University of Southern California Dissertations and Theses

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

(USC Thesis Other)

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

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content PUBLIC OPINION AND INTERNATIONAL AFFAIRS:
A MULTI-METHOD APPROACH TO FOREIGN POLICY ATTITUDES
by
Evgeniia Iakhnis
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulllment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POLITICAL SCIENCE AND INTERNATIONAL RELATIONS)
December 2020
Copyright 2020 Evgeniia Iakhnis
Acknowledgments
First and foremost, I extend my gratitude to my dissertation committee: Patrick James, Brian
Rathbun, and Nathanael Fast. Pat and Brian have formed an ideal team of dissertation advisors
for me. Pat has become my advisor early on and gave me an amazing opportunity to work with
him on a variety of projects and learn best research practices. Serving as Pat's research assistant, I
got the best possible preview of how the life of an academic would be. Pat always nds the perfect
balance between pushing me to work hard and providing me with encouragement and support when
I need it most. Besides, he has always given me the freedom to discover what I want to do and
supported me in all my choices.
Brian has been the perfect complement to Pat and my intellectual mentor. Thanks to Brian
I became interested in the topic of public opinion and foreign policy attitudes in international
relations that formed my entire research agenda. Without doubt, Brian has had the highest impact
on how I view the discipline overall. He saw potential in me early on and taught me how to
produce high quality research. From Brian, I learned how to conduct a project from A to Z, how to
collaborate in academia, and how to go through the weeds of the publication process. It is worth
mentioning that Brian is incredibly straightforward and honest in his criticism { his timely and
most rigorous feedback helped me improve and move forward in my career. And, denitely, I wish
I could always be as excited about my research projects as Brian is!
Finally, I am very grateful to Nate for serving as an external member of my committee. I truly
enjoyed participating in several sessions at his Hierarchy, Networks, and Technology Lab (HiNT
Lab) and always found the discussions fascinating and thought-provoking. From these sessions
I learned how psychological insights can contribute to our understanding of organizations and
familiarized myself with research practices outside of the eld of political science and international
ii
relations.
Beyond my committee, I have learned a great deal from many other scholars at USC. One
of the biggest credits goes to Pablo Barbera who gave me an opportunity to join his Networked
Democracy Lab. As a member of this lab, I got an opportunity to participate in some of the
most exciting research projects, met some amazing external speakers, and became a part of a
small community of quantitative researchers here at USC. Besides, thanks to Pablo, I discovered
opportunities of being a researcher outside of academia and followed that path. I would also like
to thank Nicholas Weller, Morris Levy, James Lo, and Gourab Mukherjee for further contributing
to my methodological training at USC. I also extend my thanks to Veri Chavarin for being always
so helpful and responsive. The department would not be the same without you!
I am also very grateful to my colleagues for all the support and camaraderie. I thank all
members of my wonderful cohort for making the rst two years of this PhD so enjoyable. And
even though we do not see each other as frequently now, I hope we can stay in touch after we all
graduate. I also thank Daniela Maag, D avid Somogyi, Shiming Yang, Mery Farinas, Xinru Ma,and
Peter Knaack for being the best post-communist kids ever and exploring the Los Angeles culinary
scene together.
Beyond people at USC, I am innitely grateful to my large support group back in Russia. First,
my mother Irina and my father Iakov have been the most supportive when I told them of my
decision to go do a PhD a thousand miles away from home. I know that you are always there for
me, regardless how far, and I work hard to make you proud! I am also grateful to my dear friends
Vera, Marina, and Natasha who are always inspiring and encouraging me to do better. Besides my
family back in Russia, I also need to thank my newly acquired family here in Los Angeles: Nadia,
Badawy, Yasmine, Heba, and Mona. Thank you for making me feel like home away from home and
being always so innitely kind and supporting.
And nally, none of this would have been possible without Adam. If not for you, I would have
probably dropped out of this program after the rst semester and never fullled my full potential.
You always believed in me even when I myself did not. You always supported and guided me in
the right direction. All my successes are your successes, and I would not have gone so far without
your constant support, patience, and encouragement. Thank you for everything.
iii
Contents
Acknowledgments ii
List of Tables vii
List of Figures ix
Abstract xi
1 Introduction 1
1.1 How do People Attribute Blame in International Relations? . . . . . . . . . . . . . . 3
1.2 How do People Form Moral Judgements about Other Countries? . . . . . . . . . . . 4
1.3 How do Political Leaders Interact on Social Media? . . . . . . . . . . . . . . . . . . . 5
2 Attribution Biases in Perceptions of International Aairs 7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Attribution theory in International Relations . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Group Categorization and Judging Intentions . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Individual dierences in attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Attributions and Policy Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.6 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7.1 Group Categorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.7.2 The Heterogeneous Pattern of Attributions . . . . . . . . . . . . . . . . . . . 21
2.7.3 Attributions and Policy Preferences . . . . . . . . . . . . . . . . . . . . . . . 24
iv
2.8 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3 External Threat and Moral Judgement 30
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2 Threat and moral judgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Ideology and moralization of threats . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Study 1: Moral Rhetoric about Adversarial Countries on Twitter . . . . . . . . . . . 37
3.4.1 Data and Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 Study 2: An Experimental Test of Relationship between Threat and Moral Judge-
ment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.1 Data and Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4 Networks of Power: Analyzing World Leaders' Interactions on Social Media 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 How do Leaders Interact on Social Media? . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.1 Twitter Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.3.2 Retweet and Mention Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4.1 Leaders' clustering and interaction patterns . . . . . . . . . . . . . . . . . . . 68
4.4.2 Heatmaps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.4.3 Predictors of Social Media Centrality . . . . . . . . . . . . . . . . . . . . . . . 72
4.4.4 Checking Network Stability Over Time . . . . . . . . . . . . . . . . . . . . . 77
4.5 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5 Conclusion 81
Bibliography 87
v
Appendices 104
A Appendix for Chapter 2 105
A.1 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
A.2 Foreign Policy Attitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
A.3 Saturated Interaction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.4 Eect of Ideology on Attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
A.5 Replication of results for attentive respondents only . . . . . . . . . . . . . . . . . . 111
A.6 Survey Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
A.7 Robustness Check: Controlling for Foreign Policy Orientations . . . . . . . . . . . . 115
A.8 Nonparametric Mediation Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
B Appendix for Chapter 3 118
B.1 Survey Instrument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
B.2 Survey Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
C Appendix for Chapter 4 121
C.1 Examples of Mention Tweets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
C.2 Robustness Checks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
C.3 Retweet Network results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
vi
List of Tables
2.1 Categorization of International Actors: Shared Characteristics . . . . . . . . . . . . 13
2.2 Eect of Country Type on Dispositional Attributions . . . . . . . . . . . . . . . . . . 20
2.3 Eect of Ideology on Attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Eect of Attributions on Policy Preferences . . . . . . . . . . . . . . . . . . . . . . . 26
3.1 Twitter Data Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Top 15 Stemmed Moral Words from Liberal and Conservative Users . . . . . . . . . 43
3.3 Eect of Ideology on Perceptions of Leaders' as Trustworthy and Aggressive . . . . 53
3.4 Eect of Ideology on Perceptions of Leaders' as Greedy and Resolute . . . . . . . . . 54
4.1 Twitter Data Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2 Distribution of Accounts by Type of Actor, Regime, and Region . . . . . . . . . . . 65
4.3 Descriptive statistics of the Leaders' Retweet and Mention Network. . . . . . . . . . 66
4.4 Mention Network Assortativity Coecients . . . . . . . . . . . . . . . . . . . . . . . 70
4.5 Mention Network Centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
A.1 Saturated Interaction Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
A.2 Eect of Ideology on Attributions (Attentive Respondents Only) . . . . . . . . . . . 111
A.3 Survey sample characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
A.4 Eect of Ideology on Attributions (weighted) . . . . . . . . . . . . . . . . . . . . . . 113
A.5 Eect of Attributions on Policy Preferences (weighted) . . . . . . . . . . . . . . . . . 114
A.6 Eect of Ideology on Attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
B.1 Survey sample characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
vii
C.1 Alternative Specications of the Model . . . . . . . . . . . . . . . . . . . . . . . . . . 126
C.2 Retweet Network Assortativity Coecients . . . . . . . . . . . . . . . . . . . . . . . 127
C.3 Retweet Network Centrality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
viii
List of Figures
2.1 Dispositional Attribution by Treatment Group . . . . . . . . . . . . . . . . . . . . . 19
2.2 Conditional Eect of Ideology on Attributions . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Eect of Dispositional Attributions on Support for Harsh Response . . . . . . . . . . 27
3.1 Percentage of Posts about Adversarial Countries by Ideology . . . . . . . . . . . . . 41
3.2 Percentage of Posts about Adversarial Countries by Moral Content . . . . . . . . . . 42
3.3 Percentage of Posts by Moral Content and Ideology . . . . . . . . . . . . . . . . . . 44
3.4 Breakdown of Negative Moral Content by Type . . . . . . . . . . . . . . . . . . . . 46
3.5 Eect of Threat on Moral Perceptions of Country . . . . . . . . . . . . . . . . . . . . 50
3.6 Eect of Threat on Moral Perceptions of Country's Leaders . . . . . . . . . . . . . . 51
4.1 Volume of unique users and tweets by month . . . . . . . . . . . . . . . . . . . . . . 64
4.2 Leaders' Mention Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3 Clustering by Specic Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4 Clustering by Specic Type of Regime . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Distribution of Weighted Indegree . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.6 Eect of Regime on Probability of Network Centrality . . . . . . . . . . . . . . . . . 77
4.7 PageRank (left) and Weighted in-degree (normalized by sum of in-degree per time
period) (right) statistics across the whole time period . . . . . . . . . . . . . . . . . 78
4.8 PageRank (left) and Weighted in-degree (normalized by sum of in-degree per time
period) (right) statistics form 2015 onwards, where more than 86% of the mention
data is represented . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
ix
A.1 Conditional Eect of Ideology on Attributions . . . . . . . . . . . . . . . . . . . . . . 110
A.2 Mediating Eect of Ideology on Causal attributions . . . . . . . . . . . . . . . . . . . 117
C.1 Examples of Mentions by World Leaders . . . . . . . . . . . . . . . . . . . . . . . . . 122
C.2 Examples of Mentions by World Leaders . . . . . . . . . . . . . . . . . . . . . . . . . 123
C.3 Word Cloud of 200 Most Common Words in Mention Tweets . . . . . . . . . . . . . 124
C.4 Retweet Network Type of Government Clustering . . . . . . . . . . . . . . . . . . . . 128
C.5 Retweet Network Region Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
x
Abstract
This dissertation uses a multi-method approach to explore how the public forms foreign policy at-
titudes and how political leaders use online platforms to form public opinion at home and abroad.
The introduction begins with a brief overview of the literature on public opinion, introduces the
studies, and presents a case for combining experimental and computational methodologies to ad-
dress various questions in international relations research. Chapter 2 uses attribution theory to
explain how people make inferences about complex and ambiguous international events. It uses
experimental design to show that both identity of the country involved and individual predisposi-
tions of the inferring subjects aect attributions of blame. Chapter 3 investigates how people form
moral judgement about adversarial foreign nations and the role of intergroup threat in this process.
Using a combination of computational and experimental methodology, it shows that perception of
intergroup threat might lead to the formation of moral judgment about the source of threat, even
when no information about morality of the group is available. Chapter 4 examines how political
leaders use online platforms to form public opinion domestically and internationally. Using a large
sample of political tweets, I explore how leaders strategically interact with each other on social
media to demonstrate collaboration, attract supporters abroad, shape the con
ict narrative, and
demonstrate resolve. The dissertation concludes with a review of the ndings and suggestions for
future research.
xi
Chapter 1
Introduction
This article-style dissertation consists of three studies tied together by the theme of foreign pol-
icy attitudes. In democratic countries, public opinion plays a central role in decision making,
representation, and accountability; yet, a vast body of early research on the topic assumed that
foreign policy attitudes of the mass public were incoherent, disorganized, and insignicant (Almond,
1960, Lippmann, 1955, Converse, 1964, Erskine, 1963, Free and Cantril, 1968). According to the
\Almond-Lippmann consensus" that emerged in the the aftermath of the Second World War, public
opinion on foreign policy is volatile, lacks coherence or structure and has little impact on foreign
policy (Holsti, 1992). Speaking about foreign policy attitudes among the American public, Almond
depicted public opinion as a volatile (\Often the public is apathetic when it should be concerned,
and panicky when it should be calm" (Almond, 1956)) and poorely informed (\The world outside
is still very remote for most Americans; and the tragic lessons of the last decades have not been
fully digested" (Almond, 1960)).
In sharp contrast with this early cynicism, nowadays there is mounting evidence that mass
opinion about foreign policy is considerably more stable and might prove in
uential for international
relations. Mass foreign policy attitudes are coherent and reasonable (Page and Shapiro, 1983, 1992,
Shapiro and Page, 1988, Russett, 1990); they are well-structured in stable foreign policy orientations
(Wittkopf, 1990, Herrmann et al., 1999, Rathbun, 2007); rooted in core moral and personal values
(Kertzer et al., 2014, Rathbun et al., 2016) and might even have genetic underpinnings (McDermott
et al., 2009). In terms of their impact, we now have evidence that foreign policy attitudes often
play a role in electoral campaigns (Aldrich et al., 1989, Anand and Krosnick, 2003, Gelpi et al.,
1
2006, Gronke et al., 2003, Fiorina, 1981), constrain democratic leaders on their powers to use force
(Morgan and Campbell, 1991, Hildebrandt et al., 2013, Hartley and Russett, 1992), and aect policy
decisions (Mueller, 1973, Sobel, 2001, Russett, 1990). Public opinion might be especially in
uential
in relation to con
icts or crises that tend to attract the attention of a public that is usually
not closely attuned to the international events (Baum and Potter, 2008, Sobel, 2001). However,
international crises are usually highly complex, involve several actors, and various situational and
dispositional factors and, until now, we do not fully understand how people perceive ambiguous
international events and form reasoned positions in the conditions of limited attention and cognitive
constraints. This dissertation seeks to contribute to our understanding of this important topic.
In this dissertation I follow a psychological approach to foreign policy attitudes that takes into
account both situational and dispositional factors in the process of belief formation. Rather than
treating public as homogenous, I consistently pay attention to individual characteristics, such as
political ideology and individual moral values. In chapter 2, I use attribution theory developed in
social psychology to explain how people make inferences about complex and ambiguous interna-
tional events. Using experimental design, I show that attributions about international events are
primarily in
uenced by two factors: the identity of the country involved and individual character-
istics of the inferring subjects. Only considering both factors we can built a complete picture of
how people interpret the ambiguous world around them.
In chapter 3, I continue exploring the topic of belief formation using both situational and dispo-
sitional predictors. Specically, I investigate how people form moral judgement about adversarial
foreign nations and the role of intergroup threat in this process. Using a large dataset of political
tweets and a survey experiment, I show that perception of intergroup threat might lead to the
formation of moral judgment about the source of threat, even when no information about morality
of the group is available. Besides, on the dispositional side, I demonstrate an individual variation in
the people's propensity to moralize threats. Finally, in chapter 4, I take a dierent approach and ex-
plore how political leaders use online platforms to form public opinion at home and abroad. I study
how leaders strategically interact with each other on social media to demonstrate collaboration,
attract supporters abroad, shape the con
ict narrative, and demonstrate resolve.
In the remainder of the chapter, I brie
y introduce the empirical chapters. Taken together, the
chapters advance our understanding of public opinion on foreign policy and highlight the importance
2
of this research area in international relations studies. Besides, methodologically, this dissertation
advances a multi-method approach to the study of international relations. Chapter 2 relies on a
methodology of a survey experiment that has traditionally been used to explore public attitudes,
including foreign policy attitudes in IR (Kertzer and Tingley, 2018). However, while experimental
studies are unparalleled for testing causal relationships, they have serious limitations related to
achieving acceptable levels of external validity and generalizability (McDermott, 2011). In order
to bridge the gap between internal and external validity, in Chapter 3 I use a combination of an
online survey experiment and an observational big data study. First, I analyze a large sample of
real discussions about adversarial countries on Twitter to trace how people talk about threatening
groups in natural conditions. Then, I conduct a survey experiment to test the causal nature of the
relationship between intergroup threat and moral judgement. Taken together, the ndings from
both studies shed light on the processes of moral judgement under conditions of threat.
Finally, in Chapter 4, I continue with big data approach and analyze a large sample of online
communication between political leaders on social media. I use a novel methodology of network
analysis that is particularly suitable for modelling multi-state relations.(Cranmer and Desmarais,
2016). Network analysis allows for most eective treatment of relational data through the forma-
tion of interconnected relational systems. While this approach is purely descriptive and does not
allow for testing of causal relationship, it provides the most accurate descriptive representation
of the interactions among leaders on social media platforms. Overall, in this dissertation, I show
that experimental and computational approaches can be combined eectively to address various
questions in international relations and public opinion research.
1.1 How do People Attribute Blame in International Relations?
We know that public is generally uninformed about foreign policy and does not pay close attention
to international events (Holsti, 2004, Aldrich et al., 2006). Considering people's limited knowledge
about international relations and severe cognitive limitations, how do people comprehend what
is happening and make inferences? How do people judge intentions from ambiguous international
actions, and form subsequent policy preferences? Attribution theory developed in social psychology
might provide answers to these kinds of questions (Heider, 1958, Jones and Davis, 1965, Jones
3
and Nisbett, 1971, Pettigrew, 1979). The study of attribution theory has also had a profound
eect on our understanding of international relations (Heradstveit, 1979, Heradstveit and Bonham,
1996, Hirshberg, 1993, Mercer, 1996, Dickson, 2009, Larson, 1985). Yet the current research on
attributions in international relations suers from serious shortcomings. In chapter 2, I argue that
attributions about international events are primarily in
uenced by two factors that have not been
suciently theorized in the existing literature: the identity of the country involved and individual
predispositions of the inferring subject. Existing research relies on a crude division between a home
country and countries, treating all other states homogeneously as a uniform outgroup (Mercer,
1995, Larson, 1985, Heradstveit, 1979, Hirshberg, 1993). Using an experimental design, I show that
a binary categorization of international actors into ingroup and outgroup is insucient. Instead, I
develop and test a more ne-grained categorization scheme based on shared characteristics between
countries, and demonstrate that countries with strong similarities to the home country are perceived
as part of the ingroup with corresponding attributional patterns.
Besides, I also argue that the existing assumption of unitary audiences prevalent in attributional
research is insucient; rather, dierent types of individuals have dierent reaction to the same
events. From existing research on political psychology, we know that ideological leanings aect
causal attribution process (Williams, 1984, Zucker and Weiner, 1993). I provide evidence that
conservatives display very dierent attributional patterns than liberals: those who score high on
conservatism tend to discriminate against outgroups and favour ingroups when making causal
attributions. Finally, I provide compelling evidence that attributions matter since they translate
into policy support. Overall, the ndings of this chapter account for inconsistent results in previous
research on attributions and demonstrate the unrealized potential that attribution theory might
hold for our understanding of foreign policy attitudes.
1.2 How do People Form Moral Judgements about Other Coun-
tries?
Perception of a target as moral or immoral has been shown to have clear behavioral consequences,
including support for aggressive and belligerent retaliatory policies (Opotow, 1990, Staub, 1990).
Despite that, we still know very little about how people form negative moral judgement and what
4
the role of intergroup threat is in this process. Existing literature has shown that threat might
in
uence public opinion in many circumstances (Herrmann et al., 1999, Arian, 1989, Friedland
and Merari, 1985, Huddy et al., 2005); however, the possible causal relationship between integroup
threat and moral evaluations remains largely unexplored. In chapter 3, I ll this gap and argue that
perception of intergroup threat might lead to the formation of moral judgment about the source of
threat, even when no information about morality of the group is available. Using a combination of
observational data and a survey experiment, I show that people have the propensity to see source
of threat as less moral even if they have no other reasons to make negative moral evaluations about
it.
Besides, I also demonstrate that there is an individual variation in the people's propensity
to moralize threats. People identifying with conservative and right-wing ideology are generally
higher in authoritarianism, social dominance, and system justication that make them more prone
to intolerance, prejudice, and hostility toward a wide variety of out-groups (Duriez and Van Hiel,
2002, Jost et al., 2003, 2009a, Sidanius and Pratto, 2001a). Besides, according to the negativity bias
perspective, conservatives are especially responsive to negative stimuli and events (Hibbing et al.,
2014, Bonanno and Jost, 2006, Oxley et al., 2008, Amodio et al., 2007). In line with these theories
I nd that people with conservative leanings tend to deny morality to the threatening groups more
strongly than liberals. When discussing adversarial foreign countries on Twitter, conservatives
tend to use more negative moral rhetoric. Also, when being primed with external threat in a
survey experiment, the more conservative the respondents, the lower they evaluate the country's
leaders in terms of their trustworthiness and higher in terms of their aggressiveness. Overall, the
ndings of this chapter help connect the research on intergroup threat with the study of morality
in international relations and demonstrate the vast unrealized potential that this research agenda
might hold for our understanding of foreign policy attitudes.
1.3 How do Political Leaders Interact on Social Media?
World leaders have been increasingly using social media platforms as a tool for political communi-
cation. Presidents, prime ministers, ministers of foreign aairs and other high-positioned politicians
around the world use Twitter and Facebook to broadcast messages to both domestic and interna-
5
tional audiences and aect public opinion at home and abroad. Previous work has examined how
political leaders use social media strategically to divert public attention from domestic problems,
bolster regime legitimacy, and suppress domestic opposition (Barber a et al., 2018, Gunitsky, 2015,
Pearce, 2015). However, despite the growing research on governmental accounts on social media,
virtually nothing is known about interactions among world leaders. Online communication between
leaders represents a new form of diplomacy that bypasses traditional channels and unfolds openly in
front of domestic and international publics (Su and Xu, 2015, Strau et al., 2015). Besides, online
interactions between leaders might shed light on important con
ictual or collaborative relationships
that emerge between world leaders online. Previous research shows that direct social media inter-
actions were used to shape the con
ict narrative, demonstrate resolve, attract supporters abroad,
and in
uence public opinion (Zeitzo, 2018).
Using a novel, cross-national dataset of Twitter communication for leaders of 193 U.N. member
countries for the period of 2012-2017, I construct retweet and mention networks to explore the
patterns of leaders' communication. I have three key ndings. First, I conclude that leaders'
interaction on social media closely resemble their interaction in the oine world. Similarly to
the oine diplomatic communities, I show that leaders form mention/retweet communities along
regional lines and similar levels in political hierarchy. Second, I nd that the regime type plays
a key role in the way Twitter communities are formed. Consistent with the expectations of the
democratic peace theory, I demonstrate that leaders from democratic states are more likely to
engage with other democratic leaders. Finally, I explore the popularity patterns among the world
leaders and conclude that leaders of democratic countries tend to occupy more central positions
in the network, with leaders of non-democratic countries present at the periphery. Overall, this
chapter yields new insights on how social media is used by government actors, and have important
implications for our understanding of the impact of new technologies on the new forms of diplomacy.
Overall, taken together, the chapters of this dissertation advance our understanding of public
opinion on foreign policy and highlight the importance of both situational and dispositional pre-
dictors in the process of belief formation. Besides, methodologically, this dissertation advances a
multi-method approach to the study of international relations. Across the three chapters I show
that experimental and computational approaches can be combined eectively to address various
questions in international relations and public opinion research.
6
Chapter 2
Attribution Biases in Perceptions of
International Aairs
2.1 Introduction
How do people attribute blame in international relations? And how do attributions aect people's
decision to support certain policies including the use of force? In this study, I argue that attributions
of blame are key for understanding the foreign policy attitudes of the mass publics. By aecting
inferences about other countries' intentions, attributional patterns translate into support of certain
policies, including the most aggressive ones (Sadler et al., 2005). However, despite a strong link
between causal attributions and policy preferences, we still lack a convincing theory of when and
why people tend to stress internal, dispositional factors over external, situational forces. Here,
I argue that attributions about international events are primarily in
uenced by two factors that
have not been suciently theorized in the existing literature: the identity of the country involved
and individual predispositions of the inferring subject. When presented with information about an
ambiguous event when a civilian air plane was shot down by a foreign country, people's attributions
vary dramatically depending on the identity of the country in question. Besides, subjects dier in
their reaction to similar events.
Previous studies rely on a crude division between a home country and other countries, treating
all other states homogeneously as a uniform outgroup (Mercer, 1995, Larson, 1985, Heradstveit,
7
1979, Hirshberg, 1993). However, this does not capture the full range of relationships among
nation-states, relationships that likely aect attributions. Using insights from image theory and
democratic peace theory, I propose a novel, comprehensive theory of categorization based on the
shared characteristics between countries. The type of attributions change systematically when we
move from non-allies without shared characteristics to strong allies with common enemy, shared
experience of ghting together in previous con
icts, democratic regime, and Western culture. As
we move from non-allies to strong allies, we see that the percentage of respondents who make
dispositional attributions drops dramatically from 62.7% to 34%. We attribute negative actions on
the part of those closest to us by reference to situation and those furthest from us by reference to
disposition.
I also argue that the existing assumption of unitary audiences prevalent in attributional research
is insucient; rather, dierent types of individuals have dierent reaction to the same events. In
particular, liberals and conservatives have very dierent attributional patterns: those who score
high on conservatism tend to discriminate against outgroups and favour ingroups when making
causal attributions. More liberal individuals exhibit greater egalitarianism. Their attributional
patterns are less aected by the identity of the country in question.
Finally, I present compelling evidence that attributions matter since they translate into policy
support: when people make dispositional attributions, they are roughly 20 points more likely to
endorse harsh policies than people who make situational attributions. By aecting inferences about
other countries' intentions, attributional patterns in
uence policy preferences, potentially aecting
policy decisions.
2.2 Attribution theory in International Relations
Considering all the complexities of international crises and people's limited knowledge about world
politics, how do they comprehend what is happening and make inferences? How do people attribute
blame for international actions and form subsequent policy preferences?
Attribution theory developed in social psychology might provide answers to these kinds of ques-
tions. According to attribution theory, outcomes can be either attributed to situational factors that
correspond to external factors out of an actor's control, or dispositional factors that refer to internal
8
reasons, such as beliefs, motives, or character traits (Heider, 1958). While complex international
interactions are usually driven by the combination of systemic pressures and individual decisions,
people simply cannot process the whole body of relevant information and understand situation
comprehensively due to severe cognitive limitations (Tetlock, 1998). Instead, they tend to rely on
cognitive shortcuts by emphasizing either a nation's nature or situational forces and deemphasizing
other factors. Thus, the key question is: when and why people emphasize dispositional factors over
situational forces?
Here I argue that attributions about international events are primarily in
uenced by two factors
that have not been suciently theorized in the existing literature on attributions in IR: the identity
of the country involved and the individual predispositions of the inferring subject. The identity
of the country involved plays a key role as inferences are generally susceptible to a group-serving
attribution bias. Group-serving bias is a motivated attribution bias driven by people's psychological
and emotional needs to preserve one's ego and present oneself in the best light (Fiske and Taylor,
1991). Group-serving bias propels people to explain positive actions on the part of the ingroup
through dispositional (internal) factors and negative actions through situational (external) factors
(Pettigrew, 1979). Conversely, when judging other groups, positive actions are perceived as caused
by situational factors while negative actions are viewed as internally driven.
Group-serving or ethnocentric bias has been shown to have a strong and robust eect on inter-
pretation of other countries' behavior (Larson, 1985, Heradstveit, 1979, Mercer, 1996, Hirshberg,
1993). However, existing literature on attributions in IR tends to treat all foreign states homo-
geneously, as a uniform outgroup. For example, Mercer (1996) argues that only a home country
constitutes an ingroup, while all other countries are considered an outgroup in world politics with
respective attributional patterns. There are reasons to question this assumption, such as frequent
reference to \special relationship" between some countries (such as the United States and Great
Britain (Riddell, 2004)).
Besides anecdotal evidence, some academic research also indicates that such crude assumption
is not justied. While not using attribution theory explicitly, image theory scholars provide strong
evidence that not all foreign countries are perceived in the same way (Herrmann and Fischerkeller,
1995, Herrmann et al., 1997). We also have strong evidence that countries with dierent regime
types activate dierent inferential processes. For example, democratic peace literature shows that
9
mass publics make dierent inferences about democratic and non-democratic countries, at least in
terms of threat perception (Tomz and Weeks, 2013). While existing literature oers some indication
that not all foreign nations are perceived in the same way, it does not provide a unied theory of
categorization of foreign nations. Most survey experiments draw a distinction between allies and
adversaries, thus treating allies as a single, uniform group (Herrmann et al., 1997, Tomz and Weeks,
2013). At the same time, the literature on alliance formation indicates that alliances are far from
uniform. Thus, in order to understand attributions, we need a better conceptualization of how
individuals in the mass public might categorize foreign nations.
Besides country-identity, the process of causal attribution in IR cannot be fully understood
without reference to individual predispositions. From existing research on political psychology, we
know that ideological leanings aect causal attribution process (Williams, 1984, Zucker and Weiner,
1993). For example, Grin and Oheneba-Sakyi (1993), show that conservatives attribute poverty
to individual dispositions whereas liberals attribute poverty to situational sources. However, to the
best of my knowledge, no IR studies have yet explored possible heterogeneous treatment eects
due to specic individual predispositions of blame attribution. Instead, studies generally implicitly
assume that individuals react identically towards ingroups and outgroups.
Therefore in the following sections, I improve on the existing application of attribution theory
in IR by relaxing two assumptions prevalent in the existing literature: binary categorization and
homogenous audience.
2.3 Group Categorization and Judging Intentions
Following research on attribution, I conceptualize ingroup and outgroup based on the expecta-
tions that a group generates: an outgroup is a group that mostly produces negative expectations,
while ingroup is a group that mostly produces positive expectations. This denition allows for
re-conceptualization of ingroup-outgroup categorization from the traditional dichotomous under-
standing to a continuum that ranges from countries with the most negative expectations (strong
outgroup) to countries with the most positive expectations (strong ingroup).
I argue that an individual's expectations about the behavior of other countries are based on
a sense of the perceived shared characteristics between those countries and one's home country.
10
Here, I suggest a theory of ingroup-outgroup categorization that should be applicable to a set of
Western, democratic countries where mass public attitudes are likely to matter most. Later, I will
discuss how this theory can be adopted to countries that do not t this denition.
Uncontroversially, enemies are most likely to be considered an outgroup and should produce the
least positive expectations. According to the image theory cited above, enemy and ally are distinct
and separate schematas primarily distinguished by the attributed motivation (Herrmann et al.,
1997, Boulding, 1959). Distinct images of enemies and allies aect the interpretation of incoming
information as subjects attach schema-consistent meaning to otherwise ambiguous actions. In
order to eliminate unwanted connotations with a word \enemy" (people might perceive enemies as
not simply countries with dierent characteristics, but as countries with ongoing military con
ict
or history of previous military aggression), here I adopt a more neutral concept of \non-ally".
While neutral, this concept also implies fundamental dierences (i.e. lack of shared characteristics)
between countries; therefore, non-allies are likely to produce mostly negative expectations and to
be perceived as an outgroup.
If non-allies generate mostly negative expectations, what about the allies? On the one hand,
existing research on image theory and the democratic peace literature demonstrate that we cannot
simply assert that all other states except for the home country are considered to be in the outgroup
(Herrmann et al., 1997, Tomz and Weeks, 2013). At the same time, existing research on attributions
in IR indicates that we cannot simply assume the opposite either: allies do not produce strictly
positive expectations and are not always considered as part of the ingroup. Previous attempts
to widen the ingroup category to include the members of a military alliance led to inconsistent
ndings. For example, in the context of the Gulf War, Heradstveit and Bonham (1996) found that
Arab elites dierentiated between the behavior of the U.S and other members of the Coalition.
Similarly, Mercer (1996) shows that allied countries in three crises from 1905 to 1911 considered
each other as an outgroup.
Such inconsistencies in previous research indicate that not all allied countries generate similar
expectations, and expectations can dier drastically depending on the degree to which allies share
characteristics. Allies with strong similarities to a home country should generate strong positive
expectations and be considered as part of the ingroup, while allies with weak similarities would
generate less positive expectations, perhaps even remaining in the outgroup category. Therefore I
11
develop a more nuanced typology of allied relationships that might aect attributions.
Allies with weak similarities to a home country are instrumental allies. Alliances formed and
sustained in response to a common threat help to enhance security by providing credible signals of
their future intentions (Morrow, 1994) or by exploitation of joint production economies (Conybeare,
1992, Lake, 1999); however, they are based on a simple need to deter potential adversaries rather
than any strong tie between the countries. The principle of the \enemy of my enemy is my friend"
does not usually result in a strong, sustainable friendship as it lacks any other connective tissues.
Therefore, transactional allies might not produce strong positive expectations with corresponding
attributional patterns.
If all alliances were based on purely security considerations, then all allies would be perceived
as a uniform outgroup. However, existing scholarship on alliance formation demonstrates that not
all alliances are motivated solely by security considerations, they are also based on shared char-
acteristics between countries. For example, in addition to security considerations, alliances can
be formed (or deepened after they are formed) on the basis of shared experience. The shared ex-
perience of ghting a common enemy is likely to strengthen an alliance by providing information
about the reliability of its members. Information that a country previously honored its commit-
ment decreases the expected costs of opportunism and, thus, increases the probability for strong
cooperative agreements (Lake, 1999, Risse-Kappen, 1995a). Therefore, we can expect that allies
with shared experience of military engagement will be treated by the public with stronger anity
than transactional allies without any shared experience.
Besides security considerations and shared experience, countries can also form alliances on the
basis of common democratic norms. Two mechanisms have been explored: the liberal institu-
tionalist tradition emphasizes the role of democratic institutions in enhancing the prospects of
collaboration (Bueno de Mesquita and Lalman, 1994, Morgan and Campbell, 1991) while social
constructivism tradition focuses on the role of ideas or norms held by democracies (Risse-Kappen,
1995a, Maoz and Russett, 1993b). Regardless of the initial mechanism of alliance formation, a
collective identity is likely to emerge when democracies interact in an institutionalized setting of an
alliance (Risse-Kappen, 1995a). Therefore, considering the deeper nature of democratic alliances,
we can expect that democratic allies should generate stronger positive expectations than purely
transactional allies or allies with shared military experience.
12
Finally, besides the commonality of regime, allies can be connected by an even stronger tie
{ that of shared culture. Culture similarity between countries is important as it might in
uence
expectations about possible norms of exchange (Herrmann and Fischerkeller, 1995, Jackson, 1993).
Strong cultural ties between allies might enhance the perception of trustworthiness and the per-
ceived likelihood that commitment will be honored. There is some indication that special character
of the relationship between the U.S. and its European allies can be explained through shared West-
ern world views and \individualistic and secular scientic premises" (Goldstein and Keohane, 1993,
8).
Therefore, based on shared characteristics between countries, we can identify a continuum that
ranges from non-allies (most negative expectations, strong outgroup) to strong allies (most positive
expectations, strong ingroup), with some intermediary cases in-between (Table 2.1).
1
According to
attribution theory, this perceptive dierence in expectations should have a strong and robust eect
on interpretation of countries' behavior. Negative expectations about outroup countries should
compel people to explain negative behavior with dispositional factors, while positive expectations
about ingroup countries should propel people toward situational explanations.
Table 2.1: Categorization of International Actors: Shared Characteristics
Categories Similarities Attributions
Non-ally Weak Dispositional
Transactional ally
Transactional + Shared experience
Transactional + Shared experience + Democracy
Transactional + Shared experience + Democracy + Culture Strong Situational
Following this logic, my main hypothesis is:
H1. Category Attribution Hypothesis: The more shared characteristics between countries, the more
likely the respondents are to use situational attributions for another country's negative behavior.
1
The specic characteristics were selected out of a wide range of possible similarities between countries. The goal
of this chapter is not to show which exact feature shifts expectations; rather, to show that attribution patterns change
incrementally as we move from non-allies to weak allies and, then, strong allies with shared characteristics.
13
2.4 Individual dierences in attributions
While I argue that shared characteristics aect our expectations and inferences about countries'
intentions on average, I also suggest that people's attributional patterns will vary due to dier-
ences in individual characteristics. Existing research on attribution theory in IR treats audiences
as homogeneous: people are strongly aected by group-serving attribution biases regardless of
individual-level characteristics. Here I argue that while attribution biases aect people on average,
we need to relax the concept of a unitary audience and explore individual characteristics that aect
people's inferences.
Out of the plethora of individual characteristics that might aect attributional patterns, here
I focus on political ideology as it plays a paramount role in dening people's intergroup attitudes.
Conservatives and liberals are starkly dierent in their attitudes towards group membership. Con-
servatism is attuned to negative outcomes (\dangerous world" beliefs (Duckitt et al., 2002)) and,
thus, tends to emphasize society's protection and security from external threats and dangers. Such
protection orientation seems to lead to a particular interest in (and unease about) ingroup-outgroup
membership (Jano-Bulman, 2009). Being highly attuned to threats, conservatives are especially
interested in who can be trusted, thus, constructing rigid ingroup{outgroup (i.e. us versus them)
boundaries (Brewer et al., 2004). As a consequence, conservatives put a special moral emphasis on
ingroup loyalty and provide strong, and even sometimes, unconditional support for the one's group
(Haidt and Graham, 2007a). For example, it has been shown that conservatives particularly value
patriotism, and specically, blind patriotism and nationalism which entail uncritical support for
one's national ingroup (Kosterman and Feshbach, 1989).
At the same time, due to their emphasis on external threats and negative outcomes, conser-
vatives take a \tough" approach on outsiders. People with strong conservative leanings are more
prone to intolerance, prejudice, and hostility toward a wide variety of outgroups (Sidanius and
Pratto, 2001b, Duriez and Van Hiel, 2002, Jost et al., 2003, 2009b). In sum, conservatives tend to
treat the members of their ingroup and outgroups very dierently: you are with us or against us.
Thus, we would expect people with strong conservative leanings to make dierent attributions for
the behavior of non-allies and strong allies: they are more likely to use dispositional explanations for
negative behavior of non-allies and situational explanations for negative behavior of strong allies.
14
In contrast with conservatism, liberalism is focused on positive outcomes and gains rather
than danger and losses, thus, ingroup{outgroup boundaries are not the focus of the liberals' at-
tention (Jano-Bulman, 2009). Liberalism is strongly associated with a moral foundation of fair-
ness/reciprocity that stresses obligations to treat everyone equally, fairly, and justly (Graham et al.,
2009). Liberals tend to put a special emphasis on the common humanity of ingroup and outgroup
members, thus often deemphasizing the intergroup dierences. Therefore, we can expect people
with liberal leanings to display less divergence in attributions for the behavior of non-allies and
strong allies.
Following this line of logic:
H2. Ideology hypothesis : People high on conservatism should be more likely to use dispositional
explanations for negative behavior of non-allies and situational explanations for negative behavior
of strong allies. Conversely, people high on liberalism should be less likely to make dierent attri-
butions for negative behavior of non-allies and allies.
2.5 Attributions and Policy Preferences
All the theorizing so far has focused on the process of causal attribution and factors that might
aect inferences that people make about social circumstances. However, when applying attribution
theory to the study of IR, we are not only interested in the process of causal attribution, but also
in the ways attribution patterns might aect policy preferences. Here I argue that by aecting
inferences about other countries' intentions, attributional patterns might consequently in
uence
policy preferences.
Existing research indicates that dispositional explanation of negative events is associated with
more severe response (Sadler et al., 2005, Pronin et al., 2006). For example, Sadler et al. (2005)
showed that participants who placed the blame for attacks on terrorists rather than circumstances
were more supportive of military interventions and increased domestic surveillance and less sup-
portive of the policy of increased humanitarian aid. However, the link between attributions and
policy preferences is yet to be shown in relation to other countries' behavior. While there are
15
indications that prior expectations about an actor might lead to certain policy preferences (Holsti,
1967, Shimko, 1991), existing research does not trace the exact mechanism underlying this linkage.
Previous authors focused on threat and aect as explanatory paths between images and policy
preferences (Herrmann et al., 1997, Herrmann, 1986, Holsti, 1967), ignoring the role of causal attri-
butions. At the same time, we know that certain emotions, such as anger, are closely connected to
attributions of responsibility. Two paths have been theorized and tested: dispositional attributions
lead to increase in anger levels that consequently leads to punitive preferences or, reversely, elevated
levels of anger might lead to more dispositional attributions that lead to more punitive preferences
(Lerner et al., 1998, Averill, 1983). Thus, whatever the exact mechanism, we should expect that
attributions serve as an important intermediary link between the identity of the country and policy
preferences:
H3: Policy preference hypothesis: Respondents explaining negative behavior of countries in
dispositional terms will be more likely to advocate harsher response than respondents explaining
behavior in situational terms.
2.6 Method
I explore these hypotheses with data gathered in the spring of 2017 on Amazon Mechanical Turk
online marketplace (MTurk). A total of 1352 American adults were recruited for a compensation
of $0.65.
2
To be included in the sample, participants had to complete the entire survey that took
an average of 5.8 minutes of their time.
In recent years, MTurk has become a popular resource for conducting survey experiments across
the social and behavioural sciences. MTruk is a very cost eective way to recruit large samples
necessary for examining heterogeneous treatment eects as I aim to do in this study. Despite
the obvious advantages of MTurk, we do have concerns about data quality. First, MTurk subject
are professional workers who have an incentive to complete a task as quickly as possible to receive
payment. As a result, subjects might be simply clicking through questions without paying sucient
2
In order to participate, participants had to be at least 18 years of age, must be located in the U.S., have approval
rating of greater than or equal to 95%, and number of approved tasks greater than or equal to 100.
16
attention. In order to address this concern, I followed the recommendations of Berinsky et al.
(2012). The survey included an attention check question, along with a prompt to \study the
situation carefully" and a warning that \certain factual questions about the situation" will be
asked. Overall, 92.5% of the sample answered attention check question correctly, suggesting that
the majority of the participants were paying sucient attention to the survey. I also show that my
experimental results are robust to the exclusion of less attentive respondents (see Appendix A.5).
Another concern with using MTurk sample has to do with external validity. Previous studies
indicated that MTurk subject pool tends to be more representative and diverse than most conve-
nience samples used in political science (Berinsky et al., 2012, Buhrmester et al., 2011). In addition,
prior work has been able to replicate established ndings with MTurk respondents (Berinsky et al.,
2012). However, MTurk samples do tend to dier from general population in some important ways
demographically.
3
Looking at basic demographics presented in Appendix A.6, my sample is sig-
nicantly younger and more educated than the general population (as compared to the 2010 U.S.
census). To make sure that these dierences in demographics do not aect generalizibility of my
ndings, I employ entropy balancing to reweight the data to known population parameters (Hain-
mueller and Xu, 2013). The analysis provided in Appendix A.6 compares weighted and unweighted
samples and shows that the substantive results do not change when weights are introduced. Thus,
the ndings seem to be robust to the demographic composition of the sample.
After the battery of standard demographics, respondents were presented with a ctional vignette
about an ambiguous international event. The vignette describes a situation when a civilian air plane
was shot down by a ctitious foreign country (country A).
4
I selected this scenario for two primary
reasons. First, there have been instances of such tragic incidents in relatively recent history (such
as, shooting down of Korean Air Lines Flight 007 by the Soviet Union in 1983 or shooting down
the Iran Air Flight 655 by the United States in 1988). Therefore, the vignette achieves a certain
level of realism important for external validity reasons. Second, such cases are usually highly
ambiguous and open to varying plausible interpretations (Entman, 2004), which might lead to
dierent attributional patterns.
3
Participants recruited through MTurk tend to score higher in education and attention to study tasks, and score
lower in age, conservatism, employment, and religiosity (Berinsky et al., 2012, Paolacci and Chandler, 2014).
4
Fictitious country was selected so as to avoid possible associations with a real world country, which might confound
the results. Moreover, using a real-world referent might create a heterogeneous eect if the amount of knowledge
about the country used is correlated with specic predictor variables at the individual level.
17
The description of the ctitious country A was manipulated experimentally to correspond to
the continuum of categories that I outline in this chapter, ranging from a nonally to a strong ally
with shared characteristics, with intermediary cases in-between. This was the only manipulation.
Together with the control group that did not receive any description of country A
5
, the experiment
consisted of six groups. Full information about vingette and treatment groups is provided in
Appendix A.1.
Having read the scenario, the participants were asked to select the most plausible cause for
the incident and indicate how condent they were about their inference.
6
The rst choice, techno-
logical malfunction, represents the situational explanation. The second option, intentional attack,
represents the dispositional explanation.
2.7 Results
I present my results in three phases. First, I look at treatment eects to see whether dierent
categories of countries produce distinct attributional patterns. Second, I use political ideology
to show how dierent types of people reacted to similar stimuli. Finally, I discuss the eect of
attributions on policy preferences expressed by the respondents.
2.7.1 Group Categorization
The category attribution hypothesis posits that dierent categories of countries would elicit dierent
inferences among the general public. In particular, I hypothesize that shared characteristics between
countries would aect attributions, such that the countries with stronger similarities would generate
more situational attributions for identical negative behavior. Figure 2.1 shows the percentages of
respondents who attributed the outcome to dispositional factors for every experimental group with
95% condence intervals. The gure provides visual support of my rst hypothesis: as we can
5
This treatment simply read: \A civilian airplane was shot down by country A" without any description of the
country.
6
One issue with such design could arise if certain vignettes made people think about specic real-world countries
or events, while others did not, thus violating information equivalence condition (Dafoe et al., 2017). If that were
the case, the causal attributions might be driven by the image of that specic country or the details of that specic
event. In order to diagnose this issue, subjects were asked whether they had any specic country in mind when they
read the vignette. A 2-sample test for equality of proportions with continuity correction showed no imbalance on
this attribute across experimental conditions, indicating that images of specic countries or real-world events did not
confound the treatment.
18
see, the number of people inferring intentional attack as opposed to a technological malfunction
decreases dramatically as we move from a country that is not an ally of the United States to a
country that is a close ally with a common enemy, shared experience of ghting together, democratic
regime, and Western culture.
Despite the very limited information provided in the vignette, most of the respondents made
causal attributions with high degree of condence: only 12% of the sample answered that they were
\Somewhat uncondent" or \Strongly uncondent" about their inference. This result indicates that
ingroup-ougroup categories serve as powerful cognitive shortcuts allowing people to comprehend
ambiguous situations with a high degree of perceived certainty.
Figure 2.1: Dispositional Attribution by Treatment Group
Note: Figure displays the percentage of respondents who attributed the outcome to dispositional factor for
every experimental group with 95% condence intervals. It shows that dispositional attributions decrease
incrementally when we move from non-allies with the most negative expectations to strong allies with the
most positive expectations. Ally 1 { transactional ally; Ally 2 { ally with shared experience; Ally 3 {
democratic ally; Ally 4{ ally with shared culture
19
To evaluate the treatment eects more rigorously, Table 2.2 presents the dierence between
the proportions of respondents who attributed the outcome to dispositional factors. It shows that
approximately 61.06% of the respondents in a control group believe that the incident happened due
to dispositional reasons (intentional attack). Interestingly, the dierence between the respondents
in a control group, who did not receive any cues about the country involved, and the respondents
in a nonally group, who received an indication that the country is not an ally of the United States,
is small and not statistically signicant. This seems to indicate that, in absence of any additional
information, respondents have negative expectations about foreign countries and treat them as an
outgroup unless otherwise indicated.
Table 2.2: Eect of Country Type on Dispositional Attributions
Control 61.06 61.06 61.06 61.06 61.06
Nonally 62.72
Ally1 53.13
Ally2 43.17
Ally3 37.17
Strong Ally 34.08
Dierence 0.94 -7.93 -17.89*** -23.89*** -26.98***
95% C.I. -7.72, 11.03 -17.49, 1.62 -27.38, - 8.39 -33.29, -14.5 -36.32, -17.64
Notes: The table gives the percentage of respondents who attributed the outcome to disposi-
tional factor. 95% condence intervals for the dierences were obtained from a 2-sample test
for equality of proportions with continuity corrections.
When presented with information that a country A is a transactional ally of the United States, a
smaller share of the respondents attribute the outcome to the disposition of the country. However,
as hypothesized above, transactional alliances do not produce strong positive expectations due
to the weak ties between the countries. Respondents are practically divided in their attributions
(53.13% make dispositional inferences), and the dierence with the control group is not statistically
signicant. As expected, weak transactional allies do not generate strong positive expectations and
are not considered as an ingroup by most of the respondents.
The real change in attributional patterns happens when countries are described as having strong
ties to the United States. Adding information about shared experience of ghting together decreases
the proportion of people inferring intentional attack down to 43.17%. This result conrms my
theoretical expectations that an ally, that previously honored its military commitment, is perceived
20
by the public as more reliable and trustworthy. Adding information about democratic regime
decreases the proportion of people using dispositional attributions to an even lower level of 37.17%.
This result is also consistent with existing claims that alliances formed on the basis of common
democratic norms are perceived as particularly strong. Finally, we can see that adding a cultural
dimension to the description of the country decreases the percentage of people inferring intentional
attack to 34.08% of respondents. Thus, the pattern of attributions is practically reversed when we
compare control group with the strongest type of ally.
Overall, the results presented in Table 2.2 provide support for my main hypothesis that foreign
countries are not all simply perceived as an outgroup. Rather, shared characteristics between coun-
tries aect ingroup-outgroup categorization. In addition, the results conrm my conceptualization
of ingroup-outgroup categorization as a continuum: attributions change incrementally when we
move from non-allies with the most negative expectations to strong allies with the most positive
expectations.
2.7.2 The Heterogeneous Pattern of Attributions
So far, I have demonstrated that people make dierent causal attributions depending on shared
characteristics between countries. However, as we can see from Table 2.2, the substantial numbers
of respondents make dispositional attributions even when another country is a very strong ally
of the United States. Rather than assume therefore that country identity alone is sucient to
explain variation in attributional patterns, we need to explore how attributions vary as a function
of individual characteristics. Specically, I hypothesized above that such individual-level dierences
as ideological leanings have a strong eect on people's inferences.
In order to test this hypothesis, I estimate a series of logistic regression models with the at-
tributed cause as a dependent variable (0 for situational attribution, 1 for dispositional attribution).
The results of this analysis are presented in Table 2.3.
7
The coecients on dummy variables in
Model 1 present additional support that strong allies produce situational attributions (control group
is a reference category), controlling for such basic demographics as age, race, gender, education,
party identication, and ideology.
8
The coecients for Ally 2, Ally 3, and Strong Ally are negative
7
Logistic regression was used for all models in Table 2.3 due to the binary nature of the dependent variable. All
models passed the multicollinearity test with no severe correlation between variables detected.
8
Age is coded through four categories. Race is dichotomized between white (1) and non-white respondents (0).
21
and highly signicant (p<.001), indicating that the propensity of inferring intentional attack in
these groups was signicantly lower than in the control group.
Table 2.3: Eect of Ideology on Attributions
Dispositional Attribution
Model 1 Model 2
(1) (2)
Nonally 0.091 (0.196) 0.268 (0.325)
Ally 1 0.307 (0.193) 0.203 (0.318)
Ally 2 0.691

(0.194) 0.526

(0.315)
Ally 3 0.969

(0.197) 0.465 (0.332)
Strong ally 1.125

(0.199) 0.602

(0.328)
Age:25-44 0.036 (0.169) 0.018 (0.164)
Age:45-64 0.530

(0.208) 0.542

(0.203)
Age:65+ 0.273 (0.362) 0.319 (0.361)
White 0.183 (0.135)
Male 0.034 (0.118)
High school 0.068 (0.863)
Some college 0.246 (0.842)
College degree 0.145 (0.840)
Graduate school 0.293 (0.846)
Party ID 0.440 (0.321)
Ideology 0.757

(0.327) 0.855

(0.490)
Interactions
Nonally x Ideology 0.837 (0.703)
Ally1 x Ideology 0.316 (0.695)
Ally2 x Ideology 0.504 (0.674)
Ally3 x Ideology 1.278

(0.682)
Strong ally x Ideology 1.373

(0.696)
Constant 0.751 (0.860) 0.225 (0.259)
N 1,354 1,354

p < .1;

p < .05;

p < .01
Notes: Table displays logit coecients; standard errors in parenthesis.
Dependent variable is coded 1 for dispositional attribution and 0 for situ-
ational attribution. The reference categories are country identity, control
group, 18- 24 years old, non-white, female, less than high school. All non-
dichotomous measures have been rescaled from 0 to 1. All analyses are
unweighted, for weighted analyses see Appendix A.6.
Besides the average eect of treatment, Model 1 also shows that ideology has a signicant im-
Gender is dichotomized between males (1) and females (0). Education is coded through six categories with higher
categories corresponding to high levels of education. Party ID variable ranges from \Strongly Democrat" to \Strongly
Republican" with higher values corresponding to more Republican leanings. Ideology ranges from \Extremely Liberal"
to \Extremely Conservative" with higher values corresponding to more Conservative leanings.
22
pact on the respondents' attributional patterns. However, the relationship between ideology and
attribution patterns was theorized to be more complex than a simple positive association between
conservatism and dispositional attributions. On the one hand, due to a strong propensity for
outgroup discrimination, people high on conservatism should be more likely to use dispositional ex-
planations for the behavior of non-allies. However, on the other hand, conservatism is also typically
correlated with strong ingroup favouritism, which should lead conservatives to use situational ex-
planations for the behavior of strong allies. As a result, ideological aliation should have dierent
eects on the attributions of blame depending on the type of country in question.
Model 2 of Table 2.3 tests this hypothesis through interaction terms that compare the eect
of ideology for each experimental group.
9
Negative coecients for interaction terms indicate that
respondents who score high on conservatism are much less likely to make dispositional attributions
about the behavior of a strong ally than a country that is not an ally of the United States. This
interaction eect is further visualized in Figure 2.2.
10
As we can see, the gure provides strong evi-
dence for the interaction eect hypothesized above. It shows that ideology aects attributions very
dierently depending on the country in question. When making attributions about the behavior of
a non-ally, people who identify as conservatives are much more likely to explain the outcome with
dispositional factors. Moving from the most liberal to the most conservative on the ideology scale
increases the probability of making dispositional attributions from 49.4% to 84.1%. However, the
pattern is reversed for the behavior of a strong ally. In this case, the more conservative the people
are, the less likely they are to use dispositional attributions. Moving from liberal to conservative on
the ideological scale decreases the probability of inferring intentional attack from 44.5% to 34.4%.
People who self-identify as strong liberals do not seem to be signicantly aected by the identity
of the country involved in the incident, making situational attributions for both strong allies and
explicit non-allies. While seemingly surprising, this nding is consistent with extensive literature
on political ideology that posits that liberals exhibit greater egalitarianism and are signicantly less
likely to hold prejudicial attitudes at either conscious or unconscious levels (Jost, 2006, Kerlinger,
1984). Strong conservatives show the opposite pattern, consistent with a large body of research
9
Interaction model here includes only signicant variables from Model 1. See Appendix A.3 for the saturated
interaction model.
10
For the sake of space and ease of presentation,the gure depicts results only for a non-ally and the strongest ally.
For full results, see Appendix A.4.
23
Figure 2.2: Conditional Eect of Ideology on Attributions
Note: Figure displays interaction eects of ideology and treatment group on probability of making a dispo-
sitional attribution for a non-ally and the strongest ally. Estimates are based on Model 2 from Table 2.3
with 95% condence intervals. Ideology ranges from \Extremely Liberal" to \Extremely Conservative" with
higher values corresponding to more Conservative leanings. Control variables are set at median values.
that shows that conservative orientations are generally associated with higher ingroup favouritism
and higher hostility toward a wide variety of outgroups (Altemeyer, 1998, Duckitt et al., 2002,
Sidanius and Pratto, 2001b).
11
2.7.3 Attributions and Policy Preferences
The results presented above demonstrate that country identity and individual characteristics do
indeed aect peoples' attribution about blame. However, do these attributions translate into sup-
port for certain policies? In order to answer this question, I asked the participants which steps the
United States should take to address the situation in the vignette. Participants were given a choice
11
As a robustness check, I also test whether eect of ideology holds when controlling for more specic attitudes
toward international aairs: militant internationalism (MI), cooperative internationalism (CI), and isolationism (Wit-
tkopf, 1990, Kertzer et al., 2014). Appendix A.7 shows that the eect of ideology is robust but smaller when controlling
for MI due to the slight post-treatment bias: considering that ideology underlies foreign policy orientation, eect of
ideology is partially mediated by MI (Kertzer et al., 2014). See Appendix A.8 for details on nonparametric mediation
analysis (Imai et al., 2011).
24
of four options, two of which correspond to \soft" or passive response (\Do nothing" and \Demand
immediate investigation of the incident"), and two to \harsh" or active response (\Impose economic
sanctions" and \Use military force"). Respondents could choose as many options as they saw t. I
constructed a dependent variable with two levels, with those selecting an exclusively soft response
receiving a score of 0 and people selecting a hard response receiving a score of 1.
I present the results of the analysis in Table 2.4, which I use to generate a predicted probability
plot in Figure 2.3. The model includes a dummy variable for the attributed cause along with
a list of control variables that are also expected to aect policy preferences.
12
Besides ideology
and the usual demographic variables, the model controls for foreign policy orientations { militant
internationalism (MI), cooperative internationalism (CI), and isolationism { that have been shown
as strong predictors of policy preferences (Chittick et al., 1995, Rei
er et al., 2011).
13
Following
Kertzer et al. (2014), scores for CI, MI, and isolationism were constructed through principal-axis
factoring with varimax rotation.
14
As expected, militant internationalism that measures hawkishness is a strong predictor of the
support for a harsh response, such as the imposition of economic sanctions or the use of military
force. However, the coecient for attribution is also positive and highly signicant, indicating that
people who explain the crash by reference to dispositional and intentional factors are much more
likely to support a hard line response. Attributions play an important role even controlling for
foreign policy orientations.
Figure 2.3 illustrates the marginal eect of attributions on policy preference, while keeping
all other explanatory variables constant. Changing attributions from dispositional to situational
decreases the probability of supporting a harsh policy response from 25.52% to 4.93%. In other
words, when people make dispositional attributions of blame they are roughly 20 points more likely
to endorse harsh policies than people who make situational attributions.
12
Logistic regression was used due to the binary nature of the dependent variable. All variables are standardized
to range from 0 to 1.
13
Following Kertzer et al. (2014), the foreign policy instrumentation consisted of 20 questions, listed in Appendix
A.2.
14
Factor scoring provides researchers with less noisy estimates than additive scores as it allows the extraction of
the sole dimension of interest (Kertzer et al., 2014). Principal-axis factoring identied one latent dimension for each
of three foreign policy orientation scales and produced a single score, that was later normalized to range from 0 to 1.
25
Table 2.4: Eect of Attributions on Policy Prefer-
ences
Harsh Response
Dispositional Attribution 1.889

(0.179)
Age:25-44 0.139
(0.218)
Age:45-64 0.365
(0.289)
Age:65+ 0.287
(0.496)
White 0.130
(0.173)
Male 0.465

(0.164)
High school 1.162
(1.011)
Some college 1.470
(0.984)
College degree 1.165
(0.979)
Graduate school 1.044
(0.987)
Party ID 0.277
(0.430)
Ideology 0.250
(0.459)
MI 1.937

(0.501)
CI 0.510
(0.459)
Isolationism 0.266
(0.419)
Constant 2.313

(1.120)
N 1,354

p < .1;

p < .05;

p < .01
Notes: Table displays logit coecients; standard
errors in parenthesis. Dependent variable is coded
1 for hard response and 0 for soft response. The ref-
erence categories are situational attribution, 18-24
years old, non-white, female, less than high school.
All non-dichotomous measures have been rescaled
from 0 to 1. The model is unweighted, for weighted
analyses see Appendix A.6.
26
Figure 2.3: Eect of Dispositional Attributions on Support for Harsh Response
Note: Figure displays eect of attributions on probability of supporting harsh response. Estimates are based
on Table 2.4 with 95% condence intervals. Control variables are set at mean and median values.
2.8 Discussion and Conclusion
The results of this study help to explain how people make inferences and attribute blame in am-
biguous international events. Using an experimental design, I contest the existing assumption that
all other international actors beyond one's home state are considered as an outgroup with respec-
tive attributional patterns. Rather, I demonstrate that ingroup-outgroup categorization should be
understood as a continuum: attributions change incrementally when we move from non-allies with
the most negative expectations to strong allies with the most positive expectations.
The nding that shared characteristics aect attributional patterns helps to account for incon-
sistent ndings in previous research on attributions. For example, Heradstveit and Bonham (1996)
found variations in the way Egyptian elites explained behavior of their allies during the Gulf War.
Despite being members of the same Coalition against Iraq, Egyptian political elites explained the
U.S. aggressive behavior through the mixture of dispositional and situational factors. If we adopt a
dichotomous understanding of countries as either allies or non-allies, such a nding would be incon-
27
sistent with predictions of the theory. However, this nding is fully consistent with the spectrum
of categories presented in this chapter. According to my theory, Egyptian elites perceived the U.S.
as a purely transactional ally \dictated more by circumstance than deliberate choice" (Heradstveit
and Bonham, 1996, 283). Transactional allies generate weak positive expectations, thus, resulting
in mixed situational and dispositional attributions for negative behavior.
Similarly, my framework provides alternative explanation for the nding by Mercer (1996) that
allied countries considered each other as an outgroup. The nding that England treated Russia
as an outgroup during 1909 Bosnian Crisis could be explained by purely instrumental nature of
pre-WWII alliances. The alliance was formed in response to a common threat { the Germans and
Austrians, with no shared experience, democratic norms or cultural ties to strengthen the alliance.
Therefore, such transactional alliance was not likely to produce strong positive expectations and
consequent attributional patterns as Mercer expects. The lack of communal feelings among allies
in the beginning of the 20th century does not prove that all allies are bound to be considered as an
out-group. Rather, stronger nature of ties that emerged after the World War II based on shared
experience, common democratic norms, and cultural similarity might lead to their inclusion as the
in-group with corresponding attributional patterns.
While my argument that categorization into ingroup and outgroup is based on the perceived
strength of ties should be applicable to any country in question, the particular characteristics of
weak and strong allies would vary depending on the country's characteristics. Here, I suggest a
pattern of ingroup-outgroup categorization applicable to a subset of democratic Western countries.
When adopting a theory to other countries, some other shared characteristics might need to be
selected. For example, when applying the theory to non-Western countries, it will be possible to
emphasize other cultural elements, such as religion (for example, Islam for Muslim countries) or
philosophical tradition (such as Confucianism for East Asian countries). Further research is needed
to determine whether my argument would hold for non-democratic and non-Western countries, or
whether other countries/regions simply do not have allies strong enough to be considered as part
of an ingroup.
Besides ingroup-outgroup categorization process, the results of my study question the exist-
ing assumption of unitary audiences prevalent in attributional research in IR. I show that people's
attributional patterns vary dramatically due to dierences in their individual characteristics. Specif-
28
ically, I show that people with conservative ideological leanings make very dierent attributions
about behavior of ingroups and outgroups. Further research is needed to fully test this nding. For
example, further research might look into actual cases of ambiguous behaviour of foreign countries
to see whether strong liberals and strong conservatives make drastically dierent attributions about
the causes of the event.
Finally, my results demonstrate why attributions of blame are important for international re-
lations research. My ndings provide strong evidence that attributions translate into support of
certain policies: when people make dispositional attributions they are 20 points more likely to
endorse harsh policies than people who make situational attributions.
Overall, the ndings of this chapter help rectify the study of attributions in international rela-
tions and demonstrate the vast unrealized potential that this research agenda might hold for our
understanding of foreign policy attitudes. Considering the evidence this study provides about the
importance of attributions, it opens broad avenues for future research on other factors that aect
attributions of blame in international relations and political science as a whole.
29
Chapter 3
External Threat and Moral
Judgement
3.1 Introduction
How do people form moral judgement about other individuals or groups? What information do
they need to evaluate other groups as \good" or \bad" and assign specic moral traits, such as
honesty, trustworthiness, loyalty, and others? Perception of a target as moral or immoral has
been shown to have clear behavioral consequences, including support for aggressive and belligerent
retaliatory policies (Opotow, 1990, Staub, 1990). Most severe consequences of threat moralization
might include political oppression, human right violations, and go as far as slavery or genocide
(Staub, 1989, 1990). Despite that, we still know very little about how people form negative moral
judgement and what the role of intergroup threat is in this process.
In this study, I argue that that perception of intergroup threat might lead to the formation of
moral judgment about the source of threat, even when no information about morality of the group is
available. When the level of integroup threat is high, people tend to rely on cognitive heuristics and
express negative evaluations of the threatening outgroup (Stephan et al., 1999). However, despite
expansive literature on threat and its in
uence on public opinion (Herrmann et al., 1999, Arian,
1989, Friedland and Merari, 1985, Huddy et al., 2005), the possible causal relationship between
integroup threat and moral evaluations remains largely unexplored. Here, I argue that people have
30
the propensity to see source of threat as less moral even if they have no other reasons to make
negative moral evaluations about it. Besides, I also argue that there is an individual variation in
the people's propensity to moralize threats. Specically, people with conservative leanings tend to
deny morality to the threatening groups more strongly than liberals.
A strong link between external threat and moral judgement is supported by two studies. In the
rst study, I use observational data to trace how people talk about threatening groups { adversarial
foreign countries.
1
Using a large dataset of political tweets related to 2016 presidential elections,
I show that people tend to pass some sort of a moral judgement when speaking about external
threats online. Besides, I also show that people with conservative leanings tend to moralize threats
more than the liberals. They tend to moralize adversarial countries more overall and, specically,
they tend to use more negative moral rhetoric when talking about these countries. Finally, in this
study I also demonstrate that conservatives moralize threats dierently than liberals. Using Moral
Foundations Theory framework (Graham et al., 2009, 2011), I show that liberals tend to emphasize
violations of individualizing moral foundations while conservatives are more focused on binding
foundations.
In the second study, I conduct an experiment to test the causal nature of the relationship
between intergroup threat and moral judgement. The results show that simply priming individuals
with intergroup threat increases moral judgement of the source of threat even when people have
no prior knowledge of or beliefs about it. Besides aecting the overall negative evaluations of a
ctitious country, threat manipulation also aects perceptions of the country's leaders along key
moral dimensions, such as aggressiveness, trustworthiness, and greed. Finally, the experiment
yields limited support to my hypothesis that political ideology might aect people's propensity to
moralize threats.
3.2 Threat and moral judgment
The perception of intergroup threat has been consistently linked with support for aggressive and
belligerent retaliatory policies against threatening groups. For example, Americans supported
overseas military action in direct proportion to the threat posed by a foreign aggressor to U.S.
1
In this project, I will focus on a specic type of an intergroup threat { a threat from foreign countries/nationals
{ that has been less explored than threats from domestic outgroups (racial, ethnic, sexual groups).
31
interests (Herrmann et al., 1999, Jentleson, 1992, Jentleson and Britton, 1998). Arian (1989) found
a direct link between the perceived likelihood of war and a preference for an increase in military
power over peace negotiations among Israelis in the 1980s. Terrorist threat is also associated with
support for aggressive military action among Israelis (Friedland and Merari, 1985). Finally, the
tendency to become more politically intolerant under conditions of external threat is also well
documented (Gibson, 1992, Marcus et al., 2005).
We also know that support for hostile behaviors towards an outgroup can result from perceiving
the outgroup as essentially and morally inferior to one's own group. Opotow (1990) argues that
morally excluded individuals or groups are perceived as \outside the boundary in which moral val-
ues, rules, and considerations of fairness apply"(p.1). Thus, harming or exploiting such individuals
or groups becomes \appropriate, acceptable, or just." Most severe consequences of moral exclusion
might include political oppression, human right violations, and go as far as slavery or genocide
(Staub, 1989, 1990).
While the extant research shows that intergroup threats and moral judgement can lead to
support of aggressive actions and policies, the relationship between these two factors is far from
clear. One possibility is that the threat and moral judgement are separate (albeit related) processes
that independently contribute to the support of aggressive policies (Maoz and McCauley, 2008).
Maoz and McCauley (2008) focus on dehumanization as a particular form of perceiving the outgroup
as morally inferior. They show that threat and dehumanization are distinguishable processes that
have an independent contribution to explaining Israeli Jewish support for aggressive retaliatory
policies toward Palestinians. While the model with threat and dehumanization as two separate
latent factors performs much better than the model that merges them into one construct, this work
does not address the possibility that these factors might cause one another.
Allowing for the causal relationship between threats and moral judgement presents two impor-
tant possibilities. First, it is plausible that the perception that the members of an outside group as
immoral or morally inferior leads to stronger perception of this group as threatening. Some empir-
ical research supports this relationship. Morality information is highly relevant in establishing the
intentions of others { it denes whether someone represents as opportunity or a threat (Brambilla
et al., 2011). In line with this argument, Brambilla et al. (2012) nds that groups perceived to be
not moral trigger high levels of perceived threat. In their consequent work, they also nd that per-
32
ceived morality aects the levels of perceived threats from both ingroup and outgroup (Brambilla
et al., 2013).
However, the opposite causal chain is also possible when the perception of the outside group as
threatening leads to negative moral connotations about the threatening group. This causal chain
is much less explored. Is it possible that threatening groups could be perceived as immoral even
when people have no prior knowledge or preconceptions about the members of these groups? Does
moral judgement of threatening groups serve as an automatic psychological mechanism that opens
the possibility for harming these groups without much remorse?
Expansive literature on threat notes that when the level of intergroup threat is high, people
typically rely on cognitive heuristics (stereotypes) and express negative emotions and evaluations
(Stephan et al., 1999). For instance, dierent types of threats have been identied as consistent
predictors of racial bias and anti-immigration attitudes (Kinder and Sears, 1981, McLaren, 2003).
The connection between threats and moral judgement is much less explored; however, several studies
suggest a causal relationship between perceived threat and dehumanization/infrahumanization
2
(Haslam and Loughnan, 2014). Esses et al. (2013) shows that media portrayals of immigrant or
refugees as threatening are likely to lead to dehumanization of these groups. Priming existential
threat of mortality might also increase infrahumanization (Goldenberg et al., 2009, Vaes et al.,
2010).
More specically, McAlister et al. (2006) show that 9/11 terrorist attack, which posed a grave
national threat, raised the level of moral disengagement among the American respondents. Moral
disengagement, among other things, includes immoralization of the adversary and acts as a psy-
chological mechanism of relieving oneself of moral predicaments related to violent military action.
Using the SEM technique, the researchers show that moral disengagement completely mediated the
eect of the terrorist attack on the support of the military campaign against terrorists.
As in the case with 9/11 attack and immigration, it seems that intergroup threat often oc-
curs before people have any morality-relevant information about the outside group. Under threat,
people tend to rely on cognitive heuristics and make negative evaluations of the outgroup without
considering all information available. I argue that these negative evaluations might concern the
2
Dehumanization is dened as perceiving a person or group as lacking humanness; Infrahumanization is dened
as perceiving an out-group as lacking uniquely human attributes relative to an in-group. Overall, infrahumanization
can be understood as a subtle variant of dehumanization. (Haslam and Loughnan, 2014).
33
moral dimensions specically, therefore:
H1: Moralization of threat: When people perceive intergroup threat, they tend to make negative
moral judgements about the source of threat.
3.3 Ideology and moralization of threats
Although the propensity to see source of threats as less moral may be universal, there might be
individual dierences in our tendency to do so. Political ideology is a major dispositional quality
that might aect people's propensity to moralize threats. Political ideology could be dened as
a set of beliefs about the proper order of society, that is commonly represented as a continuum
with conservatives on the right and liberals on the left (Jost et al., 2009a). People identifying with
conservative and right-wing ideology are generally higher in authoritarianism, social dominance,
and system justication that make them more prone to intolerance, prejudice, and hostility toward
a wide variety of out-groups (Duriez and Van Hiel, 2002, Jost et al., 2003, 2009a, Sidanius and
Pratto, 2001a).
In particular, strong correlation between conservatism and SDO might explain why conservatives
tend to deny morality to the threatening groups more strongly than liberals. Social Dominance
Orientation, that identies a preference for unequal relationships among groups of people, has been
shown as the strongest predictor of dehumanization of immigrants (Hodson and Costello, 2007),
refugees (Esses et al., 2008), and enemy war victims (Jackson and Gaertner, 2010). While right-
wing authoritarianism is also linked with dehumanization, the association is substantially weaker
implying that dehumanization and demoralization of outgroups rests on striving for intergroup
dominance rather than social conformity (Haslam and Loughnan, 2014).
While no existing research explores dierent patterns of moralization of threats specically,
we have some evidence that conservative-leaning individuals tend to demoralize ethnic outgroups
to justify their inferior economic or social status. For example, DeLuca-McLean and Castano
(2009) nd that conservative American participants infrahumanized Hispanic hurricane victims
while liberal respondents did not. Conservatives attributed less uniquely human emotions to the
hurricane victims when their names sounded Hispanic than when it sounded Caucasian. Similarly,
34
Maoz and McCauley (2008) show that Israeli Jews who hold more right-wing ideologies in the
Israeli-Arab con
ict are more likely to dehumanize Palestinians more than left-wing doves.
Besides, we can expect that conservatives would tend to moralize threats more than the lib-
erals as conservatives are more sensitive to threats overall (Jost et al., 2003, Onraet et al., 2013).
According to the negativity bias perspective, conservatives are especially responsive to negative
stimuli and events (Hibbing et al., 2014). This relationship between political orientations and
negative stimuli is corroborated by longitudinal (Bonanno and Jost, 2006), physiological (Oxley
et al., 2008) and neurological (Amodio et al., 2007) evidence. Similarly, motivated social cognition
perspective posits that conservatives should be more responsive to threat as they are especially
motivated to gain epistemic closure in the face of such threats (Nam et al., 2013) and are more
likely to perceive threats in the environment (\dangerous world" beliefs (Duckitt et al., 2002, Jost
et al., 2003)). Thus, according to both perspectives, we expect that conservatives typically dis-
play stronger physiological stress reactions to novel and threatening stimuli relevant to intergroup
relations. Moralization of the source of threat could emerge as a defensive reaction among conser-
vatives, but could be much less pronounced among less threat-prone liberals.
H2. Ideology hypothesis: People high on conservatism should be more likely to moralize external
threats than people with liberal ideological leaning.
While it is possible that conservatives tend to moralize threatening groups more than liberals,
it is also possible that conservatives tend to moralize threats dierently than liberals. Recent
work in social psychology suggests that liberals and conservatives in the United States reply on
distinct moral structures and intuitions (Graham et al., 2009, Haidt and Graham, 2007b); thus,
it is plausible that they would moralize threatening groups in a distinct manner as well. Here, I
am turning to the Moral Foundations Theory (MFT) framework that identies ve psychological
foundations of morality (Graham et al., 2009). This framework is particularly useful for thinking
about moralization of foreign threats because it has been found to shape both people's political
attitudes on both domestic and foreign policy issues (Federico et al., 2013, Kertzer et al., 2014).
Moral Foundations Theory identies ve distinct moral foundations: harm/care, fairness/reciprocity,
authority/respect, purity/sanctity, and ingroup/loyalty. First two are the \individualizing foun-
35
dations" of harm/care and fairness/reciprocity. The harm/care foundation encompasses concern
for the suering of others, while the fairness/reciprocity foundation is focused on equal treatment.
Overall, individualizing foundations are based on how well or poorly individuals treated other
individuals" (Graham et al., 2011, 366). The other three are called \binding foundations". Au-
thority/respect is focused on the maintenance of social hierarchies and assuring of social order;
purity/sanctity encompasses concerns about spiritual or bodily cleanliness, and ingroup/loyalty
stresses obligations to one's group and its protection against outgroups.
While both individualizing and binding foundations are equally \moral", dierent people and
cultures dier on the degree to which they rely on these foundations. Political liberals focus mostly
on harm/care and fairness/ reciprocity, while political conservatives value virtues based on all ve
foundations (Haidt and Graham, 2007b). As the authors posit, \A consequence of this thesis is that
justice and related virtues (based on the fairness foundation) make up half of the moral world for
liberals, while justice-related concerns make up only one fth of the moral world for conservatives."
(Haidt and Graham, 2007b, 2).
As a consequence of dierent weights given to dierent moral dimensions, liberals might per-
ceive that threatening groups violate the moral virtues they mostly care about { namely, harm/care
and fairness/reciprocity. They tend to see that threatening groups are harmful, cause suering to
others and treat others unfairly or act in an unjust way. In contrast, conservatives might perceive
that threatening groups violate both individualizing and binding moral norms. They would stress
harm and unfairness to a lesser extent, but emphasize the adversarial nature of the threatening
groups, perceive them as damaging the cohesion of the ingroup, and being overall disgusting and
impure. Following this line of logic:
H3. Moral foundations hypothesis. People high on liberalism should perceive that threatening
groups violate individualizing moral norms. Conversely, people high on conservatism should per-
ceive that threatening groups violate both individualizing and binding moral norms.
This project consists of two parts. In the rst study, I use observational data to trace how people
talk about threatening groups { adversarial foreign countries. Implicit moral judgments could be
detected observationally by looking at the lexical decisions made by the people (Ellemers et al.,
36
2019). In the second study, I conduct an experiment to test the causal nature of the relationship
between intergroup threat and moral judgement.
3.4 Study 1: Moral Rhetoric about Adversarial Countries on Twit-
ter
3.4.1 Data and Method
In order to investigate how people talk about adversarial countries, I use a large dataset of political
tweets related to 2016 presidential elections. The dataset was collected through Twitter Search
API from September 15 until November 9, 2016 and contains 43.7 million unique tweets posted
by nearly 5.7 million distinct users. It includes all tweets mentioning hashtags and keywords that
relate to the 2016 U.S. Presidential election, such as terms related to Democratic and Republican
nominees, third-party candidates, and presidential debates in general.
A special feature of the dataset is that it includes labels that classify Twitter users based on
the political leaning of the media outlets they share (Badawy et al., 2018). Users were labelled as
either liberal or conservative based on the number of tweets they produced with links to liberal
or conservative sources.
3
In other words, if a user had more tweets with links to liberal sources,
he/she would be labeled as liberal and vice versa. Then, label propagation algorithm was used to
assign labels to all remaining users (who did not share liberal or conservative media outlets) based
on the political orientation of their closest neighbors in the network.
4
The results of the algorithm
were validated through stratied 5-fold cross validation to the set of 29K seeds. Algorithm was
trained on four-fths of the seed list and tested on the remaining one-fth. The average precision
and recall scores are both over 91%. For further details on the political ideology labels, see Badawy
et al. (2018).
Considering the primary goal of this project, the next step in the analysis was to lter the
original dataset of political tweets to only include posts about adversarial countries. To determine
3
Sources are selected based on the lists of partisan media outlets compiled by third-party organizations, such as
AllSides3 and Media Bias/Fact Check (Badawy et al., 2018).
4
In a network-based label propagation algorithm, each node was assigned a label, which is updated iteratively
based on the labels of the node's network neighbors. In label propagation, a node takes the most frequent label of
its neighbors as its own new label. (Badawy et al., 2018).
37
a list of countries that Americans consider adversarial or threatening, I use data from 2017 YouGov
poll.
5
The poll was conducted from January 28 to February 1, 2017 and covered 7,150 adults from
YouGov's opt-in internet panel. The respondents were asked: \Do you consider the countries listed
below to be a friend or an enemy of the United States?" Respondents could answer \Ally of U.S.,"
\Friendly," \Unfriendly," \Enemy of the U.S." or \Not Sure" for each country listed. The list of 15
most adversarial countries was almost identical for the respondents across the aisle and included
the following countries:
Democrats: North Korea, Russia, Iran, Syria, Libya, Afghanistan, Iraq, Somalia, Yemen,
Lebanon, Pakistan, Sudan, Palestinian Territories, South Sudan, Tajikistan
Republicans: North Korea, Iran, Syria, Libya, Iraq, Afghanistan, Somalia, Palestinian Terri-
tories, Sudan, Pakistan, Yemen, South Sudan, Lebanon, Cuba, Tajikistan
Having determined the list of adversarial countries, the next step was to collect all possible
keywords related to these countries that people might use. I used CountryInfo TABARI dictionary
to gather all possible names for each country and nationality in English including some common
misspellings for hard words.
6
For every country, the last name of the head of state (and/or head of
government) was also included. For example, the keywords for Iran included Iran, Iranian, Persia,
Persian, Rouhani, Ruhani, Khamenei, Kamenei.
The resulting list of keywords related to adversarial countries was used to lter the original
Twitter dataset to contain only a subset of tweets that mentioned the above-specied country
keywords. Qualitative exploration of the resulting tweets revealed that tweets containing the name
or a leader of an adversarial country usually talk about the security threats coming from that
country. However, two countries emerged as problematic for the purposes of this analysis. First, the
vast majority of posts about Russia were concerned with the implications of Russia's interference
in the US internal aairs rather than foreign policy issues. Second, the vast majority of posts
about Cuba were focused on policy preferences of the American Cubans rather than foreign threats
5
More details on the results and methodology of the poll here and here.
6
Further information on TABARI dictionary is available here.
38
Table 3.1: Twitter Data Descriptive Statistics
Statistic Original Dataset Final Dataset
# of Tweets 43,705,293 298,058
# of Original Tweets 12,513,640 67,818
# of Distinct Users 5,746,997 126,871
# of Liberal Users >3.4M 49,673
# of Conservative Users >1M 72,987
posed by Cuba. Thus, I made a decision to exclude tweets mentioning these two countries from
the dataset. Excluding Russia and Cuba also made the list of adversarial countries identical for
Republicans and Democrats, thus, making a comparison of rhetoric by ideology more robust.
Table 3.1 reports aggregate statistics of the original dataset and the ltered dataset that includes
posts related to adversarial countries only. First, as we can see, the ltered dataset is much smaller
than the original dataset it terms of the number of tweets. Only 0.7% of all collected political
tweets contain keywords related to external threats. Similarly, in terms of the overall number of
distinct users, only 2.2% of all users discussing political topics mention adversarial countries in their
posts. What is particularly interesting though is that the distribution of liberal and conservative
users is starkly dierent in the original dataset and the dataset of external threats. The original
dataset contains 77% liberal users and 23% conservative users while the ltered dataset contains a
much higher number of conservative users than their liberal counterparts (40.5% liberals vs. 59.5%
conservatives). This dierence in distribution is statistically signicant (X
2
= 92290, p < 0.01)
and indicates that conservatives are disproportionately engaged in topics related to adversarial
countries.
7
This nding aligns well with previous research showing that conservatives are more
sensitive to threats in the environment (Jost et al., 2003, Onraet et al., 2013) and are generally
attuned to negative outcomes (Hibbing et al., 2014).
3.4.2 Results
Before moving to our main hypotheses, let us rst explore which countries are being discussed
most and least frequently by liberals and conservatives. Figure 3.1 shows the percentage of posts
mentioning specic countries for users with dierent ideological leaning. Unsurprisingly, due to
7
The results are based on Chi-square test of goodness of t that takes the original distribution of users in general
political conversation as the expected proportions.
39
constant security threats related to the Syrian Civil War and ongoing refugee crisis, the majority of
posts for both liberals and conservatives contain keywords related to Syria (44% of liberals' posts
and 37% of conservatives' posts). Iraq and Iran also emerge as countries that concern both liberal
and conservative Twitter users. Keywords related to Iraq were found in 24% of all liberals' tweets
and in 16.6% of all conservatives' tweets, while keywords related to Iran were mentioned in 16.8%
liberal tweets and 20% conservative tweets. Overall, the distribution of tweets among liberals and
conservatives seems to follow our expectations with Syria, Iraq and Iran being the most discussed
countries.
The rst hypothesis posits that outgroup threat and morality are closely linked concepts. Specif-
ically, I hypothesize that when people perceive outgroup threats, they tend to moralize the source of
the threat. In order to detect moral rhetoric in Twitter posts, I rely on the word counting method
using the Moral Foundations Dictionary (MFD) (Garten et al., 2016). Moral Foundations Dictio-
nary was created to capture the ve foundations laid out in Moral Foundations Theory (MFT)
(Graham et al., 2009) and has been used in most of the research on text-based moral rhetoric. For
instance, this dictionary approach has been used to compare the content of sermons delivered in
liberal and conservative churches (Graham et al., 2009). MFD dictionary contains of moral terms
corresponding to ve general content domains: care/harm, fairness/cheating, loyalty/betrayal, au-
thority/subversion, and purity/degradation. Each domain is divided into positive and negative
aspects, yielding a total of 10 domains with respective moral terms that can be captured using the
MFD (Garten et al., 2016). Here, I am utilizing the second version of the dictionary (MFD 2.0)
that captures each moral foundation more fully and is currently recommended for usage (Frimer
et al., 2019).
8
First, I am using Moral Foundations Dictionary to obtain the count of words with some moral
connotation. Figure 3.2 displays the percentage of posts about adversarial countries with moral
versus non-moral content. As we can see, the majority of posts contain no moral language (74.8%
of all tweets); however, a quarter of the posts does contain some moral rhetoric (25.2%). Thus, it
8
The dictionary was slightly modied to exclude words that do not have inherent moral content when discussing
foreign policy threats, such as kill, die, attack, or ght. I also exclude words that create tautologies with the
discussed topic, such as threaten, danger, and harm. The full list of excluded words is the following: protect, heal,
feed, hospitalize, alleviate, relieve, safe, harm, hurt, danger, in
ict, damage, kill, die, destroy, threat, injure, attack,
wound, assault, ght, attack, fatality, violence, war, punch. Grammatical in
ections of these words were excluded as
well. When running the analysis against the entire dictionary, the key ndings remain unchanged.
40
Figure 3.1: Percentage of Posts about Adversarial Countries by Ideology
appears that people often tend to pass some sort of a moral judgement when speaking about external
threats; however, that moral judgement might not necessarily be negative. Considering that all
moral terms in MFD dictionary are divided into positive or negative aspects (moral virtue vs.
moral vice), I can establish what share of the moral content about adversarial countries is actually
negative. Figure 3.2 shows that out of the posts with moral rhetoric, 12.5% of all posts contain
negative moral language and 12.6% of posts contain positive moral language. Thus, investigation
41
of tweets about adversarial countries presents mixed evidence in support of Hypothesis 1. While
I found that a substantial share of posts about adversarial countries contain some moral rhetoric,
only 12.5% tweets contain negative moral language and, thus, pass negative moral judgement about
the source of external threat.
Figure 3.2: Percentage of Posts about Adversarial Countries by Moral Content
While data provides limited support to my hypothesis about the human propensity to see source
of threats as less moral, this tendency might still be true for people with certain characteristics.
Specically, in Hypothesis 2 I suggest that people with conservative leaning would tend to moralize
threats more than the liberals. Table 3.2 shows 15 top moral word stems used by liberals and
conservatives.
9
Top 15 words were chosen as they account for an overwhelming majority of all
9
Stemming was used to reduce in
ected words to their base form. For example, word stem `car*' accounts for an
array of related words with similar meaning, such as caring, caringly, cared, uncaring etc.)
42
moral words used by both liberals and conservatives.
10
The word stems colored in red are classied
as negative moral words according to the Moral Foundations Dictionary. As we can see, conserva-
tives' rhetoric is dominated by negative moral content (9 out of 15 top word stems are negative).
In contrast, the liberals' rhetoric is more mixed (6 out of 15 top word stems are negative). Conser-
vatives also use more negative moral words in terms of their relative frequency: 62% of top moral
words count for conservatives are negative words versus 31.7% by liberals. Thus, a simple count
and visual investigation of the top words indicate that conservatives produce more negative moral
content than liberals when talking about threats.
Table 3.2: Top 15 Stemmed Moral Words from Liberal and Conservative Users
Liberals Count Conservatives Count
rebel 1,119 rebel 4,380
leader 793 christian 3,906
group 502 brutal 2,816
care 499 overthrow 2,493
murder 411 rape 2,410
rape 397 pray 2,011
love 381 love 1,479
father 372 murder 1,291
douchebag 308 group 1,274
govern 299 corrupt 1,112
mercy 276 hell 991
christian 245 order 647
victim 215 violent 619
fair 187 trust 617
racist 186 treason 617
In order to provide more robust test of the hypothesis, Figure 3.3 shows distribution of moral
content by political ideology. First, we can see that conservatives indeed produce more posts with
negative moral content than liberals: 13.6% of all conservatives' posts contain some negative moral
language versus 10% among the liberals. In terms of positive moral language the situation is
reversed as conservatives produce less posts with positive moral content than liberals: 12.1% of
all of all conservatives' posts contain some positive moral language versus 14% among the liberals.
Finally, the right side of the gure shows that conservatives produce slightly fewer posts without
10
Count of the top 15 moral words makes up 85.7% of the entire moral words usage by liberals and 83.4% for
conservatives.
43
any moral language than their liberal counterparts. Overall, Figure 3.3 indicates that conservatives,
indeed, tend to moralize external threats more than the liberals. They tend to moralize adversarial
countries more overall and, specically, they tend to use more negative moral rhetoric when talking
about these countries.
Figure 3.3: Percentage of Posts by Moral Content and Ideology
While the data shows that conservatives tend to moralize adversarial countries more than lib-
erals, it is also possible that conservatives moralize threats dierently than liberals. Using Moral
Foundations Theory framework, I hypothesized that people high on liberalism would perceive that
44
threatening groups violate individualizing moral norms, such as harm/care and fairness/reciprocity.
Conversely, people high on conservatism should perceive that threatening groups violate both in-
dividualizing and binding moral norms. They would stress harm and unfairness to a lesser extent,
but emphasize the adversarial nature of the threatening groups, perceive them as damaging the
cohesion of the ingroup, and being overall disgusting and impure.
Figure 3.4 shows the breakdown of negative moral content by type for conservatives and liber-
als.
11
Let us rst look at the individualizing foundations, namely, harm/care and fairness/reciprocity.
While we would expect that conservatives would stress harm and fairness to a lesser extent than
liberals, the results show that it is true only for fairness. As we can see, 32.5% of all negative moral
posts by liberals are concerned with unfairness and injustice compared to 20% by conservatives.
This dierence is quite large and follows our expectation than liberals, on average, tend to care more
about violations of the principles of fairness. Surprisingly, the same does not hold for the dimension
of harm. Contrary to our expectations, larger number of conservatives' posts stress harming nature
or behavior of adversarial countries: 27% of all negative moral posts by conservatives contain some
harm-related keywords compared to 23.6% for conservatives.
Moving to the binding foundations, the gure shows that conservatives emphasize two out
of three binding foundations more frequently than liberals. First, we can see that conservatives
stress ingroup/loyalty concerns more frequently when talking about adversarial countries: 20% of
all conservatives' posts contain keywords related to this moral foundation compared to 15.4% for
liberals. The dierence in the authority concerns is even more stark. As we can see, 35.8% of
all conservatives' tweets focus on authority concerns compared to mere 23.3% by liberals. The
only morality type that presents a pattern opposite to our theoretical expectations is sanctity
that encompasses concerns about spiritual or physical cleanliness. Contrary to our expectations,
liberals stress sanctity dimensions more frequently than conservatives: it was detected in 24.2% of
all liberals' tweets versus 18% of conservative tweets.
11
Categories are not mutually exclusive, i.e. one tweet can contain words related to dierent moral dimensions.
For that reason, percentages do not sum up to 100%.
45
Figure 3.4: Breakdown of Negative Moral Content by Type
3.5 Study 2: An Experimental Test of Relationship between Threat
and Moral Judgement
3.5.1 Data and Method
The Twitter study provides some evidence that Americans tend to make negative moral judgments
about threatening foreign countries and that this trend is especially prevalent among people with
conservative ideological leanings. However, using observational data alone, I cannot make credible
claims about causality of the relationship and the underlining mechanism. First, while I argue that
perceived threats lead to moral judgement, there is a strong alternative possibility that the causal
46
arrow is reversed. Perhaps, countries (or leaders of these countries) that are perceived to be immoral
are subsequently perceived as more threatening. Second, while I show that conservatives moralize
threatening countries more, it is possible that they do so because they have much more negative
aect toward these countries in the rst place (Mullen and Skitka, 2006, Skitka and Wisneski, 2011).
A randomized experiment that keeps the identity of the adversary constant and neutral allows us
to distinguish between these possible mechanisms.
To address these theoretical and methodological limitations of the rst study, I conducted an
original survey experiment with a sample of 1,006 American residents recruited from Amazon's
Mechanical Turk platform (55% female, 77% white).
12
Despite obvious advantages of MTurk and
general replicability of the ndings (Berinsky et al., 2012, Buhrmester et al., 2011), MTurk samples
do tend to dier from general population in some important ways demographically.
13
To make
sure that these dierences in demographics do not aect generalizibility of my ndings, I employ
entropy balancing to generate post-stratication weights based on gender, age, education and race
(Hainmueller, 2012). Appendix B.2 presents further details on the weighting process and compares
weighted and unweighted demographics.
After the battery of standard demographics, the between-subjects experiment randomly as-
signed respondents to one of eight vignettes about a ctitious country that is \dramatically in-
creasing the size of its military, including a nuclear weapons program." The experiment included
three manipulations. The rst and most important manipulation was a manipulation of threat to
the United States. Subjects were randomly assigned to read that a ctitious country has \good re-
lations with the United States" or \poor relations with the United States." Second, the experiment
also included a manipulation of motives: some subjects read that the nuclear weapons could be
used oensively since its main adversary does not have nuclear weapons. Others read that nuclear
weapons cannot be used oensively since its main adversary also has nuclear weapons. Finally, I
also manipulated whether the country had previously engaged in con
icts with that rival in the
past and had taken part of its rival's territory. While in this study I focus on the eect of threat
on moral judgement, providing explicit information about the motives and past con
ict history is
12
The survey was elded in June 2018. In order to participate, participants had to be at least 18 years of age, must
be located in the U.S., have approval rating of greater than or equal to 95%, and number of approved tasks greater
than or equal to 100.
13
Participants recruited through MTurk tend to score higher in education and attention to study tasks, and score
lower in age, conservatism, employment, and religiosity (Berinsky et al., 2012, Paolacci and Chandler, 2014).
47
important as these inferences might largely guide judgments about the morality of the aggressor
(Reeder et al., 2002). Overall, the experiment had a fully crossed 2 x 2 x 2 factorial design with
eight conditions. The full text of the vignettes is presented below:
Imagine a country with a major regional rival is dramatically increasing the size of its military,
including a nuclear weapons program. However, it is thought that its nuclear weapons [cannot
be used oensively since its main adversary also has nuclear weapons/could be used
oensively since its main adversary does not have nuclear weapons]. This country [has
been engaged in con
icts with that rival in the past and has taken part of its rival's
territory/has never fought its rival in the past]. The country has had [poor relations/good
relations] with the United States.
Having read the vignette, the participants were asked a battery of questions about the ctitious
country and its leaders to evaluate the degree of evoked moral judgement. As a measure of moral
judgement, I follow a standard approach based on self-reported traits or behaviors that indicate
the general character/personality type of another person/group (Ellemers et al., 2019) and adapt
it to the IR domain. Thus, instead of asking participants to evaluate general morality of another
person/group as suggested by Reeder et al. (2002), I asked questions about the overall perception
of the country as \good" in terms of its human rights record and level of democracy. These general
questions were followed by more specic items related to the moral characteristics of the country's
leaders. To check the eectiveness of threat manipulation, I also asked the respondents: \How
threatening do you think this country is with 0 being completely harmless and 10 being extremely
dangerous?" As expected, a country with good relations with the United States was rated as
signicantly less dangerous than the country with poor relations (t(1004) = -3.65, p < 0.01, d =
0.48). Full information about the vignette and moral judgement questions is provided in Appendix
B.1
3.5.2 Results
The randomized experiment was designed specically to test causal mechanisms suggested in Hy-
potheses 1 and 2. The moralization of threat hypothesis (H1) posits that threatening groups could
48
be perceived as immoral even when people have no prior knowledge or preconceptions about the
group and its members. First, I use the experimental data to measure the eect of threat treatment
on the overall perception of the country described in the vignette as \good" in terms of its state
of democracy and human rights record. To measure perception of democracy I asked: \Do you
expect this country to be undemocratic or democratic with 0 being completely undemocratic and
10 being completely democratic?" with a respective 10-item scale ranging from \Undemocratic" to
\Democratic." Similarly, to measure perception of human rights record I asked: \What would you
expect the human rights record of this country to be with 0 being very poor and 10 being very
good?" with a respective 10-item scale ranging from \Poor human rights record" to \Good human
rights record."
Figure 3.5 displays estimates of the mean and 95% condence intervals for evaluations of
democracy and human rights with and without threat treatment. Multivariate analysis of vari-
ance (MANOVA) was used to account for correlated multiple dependent variables (Huberty and
Olejnik, 2006) and revealed signicant main eect of threat (F(2, 1003) = 76.44, p < 0.01). At
the univariate level, the main eect of threat was signicant for both the human rights scores and
democracy scores (p< 0.01). Thus, participants rated the country as having a worse human rights
record under threat condition (M = 3.05, SD = 2) than in no threat condition (M = 4.6, SD = 2.1).
In a similar vein, the country was rated as less democratic in a threat condition (M = 3.06, SD =
2.26) than in a no threat condition (4.7, SD = 2.39). Thus, the average treatment eects provide
strong support for the causal link between threat and moral judgement: manipulating threat can
increase moral judgement of the source of threat even when people have no prior knowledge of or
believes about it.
Besides aecting the overall perceptions of a country, I expect that threat also aects perceptions
of the country's leaders along various moral dimensions. I asked the respondents to give their
estimates of the country's leaders on the following characteristics: trustworthiness, aggressiveness,
greed, and resoluteness.
14
Each characteristic was measured through 10-item scale ranging from 0
as \Not at all" to 10 as \Very".
Figure 3.6 displays estimates of the mean and 95% condence intervals for evaluations of the
14
The exact question wording was \Based on what you know about this country, please give your estimate of
whether its leaders have the following characteristics, with larger numbers indicating higher levels of that trait?"
49
Figure 3.5: Eect of Threat on Moral Perceptions of Country
Note: Figure displays means and 95% condence intervals for democracy and human rights under threat
and no threat treatment. The scores range from 0 to 10 for both variables.
leaders' aggressiveness, greed, resoluteness, and trustworthiness. MANOVA was used again to
account for correlated multiple dependent variables. The analysis revealed a signicant multivariate
eect of threat ( F(4, 1001), p < 0.01). At the univariate level, the main eect of threat was
signicant for all scores except for resoluteness. As shown in the gure, respondents rated the
country's leaders as more aggressive, greedier, and less trustworthy in a threat condition than in
a no threat condition. Thus, the moral judgement of the country's leaders also indicates a strong
causal eect between the people's perception of intergroup threat and negative moral judgement
about the source of that threat.
While the survey data provides strong evidence for the propensity to see source of threats
as less moral on average, I also hypothesized that there might be individual dierences in our
tendency to do so (H2). Specically, I suggested that political ideology might aect people's
propensity to moralize threats in the way that conservatives would tend to deny morality to the
threatening groups more strongly than liberals. In order to test this hypothesis, I estimated a series
of OLS regressions with moral evaluations of country's leaders as dependent variables. The results
50
Figure 3.6: Eect of Threat on Moral Perceptions of Country's Leaders
Note: Figure displays means and 95% condence intervals for aggressiveness, greed, resoluteness, and trust-
worthiness of the country's leaders. The scores range from 0 to 10 for both variables.
for trustworthiness and aggressiveness are presented in Tables 3.3 and the results for greed and
resoluteness ae presented in Table 3.4.
Model 1 of Table 3.3 includes dummies for threat and two other treatments (oensive moti-
vation and presence of past con
ict between countries) along with the basic demographics such
as age, ideology, party identication, education, race, and gender.
15
First, the coecient on the
15
Age is coded through four categories. Race is dichotomized between white (1) and non-white respondents (0).
Gender is dichotomized between males (1) and females (0). Education was coded through six categories with higher
categories corresponding to high levels of education. Party ID variable ranges from \Strongly Democrat" to \Strongly
Republican" with higher values corresponding to more Republican leanings. Ideology ranges from \Extremely Liberal"
to \Extremely Conservative" with higher values corresponding to more Conservative leanings.
51
threat dummy presents additional support that manipulating perception of threat produces neg-
ative moral judgement. The coecients for two other treatments are also negative and highly
signicant indicating that motive and past con
ict history aect judgements about the morality of
the aggressor as well (Reeder et al., 2002). However, the coecient for threat is much larger that
the coecients for two other treatments indicating that threat, indeed, plays the most crucial role
in the formulation of moral judgement.
Ideology does not have a signicant impact of the evaluation of the leaders' trustworthiness in
isolation, as can be seen from Model 1; however, it does seem to have a strong interactive eect
with the threat treatment (Model 2). Consistent with my hypothesis, the negative coecient of
the interaction term indicates that under threat treatment, the more conservative the respondents,
the lower they evaluate the country's leaders in terms of their trustworthiness. This nding is
consistent with a large body of research that showed that conservatives are especially responsive
to negative stimuli and events (Hibbing et al., 2014, Jost et al., 2003, Onraet et al., 2013) and are
more likely to perceive threats in the environment in the rst place (Duckitt et al., 2002).
Models 3 and 4 of Table 3.3 present analogous analysis for evaluation of the leaders' aggressive-
ness. Again, the coecient of the threat dummy is statistically signicant (p < 0.01) indicating
that threat manipulation moved the respondents toward evaluating the country's leaders as more
aggressive. Interestingly, the coecient for the oensive motive treatment is not signicant in this
case while the coecient for past con
ict treatment is signicant and substantively large. Similarly
to trustworthiness analysis, ideology does not have a signicant impact on the evaluation of the
leaders' aggressiveness; however, it has a strong interactive eect with the threat treatment (Model
2). Consistent with my hypothesis, the positive coecient of the interaction term indicates that
under threat treatment, the more conservative the respondents, the more aggressive their consider
the country's leaders (p < 0.05).
The analogous analysis for evaluation of greed and resoluteness is presented in Table 3.4. While
threat manipulation aects the negative perception of the country's leaders as more greedy (Model
1), it does not have a signicant eect on the evaluation of the leaders' resoluteness (Model 3).
Besides, threat treatment has no interactive eect with ideology as shown in Models 2 and 4.
Overall, as I can see, the randomized experiment about a ctitious country yields mixed re-
sults. While threat is shown to have a strong eect on the majority of moral dimensions (except
52
Table 3.3: Eect of Ideology on Perceptions of Leaders' as Trustworthy and Aggressive
Trustworthy Aggressive
(1) (2) (3) (4)
Threat 1.753

(0.140) 0.931

(0.311) 1.076

(0.140) 0.447 (0.313)
Oensive 0.074 (0.140) 0.111 (0.140) 0.232 (0.141) 0.261 (0.141)
Past con
ict 0.366

(0.139) 0.378

(0.139) 0.784

(0.140) 0.793

(0.140)
Age:25-45 0.819

(0.229) 0.807

(0.228) 0.098 (0.230) 0.088 (0.230)
Age:45-65 0.615

(0.244) 0.582

(0.244) 0.349 (0.245) 0.375 (0.245)
Age:65+ 0.971

(0.302) 0.922

(0.302) 0.818

(0.304) 0.855

(0.303)
Ideology 0.126 (0.071) 0.246

(0.081) 0.009 (0.071) 0.101 (0.082)
Party ID 0.107 (0.070) 0.128 (0.070) 0.062 (0.070) 0.046 (0.070)
Some College 0.180 (0.266) 0.188 (0.265) 0.141 (0.267) 0.135 (0.267)
2-year College 0.261 (0.298) 0.243 (0.297) 0.549 (0.299) 0.563 (0.299)
4-year College 0.098 (0.253) 0.078 (0.252) 0.294 (0.254) 0.309 (0.254)
Graduate Degree 0.120 (0.269) 0.113 (0.268) 0.373 (0.270) 0.379 (0.269)
White 0.293 (0.163) 0.293 (0.162) 0.294 (0.164) 0.294 (0.163)
Male 0.154 (0.145) 0.190 (0.145) 0.045 (0.145) 0.072 (0.146)
Threat x Ideology 0.235

(0.079) 0.180

(0.080)
Constant 3.822

(0.346) 3.503

(0.361) 5.610

(0.347) 5.854

(0.363)
N 1,006 1,006 1,006 1,006
R
2
0.160 0.167 0.102 0.106
Adjusted R
2
0.148 0.154 0.089 0.093

p < .05;

p < .01
Notes: Table displays OLS coecients; standard errors in parenthesis. Dependent variable ranges from
0 as least trustworthy/aggressive to 1 as most trustworthy/aggressive. The reference categories are 18-24
years old, high school or less, non-white, and female. Results weighted to match population parameters
on gender, education, age, and race. See the appendix for details.
for resoluteness in our design) as suggested in Hypothesis 1, it is not always moderated by the
person's ideology as proposed in Hypothesis 2. The possible reason for this nding is that dierent
moral judgments are formed through distinct mechanisms as suggested in Hypothesis 3. Trustwor-
thiness and aggressiveness correspond to the ve moral foundations more closely than greed and
resoluteness; thus, I see a more consistent eect of the threatening stimuli on them. Unfortunately,
the design of this experiment does not allow for robust testing of Hypothesis 3 that would require
inclusion of moral dimensions corresponding to all ve Moral Foundations; however, this limitation
should be addressed in future research.
53
Table 3.4: Eect of Ideology on Perceptions of Leaders' as Greedy and Resolute
Greedy Resolute
(1) (2) (3) (4)
Threat 1.106

(0.149) 0.739

(0.333) 0.177 (0.136) 0.248 (0.305)
Oensive 0.353

(0.150) 0.369

(0.150) 0.170 (0.137) 0.173 (0.138)
Past con
ict 0.523

(0.149) 0.528

(0.149) 0.137 (0.136) 0.136 (0.136)
Age:25-45 0.393 (0.245) 0.387 (0.245) 0.731

(0.224) 0.733

(0.224)
Age:45-65 0.567

(0.186) 0.576

(0.186) 0.884

(0.170) 0.886

(0.170)
Age:65+ 0.757

(0.260) 0.773

(0.261) 1.133

(0.238) 1.136

(0.238)
Ideology 0.243

(0.075) 0.297

(0.087) 0.130 (0.069) 0.141 (0.080)
Party ID 0.217

(0.075) 0.226

(0.075) 0.038 (0.068) 0.040 (0.068)
Some College 0.133 (0.284) 0.129 (0.284) 0.590

(0.260) 0.589

(0.260)
2-year College 0.429 (0.318) 0.437 (0.318) 0.850

(0.291) 0.852

(0.291)
4-year College 0.332 (0.270) 0.341 (0.270) 0.533

(0.247) 0.535

(0.247)
Graduate Degree 0.403 (0.287) 0.406 (0.287) 0.506 (0.262) 0.507 (0.263)
White 0.247 (0.174) 0.247 (0.174) 0.519

(0.159) 0.519

(0.159)
Male 0.368

(0.155) 0.352

(0.155) 0.066 (0.141) 0.063 (0.142)
Threat x Ideology 0.105 (0.085) 0.020 (0.078)
Constant 4.645

(0.330) 4.793

(0.351) 5.318

(0.301) 5.346

(0.321)
N 1,006 1,006 1,006 1,006
R
2
0.095 0.096 0.080 0.080
Adjusted R
2
0.082 0.082 0.067 0.066

p < .05;

p < .01
Notes: Table displays OLS coecients; standard errors in parenthesis. Dependent variables range from 0 as
least greedy/resolute to 1 as most greedy/resolute. The reference categories are 18-24 years old, high school
or less, non-white, and female. Results weighted to match population parameters on gender, education, age,
and race. See the appendix for details.
3.6 Conclusion
The results of this project help to explain how people form moral judgement about other indi-
viduals or groups. Across two studies, I nd support for my theory that perception of intergroup
threat might lead to the formation of moral judgement about the source of threat, even when no
information about morality of the group is available. Using large observational dataset of political
tweets, I show that conversation about adversarial foreign countries tends to contain strong moral
rhetoric. I also use a randomized experiment to test the causal nature of this relationship and nd
that priming individuals with intergroup threat increases moral judgement of the source of threat
even when no information about morality of that source is available. In addition to aecting the
overall negative evaluations of a ctitious country, threat manipulation aects perceptions of the
54
country's leaders along key moral dimensions, such as aggressiveness, trustworthiness, and greed.
Besides providing evidence for the people's tendency to see source of threats as less moral,
this project also explores possible individual dierences in our propensity to do so. Specically,
it provides preliminary support that people with conservative leanings tend to moralize threats
more than the liberals. When discussing adversarial foreign countries on Twitter, conservatives
tend to use more negative moral rhetoric. Also, when being primed with external threat in a
survey experiment, the more conservative the respondents, the lower they evaluate the country's
leaders in terms of their trustworthiness and higher in terms of their aggressiveness. This eect
of political ideology was not signicant for other moral dimensions though, such as greed and
resoluteness, indicating that people with dierent political leanings might be sensitive to dierent
moral dimensions rather than all moral violations uniformly. This expectation is corroborated by
observational data but is not tested in the randomized experiment. Further research including
evaluations along all ve moral dimensions is needed to fully test this nding.
Another promising avenue for future research concerns possible behavior consequences that
follow from moral judgement caused by intergroup threat. Intergroup threat has been consistently
linked with support for aggressive policies against threatening groups (Huddy et al., 2005); however,
we do not know whether moral judgement serves as a necessary a mediator in that process. Or does
threat sometimes lead to support for aggressive action without intermediate moral reaction? And
would the behavior consequences be dierent in these two scenarios? My expectation is that the
perceived external threat with consequent moral judgement might lead to more severe behaviors
as morally inferior groups are not believed to deserve a fair treatment, but this remains to be seen
in future research.
Another important extension of this project would explore whether dierent types of intergoup
threat would aect moral judgement dierently. Here I conceptualize threat as a realistic threat to
a country's national security (thus, the use of the word \enemy"). While in the existing literature
in IR and political science, threat is overwhelmingly conceptualized as a realistic threat to the
country's safety, territory, or national resources (Herrmann et al., 1999, Huddy et al., 2005); research
in social psychology identies another important type of threat { symbolic threat { that involves
perceived group dierences in values, morals, and beliefs. In political psychology, it has been
identied as a consistent predictor of racial bias and anti-immigration attitudes (Kinder and Sears,
55
1981, McLaren, 2003). Thus, considering its emphasis on values and morals, it would be interesting
to see whether symbolic threat might produce even stronger moral reaction than realistic threat.
Another type of threat that might lead to moral judgement is group esteem threat that occurs
when the actions of an outgroup potentially decrease an ingroup's esteem (Tajfel and Turner, 1979,
Branscombe et al., 1999). Support for the relationship between group esteem threat and outgroup
hostility has been found across a number of studies (Branscombe and Wann, 1994, Branscombe
et al., 2002). Future research should explore whether dierent types of intergoup threat would aect
moral judgement dierently and whether dierent people are more sensitive to dierent types of
threat.
Finally, future research on moralization of threats should consider other individual characteris-
tics that might aect moral judgement. Rather than focusing on political ideology, we could look
into more basic psychological predispositions that inform ideological leanings in the rst place. Two
obvious examples of such predispositions are Social Dominance Orientation (SDO) and Right-Wing
Authoritarianism (RWA) that are generally correlated with conservative ideology but emphasize
moral values to a dierent extent (Cohrs et al., 2005, Duriez and Van Hiel, 2002, Jost et al., 2003,
Sidanius and Pratto, 2001b). Thus, people high on SDO value dominance, power, and achieve-
ments and express prejudice due to instrumental concerns, such as striving for power and supe-
riority (Cohrs et al., 2005, Pratto et al., 1994). Thus, we could expect that people high on SDO
tend to respond to threats in amoral terms. In contrast, people high on RWA are more concerned
with ingroup norms and traditions and their prejudice is typically motivated by identity concerns
(Cohrs et al., 2005); thus, we could expect that people high on RWA will be more sensitive to moral
violations. More research is needed to understand which foundation of conservatism underpins its
tendency to moralize external threats to a higher extent.
Overall, the ndings of this chapter help connect the research on intergroup threat with the
study of morality in international relations and demonstrate the vast unrealized potential that
this research agenda might hold for our understanding of foreign policy attitudes. It opens broad
avenues for future research on threat moralization in international relations and political science as
a whole.
56
Chapter 4
Networks of Power: Analyzing World
Leaders' Interactions on Social Media
4.1 Introduction
With the ever-growing in
uence of social media on our day-to-day lives, it is not surprising that
world leaders have been increasingly using social media platforms as a tool for political communi-
cation. To be specic, these days an overwhelming 90% of governments have active social media
accounts
1
. Unsurprisingly, such drastic proliferation of governmental accounts on social media pro-
pelled a new strand of academic research focused on leaders' digital communication and diplomacy
(Barber a and Zeitzo, 2017, Munger et al., 2018, Zeitzo, 2018). From this new research agenda
we know that political leaders use social media strategically to divert attention from domestic
problems, bolster regime legitimacy, and suppress opposition (Barber a et al., 2018, Gunitsky, 2015,
Pearce, 2015).
Despite these recent advances, the important limitation of the existing literature is that it only
investigates how political leaders communicate with the public. However, besides talking to the
public, leaders also interact with each other. For example, during the 2014 Crimean Crisis the
Russian Ministry of Foreign Aairs and the US Sate Department engaged in a Twitter hashtag
war.
2
Most prominently, since his election in November 2016, the US President Donald Trump
1
http://twiplomacy.com/blog/twiplomacy-study-2017/
2
https://www.washingtonpost.com/news/worldviews/wp/2014/04/25/russia-hijacks-u-s-state-departments-ukraine-hashtag
57
brought Twitter diplomacy to a new level by actively engaging with foreign leaders and announcing
key foreign policy decisions online.
3
With the number of leaders and governments represented
online growing steadily, the leaders seem to be embracing a new form of diplomacy that bypasses
the traditional oine diplomatic channels. However, possibly due to the data limitations, until
now there has been no systematic research exploring interactions between leaders on social media
platforms.
Thus, the primary goal of this chapter is to provide a rst look into the world leaders' interactions
on social media. How and how much do leaders interact with each other on social media? Which
leaders are most interconnected, and why? And which leaders play the central role in the global
social media network? To answer these questions, I am using a novel dataset of all social media
communication between leaders of 193 U.N. member countries and match it with key political,
geographical, and sociodemographic variables. I nd that leaders' interactions on social media
closely resemble their interactions in the oine world with leaders communicating mostly within
their geographic regions or along the similar level in political hierarchy. Besides, consistent with a
large body of literature in international relations, I nd that regime type plays an important role
in the way Twitter communities are formed. Finally, I explore the patterns of centrality within
the leaders' network and nd that democratic leaders tend to occupy more central positions in
the network, possibly due to the more transparent and multi-directional communicative tendencies
they adopt.
4.2 How do Leaders Interact on Social Media?
In this chapter, I examine how leaders strategically interact with each other on social media plat-
forms, such as Twitter. The strategic usage of mass media by political leaders has been long
explored by the scholars of political communication (Scheufele and Tewksbury, 2006, Walgrave and
Van Aelst, 2006). More recently, scholars started adopting the existing theories of political com-
munication to the internet domain and investigate how political leaders use social media. While
most studies explore leaders' online behavior within particular county context (Munger et al., 2018,
Zeitzo, 2018, Golbeck et al., 2010, Grant et al., 2010, Enli and Skogerb, 2013), a few others adopt
3
http://thehill.com/policy/defense/336659-trumps-diplomacy-by-twitter-sets-off-firestorm
58
a large-n approach (Barber a and Zeitzo, 2017, Bulovsky, 2018, Barber a et al., 2018). Existing
research indicates that political leaders adopt social media strategically in response to domestic
unrest and actively use it to divert attention from domestic problems (Barber a and Zeitzo, 2017).
Rather than simply censoring, leaders have learned to use online platforms to their advantage; for
example, Gunitsky (2015) describes four mechanisms that authoritarian regimes use to co-opt so-
cial media: counter-mobilization, discourse framing, preference divulgence, and elite coordination.
Similarly, Pearce (2015) shows how authoritarian regime in Azerbaijan uses social media to harass
opposition online.
Despite the growing research on the way leaders and governments use social media, the over-
whelming majority of studies have investigated how political leaders communicate with the public.
With the number of leaders and governments represented online growing every year, there has been
no systematic research exploring interactions between leaders on social media platforms.
Why is online communication between leaders important? First, it represents a new form of
diplomacy that bypasses traditional channels and unfolds openly in front of domestic and interna-
tional audiences. This new form of diplomacy { \twitplomacy" { remains largely unexplored and
under-theorized (Su and Xu, 2015, Strau et al., 2015). Second, online interactions between leaders
might shed light on important con
ictual or collaborative relationships that emerge between world
leaders oine. It might provide valuable insights into the leaders' respective agendas, strategies,
and bargaining processes. For example, Zeitzo (2018) shows that the leaders of Hamas and Israeli
IDF engaged in a true \Twitter war" that brought the con
ict to an unprecedented level of pub-
licity. Direct Twitter interactions were used to shape the con
ict narrative, demonstrate resolve,
attract supporters abroad, and in
uence public opinion, thus, having direct impact on the con
ict.
When exploring world leaders' interactions on Twitter, I rst need to establish the most common
mode of these interactions. First, I can expect that, compared to the ordinary Twitter users,
political leaders would engage with each other at lower rates than ordinary Twitter users. It
has been found that political leaders mostly use social media platforms as a top-down channel
to broadcast information (Barber a and Zeitzo, 2017). According to Groll (2015), the fact that
leaders rarely respond to or interact with their followers shows that \world powers have embraced
Twitter more as a propaganda tool than as a two-way method of communication." If leaders are
not highly engaging with their constituents, then I can expect even lower levels of engagement with
59
other leaders. Second, and perhaps more importantly, communication between political leaders
represents a public act visible to all followers of an actor. Friendly or aggressive engagement with
a leader of a foreign country might have unexpected reaction among the followers; for example,
Donald Trump's aggressive mentioning of Theresa May on Twitter sent shock waves in both US
and UK (Smith, 2017). With the whole country watching, mentioning or retweeting other leaders
on Twitter might represent a costly act that might backre or bring negative publicity. Thus,
holding everything else constant, I expect that communication between leaders is highly strategic
and occurs more rarely than between ordinary users.
Following this thesis of strategic communication, retweeting should be especially rare among
the world leaders as they would rarely choose to display words of some foreign leader to their
constituents. Mentioning should be a more preferred way of communication as it establishes some
kind of a connection with a foreign leader/government without losing domestic legitimacy (Strau
et al., 2015). Following this logic, I hypothesize:
H1: Method of communication: Mentioning will be a more preferred way of communication
between leaders than retweeting.
Considering that mentioning and retweeting represents a costly act for a leader that might
backre or result in a loss of legitimacy, investigating the patterns of such engagement can reveal
some telling information about the leaders' usage of social media platforms. The key question here
is: When leaders do retweet or mention, whom do they engage with? One possible answer is that
leaders' interactions on social media closely resemble their interactions in the oine world. If this is
the case, I should see leaders mentioning/retweeting leaders that they interact most diplomatically
within certain geographic regions or along the similar level in political hierarchy. In other words, I
would see leaders forming mention/retweet communities by regions or type of actor (i.e. Ministers
of Foreign Aairs would form communities with other Ministers of Foreign Aairs). Following this
logic,
H2: Clustering in network: Leaders' interactions on social media would resemble their interac-
tions in the oine world with nodes clustering by region and type of actor.
60
While it seems plausible that leaders would cluster by region and type of actor, I also expect
that the regime type might play a role in the way Twitter communities are formed. Specically, I
expect that leaders of democratic countries might cluster together replicating their special relation-
ship on the international arena. A large body of literature in International Relations indicates that
democratic countries form a special community based on shared values and norms (Risse-Kappen,
1995b, Maoz and Russett, 1993a, Dixon, 1994). In his seminal \Cooperation among Democracies,"
Risse-Kappen (1995b) argues that democracies have a strong collective identity based on shared
values such as human right, the rule of law, and democratic governance. Considering strong pat-
terns of cooperation and engagement between democratic leaders and governments in real life, I
expect to detect these patterns of engagements in the online sphere as well:
H3: Interactions among Democracies: Leaders from Democratic states are more likely to engage
with other democratic leaders.
Finally, as any other network, the world leaders' network should contain a center and a periphery
(Csermely et al., 2013, Barber a et al., 2015). Traditional communication theory states that a
minority of users called in
uentials drive trends on behalf of the majority of ordinary people (Rogers,
1962). These individuals could be described as hubs, connectors, opinion leaders, or simply the
most informed, respected, and well-connected members of a certain network or society (Cha et al.,
2010, Katz et al., 1955). In
uence is frequently measured through the number of a user's retweets,
as it indicates the ability of that user to generate content with pass-along value, and a number of
mentions, as it indicates the ability of that user to engage others in a conversation (Cha et al.,
2010). Following the existing theories of communication, I expect that certain leaders will occupy
central positions in the network and receive a lion share of interactions from other users; while
leaders on the periphery might be barely engaged with at all. Not surprisingly, the top in
uentials
in ordinary Twitter networks are mostly celebrities and public gures (Cha et al., 2010); however,
what can explain popularity patterns among the world leaders?
One possible explanation is that online hierarchies might closely replicate international hier-
archies we observe in the real world (Donnelly, 2006, Lake, 2013). As some countries are more
61
powerful and dominant than others, online networks could closely re
ect this oine pecking order.
While international hierarchies are not simply based on material capabilities (Nye, 2004, Hall, 1997),
empirical analysis indicates that economic strength still plays a substantial role in the countries'
international status. For example, according to Renshon (2016), the correlation between material
capabilities and status typically hovers between 0.5 and 0.75. Therefore, more dominant countries,
as measured by their economic strength, might attract more engagement online than leaders who
represent countries with limited material resources. Following this logic, I hypothesize that:
H4: Centrality: Leaders of high income countries will occupy more central positions in the lead-
ers' network.
Another possibility is that democratic leaders occupy more central positions within social
networks due to the dierences in online behavior between democratic and autocratic politi-
cians/institutions. While autocratic leaders might be equally active on social media, they adopt
a so-called uni-directional communication style that \involves the projection of opinions with lit-
tle to no interaction in the other direction" (Bulovsky, 2018, 4). Conversely, democratic leaders
engage in multi-directional communication that involves open and circular
ow of discourse with
engagement of dierent viewpoints. According to Bulovsky (2018), such dierences in style de-
pends on the regime's incentive structure. Authoritarian power structure incentivizes leaders to
use social media accounts for national or international self-promotion rather than engaging in con-
versation. In democracies, free and fair elections push leaders to be more transparent, accountable,
and responsive (Schmitter and Karl, 1991), that results in \consistent media presence that exhibits
multi-directional communicative tendencies" (Bulovsky, 2018, 4). Thus, due to multi-directional
nature of democratic leaders' social media accounts I can expect that:
H5: Centrality: Leaders of democratic countries will occupy more central positions in the lead-
ers' network.
62
4.3 Data
4.3.1 Twitter Dataset
In order to test these theoretical predictions, I use a dataset of all Twitter communication between
world leaders (heads of state, heads of government, and ministers of foreign aairs) of 193 U.N.
member countries. For each country, a list of relevant names and institutions was identied using
the publicly available United Nations Protocol and Liaison Service website (www.un.it/protocol)
as of August 2016. Every name and institution on the list was matched with a respective Twitter
account if it exists and has been active (contains at least ten or more posts).
4
The presence of
world leaders on Twitter typically takes two forms: a personal account with posts that appear to
be written by the world leader herself or himself or an institutional account for the position, such
as the account for the presidency, prime ministry, or foreign policy ministry (Barber a and Zeitzo,
2017). In the dataset, personal accounts were distinguished from institutional by whether the name
of the prole corresponds to the world leader and whether his or her picture is used as a prole
picture.
5
Besides, the same institution or a leader might have multiple accounts: one in a local
language and one in a foreign language (mostly, English).
Both personal and institutional accounts in local and foreign languages were collected, while
parody or fake accounts were carefully excluded.
6
Using Twitter's Rest API, all Twitter commu-
nication from these accounts was gathered in a large dataset of 878,241 tweets that contains all
Twitter communication that the leaders engaged in from January 1, 2012 to June 1, 2016 or during
their tenure (for further information about the dataset see Barber a and Zeitzo (2017), Barber a
et al. (2018)). The Figure 4.1 shows the time line of the volume of the tweets and the users who
produced these tweets during the aforementioned period.
7
4
The list of accounts was partially based on the \Twiplomacy" dataset available at http://twiplomacy.com/.
Every account was veried by the team of research assistants manually. For multiple accounts with the same name,
a veried account denoted by a blue \check" sign was selected. In the absence on verication, the account with the
largest number of followers was selected.
5
Previous research showed that institutional accounts are slightly more common on Twitter than personal accounts
(Barber a and Zeitzo, 2017).
6
We assume that in most cases social media messages are posted by the leader's communication sta. However, the
specic person posting from the account is irrelevant for my analysis since I am interested in general communication
strategy rather than leaders' personal preferences or attitudes.
7
The shape of the graph re
ects the data collection procedure: the further we move from the August 2016
timestamp, the fewer accounts exist in my dataset as we move beyond most leaders' tenure. Such 'decay' occurs only
for personal accounts.
63
My theoretical argument suggests that the leaders' interactions on social media will be pre-
dicted by their regime type, geographical region
8
, and level of economic development. I use the
revised Polity IV scores (Marshall and Jaggers, 2002) to classify countries as either autocratic,
semi-autocratic, semi-democratic, or democratic.
9
Besides, to measure the levels of economic
and population development, I use GDP and rates of internet penetration from the International
Telecommunication Union (ITU).
Figure 4.1: Volume of unique users and tweets by month
Table 4.1 reports some aggregate statistics of the dataset. My dataset covers the communication
8
I use the World Bank classication of countries into regions: East Asia and Pacic (15%), Europe and Central
Asia (27%), Latin America and Caribbean (17%), Middle East and North Africa (11%), North America (1%), South
Asia (1%), and Sub-Saharan Africa (24%).
9
We use the conventional cut-os for Polity2, whereby: (autocracy: -10:-8), (semi-autocracy: -7:0), (semi-
democracy:1-7), (democracy: 8-10).
64
of 570 leaders who created over 1.1 million tweets during the data collection period (or their tenure
time), with the majority of them being original posts. Table 4.2 reports the distribution of accounts
in my dataset by type of actor, regime type, and region. As we can see, the accounts are distributed
quite evenly among heads of governments, heads of states, and ministers of foreign aairs. The
distribution of accounts by regions also mostly corresponds to the number of countries in these
regions. As for the middle part of Table 4.2, it indicates that my dataset over-represents democratic
and semi-democratic countries and under-represents autocratic states.
10
Table 4.1: Twitter Data Descriptive Statistics
Statistic Count
# of Leaders in Dataset 570
# of Tweets 1,177,497
# of Retweets 308,324
Table 4.2: Distribution of Accounts by Type of Actor, Regime, and Region
Type of Actor Count Regime Count Region Count
Head of Government 180 Autocracy 20 Sub-Saharan Africa 106
Head of State 192 Semi-autocracy 67 South Asia 24
MFA 198 Semi-democracy 123 Middle East & North Africa 66
Democracy 234 Latin America & Caribbean 99
North America 11
Europe & Central Asia 205
East Asia & Pacic 53
4.3.2 Retweet and Mention Networks
In order to explore the leaders' communication patterns, I constructed two networks corresponding
to two main types of interaction on Twitter: retweet and mention networks. Retweet network
contains nodes with a direct link between them if one user retweeted a post of another. In con-
trast, a direct link on a mention network indicates that one user mentioned another in his or her
post. While leaders retweet and mention a wide variety of dierent users including international
organizations, media outlets, and even ordinary citizens, here I am only interested in communi-
cation among heads of state, heads of government, and ministers of foreign aairs. Thus, I limit
10
Some of the obvious missing cases are China, North Korea, and Syria that simply do not maintain governmental
Twitter accounts.
65
Table 4.3: Descriptive statistics of the Leaders' Retweet and Mention Network.
Retweet Network Mentions Network
# of nodes 447 500
# of edges 3,550 7,162
Max weighted in-degree 340 780
Max weighted out-degree 6 46
Density 0.018 0.029
my analysis to the networks consisting exclusively of leaders' accounts, with other nodes removed.
Moreover, considering my interest in the patterns of international rather than domestic social media
communication, I only keep the links that connect leaders with leaders of other countries on the
list.
11
Table 4.3 shows the descriptive statistics of the retweet and mention network. The number of
nodes indicates the number of unique leaders in each network. The numbers are somewhat lower
than the overall number of leaders in the dataset (570) as any leaders who did not mention/retweet
or were mentioned/retweeted during the period of interest were excluded. The number of edges
indicates the number of unique interactions between leaders. Maximum weighted in-degree indicates
the highest number of times any leader in the network was mentioned/retweeted during the period
of data collection. Conversely, maximum weighted out-degree shows the highest number of times
any leader in the network mentioned/retweeted other leaders during the time period. Finally,
density is the number of observed edges divided by all possible number of edges between leaders.
As we can see, both networks described in the table are sparse networks; however, the mention
network is roughly twice more dense than the retweet network. Besides, it contains signicantly
higher number of nodes and edges. Such dierences in the leaders' retweet and mention networks
indicate that mentioning is a more preferred way of communication between leaders than retweeting.
Compared to ordinary Twitter users, world leaders refrain from retweeting each other, possibly,
to preserve legitimacy among their constituents. This descriptive result supports my Hypothesis 1
that mentioning will be a more preferred way of communication between leaders than retweeting.
11
It is widely common for leaders to mention/retweet other leaders of the same country. For example, ministers of
foreign aairs often retweet heads of state, and vice versa.
66
4.4 Results
Considering my descriptive nding that mentioning is a more preferred way of communication
between leaders than retweeting, I focus my further investigation of leaders' communication patterns
on the mentions network. For the sake of space, the respective results for the retweet network could
be found in the Appendix C.3. In this section I rst explore the features that explain the leaders'
clustering in the network. Do leaders form online communities resembling their oine interactions?
Second, I investigate how exactly leaders cluster within features. For example, if leaders form online
geographic communities, which regions cluster together? Finally, I explore centrality patterns in
the leaders' network. In other words, which leaders are the most important in the mention network,
and why?
Before I start answering these questions, it might be useful to explore when and for what purpose
world leaders actually mention each other. Qualitative analysis of a random sample of tweets
containing mentions indicates that leaders predominantly mention other leaders in the following
situations:
1. During state visits, summits, or other diplomatic events;
2. Congratulating and sending best wishes on national holidays;
3. Expressing condolences on tragic events, such as acts of terrorism or natural disasters;
4. Personal birthday wishes or farewells to foreign colleagues who step down from their respective
positions;
5. Expressing support for a policy decision made by a foreign leader;
6. Expressing criticism for some foreign policy act or a policy decision.
The last category is the most rare but also, perhaps, the most interesting. For example, we
see Presidential Administration of Ukraine criticizing Prime Minister of Russia for their policies in
Crimea or Minister of Foreign Aairs of Turkey attacking MFA of Sweden for an improper and inac-
curate statement. Public mentioning of another country on Twitter in such critical context allows
leaders to directly communicate with the citizens of that foreign country and broader international
67
audiences. Appendix C.1 provides further examples of tweets that illustrate these dierent themes,
along with the wordcloud that displays the most common words used in the mentions tweets.
While qualitative investigation of individual tweets is undoubtedly important, the analysis of
the whole network is needed in order to understand general patterns of interaction between world
leaders. I begin my analysis with visual examination of the world leaders' mention network. Figure
4.2 displays links between leaders with node sizes representing the weighted indegree (how many
times a leader was mentioned by other leaders).
12
We can see, for instance, that two leaders { John
Kerry and Fran cois Hollande { have the highest indegree in my network. The colors correspond
to network communities constructed by the Louvain method. The Louvain algorithm seeks to nd
the best partitions by maximizing the so-called modularity score, which is a measure usually used
to evaluate the quality of partitions being produced by community detection algorithms (De Meo
et al., 2011).
13
Overall, Figure 4.2 shows that certain nodes cluster together; however, it does not
indicate what might explain such clustering.
4.4.1 Leaders' clustering and interaction patterns
Clustering between nodes of similar types (homophily) is one of the most important aspects of
networks. Figure 4.2 shows that nodes cluster together based on their connections to other nodes;
however, in this chapter I do not only care to see nodes' clustering based on their connections, but
I aim to explore how certain node features explain this clustering. In order to see which features
explain the clustering in the given network, assortativity coecient has been used (Newman, 2018,
2002). Assortativity refers to the extent to which similar nodes are connected in a given network.
Assortativity coecient is typically quantied by computing the Pearson correlation coecient
of the degrees at each end of links in the network if the feature in question is numerical and by
using the modulairty measure if the feature is categorical (Newman, 2002). Assortativity coecient
ranges from [-1,1] for both categorical and numerical features, with higher values indicating higher
assortativity. For example, if many edges of a network connect nodes with a certain feature (for
example, many edges connect leaders from a particular geographic region), this attribute or feature
12
To make the gure more readable, I eliminated the nodes with weighted indegree below 20.
13
The modularity of a partition is a value between [-1,1] that measures the dierence between the density of a
network and the expected number of links between nodes if the network was randomly congured (Barab asi, 2013,
Newman, Newman, 2006).
68
Figure 4.2: Leaders' Mention Network
will have a high assortativity coecient.
Assortativity coecient has been widely used to determine which features explain clustering of
users on social media, such as Twitter. For example, Bliss et al. (2012) use assortativity coecient
to show that levels of happiness explain clustering of the users in a massive Twitter network. In
this case, I expect that leaders' interactions on Twitter resemble their interactions online, thus,
69
Table 4.4: Mention Network Assortativity Coecients
Feature Assortativity Coecient
Type of actor 0.41
Geographic region 0.27
Regime 0.019
Population -0.01
Income level 0.05
Internet access 0.1
the nodes should cluster by region and type of actor (Hypothesis 2). Table 4.4 displays a list of
attributes of the leaders' accounts and respective assortativity coecients. As we can see, a type
of actor, that indicates whether an account belongs to a Head of State, a Head of Government, or
a Minister of Foreign Aairs, has the highest coecient meaning that actors of the same position
tend to cluster together more. Similarly, the leaders from same geographic regions tend to form
strong online communities as predicted in my hypothesis. The other features, such as regime type,
level of economic development, and internet access do not seem to play an essential role in the
assortativity of the world leaders' network. Assortativity coecients table for the retweet network
presented in the Appendix C.3 shows a similar pattern with type of actor and geographic region
explaining clustering patterns.
4.4.2 Heatmaps
Assortativity coecients presented in the previous section indicate that world leaders do not seem
to cluster by regime type; however, that does not necessarily mean that the regime type does not
play any role in the way Twitter communities are formed. Assortativity coecient for regime type
re
ects clustering along all categories of the attribute, including autocracies and semi-autocracies
that might not interact much with each other. In order to properly test my third hypothesis that
democratic countries form a special online community on Twitter, I need to look within features
by breaking them down in particular categories.
To illustrate the concept, let us rst look at the heatmap in Figure 4.3 that breaks down the
geographic clustering into specic regions. Each box in the heatmap represents a count probability
of edges, where one node has an attribute from the X-axis and another node has an attribute
from the Y-axis (Newman, 2018). For example, looking at the heatmap of clustering by region, we
70
can see that the edges where both nodes are European leaders have the highest count probability
of 0.4 (in other words, they constitute 40% of the total number of edges in the entire mention
network). Besides, the heatmap shows that the count probability of edges between European
and Middle Eastern leaders is also quite high at around 0.063 and 0.067 between Latin American
Leaders. Overall, rather than simply concluding that leaders on Twitter cluster by region, I can
say specically whether the leaders of particular regions tend to interact the most.
Figure 4.3: Clustering by Specic Regions
Moving on to Hypothesis 3, heatmap in Figure 4.4 breaks down the clustering by regime into
specic types of government. The heatmap clearly shows that leaders of democratic countries do
cluster together on Twitter. The dark box in the middle of the heatmap indicates that the edges
where both nodes are democratic leaders have a very high count probability of 0.4. We can also
note the moderately high count probability of edges connecting democratic and semi-democratic
leaders (around 0.12). At the same time, almost white boxes of edges between autocratic and
autocratic/semi-autocratic leaders indicate extremely low levels of online interaction between these
countries. Therefore, I can conclude that the heatmap in Figure 4.4 provides support for my
hypothesis that democratic leaders are more likely to engage with other democratic leaders on
71
Twitter. Similarly to the oine world, democratic countries form a special community on this
social media platform.
Figure 4.4: Clustering by Specic Type of Regime
4.4.3 Predictors of Social Media Centrality
My nal hypotheses 4 and 5 concern the patterns of leaders' centrality on social networks. I expect
that leaders of high income and democratic countries will occupy more central positions in the
network. In order to test this relationship, I must rst select the most appropriate operationalization
of \centrality" as measurement of importance of a node in a network.
The importance of a node in a network is one of the most important questions in network science.
Considering a vast variety of applied problems (importance of an individual in a network of people; a
bank in a nancial transactions network; a web-page on the internet; a country in a trade network),
multiple centrality measures suitable for various problems have been developed. Generally there are
are two classes of centrality measures. The rst group of centrality measures captures how many
connections a node has and includes such measures as: degree centrality (how many connections
a node has and has two versions for directed networks: in-degree and out-degree); Eigenvector
72
centrality (sum of the centralities of a node's neighbors, suitable for undirected networks); and
PageRank centrality (centralities of a node's neighbors divided by their out-degrees, suitable for
directed networks). The second group of centrality measures capture the position of a node in a
network. The most notable measure from this group is betweenness centrality that calculates a
number of shortest paths between a pair of nodes that passes through the node in question.
In this chapter I use weighted in-degree centrality to capture importance of world leaders.
In-degree centrality belongs to the rst group of centrality measures. While position centrality
(second type) is important, I do not believe that it measures the kind of importance that I am
aiming to capture here. For example, states or state leaders who serve as interlocutors between two
regions of the world or dierent political campuses, would have high betweenness centrality, but
they usually are not important states in terms of militaristic, economic, and cultural/soft power.
Indegree indicates that we are mostly interested in incoming connections that occur when a leader in
question is being mentioned or retweeted by other leaders. I choose indegree over more sophisticated
measures like PageRank, as the latter usually gives too much importance to few nodes while giving
very small importance values to the rest of the nodes, making the comparison of importance for
most nodes quite dicult.
14
Finally, I use weighted indegree as it captures the number of times a
person was mentioned rather than a mere fact of mentioning, which is an important indicator of
the importance of a node or in my case.
The distribution of weighted indegree is presented in Figure 4.5. A heavy-tailed distribution
(also called power-law or scalefree) with a few accounts having a high level of indegree has been
found typical for the distribution of the number of ties of a person in a social network (Barab asi and
Albert, 1999, Caldarelli, 2007). Often such heterogeneous levels of activities follow the well-known
and widely applicable law postulated by Pareto, which states that 80% of the eects are induced
by 20% of the causes (Muchnik et al., 2013). The Pareto law seems to apply remarkably well to the
weighted indegree distribution in my network: 20% of the top accounts produce 75% of mentioning
activity. We dene these 20% of top accounts as \central" and the rest 80% as \peripheral", and
use logistic regression to explore which factors predict centrality of an account.
Table 4.5 displays the coecient estimates for a set of logistic regressions of social media cen-
14
Additionally, the random-walker model here is not necessary, since I believe that a world leader mention is a
costly signal (unlike retweets in a typical retweet network in most Twitter studies) that signals the importance of the
person being mentioned.
73
Figure 4.5: Distribution of Weighted Indegree
Note: Figure shows distribution of weighted indegree. Dotted line indicates the median.
trality on income level and regime type. The rst model introduces a set of control variables that
pertain to the type and quality of the account rather than the characteristics of a country. Ac-
count age measures how old the Twitter account is (# of days since a user signed up). Controlling
for this variable is important as older accounts would quite logically accumulate more mentions
or retweets over time. Statuses count measures how many tweets a user has produced (includes
original tweets and retweets). Burstiness is is a measure of burstiness of a system or phenomenon
74
(also called the coecient of variation) that is measured by subtracting the mean of a distribution
from its standard deviation then dividing it by the sum of the mean and the standard deviation.
This metric is bounded from [-1,1] where 1 means it is a bursty signal; 0 means neutral; and -1
means regular (periodic) signal.
15
Own language is a dichotomous variable coded as 1 if whether
an account is maintained in a country's local language and 0 for English. Actor is a categorical
variable indicating a leader's position: Head of State, Head of Government, or Minister of Foreign
Aairs (MFA). Type of account is a dichotomous variable coded as 1 for personal account and 0
for institutional account.
Model 1 presents a baseline that explains variation in centrality solely through the type and
quality of account. Unsurprisingly, the age of account is strongly correlated with online centrality
- the older the account, the more mentions it accumulates. The level of overall online activity
also matters: the more tweets a user produces, the more likely he or she is to be central in the
mention network. Ministers of Foreign Aairs tend to occupy more central positions in the network
compared to other types of accounts, possibly, due to their stronger international and diplomatic
activity. Posting in a country's local language is negatively associated with high network centrality
as such accounts are primarily maintained for domestic consumption. Finally, personal accounts
tend to be more central than institutional accounts, perhaps, due to their more engaging nature.
Now, moving to my variables of interest, Models 2 and 3 test hypotheses 4 and 5 respectively.
Model 2 looks at the eect of income level measured through GDP and rates of internet pene-
tration. GDP is tted as a categorical variable with three levels corresponding to the terciles of
the distribution (\Low income" , \Middle income", \High income").
16
Insignicant coecients in
Model 2 indicate that high income countries are not more likely to occupy central positions in the
leaders' network than countries with low or middle income. Rates of internet penetration do not
seem to be a key predictor of network centrality. H4 nds no support. Model 3 tests the eect of
regime type using the dichotomous specication of regime type in which countries with a Polity
IV score of 6 or higher are coded as democracies (Jaggers and Gurr, 1995, Barber a and Zeitzo,
2017). The signicant and positive coecient of the dummy indicates that democratic leaders are
indeed more likely to be central in the network than autocratic leaders.
17
15
For more on this metric, read Goh and Barab asi (2008).
16
In Appendix C.2, I replicate the analysis using the continuous measure and show that the results stay same.
17
In Appendix C.2, I replicate the analysis using a continuous variable of Polity IV and electoral democracy index
75
Table 4.5: Mention Network Centrality
Network Centrality (Top 20%)
(1) (2) (3) (4)
Account age 4.393

3.566

3.900

3.205

(0.685) (0.715) (0.705) (0.738)
Statuses count 1.973 3.139

2.045 2.809

(1.016) (1.275) (1.051) (1.313)
Burstiness 0.388 0.002 0.296 0.503
(2.548) (2.769) (2.479) (2.662)
Own language 0.917

0.824

0.897

0.880

(0.271) (0.285) (0.282) (0.295)
Actor: Head of State 0.370 0.573 0.324 0.570
(0.355) (0.378) (0.371) (0.390)
Actor: MFA 1.374

1.479

1.354

1.502

(0.339) (0.358) (0.352) (0.370)
Personal account 2.191

2.471

2.295

2.589

(0.295) (0.327) (0.311) (0.342)
Middle income 0.139 0.149
(0.468) (0.466)
High income 0.755 0.633
(0.641) (0.637)
Internet users 0.978 1.171
(0.976) (0.971)
Democracy Dummy 1.077

0.879

(0.314) (0.337)
Constant 4.946

6.090

5.895

6.916

(1.828) (2.047) (1.808) (1.996)
N 531 491 478 459

p < .05;

p < .01
Notes: Table displays logit coecients, standard errors in parentheses. All
non-dichotomous measures have been rescaled from 0 to 1.
The result is robust to adding income variables as shown in the saturated Model 4. Considering
that economic development might be a strong predictor of democracy (Cheibub et al., 1996), it
is useful to control for income to better isolate the relationship between democracy and centrality
(Bulovsky, 2018). Figure 4.6 shows predicted probabilities of being central in mention network for
democratic and non-democratic leaders based on the saturated model.
18
The predicted probability
of being central in the network is 9.6% higher for democratic leaders than non-democratic leaders
from the Varieties of Democracies Project (V-Dem). The results stay same regardless of the specication.
18
These predicted probabilities are derived from 1,000 bootstrapped logit models of the impact of regime on network
centrality, keeping control variables at their means and modes.
76
(p < 0.01 from a bootstrapped dierence of means test). The gure strongly suggests that demo-
cratic leaders occupy more central position within the leaders' mention network as I suggested in
Hypothesis 5.
Figure 4.6: Eect of Regime on Probability of Network Centrality
Note: Figure displays predicted probability distributions derived from 1,000 bootstraps from logistic regres-
sion models of the impact of regime type on the probability of being central in mention network, controlling
for other signicant variable in Model 4 of Table 4.5.
4.4.4 Checking Network Stability Over Time
In this chapter I treat the mention network as a static network and ignore its dynamic nature. Such
approach is common in network analysis (Goyal et al., 2018); however, some additional validation
77
Figure 4.7: PageRank (left) and Weighted in-degree (normalized by sum of in-degree per time period) (right)
statistics across the whole time period
is necessary to make sure that such static approach is justied. If a network changes signicantly
across time, treating it as static can be misleading since the analysis would only re
ect the nature
of the network at the end of the data collection, but not at various temporal points prior to that.
To eliminate this possibility, I test the stability of my mention network by examining the weighted
pagerank and the normalized weighted in-degree distributions across time.
The way the networks are constructed in gures 4.7 and 4.8 are by including all the nodes,
edges, and weights (number of mentions by node i to node j in a directed network) up to time
t with monthly intervals. In other words, every temporal network includes all the nodes, edges,
and weights that occurred in preceding months. By observing weighted pageranks and in-degree
distributions, I can see whether the distributions change over time or stay stable.
In gure 4.7 I show ve dierent metrics: 5
th
percentile, 1
st
quantile, median, 3
rd
quantile, and
95
th
percentile. The reason for showing dierent metrics for each temporal network is that both
pagerank and in-degree distributions for twitter data tend to be power-law distributions (Barab asi
and Albert, 1999). As mentioned above, a power law distribution means that most users have very
low pagerank or in-degree, while some top users have extremely high pagerank or in-degree. In
the context of a mention network, this means that few leaders receive the most mentions while
most leaders do not get mentioned or get mentioned only a few times. Thus, when assessing a
network metric distribution, it is imperative not to look only at the mean or median, but at the
top percentiles as well.
Figure 4.7 shows that my mention network is very stable from 2013, which is not that surprising,
since prior to 2013 there was little data collected (see Figure 4.1 for volume of tweets by time).
Figure 4.8 shows the same metrics, but starting from 2015. Since 86% of the mention data occurred
78
Figure 4.8: PageRank (left) and Weighted in-degree (normalized by sum of in-degree per time period) (right)
statistics form 2015 onwards, where more than 86% of the mention data is represented
after this date, it is important to take a deeper look at the metrics trends within this time frame.
All the metrics are fairly stable, although the 95
th
percentile looks slightly less stable as the other
metrics. Looking at the dierence between the upper and the lower bounds on the y-axis for this
metric, we can see that it is quite small in both the pagerank and in-degree distributions. Taken
together, gures 4.7 and 4.8 justify my assumption that the mention network can be treated as a
static network, ignoring the temporal variability.
4.5 Discussion and Conclusion
In this chapter I use a unique dataset of all Twitter communication between the leaders of U.N.
member countries to provide a rst look into the world leaders' interactions on social media. I have
three key ndings. First, I reveal that leaders' interaction on social media closely resemble their
interaction in the oine world. Similarly to the oine diplomatic communities, I show that leaders
form mention/retweet communities along regional lines and similar levels in political hierarchy.
Second, I nd that the regime type plays a key role in the way Twitter communities are formed.
Consistent with the expectations of the democratic peace theory, I nd that leaders from democratic
states are more likely to engage with other democratic leaders. Finally, I explore the popularity
patterns among the world leaders and conclude that leaders of democratic countries tend to occupy
more central positions in the network, with leaders of non-democratic countries present at the
periphery. One potential explanation of this result is that democratic leaders tend to be more
transparent and responsive online (Bulovsky, 2018). Rather than simply using social media for
credit claiming and self-promotion, they are wiling to engage in active conversation with their
79
followers. At the same time, I nd no indication that online hierarchies might be based on material
capabilities.
While my project provides a rst important step towards understanding of the leaders' inter-
actions online, many questions remained unanswered. For example, here I focus on the leaders'
network connections, rather than the content of their interactions. Content analysis may provide
greater insight into why and when leaders mention or retweet each other. Are these interactions
mostly friendly and diplomatic in nature? Or do leaders sometimes use these tools to publicly
criticize or attack their political opponents? These questions should be addressed in future work.
Furthermore, the conclusions made in this chapter are unavoidably limited by the scope of the
data at hand. First, as mentioned somewhere in the chapter, certain leaders and governments
are missing from the dataset in a non-random fashion. The dataset under-represents autocratic
states as many of them do not have an active Twitter account. Besides, the dataset covers a very
limited time period with leaders dropping out when their tenure ends. Thus, the data provides
more of a static snapshot rather than a temporal comparison of the leaders' interactions developing
and changing overtime. I hope that future data collection eorts will work to solve this issue and
provide a more comprehensive investigation of the dynamic nature of the leaders' interactions.
80
Chapter 5
Conclusion
In this dissertation, I have sought to improve our understanding of how the public forms foreign
policy attitudes and how political leaders use online platforms to form public opinion at home and
abroad. Using a multi-method approach that combines survey experiments with computational
big data research, I examine people's inferences about ambiguous international events, the eect of
intergroup threat on the formation of moral judgement, and the patterns of strategic interactions
between political leaders on social media. Across two rst chapters, I nd that both situational
and dispositional factors play a role in the process of belief formation, while the third chapter is
important to understand how leaders use online platforms to form beliefs and opinions among the
public at home and abroad. Taken together, all three studies highlight the importance of studying
foreign policy attitudes in international relations scholarship and open promising directions for
future research. In the remainder of the chapter, I brie
y summarize the main ndings of the
empirical chapters and propose how we could move these research agendas forward.
In chapter 2, I examine how people make inferences and attribute blame in ambiguous inter-
national events. Using a survey experiment, I contest the existing assumption that all foreign
international actors are considered as an outgroup with respective attributional patterns. Rather, I
show that ingroup-outgroup categorization should be understood as continuum: attributions change
incrementally when we move from non-allies with the most negative expectations to strong allies
with the most positive expectations. Besides ingroup-outgroup categorization process, the results
of this chapter question the existing assuption of unitary audiences prevalent in attributional re-
search in international relations. show that people's attributional patterns vary dramatically due to
81
dierences in their dispositional characteristics. Specically, I show that people with conservative
ideological leanings make very dierent attributions about behavior of ingroups and outgroups.
Future research on attributions of blame in IR should build on my ndings and clarify the
scope of my theory. Here, I suggest a pattern of ingroup-outgroup categorization that should be
applicable to a subset of democratic Western countries, but what about non-democratic and non-
Western countries? When adopting a theory to other countries, some shared characteristics might
be excluded and others added. For example, when adopting a theory to non-democratic countries,
the regime type should, perhaps, be excluded as a predictor of strength of ties between countries.
At the same time, common enemy and shared experience of ghting together in previous con
icts
should prime closeness of ties even for publics in authoritarian states. When applying a theory to
non-Western countries, it will be possible to emphasize other cultural elements, such as religion
(for example, Islam for Muslim countries) or philosophical tradition (such as Confucianism for East
Asian countries). Further research is needed to determine whether my argument about ingroup-
outgroup categorization would hold for non-democratic and non-Western countries, or whether
other countries/regions simply do not have allies strong enough to be considered as part of an
ingroup.
Besides, future research on attributions should consider other dispositional factors that might
aect individual inferences. It might be useful to look at core values, such as moral foundations,
that might serve as crucial mediators between ideology and attributions. Another possibility is that,
when judging international events and interactions, people might be driven not only by general ide-
ological orientations, but also by more specic dispositions towards international aairs. Existing
literature distinguishes three fundamental dispositions toward international aairs: \militant inter-
nationalism" (MI), \cooperative internationalism" (CI), and isolationism (Wittkopf, 1990, Kertzer
et al., 2014, Chittick et al., 1995). While we know that these orientations might be strong predictors
of foreign policy preferences (Herrmann et al., 1999, Chittick et al., 1995, Rei
er et al., 2011), we
are yet to explore if these predispositions might also explain the way people make inferences about
other countries' ambiguous behavior.
In chapter 3, I examine how people form moral judgement about adversarial foreign nations and
the role of intergroup threat in this process. Across two studies, I nd support for my theory that
perception of intergroup threat might lead to the formation of moral judgement about the source of
82
threat, even when no information about morality of the group is available. Using large observational
dataset of political tweets, I show that conversation about adversarial foreign countries tends to
contain strong moral rhetoric. I also use a randomized survey experiment to test the causal nature of
this relationship and nd that priming individuals with intergroup threat increases moral judgement
of the source of threat even when no information about morality of that source is available. Besides
the average eect of intergroup threat on moral judgement, I also show that people moralization
patterns vary due to their individual dierences. My research provides preliminary support that
people with conservative leanings tend to moralize threats more than the liberals.
While promising, these ndings are only the beginning of an exploration of the vast topic of
moral judgement in international relations. There are multiple potential avenues for future research
on this topic. For example, one promising avenue for future research concerns possible behavior
consequences that follow from moral judgement caused by intergroup threat. Intergroup threat has
been consistently linked with support for aggressive policies against threatening groups (Huddy
et al., 2005); however, we do not know whether moral judgement serves as a necessary mediator in
that process. Or does threat sometimes lead to support for aggressive action without intermediate
moral reaction? And would the behavior consequences be dierent in these two scenarios? My
expectation is that the perceived external threat with consequent moral judgement might lead to
more severe behaviors as morally inferior groups are not believed to deserve a fair treatment, but
this remains to be seen in future research.
Another important limitation of the project is that here I conceptualize intergroup threat as
a realistic threat to a country's national security (thus, the use of the word \enemy") while other
types of intergroup threats remain unexplored. For example, research in social psychology identies
another important type of threat { symbolic threat { that involves perceived group dierences in
values, morals, and beliefs. In political psychology, it has been identied as a consistent predictor of
racial bias and anti-immigration attitudes (Kinder and Sears, 1981, McLaren, 2003). Considering
its emphasis on values and morals, it would be interesting to see whether symbolic threat might
produce even stronger moral reaction than realistic threat. Another type of threat that might lead
to moral judgement is group esteem threat that occurs when the actions of an outgroup potentially
decrease an ingroup's esteem (Tajfel and Turner, 1979, Branscombe et al., 1999). Support for the
relationship between group esteem threat and outgroup hostility has been found across a number
83
of studies (Branscombe and Wann, 1994, Branscombe et al., 2002). Future research should explore
whether dierent types of intergoup threat would aect moral judgement dierently and whether
dierent people are more sensitive to dierent types of threat.
Finally, future research on moralization of threats should consider other individual characteris-
tics that might aect moral judgement. Rather than focusing on political ideology, we could look
into more basic psychological predispositions that inform ideological leanings in the rst place. Two
obvious examples of such predispositions are Social Dominance Orientation (SDO) and Right-Wing
Authoritarianism (RWA) that are generally correlated with conservative ideology but emphasize
moral values to a dierent extent (Cohrs et al., 2005, Duriez and Van Hiel, 2002, Jost et al., 2003,
Sidanius and Pratto, 2001b). Thus, people high on SDO value dominance, power, and achievements
and express prejudice due to instrumental concerns, such as striving for power and superiority
(Cohrs et al., 2005, Pratto et al., 1994). Thus, we could expect that people high on SDO tend to
respond to threats in amoral terms. In contrast, people high on RWA are more concerned with
ingroup norms and traditions and their prejudice is typically motivated by identity concerns (Cohrs
et al., 2005); thus, we could expect that people high on RWA will be more sensitive to moral viola-
tions. Another possibility would be to look at both RWA and LWA (Left Wing Authoritarianism)
to see if people high on dierent types of authoritarianism would emphasize dierent moral values
(Van Hiel et al., 2006, Altemeyer and Altemeyer, 1996). It has been found that both RWA and
LWA are corrected with conservatism, albeit in a slightly dierent way: while LWA is more strongly
related to economic conservatism, RWA is more strongly related to cultural conservatism (Van Hiel
et al., 2006). More research is needed to understand which foundation of conservatism underpins
its tendency to moralize external threats to a higher extent.
In chapter 4, I examine how political leaders strategically interact with each other on social
media. Using a novel, cross-national dataset of Twitter communication for 193 world leaders for
the period of 2012-2017, I analyze retweet and mention networks to explore the patterns of leaders'
communication. I use social network analysis to conclude that the leaders' interactions on social
media closely resemble their interactions in the oine world: most connections are formed along
regional lines and similar levels in political hierarchy. Besides, consistent with the democratic peace
theory that underscores a special connection between democracies, I identify political regime as the
main predictor of clustering between countries on Twitter. Finally, I explore the patterns of the
84
leaders' centrality to identify features that determine which leaders occupy more central positions
in the network. I show that leaders of democratic countries tend to occupy more central positions
in the network, with leaders of non-democratic countries present at the periphery. Overall, this
chapter yields new insights on how governmental actors use new social media tools to aect public
opinion at home and abroad.
While this chapter provides a rst important step towards understanding of the leaders' interac-
tions online, many questions remained unanswered. First, and, perhaps, most importantly, future
work is needed to examine whether leaders' communications online actually in
uences public opin-
ion in any eective way. Communicating with each other online leaders aim to attract attention
to certain policies, shape narratives, and demonstrate collaborations; however, do these attempts
play any tangible role or do they simply stay unnoticed by inattentive public? It is possible that
the level of attention to the leaders' online communication is relatively low in normal times but
spikes up in the times of international crises as suggested by Zeitzo (2018). Further research is
needed to fully adjudicate this possibility.
Besides, the conclusions made in this chapter are unavoidably limited by the scope of the data
at hand. First, this project explores world leaders' interactions on Twitter only. While popularity
of Twitter has grown in recent years, the usage of this social network varies across regions (Barber a
and Zeitzo, 2017). Besides, the leaders' activity on Twitter is largerly constrained by the particular
nature of this medium (for example, one Twitter message can have up to 280 characters). Future
work should compare leaders' interactions on Twitter with other social media websites, such as
Facebook, Instagram, Youtube, and others. Second, certain leaders and governments are missing
from the dataset in a non-random fashion. The dataset under-represents autocratic states as many
of them do not have an active Twitter account, for example, the obvious missing cases are China,
North Korea and Syria as they simply do not maintain governmental Twitter accounts. Finally,
the dataset I analyse provides more of a static snapshot rather than a temporal comparison of
the leaders' interactions developing and changing overtime. Using this data, I cannot tell whether
particular communication patterns pertain to a particular leader or to a position she or he occupies.
It would be interesting to see whether communication between leaders displays continuity even after
a tenure of a particular leader ends. I hope that future data collection eorts will work to solve
this issue and provide a more comprehensive investigation of the dynamic nature of the leaders'
85
interactions.
Finally, the last but not least, here I focus on the leaders' network connections, rather than
the content of their interactions. Content analysis may provide greater insight into why and when
leaders mention or retweet each other. Are these interactions mostly friendly and diplomatic in
nature? Or do leaders sometimes use these tools to publicly criticize or attack their political
opponents? These questions should be addressed in future work.
86
Bibliography
Aldrich, J. H., C. Gelpi, P. Feaver, J. Rei
er, and K. T. Sharp (2006). Foreign policy and the
electoral connection. Annu. Rev. Polit. Sci. 9, 477{502.
Aldrich, J. H., J. L. Sullivan, and E. Borgida (1989, March). Foreign Aairs and Issue Voting:
Do Presidential Candidates "Waltz Before A Blind Audience?". The American Political Science
Review 83(1), 123.
Almond, G. A. (1956). Public opinion and national security policy. Public Opinion Quarterly 20 (2),
371{378.
Almond, G. A. (1960). The American people and foreign policy. Praeger.
Altemeyer, B. (1998). The other \authoritarian personality". Advances in experimental social
psychology 30, 47{92.
Altemeyer, R. A. and B. Altemeyer (1996). The authoritarian specter. Harvard University Press.
Amodio, D. M., J. T. Jost, S. L. Master, and C. M. Yee (2007). Neurocognitive correlates of
liberalism and conservatism. Nature neuroscience 10 (10), 1246.
Anand, S. and J. A. Krosnick (2003). The impact of attitudes toward foreign policy goals on public
preferences among presidential candidates: A study of issue publics and the attentive public in
the 2000 US presidential election. Presidential Studies Quarterly, 31{71.
Arian, A. (1989). A people apart: Coping with national security problems in israel. Journal of
Con
ict Resolution 33 (4), 605{631.
87
Averill, J. R. (1983). Studies on anger and aggression: Implications for theories of emotion. Amer-
ican psychologist 38 (11), 1145.
Badawy, A., E. Ferrara, and K. Lerman (2018, Aug). Analyzing the digital traces of political
manipulation: The 2016 russian interference twitter campaign. In 2018 IEEE/ACM International
Conference on Advances in Social Networks Analysis and Mining (ASONAM), pp. 258{265.
Barab asi, A.-L. (2013). Network science. Philosophical Transactions of the Royal Society of London
A: Mathematical, Physical and Engineering Sciences 371 (1987), 20120375.
Barab asi, A.-L. and R. Albert (1999). Emergence of scaling in random networks. science 286(5439),
509{512.
Barber a, P., A. Gohdes, T. Zeitzo, and E. Iakhnis (2018). From riot police to tweets: How world
leaders use social media during contentious politics. Working Paper.
Barber a, P., N. Wang, R. Bonneau, J. T. Jost, J. Nagler, J. Tucker, and S. Gonz alez-Bail on (2015).
The critical periphery in the growth of social protests. PloS one 10(11), e0143611.
Barber a, P. and T. Zeitzo (2017). The new public address system: why do world leaders adopt
social media? International Studies Quarterly 62 (1), 121{130.
Baum, M. A. and P. B. Potter (2008, June). The Relationships Between Mass Media, Public
Opinion, and Foreign Policy: Toward a Theoretical Synthesis. Annual Review of Political Sci-
ence 11(1), 39{65.
Berinsky, A. J., G. A. Huber, and G. S. Lenz (2012, July). Evaluating Online Labor Markets for
Experimental Research: Amazon.com's Mechanical Turk. Political Analysis 20 (3), 351{368.
Bliss, C. A., I. M. Kloumann, K. D. Harris, C. M. Danforth, and P. S. Dodds (2012). Twitter recip-
rocal reply networks exhibit assortativity with respect to happiness. Journal of Computational
Science 3(5), 388{397.
Bonanno, G. A. and J. T. Jost (2006). Conservative shift among high-exposure survivors of the
september 11th terrorist attacks. Basic and Applied Social Psychology 28 (4), 311{323.
88
Boulding, K. E. (1959). National Images and International Systems. The Journal of Con
ict
Resolution 3 (2), 120{131.
Brambilla, M., P. Rusconi, S. Sacchi, and P. Cherubini (2011). Looking for honesty: The primary
role of morality (vs. sociability and competence) in information gathering. European journal of
social psychology 41 (2), 135{143.
Brambilla, M., S. Sacchi, S. Pagliaro, and N. Ellemers (2013). Morality and intergroup relations:
Threats to safety and group image predict the desire to interact with outgroup and ingroup
members. Journal of Experimental Social Psychology 49 (5), 811{821.
Brambilla, M., S. Sacchi, P. Rusconi, P. Cherubini, and V. Y. Yzerbyt (2012). You want to give a
good impression? be honest! moral traits dominate group impression formation. British journal
of social psychology 51 (1), 149{166.
Branscombe, N. R., N. Ellemers, R. Spears, B. Doosje, et al. (1999). The context and content of
social identity threat. Social identity: Context, commitment, content, 35{58.
Branscombe, N. R., R. Spears, N. Ellemers, and B. Doosje (2002). Intragroup and intergroup
evaluation eects on group behavior. Personality and Social Psychology Bulletin 28 (6), 744{753.
Branscombe, N. R. and D. L. Wann (1994). Collective self-esteem consequences of outgroup dero-
gation when a valued social identity is on trial. European Journal of Social Psychology 24 (6),
641{657.
Brewer, P. R., K. Gross, S. Aday, and L. Willnat (2004, January). International Trust and Public
Opinion about World Aairs. American Journal of Political Science 48 (1), 93.
Bueno de Mesquita, B. and D. Lalman (1994, February). War and Reason: Domestic and Inter-
national Imperatives (Reissue edition ed.). New Haven: Yale University Press.
Buhrmester, M., T. Kwang, and S. D. Gosling (2011, January). Amazon's Mechanical Turk A New
Source of Inexpensive, Yet High-Quality, Data? Perspectives on Psychological Science 6 (1), 3{5.
Bulovsky, A. (2018). Authoritarian communication on social media: The relationship between
89
democracy and leaders' digital communicative practices. International Communication Gazette,
1748048518767798.
Caldarelli, G. (2007). Scale-free networks: complex webs in nature and technology. Oxford Univer-
sity Press.
Cha, M., H. Haddadi, F. Benevenuto, P. K. Gummadi, et al. (2010). Measuring user in
uence in
twitter: The million follower fallacy. Icwsm 10(10-17), 30.
Cheibub, J. A., A. Przeworski, F. P. Limongi Neto, and M. M. Alvarez (1996). What makes
democracies endure? Journal of democracy 7 (1), 39{55.
Chittick, W. O., K. R. Billingsley, and R. Travis (1995). A three-dimensional model of American
foreign policy beliefs. International Studies Quarterly 39 (3), 313{331.
Cohrs, J. C., B. Moschner, J. Maes, and S. Kielmann (2005). The motivational bases of right-
wing authoritarianism and social dominance orientation: Relations to values and attitudes in the
aftermath of september 11, 2001. Personality and Social Psychology Bulletin 31 (10), 1425{1434.
Converse, P. E. (1964). The Nature of Belief Systems in Mass Publics. In Ideology and Discontent,
ed. David Apter. New York: Free Press.
Conybeare, J. A. (1992). A Portfolio Diversication Model of Alliances The Triple Alliance and
Triple Entente, 1879-1914. Journal of Con
ict Resolution 36 (1), 53{85.
Cranmer, S. J. and B. A. Desmarais (2016). A critique of dyadic design. International Studies
Quarterly 60 (2), 355{362.
Csermely, P., A. London, L.-Y. Wu, and B. Uzzi (2013). Structure and dynamics of core/periphery
networks. Journal of Complex Networks 1 (2), 93{123.
Dafoe, A., B. Zhang, and D. Caughey (2017). Information equivalence in survey experiments.
Working paper.
De Meo, P., E. Ferrara, G. Fiumara, and A. Provetti (2011). Generalized louvain method for
community detection in large networks. In Intelligent Systems Design and Applications (ISDA),
2011 11th International Conference on, pp. 88{93. IEEE.
90
DeLuca-McLean, D. and E. Castano (2009). Infra-humanization of ethnic minorities: The moder-
ating role of ideology. Basic and Applied Social Psychology 31 (2), 102{108.
Dickson, E. S. (2009). Do Participants and Observers Assess Intentions Dierently During Bar-
gaining and Con
ict? American Journal of Political Science 53 (4), 910{930.
Dixon, W. J. (1994). Democracy and the peaceful settlement of international con
ict. American
Political Science Review 88 (1), 14{32.
Donnelly, J. (2006). Sovereign inequalities and hierarchy in anarchy: American power and interna-
tional society. European Journal of International Relations 12 (2), 139{170.
Duckitt, J., C. Wagner, I. Du Plessis, and I. Birum (2002). The psychological bases of ideology and
prejudice: Testing a dual process model. Journal of personality and social psychology 83 (1), 75.
Duriez, B. and A. Van Hiel (2002). The march of modern fascism. a comparison of social dominance
orientation and authoritarianism. Personality and Individual Dierences 32 (7), 1199{1213.
Duriez, B. and A. Van Hiel (2002). The march of modern fascism. A comparison of social dominance
orientation and authoritarianism. Personality and Individual Dierences 32 (7), 1199{1213.
Ellemers, N., J. van der Toorn, Y. Paunov, and T. van Leeuwen (2019). The psychology of morality:
A review and analysis of empirical studies published from 1940 through 2017. Personality and
Social Psychology Review, 1088868318811759.
Enli, G. S. and E. Skogerb (2013). Personalized campaigns in party-centred politics: Twitter and
facebook as arenas for political communication. Information, Communication & Society 16 (5),
757{774.
Entman, R. M. (2004). Projections of Power: Framing News, Public Opinion, and U.S. Foreign
Policy. University of Chicago Press.
Erskine, H. G. (1963). The polls: Exposure to international information. The Public Opinion
Quarterly 27 (4), 658{662.
91
Esses, V. M., S. Medianu, and A. S. Lawson (2013). Uncertainty, threat, and the role of the media
in promoting the dehumanization of immigrants and refugees. Journal of Social Issues 69 (3),
518{536.
Esses, V. M., S. Veenvliet, G. Hodson, and L. Mihic (2008). Justice, morality, and the dehuman-
ization of refugees. Social Justice Research 21 (1), 4{25.
Federico, C. M., C. R. Weber, D. Ergun, and C. Hunt (2013). Mapping the connections between pol-
itics and morality: The multiple sociopolitical orientations involved in moral intuition. Political
Psychology 34 (4), 589{610.
Fiorina, M. P. (1981, October). Retrospective Voting in American National Elections. New Haven:
Yale Univ Pr.
Fiske, S. T. and S. E. Taylor (1991). Social Cognition. McGraw-Hill.
Free, L. and H. Cantril (1968). The Political Beliefs of Americans. New York: Simon & Schuster.
Friedland, N. and A. Merari (1985). The psychological impact of terrorism: A double-edged sword.
Political Psychology, 591{604.
Frimer, J., R. Boghrati, J. Haidt, J. Graham, and M. Dehgani (2019). Moral foundations dictionary
for linguistic analyses 2.0. Unpublished manuscript.
Garten, J., R. Boghrati, J. Hoover, K. M. Johnson, and M. Dehghani (2016). Morality between
the lines: Detecting moral sentiment in text. In Proceedings of IJCAI 2016 workshop on Com-
putational Modeling of Attitudes.
Gelpi, C., P. D. Feaver, and J. Rei
er (2006). Success matters: Casualty sensitivity and the war in
Iraq.
Gibson, J. L. (1992). Alternative measures of political tolerance: Must tolerance be" least-liked"?
American Journal of Political Science, 560{577.
Goh, K.-I. and A.-L. Barab asi (2008). Burstiness and memory in complex systems. EPL (Euro-
physics Letters) 81 (4), 48002.
92
Golbeck, J., J. M. Grimes, and A. Rogers (2010). Twitter use by the us congress. Journal of the
American Society for Information Science and Technology 61 (8), 1612{1621.
Goldenberg, J., N. He
ick, J. Vaes, M. Motyl, and J. Greenberg (2009). Of mice and men, and
objectied women: A terror management account of infrahumanization. Group Processes &
Intergroup Relations 12 (6), 763{776.
Goldstein, J. and R. O. Keohane (Eds.) (1993). Ideas and Foreign Policy: Beliefs, Institutions,
and Political Change. Cornell University Press.
Goyal, P., S. R. Chhetri, and A. Canedo (2018). dyngraph2vec: Capturing network dynamics using
dynamic graph representation learning. arXiv preprint arXiv:1809.02657 .
Graham, J., J. Haidt, and B. A. Nosek (2009). Liberals and conservatives rely on dierent sets of
moral foundations. Journal of Personality and Social Psychology 96 (5), 1029{1046.
Graham, J., B. A. Nosek, J. Haidt, R. Iyer, S. Koleva, and P. H. Ditto (2011). Mapping the moral
domain. Journal of personality and social psychology 101 (2), 366.
Grant, W. J., B. Moon, and J. Busby Grant (2010). Digital dialogue? australian politicians' use
of the social network tool twitter. Australian Journal of Political Science 45 (4), 579{604.
Grin, W. E. and Y. Oheneba-Sakyi (1993). Sociodemographic and political correlates of university
students' causal attributions for poverty. Psychological Reports.
Groll, E. (2015). These are the most in
uential world leaders on twitter (and why that doesn't
matter). Foreign Policy (April 28,2015). URL: https://foreignpolicy.com/2015/04/28/
these-are-the-most-influential-world-leaders-on-twitter-and-why-that-doesnt-matter/ .
Gronke, P., J. Koch, and J. M. Wilson (2003). Follow the leader? Presidential approval, presidential
support, and representatives' electoral fortunes. Journal of Politics 65 (3), 785{808.
Gunitsky, S. (2015). Corrupting the cyber-commons: Social media as a tool of autocratic stability.
Perspectives on Politics 13 (1), 42{54.
Haidt, J. and J. Graham (2007a). When morality opposes justice: Conservatives have moral
intuitions that liberals may not recognize. Social Justice Research 20 (1), 98{116.
93
Haidt, J. and J. Graham (2007b). When morality opposes justice: Conservatives have moral
intuitions that liberals may not recognize. Social Justice Research 20 (1), 98{116.
Hainmueller, J. (2012). Entropy balancing for causal eects: A multivariate reweighting method
to produce balanced samples in observational studies. Political Analysis 20 (1), 25{46.
Hainmueller, J. and Y. Xu (2013). Ebalance: A Stata package for entropy balancing.
Hall, R. B. (1997). Moral authority as a power resource. International Organization 51 (4), 591{622.
Hartley, T. and B. Russett (1992). Public opinion and the common defense: who governs military
spending in the United States? American political science review 86 (04), 905{915.
Haslam, N. and S. Loughnan (2014). Dehumanization and infrahumanization. Annual review of
psychology 65, 399{423.
Heider, F. (1958). The Psychology of Interpersonal Relations. Psychology Press.
Heradstveit, D. (1979). The Arab-Israeli con
ict: psychological obstacles to peace. Oslo: Norwegian
University Press.
Heradstveit, D. and G. M. Bonham (1996). Attribution theory and Arab images of the Gulf War.
Political Psychology, 271{292.
Herrmann, R. (1986, November). The Power of Perceptions in Foreign-Policy Decision Making:
Do Views of the Soviet Union Determine the Policy Choices of American Leaders? American
Journal of Political Science 30 (4), 841.
Herrmann, R. K. and M. P. Fischerkeller (1995). Beyond the enemy image and spiral model:
cognitive{strategic research after the cold war. International Organization 49 (03), 415{450.
Herrmann, R. K., P. E. Tetlock, and P. S. Visser (1999). Mass public decisions on go to war: A
cognitive-interactionist framework. American Political Science Review 93 (3), 553{573.
Herrmann, R. K., J. F. Voss, T. Y. E. Schooler, and J. Ciarrochi (1997, September). Images in
International Relations: An Experimental Test of Cognitive Schemata. International Studies
Quarterly 41 (3), 403{433.
94
Hibbing, J. R., K. B. Smith, and J. R. Alford (2014). Dierences in negativity bias underlie
variations in political ideology.
Hildebrandt, T., C. Hillebrecht, P. M. Holm, and J. Pevehouse (2013). The domestic politics of
humanitarian intervention: Public opinion, partisanship, and ideology. Foreign Policy Analy-
sis 9 (3), 243{266.
Hirshberg, M. S. (1993). The Self-Perpetuating National Self-Image: Cognitive Biases in Percep-
tions of International Interventions. Political Psychology 14 (1), 77{98.
Hodson, G. and K. Costello (2007). Interpersonal disgust, ideological orientations, and dehuman-
ization as predictors of intergroup attitudes. Psychological science 18 (8), 691{698.
Holsti, O. R. (1967). Cognitive Dynamics and Images of the Enemy. Journal of International
Aairs 21 (1), 16{39.
Holsti, O. R. (1992, December). Public Opinion and Foreign Policy: Challenges to the Almond-
Lippmann Consensus Mershon Series: Research Programs and Debates. International Studies
Quarterly 36 (4), 439.
Holsti, O. R. (2004). Public opinion and American foreign policy. University of Michigan Press.
Huberty, C. J. and S. Olejnik (2006). Applied MANOVA and discriminant analysis, Volume 498.
John Wiley & Sons.
Huddy, L., S. Feldman, C. Taber, and G. Lahav (2005). Threat, anxiety, and support of antiter-
rorism policies. American journal of political science 49 (3), 593{608.
Imai, K., L. Keele, D. Tingley, and T. Yamamoto (2011). Unpacking the black box of causal-
ity: Learning about causal mechanisms from experimental and observational studies. American
Political Science Review 105 (4), 765{789.
Jackson, L. E. and L. Gaertner (2010). Mechanisms of moral disengagement and their dierential
use by right-wing authoritarianism and social dominance orientation in support of war. Aggressive
behavior 36(4), 238{250.
95
Jackson, R. (1993). The Weight of Ideas in Decolonization: Normative Change in International Re-
lations. In J. Goldstein and R. O. Keohane (Eds.), Ideas and Foreign Policy: Beliefs, Institutions,
and Political Change, pp. 111{138. Cornell University Press.
Jaggers, K. and T. R. Gurr (1995). Tracking democracy's third wave with the polity iii data.
Journal of peace research 32 (4), 469{482.
Jano-Bulman, R. (2009). To provide or protect: Motivational bases of political liberalism and
conservatism. Psychological Inquiry 20 (2-3), 120{128.
Jentleson, B. W. (1992). The pretty prudent public: Post post-vietnam american opinion on the
use of military force. International studies quarterly 36 (1), 49{74.
Jentleson, B. W. and R. L. Britton (1998). Still pretty prudent: Post-cold war american public
opinion on the use of military force. Journal of Con
ict Resolution 42 (4), 395{417.
Jones, E. and K. Davis (1965). From Acts to Dispositions: The Attribution Process in Person
Perception. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology, Volume 2, pp.
219{266. New York: Academic Press.
Jones, E. E. and R. E. Nisbett (1971). The actor and the observer: Divergent perceptions of the
causes of behavior.
Jost, J. T. (2006). The end of the end of ideology. American Psychologist 61 (7), 651.
Jost, J. T., C. M. Federico, and J. L. Napier (2009a). Political ideology: Its structure, functions,
and elective anities. Annual review of psychology 60, 307{337.
Jost, J. T., C. M. Federico, and J. L. Napier (2009b, January). Political Ideology: Its Structure,
Functions, and Elective Anities. Annual Review of Psychology 60 (1), 307{337.
Jost, J. T., J. Glaser, A. W. Kruglanski, and F. J. Sulloway (2003). Political conservatism as
motivated social cognition. Psychological bulletin 129 (3), 339.
Katz, E., P. F. Lazarsfeld, and E. Roper (1955). Personal in
uence: The part played by people in
the
ow of mass communications. New York: Free Press.
96
Kerlinger, F. N. (1984). Liberalism and conservatism: The nature and structure of social attitudes,
Volume 1. Lawrence Erlbaum Assoc Incorporated.
Kertzer, J. D., K. E. Powers, B. C. Rathbun, and R. Iyer (2014, July). Moral Support: How Moral
Values Shape Foreign Policy Attitudes. The Journal of Politics 76 (3), 825{840.
Kertzer, J. D. and D. Tingley (2018). Political psychology in international relations: beyond the
paradigms. Annual Review of Political Science 21, 319{339.
Kinder, D. R. and D. O. Sears (1981). Prejudice and politics: Symbolic racism versus racial threats
to the good life. Journal of personality and social psychology 40 (3), 414.
Kosterman, R. and S. Feshbach (1989). Toward a measure of patriotic and nationalistic attitudes.
Political psychology, 257{274.
Lake, D. A. (1999). Entangling relations: American foreign policy in its century. Princeton studies
in international history and politics. Princeton, N.J: Princeton University Press.
Lake, D. A. (2013). Legitimating power: The domestic politics of us international hierarchy.
International Security 38 (2), 74{111.
Larson, D. W. (1985). Origins of Containment: A Psychological Explanation. Princeton University
Press.
Lerner, J. S., J. H. Goldberg, and P. E. Tetlock (1998). Sober second thought: The eects of
accountability, anger, and authoritarianism on attributions of responsibility. Personality and
Social Psychology Bulletin 24 (6), 563{574.
Lindberg, S. I., M. Coppedge, J. Gerring, and J. Teorell (2014). V-dem: A new way to measure
democracy. Journal of Democracy 25 (3), 159{169.
Lippmann, W. (1955). Essays in the public philosophy. Transaction Publishers.
Maoz, I. and C. McCauley (2008). Threat, dehumanization, and support for retaliatory aggressive
policies in asymmetric con
ict. Journal of Con
ict Resolution 52 (1), 93{116.
Maoz, Z. and B. Russett (1993a). Normative and structural causes of democratic peace, 1946{1986.
American Political Science Review 87 (3), 624{638.
97
Maoz, Z. and B. Russett (1993b). Normative and Structural Causes of Democratic Peace,
1946{1986. American Political Science Review 87 (03), 624{638.
Marcus, G. E., J. L. Sullivan, E. Theiss-Morse, and D. Stevens (2005). The emotional foundation
of political cognition: The impact of extrinsic anxiety on the formation of political tolerance
judgments. Political Psychology 26 (6), 949{963.
Marshall, M. G. and K. Jaggers (2002). Polity iv project: Political regime characteristics and
transitions, 1800-2002.
McAlister, A. L., A. Bandura, and S. V. Owen (2006). Mechanisms of moral disengagement in
support of military force: The impact of sept. 11. Journal of Social and Clinical Psychology 25 (2),
141{165.
McDermott, R. (2011). New directions for experimental work in international relations. Interna-
tional Studies Quarterly 55 (2), 503{520.
McDermott, R., D. Tingley, J. Cowden, G. Frazzetto, and D. D. Johnson (2009). Monoamine
oxidase a gene (maoa) predicts behavioral aggression following provocation. Proceedings of the
National Academy of Sciences 106 (7), 2118{2123.
McLaren, L. M. (2003). Anti-immigrant prejudice in europe: Contact, threat perception, and
preferences for the exclusion of migrants. Social forces 81 (3), 909{936.
Mercer, J. (1995). Anarchy and identity. International Organization 49 (02), 229{252.
Mercer, J. (1996). Reputation and international politics. Cornell University Press.
Morgan, T. C. and S. H. Campbell (1991). Domestic Structure, Decisional Constraints, and War
So Why Kant Democracies Fight? Journal of Con
ict Resolution 35 (2), 187{211.
Morrow, J. D. (1994). Alliances, credibility, and peacetime costs. Journal of Con
ict Resolu-
tion 38(2), 270{297.
Muchnik, L., S. Pei, L. C. Parra, S. D. Reis, J. S. Andrade Jr, S. Havlin, and H. A. Makse (2013).
Origins of power-law degree distribution in the heterogeneity of human activity in social networks.
Scientic reports 3, 1783.
98
Mueller, J. E. (1973). War, Presidents, and Public Opinion. Joh Wiley.
Mullen, E. and L. J. Skitka (2006). Exploring the psychological underpinnings of the moral mandate
eect: Motivated reasoning, group dierentiation, or anger? Journal of personality and social
psychology 90(4), 629.
Munger, K., R. Bonneau, J. Nagler, and J. A. Tucker (2018). Elites tweet to get feet o the streets:
Measuring regime social media strategies during protest. Political Science Research and Methods,
1{20.
Nam, H. H., J. T. Jost, and J. J. Van Bavel (2013). \not for all the tea in china!" political ideology
and the avoidance of dissonance-arousing situations. PloS one 8(4), e59837.
Newman, M. J, girvan m (2004) nding and evaluating community structure in networks. Physical
Review E 69(2), 026113.
Newman, M. (2018). Networks. Oxford university press.
Newman, M. E. (2002). Assortative mixing in networks. Physical review letters 89 (20), 208701.
Newman, M. E. (2006). Modularity and community structure in networks. Proceedings of the
national academy of sciences 103 (23), 8577{8582.
Nye, J. S. (2004). Soft power: The means to success in world politics. Public aairs.
Onraet, E., A. Van Hiel, and K. Dhont (2013). The relationship between right-wing ideological
attitudes and psychological well-being. Personality and Social Psychology Bulletin 39 (4), 509{
522.
Opotow, S. (1990). Moral exclusion and injustice: An introduction. Journal of social issues 46 (1),
1{20.
Oxley, D. R., K. B. Smith, J. R. Alford, M. V. Hibbing, J. L. Miller, M. Scalora, P. K. Hatemi,
and J. R. Hibbing (2008). Political attitudes vary with physiological traits. science 321(5896),
1667{1670.
Page, B. I. and R. Y. Shapiro (1983, March). Eects of Public Opinion on Policy. The American
Political Science Review 77 (1), 175{190.
99
Page, B. I. and R. Y. Shapiro (1992, March). The Rational Public: Fifty Years of Trends in
Americans' Policy Preferences (1 edition ed.). Chicago: University Of Chicago Press.
Paolacci, G. and J. Chandler (2014). Inside the Turk: Understanding Mechanical Turk as a par-
ticipant pool. Current Directions in Psychological Science 23 (3), 184{188.
Pearce, K. E. (2015). Democratizing kompromat: The aordances of social media for state-
sponsored harassment. Information, Communication & Society 18 (10), 1158{1174.
Pettigrew, T. F. (1979, January). The Ultimate Attribution Error: Extending Allport's Cognitive
Analysis of Prejudice. Personality and Social Psychology Bulletin 5 (4), 461{476.
Pratto, F., J. Sidanius, L. M. Stallworth, and B. F. Malle (1994). Social dominance orientation:
A personality variable predicting social and political attitudes. Journal of personality and social
psychology 67(4), 741.
Pronin, E., K. Kennedy, and S. Butsch (2006). Bombing versus negotiating: How preferences for
combating terrorism are aected by perceived terrorist rationality. Basic and Applied Social
Psychology 28 (4), 385{392.
Rathbun, B. C. (2007). Uncertain about uncertainty: understanding the multiple meanings of a
crucial concept in international relations theory. International Studies Quarterly 51 (3), 533{557.
Rathbun, B. C., J. D. Kertzer, J. Rei
er, P. Goren, and T. J. Scotto (2016). Taking foreign policy
personally: Personal values and foreign policy attitudes. International Studies Quarterly 60 (1),
124{137.
Reeder, G. D., S. Kumar, M. S. Hesson-McInnis, and D. Tramow (2002). Inferences about the
morality of an aggressor: the role of perceived motive. Journal of personality and social psychol-
ogy 83(4), 789.
Rei
er, J., T. J. Scotto, and H. D. Clarke (2011). Foreign Policy Beliefs in Contemporary Britain:
Structure and Relevance1. International Studies Quarterly 55 (1), 245{266.
Renshon, J. (2016). Status decits and war. International Organization 70 (3), 513{550.
Riddell, P. (2004). Hug Them Close: Blair, Clinton, Bush and the 'special Relationship'. Politico's.
100
Risse-Kappen, T. (1995a). Cooperation Among Democracies: The European In
uence on U.S.
Foreign Policy. Princeton University Press.
Risse-Kappen, T. (1995b). Cooperation among democracies: The European in
uence on US foreign
policy. Princeton University Press.
Rogers, E. M. (1962). Diusion of innovations. Free Press.
Russett, B. (1990, December). Controlling the Sword: The Democratic Governance of National
Security (1St Edition edition ed.). Cambridge, Mass.: Harvard University Press.
Sadler, M. S., M. Lineberger, J. Correll, and B. Park (2005). Emotions, attributions, and policy
endorsement in response to the September 11th terrorist attacks. Basic and Applied Social
Psychology 27 (3), 249{258.
Scheufele, D. A. and D. Tewksbury (2006). Framing, agenda setting, and priming: The evolution
of three media eects models. Journal of communication 57 (1), 9{20.
Schmitter, P. C. and T. L. Karl (1991). What democracy is... and is not. Journal of democracy 2 (3),
75{88.
Shapiro, R. Y. and B. I. Page (1988). Foreign policy and the rational public. Journal of Con
ict
Resolution 32 (2), 211{247.
Shimko, K. L. (1991). Images and Arms Control: Perceptions of the Soviet Union in the Reagan
Administration. University of Michigan Press.
Sidanius, J. and F. Pratto (2001a). Social dominance: An intergroup theory of social hierarchy and
oppression. Cambridge University Press.
Sidanius, J. and F. Pratto (2001b, February). Social Dominance: An Intergroup Theory of Social
Hierarchy and Oppression. Cambridge University Press.
Skitka, L. J. and D. C. Wisneski (2011). Moral conviction and emotion. Emotion Review 3 (3),
328{330.
101
Smith, D. (2017). Donald trump attacks theresa may over her criticism of his far-right retweets.
The Guardian (30 November 2017). URL: https://www.theguardian.com/us-news/2017/
nov/29/donald-trump-theresa-may-tweet-uk-us .
Sobel, R. (2001, March). The Impact of Public Opinion on U.S. Foreign Policy since Vietnam (1
edition ed.). New York: Oxford University Press.
Staub, E. (1989). The roots of evil: The origins of genocide and other group violence. Cambridge
University Press.
Staub, E. (1990). Moral exclusion, personal goal theory, and extreme destructiveness. Journal of
Social Issues 46 (1), 47{64.
Stephan, W. G., C. W. Stephan, and W. B. Gudykunst (1999). Anxiety in intergroup relations:
A comparison of anxiety/uncertainty management theory and integrated threat theory. Interna-
tional Journal of Intercultural Relations 23 (4), 613{628.
Strau, N., S. Kruikemeier, H. van der Meulen, and G. van Noort (2015). Digital diplomacy in gcc
countries: Strategic communication of western embassies on twitter. Government Information
Quarterly 32 (4), 369{379.
Su, S. and M. Xu (2015). Twitplomacy: Social media as a new platform for development of public
diplomacy. International Journal of E-Politics (IJEP) 6 (1), 16{29.
Tajfel, H. and J. C. Turner (1979). An integrative theory of intergroup con
ict. The social psy-
chology of intergroup relations 33 (47), 74.
Tetlock, P. E. (1998). Social psychology and world politics. Handbook of social psychology 4,
868{914.
Tingley, D., T. Yamamoto, K. Hirose, L. Keele, and K. Imai (2014). Mediation: R package for
causal mediation analysis.
Tomz, M. R. and J. L. P. Weeks (2013, November). Public Opinion and the Democratic Peace.
American Political Science Review 107 (04), 849{865.
102
Vaes, J., N. A. He
ick, and J. L. Goldenberg (2010). \we are people": Ingroup humanization as
an existential defense. Journal of personality and social psychology 98 (5), 750.
Van Hiel, A., B. Duriez, and M. Kossowska (2006). The presence of left-wing authoritarianism
in western europe and its relationship with conservative ideology. Political Psychology 27 (5),
769{793.
Walgrave, S. and P. Van Aelst (2006). The contingency of the mass media's political agenda setting
power: Toward a preliminary theory. Journal of communication 56 (1), 88{109.
Williams, S. (1984). Left-right ideological dierences in blaming victims. Political Psychology,
573{581.
Wittkopf, E. R. (1990). Faces of Internationalism: Public Opinion and American Foreign Policy.
Duke University Press.
Zeitzo, T. (2018). Does social media in
uence con
ict? evidence from the 2012 gaza con
ict.
Journal of Con
ict Resolution 62 (1), 29{63.
Zucker, G. S. and B. Weiner (1993). Conservatism and perceptions of poverty: An attributional
analysis. Journal of Applied Social Psychology 23 (12), 925{943.
103
Appendices
104
Appendix A
Appendix for Chapter 2
A.1 Scenario
Scenario
A civilian airplane was shot down by country A. [Control/Nonally/Ally 1/Ally 2/Ally 3/ Strong
Ally description.] All 230 passengers and crew aboard were killed. Some experts believe that the
technological malfunction of the radar was at fault as the plane was misidentied as a ghter
plane. In this case, the situation was a terrible misfortune largely out of control of country A.
However,other experts claim that the civil air defense knew that the airplane was civilian, yet tar-
geted it nevertheless as an act of provocation. In this case the shooting down of the plane was an
intentionally performed and unlawful act.
Treatment groups:
1. Control: no information about Country A.
2. Nonally: Country A is not an ally of the United States.
3. Ally 1: Country A has a common enemy with the United States.
4. Ally 2: Country A has a common enemy with the United States and shared experience of
ghting alongside the United States in previous con
icts.
5. Ally 3: Country A has a common enemy with the United States and shared experience of
105
ghting alongside the United States in previous con
icts. It is a democracy and shows every
sign that it will remain a democracy.
1
6. Strong ally: Country A has a common enemy with the United States and shared experience
of ghting alongside the United States in previous con
icts. It is a democracy and shows
every sign that it will remain a democracy. It is a Western country.
In your opinion, what is the most plausible cause for the incident?
(A) Technological malfunction
(B) Intentional attack
How condent are you about that?
1. Strongly uncondent
2. Somewhat uncondent
3. Neither condent nor uncondent
4. Somewhat condent
5. Strongly condent
Policy preference
Having identied the plausible cause(s) for the incident, participants will be asked to identify the
most appropriate response to the incident:
(A) Do nothing
(B) Demand immediate investigation of the incident
(C) Impose economic sanctions
(D) Use military force
1
This vignette follows the democracy treatment used by Tomz and Weeks (2013).
106
A.2 Foreign Policy Attitudes
All items have ve-point Likert response items ranging from \Strongly disagree" to \Strongly
agree". To mitigate survey response eects, all 16 items were presented to participants in random
order.
Cooperative internationalism ( = 0:87)
1. The United States needs to cooperate more with the United Nations.
2. It is essential for the United States to work with other nations to solve problems such as
overpopulation, hunger, and pollution.
3. Promoting and defending human rights in other countries is of utmost importance.
4. Helping to improve the standard of living is less developed countries is of utmost importance.
5. Protecting the global environment is of utmost importance.
Militant internationalism ( = 0:78)
1. The United States should take all steps including the use of force to prevent aggression by
any expansionist power
2. Rather than simply countering our opponents' thrusts, it is necessary to strike at the heart
of an opponent's power.
3. Going to war is unfortunate but sometimes the only solution to international problems.
4. There is considerable validity in the domino theory that when one nation falls to communism,
others nearby will soon follow a similar path.
5. American military strength is not the best way to ensure world peace. Reverse-coded
6. The United States must demonstrate its resolve so that others do not take advantage of it.
Isolationism ( = 0:85)
1. The U.S. should mind its own business internationally and let other countries get along the
best they can on their own.
107
2. We should not think so much in international terms but concentrate more on our own national
problems.
3. The U.S. needs to play an active role in solving con
icts around the world. Reverse-coded
4. America's conception of its leadership role in the world must be scaled down.
5. Our allies are perfectly capable of defending themselves and they can aord it, thus allowing
the United States to focus on internal rather than external threats to its well-being.
108
A.3 Saturated Interaction Model
Table A.1: Saturated Interaction Model
Dispositional Attribution
Model 1
Nonally 0.264 (0.326)
Ally 1 0.199 (0.319)
Ally 2 0.491 (0.317)
Ally 3 0.461 (0.333)
Strong ally 0.614

(0.329)
Age:25-44 0.022 (0.169)
Age:45-64 0.523

(0.209)
Age:65+ 0.246 (0.369)
White 0.180 (0.135)
Male 0.037 (0.118)
High school 0.012 (0.857)
Some college 0.146 (0.837)
College degree 0.038 (0.834)
Graduate school 0.179 (0.841)
Party ID 0.436 (0.323)
Ideology 1.171

(0.551)
Nonally x Ideology 0.880 (0.702)
Ally1 x Ideology 0.303 (0.697)
Ally2 x Ideology 0.548 (0.677)
Ally3 x Ideology 1.233

(0.684)
Strong ally x Ideology 1.307

(0.698)
Constant 0.501 (0.866)
N 1,354

p < .1;

p < .05;

p < .01
109
A.4 Eect of Ideology on Attributions
Figure A.1: Conditional Eect of Ideology on Attributions
110
A.5 Replication of results for attentive respondents only
Table A.2 replicates Table 2.3 in the main text using the sample that excludes the respondents
who failed the attention check. The table shows that the results do not substantively change when
inattentive respondents are excluded.
Table A.2: Eect of Ideology on Attributions (Attentive Respondents Only)
Dispositional Attribution
Model 1 Model 2
(1) (2)
Nonally 0.105 (0.202) 0.276 (0.339)
Ally 1 0.232 (0.200) 0.054 (0.327)
Ally 2 0.652

(0.200) 0.423 (0.328)
Ally 3 0.961

(0.203) 0.328 (0.342)
Strong ally 1.061

(0.205) 0.616

(0.337)
Age:25-44 0.026 (0.179) 0.011 (0.174)
Age:45-64 0.562

(0.216) 0.560

(0.211)
Age:65+ 0.315 (0.366) 0.350 (0.365)
White 0.260

(0.142)
Male 0.018 (0.122)
High school 0.345 (0.949)
Some college 0.510 (0.931)
College degree 0.459 (0.929)
Graduate school 0.534 (0.935)
Party ID 0.282 (0.355)
Ideology 0.638

(0.359) 0.906

(0.495)
Nonally x Ideology 0.881 (0.728)
Ally1 x Ideology 0.529 (0.710)
Ally2 x Ideology 0.644 (0.696)
Ally3 x Ideology 1.583

(0.700)
Strong ally x Ideology 1.195

(0.709)
Constant 1.042 (0.949) 0.191 (0.267)
N 1,253 1,253

p < .1;

p < .05;

p < .01
111
A.6 Survey Weighting
Table A.3 shows that my sample is younger and more educated than the general population (as
compared to the 2010 U.S. census). To make sure that these dierences do not aect generaliz-
ability of my ndings, I employ entropy balancing technique. I use the entropy package in Stata to
reweight my dataset to more closely match demographic characteristics from the national popula-
tion (Hainmueller and Xu, 2013). Following Kertzer et al. (2014), I trim the weight to reduce the
impact of extreme values.
2
Table A.3: Survey sample characteristics
Characteristic Adult Population Unweighted Sample Weighted Sample
Male 0.492 0.588 0.547
Age 18- 24 0.130 0.143 0.137
Age 25-44 0.350 0.643 0.376
Age 45- 64 0.347 0.182 0.338
Age 65+ 0.171 0.032 0.149
Less than High School 0.133 0.004 0.027
High School 0.304 0.073 0.275
Some College/University 0.196 0.294 0.242
College/University 0.271 0.435 0.335
Some Graduate/Graduate Degree 0.096 0.192 0.119
Note: Weights are trimmed at95th percentile.
Table A.4 replicates Table 2.3 in the main text using the weighted sample. As we can see, the
key results do not change substantively when the weights are introduced. The coecients for Ally
2, Ally 3, and Strong Ally are still negative and signicant, indicating that the respondents in these
experimental groups had lower propensity of inferring intentional attack. The only dierence is
that, compared to the unweighted sample, the direct eect of ideology in the weighted sample is
not statistically signicant; however, the hypothesized interaction eect still holds in Model 2.
Table A.5 replicates Table 2.4 in the main text using the weighted sample. It shows that the
eect of attributions on policy preferences stays the same when the weights are introduced. The
coecient for Dispositional Attribution is still positive and signicant, indicating that the respon-
dents who made dispositional attributions were signicantly more likely to advocate harsh response
2
Weights were trimmed with values above 3, which is slightly above 95th percentile. Sample with trimmed
weights is slightly less representative than a sample with untrimmed weights; however, trimming prevents extreme
observations from exerting too much leverage on the results (Kertzer et al., 2014).
112
to the incident.
Table A.4: Eect of Ideology on Attributions (weighted)
Cause
Model 1 Model 2
(1) (2)
Nonally 0.256 (0.235) 0.195 (0.412)
Ally 1 0.291 (0.227) 0.452 (0.392)
Ally 2 0.916

(0.228) 0.746

(0.389)
Ally 3 0.987

(0.229) 0.257 (0.410)
Strong ally 1.166

(0.236) 0.734

(0.405)
Age:25-44 0.019 (0.205) 0.007 (0.202)
Age:45-64 0.478

(0.219) 0.466

(0.212)
Age:65+ 0.222 (0.276) 0.281 (0.266)
White 0.124 (0.171)
Male 0.078 (0.141)
High school 0.028 (0.513)
Some college 0.242 (0.510)
College degree 0.028 (0.504)
Graduate school 0.283 (0.532)
Party ID 0.040 (0.365)
Ideology 0.166 (0.381) 0.507 (0.556)
Nonally x Ideology 1.051 (0.831)
Ally1 x Ideology 0.307 (0.753)
Ally2 x Ideology 0.439 (0.753)
Ally3 x Ideology 1.651

(0.789)
Strong ally x Ideology 1.017

(0.770)
Constant 0.638 (0.548) 0.391 (0.316)
N 1,354 1,354

p < .1;

p < .05;

p < .01
Notes: Table displays logit coecients; standard errors in parenthesis.
Dependent variable is coded 1 for dispositional attribution and 0 for
situational attribution. The reference categories are country identity:
control group, 18- 24 years old, non-white, female, less than high school.
All non-dichotomous measures have been rescaled from 0 to 1. The
models are weighted to match general population. Weights are trimmed
at 95th percentile.
113
Table A.5: Eect of Attributions on Policy Prefer-
ences (weighted)
Harsh Response
Dispositional Attribution 1.914

(0.210)
Age:25-44 0.325
(0.255)
Age:45-64 0.455
(0.292)
Age:65+ 0.222
(0.368)
White 0.259
(0.218)
Male 0.215
(0.194)
High school 1.159

(0.606)
Some college 1.497

(0.610)
College degree 1.105

(0.597)
Graduate school 1.724

(0.659)
Party ID 0.242
(0.485)
Ideology 0.231
(0.531)
MI 2.274

(0.596)
CI 0.804
(0.516)
Isolationism 0.317
(0.489)
Constant 2.885

(0.869)
N 1,354

p < .1;

p < .05;

p < .01
Notes: Table displays logit coecients; standard
errors in parenthesis. Dependent variable is coded
1 for hard response and 0 for soft response. The ref-
erence categories are situational attribution, 18-24
years old, non-white, female, less than high school.
All non-dichotomous measures have been rescaled
from 0 to 1. The model is weighted to match gen-
eral population. Weights are trimmed at 95th
percentile.
114
A.7 Robustness Check: Controlling for Foreign Policy Orienta-
tions
Table A.6: Eect of Ideology on Attributions
Dispositional Attribution
Model 1 Model 2
(1) (2)
Nonally 0.081 (0.197) 0.284 (0.326)
Ally 1 0.328

(0.194) 0.239 (0.320)
Ally 2 0.724

(0.195) 0.567

(0.317)
Ally 3 0.991

(0.198) 0.500 (0.334)
Strong ally 1.143

(0.200) 0.639

(0.330)
Age:25-44 0.040 (0.169) 0.028 (0.165)
Age:45-64 0.521

(0.209) 0.524

(0.204)
Age:65+ 0.269 (0.364) 0.308 (0.365)
White 0.161 (0.135)
Male 0.016 (0.118)
High school 0.270 (0.888)
Some college 0.435 (0.868)
College degree 0.330 (0.866)
Graduate school 0.463 (0.872)
Party ID 0.434 (0.328)
Ideology 0.631

(0.346) 0.748 (0.518)
MI 0.908

(0.357) 0.875

(0.354)
CI 0.398 (0.344) 0.497 (0.338)
Isolationism 0.450 (0.311) 0.401 (0.309)
Nonally x Ideology 0.862 (0.703)
Ally1 x Ideology 0.270 (0.698)
Ally2 x Ideology 0.479 (0.677)
Ally3 x Ideology 1.237

(0.687)
Strong ally x Ideology 1.317

(0.699)
Constant 0.041 (0.946) 0.763 (0.500)
N 1,354 1,354

p < .1;

p < .05;

p < .01
Notes: Table displays logit coecients; standard errors in parenthesis.
Dependent variable is coded 1 for dispositional attribution and 0 for
situational attribution. The reference categories are country identity:
control group, 18- 24 years old, non-white, female, less than high school.
All non-dichotomous measures have been rescaled from 0 to 1.
115
A.8 Nonparametric Mediation Analysis
Following Kertzer (2014), in order to check for the presence of post-treatment bias that might
erroneously suppress true eect sizes of ideology, I performed a series a nonparametric mediation
analyses (Imai et al., 2011). Using mediation package in R (Tingley et al., 2014), I check whether
the eect of ideology on causal attribution is partially mediated by CI, MI, and isolationism. I
estimate three mediation models, in which the impact of political ideology X
i
on Y is mediated
by foreign policy orientations, controlling for the other Xs as pretreatment covariates along with
the demographic characteristics from the analyzed models. Figure A.2 plots three quantities of
interest: direct aect of X
i
on Y (ADE) represents all mechanisms through which X
i
might aect
Y except from mediating mechanism M; average causal mediation eect (ACME) represents the
eect of ideology on attributions channeled through foreign policy orientations; and the total eect
which refers to the sum of direct and mediated eects.
Figure presents the mediating eect of ideology on MI, CI, and isolationism. Mediation eect
(ACME) reaches statistical signicance in top left panel only, indicating that the eect of ideology
on causal attributions is partially transmitted by militant internationalism: 18% of ideology is
transmitted through MI with the remaining 82% transmitted through other mechanisms. Two other
panels show that ideology is not channeled through cooperative internationalism or isolationism.
Therefore, smaller eect of ideology in Table A.6 of Appendix A.7 could occur due to the slight
post-treatment bias, as the eect of ideology is partially mediated by MI when both are included
in the same model.
116
Figure A.2: Mediating Eect of Ideology on Causal attributions
117
Appendix B
Appendix for Chapter 3
B.1 Survey Instrument
Participants read the following brief prompt after responding to a series of demographic questions:
\We would like you to imagine some hypothetical scenarios in international aairs and give us your
thoughts about them. Please read these summaries carefully." Then they received the following
vignette with one of the bold elements for each manipulation:
Imagine a country with a major regional rival is dramatically increasing the size of its military,
including a nuclear weapons program. However, it is thought that its nuclear weapons [cannot
be used oensively since its main adversary also has nuclear weapons/could be used
oensively since its main adversary does not have nuclear weapons]. This country [has
been engaged in con
icts with that rival in the past and has taken part of its rival's
territory/has never fought its rival in the past]. The country has had [poor relations/good
relations] with the United States.
All following items were slider questions in which respondents drag a slider handle to indicate
their preference level. Every question ranged from 0 to 10 with extreme responses provided.
What would you expect the human rights record of this country to be with 0 being very poor
and 10 being very good?
118
Do you expect this country to be undemocratic or democratic with 0 being completely undemo-
cratic and 10 being completely democratic?
How threatening do you think this country is with 0 being completely harmless and 10 being
extremely dangerous?
Based on what you know about this country, please give your estimate of whether its leaders
have the following characteristics, with larger numbers indicating higher levels of that trait?
(A) Trustworthy
(B) Aggressive
(C) Greedy
(D) Resolute
B.2 Survey Weighting
Table B.1 compares the demographics of the American population with the demographics of our
survey sample from MTurk, which is slightly more male, white, and more educated than the census
breakdown. We employ entropy balance weighting using the entropy package in Stata to generate
post-stratication weights for gender, age, education, and race (Hainmueller, 2012, Hainmueller
and Xu, 2013). To reduce the impact of extreme values, we trimmed weights greater than 5, a
value slightly below 99th percentile.
1
1
The sample values based on trimmed weights are slightly less representative than a sample with untrimmed
weights; however, trimming prevents extreme observations from exerting too much leverage on the results (?).
119
Table B.1: Survey sample characteristics
Characteristic Adult Population Unweighted Sample Weighted Sample
Male 0.492 0.547 0.517
Age 18-24 0.130 0.075 0.121
Age 25-44 0.350 0.721 0.386
Age 45-64 0.347 0.172 0.359
Age 65+ 0.171 0.032 0.134
White 0.637 0.777 0.648
High School or less 0.399 0.118 0.347
Some College 0.189 0.244 0.208
2-year College 0.098 0.134 0.099
4-year College 0.200 0.384 0.220
Graduate Degree 0.114 0.120 0.125
Note: Weights are trimmed at99th percentile.
120
Appendix C
Appendix for Chapter 4
C.1 Examples of Mention Tweets
Figures C.1 and C.2 show examples of mention tweets of six categories described in the main text.
Figure C.3 shows 200 most common words used when world leaders mention each other. It shows
that leaders mostly mention each other in the context of diplomatic visits and bilateral cooperation.
121
Figure C.1: Examples of Mentions by World Leaders
(a) State visits (b) Congratulating on national holidays
(c) Expressing condolences
122
Figure C.2: Examples of Mentions by World Leaders
(a) Personal messages (b) Support for policy decisions
(c) Criticism
123
Figure C.3: Word Cloud of 200 Most Common Words in Mention Tweets
124
C.2 Robustness Checks
In this section, we explore the robustness of our main ndings. The rst model of Table C.1 shows
that the alternative measure of income level as log GDP yields similar result: the variable is not
statistically signicant. Models 2 and 3 probe the robustness of our main nding that democratic
leaders have higher levels of network centrality. in Model 2 we use the continuous version of Polity
IV's Polity 2 scale. The signicant and positive coecient of the variable indicates that the leaders
of more democratic countries are more likely to be central in the mention network. Finally, in
Model 3 we use the electoral democracy index as measured by the Varieties of Democracies Project
(V-Dem). The advantage of this project from Polity IV and others is that it includes multiples
varieties of democracy: electoral, liberal, participatory, deliberative, and egalitarian (Lindberg
et al., 2014). Considering that our hypothesis is mostly based on the idea of electoral pressure in
democracies, we use the electoral component to check robustness of our main ndings. The highly
signicant and positive coecient of electoral democracy in Model 3 (the variable ranges from 0
for less democratic to 1 for more democratic) provides condent support for our previous results.
125
Table C.1: Alternative Specications of the Model
Network Centrality (Top 20%)
(1) (2) (3)
Account age 3.276

3.164

3.427

(0.734) (0.740) (0.729)
Statuses count 2.554

2.828

2.837

(1.284) (1.325) (1.310)
Burstiness 0.635 0.242 0.267
(2.627) (2.602) (2.688)
Own language 0.874

0.856

0.840

(0.294) (0.298) (0.295)
Actor: Head of State 0.542 0.643 0.627
(0.388) (0.394) (0.392)
Actor: MFA 1.491

1.595

1.527

(0.369) (0.376) (0.369)
Personal account 2.551

2.652

2.585

(0.338) (0.349) (0.341)
Internet users 1.421 1.164 1.087
(1.143) (0.975) (0.984)
Log GDP 0.146
(0.255)
Democracy (dummy) 0.903

(0.335)
Middle income 0.169 0.077
(0.465) (0.468)
High income 0.687 0.512
(0.636) (0.652)
Polity IV score 1.767

(0.565)
Electoral democracy 2.571

(0.845)
Constant 6.573

7.620

8.251

(2.203) (1.991) (2.115)
N 459 459 476

p < .05;

p < .01
Notes: Table displays logit coecients, standard errors in paren-
theses. All non-dichotomous measures have been rescaled from 0
to 1.
C.3 Retweet Network results
In this section, we present our ndings for the leaders' retweet network and show that the patterns of
communications between leaders on these two networks are remarkably similar. Table C.2 displays
126
Table C.2: Retweet Network Assortativity Coecients
Feature Assortativity Coecient
Type of actor 0.38
Geographic region 0.38
Regime 0.05
Population -0.01
Income level 0.10
Internet access 0.16
a list of attributes of the leaders' accounts and respective assortativity coecients (similarly to
Table 4.4 in the main text). Again, we can see that a type of actor and geographic region have
the highest values in the table, indicating that leaders of the same position and from the same
geographic regions tend to cluster together more. other features do not seem to play an essential
roles in the assortativity of the leaders retweets network.
Moving on to the heatmaps that break down features into specic categories, Figure C.5 again
shows that edges where both nodes are European leaders have the highest count probability of
0.4. The count probability of edges between Latin American leaders is also quite high at 0.2
meaning that the leaders of Latin American countries tend to retweet each other often. Heatmap
in Figure C.4 breaks down the clustering by regime into specic types of government. Similarly for
the mention network, it shows that democratic leaders are much more likely to engage with other
democratic leaders on Twitter. The dark box in the middle of the heatmap indicates that the edges
where both nodes are democratic leaders have a high count probability of 0.4.
Finally, we need to check whether the patterns on leaders' centrality on the retweet network are
similar to what we have seen with the mention network. Similarly to our main analysis, we select
weighted in-degree centrality as the our measure of centrality. Unsurprisingly, the distribution of
retweet weighted indegree is also heavy-tailed with a few accounts having a high level of indegree
and most of the accounts having very low levels. Again, we dichotomize the variable following the
Pareto law: 20% of the top accounts produce 80.5% of all retweeting activity.
Table C.3 displays the coecient estimates for a set of logistic regressions of social media
centrality on the similar set of predictors as Table 4.5 in the main text. Consistent with our main
results, Model 2 shows that high income countries are not more likely to occupy central positions
in the leaders' retweet network. Rates of internet penetration emerge as a signicant predictor in
127
Figure C.4: Retweet Network Type of Government Clustering
Figure C.5: Retweet Network Region Clustering
128
the model; however, the eect is not robust to the addition of the variable of regime (Model 4).
Democracy dummy again emerges as a highly signicant and robust predictor of leaders centrality
on Twitter.
Table C.3: Retweet Network Centrality
Retweet Network Centrality (Top 20%)
(1) (2) (3) (4)
Account age 1.951

1.188 1.294 0.757
(0.641) (0.693) (0.668) (0.715)
Statuses count 5.428

7.538

5.483

7.000

(1.146) (1.616) (1.163) (1.606)
Burstiness 1.100 1.455 0.724 1.195
(2.882) (2.888) (2.736) (2.756)
Own language 0.953

0.933

0.966

0.933

(0.262) (0.276) (0.273) (0.280)
Actor: Head of State 0.607 0.712 0.610 0.744
(0.405) (0.420) (0.424) (0.431)
Actor: MFA 2.204

2.246

2.214

2.286

(0.350) (0.363) (0.367) (0.375)
Personal account 0.008 0.208 0.011 0.205
(0.266) (0.282) (0.278) (0.291)
Middle income 0.783 0.709
(0.431) (0.433)
High income 0.682 0.694
(0.575) (0.577)
Internet users 1.807

1.744
(0.904) (0.923)
Democracy (dummy) 0.819

0.644

(0.285) (0.299)
Constant 2.719 2.944 3.149 3.253
(2.042) (2.066) (1.938) (1.972)
N 527 491 474 459

p < .05;

p < .01
Notes: Table displays logit coecients, standard errors in parentheses. All
non-dichotomous measures have been rescaled from 0 to 1.
129

Abstract (if available)

Abstract This dissertation uses a multi-method approach to explore how the public forms foreign policy attitudes and how political leaders use online platforms to form public opinion at home and abroad. The introduction begins with a brief overview of the literature on public opinion, introduces the studies, and presents a case for combining experimental and computational methodologies to address various questions in international relations research. Chapter 2 uses attribution theory to explain how people make inferences about complex and ambiguous international events. It uses experimental design to show that both identity of the country involved and individual predispositions of the inferring subjects affect attributions of blame. Chapter 3 investigates how people form moral judgement about adversarial foreign nations and the role of intergroup threat in this process. Using a combination of computational and experimental methodology, it shows that perception of intergroup threat might lead to the formation of moral judgment about the source of threat, even when no information about morality of the group is available. Chapter 4 examines how political leaders use online platforms to form public opinion domestically and internationally. Using a large sample of political tweets, I explore how leaders strategically interact with each other on social media to demonstrate collaboration, attract supporters abroad, shape the conflict narrative, and demonstrate resolve. The dissertation concludes with a review of the findings and suggestions for future research.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Information, public opinion, and international relations

PDF

Perspectives on state capacity and the political geography of conflict

PDF

Habits and policy: the socio-cognitive foundations of foreign policy stability

PDF

Thawing rivalries and fading friendships: a multi-method approach to rapprochement and alienation

PDF

Like father, like son? A succession-based explanation for conflict initiation by authoritarian regimes

PDF

When passions run high: emotions and the communication of intentions in face-to-face diplomacy

PDF

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

PDF

Selling virtue: how human rights NGOs and their donors work together to create a better world ... for themselves

PDF

Are you as old as you feel? A quantitative and qualitative analysis of leader age and foreign policy decision-making

PDF

International politics and domestic institutional change: the rise of executive war-making autonomy in the United States

PDF

Politics is something we do together: identity and institutions in U.S. elections

PDF

Neither purse nor sword: strategic minority opinions and judicial policy-making

PDF

The changing state of tolerance and tolerance in changing states

PDF

Circuit breakers: how policy entrepreneurs interrupted the electric flow with Peru’s first renewable energy legislation for the grid

PDF

Political participation, public opinion, and the law

PDF

Forced march to modernity: State-imposed cultural change and regime stability in 20th-century east Asia

PDF

Mandate and manipulate: constructing political legitimacy during the Yuan-Ming transition, 1351-1370

PDF

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

PDF

The IO contestation project: a multi-method approach towards understanding the ways states contest international organizations.

PDF

A framework for research in human-agent negotiation

Asset Metadata

Creator Iakhnis, Evgeniia (author)

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

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Political Science and International Relations

Publication Date 10/26/2020

Defense Date 04/07/2020

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

Tag experiment,Foreign policy,OAI-PMH Harvest,public opinion,survey

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor James, Patrick (committee chair), Rathbun, Brian Christopher (committee chair), Fast, Nathanael (committee member)

Creator Email iakhnis@usc.edu,xniso4ka@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-385967

Unique identifier UC11666318

Identifier etd-IakhnisEvg-9063.pdf (filename),usctheses-c89-385967 (legacy record id)

Legacy Identifier etd-IakhnisEvg-9063.pdf

Dmrecord 385967

Document Type Dissertation

Rights Iakhnis, Evgeniia

Type texts

Source 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 a...

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