, in large part, most of the studies conducted their research using the opinionlexicon

analysis method (Drus & Khalid, 2019). Since the data set this study will use are far too

large for human coding, the study will rely instead on the machine-based learning

models of the VADER sentiment tool and nCoder (Hutto, 2023; Marquart et al., 2019).

The review found that while the opinion-lexicon model was suitable for small data sets,

machine-based learning with proper training was just as accurate (Drus & Khalid, 2019).

One of the studies reviewed found that using both machine-based learning and onion-

lexicon sentiment data, Twitter feeds 43 days before the 2016 presidential election were

just as strong if not stronger predictors than polling data (B. Joyce & J.

Deng, 2017). While opinion-lexicon outperformed in smaller datasets, the Naive Bayes

Machine Learning Algorithm achieved a higher correlation coefficient the more extensive

the data set (B. Joyce & J. Deng, 2017).

Chapter 3: Methods

Designing a study using ENA to Discourse on Twitter

While there have been numerous studies on human discourse on Twitter, few

have used ENA to design qualitative and quantitative studies to gain a clearer insight

into the outcomes of specific types of rhetoric (Garson, 2020/2023). While Twitter's

sentiment analysis (SA) tool has been criticized as being inaccurate, one study found

that looking at SA results in ENA gave more meaningful insights when comparing SA

results individually (Misiejuk et al., 2021). In May 2023, while this study was being

conducted, Elon Musk eliminated academic access to the Twitter API and required

costly packages to access legacy datasets (Calma, 2023). Hence this study shifted

away from the Twitter SA tool and took on the VADER sentiment tool, which performs

arguably better (Elbagir & Yang, 2019). While the VADER is one aspect of this study, the

primary research questions are to determine if the specific rhetoric being tweeted is

polarizing. Determining this is challenging as the study will have to assume that the

respondents to specific tweets convey their intentions as polarizing or neutral. Like any

study, there needs to be some philosophical assumption in the research (Creswell &

Poth, 2016). Specifically, this is going to be the belief that ENA will create a network of

responses using the comments on each tweet that indicate both the political identity of

the user as well as their intention with the tweet.

The first research question: RQ1: Do the political elite (Members of Congress)

leverage Twitter to promote identity politics furthering political polarization in America?

Various studies have shown that the political elite, celebrities, and influencers have been

some of the most effective at spreading COVID-19 misinformation (Gisondi et al., 2022;

F. Simon et al., 2020). In their study, Simon et al. (2020) found that the political elite,

celebrities, and prominent public figures (influencers) generated around 20% of the

misinformation on social media but garnered nearly 69% of the total media engagement.

This study led to the focus of this dissertation on the political elite, as they have a

tremendous influence on Twitter users.

Following Question 1, a deeper understanding of how followers and Twitter users

are responding to the political elite will provide insight into the effectiveness and

perceived nature of the elite user's tweets. This dissertation hypothesizes that the

political elites leverage the spread of disinformation and polarizing rhetoric to further

drive cross-political animosity and, in turn strengthen their position within their base.

RQ2: Is Twitter's algorithm giving preferential status to the political elite who use

polarizing tweets to generate higher user engagement resulting in higher ad revenue?

The second research question is more challenging to study as the nature of Twitter's

algorithm is published; however, determining if users are becoming polarized is more

difficult to prove. This research aims to determine who is responsible for the divisions

created on social media. Once a satisfactory outcome of RQ2 is determined, the

subsequent question of blame will be revealed. Is the political elite tweeting or Twitter

driving the spread of the information?

Since Elon Musk's takeover of Twitter, he has pledged to make the company

even more transparent and revealed in a blog post how the recommendation algorithm

works (Twitter, 2023b). According to Twitter (2023b), the recommendation algorithm

generates a user's feed through a complex system of scoring the user's tweets and

matching them to similar tweets that the user might find engaging. One of these intricate

methods of determining these groups is a space called SimClusters, a cluster of

influential users that generally see users interacting within (Satuluri et al., 2020; Twitter,

2023b). It is hard to differentiate between SimClusters by Twitter's definition and echo

chambers; however, the original recommendation made by Twitter is being made off the

foundation of the user's first follows or tweet likes such that they are the building blocks

for their feed.

Every story has a who, what, where, when, why, and how. RQ1 seeks to answer

the who and what by looking at what the political elite are tweeting. RQ2 seeks to

answer the where and when by looking at users' engagements on Twitter and

determining if there are corresponding behaviors to specific types of tweeters (are the

loudest/craziest voices being heard the most?) RQ3 (below) seeks to answer why this

study matters and how it impacts the world.

RQ3: Were President Donald Trump's (@realDonaldTrump) and President Jair

Bolsonaro of Brazil's (@jairbolsonaro) tweets leading up to the uprising of January 6,

2021, and January 8th, 2023, respectively, responsible for the unfortunate and

subsequent events?

While Trump was banned from Twitter following his potentially triggering tweets

leading up to January 6, Elon Musk restored his account shortly after his takeover at

Twitter (Elon Musk [@elonmusk], 2022). This study will look at specific tweets from

@realDonaldTrump and determine if those who illegally entered the capitol on January

6 liked and commented on any of those tweets. Again, using the ENA web tool, the

study will model the language with nCoder and VADER sentiment analysis on the

specific tweets to determine if the former presidents could be linked to inciting

subsequent insurrections.

Following several hearings in Congress and a few court cases, many of those

defending their actions to commit treason cited the former President's tweet on

December 19, 2020, cited below, for their actions (Dreisbach, 2022).

Peter Navarro releases 36-page report alleging election fraud 'more than

sufficient' to sing victory to Trump…. A great report by Peter. Statistically

impossible to have lost the 2020 Election. Big protest in D.C. on January 6th. Be

there, will be wild! (Trump, 2020)

"Will be wild" was the triggering phrase for many of the rioters who believed that their

actions were being condoned by former President Trump (Dreisbach, 2022; Trump,

2020). While many Americans believed former President Trump's actions were

antidemocratic, nearly 70% of Republican voters believed that Trump was trying to

defend democracy (Malloy & Schwartz, 2021). The outcomes of the insurrection on

January 6 are still lingering, and whether or not democracy is still at risk is debated as

no politicians have been held accountable nearly two and half years later (S. Simon,

2021). Not unlike former President Donald Trump's tweet on December 19, former

President tweeted on the night of the election, October 30, 2022, a bible quote

translated with Google Translate read:

"Put on the whole armor of God so that you can stand firm against the Devil's

wiles, for our fight is not against humans, but against the powers and authorities,

against the rulers of this world of darkness...

Ephesians 6:11-12

- MAY GOD BLESS OUR BELOVED BRAZIL! (Bolsonaro, 2022; Google

Translate, n.d.)

While the context of this tweet is subjective in nature, journalistic publications and one

study have found that former President Bolsonaro's tweets and denial of the election

results could have led to the uprising on January 8, 2023 (Bugs et al., 2023; Dwoskin,

2023). This study aims to provide insight into determining the accountability of the

political elite who engaged with those who were inside the capitol illegally on January 6,

2021, and for those who incited insurrection in Praça dos Três Poderes, Brazil on

January 8th, 2023.

Research methodological approach and Study Design

ENA Theory

The following three research questions rely on Epistemic Network Analysis

(ENA), a technique for modeling the structure of connections in data. ENA assumes: (1)

that it is possible to systematically identify a set of meaningful features in the data (i.e.,

tweets); (2) that the data has local structure, i.e., the constituents grouped by their

ideological identity; and (3) that an essential feature of the data is the way that Codes

are connected within conversations (Bowman et al., 2021; D. Shaffer & Ruis, 2017; D.

W. Shaffer, 2018; D. W. Shaffer et al., 2016)

ENA models the connections among Codes by quantifying the co-occurrence of

Codes within conversations, producing a weighted network of co-occurrences, along

with associated visualizations for each unit of analysis in the data. Critically, ENA

analyzes all the networks simultaneously, resulting in a set of networks that can be

compared both visually and statistically.

While ENA was initially designed to address challenges in learning analytics (D.

Shaffer et al., 2009), the method is not limited to analyses of learning data. For example,

ENA has been used to analyze (a) task performance (Brückner et al., 2020; D’Angelo et

al., 2020); (b) gaze patterns (Andrist et al., 2015; Brückner et al., 2020;

D’Angelo et al., 2020); (c) team communication (Sullivan et al., 2018; Wooldridge et al.,

2018); (d) governmental communication and policy (Schnaider et al., 2021) and social

media (Dubovi & Tabak, 2021; Misiejuk et al., 2021) The critical assumption of the method

is that the structure of connections in the data is meaningful. Thus, ENA is a valuable

technique for modeling polarization in social media because it can model the relationships

among Members of Congress’ rhetoric on Twitter as they occur among their constituents.

Research Question 1: Research Design and Data Gathering

Elite polarization has had a tremendous impact on American politics as it

becomes more evident that parties are becoming increasingly ideologically driven

(Banda & Cluverius, 2018; K. T. Poole, 2007). Several studies have identified levels of

elite polarization on Twitter and concluded that politicians were using social media to

share misinformation to garner additional support for their party (Banda & Cluverius,

2018; M. Otala et al., 2021). RQ1 uses ENA to compare the polarization levels

determined by the nCoder of members of Congress with DW-NOMINATE (Lewis et al.,

2023; Marquart et al., 2019). Using DW-Nominate to help identify members of Congress

to determine political leanings will determine in their tweeting behavior matches their

voting record seen on Voteview.com (Lewis et al., 2023). DW-Nominate was created to

develop a spatial model of roll call votes in Congress to determine if individual members

of Congress were more or less liberal or conservative based on the measure or bill

passed (K. T. Poole & Rosenthal, 1985). Lewis et al. (2023) use DW-NOMINATE and

post results for each roll call vote in Congress, ultimately generating an ideology score

based on their liberal to conservative spectrum (Lewis et al., 2023). Voteview.com

(2023) allows for rankings of both houses of Congress to be downloaded into a CSV file

where each candidate was identified by their ranking as either Far Left, Median Left,

Center Left, Center Right, Median Right, and Far Right. With these assigned rankings,

ten MoCs from each category were selected from both houses, such that 120 members

of Congress were selected for their data to be collected from Twitter. To develop

consistent networks, both Independent Senators Bernie Sanders and Senator Angus

King will be categorized as Democrats as they caucus with the Democrats and

ideologically are included as Democrats in DW-NOMINATE scores on Voteview.com

(Lewis et al., 2023).

While Twitter data was accessible using the Twitter API, which is free to use and

allows Twitter users to make calls to Twitter’s data sets, request data, and even post

data for a small fee if the number of posts exceeds (Hutto, 2023). Elon Musk removed

API access for academics in May 2023, which forced the study to rely on a web

scraping tool called TwExportly instead that allowed for a maximum number of 1,000

tweets to be downloaded instead (TWExportly, 2023).

Using TwExportly, tweets from each member of 120 Members of Congress (MoC)

were downloaded, and the most recent 200 up to July 28, 2023, for each member

studied will be studied (this study will look only at data from the 118th Congress). The

data collected will include the author (MoC), the 160-character tweet and any

corresponding URL or retweet, the number of likes, the number of retweets,

corresponding hashtags, and the number of comments made on each tweet. Once the

data is collected, using the VADER sentiment analysis tool, each tweet will be scored as

positive, negative, or neutral (1 to -1 scale; (Hutto, 2023).

Once the tweets were collected, the study used nCoder, a tool created to codify

massive datasets using machine learning, to code the tweets into the codes in Table 1

below (Marquart et al., 2019). The below coding chart is the intended tool with specific

examples from members of Congress, with the Senate with the highest and lowest

scores and the most moderate Republican and Democrat (Lewis et al., 2023).

Table 1

Codes used in RQ1

Left Polarizing

Neutral/Unifying

Right Polarizing

Check Supreme Court

Supreme court ethics

Gun Violence

Judiciary Act

Tax-dodging/Wealthy

Cheating

Reproductive

health/Body’s Choice

Striking down

MAGA

Weapons Ban

Gun Violence

Improve/Improving

Innovate/Innovation

Assistance/Working With

Train/Trained

Honor

Applaud

Bi-partisan

Together/Teamed

Congrats/Celebrate

Happy

Opportunities

Patriot

Radical Left

Mainstream Media

Bidenomics/ Biden

Administration/ Joe Biden

Southern Border

Borders/Border

God/GodBless

Hunter Biden

National Security

American People

Table 2

Sample Coding in RQ1

Member of

Senate

DW-

NOMINATE

Score

Latest Tweet (as of May

17, 2023)

Sentiment

Analysis

nCoder

Polarization

Sentiments

Thomas

Hawley

Tuberville

@TTuberville

.936 (Most

Conservative

member of

Senate)

I’ve said it before and

I’ll ALWAYS say it: I’m

100% ALL-IN for

President Trump. He’s

the one to get us over

the goal line and SAVE

AMERICA from the

radical left. Stand with

President Trump today

https://bit.ly/3MnLfph

Positive

Pro-Trump

Save

America

Radical Left

Susan

Margaret

Collins

.116 (Most

Liberal

member of

the

Republican

senators)

It was an honor to

accept the Edward M.

Kennedy National

Service Lifetime

Leadership Award.

I was delighted to be

joined by so many

champions of

AmeriCorps, who

selflessly dedicate their

time and efforts to

improving the world

around them.

Positive

Unifying

Improving

Member of

Senate

DW-

NOMINATE

Score

Latest Tweet (as of May

17, 2023)

Sentiment

Analysis

nCoder

Polarization

Sentiments

Joe Manchin

III

-.06 (Most

Conservative

member of

Democratic

Senators)

GOOD NEWS:

Improving and

modernizing our roads,

bridges and highways

continues to be one of

my top priorities, and

I’m pleased the

@USDOTFHWA

is investing more than

$7.1 MILLION in

repairing roads in West

Positive

Improving

Unifying

Virginia damaged by

severe flooding.

Elizabeth

Warren

-.751 (Most

Liberal

member of

Senate)

The last time Silicon

Valley Bank’s CEO

testified to the Senate

Banking Committee, he

was lobbying for looser

regulations.

Today, he came back to

talk about how his bank

had failed—under

weaker oversight.

I’m fighting to put strong

protections in place and

prevent more crashes.

Negative

Failure

Oversight

Improving

Once the data is compiled, it will be loaded into the ENA web tool. Each data set

from each house of Congress can be evaluated in terms of their DW-NOMINATE score,

sentiment analysis, and corresponding polarization codes. If H1a is correct, using the

corresponding SA scores with the matched tweet sentiment codes will result in a

corresponding score that should be similar to their distance from 0 on DW-NOMINATE

scores (Garson, 2020; Lewis et al., 2023; Misiejuk et al., 2021). This will validate

whether the study correctly identifies whether the tweets are polarizing.

Once the sentiment scores were reviewed, the data was uploaded into the

nCoder web tool to be coded into the following codes in Table 1. The challenge with

nCoder is getting a meaningful result due to the potential overlap of specific terms and

the inability of machine learning to detect sarcasm and identify all the potential

keywords that might be used for each category. Many of the codes could also be

perceived as both left and right polarizing. However, after coding, narrowing down

keywords for each group, and training the data sets, our Left Polarizing, Right

Polarizing, and Neutral/Unifying codes achieved reliable Kappa and precision (Hallgren,

2012). For Right polarizing, a Kappa of .80 and a precision of .87 were reached,

indicating substantial agreement between the test sets created by the author and the

machine learning algorithms of the n-Coder web tool (Hallgren, 2012; Marquart et al.,

2019). For our Left polarizing, a Kappa of .89 and precision of .88 was reached and

indicating an almost perfect agreement and precision, along with a Neutral/Unifying

code reaching a Kappa of .94 and a precision of .82, suggesting again almost perfect

agreement and precision (Hallgren, 2012; Marquart et al., 2019).

Research Question 2: Research Design and Data Gathering

Once the outliers are determined in research RQ1, RQ2 will look at the levels of

engagement of the tweets posted by the members of Congress in RQ1. By adding a

ranking of tweets by likes and shares, the virality of each MoC tweet can be measured.

In this instance, MoCs were ranked using a quartile system (Highest Engagement,

Above Median, Below Median, and Lowest Engagement), where all the tweets' likes

were averaged for each MoC, then ranked by their results.

Using the same categorizations for RQ1, the DW-NOMINATE classifications for polarity

groupings, Far Right to Far Left, will be used to determine if the increased levels of

polarization reflect the level of engagement received (Lewis et al., 2023). I will also

compare (if they should vary from RQ1s results) the most polarized MoCs tweets with a

significant follower count. Responses to each of the selected tweets will also be coded

using VADER and nCoder to determine their stance affirming their desire for

crosscutting (pro-polarization) or invalidation of polarization (Hutto, 2023; Marquart et

al., 2019). To gain further insight into potential polarization themes and understand why

specific tweets gain virality over others, a more in-depth coding system was used to

determine linking themes behind each MoC polarizing or non-polarizing tweet. The

categories were Government, Social, and Economic. Government codes were when an

MoC blamed the other side, a particular policy, or a branch of government for some

fault. Social reflected attacks on social policies or religious themes alluding to isolating

and polarizing behaviors. Lastly, economics pointed to economic policies or bills

targeted as being divisive by MoC.

Once a sample of the tweets is taken from the user, and set, the codes (a partial

list is shown in Table 3 below) to be used in nCoder will be programmed. Following this,

Table 4 is an example of the coding implementation used in the study, followed by a

sample data set with like share counts.

Table 3

Codes used in RQ2

Left – Economic

Neutral/Unifying –

Economic

Right – Economic

clean energy, student

debt, tax wealthy,

working families, fossil

fuel, labor pensions,

rigged tax, social

security for all, special

interests, carbon-free

energy, extreme heat,

Clean Air Act, paid sick

days, tax cuts, blue

economy, rich,

Inflation reduction act,

innovation, innovate,

Economy works for all,

Economy works,

Small businesses,

Bipartisan Digital, upgrade

infrastructure, Trade Award,

economic growth, made in

America,

BipartisanInfrastructureLaw

, strengthen our economy,

Debt ceiling, Small

businesses, American

people, Inflation

reduction act, Big tech,

Debt ceiling deal, Right

to work, National right

work, Right to work act

Farm bill, Bidenomics,

deep faith, ban

Obamacare, taxpayer

dollars,

Left – Social

Neutral/Unifying –

Social

Right – Social

reproductive health care,

health care for all, mental

health, mental health

crisis, health crisis, health

education, women’s

rights, trans rights, sexual

identity, maternal

mortality, maternal health

crisis, weapons ban,

assault riffle ban, red flag,

reproductive health,

reproductive rights,

PACT Act, condemn hate,

honor, American heroes,

Educational Partnership,

mental healthcare, public

health, Cost-of-Living

Adjustment Act of 2023,

police reform, social

security for all,

Child sex trafficking,

Protect Children’s

innocence, Southern

border, religious liberty,

Trafficked, bypass

parents, sanctity of life,

abortion is murder, First

Amendment, second

amendment, girls safe,

censure, God bless,

conception, illegal

immigration, violate young

Left – Government

Neutral/Unifying –

Government

Right – Government

extreme maga

republicans, extremist

right, radical right,

extremist right, extremist

maga, supreme court

justices, supreme court

ethics, senate judiciary

committee, maga

republicans, conservative

court, Trump

Bipartisan infrastructure,

working together, National

security, Bipartisan bill,

RESTRICT Act, TikTok,

bipartisan infrastructure

law, Bipartisan Digital,

brave men women, both

sides, support veterans

Blame Biden, Biden crime

family, Million-dollar bribe,

Bill protects children’s,

left-wing extremists, DOJ

IS WEAPONIZED, FBI,

Hunter Biden, executive

overreach, Biden

Administration, tyrannical

FBI, DOJ is corrupt, FBI is

corrupt, FBI leadership

Table 4

Sample Coding in RQ2

User

Likes

DW-NOM

Sentiment

Codes

@RepMTG

Rep.

Margorie

Taylor

Green

R-GA

“I sold a lot state

secrets and a lot of

very important

things”

Joe Bides brain is

going and he's

literally admitting his

crimes out loud.

Impeach Biden!

It’s unreal and so

insulting to America.

https://t.co/Wt58cB

7X5T

68581

-.5

Polarizing Right

- Government

@SenWarren

Sen.

Elizabeth

Warren

D-MA

The same

Supreme Court that

overturned Roe now

refuses to follow the

plain language of

the law on student

loan cancellation.

This fight is not

over.

The President has

more tools to cancel

student debt and he

must use them.

37748

-.753

-.4

Polarizing Left -

Government &

Economic

@RepGregLa

ndsman

Rep.

Greg

Landsmans

D-OH

We’ve got

incredible leaders in

SW Ohio who are

on the ground doing

the work, and

these projects will

make a huge impact

on our communities

-.187

.31

Unifying/Neutral

- Economic

Following the VADER sentiment analysis tool, each MoC tweets were coded

using nCoder into three categories to determine which themes drove polarizing rhetoric

on Twitter. The selected categories were Social, Government, and Economic for both

left and right and for the Unifying/Neutral non-polarizing categories. (Marquart et al.,

2019). After coding and achieving the following Kappa and Precision are seen in Table

Table 5

Kappa and Precision Scores from nCoder on RQ2

Code

Kappa

Precision

Left – Government

.89

.88

Left – Social

.82

.79

Left – Economic

.77

.87

Unifying – Government

.92

1.00

Unifying – Social

.66

.83

Unifying – Economic

.54

.69

Right – Government

.84

.88

Right – Social

.92

1.00

Right – Economic

.60

.82

According to the nCoder web tool, the Kappa and Precision of the data set were

sufficient to proceed with uploading the data into the ENA web tool (Marquart et al.,

2019, 2021). To test the effectiveness of the model, the ENA web tool will provide a

Pearson and a Spearmen Score (Marquart et al., 2021). Additionally, comparing to the

result and findings from other studies should provide insight into the accuracy of the

model (Ballard et al., 2023; Russell, 2021).

Research Question 3: Research Design and Data Gathering

RQ3 seeks to answer if the tweets created by former President Donald Trump

and former President Jair Bolsonaro were significant in instigating the capitol breach on

January 6, 2022 in Washington D.C. and the insurrection in Praça dos Três Poderes on

January 8, 2023. The likes, shares, and comment data will be collected from the specific

from @realDonaldTrump on November 19th, 2020 (Trump, 2020) and from

@jairbolsonaro on October, 30th, 2022 (Bolsonaro, 2022). Unfortunately, due to the

limits of the Twitter feed, only select tweets could be downloaded from the relevant time

frame to capture the pre-event sentiments: for @realDonaldTrump, 56 tweets were

collected, and for @jairbolsonaro, 142 tweets were collected. From this, the data will be

categorized to determine if the subjects responding to the tweet had indicated any

potential for incursion into the capital leading (Capitol Breach Cases, 2021).

Similar to RQ1 & RQ2 using nCoder and sentiment analysis, the following

comment threads on the specified tweets can be matched to words and phrases

indicating potential actions taken or sentiment regarding the tweet (Trump, 2020). The

following coding system was used, Potential Threat, Moderate Threat, Actual Threat,

Religion, Rigged. Defining these codes was based on the language used that could

signify a direct threat or just an open-ended confirmation of the former president's

Tweets; the codes Rigged and Religion were added because they were central themes

to the rhetoric used by former President Bolsonaro and former President Trump. While

religion and rigged do not signify threats, confirmation of the original tweet indicated a

correlation in developing a conversation study to classify the tweets. Table 5 below

shows the codes used in the manual coding as well as an example coding of three

different responses to the tweet.

Table 6

Codes used in RQ3

Religion (Good vs. Evil)

God, our side, protection, god bless,

honor

Rigged (Fraud)

Fraud, election, count, recount, votes,

voter id, corrupt

Come together (Potential threat)

Join, come together, count me in, let’s

go,

Justice/Freedom (Moderate threat)

Protest, freedom, resistance, we will

win,

Actual threats

War, fight, guns, strap up, prepare for,

Table 7

Sample Coding in RQ3

Replying to

Tweeter

Sentimen

t Score

Code

@jairbolsonaro

@dinizacessorios

Amen!

Prepared

for war,

let's go

Brazil

-.5093

Religion/Mino

r Threat,

Major Threat

@realDonaldTrum

@Perpetualmania

Count me

in, Mr.

President

.501

Come

Together

(minor threat)

Because the sample set for the data was relatively small, it precluded adding

additional tweets from both form President Bolsonaro and Trump. This way, the

standard codes seen in the first part of RQ3 can be explored in R2 to narrow and

account for how often each invokes polarizing language. The data set included 500

tweets from each of the former two presidents, collected starting two days following

each insurrection and going backward through each subsequent insurrection. Two days

following was selected because, on January 8, 2021, Twitter suspended former

President Donald Trump's account, and while it is currently reinstated, he has not

tweeted since (Conger et al., 2021; Elon Musk [@elonmusk], 2022). After reviewing the

tweets from each president, using nCoder, four categories were created to determine

themes in their subsequent tweets. Table 6 and 7 shows the subsequent codes with

their nCoder kappa and precision scores following the coding (Marquart et al., 2019).

Positive sentiments were included to account for times when either subject was not

overtly polarizing. The other codes, Rigged/Corruption accounting for claims of election

fraud, Good vs. evil was a catch-all-category for blaming the other candidate in the lost

election or invoking some division in Bolsonaro's case, this was typically a reference to

god, and in Trump's case he would frequently blame the other candidate or socialists,

and the final code Fight/Rebel was simply looking for tweets that invoked hostile

behavior.

Table 8

Codes for RQ3 Part 2

Positive Sentiments

Vaccines, MAGA, make America great,

American people, Brazil, jobs,

employment, trade

Rigged/Corruption

Rigged, stole, stolen, steal, fake, fraud,

voter fraud, signature,

Good vs. Evil

God bless, may god, god, save

America, Socialism, Lula, Biden,

Communist

Fight/Rebel

Fight, rise up, rebel, protect, defend,

march, stop,

Table 9:

Sample Coding for RQ3 Part 2

Respondent

Comment

Likes

Coding

@jairbolsonaro

"Put on the whole armor

of God, so that you may

be able to stand against

the wiles of the devil, for

our struggle is not against

humans, but against the

powers and authorities,

against the rulers of this

dark

world..." Ephesians 6:11-

- MAY GOD BLESS

OUR BELOVED

BRAZIL! 🇺🇸

https://t.co/T2A6iqgFgB

28210

40096

.97

Good vs. Evil,

Fight/Protect

realDonaldTrum

Peter Navarro releases

36-page report alleging

election fraud 'more than

sufficient' to swing victory

to Trump

https://t.co/D8KrMHnFdK

. A great report by Peter.

Statistically impossible to

have lost the 2020

Election. Big protest in

D.C. on January 6th. Be

there, will be wild!

14224

36231

.50

Rigged/Corruption

, Fight/Protect

Table 10:

Kappa and Precision Scores from nCoder on RQ3 p.2

Code

Kappa

Precision

Positive Sentiments

.83

1.00

Rigged/Corruption

.71

.74

Good vs. Evil

.67

1.00

Fight/Protect

.71

.85

Since a sufficient Kappa and Precision were reached after using nCoder, VADER

sentiment analysis was conducted, and a sample of the following Tweets is shown in

Table 8 (Hutto, 2023; Marquart et al., 2019). The results discussed in Chapter 4 will

review the ENA web tool results for the Pearson and Spearmen scores to validate the

corresponding data sets in addition to validating the study and hypothesis H3a

(Marquart et al., 2021).

Human Subjects Considerations

The following study relies entirely on existing data sets generated by Twitter’s

feeds and historical case analysis. Due to the nature of the study, no human interactions

will be necessary to conduct any research. Twitter’s API is open and available to all

individuals by creating a free account at developer.twitter.com. Twitter’s API allows any

user to create data sets from all open and public accounts. This study will only use open

accounts and look at public tweets. Since the data is made available to the general

public, IRB subject review is not required per IRB guidelines stated in the Belmont

Report (HHS, 1979).

Proposed Data Analysis Process

Epistemic Network Analysis (ENA) must use data that can be used in machine

learning such that it is standardized to represent data when imported into the ENA tool

found at epistemicnetwork.org (D. W. Shaffer, 2014). Part 1 of Research Question

Number 1 seeks to determine if the Sentiments of the tweets are positive, neutral, or

negative. No network is being created in this analysis, as results will be returned in

variable formatting, so ENA will not be used to analyze the studies' SA. The study will

consider and assume that a positive sentiment score is unifying and a negative

sentiment score is polarizing. Once the average VADER sentiment scores are

calculated for each MoC, using the Mann-Whitney U test, and Voteview.com, the data

will be compared to see if VADER's SA score helps determine the polarization of a

particular MoC (Hutto, 2023; Lewis et al., 2023; Marquart et al., 2019, 2021; McKnight &

Najab, 2010).

Using nCoder can add the variability of the type of polarization and determine if

the specific MoC is pushing highly conservative or liberal narratives based on the coding

type used (Marquart et al., 2019). From this, the data can be interpreted to create

networks that identify groups of MoC making highly polarizing narratives on social

media. nCoder, using machine learning, can look at the massive data sets acquired

from the members of Congress's tweets and match them to phrases and words that

convey the polarization (Marquart et al., 2019). Once these networks are created, the

SA scores can be added to the networks to add validity using the method developed by

Misiejuk et al. (2021), which can help quantify the effect of potential statements made

using the ENA tool (Marquart et al., 2021).

In Research Question 2, the tweets and responses of the five most significant

contributors to the liberal and conservative polarizing language tweets will be collected

using the TwExportly and again coded using VADER and nCoder results (Hutto, 2023;

Marquart et al., 2019; TWExportly, 2023). For this study, the selected tweets reviewed

were the highest shared and commented-on tweets prior to July 28th, 2023 going back

to 200 tweets for each MoC. The nCoder results of the language used by the MoC of the

specified tweets will help provide insight into the users' reactions to the specified data.

Then like RQ1, the Misiejuk et al. (2021) method for incorporating VADER into ENA will

be added to each of the responses to verify if there is any correlation between positive

and negative sentiments to the potentially polarizing tweets (Hutto, 2023;

Shaffer et al., 2016).

Validating Study

In RQ1, data validation of the VADER sentiment scores and the DW-NOMINATE

scores were compared using comparative and regression analysis (Hutto, 2023; Lewis

et al., 2023). These regressions of the ENA results determined how close each

MoC'sMoC's VADER and nCoder scores are compared to their recent DW-NOMAINTE

pulled at the same instance of their last roll call vote (Lewis et al., 2023). Some

variances, however, don'tdon't suggest the study is invalid, as some MoC might be

leveraging polarizing rhetoric to exploit Twitter'sTwitter's algorithm but voting more

predictably. A 2010 review of Congressional member Twitter accounts revealed that the

platform'splatform's primary use was self-promotion and engaging with their base; it was

rare that MoCs used the platform to discuss policy formation (Golbeck et al., 2010). It is

becoming increasingly evident that social media has disrupted America'sAmerica's

general news and information sources. How the political elite uses social media will

determine the safety of America and other democracies across the globe, as the power

of their messages is highly effective in generating mass polarization. While this study

will provide insight into the problems surrounding Twitter, further research will be needed

to understand how the masses perceive Twitter feeds.

Chapter 4: Findings

Research Question 1: Results and Findings

As stated in this dissertation's purpose statement, this study aims to determine if

social media, specifically Twitter, is being utilized by politicians to polarize their bases to

gain political ground. Research Question 1: Do the political elite (Members of Congress)

leverage Twitter to promote identity politics furthering political polarization in America?

While our null hypothesis states there is no correlation between political elite tweets and

affective polarization in America, our hypothesis suggests that ENA would reveal a

correlation between the tweets of the political elite and the rise in affective polarization in

America.

The study compared DW-Nominate scores for Members of Congress found at

Voteview.com to the rhetoric revealed on their official Twitter accounts (Lewis et al.,

2023). The first step was ranking the members of Congress by their polarization score

taken from VoteView.com on August 28th, 2023; the politicians were then grouped by

their level of polarization: Highly polarized Republican/Democrat, Median

Republican/Democrat, and Centrist Republican/Democrat (Lewis et al., 2023).

To grab an influential sample group, ten members from each of the groupings, as

mentioned earlier's extremes and each party's medians were selected for their official

Twitter accounts to be studied. Two hundred tweets were scraped from Twitter using the

100xTools.com TWExportly tool from each of the selected 60 members of the House

and 60 members of the Senate for a total of 24,000 tweets (2023). Once all the data

was extracted and the tweets were categorized, the VADER sentiment analysis tool was

used to score each of the tweets, with 1 being a positive sentiment score, 0 as neutral,

and -1 as a negative (Hutto, 2023).

The VADER sentiment scores were then averaged for each selected member of

the Representative and Senate (Hutto, 2023). The findings, when averaged by each

group Far Democrat, Median Democrat, Centrist Democrat, Centrist Republican,

Median Republican, and Far Republican, revealed high Pearson correlation coefficients

between both Senate (R2 = .9636) and members of the House (R2 = .8945).

Figure 5

DW-Nominate Scores of Representatives vs. Sentiment Scores

Figure 6

DW-Nominate Scores of Senators vs. Sentiment Scores

In this, the absolute value of the DW-Nominate Score was used to determine

mostly level of polarization in comparison to their average sentiment scores. The study

found that MoCs with lower DW-Nominate scores (Center Republicans and Center

Democrats) were more likely to have higher sentiment scores than those who were

highly polarized based on their DW-Nominate for both parties. The average Sentiment

Score among all the tweets for Senators and Representatives was .283. The median

was significantly higher at .422, suggesting a negative skewness which was not

surprising as it seems more and more politicians are using negative language in their

regular tweets. Tying sentiment scores to congressional voting behavior is not a new

concept. While it appeared to work well with the data set using more recent tweets,

another study found exciting variations in the data where skewness changed under the

Trump presidency (Spell et al., 2020). Should Twitter’s API be opened to academics,

tweets from different presidencies should be correlated to the sentiment scores. One

hypothesis would suggest that members of the same party as the president would

generally show higher sentiment scores in their tweets.

For the second part of the study, the data set, once coded with nCoder, was

applied to ENA (Bowman et al., 2021; D. Shaffer & Ruis, 2017; D. W. Shaffer, 2018; D.

W. Shaffer et al., 2016) using the ENA Web Tool (version 1.7.0; Marquart et al., 2021) I

defined the units of analysis as all lines of data associated with a single value of Party

(Republican/Democrat) subsetted by Chamber (House of Representatives/Senate) and

each Member of Congress (MOC) reviewed. ENA algorithm uses a moving window to

construct a network model for each line in the data, showing how codes in the current

line are connected to codes that occurred previously (Ruis et al., 2019;

SiebertEvenstone et al., 2017), defined as all lines preceding the current line within a

given conversation. In this model, an infinite stanza was used as the study is only

looking at recent tweets and not connecting with tweets from the past. The resulting

networks are aggregated for all lines for each unit of analysis in the model. In this

model, an aggregated network was used in a binary summation in which the networks

for a given line reflect the presence or absence of the co-occurrence of each pair of

codes.

The ENA model included the following codes: Left Polarizing, Neutral/Unifying,

and Right Polarizing. The study defined all conversations as all lines of data associated

with a single value of Position. For example, one conversation comprised all the lines

associated with Position and Far-R. The ENA model normalized the networks for all

units of analysis before they were subjected to a dimensional reduction, which accounts

for the fact that different units of analysis may have different numbers of coded lines in

the data. A singular value decomposition was used for the dimensional reduction, which

produces orthogonal dimensions that maximize the variance explained by each

dimension (See Bowman et al., 2021, and Shaffer et al., 2016 for a more detailed

explanation of the mathematics).

Networks were visualized using network graphs where nodes correspond to the

codes, and edges reflect the relative frequency of co-occurrence, or connection,

between two codes. The result is two coordinated representations for each unit of

analysis: (1) a plotted point, which represents the location of that unit’s network in the

low-dimensional projected space, and (2) a weighted network graph. The positions of

the network graph nodes are fixed, and those positions are determined by an

optimization routine that minimizes the difference between the plotted points and their

corresponding network centroids. Because of this co-registration of network graphs and

projected space, the positions of the network graph nodes—and the connections they

define—can be used to interpret the dimensions of the projected space and explain the

positions of plotted points in the space. The model had co-registration correlations of

0.99 (Pearson) and 0.99 (Spearman) for the first dimension and co-registration

correlations of 0.99 (Pearson) and 1 (Spearman) for the second. These measures

indicate that there is a strong goodness of fit between the visualization and the original

model.

ENA can be used to compare units of analysis in terms of their plotted point

positions, individual networks, mean plotted point positions, and mean networks, which

average the connection weights across individual networks. Networks may also be

compared using network difference graphs. These graphs are calculated by subtracting

the weight of each connection in one network from the corresponding connections in

another. To test for differences, a Mann-Whitney test was applied to the location of

points in the projected ENA space for units of Independent Senator Sanders (identified

as a Democrat in the ENA tool for ease of viewing and because he caucuses with the

Democrats) and Republican Senator Lisa Murkowski. Along the X axis (MR1), a

MannWhitney test showed that Democrat (Mdn = -0.64, N = 60) was statistically

significantly different at the alpha = 0.05 level from Republican (Mdn = 0.66, N = 65 U =

404.00, p = 0.00, r = 0.79). Along the Y axis (SVD2), a Mann-Whitney test showed that

Democrat

(Mdn = 0.09, N = 60) was not statistically significantly different at the alpha = 0.05 level

from Republican (Mdn=-0.10, N =65 U = 2036.00, p = 0.67, r = -0.04). The

MannWhitney test revealed that the study’s correlation and visualization revealed there

are significant differences between Republican and Democratic members of Congress

and their use of either conservative or liberally polarizing language. However, it did

show that for non-polarizing tweets, there were minimal differences; this is not surprising

as the Senators and Members of Congress taken from the center of the DW-Nominate

Score were found to be more likely to use non-polarizing language and, even more

surprisingly, some center Democratic MoC (Rep. Golden D-MA, Senator Stabenow –

DMI, Rep. Don Beyer D-VA, Rep. Robert Garcia D-CA) and were found on the right or

the nearing the right of the spectrum seen in Figure 13. Similarly, many center-leaning

Republicans were identified as using fewer polarizing tweets and left-leaning tweets in

the polarization spectrum. Figure 8 shows Senator Lisa Murkowski R-AK, Rep. Molinaro

R-NY, Rep. Brian Fitzpatrick R-PA, Rep. Tom Kean R-NJ, as well as several others were

found in the upper left quadrant signifying uses of higher uses of non-polarizing

language and some use of Left Polarizing language.

Figure 7

ENA Results RQ1 all MoC

Figure 8

ENA Results RQ1 Centrist

When comparing the overall results from the ENA, incorporating all 60 members

from the House and Senate, each reveals interesting similarities visually to the

VoteView.com DW-nominate visualization, with the general outliers remaining similar for

Republican Party MoC but slightly different on the Democratic Side the same (Lewis et

al., 2023).

Further left or right indicates more significant levels of either Left or Right

Polarization, with the Y-axis representing increased positive sentiments and

probipartisanship as one moves up the Y-axis. For the House of Representatives,

Representative Margorie Taylor Greene, R-GA, and a DW-Nominate Score of .8 and is

one of only three representatives with a negative average VADER sentiment score at

.128 (Hutto, 2023; Lewis et al., 2023). On the Democrat party side, Rep. Greene R-GA,

is a clear outlier as she is on the extreme right. In contrast, Rep. Sean Casten D-IL was

a slight outlier on the ENA results, not unlike his DW-Nominate score as the 3rd most

liberal with a score of .673 and a positive sentiment score of just .188. The Republican

Representatives did have two strange outliers regarding scoring high in

Neutral/Unifying: Rep. David Valado, R-CA, and Rep. Adrian Smith, R-NE. After

reviewing the tweets that were scored highly, it was clear that they frequently

congratulated veterans and wished their followers happy holidays, which the n-Coder

system recognized as a Neutral/Unifying code. While these types of tweets are, in fact,

positive sentiments, they aren't rewarded with much traction on the Twitter platform;

RQ2 will reveal in greater detail why these types of tweets are significant in considering

the polarizing effects of politicians' use of Twitter.

Figures 9 and 10 show the selected members of the House, except for the

aforementioned outliers; the distribution is relatively normal compared to the

expectations seen in Voteview.com DW-Nominate results (Lewis et al., 2023). Likewise,

in Figure 9, a similar distribution for the Senators is included in the study. The study

results merit further inquiry into RQ2 to look at how the messages of polarization are

being categorized and delivered by the MoC on Twitter. The simple three-code structure

provides some insight; however, the additional coding provided in RQ will add further

clarification.

Figure 9

ENA results for RQ1: House of Representatives

Figure 10

ENA results for RQ1: Senators

In general, for both the Democrat and Republican Representatives and Senators,

the outliers are predictable based on the candidates' position in the ENA study (Figure 9

& 10) to their relative DW-Nominate Positioning see Figure 11; however, the centrist are

less predictable as there is overlap as mentioned above in the ENA but none in

DWNominate (Lewis et al., 2023). Part of this may be attributed to the catch-all

polarization categories that are simply based on terms and keywords used on Twitter.

Many of the terms at the center of each party will likely be the same.

Figure 11:

118th Congress House of Rep. DW-NOM. Distribution

Note. Adapted from “DW-Nominate Plot: Representatives.” VoteView.com, Lewis et. Al

(2023) https://voteview.com/congress/house. Reprinted with permission. Figure 12

118th Congress Senators DW-NOM. Distribution

Note. Adapted from “DW-Nominate Plot: Representatives.” VoteView.com, Lewis et. Al

(2023) https://voteview.com/congress/house. Reprinted with permission.

Figure 12 shows the average ENA results for each party's Far, Median, and

Center sub-group. Across the Democratic Senators, the distributions were also quite

similar except for Democratic Senators in the Median range who were the most

polarized of the Democratic party based on their ENA average scores. This is an

exciting occurrence; one reason for this unexpected outlier is that one of the more

prolific members of the Senate Democrats was included in the study, Sen. Charles

Schumer, D-NY, at the median. I hypothesize that because Sen. Schumer is the

appointed leader of the Democratic Senators, his voice is supposed to be the voice of

the party, and as Congress is becoming seemingly more polarized, choosing polarizing

rhetoric as the leader might be encouraged at the party level. The averages in Figure 18

indeed show that on a party level, there is increased polarization as there is zero

crossover between any of the averaged groups. The similarities in the outliers seen in

Figures 8, 9, 10, and 13 suggest that the ENA models reflecting the high levels of

polarization used by Members of Congress on Twitter are correlated DW-Nominate

Scores.

Figure 13

ENA Averages for MoC by Group Average

Comparing the results to the Hemphill et al. (2016) study, which correlated

DWNominate scores to hashtag use on Instagram, found some interesting correlations.

For example, Sen. Joe Manchin D-WB was more likely to use red hashtags (ones used

by the Republican party) and was found on the right side of the Y-Axis in the ENA study

(Hemphill et al., 2016). Additionally, Sen. Rand R-KY was an outlier on the far right in

both studies (Hemphill et al., 2016). Unfortunately, there were few other MoCs that I

could compare using the #Polar Scores as a validation tool. However, it will be

interesting should Twitter make changes that bring back the effectiveness of the hashtag

(Hemphill et al., 2016).

Research Question 2: Results and Findings

Research Question 2 follows Question 1 in so much as it looks at how individuals

respond to the tweets shared by Members of Congress online by adding value

components to the amount of shares, replies, and likes each tweet gets. Question 1

revealed that MoCs use social media to leverage their platforms to promote various

forms of polarizing rhetoric. What is unanswered by Question 1 is: How influential are

these tweets in being polarizing? RQ2 asks: Is Twitter's algorithm giving preferential

status to the political elite who use polarizing tweets to generate higher user

engagement resulting in higher ad revenue? The null hypothesis suggested that Twitter

does not give polarizing tweets preferential treatment, and the hypothesis suggests that

Twitter's algorithm favors highly polarized tweets.

The study was conducted using the same Twitter data from Question 1 but further

analyzed and modified the coding to reveal the type of polarization being used in the

studied tweets. These codes were broken down into three specific categories of the

tweeter's affective or pernicious polarizing tweets, which were as follows: Government

(Republicans/Democrats side blaming the other), Social (MoC condemning social

issues relevant to their opposition as being related to any decline) and finally the

Economy (MoC criticizing economic policies of the other parties). The same categorical

considerations were made when looking into Neutral/Unifying tweets to determine at

what level of social media engagement these tweets were either disappearing or not

being shared at scale.

Using ENA, RQ2, and the previously mentioned coding system (also seen in

Table 3) across all the 24,000 tweets, three different models were created. The second

ENA analysis took the top 200 most shared tweets of the 24,000, then were hand coded

again using the same coding system. Then finally, on the third ENA review, the codes for

all Left and Right designations were removed, Social Government and Economic

categories were combined, and the Neutral/Unifying codes all into a single category to

determine if polarizing tweets were, in fact, more likely to gain virality among the Twitter

users.

The study revealed in RQ1 that politicians at the median tended to use both

liberal (L) and conservative (R) polarizing terms while still maintaining a strong

emphasis towards neutrality/unification, as depicted in Figure 18 as Sen. Far L & R were

above the X-Axis revealing higher usage of neutral/unifying tweets than negative ones.

What was surprising was Democratic Senators, in the median of their party, on average,

were more polarized on Twitter than their constituents identified as far left by their DW-

Nominate scores (Lewis et al., 2023). This suggests that while these members of

Congress might be less polarizing in terms of their voting record, they are using

polarizing rhetoric on Twitter to engage their audience. To test this, two approaches to

ENA were used with the ENA web tool on RQ2 to determine if the users are engaging

with these polarizing tweets or if they are falling on deaf ears (Marquart et al., 2021; D.

W. Shaffer et al., 2016). I used the coding system in Table 3 below across all the 24,000

tweets downloaded in RQ1 and added a ranking based on engagement from Twitter

users by assigning averages of likes and shares from each MoC tweets. Following this,

to narrow the study, codes for all Left and Right designations were removed. They were

coded specifically for Social, Government, and Economic categories, and all the

Neutral/Unifying codes all into a combined single category to determine what type of

polarizing tweet was more likely to gain virality among the Twitter users.

Rather than looking specifically at individuals, MoCs were divided into four

categories, Highest Engagement, Above Median, Below Median, and Lowest

Engagement. The level of engagement was determined by ranking each congressional

member by their average number of likes from the subset of 200 tweets downloaded. An

infinite stanza was used, as was used in RQ1, since only recent tweets were being

reviewed and not looking at the comments following each tweet. The primary model

used in RQ2 had co-registration correlations of 0.99 (Pearson) and 0.99 (Spearman) for

the first dimension and co-registration correlations of 0.99 (Pearson) and 0.99

(Spearman) for the second. These measures indicate that there is a strong goodness of

fit between the visualization and the original model. Not unlike the results from RQ1,

similar variations of polarization found in the Far Right, Far Left, and centrist were

identified in the matrix created by the ENA model. Figure 14 reflects all the

Congressional Members studied using all nine codes coded using nCoder with networks

for Republicans and Democrats created.

Figure 14

ENA Results RQ2

By breaking down each code into the causes of polarization (Social, Economic,

and Governmental), the immediate result revealed a far less visible crossover between

even among the centrist Republicans and Democrats. Politicians move down and away

from the Y-Axis as they become more polarized and less unifying.

When narrowing the study down, looking at specific groups of politicians ranked

and categorized by their average level of engagement, the ability to determine how the

most engaged MoC’s tweets compare to the lesser engaged. To this, a quartile system

was used to count the average likes and retweets, and each MoC studied was

categorized. Those with the Highest Engagement, unsurprisingly, were the most

polarizing, seen at the low end of the Y-axis, and the lowest engaging politicians were

on the highest point of the Y-axis, seen in Figure 15.

Figure 15

ENA Result RQ2: Engagement Averages

When broken down by engagement, those with higher levels of polarization see

the highest levels of engagement, and as an MoC becomes less polarizing, they see

less engagement on Twitter. Another study found nearly identical results and suggested

that these behaviors by MoCs encourage increased funding from their donors (Ballard

et al., 2023). Additionally, Ballard et al. (2023) found that MoCs of the president's

opposite party were often more vitriol and polarizing. In the case of averages, this wasn't

necessarily the case in the study, however on an individual level, regarding the most

polarizing Republican (Rep. Greene – GA Figure 16) vs. the most polarizing Democrat

(Sen. Warren – MA Figure 17) on Twitter, the Republican MoC was the most polarizing

with a very high network score seen on the third ENA study comparing only codes for

causes of polarization.

Figure 16

ENA Results for @RepMTG

Figure 17

ENA Results for SenWarren

When comparing Figures 16 and 17, the nodes on each suggest a shift in

concerns for both ideological extremes amongst conservatives and liberals. Typically,

Republicans are viewed as expressing economic concerns and being the party of

business owners and those fiscally concerned; however, the nodes suggest that Rep

Greene R-GA's primary concern is government overreach and social concerns, versus

Sen. Warren D-MA, who is more focused on Economic issues. Further analysis was

gathered to test whether this was true for all members of Congress from each party; the

findings revealed that Republicans confirmed that there certainly was a shift towards

Government and Social concerns; however, Democrats were significantly more

scattered. Figure 18 looks at the highest engaging MoC studied, and not unlike Rep.

Greene R-GA, the Republican MoC was clustered more significantly around the Social

and Government nodes. The lesser engaged MoC from the Republican party, seen in

Figure 18, is clustered closer to Unifying and Economic, which is in keeping with the

past perceptions of the Republican party (Klein, 2020). Democrats with the highest

engagements (Figure 18) are clustered a bit more scattered, and the thin lines between

the network scores suggest that there is not a high correlation between any of the nodes

and the party average, which was the highest at .21 for Unifying/Neutral and Left

Economic.

Figure 18

Results for Republicans and Democrats

The primary objective of RQ2 was to determine if the Twitter algorithm favored

polarizing tweets. The results seen in Figure 15 provide a solid reason to assume that

polarizing tweets get higher likes and shares. The results of this study suggest that

further research on Twitter algorithms favoring polarizing rhetoric be considered.

Research Question 3: Results and Findings

The final question of the study looks at the potential negatives of polarizing

rhetoric on social media. The question: Were President Donald Trump's

(@realDonaldTrump) and President Jair Bolsonaro of Brazil (@jairbolsonaro) tweets

leading up to the uprising of January 6, 2021, and January 8th, 2023, respectively,

responsible for the unfortunate and subsequent events? The null hypothesis suggests

that Former President Donald Trump's and former Brazilian President Jair Bolsonaro's

tweets did not significantly impact the January 6, 2021, uprising in Washington D.C. or

the January 8, 2023, uprising in Brazil. The hypothesis concludes that using ENA,

former President Donald Trump and former President Jair Bolsonaro tweets used

rhetoric that may have had a significant impact on the events that took place on January

6 and January 8.

The study again uses ENA to compare two separate data sets to determine if in

fact, the former presidents' use of Twitter had a significant impact on the subsequent

insurrections. This study was met with many challenges as Elon Musk eliminated Twitter

API access in late July of 2023 while the study was taking place. This limited the ability

for individuals to retrieve legacy data from Twitter feeds and made it nearly impossible to

extract both quote tweets and comments on tweets. To best answer the question, two

small data sets were used.

The first data set comprised of responses of from both president's twitter

followers. Because there had been many comments in response to the January 6th

uprising made on @realDonaldTrump's account, only 56 quotes from former President

Trump's tweet on December 19th, 2020 were recovered that were prior to the Jan 6

date. From former President Bolsonaro I was able to manually copy 142 tweets prior to

January 8th insurrection in response to his post on October 30th, 2023 (which have

been cited as the tweets having potentially caused each insurrection). These were

manually coded using the codes seen in Table 5 (above) and be imported into the ENA

web tool (Bugs et al., 2023; Harton et al., 2022; Marquart et al., 2021). The model had

co-registration correlations of 0.98 (Pearson) and 0.99 (Spearman) for the first

dimension and co-registration correlations of 0.99 (Pearson) and 0.98 (Spearman) for

the second. These measures indicate that there is a strong goodness of fit between the

visualization and the original model. Five codes were used after ranking the responses

in groups of those affirming the respective president or descending, and then grouped

by their respective quintile-ranked VADER sentiment scores (Hutto, 2023).

The results seen in Figure 19 include the averages for those who responded to

@realDonaldTrump and @jairbolsonaro, who were identified as supporters based on

the rhetoric used in the tweet. For @realDonaldTrump (red) the most substantial

network was between the potential minimal threat and Rigged with a network score of .5

in the line on Figure 19. For @jairbolsonaro (purple), the most substantial network of .42

was between Religion and Actual threats. This is unsurprising as the former President of

Brazil uses religious language frequently in his tweets and rhetoric, invoking biblical

quotes referencing weapons and armor.

Figure 19

ENA Results RQ3 Part 1

The study, however, was less conclusive for showing threats on behalf of Donald

Trump followers, a part of this might be because many of the accounts and hashtags

used at the time invoking violence were banned from Twitter after January 6. What is

interesting is that several media outlets reported that Twitter was ineffective at managing

threats of violence in the January 8 uprising (Dwoskin, 2023; Scott, 2023; Stargardter &

Ayres, n.d.). It was reported by Dwoskin (2023) that several of the previously banned

terms following the January 6 rising in the United States, like “Stop the Steal,” were

being used in Portuguese. Whether Twitter had either failed to remove this ban or

unsuccessfully recognized the tweet due to the language variance was not revealed by

Twitter; however, further studies should take place.

The similarities between January 6 and 8 are unsurprising; however,

understanding the rhetoric leading up to and following by both former Presidents Trump

and Bolsonaro is essential in understanding if polarizing language can instigate riots.

The coding seen in Table 8 was applied to 500 of each president’s Twitter accounts

leading up to the event and two days after using nCoder. The following codes were then

uploaded to the ENA web tool for analysis. The model of the former presidents’ tweets

had co-registration correlations of 1 (Pearson) and 1 (Spearman) for the first dimension

and co-registration correlations of 1 (Pearson) and 1 (Spearman) for the second. These

measures indicate that there is an intense goodness of fit between the visualization and

the original model. The following models seen in Figure 20 were produced using the

ENA webtool coding looking at the average @realDonaldTrump and @jairbolonaro

tweet accounting for the level of engagement (low, median, and high) with a

conversation engaging the sentiment scores using a quintile system (lowest, low,

medium, high, highest) with an infinite stanza. The most substantial network for

@realDonald Trump was between Government Corruption and Rise up/Fight with a

score of .55, while @jairbolsonaro’s most substantial network was narrowly

Economy/Growth and Good vs. Evil with a score of .52, only just above Good vs. Evil

and Rise up/Fight at .51. These networks are engaging, as there is a higher use of

nonpolarizing discussion in @jairbolsonaro’s feed, this is reflective of their average

VADER sentiment scores, @jairbolsonaro averaged .159 while @realDonaldTrump

average just .070.

Figure 20

Results for RQ3 Engagement Considered

What becomes increasingly concerning is how much the levels of polarization

increase among both presidents’ tweets when one examines only the highest levels of

engagement. For @realDonaldTrump, the highest network connection at the highest

engagement was between Government Corruption and Rise up/Fight, with a network

score of 1.00 and a score of .87 between Government Corruption and Good vs. Evil

(seen in Figures 21 and 22). Similarly, @realjairbolsonaro had the most substantial

connection at Good vs. Evil and Rise up/Fight with a score of .94 in Figure 21.

Figure 21

RQ3 Part 2: Highest Engagement @realDonaldTrump

Figure 22

RQ3 Part 2: Highest Engagement @jairBolsonaro

Conclusively it is nearly impossible to say that the rhetoric of both former

President Trump and Jair Bolsonaro led to the subsequent insurrections. However, the

studies reveal that both used highly inflammatory language in their tweets. While neither

directly asked their followers to attack the capitols, they gave what many followers felt

was sufficient validation for attacking the capital. Several insurrections on trial for the

January 6 attack cited the tweets from @realDonaldTrump and comments made by the

former president at his preceding rally for their call to arms (Dreisbach, 2022; Harton et

al., 2022). While these are not enough to preclude guilty verdicts, the Washington D.C.

district attorney has filed four criminal charges against him: “conspiracy to defraud the

U.S., conspiracy to obstruct an official proceeding, obstruction of an official proceeding,

and conspiracy against the rights of the citizen” (Debussman, 2023). Former President

Jair Bolsonaro, on the other hand, was yet to be charged with any crimes in Brazil.

Chapter 5: Implications

Summary of Study and Findings

Determining the best approach for studying polarization can be challenging as

there are many contrarian viewpoints on polarization in the United States. There are

some that believe that America is not polarized, but instead, partisan sorting is causing

the illusion of polarization in America (Fiorina et al., 2008). Regardless of these

contradictory viewpoints, the research concluded in this study showed that polarization

does exist in the context of Members of Congress and their use of Twitter. To more

significant concern, the level of vitriol content being created on social media by the

MoC, or political elite, can be identified as pernicious polarization, which can only lead

to dire consequences, as seen in the January 6 and 8 insurrections (McCoy & Somer,

2019).

This study conjectures that there is an existence of elite polarization that is

evident based on the usage of Twitter by many members of Congress. The study used

data specifically from Twitter because it is utilized by nearly all the Members of

Congress and with nearly 78 million American users (Shepherd, 2023). Creating a study

using Epistemic Network Analysis was designed such that the triggers of polarization

could be identified in the rhetoric of the specified MoC while creating networks between

the types of rhetoric being used (D. W. Shaffer, 2018).

This study chose to look specifically at members of Congress and former

Presidents Bolsonaro and Trump to determine if their influence on social media,

specifically Twitter, was polarizing and, in specific cases instigating insurrections. By

framing the study in three parts, I determined if there was polarizing rhetoric on social

media, how effective polarizing rhetoric was in engaging an audience, and if potentially

harmful rhetoric could be cause for concern.

Research Question 1

RQ1 asked if the MoC studied were using polarizing language on Twitter.

Polarization was defined in two categories Left Polarizing and Right Polarizing. Then a

Neutral/Unifying category was added to catch all non-polarizing tweets. The study

reviewed tweets by coding them with the nCoder machine learning tool and loading the

ENA Webtool after completing the VADER Sentiment analysis (Hutto, 2023; Marquart et

al., 2019, 2021).

The results of the ENA study revealed less significant polarization at the center

but increased polarization at the extremes when comparing DW-NOMINATE scores to

ENA network placement (Lewis et al., 2023; Marquart et al., 2021). This is not out of the

norm as multiple studies have concluded that far right and far left members of

Congress, as well as their constituents, had shown increasingly more protective of their

partisan identity as well as likely to attack their rivals (Bail et al., 2018; Rathje et al.,

2021; Van Bavel et al., 2021).

What was noteworthy, however, was the correlation between the VADER

sentiment analysis and the level of polarization determined by the DW-NOMINATE

(Lewis et al., 2023). Politicians who were scored as having higher levels of polarization

were more likely to use hostile rhetoric. This isn’t unanticipated, but the strong linear

regressions for both Senators (r2=.96) and Representatives of the House (r2=.96)

suggest that VADER sentiment analysis could be a good gauge for monitoring social

media content (Hutto, 2023).

Monitoring and regulating content has proven to be challenging and costly for

social media companies due to the extensiveness of their user base. This theoretically

reveals VADER as an effective tool for monitoring content; ideally, social media

companies take it upon themselves to harness the power of AI to be more proactive in

protecting its users (Hutto, 2023). This was evident in the instance of the January 8

insurrection in Brazil. One journalist concluded that Telegram, Facebook, and Instagram

were all negligent in their recognition of the events as they transpired, even though all

the signs indicating an insurrection were actively being posted (Scott, 2023). It appears

that AI will offer new avenues to look at social media in real-time and provide insights

that might prevent negative actions from occurring.

Research Question 2

RQ2 took it one step further by breaking down the types of rhetoric used to

polarize. The categories selected were social, government (often identified as political,

for this study, it carries the same meaning), and economic. These tend to be the key

concerns when voters make election decisions (Budge et al., 1987). Social concerns for

both parties have been hot tokens as debates over the Supreme Court's recent overturn

of Roe v. Wade have ignited both parties into heated partisan dialogue (Swers, 2023). In

government/political, this code focuses primarily on attacks on the other side, generally

when searching through all the Tweets studied, it was more likely for Democratic

Members of Congress to mention Trump negatively and Republicans to mention Biden

negatively than it was for either party to congratulate/or praise their respective

President. The last coded category was economic; for Republicans, this was generally

targeted at "Bidenomics" and blaming President Joseph R. Biden for the inflation.

Democrats, on the other hand, were expressing their economic woes over acrimonious

tweets about the Supreme Court's rejection of Biden's student debt relief and

Republicans' lagging in working with Biden.

The categorical breakdown of types of polarizing rhetoric seen on Twitter further

provided the additional insights needed to show how political elites leverage affective

and pernicious tweets. The ENA web tool study revealed that Twitter was amplifying the

voices of the most vitriolic and hostile tweets and rewarding them. The study showed

that the most polarizing members of both parties saw much higher engagement than

those who were ideologically centrist. This is a piece of familiar information, as Rathje et

al. (2021) uncovered this in their respective study; however, what RQ2 reveals is that

politicians are leveraging consistent underlying themes to gain further momentum on

social platforms while using hostile rhetoric. Figure 23 breakdown the highest engaging

members of Congress in the study; the anticipated members like RepMTG (Rep Greene

R-GA) and RepAndyBiggsAZ are seen at the far bottom right, indicating high

polarization. However, the distribution for Democrats is not as predictable individually.

Figure 24 shows the lowest engaging MoC on Twitter and most of them are in the center

upper quadrants signifying more unifying and less polarization. The stark differences

between Figures 23 and 24 show how polarizing rhetoric is more powerful in terms of

distribution and engagement on Twitter.

Senator John Ossoff, D-GA, a Junior Senator, was categorized as a Far

Democrat because his DW-NOMINTE score was one of the ten lowest at -.454.

However, he frequently works with Republican Senators (Lewis et al., 2023). Sen.

Offsoff, who defeated a Republican incumbent in the 2020 elections, narrowly won in a

runoff. Author and Journalist Ezra Klein describes this high level of polarization driving

this competitiveness in politics (Klein, 2020, p. 250).

Figure 23

Highest Engaged Sen. & Reps.

Figure 24

Lowest Engaging Members of Congress

Former President Donald Trump in 2016 had successfully dismantled traditional

strategy by becoming a significant outsider with highly charged political views.

Democrats, on the other hand, have typically stepped away from propelling outsiders,

and much of this is because there is less amount of deviation in the party (Noel, 2016).

Comparing DW-NOMINATE scores, the median for the Republican party is .51 while the

Democrats are much closer to the center at -.37 (Lewis et al., 2023). The Republican

party is generally more divided than the Democratic party, and as a result, there are

certainly higher instances of polarization; this is both evident in voting behavior and the

rhetoric seen on Twitter. The results of this fractioning have seen an increase in new rifts

forming in the Republican Party. First, it was the Tea Party, and now it is the

Freedom Caucus (Noel, 2016). The Freedom Caucus has been relatively unsuccessful

in security a majority, they have added incredible challenges for Speaker of the House

Kevin McCarthy, but overall, they cannot gain leadership at the Presidential level or

even the highest party seats (Baer, 2023; Noel, 2016).

While the Freedom Caucasus has been less successful in the elections, RQ2

reveals its success on Twitter, so much so that a shift in Republican ideology seems to

be occurring on Twitter. House Freedom Caucasus Members, who have attached

themselves to the former President Trump, leveraged much of the xenophobic rhetoric

espoused by Trump. With the ENA web tool, an additional study using the RQ2 data set

with just the Highest Engaged Republicans and the polarizing codes identified for

conservatives/right in Figure 25 (Marquart et al., 2021). The results revealed that

Government and Social concerns were the highest tweeted among highly engaged

Republicans, with a network score of .72.

Figure 25

Republican High Engagement

Social concerns seen in the codes for Republicans on Table 3 that had high

occurrence were "Southern Border," "Child sex trafficking," "Protect Children's

innocence," and "religious liberty." Child sex trafficking has been a remarkably dividing

term as followers of the QAnon conspiracy have leveraged this narrative as an attack on

the Democratic elite (Bleakley, 2023). While it is disheartening the rhetoric being spread

by MoC on the right, it is no surprise. One study suggests that QAnon has leveraged

hashtags like "Pizzagate," a QAnon theory of child sex abuse among the liberal elite, as

a weapon to band extremists together, forming paramilitary groups like the Proud Boys

and One Percenters (Bleakley, 2023).

General societal normative beliefs about the Republican Party suggest a party

focused on less government and more economic relief (Brandt & Spälti, 2018;

Levendusky, 2009). The focus on social issues and government regulation of these

issues reflect much of the conservative ideological narratives of the Trump Republicans'

focus on their elite has shifted (Klein, 2020; Levendusky, 2009). Results like this

perhaps explain why Representative Margorie Taylor Greene, R-GA, would actively use

the term Christian as her first descriptor under her name on her Twitter profile. Viewers

of her profile will attach that descriptor to her political identity. Members of Congress are

leveraging the Christian identity to further the us vs. them narratives; much of this can

be attributed to the birtherism movement created by Trump, who frequently referred to

Former President Obama as a Muslim (Klein, 2020). This pernicious polarization creates

the strong party loyalty witnessed among Republicans with Donald Trump at the helm

(Barber & Pope, 2019). Partisan identity is no longer just political as it is becoming a

personal identity, especially for those at the extremes of each party (Iyengar &

Westwood, 2015).

Before social media, the ability to openly state one's religious identity and attach

it to their political identity required specific channels and airtime from media sources. In

the modern world, Twitter has given platforms to Members of Congress who might not

have had similar opportunities to be seen nationally. Rep Greene's highly charged

rhetoric is easily engageable and enforceable due to social media's ability to allow

followers to engage and share (Grover et al., 2019). This clustering of individuals

created by social media is mainly responsible for the appearance of more echo

chambers online (Cinelli et al., 2021).

Figure 26

Democrat High Engagement

Figure 26 reveals how the perceived norms for Democratic elite, have slightly

change and shifted. The Democratic Members of Congress among the highest engaged

posts on Twitter, were focused on Economic and Government concerns with a network

score of .55 but this is only slight above Government and Social at .52 (See Figure 26).

This shift towards economics isn’t surprising as the narrative often repeated by Senator

Warren D-MA and Sanders I-VT is over student debt relief. One study found that MoC

who supported President Biden’s Student debt relief plan saw a political benefit and a

boost in their potential votes (SoRelle & Laws, 2023). SoRelle & Laws’ (2023) survey

showed that nearly 72% of respondents supported student debt relief plan, this included

33% who identified as Republican. The study suggest that failed “red wave” in the 2022

midterms indicated significant student voter turnout who supported the student debt

relief plan (SoRelle & Laws, 2023).

Research Question 3

For RQ3, the ENA web tool was again used, to evaluate not only former

President Trump and Bolsonaro, but also to examine the rhetoric espoused by their

followers (Marquart et al., 2021). By adding responses to the former presidents’ tweets

insights into the engagement and actions of their followers was added validation of the

perceived threat of violence made. For example, @realDonaldTrump tweeted “we’ll be

wild” these famous three words taken in the context of many meant to be there for

Donald Trump on January 6th, and help prevent the confirmation of the votes for

President Biden (Harton et al., 2022; Trump, 2020).

Additional studies have established that polarization wasn’t always occurring at

the social level, but predominantly in partisan identity politics manifested by the political

elite (Mason, 2015, 2016; J. L. McCoy & Somer, 2021). Mason (2015) suggest that

American’s are more attached to their political identity than ever and that even though

politically that nation is aligned, that identity politics are driving this animosity. The study

aimed to show how politically charged rhetoric demonstrating identity politics are

frequently leveraged by Members of Congress and other influential individuals on social

media. These narratives have had significant consequences and can be linked to events

like January 6 and January 8 (Bugs et al., 2023; Harton et al., 2022). There are several

other objective reports and academic studies pointing to social media’s influence over

coups, insurrections, and uprisings there are going to be more should these polarizing

individuals continue to produce content freely (Bugs et al., 2023;

Dwoskin, 2023; Ribeiro et al., 2017; Scott, 2023; Tønnesson et al., 2022).

The top-down theories of polarization have been linked to several studies

mentioned in this paper (Banda & Cluverius, 2018; J. L. McCoy & Somer, 2021). The

results of RQ2 and RQ3 provide sufficient evidence to display that there is

unquestionably a strong correlation between user engagement on Twitter with elite

polarizing rhetoric. RQ2 indicated that when politicians use inflammatory and polarizing

rhetoric, their engagement increased considerably. The least engaged members of

Congress were those who were closest to the center ideologically and used neutral and

unifying terms like “bipartisan” or “come together”.

Part of this can be identified as the user being inclined to participate in the

acrimonious dialogue. The MAD model presented by Brady et al. (2022) reveals the

propensity of human nature on social media to engage with content that is more likely to

instigate than to bring together. However, it is now considered to be common knowledge

among these social media giants how powerful the influence of conflict is on social

media (Rathje et al., 2021). The question that remains to be answers is: who is to blame

for the algorithm ranking polarizing content higher, is it the politician who takes the bait

and continues to feed the system or the creator of the system?

The findings from RQ3 revealed that there is cause to believe that the actions

taken on social media by former President Trump and former President Bolsonaro had

inspired insurrections in both Washington D.C in various ways. While Twitter provided a

tool for the rapid dissemination of information, it also became a rallying tool for

supporting Bolsonaro and Trump. Groups of insurrectionist were able to mobilize rapidly

by using hashtags and sharing the leaders tweets to band together and ultimately lead

insurrections (Dwoskin, 2023; Harton et al., 2022). Now deciding whether Bolsonaro,

Trump, or Twitter should be liable to the following actions is arguably a moot point since

the decision was made on an individual level by each insurrectionist. This does present

a new moral quandary however, one does pass a certain form of responsibility to both

the influencer and the one providing the platform, in this case Bolsonaro, Trump, and

Twitter.

Political leaders, especially MoC and former Presidents, have responsibility to

maintain and protect democratic values in order for healthy democracies to survive.

McCoy et al. (2018) demonstrate how Trump’s anti-establishment rhetoric help lead his

way to victory in the 2016 election. Trump’s narrative of being a victim of oppression

from the establishment resonates with his followers. They are exhausted by the

“establishment” and the messages of distrust of the government that Trump voices is

codified in their belief that they are being manipulated (J. L. McCoy & Somer, 2021).

Trump has so successfully designed his narrative around distrusting the government,

that even with four criminal trials facing him in the advent of the 2024 election, his poll

numbers far outpace his fellow Republican candidates (Best et al., 2023).

Implications for Practice and Scholarship

Even though academic Twitter API access has been revoked, it is still going to be

imperative for research on tweets to be conducted. This study revealed multiple

instances of polarization being amplified by Twitter. The results of such amplification

could be attributed to violent uprisings around the world (Bugs et al., 2023; Harton et al.,

2022; Tønnesson et al., 2022). The issue of polarization seems mute when looking at

the level of harm encountered from events triggered by social media; it creates the

necessity for academics, businesses, as well as politicians to take note and begin to

work together to formulate policies and strategies to mitigate any potential future

violence. Understanding polarization and how it affects being influenced is essential.

However, there is an immediate need to monitor and protect individuals from future

harm regardless of the perceived definition of the rhetoric being used by the political

elite.

Study Limitations

When this study began, Twitter had open API access to their platform; sadly, in

May of 2023, Elon Musk removed this access, forcing academics to either pay an

exorbitant fee for legacy data collection or use scraping tools like TwExportly (Calma,

2023; TWExportly, 2023; Twitter, 2023a). While a great tool, TwExportly (2023) can only

extract tweets by username and cannot extract comments on specific tweets. This

limited the study to reviewing tweets of specific accounts instead of the entire House of

Representatives and Senate. Additionally, retrieving legacy comments from the former

president’s account @realDonaldTrump was challenging because Elon Musk’s policies

removed the ability for Twitter accounts to see more than a certain number of tweets a

day (Calma, 2023). This way, when retrieving comments for @realDonaldTrump’s “will

be wild,” I could only retrieve 56 comments written before the January 6 uprising

(Trump, 2020). Comments made after January 6 on that post would not provide insight

into the question being asked in the study.

Another limitation is the assumption that the tweets are the true intention of the

Members of Congress or the former Presidents. Rouge aids, or even misinterpretations

of the tweets, happen regularly on social media. The speed at which people can post

and share on social media often means unintended posts can become viral and taken

out of context before an individual can remove the tweet. Politicians are held to a higher

standard and must exercise caution when using social media. Former President Donald

Trump may not intend to instigate the insurrection in the capitol on January 6; however,

as a leader with a highly influential voice, he is entirely responsible for anything he

tweets or says.

Defining the origin of the polarization seen on Twitter was one of the aims of this

study, and while the study provides some vital insights, there is still much-needed

research. There is a myriad of other social platforms that have gained significant

momentum. One journalist attributed Telegram to the January 8 uprising in Brazil. In the

United States, the use of the platform Parler had potentially a significant impact on the

events that took place on January 6 (Dwoskin, 2023, 2023; Harton et al., 2022). While

potentially less significant and widespread than Twitter, Truth Social, Gettr, Telegram,

and many more might be more important to monitor as they tend to cater to fringe

groups with potentially more polarizing and hostile views (Fischer, 2022).

Short-form video is the new format that has taken social media by storm as the

platform TikTok has rocketed toward controversial success. A study on adolescent

TikTok users in China found that TikTok is addictive and highly influential (Qin et al.,

2022). One positive found that it was an effective and rapid way of informing the youth.

However, this could have negative consequences (Qin et al., 2022). This study was

limited to text-based resources for monitoring and evaluating polarization online; future

studies should consider examining narrative dialogue in the short form video social

platforms for issues surrounding polarization.

Recommendations for Practice and Future Research

While this study is a comprehensive review of how Members of Congress and

former leaders are becoming more polarized in their use of Twitter, the study opens

further research into the specifics of the types of polarization. Pernicious polarization is

the driving force behind the elite polarization. Members of Congress and former

politicians turn to their social media feeds to espouse vitriol dialogue attacking the other

side. The study gave insight into Twitter's algorithm's favoritism towards sharing the

polarized dialogues mentioned in the coding sets of RQ1, RQ2, and RQ3 and showed

how these tweets gained higher views, likes, and shares. However, the study fails to

determine if these individuals are the compelling force behind this phenomenon. The

MAD model introduces the concept of moral contagion in social media being driven by

group identity and social reinforcement. However, additional research is needed to gain

further insight into the external consequences of these types of interactions (Brady et

al., 2022).

RQ3 does provide some foundation for linking former President Trump and

former President Bolsonaro's social media accounts; however, much more needs to be

studied in this space as humans are getting more and more of their information from

social media. The American Academy of Pediatrics contained a study that revealed that

children following specific social media influencers were more likely to change their diets

based on influencers' actions (Coates et al., 2019). There is undoubtedly a need for

greater insight into this as information on social media is so quickly spread and shared

that disinformation has in the past and will continue to elicit serious harm and, in some

cases, death (Tønnesson et al., 2022).

I believe that social media companies, specifically Twitter and Facebook, need to

continue to their use of flagging content, however, rather than shadow banning it,

provide an explanation for why content is being flagged. This would help restore trust in

the context of platforms and help users understand potential bias. For example, the

World Health Organization (WHO) recommends that Twitter, Facebook, and other social

media companies limit certain information regarding COVID-19 on social media

(Cosentino, 2023, pp. 21-22). Some of the information recommended to be blocked by

the WHO was specifically linking Wuhan to the virus for fear of reprisal against Asian

communities around the world (Cosentino, 2023, p. 22). After nearly three years of

social media companies blocking information regarding the potential lab leak at the

Wuhan Institute, a Senate hearing on March 3rd, 2023, revealed that the lab leak was in

fact a potential cause for the COVID-19 epidemic (Committee on Oversight and

Accountability, 2023).

As mentioned in the limitation, further research into the new social media

platforms like Telegram, Truth Social, and Gettr should be done to understand how

these new platforms might create or inhibit pernicious polarization. The study is also

limited to the United States and briefly looks at Brazil and former President Jair

Bolsonaro's failed attempt at reelection. Populist leaders have run successful

campaigns around the world and have leveraged social media to their advantage.

Understanding what about social media is attractive to the populist follower is highly

important. While many might assume it is echo chambers that power the campaign of

former President Donald Trump, the reality is it his ability to operate outside of that

vacuum and create a co-evolved media strategy has enabled him to defy conventional

political strategies (Postill, 2018). Hitler and the Nazi party were able to defy the

establishment's role in what was traditional media at the time by leveraging radio

broadcasts in the 1930s and gaining tremendous power through a newly invented media

format (Adena et al., 2015).

Closing Comments

While many are turning to the government for regulation of social media

companies, the challenges are massive as government regulation moves at a much

slower pace than technology. For example, it took the world wide web seven years to

reach 100 million users, TikTok nine months, ChatGPT two months, and Threads hit this

milestone in just five days (granted it leveraged ownership of the existing platform

Instagram to migrate its followers; Rao, 2023). If platforms like Instagram (owned by

Meta) start getting regulated, the laws applying to them could influence their decision to

spin off new apps that might be immune from the legislation. Meta's creation of Threads

was a response to the growing frustrations of Twitter users, as Elon Musk's takeover

has brought some drastic unfavorable changes (Conger & Frenkle, 2022; Rao, 2023).

This brings a challenge for both regulators and business owners when deciding who is

responsible for managing content created on social media. The answer is still unclear;

as mentioned earlier, Section 230 has provided many companies with significant leeway

regarding the content created and posted on their respective platforms (Cramer, 2020,

p. 230; D.O.J., n.d.). This needs to change as there clearly is an influence created by

the platform’s use of their proprietary algorithms as certain content, is given favoritism,

as noted by the results of RQ2: Figure 15.

Former President Donald Trump, in a social media post, had called out Congress

for their inability to make significant changes to Section 230 and even signed an

executive order in the past to try and block it (which failed in court) (Siripurapu, 2020).

Politicians in Utah and Arkansas have passed regulations in their own state legislatures

targeted at social media with the intention of protecting children. However, some believe

the law's creators intended to harm young adolescents by forcing them to share their

private profiles with their parents (Murphy, 2023). Additionally, there are cases in Texas

and Florida heading toward the Supreme Court that are suing Meta (Facebook), Twitter,

and Alphabet (Google) for blocking first amendment rights (Chung, 2023). These

lawsuits are controversial as they target social companies for blocking information being

shared on the platforms. The COVID-19 Epidemic put more pressure on social media

companies to be monitoring their content than ever before for fear of misinformation

being tied back to their platform (Gisondi et al., 2022). This pressure revealed the

weakness of social platforms and uncovered their inability to protect their users from

misinformation (Gisondi et al., 2022). Reiterating the Simon et al. (2020) study revealed

that while only 20% of the misinformation regarding COVID-19 was created by top-down

misinformation (politicians, celebrities, and public figures), it accounts for 69% of the

total social engagement. This study reveals that, like COVID-19, the most inflammatory

and polarizing politicians are typically the most shared on Twitter. The problem is that

these politicians rarely offer solutions that provide sustainable options for societies in

general.

As of August 29th, 2023, US Special Operations and Command (USSOCCOM)

entered an agreement with AI technology company Accrete to use their software

ArgusTM to work on detecting disinformation in real time (Accrete, Inc., 2023). The

intention of USSCOM use of ArgusTM is to protect the American public from

disinformation campaigns from foreign actors, however, some speculate that this might

be more problematic as there is a potential for misuse and increase censoring limiting

free speech (US Special Operations Command Will Deploy., 2023). I believe while AI

should be used for monitoring and detecting mis/disinformation it ultimately should

provide an interstitial, or warning that the content is potentially false and provide

reasonings for the platform’s decision to create the warning. This way free speech is

protected, and consumer trust can be reestablished in the long term with social media

platforms.

The biggest problem America and the world are facing today is that elite

polarization (specifically when it is pernicious) is creating pressure for partisan identity

that extends beyond political life and into personal life (Barber & Pope, 2019; Klein,

2020; J. L. McCoy & Somer, 2021; Noel, 2016). When I was a child, I was told that it

was rude to ask someone you voted for; society generally regarded this as private

information. Apart from that might have been my upper-middle-class upbringing in an

outwardly liberal community but secretly highly conservative fiscally. The conventions of

keeping personal voting behavior private changed after the 2000 elections when Bush

defeated Al Gore. What was surprising was that while Gore won the popular vote, he did

it with less than 700 counties; Hillary Clinton in 2016 won the popular vote but lost the

electoral college with less than 500 counties (more than 1000 counties less than Bill

Clinton had won with few votes in 1990; Klein, 2020, p. 66).

The urban-rural divide narrative has long since been a part of party politics and is

extending into daily life more and more as social media has empowered individuals to

connect in echo chambers (Sasahara et al., 2021). While social media has the ability to

connect users from across the globe, the tendency of users, albeit a function of human

nature as well as the respective platform, is one where the user engages in homophily

and strengthens personal confirmation bias (Sasahara et al., 2021). This trend toward

homophily and echo chambers is a result of human nature as well as the idea of political

acrophily, the tendency of individuals to associate with others who also share extreme

political views (Goldenberg et al., 2023). This acrophily, as described by Goldenberg et

al. (2023), drives the partisan sorting that is dividing Americans. The famous Pauline

Kael's quote about not knowing anyone who voted for Nixon is most likely the case for

people who voted for opponents of other highly liberal or conservative candidates

(Brandt & Spälti, 2018). Brown & Enos (2021) study revealed that partisan isolation is

distinct from even racial and ethnic segregation and extends across all of the United

States. This separation creates and enables the acrophily encountered in the. Twitter

feeds I studied extend into the media (Goldenberg et al., 2023).

How did Americans and possibly the world i.e. Hungry, Turkey, Brazil, and

Venezuela become so acrimonious and unharmonized (McCoy et al., 2021)? Social

media has provided a platform for both hate and love, and sadly these platforms make

more money over acrimonious dialogue than shared pictures of grandchildren (Rathje et

al., 2021; Riemer & Peter, 2021). Riemer & Peter (2021) define algorithmic audiencing

as a tool created by social platforms to generate feeds that produce profits over free

speech. Part of this audiencing can be attributed to the acrophily witnessed in Brazil and

the United States. If the least shared members of Congress had the same views as

@realDonaldTrump, would January 6 have occurred? There is no way to answer this

with certainty, as some believe that cross-party exposure increases polarization (Bail et

al., 2018).

The ultimate problem this dissertation seeks to address is preventing pernicious

social media rhetoric from turning into violent uprisings. Many parties share

responsibility regarding what happened on January 6 and January 8. Political elites,

Members of Congress, politicians, and world leaders can influence and espouse rhetoric

on social platforms to a like that has never been seen before in history. The speed and

virality at which one can disseminate messages to the world becomes greater daily.

Fixing this problem requires many different parties to work together to provide a solution

that prevents further harm to individuals.

I argue that all parties involved have a responsibility to preserve democracy and

reduce hard. Platforms like Twitter, Facebook, TikTok, and many others are going to be

tasked with even more significant challenges of monitoring content as AI has given rise

to the deep fake. This technology is particularly worrisome as individuals will have

difficulty discerning between reality and fiction. Political leaders, politicians, and

influencers are also responsible for maintaining a moral high ground. Their ability to

transform their followers into insurrectionists is concerning. In the end, however, it is the

responsibility of the individual to learn and be mindful of the ideological minefield the

world has become. Individuals like Alex Jones create and reiterate narratives like lizard

people to purposely dehumanize his opponents (Van den Bulck & Hyzen, 2020). This

dehumanization is dangerous as individuals believe that compromise is impossible. The

Students also viewed