University of Southern California Dissertations and Theses

Three text-based approaches to the evolution of political values and attitudes

(USC Thesis Other)

Three text-based approaches to the evolution of political values and attitudes

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content THREE TEXT-BASED APPROACHES TO THE EVOLUTION OF POLITICAL VALUES AND
ATTITUDES
by
Meiqing Zhang
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(COMMUNICATION)
December 2023
Copyright 2023 Meiqing Zhang

Dedication
To my dear grandmother Wang Zeshu, "tree", your own troubled hero.
ii

Acknowledgements
I would like to express my deepest appreciation to my committee. My advisors Professor Patricia Riley
and Professor Emilio Ferrara supported me through my doctoral training and research, including some
unconventional learning paths across different USC departments. They were always there to help me
find resources, give valuable feedback, and provide a comfortable space for me to develop and be myself.
They have profound beliefs in my abilities and exposed me to excellent opportunities. I am also deeply
indebted to my mentors and committee members Professor Pablo Barberá and Professor Thomas Hollihan.
They generously advised my research, offered careful and insightful comments, and had been wonderful
instructors whose courses were instrumental in the development of my dissertation. I am extremely lucky
to have all of them on my committee, who are also genuinely kind persons.
I would also like to extend my sincere thanks to Professor Larry Gross at Annenberg, Professor Dennis
Chong at POIR, and Professor Mohammad Soleymani at ISI, Professor Sergey Lototsky, whose teaching
and advice at different points of my doctoral studies greatly inspired my dissertation projects and vision
for future research.
I am extremely grateful to the research grants I received from Annenberg summer and dissertation
research funds. The overall financial support for research and professional development from my department allowed me to acquire desirable academic and computing resources with ease.
The completion of my second study was only possible thanks to the contribution from my colleagues.
I am indebted to Julie Jiang, who helped run her signature ideology classifiers on the dataset of this study,
iii

which is crucial to the analysis. Many thanks to Emily Chen, whose meticulous Twitter data collection
and generous support on this dataset set the cornerstone of this project.
I am grateful to my friends and cohort members at Annenberg School: Cerianne Robertson, Junyi Lv,
Ana Howe Bukowski, Mingxuan Liu, Jessica Hatrick, Amber Lynn Scott, Herbert Chang, Ally Arrieta,
Simogne Hudson, Lichen Zhen, Nan Yuanfeixue, Jack Tang, Ashley Phelps, Sarah Wise, Suk Young Choi,
as well as alumni who eased me into this community in the early stage of my studies: Ming Curran, David
Jeong, Calvin Kim, and James Lee. The list of kind, supportive people whom I have had the pleasure
of encountering at different stages of my program goes on. Outside of Annenberg, I wish to extend my
gratitude to all the lab members and alumni of MINDS (Emilio’s lab) at ISI. It was my great pleasure to
cross paths with you all - brilliant scholars and collegial colleagues. Special thanks to professors who
helped me through my past academic journey: Professor Dan Slater, Professor David Bachman, Professor
Gary Hamilton, Dr. Evelyn Engesser, Professor Gordon Arlen, as well as my friends Jackie, Jade, Goutham,
Jacob, Bo, Sheep, among many, many other wonderful people in my life, even though I am not able to list
all of you.
No words would suffice to express my gratitude to my family, in particular my parents and my partner
Rom. Without your love and support, sometimes at the expense of your own needs and comfort, this
journey would not have been as rewarding.
Finally, I owe my dissertation and research paths in media and politics to those who fought for freedom,
equality, and democratic ideals courageously.
iv

Table of Contents
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
Chapter1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Chapter2: The semantic evolution of liberty and equality . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Tensions between liberty and equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Change and continuity of the liberal consensus . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Incremental movement towards equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Expanding the definition of equality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.7 Studying semantic change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.8 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.9 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.9.1 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.10 Analytic strategies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.11 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.11.1 Face validity checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.11.2 Results for H1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.11.3 Results for H2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.12 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.13 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
Chapter3: Moralization of political discourse on Twitter in the lead-up to the US 2020 election . . 29
3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Stage 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.1 Process of moralization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.2 Moralization of political attitudes and communication . . . . . . . . . . . . . . . . 32
3.3.3 The case of US 2020 election campaign trail . . . . . . . . . . . . . . . . . . . . . . 33
v

3.3.4 Data and measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.4.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3.4.2 Detection of moral expression in tweets . . . . . . . . . . . . . . . . . . 35
3.3.4.3 Subtopics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.4.4 User ideology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.4.5 Misinformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3.5 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.4 Stage 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.5 The role of affect and emotions in moral judgments . . . . . . . . . . . . . . . . . . . . . . 43
3.6 Affective polarization and moralistic politics . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.6.1 Data and Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6.1.1 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6.1.2 Negative partisan affect . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6.1.3 Measure of specific moral dimensions . . . . . . . . . . . . . . . . . . . . 50
3.6.2 Empirical strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.6.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Chapter4: Partisan bias in inflation news narratives . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Media and political conditioning of economic assessments . . . . . . . . . . . . . . . . . . 65
4.4 Partisan cues, ideological anchors, and political activation . . . . . . . . . . . . . . . . . . 67
4.5 Priming and framing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.6 Cable television as a venue to study inflation narratives . . . . . . . . . . . . . . . . . . . . 71
4.7 Data and measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7.1 Cable television news dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7.2 Inflation data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.7.3 Inflation expectations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.8 Analytic strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.8.1 Detection of narratives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.9 News consumption and inflation expectations . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.10 Cable news narratives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.11 Comparison with inflation narratives from surveys . . . . . . . . . . . . . . . . . . . . . . 86
4.12 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
Chapter5: Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
vi

List of Tables
2.1 Most similar terms to "liberty" across decades . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Most similar terms to "equality" across decades . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3 Dictionary of democratic values and progress . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.4 Dictionary of individualism and egalitarianism . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1 Model performance – prediction of moral content . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Daily proportion of moral content in the first and last month . . . . . . . . . . . . . . . . . 39
3.3 Sample tweets from liberals and conservatives that were rated high on the moral
dimensions of cheating and degradation. Stop-words were removed. . . . . . . . . . . . . . 54
3.4 Regression table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.1 Logistic regression table of partisanship and the likelihood of hearing favorable and
unfavorable news about business conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.2 Weighted least squares regression table of inflation expectations . . . . . . . . . . . . . . . 79
4.3 Top narratives of inflation drivers discussed by cable news networks. Verbs and entities
presented in root forms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4 Top causes of inflation for Republicans and Democrats. Source: YouGov . . . . . . . . . . . 83
4.5 Sample statements about federal government spending. Source: LexisNexis news transcripts 84
4.6 Sample statements about corporate greed. Source: LexisNexis news transcripts . . . . . . . 85
4.7 Top news partisans heard about business conditions . . . . . . . . . . . . . . . . . . . . . . 89
vii

List of Figures
2.1 Similarity between equality and groups of words that indicate black identity. Y-axis
represents cosine similarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Similarity between equality and groups of words that indicate female identity. Y-axis
represents cosine similarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Average similarity between liberty (red)/equality (blue) and a dictionary of democratic
values and progress. Error bars represent 95% confidence intervals derived from
bootstrapping. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Similarity between equality and individualism (light blue), egalitarianism (navy). Error
bars represent 95% confidence interval derived from bootstrapping. Y-axis represents
cosine similarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Similarity between liberty and individualism (light blue), egalitarianism (navy). Error bars
indicate 95% confidence interval derived from bootstrapping. Y-axis represents cosine
similarity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6 Similarity between equality/liberty and “social”, “freedom” . . . . . . . . . . . . . . . . . . 28
3.1 Bootstrapped daily average proportion of moralized content under the topics of (a)
COVID-19, (b) mail-in voting and (c) voter fraud compared to the entire US 2020 election
corpus. Shaded areas represent 95% confidence intervals. The "overall" trend lines in blue
represent the average proportion of moralized content calculated from the entire corpus. . 40
3.2 (a) Proportion of moralized content by month. Shaded areas indicate 95% confidence
intervals based on daily average. The brown line represents the topic of mail-in voting,
blue COVID-19, and green voter fraud. Red is average of the entire corpus (baseline).
(b) Difference in moral sentiment between misinformation and reliable information. The
y-axis displays the daily value of δ (3.1) over the study period . . . . . . . . . . . . . . . . 41
3.3 Monthly density distribution of percent moralized content by user ideology and topic . . . 42
3.4 Distribution of five moral dimensions in voter fraud-related topic . . . . . . . . . . . . . . 53
3.5 Distribution of five moral dimensions in voter fraud-related topic by month . . . . . . . . 53
viii

3.6 Monthly distribution of cheating and degradation by ideology . . . . . . . . . . . . . . . . 54
3.7 Monthly Distribution of negative partisan affect . . . . . . . . . . . . . . . . . . . . . . . . 54
3.8 Average marginal effects of negative partisan affect on propensity for moral expression.
Left panel displays the average marginal effects of negative affect towards the opposing
candidate for liberals and conservatives respectively. Right panel displays the average
marginal effects of negative affect towards the opposing party. . . . . . . . . . . . . . . . . 56
3.9 Average marginal effects of negative partisan affect on propensity for moral expression,
varying dependent variables. Error bars of different colors represent the average marginal
effects of negative partisan affect on different measures of the dependent variable. Moral
was the same measure as 3.8. Cheating and degradation measure the respective moral
categories, which are the most prominent ones in this discourse domain. . . . . . . . . . . 57
3.10 Average marginal effects of negative affect towards the opposing party on moral
expression over three time periods. The nine-month study period was divided into three
intervals: T1 covers the first three months (March, April and May); T2 covers June to
August; T3 covers September to November. . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.11 Average marginal effects of negative affect towards the opposing party on expression of
degradation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1 Average 12-month inflation expectations by party identification, January 2020 to December
2022. Grey dashed line marks November 2020 when Biden was named the president-elect,
on track to replace Trump in the White House. Shaded areas indicate standard errors.
Source: Surveys of Consumers – University of Michigan. . . . . . . . . . . . . . . . . . . . 65
4.2 Example agent-verb-patient triplet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.3 Narrative extraction pipeline using RELATIO. Flowchart developed based on Sipka et al.
(2022) and Ash et al. (2023) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4 Proportions of Democrats, Republicans and Independents/unknowns who heard favorable
and unfavorable news about business conditions before and after the 2020 election. Grey
dashed line indicate the month of 2020-11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Counts of inflation-related transcripts and "inflation" mentions by cable news networks.
Grey dashed is a time series of year-over-year inflation rate. . . . . . . . . . . . . . . . . . 83
4.6 Top entities three cable news networks’ inflation coverage . . . . . . . . . . . . . . . . . . 86
ix

Abstract
This dissertation consists of three independent studies that explore and investigate different development
processes of political values, emotions, and cognition. It draws on literature from communication, political
science, and psychology to motivate research that engages with how people have become ideologically,
affectively and cognitively divided. The first study entertains the hypothesis of the rising status of equality
as a core democratic value over recent decades. The tensions between liberty and equality constitute an
important aspect of the ideological polarization in the contemporary United States. I tracked the evolving
meanings of liberty and equality using the NLP method of diachronic word embeddings. The meaning of
equality has gradually risen in semantic proximity to core democratic values relative to liberty and moved
towards an egalitarian vision of the concept over the past six decades. The second study examines how
social media political discourse became moralized over the emerging issues of voting-by-mail and voter
fraud during the US 2020 presidential election campaign. Using a BERT-based model and a large-scale
dataset, I detected synchronized temporal patterns of moral expression between liberals and conservatives
with regard to voter fraud discussions, pointing to the role of partisanship in moral amplification. As it
drew close to the election, negative partisan affect and moralized expression showed an increasingly positive association. The last study examines the widening partisan divide in inflation expectations. It presents
evidence that the tone of media coverage individuals received moderates the effects of party identification
on inflation expectations. In a follow-up narrative analysis of cable news coverage, I discuss how partisan
x

media narratives could have shaped people’s inflation assessments. The rising phenomenon of partisanbased economic reasoning complicates the information factor of economic voting and may cause people
to vote in accordance to their partisan views rather than to reflect their own. This dissertation experiments with a variety of NLP methodologies applied in computer science and social sciences and explores
the promise and pitfalls of addressing political communication research inquiries through computational
content analysis.
xi

Chapter 1
Introduction
ch:introduction
Political polarization in the United States raised extensive academic, journalistic, and societal attention
in recent decades thanks to how it endures and self-reinforces in a political system once characterized
by gravity towards moderation and the political center (Pierson & Schickler, 2020). Elite polarization has
deepened since the latter part of the twentieth century, as the intra-party ideological heterogeneity has
been shrinking and the inter-party distance increasing (Fiorina, 2017; Fiorina et al., 2011; Levendusky,
2009; Noel, 2014). The Democratic and Republican Party have become ideologically coherent and distinct
political organizations (Levendusky, 2009; Noel, 2014), cross-party voting in Congress less common, and
citizens better able to see the differences between the two parties. Evidence of polarization is particularly
strong among the politically engaged class (Abramowitz & Saunders, 2008), including politicians, donors,
party and issue activists and partisan media commentators, who constitutes a minority population but
exerts an outsized influence on political discourse.
Although conclusions on the nature and extent of mass polarization are mixed, voters have been effectively sorted into these two ideologically polarized parties (Baldassarri & Gelman, 2008; Fiorina, 2017;
Levendusky, 2009; Noel, 2014), turning partisanship into a "mega identity" (Mason, 2018; Pierson & Schickler, 2020). It erodes the cross-cutting social ties and differences (Mason, 2018) crucial for robust political
pluralism and democratic stability (Baldassarri & Gelman, 2008; Dahl, 2005; Lipset et al., 1960; Lipset &
1

Rokkan, 1967; Mason, 2018). Many studies on political polarization contend that the division between
the Democratic and Republican Party today represents all dimensions of political conflict, from economic,
racial to religious cleavages (see for example Abramowitz, 2018; Levendusky, 2009). Political opinions
of the general public, despite displaying less ideological coherence, were increasingly aligned with party
elites along multiple lines of political cleavages (Baldassarri & Gelman, 2008; Barber & Pope, 2019; Layman
et al., 2006). They were less radicalized than sorted along party lines (Baldassarri & Gelman, 2008).
A confluence of institutional, social factors and historical contingencies has shaped the polarized party
system as we know it today. It started from a critical party realignment on the issue of race after the civil
rights legislation (Carmines & Stimson, 2020) and solidified under the a series of institutional transformations that nationalized politics and policy making, changed the incentive structures of interest groups,
undermined the autonomy of state parties, and reshaped media landscape (Pierson & Schickler, 2020).
This dissertation also concerns the evolution of political polarization in the US context, but different
from an institutional perspective, it explores how values, emotions and attitudes evolve through three
separate cases on different timelines.
These three essays, although each has its distinct study background, explore different facets of value
and opinion change that hold implications for political polarization. They are also an experiment of computational content analysis, and each study applied a different NLP technique to approach its inquiries.
Chapter 2 presents the first study. It addresses a potential shift in democratic priorities over a long historical time span. I tracked the semantic change of two crucial democratic concepts, liberty and equality,
over two centuries from the early 19th century to the 2010s using diachronic word embeddings. Specifically, we ask these questions: is equality rising as a more important democratic value relative to liberty
under the influence of American progressives? Has the meaning of equality gradually conveyed an egalitarian vision of the concept relative to an individualistic vision? The research problem is motivated by the
observations from some scholars that the longstanding liberal consensus shared by the better-educated
2

Americans appears to be subtly drifting apart, with liberals moving to prioritize the value of equality and
equity over established civil libertarian norms (Chong et al., 2022). The ambition of studying a change at
the political cultural level can hardly fit within the scope of one study. Such value shifts are difficult to
detect, define, quantify and validate. Nonetheless, it is worth approaching with research efforts, because
the tensions between freedom and equality stealthily underlie many politically polarizing issues in the
contemporary era. An evolving and expanded vision of equality and its challenge to the liberal consensus
are deeply involved in the various debates from cancel culture, identity politics, to economic inequalities.
Chapter 3 brings the focus back to the contemporary age and zooms in on the dynamics of opinion
conflict within the short span of a presidential campaign. In contrast to the previous study’s subject matter
of underlying democratic values, this essay examines online expressions around concrete, novel issues that
arose during the US 2020 elections. Using a large-scale Twitter dataset, we looked at whether and how the
topics of mail-in voting, covid-19 and voter fraud became moralized as the 2020 election drew close. In
a decentralized and loosely regulated social media environment where opinions, incivility and emotions
abound, negative partisan affect and moral sentiment feed off each other, reinforcing affective polarization.
Negative preconceptions of the opposing party’s moral characters make conspiracy theories easier to sell,
further intensifying partisan animosity. We trained a BERT-based model to detect moral sentiment.
Chapter 4 shifts the study context from the modern digital information environment to traditional
institutional media. It takes an interest in the growing partisan differences in inflation and economic
assessments, which have fewer reasons to diverge due to its reliance on hard facts and professional opinions rather than personal morals and preferences. However, partisans hear different economic news, and
what they hear makes an impact on their short-run inflation expectations. It shows that, more than the
relatively stable partisan identification, over-time partisan information consumption also plays a role in
creating partisan differences in inflation expectations. I analyzed three years of mainstream cable news
content on inflation through narrative extraction and showed how partisan media presented ideologically
3

aligned narratives, which can be persistently recycled to activate the interpretive frames of their target
audience. Although conflict over cultural issues often seems to be the harder ones to bridge since they
involve morality, religion, and identity – factors not to compromise on. This study illustrates that the
interpretations of economic issues, driven by narratives rather than substantive information, can also live
in different worlds of truths.
Chapter 5 closes this dissertation with concluding remarks.
4

Chapter 2
The semantic evolution of liberty and equality
ch:lib_equ
2.1 Abstract
This paper traces the semantic co-evolution of liberty and equality from the early 19th century to the 2010s
using the NLP method of word embeddings. The relationship between these two fundamental democratic
principles is an important aspect of the value conflicts underlying many of the emerging political divisions
in the present age. I took a genealogy approach to study the change of political values through the examination of value-loaded concepts in their respective contexts. Diachronic word embedding models were
trained on a large English corpus over the last two centuries. By studying the evolving meanings of liberty
and equality, I argue that equality is rising on the democratic value hierarchy relative to liberty, meanwhile
a more egalitarian vision of equality has overtaken an individualistic vision over the past several decades.
The implications of a potential reordering of democratic value priorities are a vehicle to understand the
contemporary ideological polarization.
5

2.2 Introduction
A curious value shift seems to be brewing among the better-educated citizens in the United States. Although much remains to be defined, both academic and journalistic endeavors have uncovered a somewhat drastic reversal of public attitudes toward free speech and civil libertarian norms. In a manifestation
of this change, many liberals, once the staunchest supporters of free speech and due process from the
establishment of the ACLU during World War I throughout the civil rights movement (Walker, 1999), are
now found to be reconsidering the boundaries of permissible speech (Chong et al., 2022) and whether free
speech is the means to serve progressive ends (Seidman, 2018). Meanwhile, conservatives have increasingly invoked “freedom of speech” to fight the progressive agenda in court cases and public discourse. This
opposition is reflected in issues such as the right of merchants to reject providing services to gay couples,
or opposition to legal requirements to provide abortion information (Liptak, 2018). Through a recent national survey, Chong and collaborators found that liberal ideology and educational attainment were more
likely to predict intolerance of offensive speech, in particular speech that was believed to undermine the
rights and dignity of minorities. This alignment is in sharp contrast to evidence from the 1950s and 1970s,
when liberal orientation and college education were consistently associated with greater tolerance of objectionable speech in various forms, including the assemblies of Nazi groups that run counter to liberal
ideals (McClosky & Brill, 1983). Notably, support for racial equality and egalitarian values contributed to
this revisionist attitude towards freedom of expression (Chong et al., 2022). This relationship embodies an
increasingly institutionalized liberal stance that prioritizes the protection of historically underprivileged
communities from harm over the deeply entrenched value of individual and civil liberty. A typical school
of thought in this camp believes, for example, that racist speech shall not be dignified as an idea worth
reasoning with, and that “free speech” is weaponized to shield speakers of racist speech from criticism
(Titley, 2020). This public sentiment arguably signals a shift in attitudes towards one of the fundamental
values in American democracy. The right of free speech, once believed to give voice to the voiceless, is
6

now considered by many as a means to protect the powerful, whose exercise of free speech could victimize
the powerless (Liptak, 2018).
Freedom of expression is a cornerstone of the American political tradition “conceived in liberty” (Rothbard, 2011) and intimately connected to other constitutive elements of the American conception of liberty,
such as property rights, individual liberty, and limited government intervention (Scott, 2014; Seidman,
2018). Liberals’ reevaluation of an unconditional commitment to free expression communicates signs of
rising tensions between liberty and equality in modern American liberalism. Through the inclusion of
democratic claims from historically marginalized groups, modern liberalism expanded its vision of equality and gained a shade of collectivist sympathy that contradicts several libertarian doctrines of classical
liberalism (Shils, 2019; Starr, 2007). For modern liberals, is equality rising on the democratic value hierarchy relative to liberty, which encompass traditional ideals from freedom of speech, economic freedom, to
civil libertarianism?
This paper engages with a hypothesis of rearranged democratic value priorities by taking a retrospective look at the evolving political status and interpretations of two core democratic principles: liberty and
equality.
I examine the evolving concepts of liberty and equality in American English through the natural language processing approach of word embeddings using historical text corpora. Studying cultural change
through the lens of semantic change has been practiced in conceptual history, genealogy, computational
linguistics, among others. In search for the evolving meanings of principal concepts deployed in political
and social thought, conceptual historians discover the values, interests, and preferences being shared or
contested in their respective contexts over time (Richter, 1995a). In methodologically different veins, initiatives were also taken in NLP that revealed cultural change through the study of time-variant semantics
(Hamilton, Leskovec, and Jurafsky 2016; Giulianelli, Del Tredici, and Fernández 2020; Schlechtweg et al.
7

2019). This paper follows these footsteps and tracks the semantic co-evolution of the political concepts of
liberty and equality using diachronic word embeddings.
The upcoming sections are structured as follows: the literature review opens with a discussion about
the tensions between liberty and equality in a liberal-democratic setting, and then moves on to argue why
an exceptionally liberty-oriented political tradition may have been developing a growing taste for equality
in the US context. The next section discusses the methodological grounds for studying the relationship
between liberty and equality through word meanings. Finally, data, method, and results were presented
in order before concluding remarks.
2.3 Tensions between liberty and equality
I choose the contrast of liberty versus equality as an entry point to examine a hypothesized value change,
because they are both first-order liberal-democratic ideals yet not always complementary, and their ebbs
and flows are relevant to the value shift this study considers.
Although liberty and equality are both cherished liberal-democratic ideals, tensions arise from their
incompatible aspects and scenarios that heighten their incompatibility. For instance, a laissez-faire political economy can aggravate inequality, and an equality-of-outcome political agenda might require greater
restrictions on individual liberty and private interest (Steiner, 2017). If economic and political inequality
stemming from a free-market economy becomes a pressing issue in a specific context, democratic demands
to sacrifice certain aspects of economic liberty can be expected to grow. Similarly in the phenomenon discussed in the opening of this paper, when American progressives hold that protecting minority groups
from speech-caused harm takes precedence over the creed of a right to free speech, they are willing to
censor such speech at the cost of individual freedom and the risk of inadvertently removing legitimate
debate (Kemp & Ekins, 2021).
8

The debate resulting from the trade-offs between liberty or equality is a timeless story. The libertyoriented narrative originated with Locke and the equality-oriented narrative traced back to Rousseau,
constitute an enduring intellectual conversation that sustained the last three centuries and remains vital
today (Capaldi & Lloyd, 2016). Capaldi and Llyod called them narratives as opposed to arguments, because narratives tell a story and structure our thought and thereby can survive repeated refutations over
history. The liberty-equality divide consists of two successful and evolving narratives taken on by a series
of canonical authors, deeply embedded in Western political discourse and still relevant to today’s public
policy debates (Capaldi & Lloyd, 2016).
It was famously argued that American political culture was predominantly rooted in the Lockean liberal tradition (Hartz, 1955) and received little influence from Rousseau (Schachner, 1951)
1
. Hartz (1955)
argued that America was built on the Lockean liberal consensus of property rights, individual freedom,
free markets, free trade, and democracy, because it does not have a feudal history and the equivalent of the
traditional European conservatism. Furthermore, the American revolution and its struggle for the meaning
and reality of political liberty profoundly baked a Lockean, individualistic vision of liberalism into the nation’s founding political culture and institutions (Dworetz, 1989; Klosko, 2017; Lipset, 2013; Rauser, 1998).
The prevailing Lockean liberal consensus tended to assimilate or marginalize ideological antagonists, especially European socialism (Kann, 1980). It is the series of conditions formative for a unique American
character that makes the United States “exceptional” (Hartz, 1955; Lipset, 2019).
1Although some of the Founding Fathers, such as Benjamin Franklin and Thomas Jefferson, shares certain paralles with
Rousseau and the French philosophes aside from Locke: https://iep.utm.edu/american-enlightenment-thought/, also
https://almostchosenpeople.wordpress.com/2010/01/07/jefferson-and-rousseau-on-democracy/
9

2.4 Change and continuity of the liberal consensus
Some earlier literature contended that the basic American values derived from the nation’s founding moments were stable (Lipset, 1964), and found that freedom was persistently high on American value priorities (Rokeach & Ball-Rokeach, 1989). Individual freedom, political and civil liberties against state authority
constitute a big part of the nation’s founding identity, shared by liberals and conservatives back in time,
referred to as “constitutional liberalism” by Starr (2007). Values forged into political institutions and norms
are expected to stick and hold greater resilience against change. For example, paradoxically, the individualistic Lockean beliefs had been consistently used to justify welfare state policies by American political
leaders (Klosko, 2017). However, conditions do exist that would set in motion changes to the distinctive
constitutional liberalism over a long course of history.
Alexis De Tocqueville (2006) in his Democracy in America argued that citizens in democratic countries
would always love equality more than liberty. He explained that “the advantages which freedom brings
are only shown by length of time; and it is always easy to mistake the cause in which they originate. The
advantages of equality are instantaneous, and they may constantly be traced from their source. Political
liberty bestows exalted pleasures, from time to time, upon a certain number of citizens. Equality every
day confers a number of small enjoyments on every man.” On the contrary, the ills brought by freedom
are imminent and felt by all, whereas the negatives brought by extreme equality are slowly revealed in a
process of creeping normality (De Tocqueville, 2006).
These century-old observations seem to stand the test of time and apply to the United States, as rounds
of progressive movements since the 20th century spurred greater acceptance of egalitarian values. Even
right-wing populism driven by dissatisfaction with inequality or senses of unfairness had gained momentum (Hochschild, 2018; Lieberman et al., 2019; Norris & Inglehart, 2019; Rodriguez-Pose et al., 2021).
One indicator of the increased cultural weight of equality preferences is the evolving meanings of
American liberalism. Starr (2007) noted that the modern, democratic liberalism incorporated democratic
10

claims from historically marginalized groups and “a more comprehensive vision of equality” (p.16) as it
branched off constitutional liberalism. Unlike constitutional liberalism, which was shared by all sides,
modern democratic liberalism distinguishes the beliefs of liberals from those of the conservatives. Shils
(2019) also commented that liberals had moved from guarding against state authority towards guarding
against all forms of authority. This expanded list of authorities that modern liberalism became critical of
should include the perceived power of modern corporations and the privileged class. At the same time, this
current of modern liberalism showed greater sympathy for collectivist ideas and faith in expanding government action for the common good (Shils, 2019). The motivating rationale of this transition emphasizes
ideals to advance the status of the poor and minority groups and seek equality beyond equal opportunity
and equality before the law. Therefore, despite inheriting the name of liberalism, modern democratic liberalism has to a greater extent embraced equality and egalitarianism relative to liberty and individualism,
the more orthodox liberal principles. Rokeach (1973) argued in the 1970s that the relative rankings of the
values of freedom and equality were at the core of the ideological differences in the US context. The conservatives prioritized the value of freedom, and liberals ranked equality at least as highly as liberty (Rokeach,
1973) This value divide continued to grow as liberal Democrats were found to be increasingly egalitarian
over a course of 12 years since 2008 (Carman et al., 2020), as Democrats are a coalition of diverse social
groups advocating for concrete government action (Grossmann & Hopkins, 2015; Lupton et al., 2017).
2.5 Incremental movement towards equality
How do American liberals’ changing beliefs and commitments play a role in the hypothesized value change
we are studying? Historians and observers of American development patterns identified repeated pendulum swings in American history between periods of democratic, public-oriented, and inclusive reforms
and periods of conservatism and restoration, i.e. between liberalism and conservatism (Schlesinger, 1999;
Schlesinger & Hart, 1950). Liberal periods were dominated by calls for expanding democracy, politically,
11

economically, and socially. Conservative periods tended to contain democracy and maintain the status quo.
History alternated between these two competing motivations starting from the initial liberal movement
to create the Constitution in 1776. Because the reforms achieved during each liberal period were mostly
preserved during the succeeding conservative period, the new cycle of liberalism would start at a higher
level instead of reverting to the original point. Over the long run, this spiral course of history accumulated change from each cycle, and incrementally moved towards a more democratic, public-oriented, and
inclusive direction (Schlesinger, 1999; Schlesinger & Hart, 1950). Therefore, liberals’ evolving appetite for
equality and waves of progressive movements through the Civil War, the Progressive Era, the New Deal,
the Civil Rights Era, and counting, could have had cumulative impact on long-term cultural change.
McClosky and Zaller (1984)’s book on public attitudes towards democracy and capitalism described a
similar spiral movement across time. The authors perceived an internal value conflict between the American commitment to democracy and capitalism. They share historical and philosophical roots, but as capitalism increased economic inequality, political inequality increased accordingly, and its persistence undermined democratic governance. This inherent conflict was built into the ideological division between liberals and conservatives, which caused national moods to swing between pro-capitalism and pro-democracy
sentiment with a long-term direction towards preferences for democracy (McClosky & Zaller, 1984). Thus,
a similar story was told about how liberals’ preferred direction of reform, greater equality or democratic
expansion, drove the long-term value trending.
2.6 Expanding the definition of equality
The waves of social movements championed by liberals in their respective contexts since the late 19th and
early 20th century strengthened the value of equality in liberal causes and expanded the liberal vision of
equality. First, an individualistic vision of procedural equality was increasingly challenged during the last
century. Ellis (1992) pointed out that both an individualistic vision of equality and an egalitarian vision of
12

equality were rooted in early American political thought. The former emphasizes equal opportunity and
equal process, and the latter demands equality of outcomes. Until the Jacksonian Era, the two visions of
equality were peacefully aligned: those who favored equal results believed equal process in free-market
capitalism was the means to achieve that end (Ellis, 1992). However, Ellis (1992) went on to elaborate
that this belief started to be challenged after the Civil War, slowly fell apart throughout Populism, Progressivism, and the New Deal, and was replaced by a downright rivalry between the individualistic and
the egalitarian vision of equality after the New Left Movement established a results-oriented definition of
equality. The emerging new liberalism was drawn to welfare state systems and almost divorced from the
belief of achieving personal liberty through private property and free market capitalism (Gaus, 1983).
Apart from the changing conception of equality with regard to wealth, the rise of identity politics and
group-based rights movements in the late 20th century further expanded the vision of equality and had
a lasting impact on contemporary political discourse. Theorists of these identity-based social movements
raised criticisms against the individualistic reading of equality in liberal theory that maintained equal
treatment of individuals and a stance of state neutrality (Moore, 2009). Instead, advocates for the politics
of identity contended for equal treatment of all social and cultural identities in a system where a diverse set
of minority identities were subjected to structural biases (Moore, 2009). This expanded reading of equality
was at the core of the politics of recognition, in which equality was understood as the end of hierarchical
relationships between social groups (Laden, 2009). It further embraced and enriched an egalitarian vision
of the equality principle on the socio-cultural dimension.
This pattern was corroborated by a longitudinal national survey which revealed that Democrats’ values
had become “more egalitarian and less individualistic”, while Republicans remained “mostly static” over a
span of 12 years up to 2020 (Carman et al., 2020).
13

Therefore, walking through an evolutionary course of equality being elevated in modern liberalism and
reinterpreted through waves of demands for equality, economically, socially, and culturally, my hypotheses
regarding the conceptual evolution of liberty and equality are laid out as follows:
H1: The meaning of equality should increasingly represent democratic values and progress relative
to liberty over more recent decades.
H2: The meaning of equality should increasingly embody an egalitarian vision of the concept.
H1 is implied from the idea that the value of equality gained increasing favorability over the last century among liberals, whose preferences tended to define the direction of political progress on a long time
horizon. H2 represents the judgment that, alongside the rise of equality preferences, the egalitarian vision
of this political ideal rose in prominence over its individualistic interpretation.
2.7 Studying semantic change
Semantic change can be a symbolic mirror of cultural change, as cultural values and beliefs are sometimes
encoded into the meanings of value-laden words. For example, the commonly understood meaning of
diversity in the US context is inseparable from the classification of social groups based on race, ethnicity,
gender, religion, among others. It reflects the political salience of these social categorizations unique to
its own past and present. As aforementioned, there are disciplinary and methodological precedents for
studying the change and continuity of values, interests, and preferences through the evolving meanings of
abstract concepts. Charting the history of political and social concepts has its own discipline in conceptual
history (or Begriffsgeschichte) in German language scholarship, whose research inquiries include a history
of the concept of “liberty” during the French Revolution, or how “freedom” became a political concept in
Athenian democracy (Richter, 1995a). It treats value systems and ideas as historically contingent rather
14

than invariant, which contrasts with the emphasis on formal logic and universal rules to study how language relates to reality in the early school of the philosophy of language, which had a stronger influence
in the 20th century Anglophone countries (Richter, 1995b). Another continental strand of thought that
deals with the evolution of discourse is the genealogy approach. Foucault (1980), who developed this approach from Nietzsche, held the belief that the struggle for humanity’s dominant system of rules had been
recorded in a series of interpretations, especially in the interpretations of metaphysical concepts, morals
and ideals. The genealogy approach is meant to uncover the history of events through which aspects of a
concept was formed and different interpretations emerged (Foucault, 1980). Compared to conceptual history, the Foucauldian genealogy approach of discourse also makes efforts to identify the role of historical
agents in the adoption of political and social languages (Richter, 1995a).
Tracking the evolution of concepts is an unwieldy task and relies on a vast amount of historical text
data, often tedious and costly to be conducted manually. The German language conceptual historians
have collectively compiled monumental resources on the history of political and social concepts such as
HWP (Historisches Wörterbuch der Philosophie), GG (Geschichtliche Grundbegriffe), and Handbuch. The
availability of digitized corpora and development in modern natural language processing (NLP) techniques
has allowed for automated methods to model semantic change in large-scale historical text data. Word
embeddings in particular were utilized for this objective since its inception (Hamilton et al., 2016b).
Word embeddings represents words as vectors by their collocations (Mikolov et al., 2013). The vector
representation of words by co-occurring words relates to Wittgenstein’s later philosophy and Firth’s approach to word meanings – to infer the meaning of a word by the company it keeps (Firth, 1957; Skelac
& Jandrić, 2020). A word is represented by its co-occurrences as a vector, so words that appear in similar
contexts will have a closer geometric distance to each other in the vector spaces. This method rests on
the distributional hypothesis in linguistics that words that occur in the same contexts tend to have similar
meanings.
15

Among its applications in modeling semantic change, Kulkarni and collaborators used word embedding models learned from the context of individual time periods to show that the word “gay” moved from
a vector-space position proximate to “cheerful” and “dapper” in the 1900s towards a position closest to
“homosexual” and “lesbian” in the 21st century (Kulkarni et al., 2015). In addition, the authors found that
the meaning of “gay” remained stable for the first 50 years, started to shift between 1950 and 1975, and
accelerated its change from 1975 till now (Kulkarni et al., 2015). Rodman (2020) tracked the interrelationship between “social” and “equality” over the past 161 years using the diachronic distances between this
word pair learned from a smaller corpus of newspaper coverage (Rodman, 2020). Rodman found that “social” and “equality” started with closer proximity when “social equality” was a euphemistic expression of
aversion to black-white intimacies. However, the proximity between the pair of words steadily declined
before the Civil Rights movement in the 1950s, and then rose again after the movement, this time not as a
racial euphemism, but motivated by social justice and economic equality discourse (Rodman, 2020).
This study traces the semantic coevolution of liberty and equality in a similar methodology of diachronic word embeddings under the assumptions of how conceptual meanings map to cultural values
held by conceptual history and the genealogy approach. Liberty and equality, the most important principles underpinning modern liberal democratic institutions offer a unique lens into the long-term political
cultural patterns. The next section introduces the data and analytical strategies for testing my hypotheses.
2.8 Data
To test my hypotheses regarding the semantic trends of liberty and equality, I obtained the diachronic text
data commonly used to study cultural change: the Corpus of Historical American English (COHA) and
the Corpus of Contemporary American English (COCA) constructed by Mark Davies. Word embeddings
perform better on larger corpora (Lai et al., 2016). While English corpora tend to be smaller, COHA and
COCA are 100 times larger than any other structured corpus in American English, but also provide the
16

benefits of smaller corpora (Davies, 2012a, 2012b). COHA and COCA share the same architecture and
are genre-balanced across time. COHA contains 400 million words maintains the same genre and subgenre balance between fiction, popular magazines, newspapers, and non-fiction books decade-by-decade
2
. For example, the selection of non-fiction book categories (e.g. history, religion) were balanced across the
Library of Congress classification system, and the balance is sustained across decades (Davies, 2012b). This
sampling procedure ensures reliable comparison of American English language and culture across decades
(Davies, 2012b). COCA contains over one billion words and the texts are evenly divided between spoken,
fiction, magazines, newspapers, academic journals, blogs, other web pages, and TV/Movie subtitles and
the source balance is sustained from year to year 3
. The three decades of COCA include a few more genres
compared to COHA due to its contemporariness, but overall they were both constructed for over-time
comparability and share multiple core genres, with COCA offering richer information. Combined, they
provide texts from representative sources intended to reflect the language, culture and society of a given
time period from the 1810s to 2010s.
The author did not choose another popular option for large-scale historical corpora: Google Books
Ngrams. It accounts for 6% of all books published over the past five centuries in eight languages (Lin et
al., 2012) and is larger than the COHA. However, it is not genre-balanced and does not include metadata.
Therefore, it is criticized by some as not reflective of cultural and linguistic change (Koplenig, 2017).
2.9 Method
2.9.1 Model
Diachronic word embeddings were performed on the combined COHA and COCA corpora to trace the
semantic evolutions of interest. Word embedding models were trained with an objective to maximize the
2
https://www.english-corpora.org/coha/
3
https://www.english-corpora.org/coca/
17

probabilities that a word co-occur with its co-occurring words (Kulkarni et al., 2015) and therefore carry
more contextual information than word frequencies. Diachronic word embeddings train sequential word
embedding models on timestamped corpora over successive time units (Hamilton et al., 2016b). I divided
the corpora into decade-long bins from the 1810s to the 2010s and learned one word embedding model for
each decade. Although this diachronic analysis approach is not best-suited for small corpora (Rodman,
2020), it is standard practice applied to large corpora like COHA and Google N-Grams (see for example
Hamilton et al., 2016a, 2016b; Kulkarni et al., 2015). This amounted to 21 models that map to 21 successive
decades. I used the word2vec model offered by the python package gensim 4
.
During model training, I tested both the continuous bag of word (CBOW) and skip-gram techniques.
They are both shallow neural networks but with different architectures. CBOW turned out to perform
better and was therefore chosen by this project. Negative sampling was also used. The dimension of
each word vector was 300, within a standard range for yielding the most useful results (Pennington et
al., 2014) and at an optimal point that balances performance and computational efficiency (Rodriguez &
Spirling, 2022). The window was set to 20, meaning the 20 words both before and after a target word were
considered as collocations during model learning. This is larger than the recommended and popular size
for this parameter and offers better performance (Rodriguez & Spirling, 2022). Bigrams were built as model
inputs: if two words co-occur more frequently than chance allows, they were treated as a bigram phrase.
The training epochs were between 30 to 50, until the loss function stopped to decrease.
2.10 Analytic strategies
The diachronic models output word vectors for each of the 21 decade-long corpora from the 1810s to the
2010s. The geometric relationships between different words were calculated to capture their semantic
relationships. The semantic similarity between each pair of words was measured by the cosine similarity
4
https://github.com/RaRe-Technologies/gensim
18

of their vector representations (Hamilton et al., 2016b; Turney & Pantel, 2010). Cosine similarity measures
the similarity of two vectors and takes a value between −1 and 1. A higher value indicates a smaller angle
between two vectors in an inner product space and greater similarity between them. To test whether
liberty or equality connotes democratic values and political progress for our first hypothesis, I compared
their respective similarities to a group of terms that indicate desirable political outcomes. The terms were
curated from an expert survey of terms that represent democratic values and desirable political outcomes,
such as reform, democracy, justice, as well as a list of positive words on dictionary.com, such as improve,
moral and happiness (Appendix 2.3). Our assumption is that the more desirable liberty or equality is as
a democratic principle, the closer semantic distances it has to this dictionary of democratic values and
progress under the logic of word embeddings. An extended and an abbreviated dictionary of political
preferences are also constructed for sensitivity analysis.
Dictionaries of individualism and egalitarianism were also constructed to test my second hypothesis:
whether the meaning of equality has moved towards an egalitarian interpretation of the concept. If my
prediction holds, equality should move closer to the group of terms that describe egalitarianism over time.
The selection of these terms was based on a snowball approach: first, notable propositions and keywords
in support of individualism or egalitarianism were identified based on encyclopedias of philosophy, such
as individual freedom, property rights, and self-sufficience for individualism, and equal protection, wealth
equality, racial equity and so forth for egalitarianism 5
; next, keywords from the first step were fed into
our model vocabulary in search for additional relevant terms and phrases. For example, using the seed
word "free", I found individualism-related bigrams "free will", "free press", "free expression", "free trade",
"personal freedom", "free enterprise" in model vocabulary (see Appendix 2.4). If equality showed greater
proximity to the egalitarianism dictionary than to the individualism dictionary over time, then this hypothesis would be supported.
5
Stanford Encyclopedia of Philosophy (https://plato.stanford.edu/); Encyclopedia Britannica (https://www.
britannica.com/browse/Philosophy-Religion)
19

Figure 2.1: Similarity between equality and groups of words that indicate black identity. Y-axis represents
cosine similarity. fig:ch2_race
2.11 Results
2.11.1 Face validity checks
To confirm the face validity of our unsupervised models, I checked their performance on commonsense
expectations. For example, it is reasonable to expect “liberty” to be more similar in meaning to “freedom”
than is “equality”. Equality should have more of a social aspect than does liberty. Results from the models turned out as I expected: Liberty was consistently more similar to “freedom” than was equality over
our study period of 21 decades, and equality was consistently more similar to “social” than was liberty
(Appendix 2.6).
As was discussed in literature review, identity-based social movements during the 20th century raised
the demands for equality, which impacted how equality was being valued among liberals. I looked at the
proximities between equality and groups of words that indicated black and female identities respectively,
since the advocacy for racial and gender equality was among the most influential identity-based activism.
The semantic proximities were obviously more positive during the Civil War Era around the 1860s and
from the Civil Rights Movement Era onward (2.1). Similarly, equality and the female identity exhibited
positive similarity measures right around the first, second, and third waves of feminism respectively (2.2).
20

Figure 2.2: Similarity between equality and groups of words that indicate female identity. Y-axis represents
cosine similarity. fig:ch2_gender
2.11.2 Results for H1
To test H1, the average similarity scores between liberty to our dictionary of democratic values and
progress were computed across time. They were compared to the similarity scores between equality and
the same dictionary in each decade (2.3). This is to compare the positive political connotations of the terms
liberty and equality over time. For much of the study period, equality was not semantically closer to the
terms that convey democratic values and progress. However, since the 1960s, at the tail end of the civil
rights movement, equality had become consistently closer to terms of democratic values and progress compared to liberty. For sensitivity analysis, I prepared extended and abbreviated versions of the dictionary
by adding or excluding certain terms (see Appendix 2.3). The extended version of the dictionary included
more generic positive terms such as opportunity, potential and joy, whereas the abbreviated version of the
dictionary removed terms that might be inherently pro-equality, such as emancipation, a Marxist term.
The results that equality increasingly connoted political progress and positive meanings in general than
did liberty starting from the 1960s were robust to dictionary adjustments.
Pairwise word similarity time series (Hamilton et al., 2016b) of core terms from our dictionary were
also constructed, including “change”, “progress”, “hope”, “future”, “justice”, “value”, “life”, “happiness”. As
21

Figure 2.3: Average similarity between liberty (red)/equality (blue) and a dictionary of democratic values
and progress. Error bars represent 95% confidence intervals derived from bootstrapping. fig:ch2_dem_vals
a common pattern, their proximity to equality eventually overtook their proximity to liberty over the past
21 decades.
These results are consistent with my expectation in H1 that the term equality had increasingly implied
democratic values, progress and general positivity relative to liberty since the middle of the 20th century.
2.11.3 Results for H2
In a diagnostic attempt to test my second hypothesis, the most similar terms to liberty and equality in every
decade were obtained to approximate the evolving meanings of these two political concepts. The meaning
of liberty remained similar over the past two centuries. It featured the core elements found in the United
States Declaration of Independence: the inalienable rights of life, liberty, and property. Table 2.1 presented
the most similar terms to liberty in very different historical periods as examples (the Jacksonian era, Civil
Rights Era and the present respectively) 6
. It characterized a Lockean vision of the intimacy between
political, religious, and economic liberty (Capaldi & Lloyd, 2016). This indicates that the dominant vision
of liberty may have always been an individualistic one. Over much of the 21 decades, the most similar
terms to equality likewise resembled an individualistic vision, embodying themes such as the inalienable
rights, equal opportunity, individual freedom, and liberty. However, starting from the 1950s and 1960s, it
6Due to the limited space, only figures of three representative decades are inserted here, although all 21 decades are available.
22

Decade Most similar terms to "liberty"
1840s freedom, right, free, guaranties, inalienable_right, civil_liberty, inalienable, government, constitution, civil_religious
1960s freedom, inalienable, private_property, free_speech, guarantee, constitution, freedom_speech, rule_law, constitutional_right, constitutional
1970s freedom, declaration_independence, free_speech, patriotism, constitution, first_-
amendment, monarchy, founding_father, republic, tyranny
2010s freedom, individual_liberty, founding_father, declaration_independence, constitution, tyranny, unalienable_right, freedom_speech, liberty_property, inalienable_right
Table 2.1: Most similar terms to "liberty" across decades tab:ch2_lib
Decade Most similar terms to "equality"
1840s equal_right, right, right_privilege, individual, equal, inequality, liberty, favored,
principle, property
1940s equal, liberty, freedom, abolition, self-government, principle, segregation, inalienable, equal_opportunity, subordination
1950s equal, inequality, fourteenth_amendment, separate_equal, equal_protection, discriminati, suffrage, principle, racial, discrimination
1960s congress_racial, n_word, equal_opportunity, inequality, equal, educational_opportunity, racial, civil_right, racial_discrimination, racial_equality
1970s equal_right, equal, educational_opportunity, equal_opportunity, society, feminist,
polygamy, inequality, equal_protection, injustice
2010s equal, equal_opportunity, gender_equality, human_dignity, freedom, social_justice, racial_equality, civil_right, inequality, egalitarian
Table 2.2: Most similar terms to "equality" across decades tab:ch2_equ
started to gain a racial-equality theme. From the 1970s onward, new neighboring words emerged such as
society, feminist, injustice, social justice, and oppression. By the 2010s, “egalitarian” rose to one of its most
similar terms, along with racial equality and gender equality (2.2). This is evidence that an egalitarian
vision of equality gained traction from the 1970s and became more dominant in the recent decades. The
evolution of the dominant themes proximate to “equality” is consistent with our prediction of a trend
towards an egalitarian interpretation of equality.
The average similarity scores from equality to our concept dictionaries of individualism and egalitarianism were computed over the past century to test H2. Data from the 19th century was omitted because
core conceptual terms such as “egalitarian” did not appear frequently enough in our corpus to make their
way into model vocabulary until later decades. The over-time similarity score comparison shows that,
23

Figure 2.4: Similarity between equality and individualism (light blue), egalitarianism (navy). Error bars
represent 95% confidence interval derived from bootstrapping. Y-axis represents cosine similarity. fig:ch2_equ
in a similar trend, equality had become semantically closer to terms that represent egalitarianism than
to terms that represent individualism since the 1960s, and the separation became complete in the 2010s
(2.4). In contrast, liberty had always been semantically closer to individualism relative to egalitarianism.
No changes of leads took place within the same window of a hundred years (2.5). These results are in
support of H2 which argues an egalitarian vision of equality was gradually replacing an individualistic
vision thereof.
2.12 Discussion
The results based on term and concept similarities are consistent with our hypotheses that the meaning
of equality was increasingly linked to democratic values and progress relative to liberty and moving towards an egalitarian interpretation most evidently after the Civil Rights Movement Era. Conclusions about
macro-level phenomena from semantic studies should be drawn with caution. These findings based on semantic evolution are not direct evidence, but at least congruent with a broader argument for incremental
24

Figure 2.5: Similarity between liberty and individualism (light blue), egalitarianism (navy). Error bars
indicate 95% confidence interval derived from bootstrapping. Y-axis represents cosine similarity. fig:ch2_lib
cultural preferences towards equality, motivated by the political goals of advancing the rights, well-being,
and dignity of historically marginalized groups.
A political tradition built on liberty is rare, hence a part of “American exceptionalism”. If Tocqueville
was right, liberty does not hold the same type of broad-based appeals that equality does. It is after all a
principle used to justify the protection of nonconforming individuals and minorities from the tyranny of
various forms. Equality, on the other hand, can echo through a vast crowd who fall on the wrong end
of inequality. It is worthwhile to point out that egalitarianism is also an American Revolution ethos and
an exceptional element of the cultural tradition, alongside status-achieving (Lipset, 2019). Egalitarianism
may even be a higher order value. When liberty was believed to serve egalitarian goals, both were prided
ideals. When the relationship between liberty and egalitarianism turned fraught, the preferred direction
of change was towards expanding egalitarianism. In this sense, liberty was an instrumental value.
During the civil rights movement, identity-based rights movements championed arguments embedded
in ideals of liberty in their demands for recognition, such as the expansion of First Amendment rights and
25

self-expression. But from the end of the 20th century to the contemporary period, a new conception of
what better serves group rights is gaining traction. Equality and equal status are espoused as both the
means and ends. Much more about this quiet evolution of democratic priorities awaits to be observed,
defined, qualified, validated, and explained.
The rising significance of equality as a democratic virtue and its evolving meanings for some social
groups presents both pressure for greater, more inclusive democracy and challenges to democratic stability,
as the tension between newer ideals and older liberal consensus risks deepening an already polarized
society, demonstrated by some of the late controversies around free speech I discussed in the introduction.
This study identified a key inflection point of this movement of thought in the 1960s. Future research
should dive deep into this critical juncture through close reading of texts, identify the actors and driving
forces of a potentially impactful political cultural trend for times to come.
The application of diachronic word embeddings will also benefit from more rigorous alignment of
models that were trained on different corpora. Although I confirmed satisfactory face validity of the diachronic models, the best practice is to align them across time for improved comparability. From there, we
would be able to meaningfully compute the rate of semantic change and other interesting metrics for our
concepts of interest.
2.13 Appendix
26

Main dictionary
change progress justice value life happiness hope future liberation emancipation moral improve reform achieve democracy liberal_democracy human_right human_dignity ideal optimism strength
Extended dictionary
(More generic positive terms were added)
change progress justice value life happiness hope future liberation emancipation moral improve
reform achieve democracy liberal_democracy human_right human_dignity human_race ideal optimism strength wonder faith vision ambition spirit dignity potential faith opportunity wonder
joy leadership grace
Abbreviated dictionary
(Removed terms that could have endogenous connotations of equality, such as emancipation – a
Marxist term)
change progress value life happiness hope future moral improve reform achieve democracy liberal_democracy human_right ideal optimism strength
Table 2.3: Dictionary of democratic values and progress tab:dict1
Individualism(38 words/phrases)
individual individualism individuality individualist individuali individual_liberty personal_freedom capitalism free_market free_trade liberty_property civil_liberty free-thinker freethinker freedom_speech freedom_press freedom_expression free_will freewill free_enterprise self-sufficien
self-sufficie self-sufficient self-suffic self-suffici self-worth self-interest religious_libertyselfreliance self-reliant self-relian private private_interest privacy equal_opportunity property private_property market-based libertarian
Egalitarianism (36 words/phrases)
equal_participation equal_share equal_right equal_distribution inequality unequal racial_equality equal_protection equal_opportunity poverty_inequality economic_inequality equal_employment income_inequality wealth_inequality marriage_equality gender_inequality equity egalitarian egalitarianism social_justice social_equality racial_equality gender_equality inequity racial_-
inequity gender_equity fair_share free-for-all community communal commune communitarian
social_welfare public_welfare welfare_reform welfare_state
Table 2.4: Dictionary of individualism and egalitarianism tab:dict2
27

(a) social fig:ch4_social (b) freedom fig:ch2_free
Figure 2.6: Similarity between equality/liberty and “social”, “freedom” fig:ch2_app3
28

Chapter 3
Moralization of political discourse on Twitter in the lead-up to the US
2020 election
ch:moral
3.1 Abstract
This chapter characterizes the dynamics of social media moral expression leading up to and shortly after
the US 2020 election using a large-scale Twitter dataset (26 million users and 848 million tweets). The
study is divided into two parts. The first part presents exploratory analyses on the moralization process
of topics unique to the US 2020 election, including mail-in voting, COVID-19 and voter fraud. I applied a
transformer-based model to detect moral content in social media posts. I identified rise and fall of moral
sentiment with regard to mail-in voting discussions and synchronized patterns of moral expression by
liberals and conservatives over voter fraud disputes. The second part considers the relationship between
affective polarization and moral expression. Through regression analysis, we show a clear positive association between negative partisan affect and moral expression in the discussions of electoral fairness. This
relationship was further strengthened in the final stage of the election cycle. This study joins the rich
literature on the intimate correspondence between affect and moral judgments and lends support to the
hypothesis that negative partisanship plays a role in amplifying moral expression.
29

3.2 Introduction
Analysts have noted the salience of moralistic rhetoric in contemporary American politics, from the focus
of identifying good and evil in interpreting nuanced policies to regular moral condemnation of political
opponents in political campaigns 1
. Although moralized visions and practices of political relations have a
longer tradition in American foreign policy since Woodrow Wilson pioneered a style of "moral diplomacy"
(Berggren & Rae, 2006; Kennedy, 2013), moralistic rhetoric in domestic politics constitutes a relatively new
political force and increasingly reflects divergent moral philosophies of different social segments. Moralized political discourse communicates political messages in moral terms, elevates the framing of issues and
entities in the moral plane, and insists on the speakers’ own beliefs of righteousness (Weiner, 2019). It interprets concrete policy issues or partisan objects through the lens of categorical right and wrong, which
leans towards polarized thinking. Therefore, moralized attitudes resist the political instrument of compromise and negotiation (Mooney & Schuldt, 2008; Ryan, 2017; Weiner, 2019), produce strong emotions
and intolerance to differing views (Cole Wright et al., 2008), and provoke antagonistic political participation, even violence (Mooijman et al., 2018; Skitka & Morgan, 2014) – a host of characteristics that define
political polarization (Clifford, 2019). A moralized political climate, under which people reason through
uncompromising principles rather than costs and benefits (Ryan, 2017), squeezes the room for deliberation and negotiation. It shares many symptoms and consequences with political polarization, yet the
moralization of politics is an independent phenomenon that warrants its own attention.
This chapter studies the moralization process of political discourse during the 2020 US presidential
election by observing the dynamics of moral expression on social media. Previous research demonstrated
that changes in moral attitudes on political issues or candidates can happen along the path of an election
based on longitudinal survey data (Brandt et al., 2015; Clifford, 2019). This study engages with this line of
inquiry using large-scale social media data to detect broader patterns.
1
see for example Cassese, 2021; Weiner, 2019
30

The study is divided into two stages: the first stage is exploratory. We take the moral temperature of
several focal issues unique to the 2020 election. In the second stage, we zoom in on intriguing results from
the first stage and examine the relationship between moral expression and affective polarization.
3.3 Stage 1
3.3.1 Process of moralization
Moralization describes the process whereby a morally neutral entity acquires a moral status (Rozin, 1999;
Rozin et al., 1997). Once an entity becomes moralized, individual and social censure is licensed, and preferences are transformed into durable values (Rozin, 1999). For example, when an individual moralizes meat
consumption, they could turn into lifelong vegetarians. It is therefore a profound cognitive and affective
process that could cultivate or alter preferences and norms. More attention has been paid to the impact of
moralized attitudes, yet how the process of moralization takes place is an intriguing and less understood
phenomenon, for example, the contributing and hindering factors of moralization, the shape of moralization, etc. (Feinberg et al., 2019). Rhee et al. (2019) introduced a useful framework that differentiates varying
forms of moralization. In a process of "moral recognition", an entity transitions from being morally neutral
to a moral issue. In a process of "moral amplification", an already morally relevant entity sees a change in
the degrees of moral concerns being attached to it (Rhee et al., 2019). This framework distinguishes the
change from zero to one from a change in degrees.
The mechanisms of moralization are relatively understudied and could be better explored in specific
domains (Rhee et al., 2019). Commonly recognized mechanisms include moral emotions, such as guilt,
anger and disgust, and moral piggybacking, the psychological process of connecting an issue with an
individual’s existing moral convictions (Feinberg et al., 2019; Rozin, 1999).
31

3.3.2 Moralization of political attitudes and communication
The processes of moralization will be better understood within their specific domains (Rhee et al., 2019).
In the political context, much remains to be known regarding the characteristics and mechanisms of moralization. Morality is not inherently relevant to political beliefs. There exists decent individual variation
in their propensity to moralize political issues and figures (Clifford, 2019; Ryan, 2014). Some individuals
are more likely to base their political opinions on fundamental right and wrong than others, and those
who view politics through a moral lens display greater desire for social distancing from opposing partisans (Garrett & Bankert, 2020). Moral convictions about a political party can take roots if the party or its
leaders violate an individual’s moral principles at the core of their identity (Garrett, 2016).
The admission of cultural and moral issues into American partisan politics in recent decades expands
the dimensions of moral concerns in domestic political preferences. At the top of the most morally dividing matters, people can find salient partisan issues on the socio-cultural dimension like abortion and
gay rights (Saad, 2010). Over the interpretations of these issues, we see a clash between traditional moralism commonly associated with Protestant culture and evangelical groups and a modern, secular strand of
moralism that deeply values dignity, safety and equal rights of women and LGBTQ groups. It is not hard
to find a clash of moral principles behind every hot-button issue. The moral basis of ideological differences
has been validated by the popular paradigm of Moral Foundations Theory (Graham et al., 2013; Graham
et al., 2009; Graham et al., 2011; Haidt, 2012; Haidt & Graham, 2007), which provides evidence that liberals,
conservatives, libertarians operate under different moral "foundations" (Graham et al., 2009; Iyer et al.,
2012).
Morally framed communication also helps the audience map political entities to moral codes, and it
is a universal mechanism of moralization (Rhee et al., 2019). Moral messages are often evocative, clear
and adamant, with ambiguities and nuances removed. Hence a powerful device to mobilize and recruit
followers. Moral arguments, such as who is more "deserving" of social welfare, are more effective in
32

motivating a constituency than economic reasoning (Cohen, 2017). Visual information that primes moral
shock to aborted fetuses increases moral convictions about abortion (Wisneski & Skitka, 2017). Persuasive
frames that appeal to moral emotions (in particular anger and disgust) evidently moralize public opinion
(Clifford, 2019).
Social media amplifies the effects of morally framed communication. Online social networks encourage the diffusion of morally and emotionally charged content (Brady et al., 2017; Carpenter et al., 2020).
Moral content is more likely to capture attention, and social media affordances capitalize on this human
propensity (Brady et al., 2020). Information consumers are exposed to moral content on social media platforms with a much higher probability than from other information sources (Crockett, 2017). Instant and
continuous social feedback from social media amplifies moral outrage (Brady et al., 2021). Operating under a social media environment, political elites would effectively leverage the contagious effects of moral
emotional language to spread their messages (Brady et al., 2019).
3.3.3 The case of US 2020 election campaign trail
The US 2020 election campaign trail provides a unique opportunity to observe the process of moralization
due to unexpected controversies that emerged in this election cycle. The inordinate attention given to
mail-in voting saw the politicization of a previously neutral topic of how voting is administered (Clinton
et al., 2020; McGhee et al., 2022). Although relevant research does not support that universal mail-in voting
would have any partisan effects on voter turnout (McGhee et al., 2022; Thompson et al., 2020), Trump and
Republicans believed that the expansion of voting by mail would benefit the Democratic Party (West,
2020). Trump himself made regular remarks on Twitter and offline denigrating the legitimacy of voting by
mail (Niebler, 2020; Peeples et al., 2020), after the outbreak of the pandemic COVID-19 motivated greater
demands for expansion of mail-in voting (Clinton et al., 2020; Herrnson & Stewart III, 2022; Herrnson &
Stewart III, 2023).
33

The unusual voting by mail controversies during the 2020 election cycle involve three key elements:
COVID-19 which provided the basis for expanding postal voting in many states, mail-in voting itself, and
voter fraud claims perpetuated by Trump. Poll data suggests that the majority of voters were aware of
a discussion surrounding mail-in voting due to COVID-19 and whether it undermines the legitimacy of
the election (Mitchell et al., 2020). Beliefs in the legitimacy of mail-in voting revealed partisan divide
(Clinton et al., 2020; Mitchell et al., 2020; Niebler, 2020). And this partisan divide took shape rapidly after
an initial short period of bipartisan support for mail-in voting and Republicans became less concerned
about COVID-19 (Clinton et al., 2020).
Mail-in voting as a technical aspect of administering elections was not in and of itself a moral issue.
However, it was brought into the conversation of election integrity and subsequently interpreted on a
moral dimension of fairness and cheating. Our initial inquiry is whether the topics of COVID-19, mail-in
voting, and electoral fraud – three key components of this controversy – became increasingly moralized
during the presidential campaign. In addition, liberals and conservatives likely display different patterns
of moral expression in these topics. Therefore, it is worthwhile to also consider the partisan differences.
RQ1: How do the levels of moral expression change in the discussions of mail-in voting, COVID-19,
and voter fraud over the 2020 US election campaign trail?
RQ2: how do liberals and conservatives differ in their levels of moral expression related to the topics
of mail-in voting, COVID-19, and electoral fraud?
To complement the analysis of these three themes, we examine the role of misinformation in moralization. The voter fraud claims have been rendered baseless and repeatedly debunked by experts (Hansen,
2020; West, 2020). The 2020 election trail was plagued with viral misinformation and conspiracy theories from mail-in voting fraud to QAnon 2
. Misinformation undermines public trust in electoral processes
2
https://www.nytimes.com/article/what-is-qanon.html
34

and democracy (Green et al., 2022; Sanchez et al., 2022). It significantly impacted the target audience’s
perceptions of fairness and cheating in elections. However, a pessimistic finding reveals that lies travel
faster and cascade deeper than truth on Twitter, possibly due to their ability to elicit surprise, fear and
disgust (Vosoughi et al., 2018) – negative emotions are often antecedents to the formation of moral convictions (Wisneski & Skitka, 2017). Based on how misinformation is framed at greater liberty to provoke
moral emotions, we predict misinformation to contain a greater share of moral sentiment than reliable
information.
H1: Misinformation is more moralized than trustworthy information.
3.3.4 Data and measures
3.3.4.1 Data
We used a publicly available Twitter dataset on the 2020 US presidential election that tracked keywords
and accounts specific to this election (Chen et al., 2021). Our study period starts from March 1, 2020 to
November 30, 2020, covering a substantial portion of the 2020 election campaign trail. The dataset contains a very large sample of political conversations on Twitter regarding the presidential election, having
collected all real-time tweets that came through Twitter’s streaming API, which returns approximately 1%
of all tweets being published (that match the keywords) at any given time (Chen et al., 2021). The sample
size amounts to 848,199,398 tweets. The percentage fluctuations of candidate mentions tally well with
real-world events during this time period, from the Democratic primaries, presidential debates to Election
Day (Chen et al., 2021).
3.3.4.2 Detection of moral expression in tweets
This study takes advantage of an existing annotated moral foundations Twitter corpus (Hoover et al.,
2020), which is used to train a deep learning classifier (BERT-based model) to classify and predict moral
35

model precision recall F1-score
SVM-MFD2 0.79 0.84 0.81
TweetBERT 0.76 0.91 0.83
BERT-base 0.81 0.87 0.84
Table 3.1: Model performance – prediction of moral content tab:ch3_acc
content in the aforementioned Twitter dataset. This annotated corpus contains 35,108 tweets related to
seven different social and political issues. It also includes the 2016 presidential election as one of the
seven discourse domains (Hoover et al., 2020). According to Hoover and colleagues, the authors of this
dataset, each tweet was labeled by at least three annotators on the five moral foundations according to the
Moral Foundations Theory (Graham et al., 2013). I generated a binary variable of moral expression from
this dataset by collapsing all five moral dimensions and ten categories (moral virtues and vices). Tweets
that have the majority vote going towards at least one moral dimension were labeled 1, and those being
considered as nonmoral by the majority vote were labeled 0. Then a pre-trained BERT model from Hugging
Face library 3 was fine-tuned (weights on pre-trained model are trained on new data) on this annotated
Twitter corpus. It outperforms other models I implemented, including an SVM model trained on word
counts of Moral Foundations Dictionary 2.0 (MFD2)4
and another transformer-based model fine-tuned on
TweetBERT (Qudar & Mago, 2020).
The resulting model was applied to our US 2020 election dataset. To test its performance on the new
dataset, we manually annotated 200 tweets, and it achieved an accuracy score of 0.80 5
. Our measures
of moral expression should be best interpreted as expressions of both moral judgments, attitudes and
sentiment, since it is generated from parsing organic, unstructured short social media content as opposed
to survey scales which are designed to capture a precise construct.
3
https://huggingface.co/docs/hub/models-libraries
4The SVM model implementation followed the method that Hoover et al., 2020 used to test the performance of their annotated
moral foundations Twitter corpus. I ran linear SVM regression with L2 regularization on MFD2 word counts. MFD2 is available
at available at https://osf.io/ezn37/
5
Precision is 0.71, recall 0.93
36

3.3.4.3 Subtopics
Subtopics of mail-in voting, COVID-19 and electoral fraud were identified with relevant keywords. We
selected keywords from a pilot study that helped us gather topic-specific terms from performing different
textual analysis on a subcorpus.
3.3.4.4 User ideology
User ideology was predicted by a state-of-the-art and efficient framework, "Retweet-BERT", developed
by Jiang et al. (2023). The model architecture takes into account both Twitter user profile description
and retweet network structure, combining the advantages of content-based and network-based methods
built for similar tasks. The model achieved a 96% cross-validated accuracy score and outperformed other
popular approaches on a relevant COVID-19 Twitter dataset (Jiang et al., 2023)
6
.
3.3.4.5 Misinformation
Misinformation and trustworthy information were approximated by the source reliability of URL domains
quoted in tweets. Therefore, only tweets that contain URL links were analyzed for this task. Source reliability was rated by third party programs commonly used for differentiating disinformation/misinformation
from reliable information 7
. We used the fake news and unreliable domains provided by (Zimdars &
Mantzarlis, 2016)
8
and Politifact 9
to filter for misinformation, or precisely, information from unreliable
news sources.
6Original author of this paper, Julie Jiang, helped run the user ideology classification algorithm.
7
see for example Sharma et al., 2020
8The list is dynamically updated
9
available at https://www.politifact.com/article/2017/apr/20/politifacts-guide-fake-news-websites-and-what-they/
37

3.3.5 Empirical strategy
We estimated the daily proportion of moralized tweets and 95% confidence interval using moving block
bootstrap (Kunsch, 1989), which is an appropriate non-parametric estimation method for time series data
that contains autocorrelation. We set the moving block to five days. For each moving block, we drew a
N = 1, 000 sample to obtain the sample mean, and repeat the process 1000 times to calculate sampling
means and confidence intervals.
Using the same method, we also estimated the daily proportion of moralized tweets in misinformation
and trustworthy information, and presented their differences in :
δ = rm − rt
(3.1) eq_mis
where rm represents the proportion of moralized content in misinformation, and rt represents the
proportion of moralized content in trustworthy information. If delta > 0, it implies misinformation is
more moralized than trustworthy information.
3.3.6 Results
The results show that topics related to mail-in voting, COVID-19 and voter fraud were clearly more moralized than average throughout the election campaign trail 3.1. The exception is mail-in voting. Its proportion of moralized content experienced volatility in the early stage of the election campaign, when the topic
of mail-in voting was suddenly politicized. Our aggregated time series of daily proportion of moral content
on social media is limited in differentiating the construct of moral recognition from that of moral amplification (Rhee et al., 2019). Nonetheless, the spikes of moralized expression in April and May when Trump
initiated attacks on the legitimacy of mail-in voting on Twitter fit an intuitive pattern of moral recognition.
38

March November
mean SD mean SD
overall 0.605 0.023 0.608 0.036
COVID 0.736 0.026 0.730 0.033
mail-in voting 0.681 0.129 0.706 0.040
fraud 0.795 0.045 0.812 0.019
Table 3.2: Daily proportion of moral content in the first and last month tab:ch3_sum
The topic did not start out as more moralized than average, yet Trump connecting it to a moral violation
of electoral cheating attached moral considerations to an inherently non-moral issue, suggesting a mechanism of moral piggybacking (Rozin, 1999). The topic of COVID-19 had been the most stationary. It was
consistently above the average but did not show patterns of amplification. As Table 3.2 indicates, the average daily moral sentiment of COVID-related posts in the last month of our observation period and final
stage of the election cycle (November) remained at the same level as it was in the first month of our observation period (March). Same static trend goes for the parent topic of the 2020 election, our entire sample.
But for mail-in voting and voter fraud topics, the average moral sentiment experienced a clear increase
comparing the start and end of our study period 10. A simple regression that fit the moving-block average
moral sentiment proportion on time trend indicated a statistically significant upward trend of the fraud
topic (p < 0.01), although no statistically significance were detected for COVID-19 and mail-in voting
topics.
The moral expression related to voter fraud slightly went up as the election approached and after the
moralization of mail-in voting. Despite this topic appearing to be inherently moral with a theme of fraud,
there were non-moral comments referencing Trump’s voter fraud claims without moralizing. For example,
some users cited articles or laid out evidence to explain why rigging the election would be unlikely. Others
made fun of the voter fraud claims (e.g. joking that themselves were committing voter fraud through mailin voting), which conveyed a disparaging attitude towards "voter fraud" claims but did not directly express
10Due to our large sample size, even a minor increase in average is significant
39

(a) COVID-19 fig:covid (b) mail-in voting fig:mail
(c) voter fraud fig:fraud
Figure 3.1: Bootstrapped daily average proportion of moralized content under the topics of (a) COVID19, (b) mail-in voting and (c) voter fraud compared to the entire US 2020 election corpus. Shaded areas
represent 95% confidence intervals. The "overall" trend lines in blue represent the average proportion of
moralized content calculated from the entire corpus. fig:three topics
moral judgments. At the early stage of the campaign trail, not all tweets classified as moral were responding
to Trump’s voter fraud claims. Around March and April, there were tweets posted by liberals questioning
the electoral fairness in the Democratic primaries (e.g. from non-Biden supporters) or expressing the idea
that Trump stole the 2016 election. As time moved forward, the topic became more concentrated on the
controversy that stemmed from Trump’s objections to voting-by-mail expansion. Conservatives joined the
chorus of Trump’s mail-in voting fraud accusations, and liberals disputed Trump’s claims or argued that
Trump was trying to steal the election.
To answer our RQ2 about the temporal distribution of moralized content by user ideology, we plotted
the monthly distribution of moralized content for liberals and conservatives (3.3). Liberals were consistently more moralized when discussing COVID-19. Conservatives were consistently more moralized in
mail-in voting-related comments. These patterns are congruent with the respective positions liberals and
40

(a) Monthly distribution of moral content fig:ch3_trend
(b) Difference in moral sentiment between misinformation and reliable information. fig:ch3_misinfo
Figure 3.2: (a) Proportion of moralized content by month. Shaded areas indicate 95% confidence intervals
based on daily average. The brown line represents the topic of mail-in voting, blue COVID-19, and green
voter fraud. Red is average of the entire corpus (baseline). (b) Difference in moral sentiment between
misinformation and reliable information. The y-axis displays the daily value of δ (3.1) over the study
period fig:stage1
conservatives take on these issues. Liberals were more concerned about people not taking COVID-19 and
public health seriously (Clinton et al., 2020), while conservatives who subscribed to Trump’s theory were
worried about electoral cheating via postal votes.
However, liberals and conservatives were similarly moralized about the electoral fraud topic. Towards
the end of the election cycle, we saw synchronized decreasing temporal variability of moralized content
for liberals and conservatives. The proportion of moralized content gradually trended up and towards a
narrower spread, indicating an increasingly stable level of aggregated moral expression. Accordingly, we
believe the moralized discourse about voter fraud exhibits a pattern of consolidation among both liberals
and conservatives.
Tweets that cited unreliable news sources displayed greater moral sentiment during most of the campaign, which confirms our expectation in H1. It implies that misinformation was more moralized than
trustworthy information in general. Notable themes of misinformation include conspiracy theories such
as QAnon and deep state, in which Trump supporters lamented that a coalition of CIA, FBI, Department
of Justice and mainstream media were conspiring to take him down.
41

(a) COVID-19 fig:ch3_covid (b) mail-in voting fig:ch3_mail (c) electoral fairness/fraud fig:ch3_fraud
Figure 3.3: Monthly density distribution of percent moralized content by user ideology and topic fig:three graphs
3.3.7 Discussion
We examined the patterns of moral expression on Twitter over the course of the 2020 presidential election
campaign and focused on three topics specific to this election cycle. Our initial analysis shows a plausible
moral recognition process for the topic of mail-in voting. As an intrinsically non-moral topic, it became
rapidly moralized during the months when Trump started tying it to a moral concern of election fairness.
The proportion of moral expression had come down since this rapid spike. Instead, we see the steady
amplification of moralized content in discussions about voter fraud and election fairness. These time
series also shed light on the shape of moralized expression at aggregate level. For topics that experienced
an increase of moralization comparing to the start of the timeline (mail-in voting and fraud, see 3.2), they
also displayed smaller variance of daily moral expression over time. The increased average and decreased
temporal variability represent a consolidation of moral temperature for mail-in voting and voter fraud
leading up to the end of the election cycle.
42

3.4 Stage 2
Based on our exploratory analysis in the first stage, the topic of election fraud displays an interesting
shape of moral amplification, with a gradual increase in average and a synchronized decrease in temporary
variability for both liberals and conservatives. On the topic of voter fraud, a majority of moralized tweet
samples claimed that their partisan rivals were fraudulent, about to steal an election or have stolen an
election. Such morally and emotionally charged expressions came from both sides of the political spectrum.
This prompts us to probe into this topic and examine the relationship between partisanship and relevant
moral expression in the second stage of this study. Following a line of research that demonstrates the role
of affect in how moral evaluations form, we pay particular attention to partisan affect against a political
background of growing affective polarization in the United States (Abramowitz & Webster, 2016; Iyengar
et al., 2012; Iyengar & Westwood, 2015).
3.5 The role of affect and emotions in moral judgments
Different paradigms supply different explanations for the formation of moral attitudes and moral judgments, mainly revolving around the reason-or-emotion debate (Monin et al., 2007). Moral psychology
was traditionally dominated by a rationalist perspective derived from cognitive development research and
attributed moral judgment to reasoning and reflection (Haidt, 2001; Kohlberg et al., 1969; Piaget, 1997),
but the role of affect and emotions in moral cognition has been increasingly examined and credited (Forgas & Smith, 2007; Greene et al., 2001; Monin et al., 2007) in the recent two decades. According to a
rationalist approach, moral judgment results from moral reasoning (Kohlberg et al., 1969; Piaget, 1997).
Affect sometimes serves as inputs for moral reasoning but is not a necessary condition or direct cause of
moral judgment (Haidt, 2001). In contrast, an intuitionist perspective highlights the role of intuition and
emotions. Proponents of this approach argue that moral judgments are automatically routed from moral
43

intuitions and feelings, similar to System 1 fast thinking mode proposed by Kahneman (2011); a reasoning
process takes place ex post facto and feeds into future moral intuitions (Haidt, 2001; Hume, 1998). The
contemporary mainstream scholarship at least considers all of affect, intuition and reason in the formative
process of moral judgments (Ditto et al., 2009), with disagreements over their respective causal weights
and positions on the moral evaluation chain.
Apparently, reason, emotions and their interactions are collectively important for moral decision making (Monin et al., 2007). To settle the reason-versus-emotion intellectual conflict, Monin et al. (2007) argues that different "moral encounters" (prototypical situations) activate different moral decision making
processes. In situations where people are presented with complex moral dilemmas and competing moral
claims, such as the trolley problem (Thomson, 1984)
11, rational deliberation and reasoning have been studied as the primary source of moral judgments. On the other hand, emotions are found to motivate moral
judgments when people react to moral transgressions (Monin et al., 2007). Negative affect especially predisposes people to see harm and amplifies moral judgment (Horberg et al., 2011; Schein & Gray, 2018;
Valdesolo & DeSteno, 2006).
This study of voter fraud-involved discussions in a polarized social media environment fits a prototypical situation where people react to what they perceive as moral violations (Monin et al., 2007), from
a candidate committing fraud to spreading conspiracy claims. Emotions have been shown to play an important part in spreading moralized ideas across online social networks (Brady et al., 2017). Therefore, the
role of affect and emotions in arousing moral concerns lends a helpful lens to understanding our research
subject. Nichols (2002) notes that an affective mechanism underlies our cognitive capacity to distinguish
moral violations from non-moral norm violations. Moral judgments are arguably formed upon affective
responses, especially negative affect (Cameron et al., 2015; Schein & Gray, 2018). People’s affective responses to what they deem as wrong often come before they can expound on the reasons behind their
11A thought experiment that presents the ethical dilemmas of saving one person who did not make mistakes versus multiple
people who did.
44

moral judgments (Haidt et al., 1993). Distinct moral emotions, such as anger, disgust, sympathy, are developed from core affect through individuals’ contextualized interpretations thereof, and in turn linked to
specific (e.g. harm, purity and fairness) or domain-general moral concerns (Cameron et al., 2015; Rozin
et al., 1999). Quick, automatic affective reactions also nudge our information organization towards a conclusion that satisfies our preferences, and activate motivated moral reasoning, i.e., assembling arguments
congruent with previously arrived moral conclusions (Ditto et al., 2009). We are motivated to justify these
conclusions, because we have an "affective stake" in a world that "makes good moral sense" (Ditto et al.,
2009; Lerner & Miller, 1978). In a nutshell, affect is a crucial component in both moral emotions and reasoning. As for our study of online moral expression set in the US political context, the role of affective
polarization is particularly relevant.
3.6 Affective polarization and moralistic politics
The rise of affective polarization and especially negative partisanship constitutes one of the most consequential trends in American politics over the last forty years (Abramowitz & Webster, 2016; Abramowitz
& Webster, 2018; Iyengar et al., 2019; Iyengar et al., 2012). Affective polarization is characterized by negative sentiment towards the opposing party and positive feelings towards own party, and it describes a
partisan identity-based cleavage in contrast to ideological or issue position polarization (Abramowitz &
Webster, 2016; Bankert, 2021; Iyengar et al., 2012; Mason, 2013). Negative partisanship, whereby partisans
dislike the opposing parties and candidates more than they like their own, is an outstanding symptom
of affective polarization in contemporary American politics. Partisans increasingly oppose their children
marrying members from the opposing party and resist socializing across party lines (Iyengar et al., 2019;
Iyengar et al., 2012), to an extent that exceeds race-based discrimination (Iyengar & Westwood, 2015).
Negative partisanship is anger-infused, strengthens voters’ party loyalty, makes partisan identities more
salient (Abramowitz & McCoy, 2019; Abramowitz & Webster, 2016; Bankert, 2021), and inflates negative
45

perceptions of the other party (Moore-Berg et al., 2020). The negativity bias is symmetrical across both
major parties (Moore-Berg et al., 2020) and automatic (Iyengar & Westwood, 2015). Many factors have
contributed to the growth of negative partisanship and affective polarization, including exposure to political campaign messages(Iyengar et al., 2012), partisan media consumption, politicization of racial divide
(Abramowitz & Webster, 2018), misconceptions of partisan opponents (Edsall, 2023; Moore-Berg et al.,
2020), and personality types (Abramowitz & Webster, 2016; Webster, 2018).
Negative partisan affect as a politically salient emotion carries particular relevance to our inquiry
into the relationship between affect and moral expression in the run-up to a presidential election. In the
previous section, we have discussed the role of affect and emotions in the formation of moral judgments.
The automatic affective responses when we see moral violations guide us to an intuitive conclusion ex
ante that motivates moral reasoning. This cognitive process explains our moral judgments on people and
groups conditional on our feelings towards them: we tend to attribute morally inferior qualities to people
and groups that we feel negatively about and good moral characters to ourselves and groups we identify
with (Ditto et al., 2009). Negative partisan affect similarly fuels moralized attitudes towards the opposing
party. As a type of social identity-based affect, it primarily influences group-based moral evaluations.
Partisans perceive their own group as morally superior compared to their opposing camps (Cassese, 2021;
Haidt, 2012). The more Democrats or Republicans dislike the other party, the more likely they impute
negative traits, such as mean, selfish, closed-minded, hypocritical, fascist, to their opposing party (Iyengar
et al., 2012). In fact, moral superiority builds ingroup self-concept more than other group merits (e.g.
sociability and competence) (Leach et al., 2007). The still more destructive consequence of group-based
moral bias includes propensity for moral disengagement. Moral disengagement means people deactivate
their moral self-regulation and self-justify harmful actions to others who are considered unworthy of moral
considerations (Bandura, 1999; Bandura et al., 1996; Graça et al., 2016). An alarming example is the rise
of dehumanizing rhetoric and prejudice among partisan rivals in the recent two election cycles (Cassese,
46

2021; Landry et al., 2023; Moore-Berg et al., 2020). Partisans were inclined to consider their opposing party
members as less moral, less human and therefore less deserving of moral considerations (Cassese, 2021).
Partisan identity-based affect can activate motivated reasoning (Peterson & Iyengar, 2021; Taber &
Lodge, 2006) and motivated moral reasoning (Ditto et al., 2009) about political issues and entities. Affective
polarization shapes partisan differences in the interpretation and evaluation of political issues. Resentment
against the outparty (a partisan’s opposing party) drives individuals’ adoption of partisan-based issue positions. When a new issue such as COVID-19 becomes politicized and partisan elites take distinctive sides,
those who hold pre-existing animosity against the opposing party are quickly sorted into issue positions
aligned with their own party’s elites (Druckman et al., 2021). Botvinik-Nezer et al. (2023) finds that the
desire for a preferred candidate to win an election impacts individuals’ posterior beliefs in the procedural
legitimacy of the election. If people’s preferred candidate loses, they are more likely to downgrade their
trust in electoral fairness; vice versa (Botvinik-Nezer et al., 2023). This behavioral pattern is consistent
with the so-called "moral mandate effect": individuals judge the procedural and outcome fairness of an
event based on whether that outcome accords with their moral point of view, and anger about the morally
incongruent outcome triggers the retrospective updates on their beliefs about fairness (Mullen & Skitka,
2006). These examples further demonstrate that people’s differentiated affect towards inparty and outparty
impact their judgments of good and bad and right and wrong.
While negative affect towards partisan objects leads to certain moral convictions about them, the reverse relationship also seems to hold. Moralized attitudes about political parties elicit strong negative
emotions against the opposing camp (Garrett, 2016; Ryan, 2014). Further, the morality dimension has been
understood as an underlying factor of rising affective polarization. In a social media setting, moral outrage activates group antagonism and dehumanization, and these psychological tendencies result in greater
affective polarization (Carpenter et al., 2020). Garrett and Bankert (2020) argue that propensity to moralize politics is associated with stronger partisan bias and hostility towards the opposing party, even after
47

controlling for partisan strength. Being convinced that a partisan object violates oneś foundational moral
beliefs gives rise to partisan moral convictions, which then generate negative emotions against the said
partisan object (Garrett & Bankert, 2020). Like the Moral Foundations Theory (Graham et al., 2013), Garrett and Bankert (2020)’s work views individuals’ basic perceptions of right and wrong as roots of political
ideology or partisan attitudes. Political beliefs stem from a "foundational" layer of moral worldviews.
Evidently, the links between emotion and morality are not unidirectional. In the domain of affective
polarization and moralistic politics, these two elements are also mutually reinforcing. The basic layer of
our moral worldviews shapes our feelings about political parties, conditional on political experience and
knowledge, and these feelings in turn shape our moral evaluations about political parties.
Affect and emotions towards political parties and leaders could intensify along a presidential campaign
trail, especially negative partisan affect, as voters are exposed to repetitive negative messages of the opposing party (Iyengar et al., 2012). Priming negative partisan affect also fans moralized attitudes about
opposing parties and leaders. Clifford (2019) finds that persuasive frames that appeal to certain emotions
– anger and disgust – contribute to moralized political attitudes. Over the course of a presidential election
trail, we should also expect a mutually reinforcing relationship between negative partisan affect and moralized political attitudes. Brandt et al. (2015)’s longitudinal studies of moralization on the 2012 presidential
campaign trail demonstrate that affect is both an antecedent and consequence of moralized attitudes towards presidential candidates. The positive affect of enthusiasm for oneś preferred candidate and negative
affect of hostility against the non-preferred candidate significantly predict changes in moral convictions
about the respective candidates (Brandt et al., 2015). Meanwhile, stronger moral convictions also predict greater affective reactions over time (Brandt et al., 2015). Expecting a similarly positive relationship
and mutually reinforcing dynamics between negative partisan affect and moralized expression online, we
hypothesize:
48

H2: Display of negative partisan affect, including negative affect towards the opposing presidential
candidate and party, predicts a greater propensity for moral expression.
H3: The linkage between negative partisan affect and moral expression is heightened as it draws close
to the general election.
Although we only test for association rather than causation between the two variables, on an election
campaign trail, partisans likely bring pre-existing negative affect towards the opposition, which motivates
moralized expression.
3.6.1 Data and Measures
3.6.1.1 Data
We took the subset of data under the topic of electoral fairness and fraud and selected only tweets by users
who received a clear ideology label based on our user ideology classifier. It totals 58,100 users and 2,431,774
tweets.
3.6.1.2 Negative partisan affect
Negative affect is a generalized term for emotional distress encompassing anger, contempt, fear, disgust,
hostility and other negative emotional states (Stringer et al., 2013; Watson et al., 1988). Negative partisan
affect embodies the same range of negative feelings towards an individual’s out-party.
To approximate this measure, we first used Brady et al. (2017)’s dictionary of negative affect to generate a measure of negative affect, as Brady and collaborators constructed the dictionary for a similar task
that examined moral and emotional expression in tweets 12. Words from the dictionary measure negative emotions (n = 374) based on LIWC’s categories of emotion (e.g. anger), and distinctively emotional
words (instead of moral-emotional) can be selected based on their dictionary of morality (Brady et al.,
12dictionaries available at https://osf.io/59uyz/
49

2017). We normalized the negative affect term frequency by tokenized text length, generating a measure
with a theoretical range between 0 and 1, although the mode falls close to 0.
The partisan negative affect measure was divided into 1) negative affect against the opposing party
(Democrats for conservatives and Republicans for liberals) and 2) negative affect against the opposing
presidential candidate (Biden for conservatives and Trump for liberals).
NegativeP arty = IP ×
wneg
ω
(3.2)
NegativeCandidate = IC ×
wneg
ω
(3.3)
Where IP and IC are indicator variables that represent mentions of the opposing party and the opposing candidate respectively. They are multiplied by the relative (normalized) term frequency of negative
affect. wneg represents the occurrences of negative affect terms, and ω is the count of all tokens.
3.6.1.3 Measure of specific moral dimensions
To test H2 and H3, we also created measures of specific moral foundations. Measures of specific moral
foundations were created using a dictionary method. We used the widely applied Moral Foundations
Dictionary 2.0 (MFD2) 13. The reasons for using a dictionary method is because of its cost-effectiveness
compared to training new multi-class classifiers, given how large the dataset is and computationally expensive it would be to apply classifiers for every moral virtue and vice. Also, it was hard to get a high
accuracy rate for a specific moral foundation, due to the imbalance between the ratios of positive and
negative cases and the difficulty of the task (to recognize specific moral dimensions). The original corpus
developers also trained different classifiers for each moral dimension, and it was common for the accuracy
measure (F1 score) to fall somewhere between .40 to .70 (Hoover et al., 2020). Since we already have an
13available at https://osf.io/ezn37/
50

established dictionary MFD2, and they proved to be predictive features in our previous training task. We
took the count of occurrences of MFD2 terms and normalized it by the length of tokenized text. This gave
us a continuous measure of each specific moral category (e.g. fairness, harm) that has a theoretical range
between 0 and 1.
3.6.2 Empirical strategy
We ran generalized linear regression to test the marginal effects of negative partisan affect on propensity
for moral expression. Tweets were aggregated at the user-level. Because we took the average of every
user’s proportion of moralized tweets and negative partisan expression, only those who posted at least
ten tweets over the study period were kept in the dataset. The final sample size is N = 37, 837 users.
Filtering for relatively active users avoids user-level measures being skewed by those who posted only
one or two tweets. The models regress proportion of moral expression on ideology, measures of negative
partisan affect, and their interaction terms. Control variables such as length and frequency of tweets were
also included for different variants of models. The model is specified as follows:
MoralExpression = β0 + β1Ideology + β2NegativeCandidate + β3NegativeP arty
+ β4Ideology × NegativeCandidate + β5Ideology × NegativeP arty + ϵ (3.4) eq_reg
The dependent variable Y represents moral expression, measured by the proportion of an individual
user’s moralized tweets over a study period. There are three types of dependent variables: 1) "moral"
represents the percentage of a user’s tweets classified as moral expression by our BERT-based classifiers
within a study period; 2) "cheating" represents the averaged relative frequency of words that indicate the
51

cheating category based on MFD2 within a study period; 3) "degradation" is similar to "cheating" but for
the "degradation" category. Each data point corresponds to an individual user.
As aforementioned, NegativeCandidate represents the negative affect towards the opposing partyś
presidential candidate. NegativeP arty represents the negative affect towards the opposing party. Due to
the relatively small magnitudes of these measures (averaged at 0.022 and 0.013) and their variance, they
were re-scaled by multiplying by 10 to avoid drastic scales of regression coefficients.
The dependent variable moral expression was standardized, and both rescaled measures of negative
partisan affect were centered to their means for convenience of interpretation.
To test H2, we picked the best model from the full study-period regression analysis based on the proportion of variance explained by independent variables (R2
) and run it for every three months. As our data
covers nine months, we have three time intervals (labeled as T1, T2 and T3). We choose a three-month
window as an interval unit because each three months mark an important stage of the election. A quarterlong interval also ensures a larger sample of users who posted enough tweets within the time frame than
a shorter interval. The first three months, from March to May, attention was on Democratic primaries
until Bernie Sanders dropped out of the race in April. This is also the time when Trump first suggested
potential voting fraud via mail-in ballots on Twitter and during White House briefing (Parks, 2020). Then
within T2, from June to August, both major parties’ national conventions took place. Trump continued his
voter fraud claims (Shabad, 2020). The final three months cover the post-Labor Day ramp up to Election
Day and the entirety of November, when the results slowly came in and wild "stolen election" accusations
fired up immediately after the Election Day (Qiu, 2020).
3.6.3 Results
Because we expect moral expression in this discourse domain is associated with, if not driven by, negative partisan affect, the moral expression we capture is likely biased towards negativity (moral vices). In
52

Figure 3.4: Distribution of five moral dimensions in voter fraud-related topic fig:ch3_dim_dist
Figure 3.5: Distribution of five moral dimensions in voter fraud-related topic by month fig:ch3_dim_by_month
a preliminary step, we looked at the distribution of moral dimensions based on the Moral Foundations
Theory (Graham et al., 2013), which outlines interpretable, discrete moral content areas (Cameron et al.,
2015). Unsurprisingly, two categories that describe moral vices – cheating and degradation – feature most
prominently in a topic that discusses voter fraud and electoral fairness (see 3.4 and 3.5).
The condemnation of cheating and degradation in the US 2020 election revolves around electoral fraud
and the moral degradation of opponents (e.g. opposing party or candidate being corrupt, see sample tweets
in 3.3). Conservatives were on average more likely to express ideas related to degradation than were liberals
across all nine months. Liberals also led in the condemnation of cheating in the early stage. However,
conservatives overtook liberals in this moral category halfway through our study period in July, most
likely influenced by Trump’s narrative of voter fraud.
53

Moral dimension Ideology Example tweet
cheating liberal trump steals election
@realdonaldtrump trump did steal children’s cancer charity
cheating conservative #electionfraud #bidencheated #breaking biden cheated......pass
#bidencheated
cheating. caught. #voterfraud
degradation liberal new russian actors spreading misinformation mail-in voter fraud
since march
this scumbag @realdonaldtrump relentless unapologetic comes
corruption fraud vote scumbag send straight jail
degradation conservative #democratsarecorrupt #democratvoterfraud #rigged #stopthecount #bidencrimefamily
democrats are corrupt #voterfraud #bidencorruption
Table 3.3: Sample tweets from liberals and conservatives that were rated high on the moral dimensions of
cheating and degradation. Stop-words were removed. tab:ch3_sample
Figure 3.6: Monthly distribution of cheating and degradation by ideology fig:ch3_dim_by_ideo
The monthly distribution shows that expression of negative partisan affect did not necessarily grow
over time on the election campaign trail. Based on our monthly aggregates, liberals’ negative sentiment
against Trump was the most salient category of negative partisan affect, especially during the Democratic
primaries when the anti-trump coalition was a prominent theme and Trump also started denigrating mailin voting around the same time.
Figure 3.7: Monthly Distribution of negative partisan affect fig:ch3_neg_aff
54

Results from generalized linear regression intuitively confirmed our first hypothesis in this study stage
about the relationship between negative partisan affect and propensity for moral expression. Negative
affect towards the opposing party predicts higher likelihood of moral expression for both liberals and
conservatives with statistical significance (see right panel of 3.8). Compared to liberals’ negative affect
towards the Republicans, conservatives’ negative affect towards the Democrats is a stronger predictor of
moral expression (One unit increment of negative partisan affect, meaning 10% increase of negative affect
word usage from the average, increases the propensity for moral expression by 1.25 standard deviation
for conservatives, compared to an 0.57 increase of standard deviation for liberals). This asymmetry is
consistent with evidence from the previous election cycle that demonstrated greater contagion of moralized
content posted by Republican elites (Brady et al., 2019). However, liberals’ negative affect towards Trump
more significantly predicts moral expression than does conservatives’ negative affect towards Biden (see
left panel of 3.8). This is expected, because liberals’ moral outrage against Trump since his ascension to
presidency is well-documented (Jennings, 2019; Meyer & Tarrow, 2018; Tavernise, 2017), and a united front
to vote Trump out of the office was a priority for many starting from the outset of the 2020 presidential
election cycle (Brenan, 2019; Martin & Burns, 2020). Democratic political elites also used more moral
language after Trump was elected in 2016 (Wang & Inbar, 2021).
For the next hypothesis, we expect that the link between negative partisan affect and moral expression
became heightened as the presidential election drew close. Our results show that in the final three months
(T3), at the height of an electoral competition, the association between negative partisan affect and moral
expression for both liberals and conservatives was strengthened and statistically significant, compared to
the previous two time intervals (See 3.10). During T3, every unit of increase in negative partisan affect
expression leads to a 0.48 increase in standard deviation of moral expression for liberals (compared to a 0.08
increase during T1), and a 0.74 increase in standard deviation for conservatives (compared to a 0.36 increase
in T1). As the polarizing controversies about fraud accusations heated up leading up to the Election Day
55

Figure 3.8: Average marginal effects of negative partisan affect on propensity for moral expression. Left
panel displays the average marginal effects of negative affect towards the opposing candidate for liberals
and conservatives respectively. Right panel displays the average marginal effects of negative affect towards
the opposing party. fig:ape
Dependent variable: moral
(1) (2) (3) (4)
liberal 0.113∗∗∗ 0.131∗∗∗ 0.232∗∗∗ 0.207∗∗∗
(0.024) (0.025) (0.026) (0.025)
NegOpposCandidateCentered 0.173∗∗∗ -0.389∗∗∗ -0.547∗∗∗ -0.192
(0.058) (0.118) (0.118) (0.121)
NegOpposCandidateCentered 0.173∗∗∗ -0.389∗∗∗ -0.547∗∗∗ -0.192
(0.058) (0.118) (0.118) (0.121)
liberal:NegOpposCandidateCentered 0.832∗∗∗ 0.866∗∗∗ 0.505∗∗∗
(0.136) (0.135) (0.138)
liberal:NegOpposPartyCentered -1.139∗∗∗ -1.041∗∗∗ -0.679∗∗∗
(0.171) (0.169) (0.172)
length -0.041∗∗∗ -0.046∗∗∗
(0.002) (0.002)
TweetCount 0.000∗∗
(0.000)
Mail 0.629∗∗∗
(0.058)
COVID 0.370∗∗∗
(0.103)
Intercept -0.042∗∗∗ -0.118∗∗∗ 0.974∗∗∗ 1.020∗∗∗
(0.010) (0.017) (0.051) (0.052)
Observations 37,837 37,837 37,837 37,837
R2 0.017 0.020 0.042 0.047
Adjusted R2 0.016 0.020 0.042 0.047
Residual Std. Error 0.992(df = 37833) 0.990(df = 37831) 0.979(df = 37829) 0.976(df = 37828)
F Statistic 148.878∗∗∗ (df = 3.0; 37833.0) 121.939∗∗∗ (df = 5.0; 37831.0) 175.110∗∗∗ (df = 7.0; 37829.0) 174.013∗∗∗ (df = 8.0; 37828.0)
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01 tab:ch3_reg
Table 3.4: Regression table
56

Figure 3.9: Average marginal effects of negative partisan affect on propensity for moral expression, varying dependent variables. Error bars of different colors represent the average marginal effects of negative
partisan affect on different measures of the dependent variable. Moral was the same measure as 3.8. Cheating and degradation measure the respective moral categories, which are the most prominent ones in this
discourse domain. fig:ch3_q_all_dv
57

Figure 3.10: Average marginal effects of negative affect towards the opposing party on moral expression
over three time periods. The nine-month study period was divided into three intervals: T1 covers the first
three months (March, April and May); T2 covers June to August; T3 covers September to November. fig:ch3_q_ape
and continued to deteriorate post-election, negative affect towards the opposing party could be playing a
bigger role in driving moral expression during this period, when the daily trend of moral expression went
stably higher. However, the marginal effects of negative affect towards the opposing candidate were not
statistically significant (except for liberals against Trump in T1) and showed slight decrease towards the
end of the election cycle (from 0.30 in T1 to 0.09 in T3 for liberals against Trump and from 0.02 in T1 to
-0.08 in T3 for conservatives against Biden).
As we turn to the two prominent moral categories as dependent variables, conservatives’ expression
of cheating and degradation responded increasingly significantly to negative partisan affect over time.
The average marginal effects of negative partisan affect on condemnation of cheating for conservatives
increased from 0.14 in T1 to 0.36 in T3, and notably on condemnation of degradation from 1.84 in T1
to 3.39 in T3 (illustrated in 3.11). The marginal effects of negative affect against Biden for conservatives
58

Figure 3.11: Average marginal effects of negative affect towards the opposing party on expression of degradation fig:ch3_q_ape_dgd
also increased from T1 to T3, especially in the case of cheating expression. This makes sense since the
accusations of electoral fraud from Trump supporters reached the highest level in the final three months
of our study period (T3). The narratives of Democrats being corrupt or Democrats destroying America
were common occurrences. Although not as significantly as it is for conservatives, the positive association
between negative partisan affect and expression of cheating and degradation grew stronger for liberals as
well, most so in the case of negative affect against Trump and condemnation of degradation (marginal
effects grew from −0.10 in T1 to 0.32 in T3, p < .05).
3.6.4 Discussion
The scope of this study does not cover data on the pre-existing negative partisan affect and moralized
political attitudes, which would be better measured by validated survey scales. Therefore, we are unable
59

to verify the direction of impact. We have confirmed, however, a clear positive relationship between expression of negative partisan affect and moralized ideas, which was strengthened in the final stage of the
campaign trail. Based on previous discussion of literature, these two variables could be mutually reinforcing on an election trail, when both partisan camps bring existing negative partisan affect and moral
convictions.
The political discourse surrounding election fraud we discussed in the first study stage shows that some
liberals and conservatives held prior beliefs that the other party would cheat in an election, even before
Trump started the voter fraud conspiracy claims. Their negative affect was converted into or stemmed
from the moral beliefs that the other party was fraudulent, corrupt, and anti-democratic. This view was
further amplified as the election drew close and after the voter fraud accusations blew up. Moral expression
became the norm in this discourse domain (electoral fraud) at the end of the study period, and it revealed
polarizing interpretations of the election results and voter fraud claims. From a partisan’s perspective,
this could be a time period when their prior beliefs about the opposing party’s moral characters were
confirmed.
This study further demonstrates the relationship between emotion and morality in the political realm
through a dynamic lens. Various concerns regarding the negative consequences of affective polarization
have been raised. And how it produces and reinforces moralistic politics is an equally important political implication. The potential of motivated moral reasoning and negativity bias about the opposition’s
moral characters would elevate the difficulty of compromise and continue the trend of diverging partisan
worldviews.
3.7 Conclusion
This chapter has tracked the moralization patterns of online political discourse over the course of a presidential campaign trail. We observed the spikes in moral expression related to the previously non-moral
60

issue of voting by mail after Trump piggybacked it onto the problem of voter fraud. Discourse around
voter fraud saw steady growth in moral expression for liberals and conservatives ever since, demonstrated
by increased average and decreased temporary variability of moralized content.
In the second phase of the study, we tested the relationship between affective polarization and moral
expression in comments about election fraud. We confirmed a positive association between negative partisan affect and moral expression and the increased strength of this relationship in the final stage of the
election cycle. It lends support to the hypothesis that existing negative attitudes against partisan opponents amplify moral expression over the course of an election campaign trail. The effects are stronger for
conservatives, whose negative affect against the Democrats incrementally predicts their moral condemnation of "cheating" and "degradation" over three successive periods. The dislike for opposing parties and
leaders makes it easier to accept the conclusions that they have questionable morals.
Our study that confirms a growing tie between negative partisan affect and negative moral expression
agrees with the recent observation regarding a spiral of negativity bias and negative stereotypes about
partisan rivals (Edsall, 2023). Growing evidence points to the phenomenon of a political divide based on
misperceptions (Baharaeen, 2023; Edsall, 2023). Democrats and Republicans hold biased understanding
of the demographic composition of the opposing party (Edsall, 2023), overestimate the distance of issue
positions between them (Baharaeen, 2023; Levendusky & Malhotra, 2016) and overestimate partisan opponentswillingness to violate democratic norms and commit violence (Braley et al., ´ 2023; Mernyk et al.,
2022). The widespread beliefs of voter fraud post-2020 election illustrate this point, leading to undemocratic collective action to stop the misconceived election fraud. Social media further adds to the mutual
misunderstanding, as it tends to inflate perceptions of moral outrage and intergroup hostility (Brady et
al., 2023). An information environment that reinforces negative partisan stereotypes and misperceptions
provides a fertile ground for moral outrage and condemnation. In our case study of the voter fraud topic
61

leading up to the 2020 election, such moral condemnation expressed beliefs that the other party was corrupt and willing to cheat. A spiral of negative moral perceptions from opposing parties holds implications
for public trust in the electoral system and process in the long run.
Several limitations can be identified in this study. First, social media represents a biased sample of political discourse. Despite the large sample size, Twitter data likely overrepresented voices of the politically
engaged and opinionated group. Political expression on social media also has a questionable correspondence with people’s private beliefs (Joseph et al., 2021). Second, our automated measures of moral content
are prone to errors if a user only embeds their moral expression in a string of hashtags instead of a complete statement. The proxy of negative partisan affect could also benefit from more accurate and scalable
measures than a dictionary-based approach. Finally, observational studies based on content data are not
the best at capturing conceptually meaningful distinctions such as moral judgments and moral emotions,
as well as de facto changes in beliefs and attitudes. It is a temperature and mood check based on the rise and
fall of moral expressions. We are also limited in making causal claims based on this study. Future research
could use causal inference methods to better understand the relationship between negative partisan affect
and moralized political attitude.
A key takeaway from this study is the association between affective polarization and beliefs of novel
conspiracies such as voter fraud, which can serve to satisfy a psychological need and validate the idea
that "people I dislike have inferior moral character". It motivates future research questions and hypotheses
regarding the relationship between the propensity to moralize one’s opposing party and the likelihood to
accept conspiracy theories and misinformation. These questions can be explored through deeper analysis
of users who posted misinformation based on this dataset. New studies can better address these questions
through surveys and interviews. Further, this implication invites research designs that test the effects of
counterframing strategies that "de-moralize" partisan objects and attack partisan-based morality bias (see
for example Puryear et al., 2022).
62

Chapter 4
Partisan bias in inflation news narratives
ch:friends
4.1 Abstract
As the American inflation rate reached its highest record in forty years, inflation became a politically
contested issue in the recent two years. This chapter investigates the phenomenon of rising partisan
differences in inflation expectations and traces one of its potential sources in partisan media narratives.
Democrats and Republicans used to have similar economic assessments but have demonstrated a growing gap in their short-term economic expectations in recent decades, conditioned by who is the occupant
of the White House. I argue that consistent partisan news consumption, not simply the relatively stable
party identification, is an important variable to explain this widening gap. In this study, I first presented
evidence that Democrats and Republicans were significantly likely to hear different tones of economic
news, conditional on party affiliation of the president. Further, the tones of news partisans heard demonstrated an amplifying effect on their inflation expectations. Based on these findings, I compared narratives
of inflation drivers presented in three major cable news networks, Fox News, CNN and MSNBC, over a
three-year span through computational methods of narrative extraction. Partisan cues were prominently
featured in Fox News, to a lesser extent in MSNBC. Fox News and MSNBC both presented ideologically
aligned frames of inflation drivers, with Fox News distinctively blaming inflation on excessive government
spending and MSNBC on corporate greed. These two narratives presented in partisan cable TV could in
63

part explain similar patterns in popular beliefs on what caused inflation. Fox News presented the densest
partisan cues and ideological anchors while CNN appeared the most balanced. This study bridges media
effects and literature on the political conditioning of economic perceptions and introduces the potential
moderating factor of partisan news consumption in the formation of public economic perceptions. Diverging partisan media narratives and their impact on voters’ economic reasoning have implications for the
role information plays in a rational voter’s preference formation.
4.2 Introduction
The American inflation rate reached its highest level in 40 years entering the year of 2022. The year-overyear increase of Consumer Price Index for All Urban Consumers (CPI-U) climbed to 9.1% in June 2022, the
largest 12-month increase since 1981. 1 The timing coincided with the 2022 midterm election campaign
trial, making inflation a top voter concern prior to the midterm elections. 2 The issue of inflation thus
gained significant national attention and political weights in the 2021-2022 period and stoked partisan
efforts to influence the narratives especially before the midterm elections. 3
As much as economic news is predicated on hard data, ordinary voters’ economic assessments have
increasingly fallen susceptible to politically motivated narratives. Democrats and Republicans used to have
similar economic assessments under different administrations (Krugman, 2022), but empirical evidence in
the recent decade reveals a persistent partisan gap in economic expectations (Bachmann et al., 2021; Curtin,
2018; Gerber & Huber, 2010). The effects of partisan bias on economic evaluations are particularly strong
around elections (Gerber & Huber, 2010; Mian et al., 2021). Data from Surveys of Consumers shows clear
1
From U.S. Bureau of Labor Statistics: https://www.bls.gov/opub/ted/2022/consumer-prices-up-9-1-percent-over-the-yearended-june-2022-largest-increase-in-40-years.htm
2
see for example, a Gallup survey in early 2022 showed one out of five Americans regarded inflation as the top
national concern: https://news.gallup.com/poll/391220/inflation-dominates-americans-economic-concerns-march.aspx; in another 2022 Economist/YouGov poll, "the prices of goods and services" were considered the best measure of the wellbeing of the national economy by 54% of the subjects, compared to "unemployment rate" by only 19% of the subjects:
https://docs.cdn.yougov.com/vkklxli0e8/econTabReport.pdf
3
see for example: https://thehill.com/business/3460548-republicans-democrats-push-dueling-economic-narratives-ahead-ofmidterms/
64

Figure 4.1: Average 12-month inflation expectations by party identification, January 2020 to December
2022. Grey dashed line marks November 2020 when Biden was named the president-elect, on track to
replace Trump in the White House. Shaded areas indicate standard errors. Source: Surveys of Consumers
– University of Michigan. fig:ch4_inflation_expectation
patterns of partisan inflation assessments (see 4.1). Democrats on average held slightly higher inflation
expectations than Republicans prior to the 2020 presidential election, and this trend has been drastically
reversed since Biden was elected, followed by the U.S. entering record levels of inflation.
In this chapter, we investigate this phenomenon and trace its potential origin in mainstream media
narratives. We start by validating the presence of partisan differences in the types of economic news they
hear, and then compare how inflation narratives diverged in partisan media coverage.
4.3 Media and political conditioning of economic assessments
Krugman (2022) in one of his New York Time opinion pieces on inflation noted that consumer sentiment on
the national economy was downright negative while being optimistic about personal finances. Evaluations
of personal finances are inseparable from individuals’ lived experience. The optimism in this regard was
supported by numbers that speak to rapid wage growth, albeit outpaced by inflation, and low unemployment rates. People’s evaluations of the national economy, however, come from what they hear. He argued
that the differential evaluations of personal finances and national economy can be explained by the "power
65

of narratives" (Krugman, 2022). In the story of inflation expectations, politically motivated narratives serve
as an important factor to understand the partisan differences we discussed in the introduction section. Extant research on the political conditioning of economic attitudes found that the economic narratives from
opposition parties during times of high inflation and unemployment would particularly undermine voters’
confidence in the national economy (Mutz, 1994; Pardos-Prado & Sagarzazu, 2016). Based on a case study
of Spain, Pardos-Prado and Sagarzazu (2016) further corroborated that voters’ subjective evaluations of
national economy are not simply a function of stable traits like party identification; they are also affected
by dynamic partisan cues communicated from political elites.
Nonetheless, media narratives are not a commonly included variable in research on the political conditioning of economic assessments, although mass media can affect policy preferences (Dalton et al., 1998;
Noelle-Neumann, 1974; Zaller, 1996), political learning (Damstra, 2019; Prior, 2007), and plays an important part in the "spiral of political reinforcement" (Berelson et al., 1986). Media effects could also be
significantly underestimated due to measurement errors (Bartels, 1993). In this study, we examine the
narratives advanced by mass media because, as the most important source of information, news media
mediates citizens’ exposure to and participation in political discourse (Bennett & Entman, 2000; Shehata
& Strömbäck, 2014). Narratives and discourse, whether flowing from government to citizens (Fan, 1988;
Zaller, 1992), or the other way around (Stimson et al., 1995) (esp. with the contemporary "populist turn"
of media and politics (Davis, 2019)) are collected, organized, edited, and transmitted by various forms of
media. Institutional media passes on political messages from elites, but its role extends beyond a neutral
transmitter. It more directly influences the masses with its language use (Simon & Jerit, 2007) and message construction dictated by media logic (e.g. according to what is newsworthy) (Altheide, 2004, 2015).
Even political parties and elites are conditioned to package their terms under media logic in the age of
mass media (Meyer & Hinchman, 2002). As Dupoirier et al. (2021) recommended, how individuals cognitively process an emerging issue according to their a priori political anchoring should be studied through
66

the social communication of such anchors, especially news media (Dupoirier et al., 2021). Mass media is
the starting point of social communication and influences the messages being transmitted across social
networks (Lazarsfeld et al., 1968).
The types of media partisans consume can change their perceptions of politically salient issues (Broockman & Kalla, 2022; Guess et al., 2021). Inflation became one of such issues as it rose in 2021 and continued
to hit historical levels prior to midterm elections in 2022. Meanwhile, striking partisan patterns in news
media consumption already exist in the American political context. For example, CNN, the most trusted
source among Democrats, is the most distrusted for Republicans; in perfect contrast, Republicans’ most
trusted source Fox News is the most distrusted source for Democrats (JURKOWITZ et al., 2020). It is generally well-known in the contemporary age that Democrats and Republicans trust and consume different
news sources, and this divide widens over time (JURKOWITZ et al., 2020). Therefore, we anticipate that
Democrats and Republicans hear different news about the economy and that news consumption amplifies
partisan differences in inflation expectations.
H1a: Democrats and Republicans hear different economic assessments from news. Republicans are
more likely to hear negative reports on inflation and the economy.
H1b: The types of news partisans hear amplifies existing partisan differences in inflation expectations.
4.4 Partisan cues, ideological anchors, and political activation
Messages from political elites and media can activate partisan identification and beliefs, making them
relevant in the evaluations of an emerging issue such as inflation. As Democrats and Republicans have
developed into two ideologically coherent and distinct parties over the past few decades (Noel, 2012), and
American voters are sorted into their own party’s package of issue positions (Levendusky, 2009), voters
receive clear, repetitive party signals regarding where their party stands (Levendusky, 2010). Partisan
67

cues in this polarized context offer a low-effort heuristic for audiences to process political information
(Campbell et al., 1960; Kam, 2005; Peterson, 2019; Rahn, 1993) and adopt value-consistent preferences
(Bakker et al., 2020).
For voters who do not engage with politics ideologically, they look to social groups with whom they
identify for clues, including race, gender, religion and political parties (Berelson et al., 1986; Conover, 1988;
Converse, 1964; Dawson, 1995; Kinder & Kalmoe, 2017; Westwood & Peterson, 2020). Partisan identification, as group-based psychological attachments, harbors a sense of belonging with a political party and
the image it conjures (Bartels, 2002; Green et al., 2004; Greene, 2004; Groenendyk, 2013). It remains stable
even after party platform reshuffling (Green et al., 2004) and sometimes trumps attitudes based on ethnic
identity (Kam, 2007). Partisans employ motivated reasoning (Bakker et al., 2020; Gaines et al., 2007; Petersen et al., 2013; Taber & Lodge, 2006) or rely on partisan cues and stereotypes in political information
processing (Barber & Pope, 2019; Rahn, 1993). Partisan cues are the most effective in activating party identification and partisan-motivated reasoning when a novel issue is being incorporated into partisan politics
(Nicholson & Hansford, 2014).
Although ideology and partisanship are often aligned in the contemporary U.S. context, ideology functions differently from partisan cue-taking in political information processing. A political ideology ties together political ideas, especially an individual’s various issue positions (Noel, 2014). These idea elements
are not tied together arbitrarily but through a belief system with logical constraints (Converse, 1964). And
by that standard, most people are not ideologues, for they either do not commit to or cannot articulate the
set of abstract principles that coherently order their policy preferences (Barber & Pope, 2019; Converse,
1964; Kinder & Kalmoe, 2017; Sniderman et al., 1991).
Nonetheless, voters who are not ideologues can still take ideological cues from political parties, media,
as well as other types of opinion leaders. Political parties would instrumentally package issues in ideological terms so that voters understand the relations among issues and the differentiation between parties
68

better (Levendusky, 2009). Citizens who are unable or unmotivated to master a political ideology can still
behave quasi-ideologically with reasoning aids from opinion leaders (Sniderman et al., 1991). They may
affectively arrive at a policy preference first, and then acquire the rational explanations for the preference
from elites or society. The acquired explanations are “scripted” causal attributions of issues (Sniderman et
al., 1991), providing ideological anchors for individuals to connect causes and conclusions. In other words,
rather than following the sequence of ideological reasoning that starts from abstract principles to specific
policy preferences, the less politically sophisticated reason backwards by filling in the logical link to their
preferences with prepackaged attributions that are made available to them (Sniderman et al., 1991).
Therefore, media and elite messaging is important in the formation of public policy preferences and
political cleavages. For one thing, it releases signals regarding where a voter should stand if they identify
with a political party, i.e. partisan cues. Furthermore, it connects idea elements that are otherwise not
connected in an average citizen’s mind. By supplying "scripted causal attributions", persuasive messages
help people rationalize their pre-existing preferences derived from affect and intuitions, making the crucial connections between idea elements and preferences. Effective elite messaging “activates” preexisting
attitudes and makes them relevant to elections, i.e. “political activation” (Bartels, 2017; Sides et al., 2019).
Narratives about national economic and business conditions can persuade partisan audiences through
both partisan cues and ideological anchors, activating their existing attitudes against the opposing party’s
positions, platform or stereotypes.
4.5 Priming and framing
Media messages could facilitate political activation through framing and priming. Both describe instruments of persuasion through the selective use of information among possible alternatives. Priming influences the audience by making certain considerations accessible in memory. In political communication, it
is defined as “changes in the standards that people use to make political evaluations” (Iyengar and Kinder,
69

1987, p.63). Priming occurs when people are made to realize they should use specific considerations to evaluate the performance of political leaders and governments (Scheufele & Tewksbury, 2007), for instance. It
is considered a memory-based model. When it comes to our subject of interest, the media could prime partisan identity to make it a salient consideration in the assessments of inflation. Whereas most economists
would explain the causes of inflation from textbook supply-side and demand-side factors (Andre et al.,
2023; Burns, 2022), households often name government incompetence as a key driver of inflation (Andre
et al., 2023). It is reasonable to speculate that households received partisan cues from their information
sources.
Framing influences the audience’s cognition through a different mechanism. Rather than making considerations accessible in people’s memory, framing activates existing interpretive frameworks of the audience and makes them applicable in issue evaluations (Cacciatore et al., 2016; Goffman, 1974; Nelson
et al., 1997; Price et al., 1997; Scheufele & Tewksbury, 2007; Tversky & Kahneman, 1981). Cacciatore et
al. (2016) traced the origin of framing theory and reported a distinction between the psychology-rooted
definition and the sociology-rooted definition of framing and frames. The psychology-rooted definition
of frames characterizes the presentation of logically equivalent information (Cacciatore et al., 2016), such
as the same information framed as gains versus as losses (Tversky & Kahneman, 1981). The sociologyrooted definition of frames is built around the selection and highlighting of certain aspects of information,
rather than choosing between logical equivalents of the same information (Cacciatore et al., 2016). This
has expanded the definition of a frame to a cognitive template that structures information in our minds,
similar to the concepts of "schemas" and "scripts" (Cacciatore et al., 2016). It is difficult to manipulate logical equivalents in a text-based analysis this study conducts, say presenting death toll as lost lives versus
saved lives (Tversky & Kahneman, 1981). The definition of frames we use is more aligned with the concept
of schemas, which are cognitive structures derived from the organization of prior knowledge and ready to
interpret new information (Conover & Feldman, 1984; Fiske & Linville, 1980; Hunzaker & Valentino, 2019).
70

In the context of inflation narratives, frames can activate existing schemas that correspond to different
economic ideologies in the left-right party politics. Not unlike how liberals and conservatives attribute
poverty to different entities (individual attributes or social circumstances) (Hunzaker & Valentino, 2019),
they also have regular go-to targets of blame for economic issues broadly fitting under capitalism/big
corporations and big government respectively. These knowledge structures are consensual schemas shared
by many (Conover & Feldman, 1984; Fiske & Kinder, 2017). These particular examples of liberal and
conservative schemas correspond to "scripted causal attributions" (Sniderman et al., 1991) we discussed in
the previous section. Once the general public acquires these scripted attributions from opinion leaders,
they acquire cognitive frameworks that can be activated to evaluate new information.
RQ1: How do partisan media discuss the attributions of inflation differently? Do they emphasize
ideologically aligned interpretive frameworks?
Since my focus is on partisan media, we expect the presence of partisan cues, especially in rightleaning media, who are motivated to criticize the economic performance under the incumbent Democratic
administration.
H2: Partisan cues are salient in partisan media’s narratives of inflation drivers. In particular, rightleaning media are likely to prime Biden’s responsibility for inflation.
4.6 Cable television as a venue to study inflation narratives
To compare inflation narratives between media of different political leanings, I conducted content analysis
on cable television news. We have two primary reasons to select cable networks as my main data source:
1) the format of television news facilitates narrative generation compared to other institutional media; 2)
the user profile of cable television matches the target audience of political persuasion and activation we
discussed above.
71

Mass media mediates political discourse in the formats of information and entertainment (Altheide,
2004). Television upon its arrival vastly enabled this "infotainment" feature of mass media with its dynamic
audiovisual capability, fast pace, less in-depth news coverage, as well as a host of comedies and talk show
programs. Its programming is oriented towards generating narratives rather than sound arguments and
hard facts (Altheide, 2004). News consumers who primarily acquire information from television were found
to be less knowledgeable about political facts than consumers of newspapers and magazines (Chaffee &
Kanihan, 1997; McLeod & McDonald, 1985). This increases their chances to be persuaded than those who
formed their attitudes on well-considered grounds (Zaller, 1992).
Television had been the primary source of news diet for American adults since the 1960s (Chaffee &
Kanihan, 1997; Roper, 1983) and remains the most preferred source of information among traditional media
channels (compared to print and radio), with 65% adults checking television for news at least sometimes
(Matsa & Naseer, 2022). Not to mention that news content produced by TV broadcasters continues to
circulate in the digital spaces, from video clips uploaded by TV networks to their official Youtube channels
to cuts edited and shared by users on Tiktok and Twitter. Among TV broadcasters, cable news channels
such as Fox News, CNN, and MSNBC hold a relatively more partisan reputation than network TVs such as
ABC, CBS, and NBC, whose broadcast licenses were once subjected to the Fairness Doctrine 4
. Republicans
also trust ABC, CBS, NBC more than CNN and MSNBC 5
. Partisan TV viewers are on the rise despite the
decline of overall TV viewership, and they are less likely to break out of their partisan news diets (Muise
et al., 2022). Partisan cable TV has observable effects on viewers’ political opinions, as latest research
by Broockman and Kalla (2022) demonstrates that switching from Fox News to CNN changed viewers’
opinions on major issues after a month. Due to their loyalty, partisan cable TV viewers are more likely to
be familiar with the common narratives and frames presented by their preferred channel. Thus, framing
would have greater effects on these audiences (Nelson et al., 1997).
4
https://www.usatoday.com/story/news/factcheck/2020/11/28/fact-check-fairness-doctrine-applied-broadcast-licenses-notcable/6439197002/
5
https://www.pewresearch.org/journalism/2020/01/24/u-s-media-polarization-and-the-2020-election-a-nation-divided/
72

4.7 Data and measures
4.7.1 Cable television news dataset
Cable television was selected to represent partisan media content. Data from major cable television networks was collected to represent media content. I selected the top three cable television news networks:
Fox News, CNN and MSNBC 6
. These three networks represent the US political spectrum from right to
left, and had also displayed growing ideological distance over the decade of 2010 to 2020 (Kim et al., 2022).
I downloaded the transcripts of Fox News, CNN and MSNBC between January 1 2020 and December 31
2022 from the LexisNexis Academic database via the university library, covering a three-year length of
time and the accelerating process of inflation. To filter for news stories that focus on inflation instead of
mentioning it in passing, I restricted the search to transcripts in which "inflation" occurs at least five times.
The resulting sample size is N = 3329 records.
4.7.2 Inflation data
Inflation measures were included as complementary information to understand the temporal patterns of
inflation narratives. Monthly Consumer Price Index (CPI) series was downloaded from the U.S. Bureau of
Labor Statistics 7
. Year-over-year CPI growth rate was computed as the measure of inflation rate.
4.7.3 Inflation expectations
Data of consumers’ inflation expectations along with demographic information was obtained from Surveys of Consumers at the University of Michigan. Surveys of Consumers publish monthly indicators of
consumer sentiment and expectations based on a nationally representative sample 8
.
6Cable News Fact Sheet: https://www.pewresearch.org/journalism/fact-sheet/cable-news/ by Pew Research Center Fact
Sheets: State of the News Media
7The time series downloaded is Consumer Price Index for All Urban Consumers (CPI-U), Seasonally Adjusted:
https://beta.bls.gov/dataViewer/view/timeseries/CUSR0000SA0
8Available at https://data.sca.isr.umich.edu/
73

The measure we used was short-run (one-year) inflation expectations. Survey participants were asked:
"During the next 12 months, do you think that prices in general will go up, or go down, or stay where they
are now? By about what percent do you expect prices to go (up/down) on the average, during the next 12
months?" 9
4.8 Analytic strategy
We used a difference-in-difference design to test the relationship between partisanship and the likelihood
of hearing negative news about the economy. The 2020 election (November 2020) is treated as an intervention, after which the White House switched parties (or was anticipated to switch parties soon). We
regard self-identified partisans (Democrats and Republicans) as treatment groups, because their respective party’s control of the presidency changed in the "post-treatment" period. Meanwhile, self-identified
independents or those who did not know their party affiliations belong to the control group. We looked at
whether Republicans were more likely to be exposed to unfavorable news about the economy compared to
Democrats post-2020 election. The dependent variable is the log odds of hearing favorable and unfavorable
news, which is provided by Surveys of Consumers.
To validate the interaction effects of news reception and partisanship on consumers’ short-term inflation expectations, we followed a similar difference-in-difference design except that the dependent variable
is inflation expectations. Partisans are considered as treated groups and the rest are in the control group.
An intervention occurred in November 2020. Specified models are in Table 4.2.
4.8.1 Detection of narratives
To answer RQ1 and examine how information about inflation was presented in partisan media, I conducted
content analysis on cable news transcripts. I applied Ash et al. (2023)’s open-source package RELATIO for
9Variable label "PX1": https://sda.umsurvey.org/sca/Doc/sca.htm
74

Figure 4.2: Example agent-verb-patient triplet
the task of narrative extraction. RELATIO was designed to extract the narrative structures in the form of
"who does what to whom" using unsupervised learning (Ash et al., 2023, p.1). It has been productively
applied to quantify and represent political and economic narratives (Ash et al., 2023; Lange et al., 2022;
Sipka et al., 2022).
RELATIO uncovers narrative structures through semantic role labeling (Ash et al., 2023). Semantic roles
are old linguistic models that describe the thematic purposes of entities in an argument, for example, the
causer, experiencer, or result of an event (Baker et al., 1998; Jurafsky & Martin, 2021). The NLP approach
of semantic role labeling combines syntactic analysis with semantic inferences (concepts and relations
between concepts) (Bonial et al., n.d.). It is a useful tool for frame and argument analysis. RELATIO was
built on AllenNLP (Gardner et al., 2018)’s semantic role labeling models and extracts narratives from the
simple semantic role structure composed of AGENT, VERB, PATIENT, where a subject “AGENT” initiate an
action to a target “PATIENT”.
Not all statements contain all three semantic roles. For example, some statements do not have a patient. In my media narrative analysis, I filtered for sentences with complete agent-verb-patient sequences.
Additionally, based on my interest in the rhetoric of inflation attributions, I selected verbs that signal causation (such as cause, drive and fuel). The interpretive frames of interest were therefore operationalized
into the sequence of agent-(causal)verb-patient. In my focused dataset 10, it identifies most of the claims
about inflation drivers. Narrative extraction helps to capture relations between entities, which are more
interpretable and give us a fuller story arc than NLP models that treat words and phrases as discrete entities.
10I preprocessed the inflation news dataset to keep only paragraphs where the word "inflation" occurs, since it is not uncommon
for multiple news topics being featured in a single episode.
75

Figure 4.3: Narrative extraction pipeline using RELATIO. Flowchart developed based on Sipka et al. (2022)
and Ash et al. (2023)
fig:ch4_pipe
(a) unfavorable news fig:ch4_unfav (b) favorable news fig:ch4_fav
Figure 4.4: Proportions of Democrats, Republicans and Independents/unknowns who heard favorable and
unfavorable news about business conditions before and after the 2020 election. Grey dashed line indicate
the month of 2020-11 fig:ch4_news
4.9 News consumption and inflation expectations
Our H1a tests whether partisans tend to hear unfavorable news about the economy during the opposing
party’s administration. Descriptive evidence (4.4) already illustrates that Republicans were more likely to
hear unfavorable news about the economy after the transition of power in White House in January 2021,
and Democrats were less likely, relative to Independents. These patterns were reversed during the Trump
administration. In a similar vein, a consistently greater proportion of Democrats and a smaller proportion of Republicans heard favorable news about the business conditions after the presidential transition,
contrary to the patterns during the Trump administration.
Logistic regression results also support this finding. Republicans were significantly more likely, and
Democrats less likely, to hear unfavorable news about economic conditions after the election in November
76

2020 and an anticipated power transition (p < 0.01). On the other hand, Democrats were significantly
more likely, and Republicans less likely, to hear favorable news about economic conditions in the posttreatment period (p < 0.01). The results hold after controlling for whether an individual received college
education or adding month fixed effects. These patterns were reversed in the pre-treatment period (prior
to November 2020, see 4.1).
Dependent variable
Heard unfavorable news Heard favorable news
(1) (2) (3) (4)
Intercept 0.445∗∗∗ 0.283∗∗∗ -1.548∗∗∗ -1.777∗∗∗
(0.045) (0.050) (0.057) (0.069)
Post -0.205∗∗∗ -0.199∗∗∗ 0.189∗∗∗ 0.200∗∗∗
(0.052) (0.052) (0.066) (0.067)
DEMOCRAT 0.640∗∗∗ 0.255∗∗ -0.793∗∗∗ -0.605∗∗∗
(0.071) (0.101) (0.103) (0.167)
REPUBLICAN -0.403∗∗∗ -0.379∗∗∗ 0.575∗∗∗ 0.754∗∗∗
(0.068) (0.087) (0.081) (0.106)
Post × REPUBLICAN 0.834∗∗∗ 0.782∗∗∗ -1.072∗∗∗ -1.157∗∗∗
(0.081) (0.099) (0.100) (0.125)
Post × DEMOCRAT -1.025∗∗∗ -0.624∗∗∗ 1.474∗∗∗ 1.224∗∗∗
(0.082) (0.115) (0.112) (0.176)
COLLEGE 0.320∗∗∗ 0.417∗∗∗
(0.046) (0.058)
DEMOCRAT × COLLEGE 0.546∗∗∗ -0.356∗
(0.123) (0.193)
Post × DEMOCRAT × COLLEGE -0.652∗∗∗ 0.358∗
(0.131) (0.196)
REPUBLICAN × COLLEGE -0.038 -0.316∗∗
(0.111) (0.127)
Post × REPUBLICAN × COLLEGE 0.146 0.163
(0.124) (0.150)
Observations 19,375 19,375 19,375 19,375
R2
Adjusted R2
Residual Std. Error 1.000(df = 19176) 1.000(df = 19171) 1.000(df = 19176) 1.000(df = 19171)
F Statistic (df = 5; 19176.0) (df = 10; 19171.0) (df = 5; 19176.0) (df = 10; 19171.0)
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Table 4.1: Logistic regression table of partisanship and the likelihood of hearing favorable and unfavorable
news about business conditions tab:ch4_logit
As it follows, my next question asks whether hearing different news about the economy amplified the
partisan differences in inflation expectations. I followed a similar difference-in-difference design, except
77

that the dependent variable is prospective inflation expectations (for the next 12 months). Weighted least
squares regression results first confirmed the earlier descriptive evidence that Republicans were significantly more likely to overestimate inflation rates compared to Democrats and Independents post-2020
election (Model 1 and 2 in Table 4.2). On the contrary, during the Trump administration, they underestimated inflation rates relative to Democrats and Independents (p < 0.01). Unsurprisingly, Democrats
underestimated inflation rates relative to Republicans and Independents during the Biden administration
(p < 0.01).
I further demonstrated the interaction effects of the tones of news that people were exposed to (favorable or unfavorable) on their inflation expectations (see model 3 and 4 in 4.2). Intuitively, Republicans who
heard unfavorable news about the economy were significantly more likely to expect a higher inflation rate
than Democrats during the Biden administration. In the same post-treatment period, however, Republicans
who heard favorable news about the economy tended to predict a lower inflation rate with statistical significance. These findings confirmed the amplification effects of news consumption on partisan differences
in inflation assessments.
4.10 Cable news narratives
We have established that partisanship plays a role in whether people hear favorable or unfavorable economic news. Moreover, the type of news partisans hear affects their inflation expectations. Next, I look at
what information could have been represented along party lines through cable television news.
To compare partisan news narratives about inflation, I extracted narratives in the form of an agentverb-patient sequence for each of the three cable TV networks and filtered for narratives whose verbs signal
causal attributions such as "drive", "cause", "bring", "blame" and "fuel". The identified narratives, sorted by
their frequencies, are presented in Table 4.3. These narratives capture causal attributions presented or
discussed by anchors and guests in each network’s programs.
78

Dependent variable: Inflation expectations
(1) (2) (3) (4)
Intercept 3.189∗∗∗ 3.087∗∗∗ 3.165∗∗∗ 3.298∗∗∗
(0.076) (0.099) (0.193) (0.199)
Post 2.125∗∗∗ 3.462∗∗∗ 3.534∗∗∗ 3.950∗∗∗
(0.109) (0.149) (0.149) (0.185)
DEMOCRAT 0.225 0.042 0.201
(0.153) (0.308) (0.306)
DEMOCRAT × Post -3.059∗∗∗ -2.631∗∗∗ -2.586∗∗∗
(0.214) (0.380) (0.377)
REPUBLICAN -0.681∗∗∗ -0.579∗∗∗ -0.514∗
-0.394
(0.134) (0.148) (0.292) (0.290)
REPUBLICAN × Post 3.354∗∗∗ 2.016∗∗∗ 1.473∗∗∗ 1.253∗∗∗
(0.212) (0.235) (0.415) (0.415)
Favorable News -1.805∗∗∗ -1.474∗∗∗
(0.250) (0.250)
Unfavorable News 0.393∗ 0.637∗∗∗
(0.222) (0.221)
Favorable News × DEMOCRAT 1.963∗∗∗ 1.794∗∗∗
(0.508) (0.507)
Favorable News × REPUBLICAN 1.058∗∗∗ 0.809∗∗
(0.377) (0.377)
Unfavorable News × DEMOCRAT -0.270 -0.328
(0.355) (0.354)
Unfavorable News × REPUBLICAN -0.271 -0.424
(0.354) (0.352)
Post × Favorable News × DEMOCRAT -1.334∗∗ -1.154∗∗
(0.531) (0.530)
Post × Favorable News × REPUBLICAN -1.644∗∗∗ -1.405∗∗∗
(0.525) (0.524)
Post × Unfavorable News × DEMOCRAT -0.119 -0.018
(0.405) (0.404)
Post × Unfavorable News × REPUBLICAN 0.864∗ 1.059∗∗
(0.460) (0.460)
COLLEGE -0.659∗∗∗
(0.130)
Post × COLLEGE -0.869∗∗∗
(0.191)
Observations 19,375 19,375 19,375 19,375
R2 0.059 0.080 0.093 0.102
Adjusted R2 0.059 0.080 0.092 0.101
Residual Std. Error 6.746(df = 19371) 6.669(df = 19369) 6.625(df = 19359) 6.593(df = 19357)
F Statistic 433.373∗∗∗ (df = 3.0; 19371.0) 315.106∗∗∗ (df = 5.0; 19369.0) 119.007∗∗∗ (df = 15.0; 19359.0) 109.872∗∗∗ (df = 17.0; 19357.0)
Note: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Table 4.2: Weighted least squares regression table of inflation expectations tab:ch4_wls
79

In the Fox News content, the attributions embody a clear theme: "President Biden (and the Democrats)
caused inflation", as well as "Biden’s spending proposals fuel inflation". This narrative closely follows
the growth of inflation rates based on a time series model I fitted, with the one-month lagged inflation
rate being statistically significant (p < 0.01)
11. They took issue with Biden’s Infrastructure Bill, Build
Back Better agenda (including American Rescue Plan, American Jobs Plan and American Families Plan)
and climate and energy bill which was estimated to cost trillions of dollars. The Biden administration’s
inclination to "spend trillions of dollars" is a repetitive script aired on Fox News. It is unsurprising given
the salience of "Biden" in their inflation news coverage.
These attributions convey clear partisan and ideological cues. Fox News’ consistently negative portrayal of Biden delivered partisan cues, which could effectively prime its audiences’ existing negative feelings about Biden and the Democrats. These narratives not only attributed inflation to Biden and his policies, but they also emphasized that Biden lacked accountability for the issue of inflation (as is evident in
the narrative "he blames everyone" from Putin to oil companies). The network’s frequent reference to
"trillion", "deficit" and "spending" rendered a "scripted" attribution of inflation to its audience. This recurring script served to invoke and reinforce a common right-leaning economic belief that government
spending exacerbates inflation. Although government spending could have played a role in exacerbating
short-term inflation 12, Fox News’ presentation of this possible cause skipped a few intermediate steps
compared to others, such as "stimulus package fueled demand" which was featured in CNN’s narratives
and was relatively quiet about the fact that Trump administration passed trillions of stimulus packages as
well.
11I fitted an ARDL Model that took the narrative of "Biden (or his spending policy) causes inflation" as the dependent variable
and regressed it on its lags and lagged inflation rate.
12see for example survey of experts in Andre et al. (2023) and https://fivethirtyeight.com/features/what-democrats-andrepublicans-get-wrong-about-inflation/
80

Compared to Fox News which repetitively exploited a unified theme and target, CNN’s narratives
appeared more balanced. They brought in various attributions commonly addressed in inflation discourse
and discussed more textbook factors such as interest rates and supply chain issues.
Partisan cues are not obvious among the narratives I identified in MSNBC News, although it paid
some attention to how Republicans and right-wing media attributed inflation to Biden and the Democrats,
and "Republicans" are the top partisan entities they discussed (4.6c). They also touched upon a wider
range of supply and demand-side factors. But on the other hand, MSNBC News served some left-leaning
ideological anchors in inflation narratives by directing their attacks at corporate America, big oil companies
and monopolies. MSNBC News was more adamant in their claims that corporate greed and price gouging
caused inflation, which is not commonly shared by experts 13. On the same topic, CNN presented more
balanced perspectives (both for and against this interpretation, 4.6).
The "government spending" and "corporate greed" attributions are the two ideologically aligned interpretive frameworks I identified in partisan cable news. They also appeared to be the top two factors that
people generally attribute inflation to based on a June 2023 YouGov survey 14. They overshadowed less
politicized factors such as supply chain issues and demand pressure (4.4).
Furthermore, it is worthwhile to show that the leading right-leaning schema "federal government
spending" and leading left-leaning schema "corporate greed" were presented differently in the three cable
news networks (4.5, 4.6). On federal government spending, Fox News took a firm stance against Biden’s
spending programs while highlighting the magnitude of these proposals (through salient word choice of
"trillion" in its accounts of inflation drivers). CNN cited reasons why Biden’s spending bills would not
raise inflation, but various perspectives were represented in its news programs, including those who argued Biden’s well-intended spending bills could exacerbate inflation. MSNBC News generally refuted the
13See Andre et al. (2023), https://news.northeastern.edu/2022/08/02/what-is-greedflation-and-is-it-driving-higher-prices/ and
https://www.kentclarkcenter.org/surveys/inflation-market-power-and-price-controls/
14https://today.yougov.com/topics/economy/articles-reports/2023/07/07/more-americans-now-blame-inflation-corporations
81

Fox News
the government drive price
president biden cause inflation
policy cause inflation
the deficit/trillion (spending) cause inflation
this bill fuel inflation
he blame putin
we blame president biden
they blame everyone
they blame the oil company
biden’s inflation cause more and more economic anxiety
gas price drive inflation
the democrats cause inflation
CNN
high interest rate bring down inflation
the fed cause a recession (if they do too much)
inflation drive price (for food and other consumer goods)
record high energy price drive inflation
the war/russia’s war/putin’s invasion drive (gas) price/inflation
president biden blame the war/putin
they/some/americans blame president biden
republican blame the democrats
supply chain issues/supply bottleneck cause inflation
pandemic cause some (inflation)
the stimulus package fuel demand
wage cause inflation
MSNBC
people blame president biden
gas price drive inflation
the war/putin’s invasion drive price/inflation
the supply factor/global supply chain issue/supply constraint drive inflation
corporate profits/monopoly/price gouging drive inflation
high energy price drive inflation
many/other factors drive inflation
inflation drive inflation and immigration vote
demand drive inflation
republicans blame the democrats
this administration’s policy drive our remarkable progress
headwind (war/pandemic) cause supply chain issue
Table 4.3: Top narratives of inflation drivers discussed by cable news networks. Verbs and entities presented in root forms. tab:ch4_attr
82

Republicans Democrats
federal government spending 77 31
large corporations maximizing profits 44 76
Democrats 74 14
Table 4.4: Top causes of inflation for Republicans and Democrats. Source: YouGov tab:ch4_yougov
(a) transcripts (b) mentions of "inflation"
Figure 4.5: Counts of inflation-related transcripts and "inflation" mentions by cable news networks. Grey
dashed is a time series of year-over-year inflation rate. fig:ch4_count
attribution to government spending, their language conveyed stronger emotional valence against the conservatives (e.g. "right-wing hysteria") compared to that of CNN. But MSNBC also aired mildly opposing
arguments (such as the last sample statement, which appeared in a program hosted by a business news
anchor from its politically neutral sibling of NBC).
Partisan cues are present in these politicized interpretive frames compared to the more neutral and
textbook attributions such as supply chain issues and pressure of pandemic pent-up demands.
Finally, the frequencies of entities being presented in inflation news display interesting patterns of
partisan cues (4.6). Fox News involved Biden and the Democrats in inflation narratives at an overwhelming
frequency. Considering mentions of "inflation" in Fox News only come second to those in CNN during
the study period (4.5), the outstanding volume of "Biden" in Fox News shows revealing signs of repetition
priming. Echoing Fox News’ attention to its ideological/partisan opposites, MSNBC featured "Republicans"
relatively frequently with an interest in how Republicans were going to play up the issue of inflation as
midterm elections approached. As the sample statements illustrated, partisan cues in the form of negatively
portrayed opposing parties and leaders are more evident than mentions of the politically aligned party.
83

Network Sample statement: government spending
Fox News We’re seeing record high inflation of 8 percent and the Biden budget numbers laughably assume that the trillions they’ve spent already hasn’t caused inflation.
Fox News But when it comes to rising gas prices, he is not addressing at all the fact that his
egregious spending has driven inflation, that we have slow down energy production
here in the United States, that the red tape and the bureaucracy for permitting has
made it so difficult to be energy independent in the United States.
Fox News Excess federal spending means higher inflation.
Fox News The nonpartisan Tax Foundation also determined, quote, by reducing long-run economic growth, the bill worsens inflation by constraining the productive capacity of
our economy.
CNN But – and this is really crucial because right now President Biden and the Democrats
are saying that spending another at least $1.9 trillion will help alleviate inflation.
CNN But when you look at most economists, their read is, look, because spending is so
spread out over a matter of years, it’s not going to have a significant impact on inflation.
CNN But inflation can quickly get out of control when governments print too much money
to pay for spending. When not enough real value underlies that paper, prices surge,
that‘s called hyperinflation.... it’s the Fed‘s job to make sure the United States keeps
the inflation rate on track.
CNN And Republicans are arguing that spending more money, especially after all the pandemic stimulus payments, is only going to make inflation worse.
MSNBC No one raised inflation when – in 2017, when Republicans gave $2 trillion away to
billionaire corporations in the top 1 percent of this country, and everybody – every
working family was left behind.
MSNBC For all the right-wing hysteria over Joe Biden‘s handling of the economy, you would
never know that it‘s actually doing pretty great....Inflation Republicans and the
wealthy, and especially the super rich – and, by the way, they did gangbusters during
the pandemic – well, they like to blame inflation on increased government spending.
MSNBC I mean, the Build Back Better is a – is not that – it looks like a lot of money – a trillion
here. You always sound like Dr. Evil, $1 trillion, but the U.S. economy is enormous,
and the amounts of money that we‘re talking about largely paid for with new taxes
are just not enough to be a source of inflation.
MSNBC I have to say the irony of, because this is well-intentioned spending you mentioned,
and I think it‘s to fix the problem, frankly, that does not exist anymore. Given what
I just said, if I was Darth Vader and I wanted to destroy the U.S. economy, I would
actually do aggressive spending in the middle of an already hot economy, which is
exactly what we have.
Table 4.5: Sample statements about federal government spending. Source: LexisNexis news transcripts tab:ch4_spend
84

Network Sample statement: corporate greed
Fox News He’s blaming it on corporate greed was his latest scenario.
Fox News He said, "You know, first it was transitory inflation; now it’s corporate greed; now
we’re blaming Putin.
Fox News If oil companies are making record profits and energy markets destabilized by Putin’s
invasion of Ukraine, President Biden thinks they need to lower prices at the pump.
CNN I mean, corporate greed is real but it is not new. So, it is not easy to argue that
corporations have gotten greedy or than they were a year ago. What is possibly true
is that inflation may be giving some corporations kind of cover to be more rapacious
than usual. And I wouldn’t dismiss that as a factor, but it’s certainly not the main
cause of the inflation we’re seeing.
CNN The answer has to do with supply and demand. And I just think it is at best a distraction and a relevant sort of demagogic – demagoguery claim to talk about corporate
greed.
CNN It has to do with the supply and demand factors that I was just talking about. Companies tend to do quite well when demand is high. That’s why you’re seeing these
things coincide, record profits for corporations and higher prices
CNN Inflation is a global problem caused, A, by the breaking of supply chains because of
the pandemic, by the war in Ukraine, and, as I said, a significant part of inflation has
to do with corporate greed.
MSNBC And that‘s not all, there‘s also some corporate price gouging that‘s going on where
companies are using all of this uncertainty in inflation, to basically exploit the situation and make you pay more than you otherwise would, raising prices.
MSNBC But I also have a track record of standing up to corporate power, standing up to
big pharma, who‘s cheating us with prescription costs, making Medicare costs more,
standing up to monopolies that are driving up inflation with their price gouging,
standing up to big oil and their greed at the pump.
MSNBC We fight for these underdogs who are getting who are literally the – they are being
preyed upon in this marketplace because corporate actors have too much power over
both laborers and consumers who are jean-jacked on prices.
Table 4.6: Sample statements about corporate greed. Source: LexisNexis news transcripts tab:ch4_greed
85

(a) Fox News fig:fox (b) CNN fig:cnn
(c) MSNBC fig:msnbc
Figure 4.6: Top entities three cable news networks’ inflation coverage fig:ch4_entities
The presence of partisan cues and ideological frames in the more slanted media (Fox News and MSNBC
compared to CNN 15), and Fox News’ recurring frames of Biden and his policy in particular provide support
for H2.
4.11 Comparison with inflation narratives from surveys
We identified and presented narratives in cable news on the attributions of inflation. The presence of
partisan cues and ideologically aligned frames, evident in Fox News and to some extent in MSNBC, is
a proxy view of how Democrats and Republicans might have heard different news about inflation and
national economy. Different narratives on inflation drivers, whether reaching people via cable news, peers,
15Based on Allsides’ Media Bias Chart, Fox News received a rating of "Right", MSNBC "Left", while CNN received "Lean-left":
https://www.allsides.com/media-bias/media-bias-chart
86

or social media, hold potential explanations for the partisan differences in inflation expectations that I
presented at the beginning of this chapter.
Although the data I collected does not directly link the variations in media narratives to people’s inflation expectations, due to a lack of information on what specific news sources they consume, additional
survey results on inflation narratives reveal interesting parallels with what I found in cable news content.
Andre et al. (2023) conducted a survey of economists and households on the causes of inflation. They
found that experts placed greater focus on textbook supply and demand-side factors and specified more
complicated causal chains. In comparison, households’ accounts of inflation drivers were simpler and more
politicized, mentioning factors such as government mismanagement or price gouging, which were absent
in experts’ explanations (Andre et al., 2023). We identified these two prominent narratives in Fox News and
MSNBC respectively (4.3). Furthermore, experts went into details on both supply-side and demand-side
factors, whereas households’ narratives mostly omitted demand-side factors, such as government spending, loose monetary policy, pent-up demand, and demand shift (Andre et al., 2023). The lack of discussion
of demand-side factors is also aligned with the patterns in cable news narratives we identified. CNN and
MSNBC both covered supply chain disruptions more than demand-side factors as causes of inflation. An
interesting aspect of this pattern in Andre et al. (2023)’s survey is that fewer households (17%) listed government spending as a driver of inflation than did experts (50%). According to my media content analysis,
government spending is only a salient topic in Fox News and often discussed in a partisan tone. The Fox
News presentation of government spending may correspond better with the "government mismanagement" attribution among households in Andre and colleges’ survey. They similarly found correlational
evidence between beliefs in government mismanagement and overspending and higher inflation expectations (Andre et al., 2023). At the same time, it is plausible that households who consumed left-leaning
media were less likely to be exposed to the narrative that government spending fueled inflation and more
87

likely to hear that "Republicans blame Biden and his spending policies for inflation", therefore discounting
the credibility of this explanation.
Additional evidence from the Surveys of Consumers supports a partisan difference in exposure to
the narrative that the government "mismanaged" the economy. Table 4.7 ranked the top ten types of
news survey participants heard about business conditions by partisanship. The most common news that
Republicans and Democrats heard were similar such as price inflation, drop and increase of employment,
opening and closing of factories. The outstanding difference is that many more Republicans heard the
news that "Government not improving business conditions" than Democrats. It is among the top ten
types of business news Republicans and Independents heard, but completely absent from Democrats’ top
ten list of news they heard about business conditions, indicating the partisan reach of this narrative. Its
potential influence on Independents is consistent with general negativity bias in economic news coverage
and reception (Damstra, 2019), as well as findings from other studies showing that an opposition party’s
criticism of economic issues increases negative public perceptions of the economy when the opposition
party is believed to own economic issues (Pardos-Prado & Sagarzazu, 2016)
16
.
4.12 Discussion
This chapter validated the role of information in partisan differences of inflation expectations and identified partisan differences in the types of news people heard about the economy. To find potential sources
of this information asymmetry, I analyzed three years of partisan media narratives on inflation. Coverage
from Fox News and MSNBC both displayed ideologically aligned frames that explained causes of inflation.
Fox News also displayed denser partisan cues with repetitive mentions of Biden and his spending proposals. CNN’s inflation news coverage appeared to be relatively balanced in presenting different narratives
16People on average tend to believe Republicans have a better plan for the economy, but the gap is covertly shrinking as of
late: https://today.yougov.com/topics/economy/articles-reports/2023/07/07/more-americans-now-blame-inflation-corporations
88

Republicans Democrats Independents
1 Has heard of no changes Has heard of no changes Has heard of no changes
2 (UNFAV) Closing of plants (UNFAV) Closing of plants (UNFAV) Closing of plants
3 (UNFAV) Prices high, inflation (UNFAV) Drop in employ, less
overtime
(UNFAV) Prices high, inflation
4 (UNFAV) Drop in employ, less
overtime
(UNFAV) Prices high, inflation (UNFAV) Drop in employ, less
overtime
5 (UNFAV) Gov’t not improving
business conditions
(FAV) Employ is high, plenty of
jobs
(UNFAV) Decline in specific industries
6 (UNFAV) Production decreasing,
GNP down
(FAV) Opening of plants, factories,
stores
(FAV) Opening of plants, factories,
stores
7 (FAV) Employ is high, plenty of
jobs
(UNFAV) Decline in specific industries
(UNFAV) Gov’t not improving
business conditions
8 (FAV) Opening of plants, factories,
stores
(UNFAV) Production decreasing,
GNP down
(UNFAV) Production decreasing,
GNP down
9 (UNFAV) Decline in specific industries
(FAV) Improvements in specific industries
(FAV) Employ is high, plenty of
jobs
10 (FAV) Improvements in specific industries
(UNFAV) Stock market decline (FAV) Improvements in specific industries
Table 4.7: Top news partisans heard about business conditions tab:ch4_topnews
amongst the three. The cable news narratives in left and right-leaning networks we detected share important similarities with household narratives of inflation based on surveys we found in other studies (Andre
et al., 2023; Andre et al., 2021).
My study complements literature on partisan bias in inflation expectations (Bachmann et al., 2021;
Bartels, 2002; Mian et al., 2021) by showing the moderating effects of news reception. It bridges this body
of literature with recent studies on how inflation narratives among households were more politicized than
those of the experts (Andre et al., 2023; Andre et al., 2021) through analyzing the content of possible news
sources for households. Cable news networks, to varying degrees, released cues and frames that would
match their target audiences’ existing interpretive frameworks. The divergence of media narratives we
identified could have potentially served to activate different partisan identities and interpretive frames
about inflation. The most influential media source on the right, Fox News, repetitively emphasized the
government’s role in causing inflation, which was found to be correlated with higher inflation expectations
(Andre et al., 2023). Compared to content analysis that rely on the "salience" of certain lexicons, I examined
narratives that specified a causal attribution for inflation.
89

Furthermore, the media narratives on inflation drivers in the three networks in my analysis also conform to the incumbency-opposition distinction noted by a number of studies of subjective economic assessments (Bachmann et al., 2021; Pardos-Prado & Sagarzazu, 2016). For example, voters’ misperceptions
of inflation are smaller when the party they support occupies the White House (Bachmann et al., 2021). My
comparison of partisan news content mostly during the Biden administration shows that CNN and MSNBC
both introduced more substance and various supply-side factors that contributed to inflation, whereas Fox
News’ attention was fixated on Biden and his spending plans. Admittedly, it could be the case that CNN
and MSNBC (especially CNN) on average provide more information than Fox News. Future research could
compare partisan media narratives on economic conditions when the Republican Party holds the White
House.
This study has several limitations. First, it only captured the media aspect of social communication and
cable news content. Although I selected cable networks for carefully considered reasons, the heterogeneity of news sources in today’s information environment offers a variety of novel, traditional and fringe
channels for partisan news consumption, from social media, podcasts, to press and social interaction. On
a relevant note, I pointed out parallels between media narratives that I summarized and households’ narratives that others identified, without establishing a direct relationship between these two types of narratives. Consumer sentiment surveys that incorporate more thorough questions on people’s economic news
sources would help us select, compare a more representative sample of news content. Such information
would also allow us to better test the correlations between narrative patterns derived from media content
analysis with public perceptions and therefore improve our understanding of the political conditioning of
subjective economic evaluations (Evans & Andersen, 2006; Pardos-Prado & Sagarzazu, 2016; van der Eijk
et al., 2007).
90

Finally, the narrative extraction techniques I implemented are effective in detecting the semantic roles
within a single statement but fall short of finding a narrative embedded in multiple interconnected statements (Lange et al., 2022). This resulted in under-detected narratives and increased efforts of manual
correction. Optimized packages that incorporate the feature of inferring causality between statements
might provide a better solution 17
.
The findings from this study suggest a revised consideration of the role information plays in economic
preference formation. Partisan differences in subjective economic assessments are related to questions
such as why people vote against their presumed economic interests, and what explains the decline of classbased voting (Evans & Tilley, 2012). The rising phenomenon of partisan media consumption and economic
reasoning complicates how the role of information should be conceived in a rational voter model, which
takes into account the information effect but often associates higher media exposure or high information
voters with more accurate economic assessments (Aidt, 2000; Goidel & Langley, 1995; Holbrook & Garand,
1996; Nordin, 2014). This study highlights the widening gap between "objective" economic realities and
varied subjective perceptions. Voters could detach their experience of personal financial situations from
their evaluations of the national economy, which are subjected to the influence of conflicting narratives in
the contemporary information environment. Same economic evidence could lead to different interpretive
frames written with existing ideological formulas. The nuanced implications of information acquisition
pose a challenge to a uniform approach to understanding how media coverage translates into aggregated
economic evaluations. More than producing high and low information voters, media consumption also
feeds into the formation of voters’ heterogeneous endogenous preferences (Becker, 2009; Bowles & Gintis,
2006; Gintis, 1974). Insights into the dynamic, heterogeneous influence from media and elites on individual
economic perceptions constitutes an important avenue to move economic voting research forward.
17For example, Lange et al. (2022) adapted the original RELATIO library for their work: https://github.com/K-RLange/
relatio/tree/master
91

Chapter 5
Conclusions
ch:conclusions
This dissertation presented three independent studies using different forms of text data to answer questions
about the evolution of political values and attitudes. They glimpse into different aspects of the contemporary political polarization in the United States and processes that lead towards a divergence of opinion.
The first study considers the evolving meanings of two core democratic values. An expanded vision of
equality since the 1960s, following a series of political and social changes, incorporates greater concerns
for a larger diversity of social groups. Its rise on the democratic value hierarchy poses a challenge to the
established liberal consensus that prioritizes individual freedom. The growing tension between liberty and
equality underlies many contested cultural issues today, especially issues that trigger conflicting debates
over free speech, from cancel culture, identity politics, to platform governance.
If the first study has implications for political divides over cultural issues, the third study, the fourth
chapter on inflation narratives, demonstrated how economic issues are also capable of breeding a clash
of opinions. Curiously, this divide does not necessarily involve the varied economic interests of different
social groups and disagreements over redistribution. It tells the story of people forming economic judgements conditional on partisan identity and a cautionary tale of how individuals need to navigate through
narratives in search for information.
92

The second study is located at a present-day information environment that is often criticized for its
influence on public opinion. The study of the moralization of novel issues on the 2020 election trail on
Twitter suggested links between negative partisanship and moral expression about a conspiracy theory.
It naturally leads to the future inquiry of whether negative preconceptions of the opposing party’s moral
characters predict a higher likelihood of belief in conspiracy theories. This hypothesis joins many in
depicting a political polarization based on the mutual misperceptions (Edsall, 2023).
This dissertation journeyed through research avenues of certain "soft infrastructure" for democratic
stability, such as ideas, values, identities, and media. The tension between liberty and equality approached
by the first study involves deep-seated, moralized principles, which are difficult to resolve. It is a developing
case to observe in times to come. When the liberal consensus is radically challenged, will a new consensus
be renegotiated and arrive somewhere in the middle?
The second and third studies represented cases of two worlds of truths when party loyalty trumps
democratic principles (Graham & Svolik, 2020). To moderate and attenuate the intensifying political polarization, it is worth considering what communication strategies would depress partisan identity and
instead prime context-specific common identities.
∗ ∗ ∗
93

Bibliography
Abramowitz, A. (2018). The great alignment: Race, party transformation, and the rise of donald trump. Yale
University Press.
Abramowitz, A., & McCoy, J. (2019). United states: Racial resentment, negative partisanship, and
polarization in trump’s america. The ANNALS of the American Academy of Political and Social
Science, 681(1), 137–156.
Abramowitz, A., & Saunders, K. L. (2008). Is polarization a myth? The Journal of Politics, 70(2), 542–555.
Abramowitz, A., & Webster, S. (2016). The rise of negative partisanship and the nationalization of us
elections in the 21st century. Electoral Studies, 41, 12–22.
Abramowitz, A., & Webster, S. W. (2018). Negative partisanship: Why americans dislike parties but
behave like rabid partisans. Political Psychology, 39, 119–135.
Aidt, T. S. (2000). Economic voting and information. Electoral Studies, 19(2-3), 349–362.
Altheide, D. L. (2004). Media logic and political communication. Political Communication, 21(3), 293–296.
Altheide, D. L. (2015). Media logic. The international encyclopedia of political communication, 1–6.
Andre, P., Haaland, I., Roth, C., & Wohlfart, J. (2023). Narratives about the macroeconomy.
Andre, P., Roth, C., Haaland, I., & Wohlfart, J. (2021). Inflation narratives.
Ash, E., Gauthier, G., & Widmer, P. (2023). Relatio: Text semantics capture political and economic
narratives. Political Analysis, 1–18.
Bachmann, O., Gründler, K., Potrafke, N., & Seiberlich, R. (2021). Partisan bias in inflation expectations.
Public Choice, 186, 513–536.
Baharaeen, M. (2023). Much of america’s political divide is an illusion. Persuasion. Retrieved June 20,
2023, from https://www.persuasion.community/p/much-of-americas-political-divide
Baker, C. F., Fillmore, C. J., & Lowe, J. B. (1998). The berkeley framenet project. COLING 1998 Volume 1:
The 17th International Conference on Computational Linguistics.
94

Bakker, B. N., Lelkes, Y., & Malka, A. (2020). Understanding partisan cue receptivity: Tests of predictions
from the bounded rationality and expressive utility perspectives. The Journal of Politics, 82(3),
1061–1077.
Baldassarri, D., & Gelman, A. (2008). Partisans without constraint: Political polarization and trends in
american public opinion. American Journal of Sociology, 114(2), 408–446.
Bandura, A. (1999). Moral disengagement in the perpetration of inhumanities. Personality and social
psychology review, 3(3), 193–209.
Bandura, A., Barbaranelli, C., Caprara, G. V., & Pastorelli, C. (1996). Mechanisms of moral disengagement
in the exercise of moral agency. Journal of personality and social psychology, 71(2), 364.
Bankert, A. (2021). Negative and positive partisanship in the 2016 us presidential elections. Political
Behavior, 43(4), 1467–1485.
Barber, M., & Pope, J. C. (2019). Does party trump ideology? disentangling party and ideology in america.
American Political Science Review, 113(1), 38–54.
Bartels, L. M. (1993). Messages received: The political impact of media exposure. American political
science review, 87(2), 267–285.
Bartels, L. M. (2002). Beyond the running tally: Partisan bias in political perceptions. Political behavior, 24,
117–150.
Bartels, L. M. (2017). The ‘wave’ of right-wing populist sentiment is a myth. Washington Post Monkey
Cage. Retrieved August 1, 2023, from https://www.washingtonpost.com/news/monkeycage/wp/2017/06/21/the-wave-of-right-wing-populist-sentiment-is-a-myth/
Becker, G. S. (2009). Accounting for tastes. Harvard University Press.
Bennett, W. L., & Entman, R. M. (2000). Mediated politics: Communication in the future of democracy.
Cambridge University Press.
Berelson, B. R., Lazarsfeld, P. F., & McPhee, W. N. (1986). Voting: A study of opinion formation in a
presidential campaign. University of Chicago Press.
Berggren, D. J., & Rae, N. C. (2006). Jimmy carter and george w. bush: Faith, foreign policy, and an
evangelical presidential style. Presidential Studies Quarterly, 36(4), 606–632.
Bonial, C., Bonn, J., Conger, K., Hwang, J. D., & Palmer, M. (n.d.). Propbank: Semantics of new predicate
types.
Botvinik-Nezer, R., Jones, M., & Wager, T. D. (2023). A belief systems analysis of fraud beliefs following
the 2020 us election. Nature Human Behaviour, 1–14.
Bowles, S., & Gintis, H. (2006). Social preferences, homo economicus, and zoon politikon. The Oxford
handbook of contextual political analysis, 172–186.
95

Brady, W. J., Crockett, M. J., & Van Bavel, J. J. (2020). The mad model of moral contagion: The role of
motivation, attention, and design in the spread of moralized content online. Perspectives on
Psychological Science, 15(4), 978–1010.
Brady, W. J., McLoughlin, K., Doan, T. N., & Crockett, M. J. (2021). How social learning amplifies moral
outrage expression in online social networks. Science Advances, 7(33), eabe5641.
Brady, W. J., McLoughlin, K. L., Torres, M. P., Luo, K. F., Gendron, M., & Crockett, M. (2023).
Overperception of moral outrage in online social networks inflates beliefs about intergroup
hostility. Nature human behaviour, 1–11.
Brady, W. J., Wills, J. A., Burkart, D., Jost, J. T., & Van Bavel, J. J. (2019). An ideological asymmetry in the
diffusion of moralized content on social media among political leaders. Journal of Experimental
Psychology: General, 148(10), 1802.
Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the diffusion of
moralized content in social networks. Proceedings of the National Academy of Sciences, 114(28),
7313–7318.
Braley, A., Lenz, G. S., Adjodah, D., Rahnama, H., & Pentland, A. (2023). Why voters who value
democracy participate in democratic backsliding. Nature Human Behaviour, 1–12.
Brandt, M. J., Wisneski, D. C., & Skitka, L. J. (2015). Moralization and the 2012 us presidential election
campaign. Journal of Social and Political Psychology, 3(2), 211–237.
Brenan, M. (2019). Electability of democratic nominee outranks issue stances. Retrieved June 10, 2023,
from https://news.gallup.com/poll/259454/electability-democratic-nominee-outranks-issuestances.aspx
Broockman, D., & Kalla, J. (2022). Consuming cross-cutting media causes learning and moderates
attitudes: A field experiment with fox news viewers.” https://doi.org/10.31219/osf.io/jrw26
Burns, T. (2022). Republicans, democrats push dueling economic narratives ahead of midterms. The Hill.
Retrieved August 1, 2023, from https://thehill.com/business/3460548-republicans-democratspush-dueling-economic-narratives-ahead-of-midterms/
Cacciatore, M. A., Scheufele, D. A., & Iyengar, S. (2016). The end of framing as we know it. . . and the
future of media effects. Mass communication and society, 19(1), 7–23.
Cameron, C. D., Lindquist, K. A., & Gray, K. (2015). A constructionist review of morality and emotions:
No evidence for specific links between moral content and discrete emotions. Personality and
social psychology review, 19(4), 371–394.
Campbell, A., Converse, P. E., Miller, W., & Stokes, D. E. (1960). The american voter. University of Chicago
Press.
Capaldi, N., & Lloyd, G. (2016). Liberty and equality in political economy: From locke versus rousseau to the
present. Edward Elgar Publishing.
96

Carman, J., Rosenthal, S., Maibach, E., Kotcher, J., Wang, X., Marlon, J., & Leiserowitz, A. (2020). Shifting
views of an ideal society in the u.s. Yale University and George Mason University. Retrieved April
20, 2023, from https://climatecommunication.yale.edu/publications/shifting-views-of-anideal-society-in-the-u-s/
Carmines, E. G., & Stimson, J. A. (2020). Issue evolution: Race and the transformation of american politics.
Princeton University Press.
Carpenter, J., Brady, W., Crockett, M., Weber, R., & Sinnott-Armstrong, W. (2020). Political polarization
and moral outrage on social media. Conn. L. Rev., 52, 1107.
Cassese, E. C. (2021). Partisan dehumanization in american politics. Political Behavior, 43, 29–50.
Chaffee, S. H., & Kanihan, S. F. (1997). Learning about politics from the mass media. Political
communication, 14(4), 421–430.
Chen, E., Deb, A., & Ferrara, E. (2021). # Election2020: The first public twitter dataset on the 2020 us
presidential election. Journal of Computational Social Science, 1–18.
Chong, D., Citrin, J., & Levy, M. (2022). The realignment of political tolerance in the united states.
Perspectives on Politics, 1–22.
Clifford, S. (2019). How emotional frames moralize and polarize political attitudes. Political Psychology,
40(1), 75–91.
Clinton, J. D., Lapinski, J. S., Lentz, S., & Pettigrew, S. (2020). Trumped by trump? public support for vote
by mail voting in response to the covid-19 pandemic. Public Support for Vote By Mail Voting in
Response to the COVID-19 Pandemic (June 16, 2020).
Cohen, P. (2017). On health and welfare, moral arguments can outweigh economics. The New York Times.
Retrieved June 15, 2023, from
https://www.nytimes.com/2017/05/07/business/economy/congress-benefits-fairness.html
Cole Wright, J., Cullum, J., & Schwab, N. (2008). The cognitive and affective dimensions of moral
conviction: Implications for attitudinal and behavioral measures of interpersonal tolerance.
Personality and Social Psychology Bulletin, 34(11), 1461–1476.
Conover, P. J. (1988). The role of social groups in political thinking. British Journal of Political Science,
18(1), 51–76.
Conover, P. J., & Feldman, S. (1984). How people organize the political world: A schematic model.
American Journal of Political Science, 95–126.
Converse, P. E. (1964). The nature of belief systems in mass publics. in ideology and discontent, ed. david
apter. new york: Free press.
Crockett, M. J. (2017). Moral outrage in the digital age. Nature human behaviour, 1(11), 769–771.
Curtin, R. (2018). Consumer economic expectations: Persistent partisan differences.
97

Dahl, R. A. (2005). Who governs?: Democracy and power in an american city. Yale University Press.
Dalton, R. J., Beck, P. A., & Huckfeldt, R. (1998). Partisan cues and the media: Information flows in the
1992 presidential election. American Political Science Review, 92(1), 111–126.
Damstra, A. (2019). Disentangling economic news effects: The impact of tone, uncertainty, and issue on
public opinion. International Journal of Communication, 13, 20.
Davies, M. (2012a). Examining recent changes in english: Some methodological issues. Handbook on the
history of English: Rethinking approaches to the history of English, 263–87.
Davies, M. (2012b). Expanding horizons in historical linguistics with the 400-million word corpus of
historical american english. Corpora, 7(2), 121–157.
Davis, A. (2019). Political communication: A new introduction for crisis times. John Wiley & Sons.
Dawson, M. C. (1995). Behind the mule: Race and class in african-american politics. Princeton University
Press.
De Tocqueville, A. (2006). Democracy in america (H. Reeve, Trans.). Project Gutenberg.
https://www.gutenberg.org/files/816/816-h/816-h.htm
Ditto, P. H., Pizarro, D. A., & Tannenbaum, D. (2009). Motivated moral reasoning. Psychology of learning
and motivation, 50, 307–338.
Druckman, J. N., Klar, S., Krupnikov, Y., Levendusky, M., & Ryan, J. B. (2021). Affective polarization, local
contexts and public opinion in america. Nature human behaviour, 5(1), 28–38.
Dupoirier, S., Demarque, C., Souville, M., Apostolidis, T., & Lampropoulos, D. (2021). The
socio-representational construction of universal basic income in the french context: Effects of
politico-ideological anchors. Papers on Social Representations.
Dworetz, S. M. (1989). The unvarnished doctrine: Locke, liberalism, and the american revolution. Duke
University Press.
Edsall, T. (2023). The politics of delusion have taken hold. The New York Times. Retrieved June 8, 2023,
from https://www.nytimes.com/2023/05/31/opinion/politics-partisanship-delusion.html
Ellis, R. J. (1992). Rival visions of equality in american political culture. The Review of Politics, 54(2),
253–280.
Evans, G., & Andersen, R. (2006). The political conditioning of economic perceptions. The Journal of
Politics, 68(1), 194–207.
Evans, G., & Tilley, J. (2012). How parties shape class politics: Explaining the decline of the class basis of
party support. British journal of political science, 42(1), 137–161.
Fan, D. P. (1988). Predictions of public opinion from the mass media: Computer content analysis and
mathematical modeling (Vol. 12). Greenwood Publishing Group.
98

Feinberg, M., Kovacheff, C., Teper, R., & Inbar, Y. (2019). Understanding the process of moralization: How
eating meat becomes a moral issue. Journal of personality and social psychology, 117(1), 50.
Fiorina, M. P. (2017). Unstable majorities: Polarization, party sorting, and political stalemate. Hoover press.
Fiorina, M. P., Abrams, S. J., & Pope, J. C. (2011). Culture war?: The myth of a polarized america (3nd ed.)
New York: Longman.
Firth, J. (1957). A synopsis of linguistic theory, 1930-1955. Studies in linguistic analysis, 10–32.
Fiske, S. T., & Kinder, D. R. (2017). Involvement, expertise, and schema use: Evidence from political
cognition. In Personality, cognition and social interaction (pp. 171–190). Taylor; Francis.
Fiske, S. T., & Linville, P. W. (1980). What does the schema concept buy us? Personality and social
psychology bulletin, 6(4), 543–557.
Forgas, J. P., & Smith, C. A. (2007). Affect and emotion. The Sage handbook of social psychology, 146–175.
Foucault, M. (1980). Language, counter-memory, practice: Selected essays and interviews. Cornell
University Press.
Gaines, B. J., Kuklinski, J. H., Quirk, P. J., Peyton, B., & Verkuilen, J. (2007). Same facts, different
interpretations: Partisan motivation and opinion on iraq. The Journal of Politics, 69(4), 957–974.
Gardner, M., Grus, J., Neumann, M., Tafjord, O., Dasigi, P., Liu, N., Peters, M., Schmitz, M., &
Zettlemoyer, L. (2018). Allennlp: A deep semantic natural language processing platform. arXiv
preprint arXiv:1803.07640.
Garrett, K. N. (2016). The moralization of politics: Causes, consequences, and measurement of moral
conviction (Doctoral dissertation). The University of North Carolina at Chapel Hill.
Garrett, K. N., & Bankert, A. (2020). The moral roots of partisan division: How moral conviction
heightens affective polarization. British Journal of Political Science, 50(2), 621–640.
Gaus, G. F. (1983). Public and private interests in liberal political economy, old and new.
Gerber, A. S., & Huber, G. A. (2010). Partisanship, political control, and economic assessments. American
Journal of Political Science, 54(1), 153–173.
Gintis, H. (1974). Welfare criteria with endogenous preferences: The economics of education.
International Economic Review, 415–430.
Goffman, E. (1974). Frame analysis: An essay on the organization of experience. Harvard University Press.
Goidel, R. K., & Langley, R. E. (1995). Media coverage of the economy and aggregate economic
evaluations: Uncovering evidence of indirect media effects. Political Research Quarterly, 48(2),
313–328.
99

Graça, J., Calheiros, M. M., & Oliveira, A. (2016). Situating moral disengagement: Motivated reasoning in
meat consumption and substitution. Personality and Individual Differences, 90, 353–364.
Graham, J., Haidt, J., Koleva, S., Motyl, M., Iyer, R., Wojcik, S. P., & Ditto, P. H. (2013). Moral foundations
theory: The pragmatic validity of moral pluralism. In Advances in experimental social psychology
(pp. 55–130). Elsevier.
Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral
foundations. Journal of personality and social psychology, 96(5), 1029.
Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (2011). Mapping the moral domain.
Journal of personality and social psychology, 101(2), 366.
Graham, M. H., & Svolik, M. W. (2020). Democracy in america? partisanship, polarization, and the
robustness of support for democracy in the united states. American Political Science Review,
114(2), 392–409.
Green, D. P., Palmquist, B., & Schickler, E. (2004). Partisan hearts and minds: Political parties and the social
identities of voters. Yale University Press.
Green, J., Hobbs, W., McCabe, S., & Lazer, D. (2022). Online engagement with 2020 election
misinformation and turnout in the 2021 georgia runoff election. Proceedings of the National
Academy of Sciences, 119(34), e2115900119.
Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fmri investigation
of emotional engagement in moral judgment. Science, 293(5537), 2105–2108.
Greene, S. (2004). Social identity theory and party identification. Social science quarterly, 85(1), 136–153.
Groenendyk, E. (2013). Competing motives in the partisan mind: How loyalty and responsiveness shape
party identification and democracy. OUP USA.
Grossmann, M., & Hopkins, D. A. (2015). Ideological republicans and group interest democrats: The
asymmetry of american party politics. Perspectives on Politics, 13(1), 119–139.
Guess, A. M., Barberá, P., Munzert, S., & Yang, J. (2021). The consequences of online partisan media.
Proceedings of the National Academy of Sciences, 118(14), e2013464118.
Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral judgment.
Psychological review, 108(4), 814.
Haidt, J. (2012). The righteous mind: Why good people are divided by politics and religion. Vintage.
Haidt, J., & Graham, J. (2007). When morality opposes justice: Conservatives have moral intuitions that
liberals may not recognize. Social justice research, 20(1), 98–116.
Haidt, J., Koller, S. H., & Dias, M. G. (1993). Affect, culture, and morality, or is it wrong to eat your dog?
Journal of personality and social psychology, 65(4), 613.
100

Hamilton, W. L., Leskovec, J., & Jurafsky, D. (2016a). Cultural shift or linguistic drift? comparing two
computational measures of semantic change. Proceedings of the conference on empirical methods
in natural language processing. Conference on empirical methods in natural language processing,
2016, 2116.
Hamilton, W. L., Leskovec, J., & Jurafsky, D. (2016b). Diachronic word embeddings reveal statistical laws
of semantic change. arXiv preprint arXiv:1605.09096.
Hansen, J. M. (2020). Republicans claim voter fraud. how would that work? The New York Times.
Retrieved June 16, 2023, from
https://www.nytimes.com/2020/11/04/opinion/trump-voter-fraud-pennsylvania.html
Hartz, L. (1955). The liberal tradition in america: An interpretation of american political thought since the
revolution (Vol. 46). Houghton Mifflin Harcourt.
Herrnson, P., & Stewart III, C. (2022). Covid-19 and voter turnout and methods in the 2020 us general
election. Available at SSRN 4314257.
Herrnson, P. S., & Stewart III, C. (2023). The impact of covid-19, election policies, and partisanship on
voter participation in the 2020 us election. Election Law Journal: Rules, Politics, and Policy.
Hochschild, A. R. (2018). Strangers in their own land: Anger and mourning on the american right. The New
Press.
Holbrook, T., & Garand, J. C. (1996). Homo economus? economic information and economic voting.
Political Research Quarterly, 49(2), 351–375.
Hoover, J., Portillo-Wightman, G., Yeh, L., Havaldar, S., Davani, A. M., Lin, Y., Kennedy, B., Atari, M.,
Kamel, Z., Mendlen, M., et al. (2020). Moral foundations twitter corpus: A collection of 35k tweets
annotated for moral sentiment. Social Psychological and Personality Science, 11(8), 1057–1071.
Horberg, E. J., Oveis, C., & Keltner, D. (2011). Emotions as moral amplifiers: An appraisal tendency
approach to the influences of distinct emotions upon moral judgment. Emotion Review, 3(3),
237–244.
Hume, D. (1998). An enquiry concerning the principles of morals: A critical edition (Vol. 4). Oxford
University Press.
Hunzaker, M. F., & Valentino, L. (2019). Mapping cultural schemas: From theory to method. American
Sociological Review, 84(5), 950–981.
Iyengar, S., & Kinder, D. R. (1987). News that matters: Television and american opinion. University of
Chicago Press.
Iyengar, S., Lelkes, Y., Levendusky, M., Malhotra, N., & Westwood, S. J. (2019). The origins and
consequences of affective polarization in the united states. Annual review of political science, 22,
129–146.
101

Iyengar, S., Sood, G., & Lelkes, Y. (2012). Affect, not ideology: A social identity perspective on
polarization. Public opinion quarterly, 76(3), 405–431.
Iyengar, S., & Westwood, S. J. (2015). Fear and loathing across party lines: New evidence on group
polarization. American journal of political science, 59(3), 690–707.
Iyer, R., Koleva, S., Graham, J., Ditto, P., & Haidt, J. (2012). Understanding libertarian morality: The
psychological dispositions of self-identified libertarians.
Jennings, S. (2019). Op-ed: Dear liberals: Your outrage helps trump far more than it hurts him. Los Angeles
Times. Retrieved June 10, 2023, from
https://www.latimes.com/opinion/op-ed/la-oe-jennings-trump-haters-20190714-story.html
Jiang, J., Ren, X., & Ferrara, E. (2023). Retweet-bert: Political leaning detection using language features
and information diffusion on social networks. Proceedings of the International AAAI Conference on
Web and Social Media, 17, 459–469.
Joseph, K., Shugars, S., Gallagher, R., Green, J., Mathé, A. Q., An, Z., & Lazer, D. (2021). (mis) alignment
between stance expressed in social media data and public opinion surveys. arXiv preprint
arXiv:2109.01762.
Jurafsky, D., & Martin, J. H. (2021). Speech and language processing (draft). URL: https://web. stanford.
edu/˜ jurafsky/slp3.
JURKOWITZ, M., MITCHELL, A., SHEARER, E., & WALKER, M. (2020). U.s. media polarization and the
2020 election: A nation divided. Pew Research Center.
https://www.pewresearch.org/journalism/2020/01/24/u-s-media-polarization-and-the-2020-
election-a-nation-divided/
Kahneman, D. (2011). Thinking, fast and slow. macmillan.
Kam, C. D. (2005). Who toes the party line? cues, values, and individual differences. Political behavior, 27,
163–182.
Kam, C. D. (2007). Implicit attitudes, explicit choices: When subliminal priming predicts candidate
preference. Political Behavior, 29, 343–367.
Kann, M. E. (1980). Challenging lockean liberalism in america. the case of debs and hillquit. Political
Theory, 8(2), 203–222.
Kemp, D., & Ekins, E. (2021). Poll: 75% don’t trust social media to make fair content moderation decisions,
60% want more control over posts they see.
Kennedy, C. (2013). The manichean temptation: Moralising rhetoric and the invocation of evil in us
foreign policy. International politics, 50, 623–638.
Kim, E., Lelkes, Y., & McCrain, J. (2022). Measuring dynamic media bias. Proceedings of the National
Academy of Sciences, 119(32), e2202197119.
102

Kinder, D. R., & Kalmoe, N. P. (2017). Neither liberal nor conservative: Ideological innocence in the american
public. University of Chicago Press.
Klosko, G. (2017). The transformation of american liberalism. Oxford University Press.
Kohlberg, L., et al. (1969). Stage and sequence: The cognitive-developmental approach to socialization.
Handbook of socialization theory and research, 347, 480.
Koplenig, A. (2017). The impact of lacking metadata for the measurement of cultural and linguistic
change using the google ngram data sets—reconstructing the composition of the german corpus
in times of wwii. Digital Scholarship in the Humanities, 32(1), 169–188.
Krugman, P. (2022). Inflation and the power of narrative. The New York Times. Retrieved June 16, 2023,
from https://www.nytimes.com/2022/01/25/opinion/inflation-economy.html
Kulkarni, V., Al-Rfou, R., Perozzi, B., & Skiena, S. (2015). Statistically significant detection of linguistic
change. Proceedings of the 24th international conference on world wide web, 625–635.
Kunsch, H. R. (1989). The jackknife and the bootstrap for general stationary observations. The annals of
Statistics, 1217–1241.
Laden, A. S. (2009). Contemporary debates in political philosophy. In T. Christiano & J. Christman (Eds.).
Blackwell Publishing Ltd.
Lai, S., Liu, K., He, S., & Zhao, J. (2016). How to generate a good word embedding. IEEE Intelligent
Systems, 31(6), 5–14.
Landry, A. P., Schooler, J. W., Willer, R., & Seli, P. (2023). Reducing explicit blatant dehumanization by
correcting exaggerated meta-perceptions. Social Psychological and Personality Science, 14(4),
407–418.
Lange, K.-R., Reccius, M., Schmidt, T., Müller, H., Roos, M. W., & Jentsch, C. (2022). Towards extracting
collective economic narratives from texts. Ruhr Economic Papers.
Layman, G. C., Carsey, T. M., & Horowitz, J. M. (2006). Party polarization in american politics:
Characteristics, causes, and consequences. Annu. Rev. Polit. Sci., 9, 83–110.
Lazarsfeld, P. F., Berelson, B., & Gaudet, H. (1968). The people’s choice: How the voter makes up his mind in
a presidential campaign. Columbia University Press.
Leach, C. W., Ellemers, N., & Barreto, M. (2007). Group virtue: The importance of morality (vs.
competence and sociability) in the positive evaluation of in-groups. Journal of personality and
social psychology, 93(2), 234.
Lerner, M. J., & Miller, D. T. (1978). Just world research and the attribution process: Looking back and
ahead. Psychological bulletin, 85(5), 1030.
Levendusky, M. (2009). The partisan sort: How liberals became democrats and conservatives became
republicans. University of Chicago Press.
103

Levendusky, M. (2010). Clearer cues, more consistent voters: A benefit of elite polarization. Political
Behavior, 32, 111–131.
Levendusky, M., & Malhotra, N. (2016). (mis) perceptions of partisan polarization in the american public.
Public Opinion Quarterly, 80(S1), 378–391.
Lieberman, R. C., Mettler, S., Pepinsky, T. B., Roberts, K. M., & Valelly, R. (2019). The trump presidency
and american democracy: A historical and comparative analysis. Perspectives on Politics, 17(2),
470–479.
Lin, Y., Michel, J.-B., Lieberman, E. A., Orwant, J., Brockman, W., & Petrov, S. (2012). Syntactic annotations
for the google books ngram corpus. Proceedings of the ACL 2012 system demonstrations, 169–174.
Lipset, S. M., et al. (1960). Political man. the social bases of politics (Vol. 492). London.
Lipset, S. M. (1964). Canada and the united states-a comparative view. Canadian Review of
Sociology/Revue canadienne de sociologie, 1(4), 173–185.
Lipset, S. M. (2013). Continental divide: The values and institutions of the united states and canada.
Routledge.
Lipset, S. M. (2019). A changing american character? In American social character (pp. 98–133). Routledge.
Lipset, S. M., & Rokkan, S. (1967). Cleavage structures, party systems and voter alignments. lipset. In
S. M. Lipset & S. Rokkan (Eds.), Party systems and voter alignments: Cross-national perspectives
(pp. 3–64). Free Press, New York.
Liptak, A. (2018). How conservatives weaponized the first amendment. The New York Times.
Lupton, R. N., Myers, W. M., & Thornton, J. R. (2017). Party animals: Asymmetric ideological constraint
among democratic and republican party activists. Political Research Quarterly, 70(4), 889–904.
Martin, J., & Burns, A. (2020). Democratic candidates focus on all-consuming hunger to unseat trump.
The New York Times. Retrieved June 10, 2023, from
https://www.nytimes.com/2020/02/01/us/politics/2020-electability.html
Mason, L. (2013). The rise of uncivil agreement: Issue versus behavioral polarization in the american
electorate. American Behavioral Scientist, 57(1), 140–159.
Mason, L. (2018). Uncivil agreement: How politics became our identity. University of Chicago Press.
Matsa, K. E., & Naseer, S. (2022). News platform fact sheet. Pew Research Center.
https://www.pewresearch.org/journalism/fact-sheet/news-platform-fact-sheet/
McClosky, H., & Brill, A. (1983). The dimensions of tolerance: What americans believe about civil liberties.
Russell Sage Foundation.
McClosky, H., & Zaller, J. (1984). The american ethos: Public attitudes toward capitalism and democracy.
Harvard University Press.
104

McGhee, E., Paluch, J., & Romero, M. (2022). Vote-by-mail policy and the 2020 presidential election.
Research & Politics, 9(2), 20531680221089197.
McLeod, J. M., & McDonald, D. G. (1985). Beyond simple exposure: Media orientations and their impact
on political processes. Communication Research, 12(1), 3–33.
Mernyk, J. S., Pink, S. L., Druckman, J. N., & Willer, R. (2022). Correcting inaccurate metaperceptions
reduces americans’ support for partisan violence. Proceedings of the National Academy of Sciences,
119(16), e2116851119.
Meyer, D. S., & Tarrow, S. (Eds.). (2018). The resistance: The dawn of the anti-trump opposition movement.
Oxford University Press.
Meyer, T., & Hinchman, L. (2002). Media democracy: How the media colonize politics. Cambridge: Polity.
Mian, A., Sufi, A., & Khoshkhou, N. (2021). Partisan bias, economic expectations, and household
spending. Review of Economics and Statistics, 1–46.
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in
vector space. arXiv preprint arXiv:1301.3781.
Mitchell, A., Jurkowitz, M., Oliphant, J. B., & Shearer, E. (2020, September 16). Legitimacy of voting by
mail politicized, leaving americans divided. Retrieved June 16, 2023, from
https://www.pewresearch.org/journalism/2020/09/16/legitimacy-of-voting-by-mail-politicizedleaving-americans-divided/
Monin, B., Pizarro, D. A., & Beer, J. S. (2007). Deciding versus reacting: Conceptions of moral judgment
and the reason-affect debate. Review of general psychology, 11(2), 99–111.
Mooijman, M., Hoover, J., Lin, Y., Ji, H., & Dehghani, M. (2018). Moralization in social networks and the
emergence of violence during protests. Nature human behaviour, 2(6), 389–396.
Mooney, C. Z., & Schuldt, R. G. (2008). Does morality policy exist? testing a basic assumption. Policy
Studies Journal, 36(2), 199–218.
Moore, M. (2009). Contemporary debates in political philosophy. In T. Christiano & J. Christman (Eds.).
Blackwell Publishing Ltd.
Moore-Berg, S. L., Ankori-Karlinsky, L.-O., Hameiri, B., & Bruneau, E. (2020). Exaggerated
meta-perceptions predict intergroup hostility between american political partisans. Proceedings
of the National Academy of Sciences, 117(26), 14864–14872.
Muise, D., Hosseinmardi, H., Howland, B., Mobius, M., Rothschild, D., & Watts, D. J. (2022). Quantifying
partisan news diets in web and tv audiences. Science advances, 8(28), eabn0083.
Mullen, E., & Skitka, L. J. (2006). Exploring the psychological underpinnings of the moral mandate effect:
Motivated reasoning, group differentiation, or anger? Journal of personality and social psychology,
90(4), 629.
105

Mutz, D. C. (1994). Contextualizing personal experience: The role of mass media. The Journal of Politics,
56(3), 689–714. Retrieved August 4, 2023, from http://www.jstor.org/stable/2132188
Nelson, T. E., Oxley, Z. M., & Clawson, R. A. (1997). Toward a psychology of framing effects. Political
behavior, 19, 221–246.
Nichols, S. (2002). Norms with feeling: Towards a psychological account of moral judgment. Cognition,
84(2), 221–236.
Nicholson, S. P., & Hansford, T. G. (2014). Partisans in robes: Party cues and public acceptance of supreme
court decisions. American Journal of Political Science, 58(3), 620–636.
Niebler, S. (2020). Vote-by-mail: Covid-19 and the 2020 presidential primaries. Society, 57, 547–553.
Noel, H. (2012). The coalition merchants: The ideological roots of the civil rights realignment. The
Journal of Politics, 74(1), 156–173.
Noel, H. (2014). Political ideologies and political parties in america. Cambridge University Press.
Noelle-Neumann, E. (1974). The spiral of silence a theory of public opinion. Journal of communication,
24(2), 43–51.
Nordin, M. (2014). Do voters vote in line with their policy preferences?—the role of information. CESifo
Economic Studies, 60(4), 681–721.
Norris, P., & Inglehart, R. (2019). Cultural backlash: Trump, brexit, and authoritarian populism. Cambridge
University Press.
Pardos-Prado, S., & Sagarzazu, I. (2016). The political conditioning of subjective economic evaluations the
role of party discourse [Publisher: Cambridge University Press]. British Journal of Political
Science, 46, 799–823. https://doi.org/10.1017/S0007123414000428
Parks, M. (2020). Fact check: Is mail ballot fraud as rampant as president trump says it is? NPR. Retrieved
June 10, 2023, from
https://www.npr.org/sections/coronavirus-live-updates/2020/04/07/829323152/fact-check-ismail-ballot-fraud-as-rampant-as-president-trump-says-it-is
Peeples, L., et al. (2020). Covid and the us election: Will the rise of mail-in voting affect the result. Nature,
10.
Pennington, J., Socher, R., & Manning, C. D. (2014). Glove: Global vectors for word representation.
Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP),
1532–1543.
Petersen, M. B., Skov, M., Serritzlew, S., & Ramsøy, T. (2013). Motivated reasoning and political parties:
Evidence for increased processing in the face of party cues. Political Behavior, 35, 831–854.
Peterson, E. (2019). The scope of partisan influence on policy opinion. Political Psychology, 40(2), 335–353.
106

Peterson, E., & Iyengar, S. (2021). Partisan gaps in political information and information-seeking
behavior: Motivated reasoning or cheerleading? American Journal of Political Science, 65(1),
133–147.
Piaget, J. (1997). The moral judgement of the child. Simon; Schuster.
Pierson, P., & Schickler, E. (2020). Madison’s constitution under stress: A developmental analysis of
political polarization. Annual Review of Political Science, 23, 37–58.
Price, V., Tewksbury, D., & Powers, E. (1997). Switching trains of thought: The impact of news frames on
readers’ cognitive responses. Communication research, 24(5), 481–506.
Prior, M. (2007). Post-broadcast democracy: How media choice increases inequality in political involvement
and polarizes elections. Cambridge University Press.
Puryear, C., Kubin, E., Schein, C., Bigman, Y., & Gray, K. (2022). Bridging political divides by correcting
the basic morality bias.
Qiu, L. (2020). Trump has amplified voting falsehoods in over 300 tweets since election night. The New
York Times. Retrieved June 10, 2023, from https://www.nytimes.com/2020/11/16/technology/trumphas-amplified-voting-falsehoods-in-over-300-tweets-since-election-night.html
Qudar, M. M. A., & Mago, V. (2020). Tweetbert: A pretrained language representation model for twitter
text analysis. arXiv preprint arXiv:2010.11091.
Rahn, W. M. (1993). The role of partisan stereotypes in information processing about political candidates.
American Journal of Political Science, 472–496.
Rauser, A. (1998). Death or liberty: British political prints and the struggle for symbols in the american
revolution. Oxford Art Journal, 151–171.
Rhee, J. J., Schein, C., & Bastian, B. (2019). The what, how, and why of moralization: A review of current
definitions, methods, and evidence in moralization research. Social and Personality Psychology
Compass, 13(12), e12511.
Richter, M. (1995a). The history of political and social concepts: A critical introduction. Oxford
University Press, USA.
Richter, M. (1995b). The history of political and social concepts: A critical introduction. Oxford
University Press, USA.
Rodman, E. (2020). A timely intervention: Tracking the changing meanings of political concepts with
word vectors. Political Analysis, 28(1), 87–111.
Rodriguez, P. L., & Spirling, A. (2022). Word embeddings: What works, what doesn’t, and how to tell the
difference for applied research. The Journal of Politics, 84(1), 101–115.
Rodriguez-Pose, A., Lee, N., & Lipp, C. (2021). Golfing with trump. social capital, decline, inequality, and
the rise of populism in the us. Cambridge Journal of Regions, Economy and Society, 14(3), 457–481.
107

Rokeach, M. (1973). The nature of human values. New York: Free press.
Rokeach, M., & Ball-Rokeach, S. J. (1989). Stability and change in american value priorities, 1968–1981.
American psychologist, 44(5), 775.
Roper, B. W. (1983). Trends in attitudes toward television and other media: A twenty four year review.
Television Information Office.
Rothbard, M. N. (2011). Conceived in liberty. Ludwig von Mises Institute.
Rozin, P. (1999). The process of moralization. Psychological science, 10(3), 218–221.
Rozin, P., Lowery, L., Imada, S., & Haidt, J. (1999). The cad triad hypothesis: A mapping between three
moral emotions (contempt, anger, disgust) and three moral codes (community, autonomy,
divinity). Journal of personality and social psychology, 76(4), 574.
Rozin, P., Markwith, M., & Stoess, C. (1997). Moralization and becoming a vegetarian: The transformation
of preferences into values and the recruitment of disgust. Psychological science, 8(2), 67–73.
Ryan, T. J. (2014). Reconsidering moral issues in politics. The Journal of Politics, 76(2), 380–397.
Ryan, T. J. (2017). No compromise: Political consequences of moralized attitudes. American Journal of
Political Science, 61(2), 409–423.
Saad, L. (2010, May 26). Four moral issues sharply divide americans. Retrieved June 15, 2023, from
https://news.gallup.com/poll/137357/four-moral-issues-sharply-divide-americans.aspx
Sanchez, G. R., Middlemass, K., & Rodriguez, A. (2022, July 26). Misinformation is eroding the public’s
confidence in democracy. Retrieved June 16, 2023, from
https://www.brookings.edu/blog/fixgov/2022/07/26/misinformation-is-eroding-the-publicsconfidence-in-democracy/
Schachner, N. (1951). Thomas jefferson: A biography. (No Title).
Schein, C., & Gray, K. (2018). The theory of dyadic morality: Reinventing moral judgment by redefining
harm. Personality and Social Psychology Review, 22(1), 32–70.
Scheufele, D. A., & Tewksbury, D. (2007). Framing, agenda setting, and priming: The evolution of three
media effects models. Journal of communication, 57(1), 9–20.
Schlesinger, A. M. (1999). The cycles of american history. HMH.
Schlesinger, A. M., & Hart, J. (1950). Paths to the present.
Scott, C. E. (2014). The five conceptions of american liberty. National Affairs, 20, 164–183.
Seidman, L. M. (2018). Can free speech be progressive? Columbia Law Review, 118(7), 2219–2250.
108

Shabad, R. (2020). Trump floats delaying the election, but he can’t do that. NBC News. Retrieved June 10,
2023, from https://www.nbcnews.com/politics/2020-election/trump-suggests-delaying-2020-
election-n1235300
Sharma, K., Seo, S., Meng, C., Rambhatla, S., & Liu, Y. (2020). Covid-19 on social media: Analyzing
misinformation in twitter conversations. arXiv preprint arXiv:2003.12309.
Shehata, A., & Strömbäck, J. (2014). Mediation of political realities: Media as crucial sources of
information. In Mediatization of politics: Understanding the transformation of western democracies
(pp. 93–113). Springer.
Shils, E. (2019). The antinomies of liberalism. In Relevance of liberalism (pp. 135–200). Routledge.
Sides, J., Tesler, M., & Vavreck, L. (2019). Identity crisis: The 2016 presidential campaign and the battle for
the meaning of america. Princeton University Press.
Simon, A. F., & Jerit, J. (2007). Toward a theory relating political discourse, media, and public opinion.
Journal of communication, 57(2), 254–271.
Sipka, A., Hannak, A., & Urman, A. (2022). Comparing the language of qanon-related content on parler,
gab, and twitter. Proceedings of the 14th ACM Web Science Conference 2022, 411–421.
Skelac, I., & Jandrić, A. (2020). Meaning as use: From wittgenstein to google’s word2vec. Guide to Deep
Learning Basics: Logical, Historical and Philosophical Perspectives, 41–53.
Skitka, L. J., & Morgan, G. S. (2014). The social and political implications of moral conviction. Political
psychology, 35, 95–110.
Sniderman, P. M., Brody, R. A., & Tetlock, P. E. (1991). Reasoning and choice: Explorations in political
psychology. Cambridge University Press.
Starr, P. (2007). Freedom’s power: The true force of liberalism. (No Title).
Steiner, H. (2017). Liberty and equality. In Theories of justice (pp. 311–326). Routledge.
Stimson, J. A., MacKuen, M. B., & Erikson, R. S. (1995). Dynamic representation. American political science
review, 89(3), 543–565.
Stringer, D. M., Gellman, M., & Turner, J. (2013). Negative affect. Encyclopedia of behavioral medicine,
1303–1304.
Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American
journal of political science, 50(3), 755–769.
Tavernise, S. (2017). Are liberals helping trump? The New York Times. Retrieved June 10, 2023, from
https://www.nytimes.com/2017/02/18/opinion/sunday/are-liberals-helping-trump.html
109

Thompson, D. M., Wu, J. A., Yoder, J., & Hall, A. B. (2020). Universal vote-by-mail has no impact on
partisan turnout or vote share. Proceedings of the National Academy of Sciences, 117(25),
14052–14056.
Thomson, J. J. (1984). The trolley problem. Yale Law Journal, 94, 1395.
Titley, G. (2020). Is free speech racist? John Wiley & Sons.
Turney, P. D., & Pantel, P. (2010). From frequency to meaning: Vector space models of semantics. Journal
of artificial intelligence research, 37, 141–188.
Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. science,
211(4481), 453–458.
Valdesolo, P., & DeSteno, D. (2006). Manipulations of emotional context shape moral judgment.
PSYCHOLOGICAL SCIENCE-CAMBRIDGE-, 17(6), 476.
van der Eijk, C., Franklin, M., Demant, F., & van der Brug, W. (2007). The endogenous
economy:‘real’economic conditions, subjective economic evaluations and government support.
Acta Politica, 42, 1–22.
Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. science, 359(6380),
1146–1151.
Walker, S. (1999). In defense of american liberties: A history of the aclu. SIU Press.
Wang, S.-Y. N., & Inbar, Y. (2021). Moral-language use by us political elites. Psychological Science, 32(1),
14–26.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief measures of positive
and negative affect: The panas scales. Journal of personality and social psychology, 54(6), 1063.
Webster, S. W. (2018). It’s personal: The big five personality traits and negative partisan affect in
polarized us politics. American behavioral scientist, 62(1), 127–145.
Weiner, G. (2019). The moralistic style in american politics. National Affairs. Retrieved June 13, 2023,
from https://www.nationalaffairs.com/publications/detail/the-moralistic-style-in-americanpolitics
West, D. M. (2020). How does vote-by-mail work and does it increase election fraud? Voter Vitals.
Retrieved June 16, 2023, from https://www.brookings.edu/policy2020/votervital/how-does-voteby-mail-work-and-does-it-increase-election-fraud/
Westwood, S. J., & Peterson, E. (2020). The inseparability of race and partisanship in the united states.
Political Behavior, 1–23.
Wisneski, D. C., & Skitka, L. J. (2017). Moralization through moral shock: Exploring emotional
antecedents to moral conviction. Personality and Social Psychology Bulletin, 43(2), 139–150.
110

Zaller, J. (1992). The nature and origins of mass opinion. Cambridge university press.
Zaller, J. (1996). The myth of massive media impact revived: New support for a discredited idea. Political
persuasion and attitude change, 17, 17–78.
Zimdars, M., & Mantzarlis, A. (2016). False, misleading, clickbait-y, and/or satirical “news” sources.
Retrieved December 1, 2021, from
https://docs.google.com/document/d/1zhaZooMfcJvk_23in201nviWJN1-LhRvGlPXJWBrPRY/edit
111

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Partisan media bias vs. structural media bias: which one dominates in a polarized environment?

PDF

Improving information diversity on social media

PDF

Socially-informed content analysis of online human behavior

PDF

The small dollar political donor: why regular folks give money to politics

PDF

Focusing the family: the rhetorical construction of "family values" in contemporary cultural politics

PDF

Analysis and prediction of malicious users on online social networks: applications of machine learning and network analysis in social science

PDF

Bound in hatred: a multi-methodological investigation of morally motivated acts of hate

PDF

Can ideologically relevant threat shift group-oriented values? Relevant threatening tweets cause Progressives to be as prejudiced as Conservatives

PDF

Political coming-of-age in the era of WeChat: understanding the ethnic media ecosystem and group politics of first-generation Chinese Americans

PDF

Democrats vs. Republicans: how Twitter users share and react to political fact-checking messages

PDF

Public opinion and international affairs: a multi-method approach to foreign policy attitudes

PDF

Set values for mean field games and set valued PDEs

PDF

How misinformation exploits moral values and framing: insights from social media platforms and behavioral experiments

PDF

How "reducing" affective polarization increases politically motivated reasoning

PDF

Essays on political economy and corruption

PDF

Curing hooliganism: moral panic, juvenile delinquency, and the political culture of moral reform in Britain, 1898-1908

PDF

Staying ahead of the digital tsunami: strategy, innovation and change in public media organizations

PDF

Voter engagement in the 2020 presidential election: what we can learn from the voter engagement initiatives in California-based and national nonpartisan organizations to increase voter participation

PDF

Forests, fires, and flights: examining safety and communication practices within aerial firefighting teams

PDF

The politics of transformation: mass media and the Northern Ireland peace process

Asset Metadata

Creator Zhang, Meiqing (author)

Core Title Three text-based approaches to the evolution of political values and attitudes

School Annenberg School for Communication

Degree Doctor of Philosophy

Degree Program Communication

Degree Conferral Date 2023-12

Publication Date 10/20/2023

Defense Date 08/25/2023

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag democratic values,machine learning,moralization of politics,natural language processing,OAI-PMH Harvest,opinion dynamics,partisan bias,political polarization

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Ferrara, Emilio (committee chair), Riley, Patricia (committee chair), Barberá, Pablo (committee member), Hollihan, Thomas (committee member)

Creator Email meiqingz@usc.edu,xiyangmili@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113761204

Unique identifier UC113761204

Identifier etd-ZhangMeiqi-12388.pdf (filename)

Legacy Identifier etd-ZhangMeiqi-12388

Document Type Dissertation

Format theses (aat)

Rights Zhang, Meiqing

Internet Media Type application/pdf

Type texts

Source 20231030-usctheses-batch-1103 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu