Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Identity salience and political violence: centering social identity in the study of intrastate conflict
(USC Thesis Other)
Identity salience and political violence: centering social identity in the study of intrastate conflict
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
IDENTITY SALIENCE AND POLITICAL VIOLENCE:
CENTERING SOCIAL IDENTITY IN THE STUDY OF INTRASTATE CONFLICT
by
Sarah A. Cueva Egan
A Dissertation Presented to
THE FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(POLITICAL SCIENCE AND INTERNATIONAL RELATIONS)
December 2022
Copyright 2022 Sarah A. Cueva Egan
ii
ACKNOWLEDGEMENTS
So many people have helped me along this journey that it would be impossible to thank them all
here.
Thank you to my dissertation committee: Laurie Brand, Gerry Munck, Stephanie Schwartz (London
School of Economics), James Lo (Meta), and Jody Agius-Vallejo. Laurie, whose patience, expertise,
and support through every twist and turn of this project has been instrumental in the production
and completion of this manuscript, has been an ideal dissertation advisor. I am privileged to have
had her as committee chair, adviser, and mentor. Gerry, thank you for always pushing me to be
precise and to consider alternative explanations. Stephanie, thank you for being a mentor and a role
model. I can’t wait to see where your career takes you. James, your enthusiasm for research and data
is infectious. You are the reason that I now love quantitative social science and find R fun. Jody,
thank you for your always insightful comments and for your investment in my success as the
external member of my committee. It is truly appreciated.
I have many colleagues to thank for their contributions to the development and improvement of
this work. My thanks to participants in the CIS Comparative Politics Workshop (March 2021) for
their comments on a rough draft of the prospectus that eventually became this dissertation. Thank
you also to participants in the Essex Summer School for Social Science Data Analysis’s Case Study
Methods course (August 2021) for their comments on the typological theory. Thank you to USC’s
Center for International Studies for their generous financial support in the funding of this research. I
am grateful to Aidan McHugh for his consistently outstanding research assistance, and look forward
to seeing what he does next. Thanks also to Alix Ziff, Jackie S.H. Wong, and Gaea Morales for their
feedback provided through the CIS Dissertation Reading Group.
I thank my incredible family and friends for their unwavering support through this process, and for
continuing to put up with me.
To the Wolfpacj, whose love and support have kept me going since 2010, thank you.
My husband, Liam, is the most patient man I have ever met (except when it comes to Mets
baseball). He has tolerated my unpredictable writing hours, been there to cheer me up when I’ve
been down, and is always willing to give me a reality check and remind me of what is most
important. Thank you for your support, for believing in me, and for loving me through dissertating.
Thank you also to our dog, Mookie, who always ensured I took frequent writing breaks to play with
him.
My parents, Brett and Sandra Cueva, have never failed to provide encouragement on my most
discouraging days, and have always been the first to pick me up when I’ve fallen (sometimes
literally). They have instilled in me an appreciation for higher education not only as a vehicle of
social and economic mobility, but also as an inherently valuable pursuit. They have been by my side
for every step of this 30-year journey, and it is to them that I dedicate this work.
iii
TABLE OF CONTENTS
Acknowledgements............................................................................................................................................ ii
List of Tables .....................................................................................................................................................iv
List of Figures.....................................................................................................................................................v
Abstract ..............................................................................................................................................................vi
Preface ...............................................................................................................................................................vii
Introduction ........................................................................................................................................................1
Chapter 1: What Is Meant By Identity Salience? Conceptualization and Measurement for
Applications to Political Science.....................................................................................................................11
Chapter 1: Existing Approaches to Conceptualizing and Measuring Identity...........................13
Chapter 1: Conceptualizing Identity Salience ............................................................................... 17
Chapter 1: Measuring Identity Salience as a Latent Variable ..................................................... 28
Chapter 1: Discussion ...................................................................................................................... 36
Chapter 2: Identity Salience and Political Violence: A Micro-Level Theory with Evidence from the
United States......................................................................................................................................................39
Chapter 2: Literature Review ............................................................................................................44
Chapter 2: A Micro-Level Theory of Identity Salience and Political Violence.........................49
Chapter 2: Empirical Analysis: Identity and Political Violence in the United States................60
Chapter 2: Results...............................................................................................................................72
Chapter 2: Discussion.........................................................................................................................76
Chapter 3: Identity-Based Political Violence: A Macro-Level Theory and Typology............................83
Chapter 3: Literature Review ............................................................................................................85
Chapter 3: Theory and Hypotheses .................................................................................................88
Chapter 3: Research Design ….........................................................................................................98
Chapter 3: Results ............................................................................................................................107
Chapter 3: Identity-Based Political Violence: Three Brief Case Illustrations..........................113
Chapter 3: Conclusion......................................................................................................................131
Conclusion………….....................................................................................................................................133
References ...................................................................................................................................................... 137
Appendices.......................................................................................................................................................153
Appendix 1: Measurement Model..................................................................................................153
Appendix 2: MTurk Survey Experiment and Supplemental Analyses......................................159
Appendix 3: IBPV Codebook and Supplemental Analyses........................................................178
iv
LIST OF TABLES
Table 2.1: Main Results (MTurk Survey Experiment).................................................................................75
Table 2.2: Main Results with Added Controls..............................................................................................76
Table 3.1: Main Results, H 1 (IBPV).............................................................................................................108
Table 3.2: Main Results for H 2 (Dependent Variable as Count Data)....................................................111
v
LIST OF FIGURES
Figure 1.1: Visual Representation of Identity Salience as a Latent Variable............................................30
Figure 2.1: Likelihood of identity salience change to lead to political violence, by threat type............54
Figure 2.2: Experimental Flow........................................................................................................................66
Figure 2.3: Experimental Treatment Condition...........................................................................................69
Figure 2.4: Frequently Used Words in Open-Ended Treatment Responses...........................................78
Figure 2.5: Sentiment Analysis of Open-Ended Treatment Responses...................................................79
Figure 2.6: Baseline Identity Salience and Political Violence Index Scores by Party Identification....81
Figure 3.1: A Visualized Typology of Identity-Based Political Violence..................................................93
Figure 3.2: Operationalization and Measurement of Key Variables.......................................................100
Figure 3.3: Predicted Probabilities of Identity Coherence*Elite Support Interaction.........................110
vi
ABSTRACT
Individual and group attachments to social identity can be important variables in shaping
micro-level and macro-level political processes and outcomes, and yet one oft-mentioned aspect of
identity remains undertheorized and ill-conceptualized: identity salience. Identity salience, as this
dissertation argues, is a key concept that must be clearly defined and measured if social scientists are
to understand the effects of identity attachments on political processes and outcomes. This
dissertation also argues that identity salience as a variable is especially important to incorporate in
studies of political violence. Article one provides a detailed conceptualization of identity salience as a
variable, defining it as the level of importance that an individual assigns to a group identity at a given
time such that it is capable of motivating political behavior. Article one also offers a theoretically
informed and novel measurement strategy for identity salience as a latent variable, which can be
adapted based on context and incorporated into quantitative models in which identity salience is a
key variable. Article two applies this model to a study of national identity salience in the United
States at the micro-level, using a survey experiment to understand the relationship between identity
salience and expressed attitudes toward political violence. Although the results indicate a strong
positive relationship between baseline levels of identity salience and permissive attitudes toward
political violence, the experimental manipulation is inconclusive. Article 3 builds upon the
foundation laid in articles 1 and 2 to propose a typological macro-level theory of identity-based
political violence. The theory proposes a causal relationship between a conflict being identity-based
and its level of severity as a function of the identity coherence of the group(s) involved and the level
of elite support each group receives. The quantitative analysis finds strong initial support for the
theory, with results indicating a statistically and substantively significant relationship between a
perpetrating actor’s levels of identity coherence and elite support on the one hand and the severity
of political violence in which they are involved on the other. Illustrative cases for each ideal type are
discussed to further demonstrate the utility of the theory. Possible future applications and extensions
of this research are then discussed.
vii
PREFACE
When I first began planning what my dissertation would look like, I did not expect a global
pandemic to derail those plans. What started as an ambitious research design that would involve a
survey experiment in Iraqi Kurdistan and qualitative interviews in conflict-affected contexts soon
became unfeasible, at least until the pandemic ended and I could safely and ethically resume travel
and interaction with others in the field. I had originally had the idea of conducting initial fieldwork
in the summer of 2020, which was then pushed to summer 2021. By the time the summer 2021
wave came about, I realized that I could no longer expect that I would be able to collect data in the
field as I had hoped. However, I still wanted to study the role of identity salience in influencing
attitudes toward and outcomes of political violence, which required me to adapt the initial plan to
something that could be done fully remotely. One of the very few benefits of writing a dissertation
during a global pandemic was that being stuck at home led me to get more creative in terms of
theory and data collection in such a way that the theory ultimately was formulated in a way that it
would become more generalizable.
At the same time as COVID-19 hindered my plans to conduct fieldwork, domestic unrest
provided a new lens through which to examine the role of identity salience in political violence. The
contentious 2020 election, the January 6 insurrection, and the aftermath of both strongly affected
and framed my thinking. These events forced me to take an at times uncomfortable look at the ways
in which existing research on political violence within the discipline—particularly among
Americans—has largely overlooked the phenomenon in the United States. Although I was at first
disappointed that I was unable to focus more internationally, both the pandemic and unrest in the
U.S. provided an invaluable perspective.
1
INTRODUCTION
Over the course of roughly eleven days in May 2021, what initially began as a dispute over
Israel’s evictions of six Palestinian families in the small East Jerusalem neighborhood of Sheikh
Jarrah resulted in a sharp escalation of Israeli-Palestinian violence that killed over 250 civilians, a
disproportionate majority of them Palestinians killed by Israeli rocket fire.
1
Angry civilians on both
sides took to the streets of Jerusalem, with days of violent clashes between Palestinians and Israelis,
rocket attacks exchanged between Israel and Hamas, and street violence leading some observers to
voice concern that a Third Intifada was imminent. Though this ultimately did not come to pass, this
event highlighted just how quickly a dispute between groups over such an issue can become
something much more serious and deadly. Of course, the Israeli-Palestinian conflict is generations
old and has been prone to flare-ups of deadly violence, making smaller-scale disputes more likely to
result in violence than would be the case between groups with shorter histories of intergroup
tensions. Still, the events in Sheikh Jarrah in the early summer of 2021 underscore important
questions about the nature of identity and its role in political violence that beg further attention
from political scientists. How exactly is it that intergroup tensions can be exacerbated to such a level
as to cause hundreds of deaths and even more injuries and property damage, and when are we more
or less likely to observe such a severe escalation? Why is it that these heightened tensions seem to be
more likely to result in more and more severe political violence when the involved parties are
defined in large part by an ethnic or religious, rather than economic, class, or partisan, identity?
What makes certain identities so persistent and susceptible to political violence waged in the name of
these identities?
1
UN Office for the Coordination of Humanitarian Affairs, Protection of Civilians Report|24-31 May 2021.
https://www.ochaopt.org/poc/24-31-may-2021
2
The continued existence and strength of ethnic, religious, and other identity-based
movements has drawn renewed attention to the importance of social identity as a variable in political
processes and outcomes. Many have been violent in nature, making a fresh and rigorous look at
identity not only a worthwhile direction of academic inquiry, but also a policy priority. Is there
something that we as political scientists are missing in our understanding of identity as a variable and
its relationship to political violence?
As many scholars of identity have long claimed, discounting its importance as a variable
removes significant analytical power and overstates the centrality of material factors in contributing
to political outcomes while obfuscating the potential effects of non-material factors. Several
important contributions have been made in conceptualizing identity as a variable (i.e. Abdelal et al.
2006, McDermott 2004; Huddy 2001; Sambanis and Kalin 2018) and in proposing alternative
measurement strategies for a variable that is notoriously difficult to measure (i.e. McDermott 2009;
Spry 2021).
2
These works have moved our understanding of identity in political science forward in
important ways, not least of which in terms of how the discipline conceptualizes the role of identity
in political violence. Identity is multi-dimensional thus very hard to quantify.
Building on the rich bodies of research on social identity and political violence, this
dissertation focuses on a specific aspect of identity that I will argue is overlooked, yet of outsized
importance, in determining whether political outcomes are violent or nonviolent: identity salience. I
will claim that identity salience is a key variable that can help explain political violence outcomes.
This stands in contrast to much of the literature on civil war and political violence, which tends to
treat identity as a contextual constant rather than as a variable. Despite frequent references to
identity salience in the literature, little effort has been made to conceptualize or measure it at the
2
This literature will be discussed in depth in all three articles.
3
microlevel or to examine its effects at the macrolevel. This dissertation takes steps toward addressing
these lacunae, with a specific focus on outcomes of political violence.
Given these gaps, the core questions that this dissertation aims to address are the following.
What is the relationship between identity salience and political violence at the microlevel, and what
influences the strength and relevance of certain identities in producing or contributing to outcomes
of political violence? More specifically, can the salience of certain types of identity change—in other
words, can identity salience be defined and operationalized as a variable—and if so, does this change
affect the likelihood and severity of outcomes of political violence? How, and through what
mechanisms? These questions are hardly new and certainly do not have easy answers. However, one
barrier to a clearer understanding stems at least in part from a lack of consensus on how to
conceptualize and measure identity salience. If we are to better understand a) how and why
intergroup tensions are activated and b) why the activation of these tensions may result in observed
changes in political behavior, then looking at the microfoundations of how and why the salience of
these group identities change is vitally important. There is a wealth of research across disciplines,
and this project aims to systematically incorporate these insights to shed more light on these
important questions.
Identity salience is a frequently mentioned but infrequently explained term used in many
studies where identity is a factor, and I argue that it is a highly complex and multidimensional
variable that has yet to be fully unpacked and operationalized. Additionally, identity and identity
salience must be more systematically incorporated into analyses of intrastate political violence.
Indeed, a recent study found that of the many variables that might contribute to the onset of civil
wars, the organization of political parties around racial, ethnic, and religious identities was one of the
two strongest predictors of whether a country would experience civil war (Walters 2022, see also
Goldstone 2010). To understand how and why identity is so important in models of political
4
violence and conflict, however, it is important to first understand the microfoundations of the
phenomenon—in other words, how and why identity could lead individuals to change their political
behavior, and more specifically, when identity will lead individuals to accept political violence as an
acceptable behavior. The first two articles of the dissertation will focus on this question. A better
understanding of identity salience not only can shed light on political violence at the microlevel but
may also help explain variation in outcomes of political violence at the macrolevel. While increases
in identity salience could lead to political violence, the severity of the outcome is not always the
same and is certainly not driven only by identity. When identity-based political violence occurs, what
determines the severity of that outcome? Article 3 begins to address these macrolevel questions.
At the heart of my argument is that the politically relevant essence of identity as a variable is
its salience, and that whether and to what extent an identity is salient will affect outcomes of political
violence. A major and theoretically central assumption of the argument is that identity is not fixed
and can vary in the short and long-term in ways that make it relevant to political outcomes.
Definitions and Scope Conditions
Though identity salience may play a role in a multitude of political outcomes, it would be
impossible to cover them all here. I instead focus on the relationship between identity salience and
political violence at the intrastate level. I will argue that there is a gap in our understanding of the
role of identity in political violence due to the incomplete conceptualization of identity salience.
Prior to defining identity salience in article 1, it is important to clarify and define a few key terms
used throughout the dissertation. For our purposes here, I consider an identity to be a cognitive
frame to which individuals may subscribe in similar ways to other ingroup members. Thus, the
potential for group consciousness and, further, group membership, is an important element of this
5
definition of identity. To limit the scope of this study, I focus only on religious, ethnic
3
, and national
identities. I do this for two main reasons. First, these types of identities are most likely to identify
members at least partially by “descent-based attributes” (Chandra 2006), rendering them the most
legible to both ingroup members and those outside of the group. In the case of religious identity,
descent-based attributes might be less legible or relevant than they would be with ethnic identities,
but these identities may be reinforced or intensified by strong institutional backing (Grzymala-Busse
2012) and everyday practice and behaviors (Hoffman and Nugent 2017). And though national
identity is not always restricted to a set of readily discernible characteristics, such as a particular
ethnicity or religion, these types of identities are potentially powerful in that they are often granted
legitimacy by a central government, which has the legal, institutional, and coercive capacity to shape
the contours of what it means to be a member of the nation and to grant or deny membership on
the basis of these criteria (Schildkraut 2014). National identities may also encompass ethnic and
religious identities, as these may be considered criteria for membership, at least implicitly, and
sometimes explicitly. In the case of American national identity, for example, some experimental
studies have investigated the extent of implicit bias regarding the racial and religious characteristics
that define a “real” American. They find that, contrary to Myrdal’s (1994) assertion that American-
ness is based on common principles and ideals (the “American Creed”), individuals frequently hold
ethnoracially and religiously restrictive ideas of what makes someone a true American (i.e., Cheryan
& Monin 2005, Devos & Banaji 2005, Yogeeswaran & Dasgupta 2010, Yogeeswaran et al. 2012,
Jacobs & Theiss-Morse 2013). This might also be the case for identities for nations without a state,
exemplified by groups that call for separation from a central state or for regional autonomy—for
example, the Basques or Catalans in Spain, the Kurds in southeastern Turkey and northern Iraq, or
Abkhazians in Georgia. These characteristics make ethnic, religious, and national identity particularly
3
I use ethnicity as a term that encompasses racial, ethnolinguistic, and ethnonational identities.
6
potent categories of social identity which might lead to changes in political behavior and shape
political outcomes, and thus will be the subject of our focus throughout the dissertation.
Salience is another concept that is important to unpack and further define. I consider salience to
be the level of importance that an individual ascribes to a particular issue or cause such that it shapes
and motivates action. This level of importance varies temporally and in terms of magnitude, and
therefore can become more or less intense over time. A useful comparison for thinking about the
meaning of salience would be political science scholarship that discusses issue salience, or the
likelihood that a particular political issue would prompt individuals to react in such a way as it drives
an individual to engage in political action. An issue would be particularly salient if, for example, it
leads individuals to vote for or against a candidate on the basis of that particular issue. A common
way of gauging the salience of a political issue, for example, involves asking individuals via a survey
what they consider the “most important problem” (or MIP) that led them to vote the way they did.
If a voter indicates that their family’s healthcare plan is the most important problem in exit polls,
then, that issue would be especially salient, and even more so if multiple voters consider that issue to
be salient. Functionally speaking, the salience of an identity is similar to the salience of an issue in
that they are both likely to shape political preferences. In the case of the voter indicating healthcare
as the MIP in our hypothetical exit poll, the salience of healthcare as an issue shapes the voter’s
preference for a particular political candidate, which then influences the behavior of casting a vote
for that candidate. Similarly, the high salience of an individual’s identity as a member of a minority
ethnic group would be likely to shape their preferences for certain political parties, candidates, or
policies, which may then lead to tangible behaviors, such as voting for a candidate from a particular
political party. Continuing with the comparison with issue salience, the salience of an identity may
also change over time. This idea of change over time will be key to the forthcoming
conceptualization and measurement strategy.
7
Finally, I consider political violence to be any action undertaken by a person or group of people
whose purpose is to inflict physical harm or damage against a perceived outgroup, whether that be
an opposing political party or candidate, religious sect, ethnic group, or the government itself.
Political violence outcomes can range in severity from property damage at the low end to the
initiation of high-casualty hostilities in the context of civil war at the high end of the spectrum, but I
only consider instances of political violence that have resulted in fatalities.
4
The term as I define it
encompasses acts with political motivations that use violent tactics—for example, shooting,
bombing, kidnapping, or any act that harms people or property. Specifically, this could take the form
of mob violence, rioting, armed insurrection or insurgency, or acts of terror committed by
individuals or groups. Political violence as I conceive of it in the dissertation does not include
random acts of violence with no clear political goal or motivation.
Plan of the Dissertation
To address the questions outlined earlier in this introduction, I examine the relationship between
identity salience and political violence at both the microlevel and macrolevel using a variety of
quantitative and qualitative methods. At the microlevel, I examine how and why identity salience
changes among individuals and test whether these changes lead to changed attitudes toward political
violence. At the macrolevel, I explore the relationship between identity and severity of political
violence outcomes after initial onset of hostilities, presenting and testing a typological theory of
identity-based political violence.
Article one lays the foundation for the rest of the dissertation, providing a novel
conceptualization of and measurement strategy for identity salience as a variable. Examining the
4
I explain exclusion criteria in more depth in article 3.
8
literature on social identity theory (i.e. Tajfel 1974) and identity as a variable in political science (i.e.
Abdelal et al. 2009), this article proposes a conceptualization of identity salience that allows for
variation over time and that can be incorporated as an independent or a dependent variable into
quantitative models. To do this, I propose a measurement model that quantifies identity salience as a
latent variable. The model relies upon individual survey responses measuring the salience of
respondents’ American national identity salience and is presented as a method that can be adapted to
measure different identities and generalized to different contexts.
Article two connects this conceptualization to attitudes toward political violence as an outcome,
outlining a microlevel theory of identity salience and political violence from which I derive and test
core hypotheses via a survey experiment on a convenience sample of American adults. This article
engages the literatures on political violence and political attitude formation, ultimately proposing a
theoretical framework that allows identity to play an active role as a primary independent variable
that can shape attitudes toward political violence. I measure individuals’ baseline levels of American
national identity salience to test the relationship between these baseline levels and attitudes toward
political violence committed on behalf of the identity being measured. To approximate change in
identity salience over time and determine whether varying levels of identity salience influence
attitudes toward political violence, I conduct an experiment that primes national identity salience and
compares political violence attitudes to those of the control group. In addition to placing identity
salience as a variable in the context of studies of attitude formation and political violence, this
approach builds upon more recent literature on measuring identity and provides a generalizable
means of testing whether there is a causal relationship between changes in identity salience and
changing attitudes. It also contributes to a growing call among scholars of political violence to more
systematically incorporate considerations of American political violence into the study of political
violence writ large, drawing upon emerging literatures within American politics that examine
9
partisanship as a social identity and its relationship to changes in attitudes toward political violence
in the United States (i.e. Egan 2019; Kalmoe and Mason 2022). The treatment of American political
violence through the same lens as is applied to studies of political violence outside of the United
States represents a concerted effort to bridge this gap between the American politics literatures on
political violence and the comparative politics literatures on political violence and civil war.
Article three outlines and tests a macrolevel theory of what I call identity-based political
violence, which proposes a causal relationship between two novel variables—identity coherence and
elite support—and the severity of political violence outcomes. This typological theory contributes to
the political violence literature by outlining a case for why political violence with identity dimensions
is distinct from non-identity based violence, and provides a framework upon which these differences
may be tested. I quantitatively test key hypotheses derived from this theory using a novel coding
approach of existing cross-national data from the Uppsala Conflict Data Program (UCDP) dyadic
dataset, outlining a detailed and replicable coding and measurement strategy for determining the
identity coherence and elite support levels of actors within political violence dyads. To my
knowledge, this is the first such attempt to do so. I supplement the quantitative evidence with three
short illustrations of typical cases for each type of violence outcome in the typology: racial violence
during the Ku Klux Klan’s so-called Second Wave in the 1920s; the sectarian insurgency during the
Troubles in Northern Ireland in the late 1960s; and the beginning of the second Lebanese Civil War
in 1975. The theory and methodology presented in article 3 represent an approach that is
simultaneously actor-based but that also situates these actors within their institutional contexts.
The scope of the dissertation is broad and its questions complex; indeed, this research perhaps
raises even more questions than it answers. At the same time, this represents a step towards
beginning to think more deeply about how to systematically incorporate identity as a variable into
studies of political violence, and the theory and measurement approaches laid out provide a
10
foundation upon which to derive and test further hypotheses. Identity certainly is not the only
important factor at work in processes of political violence, but it is an important one. The following
articles will help shed light on why this is and will provide theoretical and empirical tools that may
help put identity back at the center of the debate, while also highlighting areas of future research that
might build upon this work.
11
CHAPTER 1: WHAT IS MEANT BY IDENTITY SALIENCE? CONCEPTUALIZATION
AND MEASUREMENT FOR APPLICATIONS TO POLITICAL SCIENCE
There is a significant difference between actively identifying with a particular aspect of one’s
identity and changing one’s political behavior as a result. A devout Catholic, for example, may
prioritize their religious identity such that they attend daily mass and weekly confession, are involved
in church activities, and even choose their friends, partners, and parenting approach based on their
religion. It is another thing to cast a vote for a political candidate because of one’s religion or
position on social issues, to participate in a protest seeking change on a religiously relevant issue,
and, as a more extreme but not altogether uncommon example, to perpetrate political violence
against a religious outgroup. During Ireland’s Troubles, for example, IRA paramilitaries were
planting explosives at Protestant churches, and Unionist Protestant forces responded in kind. At the
same time, staunch Catholics and Protestants outside of Ireland were peacefully coexisting. While
this difference may seem intuitive, it is one that has thus far been inadequately captured by the
prevailing treatment of identity as a variable in political science. Mere identification with a group is
insufficient to make that identity politically relevant: it is also necessary to consider the cognitive and
emotional aspects of an individual’s identification with a group to understand what mobilizes them
to political action on behalf of that group. I argue that identity salience is a variable that captures
these aspects and that can be used to more precisely determine whether, how, and to what extent
identity might be or become politically relevant. This article is dedicated to outlining a conceptual
framework and generalizable measurement strategy that can be especially useful in understanding the
relationship between identity and political violence.
I provide in the following pages a critical review of existing research on identity as a variable and
examine current treatments of identity salience in the political science literature. A core assumption
underlying this effort is that much of the work on political violence and its causes suffers from an
12
incomplete conceptualization of identity salience. While many valuable studies provide insights into
the role of identity in political violence and civil conflict, many of these same studies do not unpack
exactly what it means for an identity to be salient, or how that salience may vary. This article first
introduces a conceptualization of identity salience and then outlines a generalizable strategy to
operationalize the measurement of identity salience as a latent variable using Bayesian item response
theory. The core claim is that identity salience is an important yet under-theorized variable that
allows for the empirical measurement of meaningful identity attachments. But do we really need a
new concept to understand identity and its effects on political outcomes, or would such an endeavor
merely add another layer to an already complex scholarly debate? A look at the existing literature on
identity suggests that such an effort would be worthwhile, particularly when it comes to
understanding the role of identity in such high-stakes outcomes as political violence.
Despite a plethora of innovative and insightful studies of identity as a variable, there is a
significant obstacle in the field’s incomplete conceptualization and measurement of identity salience.
I draw upon existing scholarship to further deconstruct this conceptual and methodological
obstacle, proposing a conceptual framework and generalizable measurement strategy for identity
salience as a variable. I identify some key areas in the literature that I attempt to build upon—and, in
some cases, push back against—in providing this updated conceptualization and measurement
strategy.
I begin by critically engaging with the relevant literature on identity as a variable, focusing
particularly on social identity theory (SIT) as it applies to the social sciences. I then examine recent
work on the measurement of identity and suggest that a new measurement strategy for identity
salience is needed. Finally, I propose a novel conceptualization of and measurement strategy for
identity salience that can be applied to further studies of political phenomena in which identity plays
a key role.
13
Chapter 1: Existing Approaches to Conceptualizing and Measuring Identity
The advent of social identity theory (SIT) in the 1970s with the work of social psychologist
Henri Tajfel eventually generated entire literatures across disciplines from social psychology to
political science that sought to understand how and why social identities develop, change, and affect
various outcomes. A social identity is one that an individual uses to categorize themselves as part of
a collective group: for example, race or ethnicity, religion, nationality, or even partisanship can be
considered social identities. A large part of social identity theory distinguishes social identities from
personal identities, which are identities that an individual uses to differentiate themselves from
others, such as one’s familial identity as a parent, professional identity as a lawyer, or even
personality characteristics. SIT assumes that social identities are uniquely relevant in shaping group
attitudes and behavior and provides a framework for thinking about how and why this is.
Indeed, SIT’s works and insights remain foundational for contemporary studies of identity
(Abdelal et al 2009, Spry 2021).
5
SIT asserts that individuals categorize themselves based on their
sense of belonging to a group, a process of social categorization by which people sort themselves
and others into groups (Tajfel 1974, Tajfel 1981, Tajfel and Turner 1979, Turner et al 1979). This
process of self-sorting and social categorization fulfills a need for self-esteem that comes from “a
positively valued distinctiveness for their own group compared to other groups.” (Turner and
Reynolds 2001, p. 134) This need for a positive social identity from in-group identification, then, can
itself lead to discrimination against members of the out-group (Tajfel 1981), even absent material or
other non-ideational threats from that outgroup (Turner 1975). These ideas have garnered robust
5
“…SIT has been one of the most successful research paradigms for studying identity, at least in terms of the quantity
and quality of research it has generated.” (Abdelal et al, p. 9)
14
empirical support from minimal group paradigm studies, first conducted by Tajfel to uncover the
minimal conditions under which in-group favoritism and out-group favoritism would be triggered.
Such studies first indicated the power of group membership in leading to discrimination against
outgroup members. Additional studies then affirmed the finding that even surface-level
identification with a group based on assignment to arbitrarily created categories is, on its own,
sufficient to lead to discriminatory attitudes and behaviors (i.e., Tajfel et al 1971, Turner et al 1979,
Hogg and Sunderland 1991).
This relational and context-dependent theory of social identity is fundamental to its utility in
applications to studies of political processes and outcomes. Constructivist scholars have long
incorporated questions of social identity into their research agendas, either implicitly or explicitly.
Recent work has recognized the importance of identity as an explanatory variable that can be
incorporated into a variety of approaches, including rationalist models and theories of political
science, while not discounting its psychological and nonmaterial components. Sambanis and Kalin
(2018) provide a nuanced conceptualization of social identity as it pertains to political science,
arguing that scholars discussing social identity as something solely influenced by material
motivations must also consider the nonmaterial, cognitive aspects of identity to fully leverage its
potential explanatory power for outcomes as diverse as patterns of tax compliance and ethnic
conflict.
If social identity is important to understanding political processes and outcomes, however,
clear conceptualization and measurement are paramount to studying how and why this is across
different contexts. A persistent problem identified by scholars of identity pursuing such a
conceptualization has been the difficulty of measuring a construct that varies over time and that
frequently means different things to different people (Lee 2008, Abdelal et al 2009, Sen and Wasow
2016). Among the more comprehensive and generalizable treatments of identity as a variable for
15
applications to political science that begins to address these criticisms is that of Abdelal et al. (2009),
who define identity as a category of collective meaning that varies along the two primary dimensions
of content and contestation. By their definition, any category of identity necessarily involves some
degree of constant ingroup contestation over its meaning, or content, though they allow that the
degree and extent of this contestation is not uniform across different types of identity (i.e., religion,
race, gender, class).
Still, such conceptualizations of identity leave out or underemphasize important dimensions that
might be causally important in affecting outcomes. Amber Spry notes that although SIT has spurred
important theoretical and methodological innovations in research on identity, there has largely been
a failure to fully account for and disentangle three important yet distinct attributes of identity: group
membership, group identity, and group consciousness (Spry 2021; see also McClain et al 2009).
Group membership, as it has been defined since the advent of SIT, refers to an individual’s
ascriptive categorization with a particular social or relational group: for example, one might be
perceived to belong to a particular group identity, such as their race or ethnicity, gender, or religion,
without actually feeling a sense of belonging with those groups. Spry acknowledges the advances in
identity research that SIT has enabled, while at the same time suggesting that the experimental work
that has arisen from it has emphasized the manipulation of group membership over group
consciousness and identity. Unlike group membership, group identity refers to an individual’s level
of psychological attachment or sense of belonging to a particular group (Frable 1997; McClain et al.
2009; Stets and Burke 2000; Tajfel 1974, Spry 2021). Group consciousness, then, is the extent to
which an individual assigns political relevance to their identity, or the extent to which an individual’s
identity might be politicized (Spry 2021).
6
6
“Group consciousness is defined as in-group identification plus a sense of political awareness about the group’s relative
position in society, plus the belief that collective action is the best means to realize the group’s shared interests.” (p. 433)
16
Whereas group membership corresponds with an individual’s sense of self, Spry notes, group
identity and consciousness are facets of identity that link individuals with the group that they
consider themselves to be part of. The more substantive idea of group consciousness refers to an
individual’s sense of their group’s relative status within society and generally leads group members to
think of collective, rather than individual, action as paramount to maintaining and advancing their
group’s status (Dawson 1994, Miller et al 1981). In other words, group consciousness is “the basis
for the link between identity and political outcomes.” (Spry 2021, p. 433) Given these conceptual
issues and the failure of existing measurement approaches to capture the relevant dimensions of
identity, Spry calls for novel approaches that better account for and distinguish between these
different components. One potential variable that can begin to capture these different dimensions of
identity is identity salience.
To discuss the “salience” of certain issues, let alone a concept as complex as identity, it is
important to understand what we mean substantively speaking when we talk about an identity being
salient. Assuming that identity groups are relatively fixed belies the fact that groups are to some
extent iteratively shaped by their constituent individuals and interactions of these individuals with
each other and with outgroups.
7
To be sure, salience has not been entirely neglected. Indeed, many
scholars have recognized the importance of salience, or identity attachment, in motivating various
forms of political action (Brewer 2001; Huddy 2001; Stryker et al 2000; Brady and Kaplan 2009;
Hatemi and McDermott 2017). Sheldon Stryker first explored how variation in the salience of
different identities—for example, occupational, familial, or political—motivated changes in
behavior. In discussing this hierarchical ranking of different identities, Stryker defines identity
salience as “a readiness to act out an identity as a consequence of the identity’s properties as a
7
See especially Brubaker and Cooper (2001) and their conceptualization of “groupness” as the extent to which an
individual attaches to a group identity.
17
cognitive structure or schema.” Others have similarly tested whether, and argued that, the salience of
an identity influences behavior (Stryker 1980; see also McCall and Simmons 1978 and Callero 1985).
Despite the existence of such accounts of identity salience within the fields of sociology and
social psychology, few within political science have attempted to unpack what identity salience
means, nor has it been comprehensively treated as a relevant variable in its own right. There is no
substantive explanation within political science of what it means for an identity to be salient, how it
becomes salient, and what the consequences of the activation and variability of identity salience are
for explaining political outcomes where identity is a factor. Still, some political scientists have
advocated for a more direct focus on the measurement of identity salience in situations where it
might plausibly be connected to political outcomes (Brady and Kaplan 2000, Brewer 2001, Huddy
2001, McDermott 2009). Indeed, some studies have found that experimental manipulation of the
salience of shared social identities has led to increased prosocial behavior
8
toward outgroups
(Charnysh et al 2015) and more favorable post-conflict negotiation outcomes (Kramer et al 1993).
Similar methods may also be applied to uncovering and better understanding the microfoundations
of political outcomes where intense in-group salience and intergroup tensions lead to violence
(McDermott 2009, p. 367).
How can we build upon these insights and create a path forward for the systematic
consideration of identity salience as a variable in political processes and outcomes? It is to a
conceptual framework of identity salience that we now turn.
Chapter 1: Conceptualizing Identity Salience
8
Prosocial behaviors are ones that benefit other members of society and that do not have an inherent benefit to the
individual engaging in such behaviors. For example, working as a volunteer, giving monetary donations, and even
abiding by certain social norms can be considered prosocial behaviors.
18
Given the importance of disentangling different dimensions of identity and the theoretical and
empirical importance of identity salience, I propose a novel conceptualization that begins to address
some of the discrepancies noted in the literature on identity in political science. In addition to
proposing a definition of identity salience, this conceptualization provides a foundation upon which
identity salience can be measured as a quantitative or qualitative variable.
At its most fundamental level, I define identity salience as the level of importance that an
individual assigns to a group identity at a given time such that it is capable of motivating political behavior.
9
Crucially, identity salience as a variable is highly context-dependent and susceptible to change in
both the short and long term. Understanding how and why individuals are drawn to, embody, and
evolve in their attachments to different aspects of their identity is necessary to understanding the
causal role of group identity in political behavior, and identity salience is a dynamic variable that
captures much of the unexplained variance in this phenomenon. When identity salience is
mentioned in the existing literature, it is treated as a one-dimensional aspect of identity and is used
to connote political relevance; however, it is not enough to claim an identity as one’s own for it to
be politically relevant. The following conceptualization accounts for and attempts to correct this by
defining a multi-dimensional variable that explains when identities become relevant such that they
may change political attitudes and behaviors.
It is important to clarify that identity salience can be both an independent and a dependent
variable, though I focus here on the role of identity salience as an independent variable. Because
identity salience as a variable is meant to account for change over time, something must cause this
change; this change can, in turn, lead to a change in attitudinal and behavioral outcomes. In this way,
identity salience can at times be thought of as operating within a feedback loop: some event or
9
When I discuss identity, I include only group identities that are also social identities: namely ethnicity/race, religion,
and nationalism. Role and personal identities—for example, one’s identity as a lawyer, a mother, a fan of a sports team,
etc.—are not considered within this framework.
19
factor leads an individual’s identity salience to increase, which in turn may affect attitudes and
behaviors—these behaviors can then increase identity salience, which renews the cycle. This is
especially true when the behavior in question involves political violence.
10
In recognition of the careful and detailed definitions of identity that have already been
developed, identity salience as I conceive of it builds upon Abdelal et al.’s (2009) analytical
framework of identity. Their framework defines identity as a social category that varies along the two
dimensions of content and contestation, with content referring to the meaning of the identity and
contestation referring to the level of agreement within the group as to the content of their shared
identity. Content may take the form of constitutive norms, social purposes, relational comparisons,
and cognitive models, some or all of which may be present within a given identity category. By their
definition, constitutive norms are the formal and informal rules that define group membership;
social purposes refer to the goals that are shared by members of a group; relational comparisons are
the ways in which ingroup members situate themselves in relation to other identity groups; and
cognitive models refer to the ways in which an identity shapes group members’ conceptions of their
material and political conditions and interests. Contestation shapes the content of the identity
category (Abdelal et al. p. 27).
Abdelal et al. explicitly exclude salience from this framework, arguing that it pertains more to
individual attachments to identity rather than group attachments: in their framework, individuals and
groups are “analytically distinct categories,” and focusing on individual attachment to a group
identity “bypasses the meaning of the group to its members.” (30) While they are right to point out
that individuals and groups are distinct units of analysis, these dimensions do not account directly
10
I discuss this in more detail in article two, especially in the context of political violence. This idea of a feedback loop is
similar to previous scholarship on political psychology and conflict, wherein the psychologically powerful effects of
being exposed to violence can reinforce certain attitudes and behaviors that contributed to the outbreak of violence in
the first instance (i.e., Canetti et al 2013, Zeitzoff 2014).
20
for the level of importance attached to a group identity by members of the group. Indeed, a high
level of shared meaning among members of the group does not automatically indicate a high level of
salience of that identity among group members, at least not to the extent that it motivates changes in
behavior in accordance with group goals. A high level of shared meaning is not necessary for a
sudden change in salience to motivate political behavior among ingroup members; at the same time,
a high level of shared meaning regarding the content of an identity increases the likelihood that: 1)
salience will increase and: 2) that an increase in salience will result in some observed change in
behavior due to an increased likelihood of mobilization around the identity. It might also be said
that an increase in identity salience can be understood as preceding an increase in agreement over
content and a decrease in degree of contestation over that meaning— when heightened identity
salience drives a decrease in contestation over meaning, the barriers to mobilization around the
group decline. If an identity group experiences an event that causes multiple individual members to
attach more importance to their identity—in other words, an event that causes the salience of that
identity to increase—then differences in interpretation of meaning might become less relevant,
leading the group towards a higher degree of agreement in terms of their constitutive norms, social
purposes, relational comparisons, cognitive models, or some combination thereof. Increased identity
salience might lower the threshold at which members of a group are willing to set aside differences
in their interpretations of content such that their attachment to the collective identity becomes more
concrete and unconditional—in other words, that particular identity becomes central to their
conceptions of themselves in relation to society and therefore becomes more influential in shaping
political behavior.
Indeed, in discussing what links identity to politics and highlighting what is missing from many
existing formulations of identity as a variable, Spry notes that an identity must be activated in some
way to lead to political outcomes (Spry 2021, p. 433). Changes in identity salience might account for
21
this activation, something that is missing in the Abdelal et al. framework and that is important to
consider when thinking about the substantive connection between identity and politics.
My conceptualization of identity salience varies along three core dimensions: primary
subscription, cognitive prioritization, and affective valence. Together, these dimensions capture the
nature and intensity of an individual’s primary group identity, something that can be measured at a
fixed point in time but that also varies significantly in response to external factors. I first define and give
examples of each dimension, explaining the importance of each in providing a full picture of identity
salience. I also explore how this fits in with existing theories and frameworks for studying identity in
political science, with a focus on its strengths in measuring change over time and in connecting the
strength of the group with its constituent individuals.
Primary Subscription
The first dimension of identity salience, primary subscription, indicates the group that someone
primarily identifies with at a particular point in time, sometimes to the exclusion of other groups of
which they are also members.
11
This is the most straightforward dimension of identity salience in
that it simply isolates which group identity an individual considers to be their most important
identity at a given time, and it is the closest conceptually to the existing literature’s references to
identity salience. While it is undoubtedly the case that individuals may subscribe to multiple group
identities at a time, and that they may assign similar levels of importance to more than one identity
category concurrently—for example, someone who identifies as Mexican-American may assign
similar levels of importance to their Mexican heritage and American citizenship—this
11
It is rarely the case that individuals subscribe to only one identity; rather, they have many different identities they
might subscribe to at one point in time. Amber Spry uses an innovative alternative measurement strategy that allows
survey respondents to allocate “identity points” across race, class, religion, and gender (2017) in the US context, for
example. Examining the effect of multiple identities, or intersectionality (Crenshaw 1991) is beyond the scope of this
dissertation but would be a worthwhile subject for future research.
22
conceptualization also assumes that at least one of these identities holds a higher level of importance
relative to others: it is rarely the case that an individual would assign the very same level of
importance or primacy to all of these identities at the same point in time. An individual who is
American, Muslim, and Black may subscribe to their national, religious, and ethnic identities at the
same time, yet they likely favor one identity over the other when they consider their group identities.
Experiencing racism may cause an individual to attach more strongly to their Black identity, whereas
encountering discrimination based on one’s religion may lead an individual to be more aware of their
identity as a Muslim than they might have been otherwise. On the other hand, one’s attachment to a
particular category of their identity may be sufficiently strong that such influences do not lead one’s
primary subscription to change. As another example outside of the US context, an individual might
be a Sunni Muslim, a Kurd, and a citizen of Iraq. However, if that individual primarily groups
himself with others based on his Kurdish identity, then this might indicate that the salience of the
individual’s Kurdish identity is higher than that of their other identities.
Though an important element of identity salience, primary subscription itself does not account
for the intensity of one’s attachment to that identity, and taken alone significantly undercuts our
ability to understand the extent to which this identity will be politically relevant such that it can
mobilize individuals to alter their political attitudes and behavior. To return to the previous example,
the individual whose social identities include an Iraqi national, a Sunni Muslim, and a Kurd may
primarily subscribe to his identity as a Kurd, but primary subscription to a Kurdish identity cannot
tell us whether he attaches to this identity with a level of intensity such that he will adapt his
attitudes and behaviors in a way that prioritizes Kurdish group interests.
In discussing primary subscription, it is important to note that although multiple identities may
be highly relevant and salient to one individual at a given time, this framework focuses on which of
these multiple identities is most relevant. Future adaptations of this framework might facilitate the
23
consideration of more than one primary identity subscription at a time, though such a study is
beyond the scope of the current project.
Cognitive Prioritization
While primary subscription itself can tell us which identity is the most salient at a given time, it is
itself sufficient to indicate the extent to which an identity is salient. This is the purview of the second
and third dimensions of identity salience, each of which captures a different aspect of intensity. The
second dimension, cognitive prioritization, is the degree to which an individual considers and
integrates their primary identity in cognitive processes
12
, such as the perception and interpretation of
different scenarios or events. Someone with high cognitive prioritization of a particular identity
frequently considers their primary subscribed identity in everyday scenarios, whereas someone
whose identity is less salient will only rarely consider their identity in these same scenarios. Put
otherwise, cognitive prioritization refers to the extent to which someone incorporates this primary
identity into thought processes, decision-making, and the interpretation of social interactions and
events. Someone whose primary subscribed identity is religious, for example, would be more likely
to consider their religious identity and affiliation in choosing which news sources to consume, whom
to associate with socially, and where to send their children to school. If someone identifies as Jewish
but does not have high cognitive prioritization, their religious identity may be nothing more than an
afterthought in everyday cognitive processes and interactions—or at least they do not prioritize
religious considerations in these processes and interactions. Cognitive prioritization, then, accounts
for the degree to which an individual’s primary identity permeates their general everyday existence,
12
I refer to the psychological consideration of cognitive processes here, which the American Psychological Association
defines as “any of the mental functions assumed to be involved in the acquisition, storage, interpretation, manipulation,
transformation, and use of knowledge,” and which may include “attention, perception, learning, and problem solving…”
(APA Dictionary of Psychology).
24
and those with high cognitive prioritization are more likely to have higher levels of identity salience.
Cognitive prioritization may be further reinforced by practices common to certain groups, such as
the recitation of the Pledge of Allegiance for American schoolchildren or undertaking the five daily
prayers for Muslims. While certainly not a prerequisite for high cognitive prioritization, commitment
to such practices is an important consideration in thinking about this dimension of identity salience.
Returning to our previous example of an individual who is simultaneously Kurdish, Iraqi, and
Sunni, what would cognitive prioritization look like? Let us assume for the purposes of this example
that the individual primarily subscribes to their identity as a Kurd. If this individual grants cognitive
priority to this identity, their Kurdishness will factor more prominently in cognitive processes,
meaning that they will grant more attention to this identity in such a way as it is apparent in their
cognitive decision-making. They may, either consciously or unconsciously, favor social interactions
with other Kurds and spend more time-consuming Kurdish language news sources focused on
Kurds and Kurdish issues. This aspect of identity salience does not account for the role of emotion
in these cognitive processes, but rather is meant to capture the extent to which an individual’s
attachment to their primary subscribed identity translates into increased attention to and thought
about the identity. Considering the level of cognitive prioritization of a primary subscribed identity is
an important step toward determining the salience of that identity.
Affective Valence
The final dimension of identity salience is affective valence. Whereas cognitive prioritization
focuses on the more rational, cognition-based manifestations of identity salience, affective valence
refers to the type and intensity of emotions that individuals attach to their salient group identities. In
thinking about this dimension, it is instructive to examine a similar concept from psychology used to
understand the type and strength of various emotions. Emotional valence refers to “the value
25
associated with a stimulus as expressed on a continuum from pleasant to unpleasant or from
attractive to aversive.” (American Psychological Association) The paradigmatic example from
psychology is the difference between happiness, an emotion of positive valence, and sadness, an
emotion that carries a negative valence. A similar approach might allow us to better understand and
conceptualize identity salience, a characteristic that varies spatiotemporally and that frequently has a
markedly emotional character. To take an example, national identity and one’s sense of it are
frequently affective— national identity is constructed and exercised in a way that corresponds with
the nature and strength of various emotions, such as pride or fear. The emotion(s) that it
corresponds to will vary in type and intensity over time and in response to certain stimuli. Just as
there are different emotions that can be associated with one’s identity, whether positive (such as
pride) or negative (such as fear), the scale of these emotions will also vary. Both type and intensity of
the emotion (or emotions) associated with an identity represent its affective valence, and changes in
the level of affective valence can in turn lead to changes in overall identity salience.
Affect is highly variable and context-dependent, and as such there are several things that may
lead to changes in affective valence. When considering the relationship between identity salience and
political violence, there is one factor that is particularly powerful in its capacity to change affective
valence: the presence of a real or perceived threat. Under a state of threat, the affective valence of a
group identity may become both more negative and more intense for group members, as they will
likely experience heightened states of fear and/or anger toward an outgroup. While individual group
members may vary in their affective valence, they will likely experience similar levels of this
dimension at the same time, thus altering the level of affective valence for the group as a whole. In
the aftermath of an attack on a place of worship, for example, group members may experience
changes in their affective valence such that they exhibit more intense emotions in association with
their status as members of their religious group. News of the persecution or repression of an
26
ethnolinguistic group in one area of a country may lead members of the same group elsewhere to
experience anger or indignation on behalf of their ingroup members, in turn leading to a stronger
and more intense attachment to their identification with that group.
The Example of Identity Salience and American Nationalism
When taken together, the three different dimensions make up identity salience. Primary
subscription represents the identity with which an individual most closely aligns at a given time,
while cognitive prioritization and affective valence measure the degree to which a person attaches to
that identity both cognitively and emotionally. Not only might an individual identify first as a
member of a certain ethnic group to the exclusion of other possible identification categories—for
example, their religious identity or national identity—but they may also experience heightened
attachment to this group and thus potentially have a lower threshold at which they might be willing
to adapt their behavior in pursuit of group goals, whether these goals are the achievement of higher
group status relative to other groups or the protection of the group from a perceived threat.
Returning to the previous example of national identity might be illustrative in thinking about
how these dimensions fit together. Prior to the terror attacks of 2001 against the US, American
national identity was arguably less salient than it was in the period following the attacks. Many
Americans altered their primary subscription to their national identity, considering themselves first
and foremost as Americans. Cognitive prioritization of an American national identity arguably
increased among many members of the group, something that was simultaneously demonstrated and
reinforced by such practices as flying American flags and displaying more explicit support for such
causes as supporting military action abroad. The tragedy also evoked intense emotions of fear and
anger among many Americans, increasing the affective valence of their national identity. Altogether,
changes in the three dimensions of primary subscription, cognitive prioritization, and affective
27
valence led to the heightened salience of American national identity in the aftermath of 9/11. This
was easily manipulated by politicians to rally support for certain actions in the name of patriotism
and protecting the homeland, actions that were facilitated by the activation of a strong national
identity. Examples such as this underscore the point that identity salience may be very important in
affecting or determining political outcomes, meriting closer and more systematic examination.
I do not assume that different types of identity have the same effects on political behavior.
However, a core idea is that when an identity increases in salience, a subsequent change in behavior
becomes more likely. One of these behaviors, which will be discussed in detail in the next chapter, is
participation in political violence.
This conceptualization adds some key considerations that are either excluded or
underemphasized in previous accounts. One unique aspect of identity salience as a variable is its
ability to capture change over time, and another is its ability to bridge the analytical gap between a
group and its constituent individuals. Of course, it is not always the case that the salience of a group
identity will increase for all individual members of the group, let alone to the same extent for each
individual. It is theoretically plausible, however, that changes in identity salience at the individual
level may lead to change at the collective level, particularly when this change is experienced by
multiple members of the ingroup over a relatively short period of time. Identity does not exist in a
vacuum, and one shortcoming of Abdelal et al.’s analytical framework is that it does not facilitate the
examination of how group identity attachments change over time for individuals. The
conceptualization presented here aims to bridge this gap. Finally, this conceptualization makes some
progress toward incorporating group membership (primary subscription) and consciousness
(cognitive prioritization and affective valence) within the same variable (Spry 2021). How, though,
can we begin to approach the measurement of such an admittedly nuanced and multidimensional
variable? Though this conceptualization is amenable to both qualitative and quantitative
28
operationalizations, I focus on a novel measurement strategy for identity salience as a latent variable
using Bayesian inference.
Chapter 1: Measuring Identity Salience as a Latent Variable
There are many possible ways to measure identity salience in accordance with the
conceptualization outlined above, ranging from qualitative interviewing to survey methods to
discourse analysis. Thus, the measurement strategy proposed here is by no means the only way to
measure identity salience; rather, the goal of the strategy I provide is to lay out a novel and
generalizable method for measuring the salience of any kind of social identity in a way that facilitates
quantitative analysis and the incorporation of identity salience as a variable in statistical models.
The complexity of identity salience as a variable and the fact that it is difficult to observe
directly means that more conventional approaches to measurement might fall short in linking
concept to construct. Measurement models, and specifically latent variable models, are an important
tool for linking theoretical concepts to data-generating processes and to the data that these processes
produce (Fariss et al 2020). I therefore propose a replicable and generalizable method of measuring
identity salience as a latent variable using a Bayesian item response model. The measurement strategy
I develop here can be used to ascertain a baseline measure of identity salience that can be adapted
and generalized to a variety of applications, whether identity salience is being used as an independent
or dependent variable.
Because there are multiple types of identity and the core aim of this model is to showcase a
method that can be adapted to different contexts, I narrow the present focus to a measurement
model of national identity in the American case. This model applies item response theory to original
survey data that I collected as part of a survey experiment via Amazon’s MechanicalTurk (MTurk)
29
platform.
13
The survey asks respondents a series of questions that aim to capture the dimensions of
identity salience expected to be theoretically important and constitutive of identity salience as I have
conceptualized it: namely, primary subscription, cognitive prioritization, and affective valence.
Survey Design and Question Wording
Given the multidimensional nature of identity salience as a variable, surveys are an ideal way
to design questions that capture each of these dimensions for each individual respondent. Indeed,
the utility of survey methods for measuring various aspects of identity is well-established (Abdelal et
al 2009, p.4). Surveys provide an opportunity to ask specific questions meant to tap into different
aspects of a respondent’s identity, and the anonymity afforded by many surveys reduces the risk of
social desirability bias that might otherwise occur when respondents are asked sensitive questions
about identity.
I fielded an anonymous survey on a convenience sample of approximately 2000 US-based
respondents on Amazon’s Mechanical Turk (MTurk) platform in June 2022.
14
The survey asked
respondents to answer a series of questions designed to uncover the salience of their American
national identity as a latent trait. In addition to collecting basic demographic information to be used
as covariates, respondents answered questions designed to ascertain their primary subscribed
identity, level of cognitive prioritization, and affective valence specific to their national identity.
Rather than directly asking respondents how important their identity was to them, these questions
were designed to gather crucial insights into identity salience as a latent trait by asking questions that
13
This measurement model uses data from the initial iteration of a survey experiment I present in the second article. For
more details on sample selection and design of the survey experiment, please see that article. This article focuses solely
on measuring identity salience as a variable and only provides as much detail as is required to present the measurement
model.
14
The study was approved by the University of Southern California’s Institutional Review Board in December 2021
(Study ID: UP-21-01039). The full survey instrument, including the informed consent disclaimer and the protocol
submitted to the IRB, are available in Appendix 2.
30
correspond with theoretically important concepts as they relate to identity salience. The directly
observed way that individuals respond to certain items about their identity and how they conceive of
and attach to it indicates an underlying latent trait of identity salience. The figure below gives a visual
representation of how the theoretical concepts of primary subscription, cognitive prioritization, and
affective valence inform the construct of identity salience as a latent variable, and how observed
responses Y to the survey items k inform the value 𝜃 for each survey respondent i.
Figure 1.1: Visual Representation of Identity Salience as a Latent Variable
To capture primary subscription, I asked respondents to rank a list of different group
identities in order of importance, from highest to lowest. Response options included religion,
race/ethnicity, nationality, gender, sexual orientation, partisanship, and class, with the understanding
Primary
subscription
Affective
valence
Cognitive
prioritization
Identity
Salience
𝜃 ij
Y ik
31
that the identity group ranked first is the respondent’s primary subscribed identity at the time of
answering. Though ranking a particular group identity lower than first does not indicate that the
salience of that identity is low, we could plausibly claim that those identities ranked lower are less
salient than those ranked higher. The survey also asked respondents to indicate their level of
agreement or disagreement with the statement that their identity as an American is central to who
they are.
Primary subscription provides important insight into which group identity an individual
considers to be most salient but is only one piece of the puzzle. Also important is how strongly
individuals attach to their primary subscribed identity, and to what extent their identity informs their
preferences and thought processes. To measure cognitive prioritization, I asked five multiple choice
questions to gauge the extent to which respondents consciously center their identity in making
certain decisions and expressing policy preferences.
15
For example, one question asked respondents
to determine how important (or unimportant) it is that K-12 education curricula “include instruction
on American history, culture, and values.” If an individual primarily subscribes to their national
identity and attaches a significant level of importance to this identity, then we could theoretically
expect the same individual to answer that this is at least moderately important. Conversely, if
national identity is ranked low in an individual’s response to the primary subscription question, then
that individual is less likely to indicate that the inclusion of instruction on American history, culture,
and values in K-12 education is highly important. Other questions used to measure cognitive
prioritization ask respondents to indicate their preferences on basic trade, border, and foreign
policies. Though nationalism and patriotism are not the same and should not be conceptualized as
such (Furia 2002, Smith and Jarkko 2001), they constitute two sides of the same coin in my
15
I explain the survey instrument and flow in article 2 but outline the questions relevant to identity salience here for the
purpose of explaining the measurement strategy for identity salience specifically. For a full explication of the
methodology and design of the survey instrument, see article 2.
32
conceptualization of identity salience. In other words, someone whose national identity is highly
salient will likely indicate high levels of patriotism while also supporting nationalist policies, either
explicitly or implicitly.
16
The survey therefore asked respondents to indicate their agreement or
disagreement with statements asserting a nationalist policy stance, such as “One of the government’s
top priorities must be to protect its borders at all costs.” Again, while responses for those with high
identity salience need not necessarily be uniform across all cognitive prioritization questions,
responses indicating support for nationalistic policies and those who believe that such things as
including American cultural education are highly important are correlated, and vice versa.
17
The final dimension, affective valence, captures the presence, strength, and nature of
emotional attachments to national identity. As with the dimensions of primary subscription and
cognitive prioritization, I expect that individuals with high national identity salience would indicate
high affective valence as well, meaning that they would express higher levels of both positive and
negative affect. These questions were designed to capture the nature and intensity of respondents’
emotional association with their national identity, borrowing from the psychology literature on the
measurement of emotion using the Positive and Negative Affect Scale (PANAS). I designed
questions intended to gauge respondents’ levels of positive and negative affect as it relates to their
American national identity, focusing on optimism and pride for positive affect and on anger and fear
for negative affect.
18
Although positive and negative affect as measured by PANAS are distinct, they
16
When I refer to patriotism here, I do not mean nationalism; rather, I mean expressed pride in one’s country that can
be independent from nationalism and that is not necessarily attached to specific policy preferences; relative to
nationalism, patriotism is likely to be more politically neutral. For example, one who expresses patriotism is not
necessarily a nationalist, but someone who is a nationalist will be more likely to both express exclusionary policy
preferences favoring Americans while also indicating high levels of patriotism.
17
A correlation test of all manifest variables shows that all items are positively correlated. The exception is that there is a
weak negative correlation between responses to the questions about fear and optimism, which comports with theoretical
expectations, as these emotions are on opposite ends of the affective scale.
18
I do not administer the PANAS test for respondents, nor do I include questions that measure all of the 20 emotions
measured through the PANAS instrument. Instead, I include only those questions that are sufficient for measuring
general affect as it might relate to identity (positive and negative).
33
are not mutually exclusive—high scores on positive affect are possible at the same time as are high
scores on negative affect (Watson et al 1988). The questions therefore measure such things as
national pride, level of optimism for the direction America is headed, and fear for future generations
of Americans, as well as attitudes toward those who violate certain norms considered important to
the ingroup, such as respecting the flag. Rather than asking about approval or disapproval, however,
this question asks respondents to indicate their agreement or disagreement with the statement that
“People who burn the American flag should be subject to a fine or a legal punishment.” Rather than
simply gauging approval or disapproval of an action that might offend someone whose national
identity salience is high, this question’s wording is intended to uncover the intensity of the offense
taken—if a respondent’s national identity salience is high and involves a certain level of emotional
intensity, then they would be more likely to respond in a relatively extreme way to someone who
burns the American flag (i.e. expressing support for legal consequences). The conceptualization of
identity salience accounts for this, but also holds that those who demonstrate both high positive and
negative affect in response to cues about their national identity harbor more intense emotional
attachments. Therefore, while individuals might express high positive affect (measured here by
national pride), those who are more intensely attached to their identity will also demonstrate high
negative affect. If someone is extremely proud to be an American, they may also indicate strong
agreement with a statement that individuals who disrespect a prominent symbol of their American
national identity should be punished, indicating high negative affect. Strong responses to the affect
questions—i.e., “strongly agree”— would indicate higher emotional intensity, and thus higher
national identity salience, than would more muted or neutral responses, such as a “somewhat agree.”
In other words, the emotions themselves are less important than their intensity, which these
questions were designed to measure.
34
The Model
The survey instrument was designed under the assumption that responses reflect an
underlying latent trait of identity salience that is not directly observable. I apply item response theory
(IRT) to understand the relationship between the latent trait and the observed data from the survey.
Item response theory is a Bayesian method used to predict the likelihood that an individual will
select a certain response based on some underlying variable that is not directly observable; in this
case, the level of the latent trait of identity salience predicts an individual’s responses to the relevant
questions. In formal terms, the measurement model measures the likelihood of an observed
response as a function of the value of item parameters 𝜷 for some unobserved quantity 𝜽. Long a
methodological mainstay of psychometric research, political scientists have increasingly recognized
the applicability of IRT and other Bayesian methods for measuring as latent variables theoretical
constructs that are difficult to observe directly, such as political ideology (i.e. Barbera 2015; Pan and
Xu 2018; Treier and Hillygus 2009), democracy (i.e. Treier and Jackman 2008, Pemstein et al 2010),
political attitudes (i.e. Blaydes and Linzer 2008), or respect for human rights (Fariss 2018 a, b).
Identity salience is one such variable that is theoretically important in political science, yet is difficult
to observe directly, and so existing attempts to account for it lack construct validity. This
measurement strategy is an attempt to rectify this.
I model survey question responses as a function of the item parameter for the latent variable
of identity salience, identifying a polytomous item response model with unconstrained
discrimination, or a graded response model (Samejina 1969). Such models are often used for ordinal
response data and operate similarly to standard two parameter logistic (2PL) IRT models for
dichotomous responses. Because the different survey questions, or items, are not expected to have
equal discrimination parameters across items—in other words, they are not uniformly predictive of
35
the latent trait—I specify unconstrained discrimination, meaning that discrimination parameters will
vary. The basic model equation is the following:
Y
ik=
𝜶
ik +
𝜷
ik
*𝜽
ik
where Y ik is a random variable that indicates the probability of observing a particular response for
each individual i for each item k. 𝜶 represents the difficulty parameter for each question in the
survey, 𝜷 represents the discrimination parameter, and 𝜽 represents the predicted value of the latent
trait of identity salience. The difficulty parameter indicates the likelihood of a certain response being
observed given a zero value of the latent trait, and can be otherwise thought of as the intercept value
in a linear model; the discrimination parameter is the strength of the relationship between the latent
trait 𝜽 and response Y, such that a one-unit increase in 𝜽 corresponds with an increase of 𝜷
(roughly analogous to the slope coefficient on an observed value x in a linear model).
19
𝜽 is our main
parameter of interest and reflects the value of the latent trait for each individual.
Prior to fitting the model, I examine the underlying structure of the data to determine
whether any items seem to be outliers so that I can remove them from the model. Dimensionality
reduction techniques, such as principal component analysis (PCA) and confirmatory factor analysis
(CFA), are useful methods for understanding the underlying structure of data. I conducted a
principal component analysis of the data to look at how each survey item loads on different
dimensions, or new linear combinations of the data. As expected, a majority of the variance in the
data can be explained by the first principal component. I also conducted a confirmatory factor
analysis (CFA), which confirmed that one factor is sufficient for explaining a majority of the
19
Note that in a graded response model, there are multiple intercepts according to the number of items in the model.
36
variance in the data. One surprising result was the relatively weak correlation of one question in
particular with the primary factor and with other questions—the question asking respondents
whether they fear for the future of America. Despite showing a weak relationship with the primary
factor, it is a positive one, and I choose to keep it in the model.
There are several software options for running a graded response model. I run the
measurement model using MCMCordfactanal in MCMCpack, an R package used for Bayesian
inference (Martin et al 2011). MCMCordfactanal takes an ordinal data factor analysis model as a
basis for generating a sample from its posterior distribution, with normal priors assumed for the
factor loadings.
20
I run the model using 100 Markov Chain Monte Carlo simulations, with normal
priors assigned to the model parameters. As expected, the posterior means are normally distributed
around a mean identity salience score of zero. Individuals with higher scores are understood to have
high identity salience, while those with lower scores are understood to have lower identity salience. I
do not report the posterior means of the latent trait here, as this is immaterial to the explanation of
the measurement model and the methodological reasoning behind it.
21
Chapter 1: Discussion
The proposed model is one measurement strategy for quantifying identity salience as a
variable. Though it employs a well-known Bayesian method, it is an easily generalizable means of
measuring the salience of different types of identity and can be adapted to a variety of contexts. As
20
This model could just as easily be run using Stan or BUGS. Given the relatively small sample size being used in this
example, it is possible to run multiple MCMC iterations sampling from the posterior distribution without vectorizing the
parameters beforehand. For larger sample sizes, it would be advisable to run the same model using another software,
such as Stan or BUGS, in order to avoid computer crashes due to insufficient memory storage.
21
I include an R Markdown file in Appendix 1 to demonstrate the main code behind the measurement model. Full
replication files can be made available upon request.
37
one example, a researcher trying to understand individual-level attachments to religious identity
could adapt the questions to be specific to that type of identity: rather than asking respondents how
important it is that their child’s education include instruction on American culture, it could ask how
important it is that their child’s education include religious instruction. Studies measuring the
salience of a particular ethnic identity might consider adding questions that focus on such things as
attitudes toward hypothetical political leaders belonging to a different ethnic group than themselves.
Some potential caveats of this method should be addressed. First, some level of knowledge
of the identity being measured is necessary if the researcher is to design questions that accurately
capture the salience of that identity, and in this way the design of a study that measures identity
salience quantitatively would likely also require some qualitative research to inform question
wording, sampling, and survey design more than in other studies using similar methods.
Additionally, it is especially important to design questions that do not bias responses by accidentally
heightening the salience of the identity being measured, an issue that pre-testing may mitigate to
some extent. Still, these limitations may also be seen as strengths if they are properly addressed, as
doing so could lead researchers to develop more sophisticated instruments that are effective at
capturing the complexity and nuance of identity salience in ways that existing approaches to
measuring similar variables have not been.
In addition to more fully addressing these issues, future related studies might adapt their
models to allow for the measurement of change over time. The use of dynamic latent variable
models, rather than the static model used here, would be useful if panel data is available.
Additionally, and in recognition of the fact that individuals rarely have only one salient identity at a
given time, future models that incorporate attachments to multiple identities at the same time would
be a welcome extension to what has been presented here. Finally, future observational studies
attempting to understand the true distribution of identity salience scores throughout a population—
38
for example, a study measuring the average level of national identity salience among all Americans—
should make every effort to prioritize representative sampling. Using a convenience sample in this
case was not problematic because it was conducted as a survey experiment. However, future
researchers who use this method to understand how attached individuals are to a particular group
identity outside of experimental contexts must be especially careful to ensure the sample matches
the population to the greatest extent possible, especially given that the accuracy and utility of this
method rely heavily upon a prior distribution that is as reflective as possible of reality. Though this is
a caveat common to any study employing Bayesian inference, it is worth noting here.
This article outlined a detailed conceptualization of identity salience and provided a novel
measurement strategy of identity salience as a latent variable. The next article builds upon the
theoretical discussion begun here and extends it to present a theory of identity salience and political
violence at the microlevel, generating observable implications that will be empirically tested using
survey experimental data.
39
CHAPTER 2: IDENTITY SALIENCE AND POLITICAL VIOLENCE: A MICRO-LEVEL
THEORY WITH EVIDENCE FROM THE UNITED STATES
A recently published book by prominent scholar of civil war and political violence scholar
Barbara F. Walter asks the question of how civil wars start, using rising political polarization and
violence in the United States as a point of departure to examine the risk factors for civil war and
how to preemptively mitigate these risk factors. Much of the book focuses on uncovering the causal
pathways between “sporadic violence” and full-scale civil war, examining the steps along the
spectrum from low-level violence to full-scale war (Walter 2022, p. 156). While important, such a
question elides the importance of examining what leads otherwise law-abiding individuals to engage
in political violence in the first instance, whether it is a sporadic act of violence or a coordinated one.
There are many explanations as to what motivates such behavior, but one that I argue is
simultaneously undervalued and of outsize importance is the role of changes in identity salience.
Building on the conceptualization of identity salience advanced in the previous article, this article
proposes a theoretical framework for considering whether, when, and how changes in identity
salience may lead to the formation of violent political attitudes among ingroup members and
presents empirical evidence testing some of the framework’s hypotheses in the case of the
contemporary US. It is important to note, however, that I do not propose this as a monocausal
explanation for political violence, nor do I assume that all identity group divisions are inherently
prone to violence; rather, this is one potential explanation for what might lead individuals to engage
in political violence on behalf of an ingroup. I propose a theoretical framework that sheds light on
this process while also accounting for previous misperceptions about and mismeasurements of
identity salience. This article focuses exclusively on political violence attitudes as the outcome.
40
Political scientists have focused extensively on the determinants of participation in civil war.
While some emphasize the role of ethnic grievance
22
in causing conflict, others point to opportunity
(or greed) as the key determinant behind whether a group or an individual will choose to fight
(Collier and Hoeffler 2004, Fearon and Laitin 2003). Still others examine the potential interaction of
these factors in prompting mobilization in civil war. In the case of the civil war in Sierra Leone,
Humphreys and Weinstein (2008) conclude that the three competing theories of rebel mobilization
that they test—grievances, selective incentives, and social sanctions—may not actually be competing
theories but might instead interact in creating the conditions in which a potential rebel may take up
arms for or against the state.
Much of this scholarship relies on the assumption that the conflict in which the individual is
choosing to participate has already been initiated, and that there is some level of coordinated
violence already ongoing. And, in conflicts where the actors have a significant identity-based
dimension, these identity attachments are largely taken as a given, with relatively little attention given
to the question of how and why these identity attachments become politically relevant in the first
place. Before hostilities are initiated and armies formed, what sparks the initial political violence that
can begin the transformation of a society at peace to one at war? Existing theories of political
violence tend to treat it as a collective outcome, obfuscating the role of micro-level processes and
outcomes. The theory I propose here focuses on such micro-level processes and outcomes as they
relate to identity salience.
I define political violence here as any act of violence where physical harm is inflicted or
attempted against another person, group, or target in connection with political goals. Political
violence varies widely in terms of severity, ranging from sporadic acts of violence at the low end to
22
Competing accounts maintain that grievances are, in fact, a key causal factor in instances of political violence with
ethnic or religious dimensions. Sambanis (2001), for example, argues that grievances are important in “identity wars” but
not in other types of civil war.
41
civil and interstate war at the high end.
23
(Kalyvas 2019, Gade 2020) Political violence may
encompass anything within the realm of what may be considered violent contentious politics—the
events that happen outside of voting, organizing, and other nonviolent and socially accepted political
action (Zeitzoff 2022).
Though this is a broad spectrum of potential outcomes, a core contention of this article and
the theory it presents is that political violence related to a social identity almost always involves a
common factor. This factor is changes in identity salience, which I claim is a determinant of political
violence at the microlevel. Rather than viewing identity as fixed, I consider it to be a causal variable
best captured as identity salience. Building on the conceptualization detailed in the previous article,
this one offers a micro-level theory of identity salience and attitudes toward political violence.
24
A few notes on scope conditions and what this theory does and does not address are in
order. First, I discuss political violence solely as an action of ordinary non-elites, rather than as a
tactic or an ongoing process among combatants in an ongoing conflict or war. Indeed, prominent
scholars and intellectuals writing on political violence have noted the importance of decoupling
violence from other cognate concepts, like war or conflict (Arendt 1970; Horowitz 2001). I seek to
do the same here. The theory I present here attempts to understand what leads individuals to change
their attitudes toward political violence in the first instance, rather than to understand such
phenomena as, for example, the logic of indiscriminate violence in civil war (i.e., Kalyvas 2006).
Relatedly, although identity can and certainly does play into conflict processes after onset, the
primary goal of this framework is to better understand the identity-based attitudinal micro-
foundations of individual-level participation in political violence. As such, the theory applies to
23
I spend more time on severity in political violence outcomes in the third article of the dissertation.
24
Articles one and two focus on identity salience and political violence at the microlevel, examining the relationship
between identity attachments among individuals and the propensity to support or engage in political violence. Article 3
of the dissertation focuses on the relationship between identity and political violence at the macrolevel and is therefore
not covered in this article’s theoretical discussion.
42
ordinary individuals who have not previously engaged in organized political violence, and focuses on
attitudes toward lower-level acts such as participation in violent riots or unstructured individual
violence.
25
The theory is agnostic as to whether these first instances of political violence would
precede further or more severe instances of political violence, though the role of identity and
identity salience at the macro-level will be considered in article three.
If identity salience does play a causal role in motivating individual participation in political
violence, as I argue that it does, how and why is this the case? Is there something especially powerful
about identity that makes it a variable capable of leading otherwise law-abiding citizens to support
political violence? And if so, do different identity types have differential effects in political violence
outcomes? I argue affirmatively on both questions and advance a theoretical framework for
understanding why.
The previous article outlined a conceptualization of identity salience and proposed a novel
measurement strategy for identity salience as a variable for general applications to political science.
The theory outlined in this article builds upon prior work on the psychosocial mechanisms of
political violence writ large and holds that identity salience is especially susceptible to change among
non-elites and can be easily manipulated by political elites to elicit a response. When the salience of a
particular identity increases, the threshold at which individuals are willing to engage in political
violence on behalf of their identity group declines, increasing the likelihood that violence will result.
Though it is important to note that changes in identity salience will not uniformly lead to violent
attitudes, I argue that the often emotional and exclusionary nature of identity and the means by
25
Emily Gade (2020) introduces a theory of what she calls unstructured individual violence, which she defines as
“political violence without formal organization or structure, (or) violence individual civilians—rather than organized
armed actors or groups—conduct.” As such, it is “sporadic, often geographically distributed, and hard to predict because
it is often not focused against key strategic targets or part of a (clearly articulated) broader political strategy.” (p. 309)
Though she provides this definition in the context of political violence in Israel and Palestine, hers is a valuable
contribution to the broader definition of political violence and can be generalized across contexts where such violence
occurs.
43
which identity is activated narrows the decision space for ingroup members in the face of real or
perceived threats from an outgroup, whether these are physical threats, like a targeted attack against
an ethnic neighborhood, or non-physical threats, like a loss of relative group status due to rapid
demographic shifts.
The psychosocial components of identity render it a factor particularly susceptible to rapid
change and immune to more rational appeals for restraint; further, individual perceptions of a threat
to one’s ingroup from an outgroup can create the impression that the costs of nonparticipation
outweigh the risks of participation. When individuals are led to attach more strongly to a particular
ingroup identity and identity salience increases as a result, these same individuals become more
willing to adopt permissive attitudes toward political violence in defense of the ingroup, if not to
express a willingness to engage in political violence themselves. When many individuals from the
ingroup experience shifts in identity salience concurrently, whether this occurs in conjunction with a
high-profile event of importance to the identity ingroup in question or as a result of some other
factor, the possibility of group-level violence increases. When activated, then, identity salience can
increase the risk of political violence by leading individuals to consider it a justified response to real
or perceived threats to their ingroup when they may otherwise consider such actions to be
unacceptable.
26
I further argue that some types of identity are more manipulable and combustible
than others.
To be sure, political violence is not the only possible outcome of increased identity salience,
and the intent of this argument is neither to essentialize identity nor to suggest that nonviolent
strategies are impossible in conflicts with significant identity dimensions. At the same time, identity
has several characteristics that render it particularly prone to emotional reaction and manipulation,
26
In using language discussing increased risks of political violence, I do not mean to suggest that this article is testing
behaviors. The key dependent variable relates to attitudes toward political violence, and makes no causal claim about the
connection between supportive attitudes toward political violence and actual violent political behavior.
44
meaning that shifts in identity salience may lead individuals to be more supportive of violence where
they may not have otherwise been supportive. More closely examining whether, how, and why
identity salience can 1) vary over time, 2) be manipulated by others, and 3) lead individuals at the
micro-level to consider political violence to be an acceptable response is the main goal of this
theory.
The remainder of the article is structured as follows. I first lay out the puzzle that the theory
attempts to address, drawing on relevant literature on political psychology and political violence in
the process. I then outline the theory and empirically test key hypotheses derived from the theory
using evidence from a survey experiment in the United States.
Chapter 2: Literature Review
Political violence is costly. Aside from the danger of loss of life and limb, participation in
political violence can lead to criminal prosecution and incarceration, persecution by the incumbent
government or regime, and risks to one’s livelihood, family, and community. It is often economically
disastrous and can lead to longer-term social and political instability, up to and including total war.
Not only is political violence costly, but it has been shown to be less effective than nonviolence in
achieving desired outcomes. Indeed, studies of nonviolent resistance have found that such
movements tend to be more successful than violent insurgencies, not least of all because the former
has what Chenoweth and Stephan call a “participation advantage.” (2011, p. 10). One recent
example of a social identity-based movement that was mostly nonviolent—racial justice protests by
Black Lives Matter in the wake of George Floyd’s murder by Minneapolis police in June 2020—
demonstrates not only the utility of nonviolent resistance movements in advancing group goals, but
also underscores the point that violence is not inevitable when identity is a main organizing
principle. Though not all outcomes of identity mobilization are violent, individuals acting on behalf
45
of a social identity group operate under a different calculus, especially when the group in question
perceives its relative group status to be under threat. In such scenarios, ingroup members may
experience precipitous changes in the salience of their identity such that the barrier to participation
in political violence is lowered—not because the risks of participation have decreased, but because
the perceived cost of nonparticipation have increased. Perceptions of such costs are heightened by
the emotional intensity of intergroup tensions, leading ingroup members to consider options
previously thought of as irrational or socially unacceptable to be a rational response to a perceived
threat.
27
The broader political psychology literature contains useful insights that can be applied to
understanding the link between identity salience and political violence, particularly those literatures
that focus on the political psychology of threat perception, emotion, and intergroup conflict. What
can we take from existing accounts to help us better understand when and why changes in identity
salience can lead to political violence?
Social identity theory (SIT) is useful in thinking about identity salience as a variable and is
especially so when considering the effects that differential group identification, or perceptions of
group status relative to other groups, can have in shaping behavior. Indeed, multiple psychological
studies have found that the perception of a threat to group status—whether this be economic,
social, political, or even physical—can affect how that social identity affects behavior (i.e., Dovidio
and Morris 1975; Hayden et al 1984; Flippen et al 1996; Branscombe et al 1999a, b). The loss or
perceived loss of relative group status can heighten individual attachment to identity, as the
awareness of a threat from an outgroup is enough to lead ingroup members who may not otherwise
strongly attach to an identity to consider that same identity to be more important. For example,
27
The calculus of nonviolent versus violent action is not equal across groups. Indeed, a recent cross-national
experimental study of the Black Lives Matter movement in the US and Palestinian resistance in Israel showed that ethnic
minorities are disproportionately perceived as violent, even when using nonviolent tactics (Manekin and Mitts 2021).
46
there was a marked increase in identification with Ukraine as a homeland after the Euromaidan
protests in 2014, during which time Russian president Vladimir Putin launched an invasion into
Crimea and began a sustained insurgency in eastern Ukraine’s majority-ethnic Russian Donbas
region (Pop-Eleches and Robertson 2018). Exposure to violence committed against a particular
social group, or groups, is one way that group members may come to believe that they and their co-
members are under threat. Related work has shown how exposure to societal violence can lead to
increased intergroup polarization by raising threat perception and affecting risk-related cognition
(Canetti et al 2013). In the context of Israel and Palestine, Sambanis and Shayo (2013) show that the
threat posed by ongoing violence has the potential to increase group identification patterns to the
point that what they call a “social identity equilibrium” shifts, leading to increased polarization and
cycles wherein ethnic identification and violence can be mutually reinforcing. In this model,
increased ethnic identification and violence exist in a sort of feedback loop. In the Sahel, exposure to
communal violence was shown to increase support for violent religious extremism (Finkel et al
2021). Another study in the context of ethnic identification in Jammu and Kashmir found that state
repression of the regional minority caused them to attach to their ethno-regional Kashmiri identity
more strongly due to the activation of an affective mechanism, whereby the salience of the ethno-
regional identity increased relative to the salience of Indian national identity (Nair and Sambanis
2019).
Clearly, threat perception and threat exposure are psychologically and politically powerful,
but why? The power of threat in large part comes from the emotional responses it evokes. Emotions
frequently associated with threat, such as anxiety, fear, and anger, have been shown to be important
in motivating various political behaviors. Albertson and Gadarian note the double-edged sword of
anxiety in democratic politics, arguing that while political anxiety can trigger productive engagement
in society under certain conditions, it can also be manipulated by partisan elites in such a way as to
47
threaten democracy; this is especially true, they argue, when anxiety “prioritizes attention to
threatening information.” (Albertson and Gadarian 2015). Similarly, Merolla and Zechmeister (2009)
demonstrated that threats of terrorism can motivate how citizens in democracies respond to these
threats with heightened mistrust against societal outgroups, among other things that bode ill for
long-term democratic health and sustainability.
Of course, not all individuals will have the same predisposition to manipulability based on
threat perception. Some individuals have certain personality traits that condition both their
sensitivity to potential threats (i.e., Baker 2021; Dallago and Roccato 2009; Sibley and Duckitt 2009;
Adam-Troian et al 2020) and their propensity to engage in political violence (Kalmoe 2014). Though
individuals may respond to threat in different ways, that they do respond is key. As Landau-Wells
argued in her introduction of Threat-Heuristic Theory (THT), the psychological and cognitive
processes underlying human responses to perceived threats are species-typical—in other words, they
are “basic features of the human mind.”
28
Intergroup emotions theory has explored the relationship between emotional reactivity and
social identity, with evidence that individuals who identify strongly with a particular social identity,
or those who might have strong identity salience, tend to react with more anger to group threat than
do others with weaker identifications (Musgrove and McGarty 2008; Rydell et al 2008; van Zomeren
et al 2008). For example, one study found that individuals identifying strongly with their American
national identity exhibited angrier emotional responses toward terrorists in the days leading up to the
2003 invasion of Iraq (Feldman et al 2012), while other studies found that individuals responded
more angrily to insulting messages about the US and Americans when those messages were written
by foreigners (Rydell et al 2008).
29
Exposing individuals whose identity salience has been
28
Landau-Wells, p. 38.
29
A more recent study in the context of exposure to anti-Chinese discrimination and attitudes toward the Chinese
authoritarian regime comes to a similar substantive conclusion (Fan et al. 2020).
48
experimentally heightened to violent frames has also been shown to increase anger among ingroup
members in response to a threat or victimization (Yzerbyt et al 2003). In a related example, British
subjects whose national identity had been made more salient through experimental manipulation
indicated higher feelings of aggression in response to photographs of the July 2005 London
bombings. These same subjects with higher national identity salience were more likely to indicate
greater support for the war on terror than were those whose gender identity had been made
experimentally salient (Fischer et al 2010). Not only can perceived threats to an ingroup prompt
angry emotional reactions among strongly identified group members, but identity and anger can be
mutually reinforcing (Thomas and McGarty 2009; Thomas et al 2009).
These emotions are powerful on their own but can be especially powerful when used as a
tool of manipulation by political elites working to achieve political ends. In democracies, political
anxiety can be leveraged by partisan elites to garner support for certain policies that emphasize
security but that may also embody anti-democratic tendencies (Albertson and Gadarian 2015). In
other contexts, emotions can serve as strategic resources when manipulated by political
entrepreneurs attempting to counter the structural resources employed by opponents (Petersen
2011). Drawing on examples from conflict in the Balkans, Petersen argues that certain experiences
leave behind powerful emotional “residues” that strategic actors can manipulate to “change the set
of actors, reshape preferences, and alter the rules” by selecting provocative actions that activate
these emotional residues. These actions might include targeted or indiscriminate bombing,
destruction of a sacred site, or any other action intended to evoke strong emotional responses.
30
The salience of social identities, then, can be enhanced by real or perceived threats that
prompt strong emotional responses. As demonstrated by SIT starting with Tajfel and Turner,
30
Petersen 2011, p. 15. On the various “action tendencies” associated with different emotions, see Frijda 1986. Jon
Elster provides an exhaustive conceptualization of emotions as causal mechanisms for human behavior. See also Elster
1998a and 1998b and McDermott 2019.
49
inciting intergroup tension is relatively easy. Several theories that have subsequently been developed
have attempted to explain why this is. Realistic Group Conflict Theory, for example, holds that
material factors and competition over material resources is key to motivating threat perception
between groups (i.e., Sherif et al. 1961; Levine and Campbell 1972). On the other hand, Symbolic
Threat Theory prioritizes competition over value systems and beliefs as the main source of
intergroup tension (Kinder and Sears 1981). Intergroup Threat Theory takes both sides, arguing that
the combination of competition over both material and non-material resources, and the perception
of threat that arises from this competition, drive intergroup tensions (i.e., Stephan et al. 2009;
Stephan and Stephan 2000; Stephan et al. 2002).
Social identity is complex and associated with many different processes and outcomes, one
of the most consequential among these being political violence. The following theoretical framework
builds on the existing literature to begin to answer how and why changes in identity salience lead to
more supportive attitudes toward political violence.
Chapter 2: A Micro-level Theory of Identity Salience and Political Violence
There is compelling empirical evidence that identity can be a powerful variable in shaping
political attitudes and behaviors. But what exactly is it about identity that makes it so potentially
powerful in shaping violent attitudes and outcomes?
At the most basic level, identity salience is a causally important variable because identity is
psychologically powerful, subject to relatively rapid shifts over short periods of time and can be
manipulated by elites to drive changes in political attitudes as a result. Identity itself may not lead
individuals to change their attitudes, but changes in the salience of that identity can certainly do so.
The theory presented here attempts to capture how and why this is. Being a member of a group, or
50
even of multiple groups simultaneously, does not preordain that such individuals will be led to act
on that identity. However, when something causes a person to attach more strongly to a particular
identity, it becomes more plausible that such a change in salience would affect attitudes relevant to
that identity. Identity salience, then, is a variable that allows the causal role of identity to be
considered in both quantitative and qualitative analyses at the individual level. Indeed, social identity
theory and empirical tests of its core precepts have shown that even minor invocations of group
identity can lead to changes in intergroup behavior, even when the group identity in question is
superficial.
31
It follows, then, that in situations where a group identity that has been long-held and
socio-politically relevant is made to be more salient, ingroup members would alter their attitudes in
pursuit of group goals. Though these changes in political attitudes do not uniformly lead to violence,
they can be especially likely to lead to permissive attitudes toward political violence under certain
conditions. The theory looks to better explain when, how, and to what extent changes in identity
salience might lead individuals to express support for political violence when they would not have
done so before.
Some individuals may a priori have higher levels of identity salience. We might plausibly
expect that a pastor of an evangelical Christian church, for example, would generally have a higher
level of religious identity salience than the general population, just as we might expect an enlisted
member of her country’s military to have a higher level of national identity salience than her non-
military co-nationals. We could also plausibly expect that such individuals would be more likely to
support political violence committed on behalf of their identity ingroup, and that proximate changes
in their identity salience would be more likely to lead to changes in attitudes than for the general
population. All else being equal, however, the theory expects that any member of a group, regardless
of their baseline level of identity salience, would be susceptible to changes in political attitudes in
31
See the literature reviews in this article and in article one for more detail.
51
response to an event that causes their identity salience to increase, especially if this increase occurs in
a relatively short time span.
32
What causes identity salience to increase? The short answer to this question is that
experiencing a certain event or stimulus may lead an individual to change their primary subscription
and/or increase cognitive prioritization and change affective valence. Various factors can affect how
strongly individuals attach to a group identity. As a relatively extreme example, a terror attack
targeting the citizens of country X by country Y would likely increase national identity salience for
citizens of country X. At the more local level, an attack on a synagogue might increase the
religious/ethnic identity salience of Jewishness for the local Jewish community. Witnessing
celebrations for a religious holiday, as another example, could increase the religious identity salience
for ingroup members, even if they are not otherwise devout. Events such as these could increase
overall levels of identity salience by: 1) enhancing identification with that ingroup (primary
subscription); 2) increasing the extent to which that individual would actively consider that identity
in cognitive processes and decision-making (cognitive prioritization), and: 3) leading the individual
to attach either positive or negative emotions to that identity (affective valence). Though it may be
clear that such examples as these might lead to an increase in identity salience, it is necessary to think
through what factors will lead something to increase identity salience, and why.
Increases in identity salience will not always lead to changes in political attitudes at the
individual level, but they are more likely to do so under certain conditions. Ceteris paribus, two key
factors condition the outcome of a change in identity salience and determine whether identity
32
Regarding the directionality of changes in identity salience, the theory focuses on increases in identity salience. Though
it is possible that something may cause the salience of an identity to decrease, this would likely follow a different logic
than would increases in identity salience. Examining the mechanisms by which identity salience may be led to decrease,
and the effects of this decrease, is outside the scope of this work but would be a worthwhile avenue of additional
research.
52
salience will increase in such a way that the associated attitudinal outcome is more likely to be
supportive of political violence: the manner by which identity salience is increased, as well as the
type of identity in question. I explain each factor in greater detail below and provide examples to
illustrate.
Key Factor 1: How is Identity Salience Being Affected? Threat-Driven vs. Non-Threat-Driven Events
Identity salience can increase in several ways, but the manner by which it increases is
particularly important for determining whether an individual acts on this identity, and if so, whether
this action involves political violence. Key in this regard is whether the stimulus that causes identity
salience to increase leads ingroup members to feel as though their group is threatened, and if so, by
what or by whom. When something causes members of a group to consider themselves to be
threatened, whether it be physically or via a loss of group status, the increase in the salience of that
identity is more likely to be such that it inspires potentially violent attitudes or behavior in service of
protecting the group from the perceived threat. To be sure, one does not need to feel as though the
ingroup is threatened to experience high or increasing levels of identity salience
33
; instead, the
introduction of a threat to one’s ingroup is a major stimulus for increasing identity salience such that
this increase is more likely to lead individuals to be willing to act on behalf of their ingroup to
protect themselves and their group members. When the action in question involves acts of political
violence, the presence of a threat is especially vital. Thus, while the absence of a threat does not
necessarily mean that identity salience will not increase, the presence of one is more likely to increase
identity salience in such a way that it provokes a potentially violent response.
33
An example of a situation where an increase in identity salience might lead to a positive, nonviolent outcome might be
when a nation experiences a tragedy that causes members of that nation to come together—for example,
commemorations of 9/11 frequently led Americans to demonstrate increased prosocial behavior toward other
Americans, even if what came later was more virulent nationalism.
53
Why is the presence of a real or perceived threat so important in thinking about identity
salience? I argue that when an individual perceives the presence of an ongoing or imminent threat to
their ingroup, they are more likely to be aware of group differences than they were previously. This
will be especially true when the threat comes from a clear outgroup, which will lead the threatened
individual to both attach more strongly to their ingroup identity and harbor more intolerance or
animosity toward the outgroup. Returning to the simplified example of a terrorist attack (a
threatening event) versus observing festivities on a religious holiday (non-threatening event), the
absence of a threat in the latter case makes it less likely that any increases in identity salience will
contribute to negative attitudes toward outgroups. In the former example, such a clear and tangible
threat as a terror attack perpetrated against a specific group by a specific outgroup would have the
dual effect of both strengthening the victimized group members’ identification with their ingroup
(primary subscription and cognitive prioritization) and increasing their animosity toward and/or fear
of a clear outgroup (affective valence). Thus, the presence of a threat can be pivotal in determining
exactly how identity salience increases and what any changed attitudes or behaviors that result may
be.
These threats may come in different forms, which are also important in conditioning
whether there is a response and, if there is, what form that response will take—namely, whether that
response is one involving political violence. In this case, threats can fall on a spectrum from
something that is largely imagined to things that pose very real risks to livelihoods, social status,
and/or the physical safety of oneself and members of one’s ingroup. While any potential threat
theoretically could increase one’s identity salience, threats that are perceived as existential to the
group and sustained over time are most influential due to their psychological impact and, relatedly,
because such threats are more liable to manipulation by elites for political gain.
54
These real or perceived threats may be either tangible or abstract. What is important to
remember when thinking about threat, however, is that whether the threat is real or perceived does
not matter insofar as the threat is considered real by the threatened individual or group. At the same
time, not all threats will be equally impactful in shaping responses, nor will they all take the same
form. For example, a group and its members may believe that there is a threat to their livelihoods,
social status, or physical safety. All such threats have the capacity to affect identity salience, but
threats that are sustained over time, tangible, and pose a threat to physical safety are most likely to
increase identity salience in such a way that political violence against the outgroup is more likely to
occur. A simplified visualization of the theoretical likelihood of a change in identity salience given
threat type is shown in figure 2.1 below.
34
Real/physical
Tangibility of Threat
Perceived/non-physical
One-off Time Horizon of Threat Sustained
Figure 2.1: Likelihood of identity salience change to lead to political violence, by threat type
An example of a sustained and tangible threat to physical safety, the type of threat that I
claim will be most likely to increase identity salience such that political violence becomes more likely,
would be a case where one group is repeatedly targeted by an outgroup. As one example, multiple
34
I do not claim that there is a direct linear relationship between threat type and likelihood of identity salience changes
to lead to political violence, nor do I test that claim; rather, this visualization is a very simplified one meant to distinguish
between types of threats and which ones are more or less likely to lead to political violence.
55
high-profile attacks on places of worship—say, attacks against Coptic Christian churches in Egypt
by Islamist militants—might effectively increase the religious identity salience of the attacked group,
particularly if the perpetrator of the attack is known and belongs to a clear outgroup. Such a threat
provides a clear and tangible risk to the lives and safety of the ingroup, and when similar risks have
emerged over a sustained period of time it is more likely that the targeted group would view this as
an acute and ongoing threat, leading them to view stronger identification with their ingroup
members as a potential survival strategy (primary subscription and cognitive prioritization) and
increasing animosity toward the targeting group (affective valence).
35
However, more abstract threats
that do not constitute a risk to physical safety can also be powerful influencers of identity salience,
so long as a threat is perceived. For example, immigration movements can be manipulated by elites
to demonize immigrants as outgroup members who pose a threat to the way of life or the continued
social dominance of the ingroup. An increasingly popular segment on Tucker Carlson’s Fox News
show promotes the idea of the so-called “Great Replacement,” or the idea that Democrats are
purposely opening national borders so that they can change the demographic makeup of the US and
secure a permanent Democratic governing majority as a result. Regardless of the nature of the threat
being posed, what is crucial is that there is something significant at stake that could be lost, whether
this be social dominance, economic wellbeing, political power, physical safety, or even the continued
existence of a group. The propagation of such threatening rhetoric, even if the threat is
manufactured and lacking any basis in fact, can be incredibly dangerous when people begin to
believe that they are under attack.
36
Acts of identity-based political violence can result.
35
It is certainly plausible that there is an endogenous relationship between exposure to political violence and changes in
identity salience, and indeed there is empirical evidence demonstrating a sort of feedback loop of violence begetting
violence (i.e., Canetti et al 2013, Zeitzoff 2014). Though the dynamics of identity salience within ongoing cycles of
political violence are worthy of study and relevant to the theory here, it is beyond the scope of the current project. I
leave this question to later research.
36
This was written before the May 2022 mass shooting in Buffalo, NY, where a white man targeted Black grocery
shoppers in a majority Black community, killing ten and injuring three others. A rambling manifesto that the shooter
posted online before the attack pointed to the idea of a “Great Replacement,” a theory also increasingly backed by far-
56
Identity salience can increase in the absence of a threat, but I claim that such increases are
less likely to contribute to violent outcomes. Invocations of group pride or cohesion—for example,
in the form of a religious holiday celebration or a patriotic speech celebrating the shared values of a
country’s citizens—could increase identity salience, but not in such a way that it is also more likely to
provoke political violence in response. A perceived threat to an ingroup might prompt anger or fear
and spur individuals to sanction or engage in relatively extreme actions like political violence, while
something non-threatening that causes ingroup members to feel a greater sense of pride is more
likely to elicit positive affective responses. At the same time, and unlike in the case of threats, there
is no perception that something important to the group is in danger. Non-threatening cues, then, are
far less likely to inspire political violence than threatening ones.
Key Factor 2: What is the Type of Identity Being Affected?
How identity salience changes is a key consideration in thinking about whether and how
changes in it may lead to violent attitudes and outcomes. The second key factor that I propose that
determines whether a change in identity salience is more likely to lead to violence relates to the type
of identity in question. In other words, substance matters. There are characteristics of different types
of identity that make them more or less liable to changes in identity salience than other types,
especially in the presence of a real or perceived threat. As I do throughout the dissertation, here I
focus on national, ethnic, and religious identity, as well as combinations thereof.
Existing scholarship has discussed the distinctive characteristics of different identity types.
Likewise, this theory expects that although national, ethnic, and religious identities are all subject to
changes in salience and can thus result in outcomes of political violence under certain conditions,
right Rep. Elise Stefanik (NY-21), the Republican congresswoman representing the district where the attack occurred
and who currently is the third-ranking member of House Republican leadership.
57
certain types tend to be more vulnerable to rapid and intractable shifts than others. Specifically, I
claim that religious identities are most subject to short-term changes in salience and are also more
likely to become entrenched as a result. Additionally, combinations of these types of identities—for
example, an ethnonationalist group or religious groups with a common ethnicity—are also more
prone to changes in identity salience for reasons that I outline below.
Not all identities are equal in their influence and potential impact. The extent to which
identity salience will lead to different outcomes varies at the individual level, but it also depends
upon certain characteristics of identities that make them prone to shifts in salience. The extent to
which one type of identity is more potentially powerful than another can be understood as a
function of both its level of institutionalization and the ways in which group membership are
expressed. Institutionalization is important because it provides a network of members connected by
structures that provide a sense of legitimacy and continuity to group members and that provides a
tangible foundation upon which the identity is embodied and practiced. For example, the
transnational institutional structure of the Catholic Church provides leadership in setting policy and
revising doctrinal principles, among other functions, and acts as a sort of governing body for group
members in a way similar to that of national governments. Ethnic groups, too, can have
organizations that advocate for their interests, whether these be transnational diaspora organizations
or dedicated political parties.
Expression of group membership is another characteristic that differs across identity types
but that is important in determining whether and how identity salience may increase. The existence
of and agreement upon common practices or means of identifying oneself with a group create a
mechanism for differentiating one’s group from another, and the continued practice of identification
sustains the salience of an identity in a way that is less likely without common practices. Again,
religious identities are made more enduring and relevant to the everyday lives of group members by
58
such practices as ritual prayer, attendance at religious institutions, the celebration of religious
holidays, and other expressions of membership and commitment to the group. These practices
provide a community of members while also serving to distinguish members from non-members,
thus increasing the salience of the identity for practitioners that might otherwise be weaker or
absent. Ethnic identities, on the other hand, do not generally require expressions of membership to
the same extent as do religious identities, but the existence of common language and cultural
practices may serve as an analog for such expressions of membership. Additionally, common
physical characteristics may provide legibility for ethnic identities in a way that religious identities do
not. National identities typically require the profession of allegiance to a particular conception of the
nation and its people, and some common practices may be found in the celebration of important
holidays, speaking a particular language in public and private spaces, or something else that
distinguishes co-nationals from outsiders.
Groups encompassing multiple identity categories at once are also subject to rapid shifts in
salience and in associated changes in attitudes and behavior. Additionally, such groups can gain
increased legitimacy by pointing to their commitment to multiple identity types, thereby broadening
their base of support. An ethnonationalist group that also subscribes to a common religion, for
example, has the advantage of pointing to the legible descent-based characteristics often conferred
by ethnicity while also claiming moral superiority by justifying its actions as ones meant to protect
the faithful and glorify their God. In addition to providing enhanced legitimacy and perhaps serving
as a recruitment and loyalty mechanism, the presence of a strong religious component is also more
likely to render an ongoing conflict more intractable once it begins. Some influential accounts
particularly point to the relative power of religious ideology in driving violence and increasing its
intensity due to such characteristics as issue indivisibility (Toft 2007; Hassner 2009; Juergensmeyer
59
2000).
37
As an example, one study found that Islamist groups in the North Caucasus are less likely to
be influenced by coercive measures like selective violence than are violent nationalist groups, lending
support to the idea that identities centered around a religious ideology may be especially resilient
(Toft and Zhukov 2015).
Just as it is important to differentiate religious identity from other social identities, it is also
important to think about how a social identity composed of two or more constituent identities may
be uniquely powerful in motivating attitudes and behavior. Previous work on convergent identities,
or identities composed of overlapping identities, has underscored its power in motivating political
action (i.e., Huddy 2013, Roccas and Brewer 2002, Mason 2012). Evidence from the US and Israel,
for example, showed that the presence of convergent identities reduced tolerance for outgroups.
Additionally, the perception of a common threat increased identity convergence in Northern
Ireland, thus increasing overlap between Catholic and Irish social identities on the one side and
Protestant and British social identities on the other (Schmid et al. 2008). Convergent identities, such
as ethnoreligious, ethnonational, religious nationalist, or other social identity combinations, are
uniquely powerful both in their ability to appeal to different bases of ingroup members and in
allowing members to draw upon multiple sources of commonality such that barriers to political
action are reduced. This is especially true under conditions of threat, whereby any prior
incompatibilities between the constituent groups would become less relevant. Therefore, the
potential for mobilization among convergent identities would likely be higher than it would be for
singular social identities. Just as identity itself is not monolithic, then, neither are different identity
categories, and recognizing when, how, and why the salience of different identities vary in response
to various factors is an important step towards addressing this.
37
An indivisible conflict or issue is defined in the literature as one in which actors will always prefer conflict over
compromise or negotiated settlement (Fearon 1995).
60
Chapter 2: Empirical Analysis: Identity and Political Violence in the United States
The insurrection at the US Capitol on January 6, 2021 shocked the nation not only for its
brazen display of violence, but for its apparent appeal to individuals who had no prior criminal
history, let alone a history of political violence or outward expressions of extremist attitudes. A
demographic profile of arrestees published by the Chicago Project on Security and Threats (CPOST)
underscored the unexpected nature of participation patterns. While some patterns in the data were
expected—for example, a disproportionate number of arrestees were white males—others were
more surprising. Socioeconomic profiles of J6 arrestees differed along some key metrics from the
general population, with approximately 24% of insurrectionists registered as business owners
compared to 11.8% of the general population. In terms of prior affiliation with extremist groups,
only 14% of the insurrectionists were identified as belonging to far-right militias, compared to 48%
of right-wing extremists (Pape 2022).
The threat of further political violence remains a serious concern among experts. One survey
conducted in February 2021 found that roughly 30% of Americans agreed with the statement that
“if elected leaders will not protect America, the people must do it themselves, even if it requires
violent actions.”
38
A YouGov poll taken two months prior to the 2020 presidential election found
that, on average, a majority of Americans (65%) thought that the “American way of life” was under
threat. The US is clearly divided along partisan lines, but what shapes these divisions, and what
exactly is under threat?
39
Understanding the correlates of permissive attitudes toward political
38
https://www.npr.org/2021/02/11/966498544/a-scary-survey-finding-4-in-10-republicans-say-political-violence-may-
be-necessa
39
Lilliana Mason and Nathan Kalmoe have researched partisanship and polarization in the US extensively, and a recently
published book has continued this research. One of their major claims is that partisanship has increasingly become a
social identity that acts similarly to other social identities, such as ethnicity and religion, and that this transformation in
the character of partisanship is a major risk factor for increasing political violence (Kalmoe and Mason 2022).
61
violence is paramount to understanding why, when, and how these attitudes are formed in the first
place.
Though monocausal explanations of complex political phenomena are rarely accurate or
useful, it is important to examine identity salience more closely as one factor that might be especially
powerful in shaping attitudes toward political violence. I empirically probe some of the key
observable implications of the theory outlined in previous pages, deriving hypotheses specific to the
American context. Using original data collected from a survey experiment conducted in June 2022
on Amazon’s Mechanical Turk (MTurk) platform, I aim to answer the following questions. First, do
higher baseline levels of national identity salience predict more permissive attitudes toward political
violence? Secondly, does experimentally manipulating the salience of American national identity lead
to observed differences in expressed attitudes toward political violence committed on behalf of that
identity?
I first provide a very brief discussion of the theory as it applies to the American case and
outline the hypotheses tested here. I then discuss the research design and methodology before
presenting the results. Following the analysis of the data, I note limitations of the design and offer
potential improvements and modifications that future research could make to test these and other
theoretical expectations.
Hypotheses
The theory advances the basic argument that changes in identity salience are likely to lead to
more permissive attitudes toward political violence, especially when the identity group in question is
perceived as being threatened. The effects of an increase in identity salience as the result of a threat
are expected to be especially large when the identity in question is religious or combines multiple
62
identity types—for example, the theory would expect that a change in the salience of an identity like
Slobodan Milosevic’s brand of Serbian ethnoreligious nationalism would be larger than it would for
an increase in, say, Austrian civic nationalism. Not only does a threat to one’s ingroup engender
more extreme attitudes among ingroup members who feel that something about their way of life or
their place in society is at stake, but the combination of two or more identity types lowers obstacles
to ingroup cohesion.
40
The combination of these two factors determines whether and to what extent
an increase in identity salience will result in increased acceptance of political violence as a means of
advancing group interests.
The broad scope of the theory makes it difficult to fully test all aspects of the theory and its
many conditional hypotheses here. However, the first step to testing the plausibility of the theory is
to test whether the causal relationships proposed by the theory are present at the lower bound—in
other words, testing whether the relationships are present in cases where the weakest effects are
expected. I test whether manipulating American national identity by exposing respondents to a
treatment that invokes a sense of threat to this identity results in the expected effect of increased
support for political violence. If we observe the expected effect in this case, we might plausibly
assume that the same expected effects would be present and stronger in other cases—for example, if
an imminent physical threat to an established ethnoreligious group were posed, we could expect the
effects to not only be present, but stronger. However, failing to reject the null hypothesis in this
study does not invalidate the theory and may be indicative of a type II error (incorrect acceptance of
the null hypothesis), a possibility that would underscore the need for more refined testing.
Because it is not possible to measure outcomes of actual political violence via a survey
experiment, I focus on attitudes toward political violence, which can serve as a sort of predictor for
40
A possible extension of this idea of convergent identities lowering obstacles to greater ingroup cohesion could be that
such a mechanism increases not only ingroup cohesion, but intergroup polarization. Though beyond the scope of this
paper, I note it here as a possibility for future research.
63
whether these microlevel attitudes will translate to behavior—when more individuals express
increasingly permissive or supportive attitudes toward political violence, it becomes more likely that
at least some of these individuals might act on such attitudes. According to the theory, higher
baseline levels of identity salience will correlate with a higher likelihood that political violence
committed on behalf of a group identity will occur. When identity salience increases, then, the
likelihood that political violence would occur increases as well. The combination of a common
ingroup identity and perceived threat to that identity from an outgroup are particularly potent
factors that affect identity salience in such a way that it may alter attitudes toward political violence.
Specifically, I test the following hypotheses:
H 1: Higher baseline levels of identity salience will predict more permissive attitudes toward
political violence carried out on behalf of an ingroup.
H 2: Increasing identity salience via a threatening experimental manipulation will cause an
increase in expressed support for political violence compared to the control group.
H 3: Respondents who are shown the threatening treatment will express higher levels of
support for political violence as their baseline identity salience levels increase.
To test these hypotheses, I conducted an online survey experiment using a convenience
sample
41
of American adults in June 2022.
42
I designed the survey instrument in Qualtrics and fielded
the study using Amazon’s Mechanical Turk (MTurk) platform.
43
The sample frame was all MTurk
“workers” who were above the age of eighteen and who were registered with a US-based IP address,
41
A power analysis indicated that the minimum sample size per group required to elicit a statistically significant (at the
0.05 level) effect size was approximately 700. To minimize wasted resources and maximize the likelihood that a
statistically significant effect size would be uncovered, I chose a total sample size of 1500, with 750 respondents
randomly assigned to each of the two groups.
42
This study was approved by the Institutional Review Board at the University of Southern California (protocol number
UP-21-01039). Documentation associated with this project’s IRB approval, including the submitted research protocol
received from the IRB approval, are included in Appendix 2.
43
The initial plan was to conduct this as an in-person survey experiment, but both COVID-19 and methodological
reasons led me to change the format to an online study.
64
an exclusion measure applied to prevent respondents based outside the United States from opting in
to take the survey.
44
The US provides an interesting test case for the theory. Identity cleavages have thus far not
fallen along explicit ethnic or religious fault lines, though the increasing espousal of racially-charged
language and alliances with identity-based groups among elites and partisans alike has increasingly
moved from the fringes to the mainstream.
Political violence has been relatively sporadic since the
end of Reconstruction, and it was not until recently that experts and commentators began more
widely and forcefully expressing concerns about the potential for increased political violence. At the
same time, the events surrounding the 2020 presidential election and the violence at the Capitol on
January 6 have given way to a more open conversation about political violence in America. Given
evidence of more permissive attitudes toward political violence among the general population, it is
worth questioning what role, if any, identity salience plays in driving these attitudes.
The survey yielded a final sample size of 1945 respondents, with a majority (about 97%) of
these completing all questions in the survey. As is typical of a convenience sample, distribution
across key demographic variables was not representative of the broader population.
45
Though the
non-representative nature of convenience samples may pose threats to external validity in non-
experimental survey research that aims to understand how a variable is distributed within a
population (Yeager et al 2011), random assignment to the treatment condition removes selection
bias and increases confidence in making causal claims (Gerber and Green 2012, Shadish et al. 2002).
44
Because the survey experiment focuses on the salience of an American national identity, it is important to minimize
the risk that bias is introduced because of non-Americans participating in the survey. Though it is possible that some
non-Americans were mistakenly included in the survey, the exclusion measures and the self-reported demographic
information given by the respondents indicate that this is unlikely, and therefore has not produced undue bias.
45
Though all major demographic groups were represented, some variables were disproportionately skewed toward
majority groups in the US. In terms of race, about 83% of the sample identified as white; for religion, 75% identified as
Christian; and in terms of education, 64% stated they were college graduates. About 55% were male versus 44% female,
53% Democrat versus 36% Republican, and a majority (about 40%) were age 25-34. Modal household income fell in the
$40,001-$60,000 range.
65
Additionally, survey experiments with convenience samples do not face the same threats to validity
when treatment effects are expected to be homogeneous (Druckman and Kam 2011). Indeed, some
studies have demonstrated that treatment effects uncovered from survey experiments on
convenience samples do not differ significantly from those taken from probability samples (i.e.,
Mullinix et al 2015), and that convenience samples obtained from MTurk in particular are often
more representative of the population than convenience samples obtained through other mediums
(Berinsky et al 2012).
Conducting a survey experiment through online platforms like MTurk, which allows for full
anonymity and does not require interaction with an interviewer, provides additional measurement
advantages by reducing social desirability bias (Yeager et al 2011, Chang and Krosnick 2009). The
sensitive nature of some of the questions in this survey instrument, particularly those that measure
attitudes toward political violence, would likely lead to more item or unit nonresponse if asked in-
person than would be the case in an online platform. The anonymity afforded via online survey
mediums in general and the MTurk platform in particular may help mitigate these issues; indeed,
response rates for the political violence questions were near 100% among this sample.
Survey Flow, Main Variables, and Experimental Treatments
I next discuss the survey flow and define the variables of interest. The survey instrument
contains 29 multiple choice and rank order questions designed to measure identity salience and
attitudes toward political violence. The full survey instrument can be found in Appendix 2, and a
diagram of the experimental flow and which blocks of questions correspond to which key variables
can be found in figure 2.2 below.
66
Figure 2.2: Experimental Flow
Respondents were asked to read and indicate their acknowledgement of a statement of
informed consent at the beginning of the survey and were reminded that their answers would remain
fully anonymous, and that participation was voluntary. The first question asks if the respondent has
been politically active at any point in the past two years, with political activity defined as
participation in a peaceful protest, campaign, rally, or march. Following this initial question is a 12-
question battery meant to measure the respondent’s baseline level of identity salience, with most of
these questions asking respondents to indicate their position on a five-point scale of agreement with
an issue or the importance of a question. This first set of questions measuring identity salience is
followed by a series of demographic questions, including gender, age, race/ethnicity, education,
religious affiliation, and annual household income. In addition to being important control variables,
these covariates might also condition susceptibility to violent attitudes. I include a screening
question as an attention check to minimize bias introduced by inattentive respondents
(Oppenheimer et al 2009, Berinsky et al 2012). About 97% of the total sample answered the screener
question correctly, indicating a high level of attention among respondents and reducing bias that
Identity Salience Questions
IV1: Baseline Identity
Salience
Treatment (Priming
national identity)
Control group
IV2: Random
Treatment
Assignment
Expressed Attitudes
Toward Political
Violence
DV: Political Violence
Index
Informed Consent
67
would come from inattentive “click-through” respondents
46
. Following the attention check, I ask
whether the respondent is registered to vote and, if so, what their party identification is.
Respondents were given the option to not disclose their demographic details and were free to skip
questions without being penalized if they chose not to answer. Still, the completion rate for those
who started the survey remained extremely high.
After the first set of questions, respondents were randomly assigned to either the
experimental treatment group or the control group. Those assigned to the treatment were shown an
image meant to prime American national identity in a way that evokes a sense of threat. Upon
viewing the image, respondents assigned to the treatment were asked to provide a caption describing
what they thought was occurring in the image, then were asked to provide a one to two sentence
explanation of why they chose the caption they had provided. Respondents assigned to the control
group were simply directed to immediately continue answering the post-treatment questions without
being shown an image. Finally, all respondents were then asked to answer seven post-treatment
questions measuring their attitudes toward political violence. Again, respondents were not required
to answer all these questions to receive compensation; still, the completion rate was near 100%.
Explanatory Variable 1: Baseline National Identity Salience
The main explanatory variables of interest are identity salience and treatment assignment.
Because this study pertains to political violence in the United States, I focus specifically on the
salience of an American national identity. By American national identity, I mean a nationalistic social
identity that tends to be exclusionary toward groups not considered to be sufficiently “American.”
47
46
The potential for “professional” survey takers on platforms such as MTurk to randomly answer questions without
actually thinking about them is a frequently cited threat to validity; however, the results of the attention check indicates
that this is not an issue for the current study.
47
As discussed earlier, there are certainly competing versions of what it means to be an American to different groups of
people. Because it is impossible to adjudicate between them all here and the purpose of this study is to study social
68
The first series of pre-treatment questions was designed to capture the dimensions of identity
salience expected to be theoretically important and constitutive of identity salience as I have
conceptualized it: primary subscription, cognitive prioritization, and affective valence.
48
As discussed
in depth in article one, these questions were designed in such a way that they would capture these
three constitutive dimensions without unintentionally causing respondents to attach more strongly
to their national identity in a way that they might not otherwise. Each respondent received a baseline
identity salience score based on their answers to the identity salience questions, with this score
representing how salient the identity being measured is for each individual without any prime or
prompts being introduced. Scores ranged from roughly -3 to 3, with values normally distributed
throughout the sample and the mean clustered around a score of approximately 0 (µ=0.06). Lower
scores indicate lower national identity salience, while higher scores indicate higher national identity
salience. The respondents with the highest baseline identity salience score answered each question
related to identity salience in the expected way, as did those with the lowest baseline identity scores,
indicating high construct validity. For example, those with the highest scores tended to respond
“strongly agree” and “extremely important” to these questions, while those with the lowest scores
responded “strongly disagree” and “not at all important.” This is as expected and indicates that the
measurements generated from the model represent identity salience as I conceptualize it.
Explanatory Variable 2: Treatment Assignment
identities that are nationalist, ethnic, or religious, I focus on an exclusionary form of national identity. To avoid the
introduction of potential bias by evoking explicitly ethnic or racial appeals, I design questions that measure national
identity salience that gauge exclusionary nationalist tendencies while not referring specifically to certain ethnic, racial, or
religious groups.
48
Because I outlined the measurement of identity salience and the questions that comprise the measure in depth
previously, I do not repeat myself here. For a detailed discussion of the measurement of identity salience and the use of
different questions to capture different aspects of identity salience, please refer to pp. 19-24 of article one.
69
The second key explanatory variable is treatment assignment, or whether a respondent was
assigned to the experimental group or the control group. The treatment was designed to prime
respondents’ national identity salience by evoking a sense of threat to a common ingroup based on
American national identity. Upon completing the pre-treatment identity salience and demographic
questions, all respondents were randomized into either the experimental group or the control group.
The control group did not receive a treatment and instead was directed to continue to the post-
treatment questions. The experimental group received a treatment intended prime the salience of an
American national identity and can be seen in Figure 2.3 below.
Figure 2.3: Experimental Treatment Condition
The image was sourced from a Google Images search using such keywords as “America”
and “threat.” After reviewing the search results, I identified this image as one that could both a)
prime the salience of American national identity given the prominent centering of the American flag
and b) induce a sense that this identity is being threatened. However, it was important to ensure that
the image be deidentified from a particular event that could bias the results, so it was lightly edited to
70
remove any potential geographic references.
49
Rather, the image is meant to prompt a sense of a
general threat to America that would be expected to affect respondents regardless of partisan
affiliation. Visual images have been shown to be effective in influencing political attitudes and
behavior (Druckman 2003, Gadarian 2014). Indeed, exposing individuals to images may encourage
them to reassess what they consider to be important about an issue, though more subtly than might
be the case with exposure to a verbal vignette or another treatment or media type (Mendelberg 2001,
Valentino et al. 2002). Rather than merely showing the image, I include a short prompt that requires
respondents assigned to the treatment to first provide a caption describing what is going on in the
image, then to explain in a maximum of one to two sentences why they chose this caption.
50
This
has two purposes. First, it prevents respondents from clicking through the image without
considering it, reducing the possibility of a null treatment effect. Second, it encourages effortful
information processing, a process by which an individual processes their perception of the image
and any thoughts associated with it such that it increases the likelihood that the effect of the image
exposure “sticks,” rather than the net effect of the exposure declining quickly after exposure (i.e.,
Hill et al. 2013, Sides and Vavreck 2013, Bartels 2014).
Finally, I intentionally excluded references to specific groups, events, or political figures not
only to avoid biased results, but as a sort of least likely test. Observing a statistically significant effect
in the expected direction between the treatment and control groups would indicate that priming
identity salience—even in a relatively innocuous way and without an explicit call to action—could be
expected to have even more significant effects when applied a) for a sustained period over time and
b) with more specific references to ingroups and threats from outgroups. In other words, observing
49
This image was taken during protests against police brutality in Minneapolis following the murder of George Floyd. I
used an online photo editor to remove the name of the store in the background so that respondents would not associate
the image with those specific events.
50
This is a forced-choice response, meaning that if a respondent assigned to treatment tries to continue with the survey
without commenting on the image, they will get a reminder that they must provide an answer in the text box provided.
71
a positive and statistically significant difference in attitudes toward political violence after
introducing this treatment would indicate that we might expect to see stronger effects when
individuals are exposed to more targeted and divisive treatments. The effect would likely be even
more pronounced in the aftermath of a realized threat, such as an act of violence or a terror attack.
As noted above, empirically testing what might occur at the lower bounds of the theory is a fruitful
way to conduct an initial plausibility probe of the theory.
Dependent Variable: Attitudes Toward Political Violence
The dependent variable measures attitudes toward political violence, which I primarily
measure as a composite score based on answers to five multiple-choice questions.
51
I use a
combination of questions to capture both general attitudes toward the idea of political violence and
the circumstances under which respondents would consider it to be justifiable, as well as individual
willingness to engage in political violence themselves. I also ask respondents to assess the likelihood
that people they know would engage in political violence, as this might approximate their own
attitudes if social desirability bias prevents them from answering truthfully about their own
willingness to engage in political violence.
52
I also ask a more generic question about whether and
when violence may be justified to achieve political goals. I additionally include a question intended
to gauge respondent attitudes toward the legal consequences faced by participants in the January 6
insurrection at the Capitol, with the expectation that individuals who answer that the penalties are
too harsh might reasonably be expected to have more permissive attitudes toward political violence
51
I also use each constituent indicator of the index individually as dependent variables and provide an overview of these
results in appendix 2. Also included in the appendix is a more detailed discussion of question wording and design for the
political violence questions.
52
Given the anonymous nature of the survey, I do not expect this to be a significant problem, but I include this
additional question anyway. Initial, albeit limited, pre-testing of the survey instrument did not indicate that respondents
were less willing to answer such a question, and I have no reason to expect that this will not be the case in the actual
survey experiment.
72
committed on behalf of American national identity. Finally, I ask whether it is ever okay to use
threats against others for political reasons, as violent rhetoric may indicate an elevated likelihood of
violent action and may itself be considered a form of political violence (Kalmoe and Mason 2022).
To create a single measure of attitudes toward political violence as a dependent variable, I
calculate a weighted index for each respondent using their responses to these five questions. Each
individual respondent is assigned a composite score based on this weighted index to measure their
attitudes toward political violence, with higher scores indicating higher levels of support for political
violence. The questions regarding the likelihood of the respondent and of people they know to
engage in political violence are each weighted more heavily than the others, as these relate more
directly to the outcome we are trying to measure.
Chapter 2: Results
I analyze the results via regression analysis using a variety of model specifications.
53
I first fit
an ordinary least squares model to test the relationship between treatment assignment and attitudes
toward political violence, with the expectation that assignment to treatment will lead to an increase
in permissive attitudes toward political violence. The basic model equation is as follows:
y = ɑ + βtTi + ε,
where y is the composite measure of attitudes toward political violence and βtTi is the coefficient on
assignment to treatment. The value of βtTi uncovers the treatment effect of assignment to the
treatment versus the control group, with the coefficient value representing the difference in means
between assignment to treatment and assignment to control.
In addition to this main model, I also fit an ordinary least squares model to measure the
relationship between baseline levels of identity salience and attitudes toward political violence
53
Replication materials can be made available upon request.
73
(without accounting for experimental treatment effects). The corresponding model equation is as
follows:
y = ɑ + βid + ε,
where y is the value of the composite measure of attitudes toward political violence and βid is the
coefficient on identity salience. I also specify models that include covariates for demographic
characteristics, previous political engagement and party identification.
The results—presented in full in figures 2.1 and 2.2 (pp. 42-43)—indicate that we can
consistently reject the null hypothesis for H1, lending support to the theoretical claim that higher
levels of national identity salience are associated with a greater likelihood of support for political
violence on behalf of the salient identity. There was a strong and statistically significant positive
effect of baseline identity salience on expressed support for political violence across all model
specifications. In the simplest model using identity salience as the sole predictor, a one standard
deviation increase in a respondent’s identity salience score, or about a one-point increase
54
, was
associated with an approximately 11-point increase in political violence support, as measured by the
composite index. The direction and significance of this effect remains consistent across all models.
Despite the strong support for H1, the results do not allow for a rejection of the null
hypotheses for H2 and H3, as the inconsistency of these results across models means that we cannot
confidently say that there is or is not an effect of assignment to treatment on supportive attitudes
toward political violence. In most of the main models, the coefficient on treatment assignment is
negative and not statistically significant, leading to insufficient evidence that the treatment meant to
increase identity salience had the anticipated effect of increasing expressed support for political
violence (H2). In addition to not achieving statistical significance, the treatment effect was
54
s=0.93
74
substantively insignificant across all model specifications. In the main model using treatment
assignment as the sole predictor of political violence index scores, the coefficient on treatment was
-0.51. Notwithstanding that the expected effect of treatment on political violence index scores was
that it would be positive, this coefficient indicates a change of less than one point on a 100-point
scale for the dependent variable. Also considering the high value of the first intercept value (in this
model, about 59.4), this is such a small number as to be almost indetectable. Further underscoring
the null effects of the treatment is the near-zero R-squared value for the model using treatment as a
predictor (0.0001)—by comparison, the R-squared for the base model using baseline identity
salience as the sole predictor is about 0.24. In other words, the effect of the treatment was essentially
null by all measures.
The results are more mixed for H3: without added controls, the coefficient on the
interaction term for baseline salience and treatment assignment is negative and statistically
insignificant, while it is strongly positive and statistically significant when key covariates, such as
race, religious affiliation, and party identification are added. This inconsistency does not allow us to
reject the null hypothesis for H3, meaning we cannot confidently assert given the available data that
an experimental manipulation of identity salience increases more quickly proportionate to baseline
levels of identity salience.
As an added robustness check, I dichotomized each of the constituent indicators of the
political violence index and ran a series of logistic regressions using these dichotomized measures as
dependent variables. Support for H1 remains consistent across all model specifications, as does the
inability to reject the null hypotheses for H2 and H3. Results of these analyses are included in
appendix 2.
75
Table 2.1: Main Results
Dependent variable:
Expressed Support for Political Violence
(1) (2) (3) (4)
Identity Salience 11.257
***
11.257
***
11.943
***
(0.462)
(0.462) (0.640)
Assigned to Treatment
-0.510 -0.440 -0.356
(0.981) (0.857) (0.858)
Salience*Treated
-1.433
(0.925)
Constant 58.538
***
59.446
***
58.754
***
58.711
***
(0.429) (0.687) (0.601) (0.602)
Observations 1,908 1,908 1,908 1,908
R
2
0.237 0.0001 0.237 0.238
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
76
Table 2.2: Main Results with Added Controls
55
Dependent variable:
Expressed Support for Political Violence
(1) (2) (3) (4)
Identity Salience 7.994
***
7.994
***
8.374
***
(0.470)
(0.471) (0.615)
Assigned to Treatment
-0.367 -0.376 -0.322
(0.819) (0.762) (0.764)
Salience*Treated 16.440
***
18.348
***
16.428
***
16.390
***
(0.934) (0.997) (0.934) (0.935)
Recent Political Activity -3.294 -3.293 -3.298 -3.342
(2.103) (2.262) (2.103) (2.104)
Voter 2.422
***
2.356
***
2.424
***
2.449
***
(0.782) (0.841) (0.782) (0.782)
Male 0.238 2.066
*
0.237 0.282
(1.039) (1.111) (1.039) (1.040)
White 5.136
***
11.965
***
5.132
***
5.129
***
(1.019) (1.007) (1.019) (1.019)
Christian
-0.789
(0.824)
Constant 48.748
***
40.263
***
48.975
***
48.907
***
(3.046) (3.268) (3.081) (3.082)
Observations 1,845 1,845 1,845 1,845
R
2
0.422 0.331 0.422 0.423
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Chapter 2: Discussion
Although the results of this study provided mixed support for the theory, the evidence was
particularly useful in two ways. First, there is empirical support for the core theoretical proposition that
there is a strong relationship between national identity salience and permissive attitudes toward political
violence committed on behalf of that identity, suggesting that further research should examine different
types of identity to determine whether and how they affect political outcomes differently. Secondly, the
55
For these models, I dichotomized some of the demographic responses to indicate whether a respondent was white,
Christian, or male.
77
mixed results on treatment effects suggest that future studies must be more careful in designing and
implementing experimental treatments intended to manipulate identity salience, particularly when the
effect being measured relates to attitudes toward political violence.
Given the strong support for H1, it would follow that experimentally increasing national identity
salience should also increase expressed support for political violence. However, the results indicate that
the treatment employed in this study did not do that, and examining the potential reasons for this could
help inform additional research and study design. Further analysis of the results indicated three potential
reasons for the null effects of treatment on attitudes toward political violence, all of which should be
taken into consideration in designing future studies related to questions of identity salience and political
outcomes, violent or otherwise.
First, it is possible that the treatment did not have the intended effect of increasing the salience
of respondents’ American national identity. Though the forced-response caption was meant to increase
the amount of time that respondents assigned to treatment spent considering the image they were
presented, the responses provided were also helpful in understanding how these respondents interpreted
the image. In addition to priming national identity salience by prominently presenting an American flag
(primary subscription), the goal of this image was to create a sense of threat to this identity (affective
valence). An analysis of these written responses indicated that reactions to the image were not uniform
across respondents and, most crucially, that not all of them reacted negatively to the image. Many
respondents indicated that they thought the image was from Black Lives Matter protests, and some
recognized the image as being connected to the events surrounding the murder of George Floyd in June
2020. Some wrote captions related to American pride and standing strong as a nation, further indicating
that some individuals were positively affected by the image and did not perceive a threat.
78
Figure 2.4: Frequently Used Words in Open-Ended Treatment Responses
Though many responses indicated a negative reaction, the diversity of open-ended responses
suggests that the treatment affected respondents differently. I expected homogeneous treatment effects
with this treatment, but the quantitative results and the open-ended responses given show that this was
not the case. A sentiment analysis of the caption titles written by respondents assigned to treatment
showed that not only was negative perception of the image not uniform across respondents, but that
roughly a third of these respondents wrote captions with positive sentiments. Indeed, the mean
sentiment score was approximately 0, a score indicating neutral sentiment.
56
If the treatment had the
intended effect of causing a negative reaction by creating a sense of threat, we would expect a majority of
sentiment scores to fall below zero. As shown in figure 2.5, sentiment scores were relatively evenly
distributed around a mean neutral score, represented by the dashed vertical line. This might be one of
56
Actual mean score was -0.06. Sentiment analysis using dictionary methods was conducted using the quanteda package
in R (Benoit et al. 2018).
79
the reasons for the null effects of the treatment. In addition to designing and implementing different
treatments—and pre-testing them to ensure that they do indeed affect identity salience in the desired
way—future studies should consider whether the type of identity in question would affect different
subsets of the population differently and implement an appropriate sampling strategy.
Figure 2.5: Sentiment Analysis of Open-Ended Treatment Responses
Another possible reason that we observed the results that we did is specific to the American
context and changing conceptions of what it means to be an American; indeed, although research
has shown that most Americans tend to agree on the normative aspects of American-ness (i.e., civic
virtues), different people experience their national identity differently depending on contextual
factors (Schildkraut 2014). More specifically, national identity has become subject to partisan sorting,
meaning that party identification in some cases conditions how individuals conceive of what it
80
means to be an American (Bonikowski et al. 2021). As shown in figure 2.6 below, an analysis of both
mean identity salience and political violence index scores by party identification indicated a
statistically significant difference between Democrats and Republicans, with Republicans
consistently indicating higher levels of both baseline national identity salience and supportive
attitudes toward political violence. To ensure that unbalanced treatment assignment by partisan
identification did not bias the results, I conducted a covariate balance test using party identification
and other key covariates, such as gender and race. The results of the balance test showed that
representations of these covariates across the treatment and control groups were balanced and did
not bias the results. A balance test of baseline identity salience score across groups indicated that
these mean scores were balanced and did not show any substantial or statistically significant
difference between those assigned to treatment and those assigned to control.
57
The significant difference by partisanship among the sample studied here, as well as general
disagreement as to what it means to be an American in the broader population, suggests that future
studies of the relationship between American national identity salience and political outcomes,
violent or otherwise, should be mindful of partisan differences when thinking about question
wording and treatment design. In the case of this study, clustered random sampling by party
identification might have been a more appropriate strategy in uncovering treatment effects.
57
Full results of the balance test can be found in appendix 2.
81
Figure 2.6: Baseline Identity Salience and Political Violence Index Scores by Party Identification
Finally, a further analysis of the results showed that a significant number of responses may
have been submitted by the same person—in other words, analysis of IP addresses and matching
open-ended responses indicated that the survey may have been compromised by one or more
people taking the survey multiple times under different worker IDs. Because the survey was
anonymous, it is not possible to know just how many of these responses were submitted by the
same people, but it should not be taken for granted that the data could have been biased as a result.
This is especially true regarding the experimental treatment: previous research has shown that low-
quality responses stemming from these MTurk-specific issues attenuate treatment effects
significantly, and it is possible that the same occurred here (Ahler et al. 2019).
The results of this study suggest that although identity salience and attitudes toward political
violence may certainly be causally connected, the multifaceted nature of identity requires careful
consideration throughout the design of research studies that attempt to manipulate identity salience
as a variable. Despite the null effects of the treatment in this study, the presence of a strong and
82
statistically significant relationship between American national identity and general attitudes toward
political violence suggests support for the core tenet of the theory: identity salience should not be
discounted as a variable in political processes, including political violence. Future research would do
well to incorporate identity salience as a variable.
83
CHAPTER 3: IDENTITY-BASED POLITICAL VIOLENCE: A MACROLEVEL THEORY
AND TYPOLOGY
In the aftermath of the violence surrounding the congressional certification of then-
President Elect Joe Biden’s electoral victory in January 2021, some scholars of political violence who
had previously focused on other countries began to discuss events in the US similarly to how they
had discussed political violence elsewhere. Some observers have pointed to the incident as an
inflection point that is leading inexorably toward large-scale political violence, from low-intensity
insurgency to all-out civil war.
58
At the same time, some scholars have urged caution in
prognosticating the onset of civil war, arguing that a focus on such extreme outcomes elides the very
real possibility and deleterious effects of political violence outcomes that fall below the high
threshold required for an event to be designated a civil war (Dayal et al. 2022).
59
The ongoing debate
in the United States over the severity of potential political violence outcomes underscores the
importance of taking a more careful approach to classifying such potential outcomes. In addition to
the need for nuance, I will argue that a more systematic incorporation of identity into the analysis of
political violence can be fundamental to understanding the severity of its outcomes. At the same
time as differentiation among outcomes is needed, I also argue that identity-based political violence
at all levels follows a similar logic.
58
In addition to sensationalist headlines heralding the beginning of a civil war, such as one published in Foreign Policy
(March 2022), recent work by civil war scholars such as Barbara F. Walter has gained increased public attention for its
comparisons of the current situation in America to such events of political violence as the Troubles in Northern Ireland
(Walter 2022).
59
“Our concern with the frame Walter and others offer — and with the attached “civil war or not” headlines — is that it
misses the wide array of other kinds of political violence the United States has not only historically experienced, but is
currently experiencing… We do not think a clearly identifiable, explosive moment of crisis that suddenly breaks with
ongoing trends is imminent — but priming people to expect a spectacular, ultimate calamity could obscure the ongoing
slow boil of political violence. Focusing on the rates, forms, and targets of political violence provides important
nuance.”
84
This article presents a theory of the causal factors of what I call identity-based political
violence. I define identity-based political violence as a violent conflict event in which at least one of
the main actors advocates for or organizes based upon goals related to a common ethnicity, religion,
or ethnoreligious group. As a part of this theoretical discussion, I outline a basic typology of
identity-based political violence that categorizes conflict events along two varying dimensions:
identity coherence and elite support. As identity coherence and elite support increase, so too does
the potential severity of the event, which may be one of three ideal-typical outcomes of identity-
based political violence: stochastic political violence, sustained insurgency, and civil war. I probe the
plausibility of the theory by leveraging a novel coding scheme including new variables for use in
conjunction with the Uppsala Conflict Data Program’s political violence datasets. In addition to
testing key hypotheses using this quantitative data, I discuss illustrative cases of the three ideal types
to demonstrate the utility of the typology. I conclude with remaining questions and
recommendations for future research.
Before continuing, it is important to clarify what this theory and typology do not consider.
In conceptualizing identity-based political violence, I exclude events that have clear international
dimensions or significant international involvement, as the direct involvement of non-domestic
actors with strategic interests of their own has the capacity to alter the contours of a conflict in such
a way as to render this typology less useful in predicting outcome severity. Though identity claims
may certainly be important in cases of interstate violence, this theory focuses solely on violence at
the intrastate level. Additionally, in discussing the role of identity coherence and elite support in
outcome severity, I do not claim that there is a specifiable level of either that will or will not lead to
violence; rather, I discuss levels of these variables as a point of reference to differentiate between
severity of violent outcomes and to better understand when we might expect to see identity-based
political violence of greater versus lesser severity.
85
Chapter 3: Literature Review
Political scientists studying conflict prior to the end of the Cold War focused primarily on
interstate conflict, with examinations of intrastate or civil conflict in its own right relatively few and
far between. Not until the end of the Cold War did this change, as international conflicts gave way
to civil wars and political violence at the domestic level. After a marked decrease in civil war
60
deaths—72% from 1990-2003, according to UCDP/PRIO data—from the early 1990s to early
2000s, there has been a substantial increase in both major and minor civil conflicts and battle deaths.
As of 2015, the number of major civil wars had increased from four to thirteen, and in 2017 minor
intrastate conflicts reached their highest number since 1994. In 2015, battle deaths per year were at
their highest level since the end of the Cold War, and the intensification of the Syrian and Yemeni
civil wars since then—in addition to other conflicts— has likely sustained these numbers. Civil wars
have additionally become more intractable, with 60% of concluded civil conflicts initiated in the
early 2000s having recurred within five years of their initial termination (von Einseidel et al. 2017).
Using Yugoslavia’s violent breakup as a guide, some scholars and policymakers focused on the role
of intergroup enmity over deep-seated ethnic grievances as the cause of the bloody civil wars that
had become more commonplace toward the end of the 20
th
century (i.e., Horowitz 2001, 2004).
These ideas were quickly challenged, however, as initial studies of the causes of civil war that tested
the “greed vs. grievance” hypothesis asserted that it was economic opportunism (greed), and not
deep-seated enmity between ethnic groups (grievance) that led to civil wars (Collier and Hoeffler
2004; Fearon and Laitin 2003). And while group identities frequently associated with intergroup
conflict may be socially constructed, they can be made very salient and real to individuals who
60
Defined here as any event involving at least one state actor and 1000 battle deaths within a yearlong period.
86
perceive themselves to be members of the group by “ethnic entrepreneurs,” or political elites who
seize upon perceived ethnic difference for the purpose of political mobilization and advantage
against opponents (Gagnon 2004; Petersen 2011). Still others have found that individuals may be led
to take up arms against their government when greed and grievance reinforce each other: in the case
of rebel mobilization in Sierra Leone, for example, grievances, selective incentives, and social
sanction interacted to lead individuals to become civil war combatants (Humphreys and Weinstein
2008).
More recent work
61
has leveraged increasingly sophisticated methodology and novel datasets
to better understand the role of ethnic and religious identity not only in civil war, but in civil conflict
and political violence writ large. The recognition of potential measurement error and the
development of new theories and innovative empirical strategies has brought renewed attention to
the role of “grievance” type factors in predicting political violence, putting more focus on identity
and how it might operate in leading to or shaping intrastate violence. The publication of the Ethnic
Power Relations dataset in 2010 revitalized the discussion surrounding the relationship between
ethnic difference and civil conflict, with evidence showing that while mere ethnic fractionalization
within a country does not itself cause political violence, ethnic diversity existing in tandem with
ethnicity-based elites and parties could (Cederman et al. 2010; Wimmer 2013). Though identity may
not intrinsically be politically relevant at first, it has the capacity to become so given mobilization by
political elites (Posner 2004). The connection between identity and its mobilization by political elites
is especially relevant when the social identity being mobilized is an exclusive one whose members
can be clearly delineated from those of outgroups. However, the potential effects of identity
mobilization may vary across regime type due to differences in the strength of institutions and the
presence of institutional constraints. When institutions are weak, particularly in hybrid authoritarian
61
For additional literature relevant to this study, refer also to the literature review section in article 2 of this manuscript.
87
regimes or partial democracies, the co-optation of exclusive identities by elites becomes even more
indicative that political violence might not be far off (Goldstone et al. 2010; Walter 2022).
Just as this scholarship has recognized that elites may use group identities as instruments for
political mobilization, other works have noted that identity is an especially potent vehicle for
mobilization due to its unique characteristics and the psychological mechanisms it might trigger
among members of the ingroup (Sambanis 2001). The type of identity being mobilized matters as
well. For example, some research has shown that contestation over religious identity may be more
likely to lead to severe and persistent conflict due to such characteristics as issue indivisibility (Toft
2006, Hassner 2009). When groups adhere to multiple types of identity simultaneously —for
example, an ethnic and religious identity—the likelihood of violence increases exponentially (Gubler
and Selway 2012). Others have argued that ethnic mobilization operates via a logic of value
rationality, as those who mobilize on behalf of social identities—ethnicity and religion—more
frequently prioritize nonmaterial goals related to their identity attachments than they do material
goals (Varshney 2003). Once political violence has occurred in contexts where intergroup divisions
are salient, violence has the potential to increase intergroup identification patterns to the point that a
cycle of polarization increases, leading to mutually reinforcing cycles of ethnic identification and
violence (i.e., Sambanis and Shayo 2013, Nair and Sambanis 2019). And among civil wars that end in
negotiated settlements, identity-based civil wars are the most likely to recur after a breakdown of the
settlement (Licklider 1995). Clearly, the causes of identity-based political violence are worth
distinguishing from those of non-identity-based political violence. The theory presented in this
article is an attempt to synthesize some of these explanations to come to a more comprehensive
understanding of outcomes of identity-based political violence in a way that systematically
differentiates between outcomes by levels of severity.
88
Chapter 3: Theory and Hypotheses
Evidence that identity-based cleavages are strongly predictive of political violence says little
about the severity of the violence that may result. A core contention of the theory I propose in this
article is that identity is powerful not only in motivating general outcomes of political violence, but
also in determining the severity of outcomes. Given the existence and apparent rise of identity-based
factionalism and cleavages throughout the world, and of violence occurring along these cleavage
lines, how can we predict whether conditions point to lower-level political violence on one end of
the spectrum, or civil war on the other? When are conditions more conducive to peace or unlikely to
lead to political violence? I argue that two key factors, identity coherence and elite support, are
instrumental in determining the occurrence and severity of political violence outcomes. At its most
fundamental level, the theory holds that an increase in identity coherence and in elite support will
lead to more severe outcomes of political violence. To specify further what different levels of
political violence might occur given different levels of these key predictors, the theory includes a
typology whose outcomes fit into the broader category of identity-based political violence. The
typology outlines three ideal types of political violence outcomes that vary in severity and that can be
understood as functions of different levels of identity coherence and elite support, the two
dimensions of the typology.
I consider a party to a political violence event to be identity-based if it explicitly advocates
for or organizes based upon goals related to a common ethnicity, religion, or a combination of these
categories. I choose this terminology because while violent outcomes are rarely driven solely by
grievances related to identity, they sometimes have a significant identity-based dimension that makes
them different—and often more intractable and combustible—than violence waged along other
motivational lines, such as resource competition or strictly political differences.
89
As an important clarifying point, much of the literature on ethnic conflict uses ethnicity and
sectarianism interchangeably, ostensibly to simplify and make clear that they are talking about
ascriptive groups. However, a core contention of this theory is that collapsing sectarianism into
ethnicity is not just harmless linguistic shorthand. Instead, it oversimplifies a complex phenomenon
and occludes the very important and substantial difference between ethnic and religious or sectarian
identities. Thus, I refer to these events as identity-based, rather than ethnic, not only because of
semantic preferences, but because it is theoretically and empirically important to be precise in
classifying the type, or types, of identities being contested in instances of identity-based political
violence. Given this understanding of identity-based political violence, I define identity coherence
and elite support before outlining the typology and its outcomes, as well as examples of events that
can be classified within each type.
Identity coherence refers to the extent to which a group is clearly committed to a shared
ethnic or religious identity, or a combination of these categories. More specifically, a group’s identity
coherence can be understood as the degree to which its members adhere to a common conception
of who they are, what they want, and who or what they oppose. Though a shared group identity is a
prerequisite for a basic level of identity coherence, it is the substance of that identity and the extent
to which it is connected to a shared goal or goals that determine the overall level of the group’s
identity coherence. As an example, the Basque separatist group ETA (Euskadi Ta Askatasuna, or
Basque Homeland and Liberty) had a high level of identity coherence at the height of its violent
activity in the latter part of the 20
th
century and initial decade of the 21
st
. In addition to sharing a
commitment to an ethnonational Basque identity, the group shared a clear goal of establishing an
independent Basque ethnostate in northern Spain and parts of France.
62
This commitment to a
62
ETA was also ideologically leftist, and a large part of their platform as a group was to establish a Basque state based on
revolutionary Marxism (https://www.cfr.org/backgrounder/basque-fatherland-and-liberty-eta-spain-separatists-euskadi-
ta-askatasuna)
90
shared ethnonational identity, a common grievance of perceived persecution by the Spanish state,
and a desire to achieve an independent Basque homeland by whatever means necessary gave ETA a
high level of identity coherence. In addition to aiding in recruitment of Basques sympathetic to the
ETA cause, the identity coherence of the group provided a powerful source of motivation for its
members to engage in illegal actions against the Spanish state and its perceived sympathizers, despite
the high costs of engaging in such actions and the low likelihood of success. In this way, then,
identity coherence is an especially important variable in determining outcomes of political violence,
as higher identity coherence provides tactical, organizational, and psychosocial resources that are
more likely to sustain longer campaigns of violence.
63
Elite support, the second key variable in the theory and associated typology, refers to the
extent to which an identity-based group has its interests represented in positions of political power.
Elite support, like identity coherence, can range from low to high. At the low end, this might come
in the form of tacit statements of support from unaffiliated political elites that expand the bounds of
what is socially acceptable, creating a permission structure for the group to operate in ways that
would have been unthinkable before.
64
At the high end, this could mean a political party that is
aligned with the group and advances its interests at the legislative level while also providing financial
resources and political legitimacy. The political wing of Hezbollah in Lebanon’s parliament could be
an example of this level of elite support. Though this article focuses primarily on elite support from
political parties, there are other vehicles that could provide a group with meaningful elite support.
Elite support may also be found in allies of identity-based groups who hold positions of political
63
The second article in the dissertation outlined a microlevel theory of identity salience and political violence, which
included a discussion of the unique power of identity in motivating otherwise high-risk or costly behaviors. Identity
coherence is one manifestation of this at the macrolevel.
64
Former president Trump’s infamous “Stand back and stand by” comment in response to a question about the far-right
paramilitary group the Proud Boys at the September 2020 presidential debate against then-candidate Biden would be one
example of this.
91
power connected to the state—aside from members of parliament or party leaders, this could also
mean high-ranking military officers, close advisers to the executive, or members of the judiciary or
powerful corporations with political connections.
65
Though the presence of sympathetic elites is certainly important to an identity group, the
group is unlikely to be effective or have the resources to achieve its goals without a higher level of
elite support. A group with high elite support, such as Hezbollah, would have a dedicated political
party that is nationally competitive and represented in the central government of the state in which
the identity group operates. High elite support provides a level of political legitimacy and access to
political and financial resources otherwise unavailable to normal constituents. Without access to
national representation through a competitive political party, however, identity-based groups may
still have elite support that can sustain their operations and provide a source of legitimacy and
resources. Returning to the example of ETA in Spain, the group’s political arm, Herri Batasuna, was
a Basque nationalist political party that was at least nominally represented in Madrid and that
provided a level of elite support that facilitated the group’s continuation of political violence over
the course of multiple decades. However, even at its peak, HB won only five seats in the Spanish
parliament, with over 170 needed for a majority—additionally, its electoral success was regionally
limited to the Basque country and Navarre (Publico). Eventually, Herri Batasuna was barred from
political participation by the Spanish government, and ETA dissolved later.
66
Even a minimal level
of elite support can be powerful in fostering the conditions for the strengthening of identity-based
movements. When there is a party that implicitly or explicitly adopts a platform or motivations of an
identity group, the likelihood of political violence’s occurring on the one hand, and the potential for
65
Elite support from these other sources, though not examined in this article, could provide a useful avenue for future
research.
66
HB was later rebranded as Batasuna, which was the party’s name at the time it was outlawed after evidence of its
financial support of ETA was revealed.
92
that violence to become more severe on the other hand, increases relative to changes in elite support
and identity coherence.
Given these definitions of identity coherence and elite support, I outline a typology within
which variation in political violence outcome severity can be situated.
67
The typology classifies
identity-based political violence outcomes into three ideal types: stochastic political violence,
sustained insurgencies, and civil wars. It is important to note that although the typology
differentiates between outcomes based on severity, severity will vary within each outcome type in
the typology. Though there are different types of political violence that can occur within these ideal
types, I condense these for the sake of conceptual and analytical clarity (figure 3.1).
68
I discuss the
operationalization and measurement of the key variables in further detail starting on page 16 (see
also Figure 3.2).
It is important to note that although this article focuses on identity-based political violence at
one point in time, it is certainly possible, and in some cases likely, that a political violence event can
become more or less severe within the typology due to ongoing changes in identity coherence
and/or elite support. However, we do not discuss the mechanisms by which these changes occur in
this article.
68
I provide examples of categories of violence that fall into each ideal type, but this is not an exhaustive list.
93
Elite support
(Low) (High)
(High)
Identity coherence
(Low)
Figure 3.1: A Visualized Typology of Identity-Based Political Violence
Type 1: Stochastic Political Violence
I classify Type 1 outcomes as stochastic political violence.
69
This is the outcome most likely
to occur when the identity-based group in question has a relatively low level of identity coherence
and elite support. A group situated at this point in the typology would likely have some level of
identity coherence—for example, they would adhere to a shared ethnic or religious identity and
might have some cohesion around common aims or goals. However, such a group would likely have
a certain level of disagreement about their goals and tactics as to render them operationally less
69
I borrow this terminology from the concept of stochastic terrorism, a term that has gained increased popularity in
American public discourse in the Trump era that refers to the omnipresent threat of random acts of political violence
incited through elite demagoguery that casts an outgroup in a particularly unfavorable light (Amman and Meloy 2021,
Kleinfeld 2021, Keats 2019).
Type 2: Sustained insurgency
Guerrilla warfare, nonstate violence,
sustained terrorism campaigns
Type 3: Civil war
1000 or more battle deaths per year
where one side is the government of a state
Type I: Stochastic political
violence
Protest-related violence, riots, electoral
violence
94
effective than other identity-based groups with higher identity coherence. At the same time, their
level of elite support would be low. They may have sympathizers among the political or military elite,
but this support would be unlikely to have been consolidated into a dedicated political party or
movement. Support from elites at this level may look like tacit support provided through under-the-
radar funding, sympathetic rhetoric, or a willingness among elites to look the other way in response
to socially unacceptable actions, including violent rhetoric or violence itself. As a result of these low
levels of identity coherence and elite support, it is unlikely that an identity-based group with low
identity coherence and low elite support will lead to violence on a large scale—generally isolated
instances of violence are a more likely outcome, though these instances may also occur around high-
profile events, such as elections or national holidays. I refer to this type of political violence as
stochastic political violence to reflect the fact that although acts of political violence perpetrated by
the group occur, they are likely to appear uncoordinated or sporadic relative to violence outcomes in
types 2 or 3 in the typology. More specific examples of type 1 political violence can be seen in such
events as intercommunal violence between pastoralist ethnic groups in sub-Saharan Africa, electoral
violence between ethnically affiliated party supporters in Kenya, and even the January 6 insurrection
in the United States. Such examples are consistent with the expectations of the theory in that the
perpetrating actors are relatively disorganized, the violence not extensively planned, and the result
low numbers of deaths compared to more severe outcomes in the typology.
Type 2: Sustained Insurgencies
At the next level up from Type 1 is Type 2, or sustained insurgencies.
70
Political violence
events at this level typically involve identity-based groups with relatively high levels of identity
70
Violence within these insurgencies may take the form of irregular or guerrilla warfare, frequent terrorist attacks, and
other acts of violence that fall short of civil war in scale and scope.
95
coherence, but their levels of elite support tend to be low to medium, precluding their being more
organized and powerful against their opponents. Type 2 events may vary in severity from low-level
insurgencies (for example, ongoing Basque separatist violence perpetrated against the Spanish state
by ETA) to high-intensity insurgencies that stop short of full-scale civil war (i.e., the Kurdish
separatist insurgency in southeastern Turkey spearheaded by the PKK). In addition to insurgencies,
identity-based political violence within this type may include state repression of an ethnic or religious
group, such as the violence perpetrated by the Burmese state against the ethnic Rohingya. Actors in
type 2 events typically have higher identity coherence than do those in type 1 events, but their level
of elite support does not rise to a level that allows them to gain the tactical and strategic upper hand
against the state that would be required for a civil war to occur. Unlike type 1 violent actors, actors
within type 2 events organize around more clearly defined identity lines, and they tend to have a
higher degree of consensus regarding shared goals and tactics. At the same time as these actors have
a high degree of identity coherence, they have a low level of elite support relative to type 3 actors.
Elite support among actors at this level may take the form of support within an established political
party, although this support may be more tacit and rhetorical than operational. As an example, Irish
republicans in the Provisional Irish Republican Army (PIRA) during the Troubles had a political arm
in Sinn Fein, though the party denied that it was in fact the political arm of the militant group. In
addition, there may be parties or political elites not officially affiliated with a movement but that
subscribe to the same identity as the insurgent actor—even if they are not explicitly supportive of
the group or its actions, the presence of actors sympathetic to the general cause and belonging to the
same identity group can contribute to the higher level of elite support characteristic of type 2 events
than is present among actors in type 1 events. Though more moderate Irish republican parties like
Fianna Fail did not expressly support violence, the presence of Irish republican voices in positions
96
of power granted a level of legitimacy to the general idea of Irish independence, albeit in a
nonviolent way.
Type 3: Civil Wars
The most severe political violence outcome in the typology is Type 3 outcomes, or civil
war
71
. To be the most likely outcome in the typology, high identity coherence must be matched by
high elite support. Actors that successfully engage a state’s government in a conventional civil war
will have a very high level of identity coherence: they will draw support almost exclusively from a
particular ethnic or religious identity group, and their mission will be supported by a wide base of
the ethnic or religious population in that state. These actors will necessarily have clearly defined
goals and a consensus about how to achieve these goals—namely, through violence and armed
conflict. Their level of elite support will also be higher than levels of elite support for actors in types
1 and 2 conflicts. They will likely have a dedicated political arm, frequently in the form of a political
party that also draws broad support from constituents within their identity group. Unlike in type 2
conflicts, these elites will also likely have enough power within the state to threaten the status quo of
the state government, but they will be unlikely to be the dominant political group. For example, the
ethnonationalist ambitions of the Tigrays in Ethiopia and their high level of elite support and
political capital in the form of the Tigrayan People’s Liberation Front (TPLF) were such that the
group’s high identity coherence and elite support helped lead to the conditions for a full-scale civil
war in the country. These examples are emblematic of the types of actors and conditions typically
necessary for involvement in political violence at the most severe level: the combination of a clear
and cohesive identity grouping with agreed upon goals and a high—yet not dominant—degree of
political institutionalization and elite support increases the likelihood that such actors will have both
71
I use the conventional definition of a civil war as a high-intensity war between an organized, armed opposition and the
government of a state.
97
the motivation and the means to pursue escalation in the name of protecting or promoting the aims
of their ingroup.
According to the theory, Type 4 political violence (low identity coherence and high elite
support) is highly unlikely. When identity coherence is low, the incentives for and ability of elites to
provide support to the group are low. The costs are too high for elites to signal support for a group
that is substantively and organizationally fragmented, and the benefits of supporting a group with
minimal coherence are too low. In this way, then, some level of identity coherence is a necessary
precondition for elite support, and an absence of identity coherence will almost always translate into
an absence of high elite support—therefore, type 4 outcomes are theoretically unlikely to occur, and
a lack of empirical evidence and cases of identity-based political violence that could be classified as
type 4 means that we will leave this box in the typology empty.
As a final point of clarification, it is not the case that all conflicts within this typology are the
same; rather, conflicts vary along a spectrum in terms of their constitutive dimensions. By necessity,
all typologies are simplifications of complex real-world events (Bennett 2013), yet they can be useful
in taking these complex phenomena and creating cognitive schemas for thinking about causal
mechanisms and processes that may lead to different outcomes (Collier et al 2012). Similarly, this
typology is useful in beginning to think about how different political violence outcomes may be
understood in relation to preexisting and changing levels of identity coherence and elite support. Of
course, there is often variation within these ideal types, but the typology presented here is a start in
proposing a conceptual and theoretical framework for thinking more systematically about the role of
identity in political violence: why some outcomes are less severe, and whether and why these
outcomes might escalate to more severe ones.
This typology is an analytical tool meant to facilitate a more systematic approach to
understanding identity-based political violence as a phenomenon distinct from other types of
98
political violence, as well as to provide a means by which to categorize different types of identity-
based conflict and violence events in terms of severity. The main assumptions of the typology, then,
are that: a) the substance of an identity matters and different identity types have different effects;
and b) once violence has occurred, the severity of violent outcomes increases relative to levels of the
two main dimensions of identity coherence and elite support. This leads us to two main hypotheses
that will be tested here:
H 1: Identity-based conflicts will be more violent than those that are not identity-based—battle deaths will increase
significantly given an identity-based conflict dyad versus a non-identity-based dyad.
H 2: The severity of the violent outcome will vary based on levels of elite support for that identity group and its identity
coherence. Higher levels of elite support and identity coherence will predict more violent outcomes, as indicated by battle
deaths.
Though there are several potential hypotheses that could be derived from this theory and
typology, the two outlined here are fundamental to testing the plausibility of the theory and will be
the focus of our empirical analysis in this article. We now discuss the data and methodology used to
conduct this plausibility probe.
Chapter 3: Research Design
To test these hypotheses and probe the plausibility of the theory, I leverage disaggregated
dyadic conflict event data initially taken from the Uppsala Conflict and Data Program’s (UCDP)
Georeferenced Events Dataset (Sundberg et al. 2013, Davies et al. 2022). Using this data, I apply a
novel coding scheme to operationalize and test the effect of two new variables, identity coherence
and elite support, on violence severity. I narrow the analysis frame from over 290,000 unique dyadic
conflict events that began between 1989 and 2022 by removing any event that occurred after 2019,
was an interstate conflict, and/or was a conflict involving certain types of transnational actors
composed of members from more than one country. I also exclude dyads where both actors are of
99
the same ethnic or religious group, as intra-ethnic conflicts are likely to follow different logics than
are interethnic ones and as such are beyond the scope of this article.
72
Of the remaining dyads, I
then aggregate total deaths per conflict dyad. Of 1577 unique dyads, I identified 380 dyads that
engaged in identity-based political violence, or violent events where the perpetrating actor in the
dyad was fighting on behalf of an ethnic or religious identity group. From these 380 dyads of
identity-based political violence, I took a random sample of 150 for further analysis.
73
The sample
represents a wide range in terms of violence severity, with the minimum death count being 25 (the
minimum threshold required for inclusion in UCDP data) and the maximum being over 12,000
deaths. The perpetrating actor (side A in UCDP data) was then coded to reflect key indicators of
identity coherence and elite support at the time of conflict initiation.
74
Although I do expect that the
identity coherence and elite support of both actors in a dyad are important in contributing to the
outcome, I consider only the perpetrating actor in predicting outcome severity due to time and space
constraints. I rely on the UCDP designation of the perpetrating actor, listed as the first side in any
UCDP conflict dyad. Future iterations of this might consider incorporating the coding of multiple
actors in testing the relationship between identity coherence, elite support, and violence severity.
Sources of information for coding came primarily from UCDP’s descriptions of conflict
dyads and actors
75
, as well as from the Ethnic Power Relations (EPR) datasets (Vogt et al. 2015). A
codebook outlining the data, coding process, and criteria in detail can be found in Appendix 3A.
72
For example, the Wanihem and Wanikade communities in Cross River State in Nigeria engaged in a conflict over land
and resources that killed 150 people but belong to the same ethnic group (the Ukelle). Because they do not organize
along explicitly ethnic lines in this dyadic episode, we do not consider them IBPV. Full details of the exclusion and
inclusion criteria can be found in the codebook within Appendix 3A.
73
To ensure that this sample size would be sufficient to detect statistically significant results and reduce the risk of a
Type II error, I conducted a power analysis using the pwr package in R. To uncover a medium effect size with high
statistical significance (p<0.001), the necessary sample size is n=138. I chose to draw a random sample of n=150 to
provide more degrees of freedom and further reduce the possibility of a type II error, even if marginally.
74
An actor’s level of identity coherence and elite support may certainly vary over time. However, due to time and space
constraints, I rely on a static approach to understand the relationship between these key variables and conflict severity.
Future research could expand this approach to look at actors across time to better understand whether and how changes
in these variables affect the evolution of violence and patterns of severity.
75
https://ucdp.uu.se/encyclopedia
100
The main advantage of using disaggregated data at this level is twofold. First, the more fine-
grained nature of dyadic data facilitates a clearer understanding of the characteristics and motivations
of armed actors than is possible by examining political violence at the level of the larger conflict.
Additionally, looking at dyad-level data allows for more variation in severity, as the threshold for
inclusion in UCDP dyadic data is lower than for other large conflict datasets, such as Correlates of
War. Finally, the focus of the data on the actor, rather than the conflict, is important in highlighting
the actor-specific characteristics that contribute to identity coherence and elite support and that are
central to the theory. This approach can be replicated on other actor-level data, which can then be
supplemented with conflict-level data.
Figure 3.2: Operationalization and Measurement of Key Variables
76
Variable Name Indicator(s) Data Source(s) Measurement
Identity coherence Identity type
Splinter group
Clear goals
Type of goal
Group representation
Recruitment from group
Broad population support
UCDP
Ethnic Power Relations
Latent variable using
Bayesian IRT
Elite support Political party
Access to state power
Regional Autonomy
UCDP
Ethnic Power Relations
Weighted Index
Severity Number of deaths UCDP Count data
76
This table is adapted from a more detailed version in the codebook. To see this original table and further detail on the
coding scheme, see Appendix 3A.
101
Key Variables
I use the data to test the relationship between identity coherence, elite support, and the
severity of political violence outcomes. Though there may certainly be other ways to operationalize
these variables, this approach represents a sound effort to create measures with high construct
validity in accordance with the theory and that will assist in the initial quantitative testing of the key
hypotheses.
Dependent Variable: Severity of Violence
I operationalize the dependent variable of outcome severity as the total number of battle-
related deaths attributed to the dyad, as measured by UCDP. Deaths include not only those
sustained by combatants, but also those of civilians, whether they were targeted specifically as part
of the dyad or killed collaterally as the result of an attack in their vicinity. A significant advantage of
using UCDP data rather than conflict data from other sources, such as the Correlates of War (CoW)
data project, is not only that it includes non-combatant deaths in the counts, but also that these data
include a much wider variation of conflict events that differ in scope and severity. Unlike the
Correlates of War data, which require a conflict event to result in at least 1000 battle deaths in order
to be classified as a civil war, the UCDP criteria set a lower inclusion threshold of 25 deaths in order
for an actor to be included. This less restrictive inclusion criterion allows not only civil wars, but also
insurgencies and smaller-scale violence to be included in the data, thus increasing the
representativeness of the data across different types of political violence that vary in scope,
magnitude, and severity.
Explanatory Variable 1: Identity Coherence
I operationalize identity coherence as a latent variable that incorporates multiple theoretically
relevant manifest variables. To maximize construct validity, I collect data on a variety of indicators
102
that the theory considers to be directly related to identity coherence and fit a measurement model
that assigns each unique actor in a dyad a single identity coherence score. These indicators represent
both the substance of the identity subscribed to by identity-based actors, as well as the degree of
commitment to this identity and the level of consensus among members of the group regarding
membership criteria and shared goals. Specifically, I measure the type of identity espoused by an
identity-based actor (i.e., ethnic, religious, or a combination of the two). To measure the degree of
consensus among group members regarding membership criteria and shared goals, I code whether a
given identity-based actor is a splinter group of a larger group (dichotomous); whether there is a
clear expression of a shared goal or goals (dichotomous), and, if so, what the type of goal is. The
categories for goal type range from protection and self-defense at the lower end of the scale to
genocide and ethnic cleansing at the highest end. When available, I include data from the Ethnic
Power Relations (EPR) ACD2EPR dataset (Wucherpfennig et al. 2012). Designed for compatibility
and easy merging with UCDP dyadic data, ACD2EPR includes three useful variables related to
identity coherence: claims, an ordinal variable measuring whether a rebel group has made an
exclusive claim to fight on the behalf of an identity group; recruitment, an ordinal variable indicating
whether a rebel group is recruiting from a specific ethnic or religious group; and support, which
indicates whether the actor/group in question is supported by at least 50% of an ethnic group.
These variables speak further to the level of agreement shared by members of an identity-based
group as to who and what they represent. The support variable is a useful indicator that better
captures whether the level of identity cohesion within a group translates to a consolidation of
support among noncombatant members of that group.
103
To calculate each unique actor’s identity coherence score, I fit a polytomous
77
item response
model that models the values of each manifest indicator as a function of the item parameter for the
latent trait of identity coherence. Item response theory assumes that the structure of the underlying
data within an IRT model can be used to create a unique score for each unit of observation—in this
case, an identity group—that both reflects this structure and systematically synthesizes the available
data in the form of a latent trait score. Unlike the variables used to measure elite support, those used
to measure identity coherence fit under the assumption of local independence: that is, the specific
indicators used to measure identity coherence are expected not to directly predict the other, but
rather can be better understood through an examination of the underlying relationship between each
indicator. That underlying relationship indicates the latent trait of identity coherence, a variable
difficult to observe directly but that can be modeled as a function of distinct but theoretically related
manifest variables. Given the assumption of local independence and the ordinal structure of the
manifest variables, I use a Bayesian item response method that calculates the latent trait value by
sampling from a simulated posterior distribution based on an ordinal data factor analysis model.
78
As
expected, the mean identity coherence score is normally distributed, with scores below zero
indicating lower identity coherence and scores above zero indicating higher identity coherence. A
more detailed description of this process, as well as reproducible code, can be found in Appendix
3B.
77
A polytomous item response model refers to models that use manifest indicators that have more than one response
option—for example, if the items that are used in the model are not dichotomous.
78
Specifically, I use the MCMCPack R package to run this model (Martin et al. 2011). Given the relatively small sample
size of the data, I run the model with 25 Markov Chain Monte Carlo simulations and assign normal priors to the
parameters, with unconstrained discrimination. A more detailed discussion of this methodology can be found in article
1’s presentation of identity salience as a latent trait.
104
Explanatory Variable 2: Elite Support
I operationalize the second main predictor, elite support, as a composite index, with three
constituent indicators representing: 1) an identity actor’s affiliation with an established political party;
2) a variable taken from EPR Core data indicating the level of access an identity actor has to state
power at the national level; and 3) a second variable from EPR Core indicating the actor’s level of
access to executive state power at the regional level. The first variable included in the index is an
ordinal variable indicating whether an identity-based group is affiliated with a political party. Those
with no political party affiliation are coded as 0, and those with implicit ties to a party are coded as 1
(an example of this would be the IRA and Sinn Fein during the Troubles). Those actors with a
dedicated political party are coded as 2—if “Tigrayans” were an actor, for example, we would code
them as 2 for this variable because of their political arm of the Tigrayan People’s Liberation Front
(TPLF). The second variable included in the index is taken from the EPR Core dataset. The ordinal
variable status measures the relative state power that an ethnic group possesses at the national level,
with values measuring whether actors are powerless, share power in the central government, or are
dominant in the central government.
79
Finally, reg_aut is a dichotomous variable indicating whether
a group has meaningful access to both de jure and de facto executive power at a regional level.
The smaller number of indicators and lower variation in values indicates that a simple
composite index calculation is more sensible for measuring elite support than is a measurement
model. In calculating the composite index, I again weight the three constituent indicators according
to their expected theoretical importance. Because the first two items in the index are of roughly
equal theoretical and empirical importance, I weight them equally in computing the index score for
each identity-based actor. I weight the regional autonomy variable slightly less, as I expect this to be
79
This only considers access to executive power. Power through access to judicial or legislative institutions is not
considered in the coding of this variable. See codebook in the Appendix 3A and/or EPR documentation for further
detail.
105
less important to an actor’s overall level of elite support than their access to power at the national
level. For example, although an actor with regional autonomy certainly would have higher levels of
elite support than would one without, access to power at the federal level is particularly crucial to the
potential for violence escalation and severity. I scale the elite support score from 0 to 1 for ease of
analysis, with a score closer to 0 representing low elite support and a score closer to 1 indicating high
elite support. I impute missing data using row means of the available elite support indicators for
each actor before calculating the composite score.
Model Estimation
Because our dependent variable is operationalized as total deaths and represents over-
dispersed count data, I specify negative binomial models to test both hypotheses H 1 and H 2.
80
To
test H 2, I also fit an ordinal logit model to test the dependent variable when it is converted to an
ordinal variable to match outcome severity as outlined in the typology. The model equation for
testing H 1, or the expectation that the presence of an identity-based actor in a conflict dyad will lead
to more severe violence, is the following:
log
deaths
= α + βx
id-based
+ ε,
where log deaths is the logged count of deaths that occurred as a result of violence within that dyad, and
β id-based is a dichotomous variable that takes a value of 1 if the conflict dyad is identity-based, a 0 if it
is not. I control for conflict duration to minimize bias related to more deaths occurring as a function
of time.
80
Though I also estimate Poisson models, a preliminary probe of model fit indicates that there is indeed overdispersion
of variance in the dependent variable—therefore, the negative binomial specification is a nominally better fit than the
Poisson.
106
I also fit a negative binomial model to test H 2, or the hypothesis that an increase in identity
coherence and elite support will lead to more severe violence outcomes. Again, I rely primarily on a
negative binomial rather than Poisson model to address the problem of overdispersion in the
dependent variable. The model equation is as follows:
log
deaths
= α + βx
identitycoherence
+ βx
elitesupport
+ ε,
with β identitycoherence representing the level of identity coherence and β elitesupport representing level of elite
support. Again, I control for duration across models.
To test the alternative, ordinal version of the dependent variable, I also fit an ordinal logit
model with the same equation as above to provide an additional test of H 2. However, the dependent
variable is noted as Y type to reflect that the outcome being modeled is ordinal, rather than count data.
Adding this model specification not only allows for a more precise match between the outcomes as
they are presented in the typology and the data itself, but it also ameliorates the issue of
overdispersion in the dependent variable that occurs when it is tested as count data. Categorizing the
dependent variable into ordinal levels of severity allows us to test a more normally distributed
outcome variable, thus minimizing bias introduced by skewness in the negative binomial model and
providing a helpful robustness check. These results can be found in Appendix 3B.
Finally, I include measures of regime type using the same model specifications as above.
Given the centrality of the idea of elite support, which relies at least partially on some degree of
electoral representation, I incorporate Polity5 scores, using the original scaling of the variable used in
the original Polity dataset—scores range from -10 to +10, with countries scoring -10 representing
strongly autocratic regime types and countries with +10 representing strongly democratic regime
types (Marshall and Gurr 2020). I estimate negative binomial and ordinal logit models incorporating
107
this data, using each country’s Polity score at the start date of the conflict dyad as a unique predictor.
I also specify the same models using an interaction term of elite support and Polity data.
Chapter 3: Results
Overall, the results of the quantitative analyses indicate strong support for the main
hypotheses. As proposed by H 1, an identity-based conflict dyad will result in more violent outcomes
than will a non-identity-based conflict dyad, and the results of the statistical analyses provide support
for this hypothesis. There is a strong and statistically significant relationship between an incident of
violence’s being identity-based and the severity of that incident, as measured by increased deaths.
The coefficient on the dichotomous predictor indicating whether an event is identity-based indicates
that a conflict’s being identity-based is associated with a 0.535 unit increase in the log count of the
dependent variable, or total deaths attributed to the event. Exponentiating the coefficient of the
main negative binomial model, we can get a clearer understanding of the substantive effects of a
conflict dyad’s being identity-based: the exponentiated coefficient, or the incident rate ratio
81
,
indicates that an identity-based conflict results in roughly 70% more deaths than does a non-
identity-based conflict. In other words, identity-based conflicts are substantively and significantly
more violent than non-identity-based ones.
81
The incident rate ratio is 1.707.
108
Dependent variable:
Total Deaths
OLS
negative
Poisson
binomial
(1) (2) (3)
Identity-Based 94.762 0.535
***
0.258
***
(63.204) (0.090) (0.004)
Duration (Days) 0.424
***
0.001
***
0.0002
***
(0.020) (0.00003) (0.00000)
Constant 182.194
***
5.104
***
5.590
***
(46.780) (0.067) (0.003)
Observations 699 699 699
R
2
0.381
Log Likelihood
-4,613.975 -259,869.700
theta
0.706
***
(0.032)
Akaike Inf. Crit.
9,233.950 519,745.500
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Table 3.1: Main Results, H 1
H 2 posited that an increase in identity coherence and elite support would lead to more severe
outcomes of political violence. The results provide strong evidence in support of this hypothesis,
and so we can confidently reject the null hypothesis that there is no relationship between a
perpetrating actor’s level of identity coherence and elite support on the one hand, and conflict
109
severity on the other. Given over-dispersion of the outcome variable, I again rely primarily on
negative binomial models to test the hypothesized relationships. In the main model using only
identity coherence and elite support as predictors of total deaths, both predictors are substantively
and statistically significant, with coefficients in the expected direction. Given a one unit increase in
the perpetrating actor’s identity coherence score, the log count of deaths increases by 0.85.
Substantively, the incidence rate ratio indicates that this translates into an increase in nearly 1.5 times
as many deaths as would have occurred without this increase in identity coherence.
82
When elite
support for the perpetrating actor increases by one unit, the log count of deaths increases by 0.649,
or a nearly 90% increase in resulting deaths (IRR = 1.91). Full results are reported in table 3.2.
83
Given the theoretical relationship between the main predictors, I also test the interaction
effect between them, the results of which are reported in table 3.2 as model 5. Exponentiating the
coefficient to get the incident rate ratio, we see that the effect of identity coherence on outcome
severity is conditional on the level of elite support, but that the these conditional effects do not
occur as theoretically expected. The negative coefficient on the interaction term indicates that as
identity coherence increases, the effect of elite support on outcome severity decreases marginally.
Independently of each other, the two predictors of identity coherence and elite support are
positively related to violence severity, but together they do not increase in tandem as the theory
would expect. Rather, as identity coherence increases, the effect of elite support on severity
decreases slightly. A visualization of the marginal effects of identity coherence given different levels
of elite support can be seen in figure 3.3 below.
82
Incidence rate ratio = 2.34
83
I report results from Poisson regressions as well, with the caveat that model fit is substantially less than that with a
negative binomial specification.
110
Figure 3.3: Predicted Probabilities of Identity Coherence*Elite Support Interaction
Though these results do not negate theoretical expectations directly, they do encourage
caution and suggest that further testing may be merited. The typological theory did expect some
relationship between identity coherence and elite support, but the outcome as outlined in the
typology depend on the levels of each of the main predictors independently of one another. While it
is certainly possible, and even expected, that levels of both would increase or decrease proportionally
to the other, this is not a prerequisite of the typology. While these marginal effects are unexpected,
one possible interpretation is that while identity coherence is powerful on its own, increasing levels
of elite support have a moderating effect. It could be the case, for example, that as elite support of
an identity group increases, so too does the incentive to moderate and engage in the political
process, rather than resorting to violence. Further testing on an expanded sample is merited to
better understand the interactive relationship between the two variables.
111
Dependent variable:
Total Deaths
negative Poisson negative Poisson negative
binomial
binomial
binomial
(1) (2) (3) (4) (5)
Identity Coherence 0.850
***
0.729
***
0.623
**
0.447
***
1.266
***
(0.188) (0.007) (0.268) (0.011) (0.272)
Elite Support 0.649
*
0.320
***
-0.779 -0.937
***
0.655
*
(0.363) (0.014) (0.575) (0.025) (0.360)
Group Size
-0.392 -0.439
***
(0.699) (0.030)
idco:elitesupport
-1.455
*
(0.746)
Constant 5.831
***
5.952
***
6.346
***
6.431
***
5.823
***
(0.153) (0.006) (0.253) (0.010) (0.152)
Observations 150 150 88 88 150
Log Likelihood -1,045.640 -67,787.330 -611.471 -36,592.970 -1,043.760
theta 0.655
***
(0.064)
0.737
***
(0.095)
0.667
***
(0.065
)
Akaike Inf. Crit. 2,097.280 135,580.700 1,230.941 73,193.940 2,095.521
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Table 3.2: Main Results for H 2 (Dependent Variable as Count Data)
112
In addition to fitting the main regression models using the count of deaths as the dependent
variable representing outcome severity, I also fit an ordinal logit model with a transformed version
of the dependent variable. Taking the duration of the conflict between actors in a dyad in years, I
calculated the rate of deaths per year, using this to categorize each dyad’s severity into low (less than
100 deaths per year), medium (between 100 and 999 battle deaths per year), and high (1000 or more
deaths per year). Though the coefficients are in the expected direction, they are not statistically
significant in the base model. The results of the ordinal logit underscore the difficulty in categorizing
the severity of political violence in a meaningful way that captures the full spectrum of political
violence outcomes. For example, UCDP does not include any dyad that does not result in at least 25
conflict-related deaths, making it more difficult to understand the correlates of lower-intensity
political violence. These results are reported in full in Appendix 3B.
Finally, I report results of models that incorporate regime type data, both as a control
variable and as a constituent part of an interaction term with elite support (Appendix 3). Despite a
reduced sample size due to unmatched Polity scores for certain dyads, identity coherence is again
strongly predictive of violence severity and remains highly statistically significant, as is a country’s
regime type (as measured by its Polity score). As expected, an increase in a country’s Polity score—a
move closer toward partial or full democracy—decreases the severity of conflict. The relationship
between elite support on its own and outcome severity becomes more opaque, however, as the
coefficient changes direction based on the inclusion of regime type as a predictor. At the same time,
the relationship between the interaction term and outcome severity shows how increased elite
support might become conducive to more severe conflict outcomes when it increases in conjunction
with a country’s Polity score—in other words, high levels of elite support may not be uniformly
predictive of outcome severity across different regime types. Though a more specific examination of
113
the conditional effects of elite support given different regime types is beyond the scope of this study,
the preliminary results presented here suggest this might be a worthwhile topic of further research.
Though the typology presented and preliminarily tested here admittedly oversimplifies
political violence outcomes in terms of severity by collapsing outcomes into three outcome types—
within which there is substantial variation—it provides a framework upon which to begin thinking
about this issue more systematically. The results of the quantitative analysis, then, have the dual
effect of demonstrating the plausibility of the theory advanced in this article, while also highlighting
remaining questions and issues to be addressed.
Chapter 3: Demonstrating Identity-Based Political Violence: Three Brief Case Illustrations
The quantitative analysis underscored the importance of systematically incorporating
multidimensional considerations of group attachments to identities, as well as the level of elite
support received by these groups, in understanding outcomes of identity-based political violence. As
effective as quantitative analyses are in demonstrating the plausibility of a theory and testing its
hypotheses, they are less effective in demonstrating the qualitative conditions and causal processes at
play. This is especially true of theories that center less easily quantifiable variables such as identity in
consideration of political outcomes. To provide examples of the theory at work in real events of
identity-based political violence, I present brief case illustrations for each ideal type of political
violence. These short illustrations represent typical cases for each type but were not included in the
sample used for quantitative analysis due to either temporal restrictions of the dataset, high
thresholds for inclusion, and/or due to the nature of random sampling that yielded the subset of
dyadic cases analyzed. Therefore, I select three cases that exemplify types 1, 2, and 3 identity-based
political violence respectively and that vary regionally and temporally. In selecting these cases, I
searched for examples of non-internationalized political violence or conflict events in countries with
114
a functional level of electoral democracy
84
, thus minimizing bias that could have been introduced by
variation on these important covariates that are not a part of the theory.
In examining these examples of identity-based political violence, I focus on the identity
coherence and elite support levels for the main perpetrators in each conflict in the period
immediately preceding and at the point of violence initiation, providing illustrations of what low to
high identity coherence and elite support look like qualitatively. For type 1, stochastic political
violence, I focus on the case of white supremacist violence perpetrated by the Ku Klux Klan in the
1920s. I rely on an analysis of Irish republicans during the Troubles in Northern Ireland to illustrate
a typical case of type 2 outcomes (sustained insurgencies). For type 3, civil wars, I discuss the case of
the Kataeb (otherwise known as Phalangists) at the beginning of the Lebanese Civil War.
For each case, I provide a very brief background of the event and introduce the main actor
to be analyzed for each. I then discuss each actor’s level of identity coherence and elite support,
demonstrating what different levels of these key predictor variables look like qualitatively. Finally, I
address key alternative explanations.
Type 1: White Supremacist Political Violence during the Second Wave of the Ku Klux Klan
Many of the larger datasets on political violence that have shaped our understanding of
political violence have historically underrepresented political violence in the United States, either
because American political violence after the civil war has not met the threshold for inclusion or
because of temporal restrictions.
85
Despite this, there are a number of noteworthy incidences of
84
As demonstrated through the quantitative analysis, regime type is an important factor affecting the outcomes of
identity-based political violence events. Because regime type is not a key predictor in the theory, however, I attempt to
minimize bias in case selection by ensuring that the state in each case offered a basic level of electoral democracy. As
such, the regime type for each of these cases as measured by Polity data equals or exceeds 6, putting each case in the
range of varying levels of electoral democracy.
85
The political explanations behind the relative dearth of research on American political violence should be considered
here as well. Considering domestic actors capable of carrying out domestic political violence is something that (at least
until the present day) has contradicted the national narrative of American exceptionalism. Institutional studies focusing
115
political violence in the United States that can be categorized within the typology as Type 1
outcomes, or stochastic political violence. To illustrate this outcome in the typology, I rely on the
case of racial political violence committed by white Americans against Black Americans over a
particularly violent period during the 1920s. This period was immediately preceded by the so-called
Red Summer, when race riots—led by disillusioned white Americans returning from war—against
Black Americans led to death and destruction in more than 26 cities across the US. Though this
violence was primarily perpetrated by individuals unaffiliated with the Ku Klux Klan, this initial bout
of racial violence and grievance provided a potent recruiting tool for the white supremacist
organization that helped lead to the so-called “Second Wave” of Klan activity. It is estimated that
nearly 1 in 30 Americans, or roughly 4 million, were members at its height, with wide geographic
distribution (Chalmers 2002). During this period, dozens of Americans were killed, and hundreds
more were brutally attacked by KKK members. Though a majority of these victims were Black,
many were white non-Protestants: for example, Jews and Catholics were also targeted by this
staunchly white Protestant group (McAndrew 2017). This relatively forgotten period of KKK
political violence constitutes a typical case of type 1 identity-based political violence, or stochastic
political violence. Events within this type are most frequently low-fatality and low-intensity, but
nonetheless lead to significant instability and may even escalate to more severe outcomes. Actors
perpetrating type 1 violence are typically organized around a core social identity, but given a
relatively loose organization and lack of consensus regarding clear goals, they are considered to have
low identity coherence. Politically, these actors may have the tacit support of elites and be able to
claim some nominal level of political power through sympathetic elected officials, but they have
on US political violence are few relative to studies on political violence in other countries as a result. In a particularly
significant case of data truncation, the CIA-funded Political Instability Task Force’s data—which Walter (2022) relied on
the make bold claims about the potential for another civil war in the United States—intentionally excludes instances of
political violence in the United States.
116
neither a dedicated political party nor a de facto political wing. As such, these actors have low elite
support. The Ku Klux Klan during the Second Wave of 1920s political violence fits type 1 in the
typology due to its relatively low levels of identity coherence and elite support, which provide
enough mobilization power and some cover of political legitimacy that facilitates disorganized, low-
level political violence.
As was the case throughout most of its post-Reconstruction history, the KKK at the
beginning of the Second Wave centered its message and goals around the protection and promotion
of native-born white Protestant Americans. Though the group had clear delineations of who was
eligible for membership
86
, its lack of a clear goal and strategy to achieve that goal diluted its identity
coherence as the typology conceives of it. On the one hand, it was clear what the organization was
for—more power for and the protection of the values of a white Christian America for native-born
Protestants—it was less clear what exactly it was organizing against. Of course, it targeted its
violence against Black Americans, immigrants, Catholics and Jews, but it had no clearly outlined or
agreed upon goals aside from intimidating these outgroups and attempting to exclude them from
gaining power, economically, politically, socially, or otherwise. The group did not have an exclusive
claim to or majority support from white Christians, nor did it have a clearly outlined plan as to how
to, for example, create a new system of government that expressly excluded non-white Christians by
law.
This lack of a clear direction and sense of purpose led some members to call for an
enhanced electoral strategy to elect “100% Americans” to national, state, and local office. Though
they saw some real success in this endeavor, the KKK’s then-alternative designation as the “Invisible
86
The main governing document of the Second Wave Klan, the Kloran, held that although non-Protestants could
technically become Klan members, they must forswear any allegiance that could conflict with the oaths taken by Klan
members. Klan founder Col. William Joseph Simmons said this regarding membership in front of a congressional
committee: “If the Klan is to secure members on an anti-Roman Catholic, anti-Jew, and anti-Negro appeal, we do not
want such members.” (McVeigh 2009)
117
Empire” underscored its level of elite support at the time (McVeigh 2009, p.25; McAndrew 2017). It
did not have a dedicated political party, precluding it from attaining any semblance of regional
autonomy or concentrated power at the federal level. At the same time, it did exert political
influence, and saw some success in advancing its preferred candidates at all levels of government, to
varying degrees. Though the organization had some political allies and was able to use its position
and connections to mobilize on behalf of its favored candidates and against its unfavored ones—
namely, non-white, non-Protestant candidates from both parties—it otherwise had minimal political
power (McVeigh 2009). As historians have noted, the KKK was neither a well-organized party
putting forth candidates for elected office, nor was it even a faction of a particular party (Chalmers
2002). The KKK occasionally saw secret members or former members run for elected office at the
local, state, and national levels, but the level of secrecy surrounding membership underscored a
degree of social and political sanction that made it difficult for the organization to consolidate elite
support in any meaningful way. Though the Klan was politically influential, only a small number of
open members of the KKK were elected to national office.
87
With the exception of KKK Grand
Wizard David Duke’s 1970s bids for state legislative positions in Louisiana, there were no candidates
that expressly ran on a platform of promoting the values of a white Christian America or that openly
touted their KKK affiliation as part of their campaign strategy (McAndrew 2017, McVeigh 2009).
What is more, the KKK favored candidates from both major parties, Democrat and
Republican. Not only did the KKK not have a dedicated political party during its early 1920s
heydays, but it also lacked an informal political alliance with a political party expressly dedicated to
promoting shared values and policies prioritizing white, native-born Protestants. The Klan
consistently supported candidates from both parties, and their criteria for support ranged from the
87
Known Klan members elected to the US Senate during the Second Wave were Rice W. Means (R-CO), Earle Mayfield
(D-TX), Hugo Black (D-AL), Tom Heflin (D-AL), and W.B. Pine (R-OK).
118
candidate’s courting of Klan support (i.e., by claiming the label of “100% American”) to the mere
fact that the candidate was running against someone the Klan opposed—namely, a Catholic or a
Jew. In one gubernatorial election in Oregon, for example, the loss of the Klan’s preferred candidate
in the Republican primary led them to support Democrat Walter Pierce, whose claim of being 100%
American and his backing of certain anti- “alien” policies favored by the Klan earned him their
favor, and ultimately victory (McVeigh 2009). During the 1924 presidential election, Klan leader
Hiram Evans appeared at both the Republican and Democratic National Conventions. At the
former, Evans successfully silenced outward opposition to the Klan, with Republican nominee (and
eventual victor) Calvin Coolidge refusing to condemn the organization. At the latter, Democratic
party delegates were split on the issue of whether to add a platform plank condemning the KKK,
though they ultimately voted not to. This lack of an organized party and an inability to capture
unanimous support from even a dedicated faction of one of the main political parties translated into
a low level of elite support, which in turn prevented members from continuing to organize around
clearer goals that might have facilitated increased violence.
At its peak of identity coherence and elite support during the Second Wave, the level of
political violence associated with the group never rose above Type 1 stochastic political violence.
Declining elite support in the latter half of the 1920s ultimately led to the group’s second decline,
which would not reverse until decades later. Still, this time period proved to represent the peak of
the group’s identity coherence and elite support, resulting in a period of stochastic political violence
that failed to escalate to type 2 violence.
Type 2: Sectarian Political Violence in Northern Ireland During the Troubles
Within the typology, type 2 political violence—sustained insurgencies—contain perhaps
some of the most high-profile examples of irregular warfare that is destructive but that does not rise
119
to the level of civil war. Actors within type 2 conflicts tend to have higher levels of identity
coherence and elite support than type 1 actors, but significantly lower than those actors involved in
type 3 conflicts. The case I rely upon to illustrate this type of political violence is the Troubles in late
1960s and early 1970s Northern Ireland, a sustained insurgency led by a nonstate actor with
relatively high identity coherence and low-medium levels of elite support. I examine the Provisional
Irish Republican Army, which operated as a sectarian paramilitary group until the official end of the
conflict in 1990 with the signing of the Good Friday Agreement. The group was united around a
Catholic sectarian identity in strict opposition to the dominant Protestant identity in Northern
Ireland. In addition to having a common religious identity that was made even more intense by the
state’s persecution of the Catholic minority, they were united around a common goal of a united
Ireland, governed not by British Protestants but by Irish Catholics. Their high levels of identity
coherence were not matched by elite support, although the presence of an unofficial political arm in
the form of Sinn Fein granted them a medium-low level of elite support relative to actors in type 1
conflicts (such as the Second Wave KKK). The IRA’s levels of identity coherence and elite support
at the early stages of the Troubles, as well as violence patterns consistent with a low-level sustained
insurgency, puts the Troubles and its main nonstate perpetrating actor squarely within Type 2 of the
IBPV typology.
Originally formed as a splinter group of the Official IRA, the Provisional IRA (henceforth
IRA) opposed its predecessor’s engagement in the political process and called more forcefully for
the end of British rule in Northern Ireland and reunification of the island of Ireland. A paramilitary
organization with broad support among Catholics, the IRA engaged in terrorist campaigns and
guerrilla warfare tactics against British troops and government targets; though most of the violence
took place in Northern Ireland, the group did carry out limited campaigns on the UK mainland as
well. Sectarian violence and the targeting of Catholic residents of Derry by predominantly Protestant
120
members of the Royal Ulster Constabulary (RUC) led to the August 1969 Battle of the Bogside,
considered by many to be the unofficial beginning of the Troubles. This violence proved a potent
mobilizing tool for the IRA and provided a steady source of recruits among Catholic republicans
hoping for the end of what they perceived to be anti-Catholic discrimination and the restoration of a
united Ireland free of British rule. The IRA primarily targeted British forces, but also clashed
frequently with opposing paramilitary groups such as the Ulster Volunteer Forces (UVF) and Ulster
Defence Association (UDA), the primary loyalist Protestant paramilitary forces of the conflict. In
addition to engagement with government forces and opposing combatants, the IRA was also known
to target civilians they considered to be sympathizers with the Crown or with the loyalists more
generally (Keefe 2019). The thirty-year conflict resulted in over 3500 deaths.
In thinking about how and why the Irish republican movement took such a violent turn at
the beginning of the Troubles, and how and why the conflict rose to the level of severity that it did,
we can consider the IRA within the context of the identity-based political violence typology as an
actor whose clear identity-based organization and goals distinguished it as one with high identity
coherence. Members of the IRA were nearly unanimously Catholic, and their sense of grievance as a
persecuted minority within Northern Ireland and common anti-British and anti-Protestant sentiment
on their own contributed to the group’s high level of identity coherence. Though initially founded as
a splinter group of the Official IRA, the Provisional IRA quickly took over as the dominant faction
within the IRA, remaining so until the ceasefire at the end of the conflict. The group had a clear goal
of reunification with the rest of Ireland and the end of British rule in Northern Ireland, with a clear
idea of who (British troops, loyalist paramilitaries, and pro-loyalist Protestants) and what they
opposed (British rule and Protestant persecution of Catholics). They recruited almost exclusively
from the ranks of republican Catholics and had a certain level of support among the pro-republican
Catholic population. All these factors contributed to a high level of identity coherence, which
121
provided the group and its members with a consistently clear raison d’être and steady stream of
recruits. There was, however, a degree of disagreement within the republican movement
surrounding political ideology; namely, the extent to which the republican movement would align
itself with left-wing socialism (Treacy 2011; Hudson 2019). However, the core focus of the
republican movement at the beginning of and during the Troubles was on the sectarian divide and
the desire for independence, and therefore the ideological debate over left-wing political alignments
was relegated more to the political realm.
88
Unlike the case of the Second Wave KKK and other type 1 actors, the IRA and other type 2
actors tend to be aligned, whether officially or unofficially, with political parties, giving them slightly
higher levels of elite support. Considering the IRA’s level of elite support according to the typology
and operationalization of elite support as a variable, then, its unofficial
89
affiliation with Sinn Fein as
its political arm granted the paramilitary group a certain degree of elite support that type 1 actors
such as the KKK lacked. Indeed, Sinn Fein was widely considered the Provisional IRA’s subordinate
“mouthpiece” throughout the conflict, despite the party and the paramilitary group’s official claims
to the contrary (Rafter 2005, p. 5), At the same time, its level of elite support was diluted by minimal
electoral success in the early stages of the conflict, as well as a lack of concentrated regional
autonomy. Indeed, the party has been considered by some analysts as secondary, or as an auxiliary
organization, to its armed counterparts in the IRA at the time of the conflict (Hopkins 2010). Still,
the group’s association with Sinn Fein gave it a tool for political organization and (limited) access to
political power at the executive level, though this did not come until later in the conflict.
88
The role of political ideology and internal debates over the same has been recognized by some as a particularly
important factor in how the conflict played out, with some going so far as to claim that IRA was a primarily left-wing
nationalist and socialist movement that turned sectarian as persecution against Catholic republicans intensified (i.e.,
Hudson 2019). This is outside the scope of this study but is certainly worthy of further research.
89
As late as 1997, official Sinn Fein party documents claimed that “Sinn Féin is not the IRA,” but that the party
“recognise(d) and acknowledge(d) the IRA’s stated intention of enhancing the democratic peace process.” (Sinn Féin
Election Manifesto, 1997).
122
Sinn Fein had its roots as a political party that was competitive and operational early in the
20
th
century. Originally founded on a platform of Irish independence from the United Kingdom, the
party has always been a staunchly republican one that has openly advocated for the reunification of
Northern Ireland with Ireland, though its influence and organizational cohesion has varied
throughout its history.
90
Following the end of the Irish Civil War in 1923, the party split into Fianna
Fail and Fine Gael, and it was not until the Provisional IRA’s split from the Official IRA that Sinn
Fein reconstituted itself as a political party. Still, the party initially followed an official policy of
abstentionism, putting forth candidates in elections but refusing to take any seats won within the
British House of Commons or the Irish Dáil Éireann.
Even so, initial electoral success for Sinn Fein was elusive, especially in the earlier part of the
conflict. The same was true for republicans in general at this point in the conflict, as loyalist parties
continued to dominate elections at all levels until well into the 1970s (CAIN).
91
As the political clout
of republican parties in Northern Ireland began to grow, however, power remained concentrated in
more moderate nationalist parties that eschewed violence and claimed that the only solution to the
problem was a political one, best exemplified by the dominant moderate republican party, the Social
and Democratic Labour Party (SDLP; Social Democratic and Labour Party Manifesto, 1975).
Despite the clear link between the IRA as a paramilitary organization and Sinn Fein as a
political arm, it was nearly unheard of for active paramilitaries to be involved in politics, and the
policy of abstentionism continued until the early 1980s—it was not until well into the conflict in
1981 that an active IRA member was elected to British parliament. As a political stunt, the IRA
90
“The objectives of Sinn Féin are to Bring the Proclamation of the Republic of Easter 1916 into effective operation
and to establish the Republic, representative of the People of Ireland, based on that Proclamation.” (Sinn Féin
Bunreacht agus Rialacha, p. 5)
91
The highest performing nationalist republican party (Nationalist Party) in the 1969 Stormont general election came in
fourth overall, winning only 7.6% of the vote and 6 seats at Stormont. Sinn Fein was not yet competitive and would not
overtake the SDLP’s electoral success until 2003, after the Troubles had officially ended in 1998 with the Good Friday
Agreement (Whiting 2016).
123
decided to advance Bobby Sands, incarcerated for his involvement in the conflict and on hunger
strike, as a candidate to fill a South Tyrone republican’s seat in the House of Commons in a sudden
by-election. Still, he did not run as a member of Sinn Fein, and he died before he was able to take
the seat (Keefe 2019, p. 213). Sinn Fein’s political organization accelerated after this initial success,
with pro-republican leader Gerry Adams taking on leadership of the party soon after and winning a
seat as MP from West Belfast in 1982. It was around this time that he began his successful bid
challenging the party’s policy of abstentionism, claiming that there would come a time when “Sinn
Féin (would) be a power in the land.” (“Sinn Féin Vice-President Gerry Adams”) This statement was
the unofficial beginning of the republican strategy of simultaneous political participation and armed
struggle, which would continue until the end of the conflict in 1998 with the signing of the Good
Friday Agreement.
92
The combination of the IRA’s high identity coherence and low to medium-low levels of elite
support can help us to better understand how and why the Troubles were a sustained insurgency:
the protracted and coordinated violence over three decades of armed conflict preclude the event
from being considered an example of type 1 political violence, yet its failure to rise to the level of a
higher fatality civil war also precludes its characterization as such. Had Sinn Fein had more power
earlier on in the conflict, it is possible that the Troubles would have led to violence on a larger scale
than was ultimately the case. Conversely, lower identity coherence of the IRA and a lack of elite
support due to the group’s affiliation with Sinn Fein would likely have meant a lower intensity
conflict more akin to the levels of violence seen in type 1 events. These distinctions are important,
as they can be useful in better understanding when identity-based political violence will occur and
the extent of its severity.
92
Danny Morrison, editor of Republican News and political strategist for Gerry Adams, summed up this strategy with this
well-known line spoken at a Sinn Fein gathering: “Will anyone here object if, with a ballot paper in one hand and an
Armalite in the other, we take power in Ireland?” (Keefe 2019, p. 214; Moloney 2002, p. 202).
124
Type 3: The Kataeb and the Second Lebanese Civil War
Few post-World War II civil wars offer as clear a picture of the role of high identity
coherence and high elite support among armed actors than does that of the second Lebanese Civil
War (1975-1990). This brutal civil war pitted neighbor against neighbor based to a significant degree
on sectarian divisions
93
and was surpassed in its intensity only by its intractability. Lasting fifteen
years, the war led to over 100,000 deaths and more than 1 million displaced. Its consequences
reverberate even today, as gridlock within Lebanon’s post-civil war consociational political system
and a protracted civil war and refugee crisis in neighboring Syria have led to instability and renewed
cycles of violence. Though there were many contributors to the outbreak of war, I focus on the role
of the Lebanese Kataeb Party (henceforth the Kataeb) in the onset and intensity of the war, a clear
example of type 3 political violence. Due to the later involvement of external actors—namely, Syria
and Israel, which both ended the war in de facto possession of different parts of Lebanon—I focus
on the earliest period of the war, starting with the beginning of hostilities in 1975, up to and
including the Black Saturday Massacre in December of that year. Limiting the temporal scope as
such allows us to focus on the state of Kataeb in the political landscape at the time immediately
preceding the beginning of hostilities and the escalation of severity to the level of type 3 violence.
The post-independence arrangement set out with the 1943 National Pact first established a
power-sharing arrangement intended to lessen tensions and facilitate peaceful coexistence between
members of different religious groups. With this agreement, it was determined that the president
would be a Maronite Christian, the prime minister a Sunni Muslim, and the speaker of parliament a
93
In referring to sectarian divisions, I do not mean that all individuals from one religion sided with each other against
individuals from a different religion in perfect uniformity. There were certain groups within Lebanon, such as the Greek
Orthodox, who did not side with the Kataeb and its Christian right coalition. Indeed, one sometimes underemphasized
aspect of the war has been the divisions not only between some Christian sects, but within the Maronite Catholic
Church itself (Henley 2008, Tegho 2018). However, such intrareligious divisions are beyond the scope of this article.
125
Shi’i Muslim. The agreement did not have the intended effect, however, as changing demographics
and shifting power dynamics resulting in part from large scale arrival of Palestinian refugees as a
result of their expulsion from their homes during the 1948 Arab-Israeli War led to increasing
sectarian power struggles. Though there were certainly intragroup disagreements among Lebanese
Christians and Muslims alike, coalitions largely crystallized along religious lines, with Sunni Muslims
and Palestinians forming an alliance that the growing Christian right coalition saw as a threat to their
maintenance of power (Salibi 1985). It was against this backdrop that the Kataeb grew and
consolidated into an established identity-based political party.
Initially founded as a Maronite Christian youth paramilitary movement in 1936, the Lebanese
Kataeb/Phalange Party (Hizb al-Kata’ib al-Lubnaniyya)
94
originated as a party aimed at protecting
and advancing the interests of Lebanese Christians (Entelis 1973b). The party first gained political
power following the first civil war in 1958, when the party’s founder and then-leader Sheikh Pierre
Jumayyil was made a member of a post-war cabinet (Stoakes 1975). The Kataeb, which began
competing in elections as early as 1952, continued operating a military wing, and their political
power only increased following Jumayyil’s cabinet appointment in 1958 (ibid.). The Kataeb did not
outwardly claim to be a Christian/Maronite party, but it was one in all but name: at the time of the
start of the civil war in 1975, at least 85% of its support came from Maronite Christians (ibid.).
The Kataeb’s high levels of identity coherence and elite support were instrumental in leading
to the onset and severity of the second Lebanese Civil War, which falls squarely within type 3 of the
typology. Although the war cannot be attributed solely to the actions and program of the Kataeb, I
focus on this actor for the fundamental role it played in the first stage of the civil war, as well as for
its role in one of the events that arguably represented one of the first major hostilities of the war and
that effectively led to Beirut’s sectarian division and all-out warfare between sectarian militias
94
Arabic: اﻟﻠ ﺑ ﻧ ﺎ ﻧ ﯾ ﺔ ﺣز ب ا ﻟ ﻛ ﺗ ﺎ ﺋ ب
126
(Tamimova 2018). The Black Saturday Massacre (Beirut Bus Massacre) of December 6, 1975 began
when members of the Kataeb’s paramilitary wing
95
began executing Muslim civilians in retaliation
for the murder of a young Christian member of the unit (ibid.). Checkpoints were soon set up across
Beirut, leading to the identity-based killing of roughly 200 individuals and precipitating a nearly two-
year cycle of retaliatory ethno-sectarian violence (Hanf 1993, p. 210; Tamimova 2018, p. 135).
As a predominantly Christian movement with clearly delineated goals and ideological
agreement, the Kataeb as an actor (and its armed wing that ultimately perpetrated the violence) had
very high levels of identity coherence at the outset of hostilities. A right-of-center party, among its
core goals was the promotion of a non-Arab national identity. In practice, promoting these goals
manifested as a struggle with pan-Arab and Palestinian influence in Lebanon (Stoakes 1975; Smyth
2021). The Kataeb’s core goal squared clearly with some basic identity-based interests of much of
the Christian population living in a small, religiously diverse state situated in a Muslim- and Arab-
majority region. Contributing to the Kataeb’s appeal to Lebanese Christians was rising pan-Arabism
and Arab nationalism during what was considered the Arab Cold War
96
, as many Christians viewed
the rising appeal of these ideologies as threatening the political status which protected Christians’
economic and political power. The party’s motto, “God, Homeland, and Family”
97
, underscores the
clarity of its shared goals and ideological commitment (Entelis 1973b). The perceived threat of
increasing pan-Arab influence in the country surely contributed to the group’s popularity among
Maronite Christians as well as small numbers of non-Christians: as of 1970, fully 90% of party
members were Christian (80% Maronite, ten percent other—Greek Orthodox, Greek Catholic,
Protestant, Armenian Catholic, etc.), with the remaining members primarily identifying religiously as
95
Specifically, the Bachir Jumayyil (BG) Brigade.
96
This did not play a role until the breakup of the United Arab Republic in 1961.
97
Arabic: "ﷲ ، اﻟ و ط ن ، اﻟ ﻌ ﺎ ﺋ ﻠ ﺔ "
127
Shi’a, Jewish, and Druze
98
(Entelis 1973a, 1973b). Nor were members were not limited to one
geographic area. Prior to the beginning of the war, members could be found throughout Lebanon
(ibid.). The perception of a clear threat facilitated both continued recruitment, especially among
young Christians joining the paramilitary wing, and widespread support for the party. Even if not in
name, the Kataeb was, for all intents and purposes, a predominantly Maronite Christian party
leading up to and at the beginning of the war. It was not the only Christian party, to be sure, but it
was a leading member of a powerful coalition of Christian political parties known as the Lebanese
Front. (Smyth 2021, Stoakes 1975).
Accounting for much of the Kataeb’s high level of elite support was its dual function as an
active political party with an explicitly recognized armed militia wing. In addition to its status as a
high-profile political party with an aim of securing power for Maronite Christians—including
through political violence—it led a coalition of other Christian political parties in parliament which
did not condemn the Kataeb paramilitary’s use of violence to achieve its goals. This consolidation of
support not only within the Kataeb party but across different Christian parties contributed
significantly to its high level of elite support, allowing it to act with broad support from Christians
and without social sanction. Contrasted with the cases of the Ku Klux Klan and the Republican
party in the 1920s and the PIRA and Sinn Fein at the beginning of the troubles in the late 1960s, it is
clear that the Kataeb was able to pursue its goals with relative impunity in no small part due to the
strong political apparatus behind it and near-unanimous support from Lebanese Christians,
Maronite and non-Maronite alike. When coupled with the guarantee that a Maronite Christian would
hold the presidency under the rules of Lebanon’s consociational political system, the party’s electoral
success granted the movement access to state power. Given the movement’s high levels of identity
coherence and elite support, particularly within the factionalized context of 1975 Lebanon and
98
A very small number (less than one percent) of members were Sunni.
128
ongoing armed clashes between its paramilitary and the Palestine Liberation Organization (PLO),
that the security situation quickly deteriorated should come as no surprise.
Alternative Explanations
In the case of the three illustrative cases presented here, there are several potential alternative
explanations for the different outcomes of political violence severity. While I do not address them
all, it is important to acknowledge variation in state capacity as a possible alternative explanation for
the observed outcomes in each case. State capacity, generally defined by political scientists as the
ability of a central government to carry out basic functions such as the maintenance of law and
order, the collection of taxes, and the provision of public goods and services, is one potential
alternative explanation for the different levels of political violence seen across these cases. Indeed,
variations in state capacity between the US in the 1920s, the UK in the late 1960s and early 1970s,
and Lebanon in the 1970s cannot be ignored in comparisons of these three cases. Due to its later
development and the influence of external actors in the creation of the state, for example, Lebanon’s
institutions that would carry out these core state functions were relatively underdeveloped in
comparison with those of the US and the United Kingdom. This extensive external influence in
effect precluded Lebanon from undergoing its own internal state-making processes identified by
political scientists as crucial to the development of strong institutions and norms that facilitate high
state capacity in strong states (i.e., Spruyt 2009, Tilly 1985). The different dimensions of state
capacity are too numerous to consider here, so I focus on two specific elements that present
especially difficult tests of the theory proposed in this article: the existence of an effective central
coercive apparatus and a strong national identity.
An effective state coercive apparatus, which may include the military, law enforcement, and
intelligence bodies, facilitates the maintenance of law and order and can quickly and decisively
129
counter threats to the state’s monopoly of violence within the state. On this measure, Lebanon was
especially weak, as its armed forces and law enforcement bodies were often divided by sect and the
legitimacy of these forces was not uniform throughout different parts of the country. The coercive
capacity of the Lebanese state leading up to the onset of the first Lebanese Civil War was certainly
less than that of the US in the 1920s and of the British central government in Northern Ireland in
the late 1960s. Indeed, any direct confrontation between the KKK and US troops or security
services would likely have led to near immediate defeat of the former, and the weakness of the IRA
in comparison to the strength of British troops and the Police Service of Northern Ireland (PSNI)
could be found in the former’s favored tactics of bombing campaigns and guerrilla warfare tactics.
By contrast, the Lebanese Armed Forces (LAF) had been plagued by sectarian political struggles
almost immediately after independence in 1943. The aftermath of the first Lebanese civil war in
1958 saw a confessional power struggle that ultimately undermined the LAF’s position of power,
which also contributed to the end of the LAF’s military intelligence apparatus, the Deuxième Bureau
(Nerguizian 2018; Nerguizian 2015).
Regardless of how powerful (or weak) the state and its security forces were in maintaining
the status quo and preventing sectarian violence, the identity coherence of the Kataeb was
instrumental in determining recruitment and retention rates among both its paramilitary wing and its
political arm and in supplying the motivation for mobilization. The appeal of each group to its
combatants due to a strong commitment to the identity group and its goals would be unlikely to be
affected by the state’s level of coercive capacity. Though I do not claim that state coercive capacity
was unimportant in these different cases—certainly, the threat of detention or worse posed by a
strong coercive apparatus can reduce mobilizational capacity—I do claim that the importance of a
group’s identity coherence can be such that the motivation for armed mobilization can sometimes be
so potent that it eclipses the potential risks posed by opposition to a powerful state. Although
130
differences in state capacity may certainly have played a role in determining the severity of violent
outcomes, its influence and the causal importance of elite support and identity coherence among
armed actors are not mutually exclusive; rather, state capacity might be considered in tandem with
the key variables proposed by the theory, and future research would do well to more fully
incorporate such considerations in examinations of identity-based political violence.
Another potential indicator of state capacity is the presence of an inclusive national identity,
which was especially lacking in the Lebanese case. Though discussions of state capacity most often
evoke thoughts of a state’s monopoly over the legitimate use of force (Weber 1965), there are other
indicators of state capacity that are less obvious but no less important. Indeed, the presence of a
civic nationalism necessitates the existence of a central state, or a “state-centered politics.”
(Greenfeld and Eastwood 2009, p. 261) Despite its de facto designation as a Christian movement,
one of the Kataeb’s main aims was attempting to foster such an identity in pursuit of its goal of a
strong central state immune from the types of external influence unfavorable to its interests.
99
However, this was not an effective strategy, and ultimately the identity became a mostly sectarian
one. This underscores two points: first, the mobilizational power of ethnic and religious identity
despite the presence of counternarratives emphasizing a national identity is significant. Secondly, this
highlights that despite low state capacity in Lebanon, there were some initial efforts by the Kataeb to
establish a secular national identity, but it was just not as effective as mobilization around a sectarian
one. This helps to highlight the unique power of the theory in accounting for the draw of identity-
based movements, despite the existence of strong national identities or dedicated attempts to create
them.
99
Namely, influence from Muslim-majority states in the region and from pan-Arabist influence from the likes of Egypt’s
Nasser. At the same time as Kataeb eschewed the potential for influence exercised by such states, it welcomed
intervention by external forces, such as Israel and Francee, insofar as interests of those states were in line with those of
the Kataeb.
131
Chapter 3: Conclusion
This article advanced a novel theory and typology of identity-based political violence and
found mixed support for its initial hypotheses. The results of the tests based on novel measures of
identity coherence and elite support as predictors of conflict severity showed two things. First,
identity-based instances of political violence within the sample of events studied were more violent
than non-identity-based instances. Secondly, the core hypothesis that increases in elite support and
identity coherence among actors leads to more severe outcomes of political violence found strong
initial support from the data tested here. At the same time, the effect of the interaction between the
two key predictors indicates that caution should be taken in making concrete claims about this
relationship, particularly in regards to elite support. Despite this, tthese results underscore the
importance of systematically incorporating theoretically and empirically sound methods of analyzing
political violence with identity-based dimensions.
Nonetheless, the theory and evidence presented here generate many questions, providing a
foundation for future research. Though I advanced what I consider to be a theoretically sound
method of measuring and testing identity coherence and elite support, this is by no means the only
possible method. Future research might further examine subsets of identity-based political violence,
such as identity-based rebel movements and identity-based political parties. They also might further
differentiate outcomes within the three types or examine the causal mechanisms that lead to changes
in levels of elite support and identity coherence. Additionally, it would be useful to more closely
examine the relationship between elite support and identity coherence. For example, can these two
things be mutually reinforcing, and if so, what are the conditions and mechanisms that lead to this?
What is the relationship between these changes and increasing polarization? These are just some of
the research questions that this theory and typology suggest for further study. More systematically
considering the correlates, causes, and effects of identity-based political violence is paramount to
132
understanding how to identify where the potential for such violence exists, as well as how to prevent
its escalation. The theory and typology provided here, as well as the promising initial quantitative
results and qualitative evidence available across a variety of cases, are an important first step in
contributing to answering such questions.
133
CONCLUSION
The theory and evidence presented in this dissertation were intended for two main purposes.
First, the evidence demonstrated that considering identity salience at the microlevel and at the
macrolevel can be useful in understanding the precursors to and causal dynamics of identity-based
political violence. Second, the work presented in the previous pages should be considered a
foundation upon which future research agendas related to identity salience and its role in various
political phenomena, but especially political violence, can be built. This brief conclusion summarizes
the main findings of the dissertation and highlights some questions for future research.
The overarching takeaway of the dissertation has been that identity, and especially identity
salience, should be considered as a central explanatory variable in certain types of political violence.
Though it is not relevant in or central to all instances of political violence, the rise of nationalism
and identity-based factionalism in the post-Cold War period necessitates an approach that allows for
the systematic incorporation of identity as a variable in studies of identity-based political violence.
Identity is important at the microlevel in motivating violent political attitudes, as demonstrated in
article 2, yet it also can be operationalized as an important variable in macrolevel studies of the
causes of political violence and the severity of its outcomes, as demonstrated in article 3.
Article one provided a conceptual framework and novel measurement strategy for identity
salience as a latent variable, using a theoretically informed survey to collect responses from
individuals on key indicators of identity salience. These data were then applied to a measurement
model, the output of which was latent trait scores measuring the identity salience of each individual
respondent.
Article 2 applied the conceptualization and measurement of identity salience as a variable to
a study of the relationship between identity salience as an independent variable and individual-level
134
attitudes toward political violence. This article developed a microlevel theory of identity salience and
political violence, hypothesizing that both higher baseline levels of identity salience and priming an
individual’s identity salience via a threatening experimental manipulation would result in more
permissive attitudes toward political violence. Using the latent trait scores computed in article one,
as well as an experimental manipulation meant to increase the national identity salience of
respondents, as key independent variables, it tested the relationship via a survey experiment
conducted via Amazon MTurk on a convenience sample of American adults. The results indicated
mixed support for the tested hypotheses: though higher baseline identity salience scores were highly
predictive of more permissive expressed attitudes toward political violence carried out on behalf of
that identity, the experimental treatment appeared to have no effect. While this article demonstrated
a strong relationship between identity salience as I conceive of it and attitudes towards political
violence, a key shortcoming was an ineffective treatment.
Article 3 examined the role of identity in political violence outcomes at the macrolevel and
proposed a theory and typology outlining this relationship. Using novel variables measuring identity
coherence and elite support as predictors, it tested the relationship between these variables and the
severity of political violence outcomes in instances of identity-based political violence. The results
showed strong initial support for the theory: identity-based episodes of political violence were
significantly more violent than were non-identity based episodes of political violence, and there was
a strong and statistically significant relationship between identity coherence and elite support on the
one hand and the severity of violence on the other. However, the relationship between elite support
and severity of outcomes was inconsistent and appeared to be conditional on the inclusion of
covariates like regime type and group size.
There are several different directions that future research might take. Though I do not
address them all here, I highlight a few. First, identity salience in this dissertation is measured as a
135
static variable—future studies of identity salience might look at change over time, given the
availability of panel data. Second, the results of the survey experiment in article two highlighted the
difficulties of designing an effective experimental manipulation of identity salience, but also showed
that there were significant differences in how people reacted to these primes based on partisan
identification. Future research might consider how individuals conceive of the same identity
differently by testing different manipulations to better understand these differences. Finally, the
relationship between identity coherence and elite support is an interesting one worth further study,
especially in the context of rising polarization. If a group’s identity coherence increases in tandem
with its level of elite support, might this exacerbate polarization when it occurs on both sides? These
are just a few of the questions raised by this work that future research would do well to consider.
136
REFERENCES
“22J Elecciones Generales 1986.” Público, Público, https://www.publico.es/resultados
elecciones/generales/1986/.
Abdelal, R., Herrera, Y., Johnston, A.I. & McDermott, R. (Eds.) (2009). Measuring
Identity: A Guide for Social Science Research. New York: Cambridge University Press.
Abdelal, R., Herrera, Y., Johnston, A.I. & McDermott, R. (Eds.) 2009. “Identity as a Variable.” In
Measuring Identity, ed. R. Abdelal, YM Herrera, AI Johnston, R. McDermott, pp. 17-32. New York:
Cambridge UP.
Adam-Troian, Jais, Eric Bonetto, Marcos Araujo, Ouissam Baidada, Elif Celebi, Marcus Dono
Martin, Fade Eadeh, Amelie Godefroidt, Slieman Halabi, Yara Mahfud, Florent Varet, and Taylan
Yurtbakan. 2020. “Positive associations between anomia and intentions to engage in political
violence: Cross-cultural evidence from four countries,” Peace and Conflict: Journal of Peace Psychology,
26(2): 217–223.
Ahler, Douglas J., Carolyn E. Roush, and Gaurav Sood. 2019. “The Micro-Task Market for Lemons:
Data Quality on Amazon’s Mechanical Turk,” Political Science Research and Methods: 1-20.
Ahmad, Aisha. 2016. Going Global: Islamist Competition in Contemporary Civil Wars. Security
Studies, 25(2): 353-384.
Albertson, Bethany, and Shana Kushner Gadarian. 2015. Anxious Politics: Democratic Citizenship in a
Threatening World. Cambridge UP.
“American Identity Movement (AIM).” Anti-Defamation League.
https://www.adl.org/education/references/hate-symbols/american-identity-movement-aim.
Amman, Molly, and J. Reid Meloy. 2021. “Stochastic Terrorism: A Linguistic and Psychological
Analysis.” Perspectives on Terrorism 15(5): 1-13.
Anderson, Benedict. 1983. Imagined Communities: Reflections on the Origin and Spread of Nationalism. New
York: Verso Press.
APA Dictionary of Psychology. https://dictionary.apa.org/cognitive-process
Arendt, Hannah. 1970. On Violence. New York: Harcourt, Brace, and World.
Baker, Melissa. “Dispositional Traits and Susceptibility to Political Anxiety.” PsyArXiv, 6 Nov. 2020.
Web.
Barbera, Pablo. 2015. “Birds of the Same Feather Tweet Together. Bayesian Ideal Point Estimation
Using Twitter Data.” Political Analysis 23(1): 76-91.
137
Bartels, Larry M. 2014. “Remembering to Forget: A Note on the Duration of Campaign. Advertising
Effects,” Political Communication 31(4): 532-544.
“Basque Fatherland and Liberty (ETA) (Spain, separatists, Euskadi ta Askatasuna),” November
2008. Council on Foreign Relations, https://www.cfr.org/backgrounder/basque-fatherland-and-
liberty-eta-spain-separatists-euskadi-ta-askatasuna.
Beauchamp, Zack. “How does this end?” Vox, 3 January 2022. https://www.vox.com/policy-and-
politics/22814025/democracy-trump-january-6-capitol-riot-election-violence
Bennett, Andrew. 2013. “Causal mechanisms and typological theories in the study of civil conflict,”
In J. Checkel (Ed.), Transnational Dynamics of Civil War (pp. 205-230). Cambridge: Cambridge UP.
Benoit K, Watanabe K, Wang H, Nulty P, Obeng A, Müller S, Matsuo A (2018). “quanteda: An R
package for the quantitative analysis of textual data.” Journal of Open Source Software, 3(30), 774.
doi: 10.21105/joss.00774, https://quanteda.io.
Berinsky, Adam J., Gregory A. Huber, and Gabriel S. Lenz. “Evaluating Online Labor Markets for
Experimental Research: Amazon.com’s Mechanical Turk,” Political Analysis 20(3): 351-368, Summer
2012.
Berinsky, Adam J., Michele F. Margolis, Michael W. Sances. “Separating the Shirkers from the
Workers? Making Sure Respondents Pay Attention on Self-Adminstered Surveys,” American Journal of
Political Science 58(3): 739-753, 2012.
Biden, Joseph R. “Inaugural Address by President Joseph R. Biden, Jr.” 20 January 2021.
https://www.whitehouse.gov/briefing-room/speeches-remarks/2021/01/20/inaugural-address-by-
president-joseph-r-biden-jr/
Blaydes, Lisa, and Drew A. Linzer. 2008. “The Political Economy of Women’s Support for
Fundamentalist Islam.” World Politics 60 (July): 576-609.
Bart Bonikowski, Yuval Feinstein, and Sean Bock. 2021. “The Partisan Sorting of ‘America’: How
Nationalist Cleavages Shaped the 2016 US Presidential Election.” American Journal of Sociology,
127(2).
Brady, Henry E., and Cynthia Kaplan. 2000. “Categorically Wrong? Nominal versus Graded
Measures of Ethnic Identity.” Studies in Comparative International Development 25, no. 3 (Fall): 56-91.
Brady, Henry E. and Cynthia Kaplan. 2009. “Conceptualizing and Measuring Ethnic Identity,” in
Measuring Identity: A Guide for Social Scientists, eds. Abdelal, Rawi, et al. Cambridge UP: 33-71.
Branscombe, Nyla R., Naomi Ellemers, Russell Spears, and Bertjan Doosje. 1999a. “The context
and content of social identity threat.” In Social Identity: Context, Commitment, Content, ed. Naomi
Ellemers, Russell Spears, Bertjan Doosje, pp. 35–58. Oxford, UK: Blackwell.
Branscombe, Nyla R., Michael T. Schmitt, and Richard D. Harvey. 1999b. “Perceiving pervasive
discrimination among African Americans: implications for group identification and well-
being,” Journal of Personality and Social Psychology 77(1):135–49.
138
Brewer, Marilynn B. 2001. “The Many Faces of Social Identity: Implications for Political
Psychology.” Political Psychology 22, no. 1: 115-125.
Brubaker, Rogers, and Frederick Cooper. 2000. “Beyond ‘Identity.’” Theory and Society 29, no. 1
(February): 1-47.
Callero, Peter L. 1985. “Role-Identity Salience.” Social Psychology Quarterly 48(3): 203-215.
Canetti, Daphna, Brian J. Hall, Carmit Rapaport, and Carly Wayne. 2013. “Exposure to Political
Violence and Political Extremism,” European Psychologist 18: 263-272.
Canetti, Daphna, Ibrahim Khatib, Aviad Rubin, Carly Wayne. 2019. “Framing and fighting: The
impact of conflict frames on political attitudes,” Journal of Peace Research 56(6): 737-752.
Cederman, Lars-Erik, Andreas Wimmer, and Brian Min. 2010. “Why do Ethnic Groups Rebel? New
Data and Analysis,” World Politics 62(1): 87-119.
Chalmers, David. 1965. “The Ku Klux Klan in Politics in the 1920s,” Mississippi Quarterly 18(4): 234-
247.
Chang, LinChiat, and Jon A. Krosnick. “National Surveys via RDD Telephone Interviewing versus
the Internet: Comparing Sample Representativeness and Response Quality,” Public Opinion Quarterly
73: 641-78, 2002.
Charnysh, Volha, Christopher Lucas, and Prerna Singh. 2015. “The Ties That Bind: National
Identity Salience and Pro-Social Behavior Toward the Ethnic Other.” Comparative Political Studies
48(3): 267-300.
Chenoweth, Erica, and Maria J. Stephan. 2011. Why Civil Resistance Works: The Strategic Logic of
Nonviolent Conflict. New York: Columbia UP.
Cheryan S, Monin B. 2005. ‘“Where are you really from?”: Asian Americans and identity denial.’
Journal of Personality and Social Psychology, 89:717–30.
Collier, Paul and Anke Hoeffler. 2004. “Greed and Grievance in Civil War,” Oxford Economic Papers
56(4).
Collier, David, Jody LaPorte, and Jason Seawright. 2012. “Putting Typologies to Work: Concept
Formation, Measurement, and Analytic Rigor.” Political Research Quarterly 65(1): 217-232.
Crenshaw, Kimberlé. 1991. “Mapping the Margins: Intersectionality, Identity Politics, and Violence
Against Women of Color,” Stanford Law Review 43: 1241-99. Women of Color at the
Center: Selections from the Third National Conference on Women of Color and the Law.
Curtis, S. McKay. 2010. “BUGS Code for Item Response Theory,” Journal of Statistical Software 36(1):
1-34.
139
Dallago, Francesca, and Michele Roccato. 2009. “Right-Wing Authoritarianism, Big Five and
Perceived Threat to Safety,” European Journal of Personality 24(2): 106-122.
Davies, Shawn, Therese Pettersson & Magnus Öberg (2022). Organized violence 1989-2021 and
drone warfare. Journal of Peace Research 59(4).
Dayal, Anjali, Alexandra Stark, and Megan A. Stewart. “Warnings of ‘Civil War’ Risk Harming
Efforts Against Political Violence.” War on the Rocks, 18 January 2022.
https://warontherocks.com/2022/01/warnings-of-civil-war-risk-harming-efforts-against-political-
violence/
Dawson, Michael C. 1994. Behind the Mule-Race and Class in African-American Politics. Princeton, NJ:
Princeton University Press.
Devos T, Banaji MR. 2005. “American = white?” Journal of Personality and Social Psychology 88: 447–66.
Dovidio, John, and William N. Morris. 1975. “Effects of stress and commonality of fate on helping
behavior,” Journal of Personality and Social Psychology 31(1):145–49.
Druckman, James and Cindy D. Kam. 2011. “Students as Experimental Participants: A Defense of
the ‘Narrow Data Base,’” in James N. Druckman, Donald P. Green, James H. Kuklinkski, and
Arthur Lupia, eds., Cambridge Handbook of Experimental Political Science, New York: Cambridge UP,
2011.
Egan, Patrick J. 2019. “Identity as Dependent Variable: How Americans Shift Their Identities To
Align With Their Politics.” American Journal of Political Science 64(3): 699-716.
Ellingsen, Tanja. 2000. “Colorful Community or Ethnic Witches’ Brew? Multiethnicity and
Domestic Conflict during and after the Cold War,” Journal of Conflict Resolution 44(2): 228-249.
Entelis, John P. 1973a. “Party Transformation in Lebanon: Al-Kata’ib as a Case Study,” Middle
Eastern Studies 9(3): 325-340.
Entelis, John P. 1973b. “Structural Change and Organizational Development in the Lebanesee
Kata’ib Party,” Middle East Journal 27(1): 21-35.
Fan, Yingjie, Jennifer Pan, Zijie Shao, and Yiqing Xu. 2020. “How Discrimination Increases Chinese
Overseas Students' Support for Authoritarian Rule.” 21st Century China Center Research Paper No.
2020-05, Available at
SSRN: https://ssrn.com/abstract=3637710 or http://dx.doi.org/10.2139/ssrn.3637710.
Fariss, Christopher J. 2018a. “Are Things Really Getting Better?: How to Validate Latent Variable
Models of Human Rights.” British Journal of Political Science 48(1): 275-282.
Fariss, Christopher J. 2018b. “The Changing Standard of Accountability and the Positive
Relationship between Human Rights Treaty Ratification and Compliance.” British Journal of Political
Science 48(1): 239-272.
140
Fariss, Christopher J., Kevin Reuning, and Michael R. Kenwick. 2020. “Measurement Models,” in
SAGE Handbook of Reseearch Methods in Political Science and Internationl Relations, edited by Luigi Curini
and Robert J. Franzese, Jr. SAGE Press.
Fearon, James. 1995. “Rationalist Explanations for War,” International Organization 49(3): 379-414.
Fearon, James and David Laitin. 2003. “Ethnicity, Insurgency, and Civil War,” American Political
Science Review 97(1): pp. 75-90.
Feldman, S., L. Huddy, and E. Cassese. 2012. “Emotions, threat, and political reasoning.” In Ron
Sun (ed.), Grounding Social Sciences in Cognition (pp. 127-156). Cambridge, MA: MIT Press.
Ferguson, Neil, and James W. McAuley. 2019. “Radicalization or Reaction: Understanding
Engagement in Violent Extremism in Northern Ireland,” Political Psychology 41(2): 215-230.
Finkel, Steven E., John F. McCauley, Michael Neureiter, and Chris A. Belasco. 2021. “Community
Violence and Support for Violent Extremism: Evidence from the Sahel,” Political Psychology 42(1):
143-161.
Finn, Daniel. 2019. One Man’s Terrorist: A Political History of the IRA. New York: Verso.
Fischer, P., A.S. Haslam, and L. Smith. 2010. “If you wrong us, shall we not revenge? Social identity
salience moderates support for retaliation in response to collective threat,” Group Dynamics: Theory,
Research, and Practice 14: 143-150.
Flippen, Annette R., Harvey A. Hornstein, William E. Siegal, and Eben A. Weitzman. 1996. “A
comparison of similarity and interdependence as triggers for in-group formation,” Personality and
Social Psychology Bulletin 22(9): 882–93.
Frable, Deborrah E. S. 1997. “Gender, Racial, Ethnic, Sexual, and Class Identities.” Annual Review of
Psychology 48(1): 139–162.
Furia, Peter. 2002. “Patriotism, Nationalism, and International Politics.” Paper presented at the
annual meeting of the American Political Science Association, Boston, September.
Gadarian, Shana Kushner. 2014. “Scary Pictures: How Terrorism Imagery Affects Voter
Evaluations,” Political Communication 31(2): 282-302.
Gade, Emily Kalah. 2020. “Social Isolation and Repertoires of Resistance,” American Political Science
Review 114(2): 309-325.
Gagnon, V.P. 2004. The Myth of Ethnic War: Serbia and Croatia in the 1990s. Itchaca, NY: Cornell UP.
Gerber, Alan S. and Donald P. Green. 2012. Field Experiments: Design, Analysis, and Interpretation. New
York: W.W. Norton & Company.
141
Goldstone, Jack A., Robert H. Bates, David L. Epstein, Ted Robert Gurr, Michael B. Lustik, Monty
G. Marshall, Jay Ulfelder, and Mark Woodward. 2010. “A Global Model for Forecasting Instability,”
American Journal of Political Science 54(1): 190-208.
Greenfeld, Liah, and Jonathan Eastwood. 2009. “National Identity,” in The Oxford Handbook of
Comparative Politics,” Oxford: Oxford UP: 256-273.
“Groyper Army and ‘America First.’” Anti-Defamation League.
https://www.adl.org/resources/backgrounders/groyper-army-and-america-first
Gubler, Joshua R. and Joel Sawat Selway. 2012. “Horizontal Inequality, Crosscutting Cleavages, and
Civil War,” Journal of Conflict Resolution 56: 206-232.
Hanf, Theodor. 1993. Coexistence in Wartime Lebanon: Decline of a State and Rise of a Nation. New York:
IB Tauris.
Hassner, Ron E. 2009. War on Sacred Grounds. Ithaca, NY: Cornell UP.
Hayden, Shelly R., Thomas T. Jackson, and Joe Guydish. 1984. “Helping behavior of females:
effects of stress and commonality of fate,” The Journal of Psychology: Interdisciplinary and Applied
117(2):233–237.
Henley, Alexander D.M. 2008. “Politics of a Church at War: Maronite Catholicism in the Lebanese
Civil War,” Mediterranean Politics 13(3): 353-369.
Hill, Seth J., James Lo, Lynn Vavreck, and John Zaller. 2013. “How quickly we forget: The duration
of persuasion effects from mass communication,” Political Communication 30: 521-547.
Hoffman, Michael and Elizabeth Nugent, “Communal Religious Practice and Support for Armed
Parties: Evidence from Lebanon,” Journal of Conflict Resolution 61(4), (2017): pp. 869-902.
Hogg, Michael A., and Jane Sunderland. 1991. “Self-esteem and Intergroup Discrimination in the
Minimal Group Paradigm.” British Journal of Social Psychology 30(1): 51–62.
Hopkins, Stephen. 2010. “Sinn Féin and the Radical Left in Europe: Natural Partners or Awkward
Allies?” Paper presented at the Political Studies Association Conference, University of Edinburgh.
Horowitz, Donald L. 2001. The Deadly Ethnic Riot. Berkeley: University of California Press.
Horowitz, Donald L. 2004. Ethnic Groups in Conflict: Theories, Patterns, and Policies. Berkeley: University
of California Press.
Horowitz, Michael (2009). Long time going: Religion and the duration of crusading. International
Security 34(2): 162–193.
Huddy, Leonie. 2001. “From Social to Political Identity: A Critical Examination of Social Identity
Theory.” Political Psychology 22, no. 1: 127-156.
142
Huddy, Leonie. 2013. “From Group Identity to Political Cohesion and Commitment,” in The Oxford
Handbook of Political Psychology, Second Edition, eds. Leonie Huddy, David O. Sears, and Jack S. Levy.
Oxford UP: pp. 749-773.
Huddy, Leonie, Stanley Feldman, Charles Taber, and Gallya Lahav. 2005. “Threat, anxiety, and
support of antiterrorism policies,” American Journal of Political Science 49(3):593–608.
Humphreys, Macartan and Jeremy Weinstein. 2008. “Who Fights? The Determinants of
Participation in Civil War, American Journal of Political Science, 52(2): pp. 436– 455.
“Identity Evropa.” Anti-Defamation League. https://www.adl.org/education/references/hate-
symbols/identity-evropa.
Jacobs, Carly M. and Elizabeth Theiss-Morse. 2013. ‘Belonging in a “Christian nation”: The explicit
and implicit associations between religion and national group membership.’ Politics and Religion 6:
373–401.
Jahnke, Sara, Katharina Abad Borger, and Andreas Beelmann. 2022. “Predictors of Political
Violence Outcomes among Young People: A Systematic Review and Meta-Analysis, Political
Psychology 43(1): 111-129.
Juergensmeyer, Mark, 2000. Terror in the Mind of God: The Global Rise of Religious Violence. University of
California Press.
Kalmoe, Nathan P. 2014. “Fueling the Fire: Violent Metaphors, Trait Aggression, and Support for
Political Violence,” Political Communication 31(4): 545-563.
Kalyvas, Stathis. 2006. The Logic of Violence in Civil Wars.” New York: Cambridge UP.
Kalyvas, Stathis. 2019. “The Landscape of Political Violence,” in The Oxford Handbook of Terrorism,
eds. Chenoweth, Erica et al. Oxford UP: pp. 12-28.
Kalyvas, Stathis, and Paul D. Kenny. 2010. “Civil Wars.” Oxford Research Encyclopedia of International
Studies.
Keats, Jonathon. January 21, 2019. “Jargon Watch: The Rising Danger of Stochastic Terrorism,”
Wired.
Keefe, Patrick Radden. 2019. Say Nothing: A True Story of Murder and Memory in Northern Ireland. New
York: Anchor Books.
Kinder, Donald R., and David O. Sears. 1981. “Prejudice and Politics: Symbolic Racism Versus
Racial Threats to the Good Life.” Journal of Personality and Social Psychology 40(3): 414-431.
Kleinfeld, Rachel. 2021. “The Rise of Political Violence in the United States.” Journal of
Democracy 32(4): 160–76.
143
Kramer, Roderick M., Pamela Pommerenke, and Elizabeth Newton. 1993. “The Social Context of
Negotiation: Effects of Social Identity and Interpersonal Accountability on Negotiator Decision
Making.” Journal of Conflict Resolution 37, no. 4 (December): 633-654.
Landau-Wells, Marika. 2018. “Old Solutions to New Problems: An Introduction to Threat-Heuristic
Theory.” Unpublished Manuscript.
Lee, Taeku. 2008. “Race, Immigration, and the identity-to-politics link.” Annual Review of Political
Science 11: 457-78.
Levine, Robert A, and Donald T. Campbell. 1972. Ethnocentrism: Theories of Conflict, Ethnic Attitudes,
and Group Behaviors. New York, NY: John Wiley & Sons, Inc.
Licklider, Roy. 1995. “The Consequences of Negotiated Settlements in Civil Wars, 1945–
1993,” American Political Science Review, 89(3): 681-690.
Manekin, Devorah, and Tamar Mitts. February 2022. “Effective for Whom? Ethnic Identity and
Nonviolent Resistance,” American Political Science Review 116(1): 161-180.
Mapping Militant Organizations. “Proud Boys.” Stanford University. Last modified January 2022.
https://cisac.fsi.stanford.edu/mappingmilitants/profiles/proud-boys
Marche, Stephen. “Why the U.S. Military Isn’t Ready for Civil War.” Foreign Policy, 4 January 2022.
https://foreignpolicy.com/2022/01/04/american-civil-war-january-6-capitol/
Marshall, Monty G., and Ted Robert Gurr. 2020. “Polity 5: Political Regime Characteristics and
Transitions, 1800-2018.” Center For Systemic Peace. www.systemicpeace.org.
Martin, Andrew D., Kevin M. Quinn, and Jong Hee Park. 2011. “MCMCpack: Markov Chain Monte
Carlo in R,” Journal of Statistical Software 42(9): 1-21.
Mason, Lilliana. 2012. “The rise of uncivil agreement: issue versus behavioral polarization in the
American electorate,” American Behavioral Scientist 57: 140-159.
Mason, Lilliana. 2018. Uncivil Agreement: How Politics Became Our Identity. Chicago: University of
Chicago Press.
Mason, Lilliana, and Nathan Kalmoe. 2022. Radical American Partisanship: Mapping Violent Hostility, Its
Causes, and the Consequences for Democracy. Chicago: University of Chicago Press.
McAndrew, Tara. January 2017. “The History of the KKK in American Politics,” JSTOR Daily.
URL: https://daily.jstor.org/history-kkk-american-politics/.
McCall, George H. and J.T. Simmons. 1978. Identities and Interaction. Revised edition. New York: Free
Press.
McClain, Paula D., Jessica D. Johnson Carew, Eugene Walton, Jr., and Candis S. Watts. 2009.
“Group Membership, Group Identity, and Group Consciousness: Measures of Racial Identity in
American Politics?” Annual Review of Political Science 12: 471-485.
144
McDermott, Rose. 2004. “Psychological Approaches to Identity: Definition, Measurement, and
Experimentation.” Paper prepared for the Harvard Identity Project Conference, Cambridge, MA, 9-
11 December.
McDermott, Rose. 2009. “Psychological Approaches to Identity: Experimentation and Application,”
in Measuring Identity: A Guide for Social Scientists, eds. Abdelal, Rawi et al. Cambridge UP: pp. 345-367.
McDermott, Rose, and Hatemi PK. 2017. “The Relationship Between Physical Aggression, Foreign
Policy, and Moral Choices: Phenotypic and Genetic Findings. Aggressive Behavior, 43(1): 37-46.
McWhirter, Cameron. 2012. Red Summer: The Summer of 1919 and the Awakening of Black America. First
St. Martin's Griffin.
McVeigh, Rory. 2009. The Rise of the Ku Klux Klan: Right-Wing Movements and National Politics.
Minneapolis: University of Minnesota Press.
Melaugh, Martin, and Fionnuala McKenna. “Results of Elections Held in Northern Ireland Since
1968.” Ulster University: Conflict Archive on the Internet (CAIN).
Mendelberg, T. (2001). The race card: Campaign strategy, implicit messages, and the norm of equality. Princeton
University Press.
Merolla, Jennifer L., and Elizabeth J. Zechmeister. Democracy at Risk: How Terrorist Threats Affect the
Public. University of Chicago Press, 2009.
Miller, Arthur H., Patricia Gurin, Gerald Gurin, and Oksana Malanchuk. 1981. “Group Con-
sciousness and Political Participation.” American Journal of Political Science 25(3): 494–511.
Moloney, Ed. 2002. A Secret History of the IRA. New York: Penguin Press.
Mullinix, Kevin J., Thomas J. Leeper, James N. Druckman and Jeremy Freese. “The Generalizability
of Survey Experiments,” Journal of Experimental Political Science 2(2): 109-138, December 2015.
Musgrove, L., and C. McGarty. 2008. “Opinion-based group memberships as a predictor of
collective emotional responses and support for pro- and anti-war action,” Social Psychology 39(1): 37-
47.
Myrdal, G. 1944. An American Dilemma. New York: McGraw-Hill.
Nair, Gautam, and Nicholas Sambanis. 2019. “Violence Exposure and Ethnic Identification:
Evidence from Kashmir,” International Organization 73(2): 329-363.
Nerguizian, Aram. 2015. “Between Sectarianism and Military Development: The Paradox of the
Lebanese Armed Forces,” in eds., Bassel F. Salloukh et al, The Politics of Sectarianism in Postwar
Lebanon. London: Pluto Press.
145
Nerguizian, Aram. 2018. “The Lebanese Armed Forces and Hezbollah: Military Dualism in Post-
War Lebanon,” Malcolm H. Kerr Carnegie Middle East Center. URL: https://carnegie-
mec.org/2018/10/30/lebanese-armed-forces-and-hezbollah-military-dualism-in-post-war-lebanon-
pub-77598.
Oppenheimer, Daniel M., Tom Meyvis, and Nicolas Davidenko. “Instructional Manipulation
Checks: Detecting Satisficing to Increase Statistical Power.” Journal of Experimental Social
Psychology 45: 867–72, 2009.
Pan, Jennifer and Yiqing Xu. 2018. “China’s Ideological Spectrum.” Journal of Politics 80(1): 254-273.
Pemstein, Daniel, Stephen A. Meserve, and James Melton. 2010. “Democratic Compromise: A
Latent Variable Analysis of Ten Measures of Regime Type.” Political Analysis 18(4): 426-449.
“Proud Boys.” Southern Poverty Law Center. https://www.splcenter.org/fighting-hate/extremist-
files/group/proud-boys
Pape, Robert. January 2022. “American Face of Insurrection.” Report by the Chicago Project on
Security and Threats (CPOST), University of Chicago.
Petersen, Roger D. 2011. Western Intervention in the Balkans: The Strategic Use of Emotion in Conflict.
Cambridge UP.
Pop-Eleches, Grigor, and Graeme Robertson. 2018. “Identity and political preferences in Ukraine—
before and after the Euromaidan,” Post-Soviet Affairs 34(2-3): 107-118.
Posner, Daniel. 2004. “The Political Salience of Cultural Difference: Why Chewas and Tumbukas
are Allies in Zambia and Adversaries in Malawi,” American Political Science Review 98(4): 529-545.
Rafter, Kevin. 2005. Sinn Féin 1905-2005: In the Shadow of Gunmen. Dublin: Gill and Macmillan.
“Red Summer: The Race Riots of 1919,” The National WWI Museum and Memorial. URL:
https://www.theworldwar.org/learn/about-wwi/red-summer.
Roccas, S., and M.B. Brewer. 2002. “Social identity complexity,” Personality and Social Psychology Review
6(2): 88-106.
Rydell, R.J., D.J. Mackie, A.T. Maitner, H.M. Claypool, M.J. Ryan, and E.R. Smith. 2008. “Arousal,
processing, and risk-taking: Consequences of intergroup anger,” Personality and Social Psychology Bulletin
34(8): 1141-1152.
Salibi, Kamal. 1985. “East Side Story/West Side Story: Contrasting Lebanese Views of the Civil
War,” in CCAS Reports. Washington, DC: Georgetown University.
Sambanis, Nicholas. 2001. “Do Ethnic and Nonethnic Civil Wars Have the Same Causes? A
Theoretical and Empirical Inquiry (Part 1),” Journal of Conflict Resolution 45(3): 259-282.
146
Sambanis, Nicholas, and Michael Kalin. 2018. “How Should We Think About Social Identity?”
Annual Review of Political Science, 21: 239-257.
Sambanis, Nicholas, Jonah Schulhofer-Wohl, and Moses Shayo. 2012. “Parochialism as a central
challenge in counterinsurgency.” Science 336: 805–8.
Sambanis, Nicholas, and Moses Shayo. “Social Identification and Ethnic Conflict”, American Political
Science Review 107 (2): 294-325 (2013).
Samejina, Fumiko. 1969. “Estimation of Latent Ability Using a Response Pattern of Graded
Scores,” Psychometrika Monograph, No. 17.
Sarkees, Meredith Reid and Frank Wayman (2010). “Resort to War: 1816 - 2007.” Washington
DC: CQ Press.
Schmid, K., N. Tausch, M. Hewstone, J. Hughes, and E. Cairns. 2008. “The effects of living in
segregated vs. mixed areas in Northern Ireland: A simultaneous analysis of contact and threat effects
in the context of micro-level neighborhoods,” International Journal of Conflict and Violence 2, 56-71.
Sherif, Muzafer, O.J. Harvey, B. Jack White, William R. Hood, and Carolyn W. Sherif. 1961. The
Robbers Cave Experiment: Intergroup Conflict and Cooperation. Oklahoma: Institute of Group Relations,
University of Oklahoma.
Schildkraut, Deborah J. 2014. “Boundaries of American Identity: Evolving Understandings of “Us,”
Annual Review of Political Science 17: 441-460.
Schmuck, Desiree, Raffael Heiss, Jorg Matthes. 2020. “Drifting Further Apart? How Exposure to
Media Portrayals of Muslims Affects Attitude Polarization,” Political Psychology 41(6): 1055-1072.
Sen, Maya, and Omar Wasow. 2016. “Race as a Bundle of Sticks: Designs that Estimate Effects of
Seemingly Immutable Characteristics,” Annual Review of Political Science 11: 457-478.
Shadish, William R., Thomas D. Cook and Donald T. Campbell. 2002. Experimental and Quasi-
Experimnental Designs for Generalized Causal Inference. New York: Houghton Mifflin Company.
Sibley, Chris G., and John Duckitt. 2009. “Big-Five Personality, Social Worldviews, and Ideological
Attitudes: Further Tests of a Dual Process Cognitive-Motivational Model,” The Journal of Social
Psychology 149(5): 545-561.
Sides, John, and Lynn Vavreck. 2013. The Gamble: Choice and Chance in the 2012 Election. Princeton,
NJ: Princeton UP.
Sinn Féin Bunreacht agus Rialacha (Constitution and Rules), 2005. Housed in Northern Ireland
Political Collection. Belfast: Linen Hall Library.
“Sinn Féin Vice-President Gerry Adams,” November 27, 1982. Irish People.
147
“Sinn Féin Westminster Election Manifesto 1997: Building a Dynamic for Change.” 1997. Dublin:
Sinn Féin.
Smith, Tom W., and L. Jarkko. 2001. National Pride in Cross-National Perspective. Chicago: University of
Chicago, National Opinion Research Center.
Smyth, Phillip. 2021. “A New Kataeb.” Middle East Political and Economic Institute. URL:
https://mepei.com/a-new-kataeb/.
“Social Democratic and Labour Party Manifesto: Speak with Strength.” 1975. Great Britain.
Northern Ireland Office (NIO). Northern Ireland Constitutional Convention Report, (20 November
1975). London: HMSO.
Spruyt, Hendrik. 2009. “War, Trade, and State Formation,” in The Oxford Handbook of Comparative
Politics, eds. Carles Boix and Susan C. Stokes. Oxford: Oxford UP: 211-235.
Spry, Amber D. 2021. “Experiments in Identity,” in Advances in Experimental Political Science. James
Druckman and Donald Green (eds.), pp. 431-446. Cambridge: Cambridge UP.
Stephan, Walter G., Kurt A. Boniecki, Oscar Ybarra, Ann Bettencourt, Kelly S. Ervin, Linda A.
Jackson, Penny S. McNatt, and C. Lausanne Renfro. 2002. “The Role of Threats in the Racial
Attitudes of Blacks and Whites,” Personality and Social Psychology Bulletin 28 (9): 1242–1254.
Stephan, Walter G., and Cookie White Stephan. 2000. “An Integrated Theory of Prejudice.” In
Reducing Prejudice and Discrimination, edited by Stuart Oskamp, 23–46. Mahwah, N.J.: Lawrence
Erlbaum Associates.
Stephan, Walter G., Oscar Ybarra, and Kimberly Rios Morrison. 2009. “Intergroup Threat Theory.”
In Handbook of Prejudice, edited by Todd D. Nelson, 43–59. New York, N.Y.: Psychology Press.
Stets, Jan, and Peter Burke. 2000. “Identity Theory and Social Identity Theory.” Social Psychology
Quarterly 63(3): 224–237.
Stoakes, Frank. 1975. “The Supervigilantes: The Lebanese Kataeb Party as a Builder, Surrogate and
Defender of the State,” Middle Eastern Studies 11(3): 215-236.
Stryker, Sheldon. 1980. Symbolic Interactionism: A Social Structural Version. Menlo Park:
Benjamin/Cummings.
Stryker, Sheldon, Timothy J. Owens, and Robert W. White, eds. 2000. Self, Identity, and Social
Movements. Minneapolis: University of Minnesota Press.
Stryker, Sheldon, and Richard T. Serpe. 1994. “Identity Salience and Psychological Centrality:
Equivalent, Overlapping, or Complementary Concepts?” Social Psychology Quarterly 57, no. 1: 16-35.
Sundberg, Ralph and Erik Melander (2013) Introducing the UCDP Georeferenced Event Dataset.
Journal of Peace Research 50(4).
148
Tajfel, Henri. 1970. “Experiments in Intergroup Discrimination.” Scientific American 223, no. 5: 96-
102.
Tajfel, Henri. 1974. “Social Identity Theory and Intrgroup Behavior.” Social Science Information/sur les
sciences sociales 13, no. 2: 65-93.
Tajfel, Henri. 1981. Social Identity and Intergroup Behavior. Cambridge: Cambridge UP.
Tajfel, Henri, Michael G. Billig, R.P. Bundy, and Claude Flament. 1971. “Social Categorization and
Intergroup Behavior.” European Journal of Social Psychology 1, no. 2 (June): 149-178.
Tajfel, Henri and John C. Turner. 1986. “The Social Identity Theory of Intergroup Behavior.” In
Psychology of Intergroup Relations, edited by Stephen Worchel and William G. Austin, 7-24, 2
nd
ed.
Chicago: Nelson-Hall.
Tamimova, May. 2018. “The Black Saturday Massacre of 1975: The Discomfort of Assembling the
Lebanese Civil War Narrative,” Contemporary Levant 3(2): 123-136.
Tegho, Julie A. 2018. “A Church at War: Clergy and Politics in Wartime Lebanon (1975-82),”
Providence. URL: https://providencemag.com/2019/09/a-church-at-war-clergy-politics-in-wartime-
lebanon-1975-82/.
Thaler, Kai. 2012. “Ideology and Violence in Civil Wars: Theory and Evidence from Mozambique
and Angola.” Civil Wars 14(4): 546-567.
Thomas, E.F. and C. McGarty. 2009. “The role of efficacy and moral outrage norms in creating the
potential for international development activism through group-based interaction,” British Journal of
Social Psychology 48: 115-134.
Thomas, E.F., C. MacGarty, and K.I. Mavor. 2009. “Aligning identities, emotions and beliefs to
create commitment to sustainable social and political action,” Personality and Social Psychology Review
13(3): 194-218.
Tilly, Charles. 1985. “War Making and State Making as Organized Crime” in Evans, Rueschemeyer,
and Skocpol, eds, Bringing the State Back In. Cambridge: Cambridge UP.
Toft, Monica Duffy. 2006. “Issue Indivisibility and Time Horizons as Rationalist Explanations for
War,” Security Studies, 15(1): pp. 34–69.
Toft, Monica Duffy. 2007. “Getting Religion? The Puzzling Case of Islam and Civil War.”
International Security 31(4).
Toft, Monica Duffy & Yuri M. Zhukov. 2015. Islamists and Nationalists: Rebel Motivation and
Counterinsurgency in Russia's North Caucasus. American Political Science Review, 109(2): 222-238.
Treacy, Matt. 2011. The IRA 1956-69: Rethinking the Republic. Manchester: Manchester UP.
149
Treier, Shawn, and D. Sunshine Hillygus. 2009. “The Nature of Political Ideology in the
Contemporary Electorate.” Public Opinion Quarterly 73(4): 679-703.
Treier, Shawn, and Simon Jackman. 2008. “Democracy as a Latent Variable.” American Journal of
Political Science 52(1): 201-217.
Trump, Donald J. “President Donald J. Trump’s RNC Acceptance Speech.” 27 August 2020.
https://www.nbcnews.com/politics/2020-election/read-full-text-president-donald-trump-s-
acceptance-speech-rnc-n1238636
Turner, John C. 1975. Social comparison and social identity: some prospects for intergroup
behaviour. Eur. J. Soc. Psychol. 5:1–34.
Turner, John C., Rupert J. Brown, and Henri Tajfel. 1979. “Social Comparison and Group Interest
in Ingroup Favouritism.” European Journal of Social Psychology 9(2): 187–204.
Turner, John C., and Katherine J. Reynolds. 2001. “The social identity perspective in intergroup
relations: Theories, themes, and controversies.” Blackwell Handbook of Social Psychology: Inter- group
Processes, eds. Rupert Brown and Samuel L. Gaertner. Hoboken, NJ: Wiley, pp. 4: 133–152.
UCDP - Uppsala Conflict Data Program. https://ucdp.uu.se/.
Ugarriza, Juan E., and Matthew J. Craig. “The Relevance of Ideology to Contemporary Armed
Conflicts: A Quantitative Analysis of Former Combatants in Colombia,” Journal of Conflict Resolution
57(3), 2013: 445-477.
UN Office for the Coordination of Humanitarian Affairs, Protection of Civilians Report|24-31 May
2021. https://www.ochaopt.org/poc/24-31-may-2021
Valentino, Nicholas A., Vincent L. Hutchings, and Ismail K. White. 2002. “Cues that Matter: How
Political Ads Prime Racial Attitudes During Campaigns,” American Political Science Review 96(1): 75-90.
van Zomeren, M., R. Spears, and C.W. Leach. 2008. “Toward an integrative social identity model of
collective action: Does relevance of group identity influence how people cope with collective
disadvantage?” British Journal of Social Psychology 47, 353-372.
Varshney, Ashutosh. 2003. “Nationalism, Ethnic Conflict, and Rationality.” Perspectives on Politics
1(01): 85– 99.
Vogt, Manuel, Nils-Christian Bormann, Seraina Ruegger, Lars-Erik Cederman, Philipp Hunziker,
and Luc Girardin. 2015. “Integrating Data on Ethnicity, Geography, and Conflict: The Ethnic
Power Relations Data Set Family.” Journal of Conflict Resolution 59(7): 1327-1342.
von Einseidel, Sebastian, Louise Bosetti, James Cockayne, Cale Salih, and Wilfred Wan. 2017. “Civil
War Trends and the Changing Nature of Armed Conflict,” United Nations University Center for
Policy Research, Occasional Paper 10.
Walter, Barbara F. 2017. “The Extremist’s Advantage in Civil Wars.” International Security 42(2): 7-39.
150
Walter, Barbara F. 2022. How Civil Wars Start and How to Stop Them. Crown/Penguin Random House.
Watson, David, Lee Anna Clark, and Auke Tellegen. 1988. “Development and Validation of Brief
Measures of Positive and Negative Affect: The PANAS Scales.” Journal of Personality and Social
Psychology 54(6): 1063-70.
Westwood, Sean J., Justin Grimmer, Matthew Tyler, and Clayton Nall. 2022. “Current Research
Overstates American Support for Political Violence,” Proceedings of the National Academy of Sciences of the
United States of America (PNAS) 119(12): 1-10.
Westwood, Sean J., Justin Grimmer, Matthew Tyler, and Clayton Nall. 2022. “Reply to Kalmoe and
Mason: The pitfalls of using surveys to measure low-prevalence attitudes and behavior,” Proceedings of
the National Academy of Sciences of the United States of America (PNAS) 119(32).
Wall Street Journal Editorial Board. “Marjorie Taylor Greene Plays Dumb,” Wall Street Journal,
February 28, 2022. URL: https://www.wsj.com/articles/marjorie-taylor-greene-plays-dumb-paul-
gosar-nick-fuentes-11646090513.
Wayne, Carly. February 2022. “Risk or Retribution: How Citizens Respond to Terror.” Unpublished
Manuscript.
Weber, Max. 1965. “Politics as a Vocation,” translated by H.H. Gerth and C. Wright Mills.
Whiting, Sophie. 2016. “Mainstream Revolutionaries: Sinn Féin as a “Normal” Political Party?”
Terrorism and Political Violence 28(3): 541-560.
Wimmer, Andreas. 2013. Waves of War: Nationalism, State Formation, and Ethnic Exclusion in the Modern
World. Cambridge: Cambridge UP.
Wimmer, Andreas, Lars-Erik Cederman, and Brian Min. 2009. Ethnic Politics and Armed Conflict.
A Configurational Analysis of a New Global Dataset (online appendix). American Sociological Review 74
(2):316–37.
Wucherpfennig, Julian, Nils W. Metternich, Lars-Erik Cederman, and Kristian Skrede Gleditsch.
2012. “Ethnicity, the State, and the Duration of Civil War.” World Politics 64(1): 79-115.
Yeager, David Scott, Jon A. Krosnick, Linchiat Chang, Harold S. Javitz, Matthew S. Levendusky,
Alberto Simpser, and Rui Wang. “Comparing the accuracy of RDD telephone surveys and internet
surveys conducted with probability and non-probability samples.” Public Opinion Quarterly, 75(4): 709-
747, October 2011.
Yogeeswaran Kumar and Nilanjana Dasgupta. 2010. ‘Will the “real” American please stand up? The
effect of implicit national prototypes on discriminatory behavior and judgments.’ Personality and Social
Psychology Bulletin 36: 1332–45.
Yogeeswaran Kumar, Nilanjana Dasgupta, and Cristian Gomez. 2012. “A new American dilemma?
The effect of ethnic identification and public service on the national inclusion of ethnic minorities.”
European Journal of Social Psychology 42: 691–705.
151
Yzerbyt, V., M. Dumont, D. Wigboldus, and E. Grodjin. 2003. “I feel for us: The impact of
categorization and identification on emotions and action tendencies,” British Journal of Social Psychology
42: 533-549.
Zeitzoff, Thomas. 2014. “Anger, Exposure to Violence, and Intragroup Conflict: A “Lab in the
Field” Experiment in Southern Israel.” Political Psychology 30(3): 309-335.
Zeitzoff, Thomas. 2022. “The Future of Political Violence in the US.” 1 March 2022, American
University School of Public Affairs, Washington, DC. Panel Discussion.
Zhou, Yang-Yang. 2020. “How Refugee Resentment Shapes National Identity and Citizen
Participation.” Unpublished manuscript:
https://static1.squarespace.com/static/53d71184e4b0653934832e81/t/5bdb592e0e2e72d1672ed27
a/1541101877512/Zhou_NatIdentity.pdf
152
APPENDIX 1: ARTICLE 1 SUPPLEMENTARY MATERIAL
1.1: RMarkdown File of Measurement Model Script
Measuring Identity Salience as a Latent Variable
Below is the R script for running a measurement model of identity salience as a latent trait. I
measure this using a 12-item scale comprised of questions that measures American national
identity salience by asking respondents about their primary ascription, cognitive prioritization
of the identity in question, and affective valence.
Reading in Data and Loading Required Packages
I run this model using a function in the MCMCpack R package. I also install the coda and psych
packages.
setwd("/Users/sarahcueva/Dropbox/Dissertation/Data/Article1_Measurement model
/Raw data/")
df <- read.csv("matrixIRT.csv")
library(MCMCpack)
## Loading required package: coda
## Loading required package: MASS
## ##
## ## Markov Chain Monte Carlo Package (MCMCpack)
## ## Copyright (C) 2003-2022 Andrew D. Martin, Kevin M. Quinn, and Jong Hee
Park
## ##
## ## Support provided by the U.S. National Science Foundation
## ## (Grants SES-0350646 and SES-0350613)
## ##
library(coda)
library(psych)
Running the model
We use MCMCpack’s MCMCordfactanal function with all 12 items included. Per the factor
analysis, I choose to set the model to use one factor with unconstrained parameters, and run
100 MCMC iterations.
set.seed(123)
posterior <- MCMCordfactanal(~culturaleduc+neighbors+natanthem+borders+proud+
optimistic+fear+buyus+fp+central+flagburn+natrank,
153
data=df, factors=1,
lambda.constraints=list(),
burnin=0, mcmc=100, thin=1, verbose=10,
L0=0.5, store.lambda=TRUE,
store.scores=TRUE, tune=1.2)
##
##
## MCMCordfactanal iteration 1 of 100
## Lambda =
## 2.72467 0.01487
## 0.92155 0.01482
## 1.02358 -0.02710
## 1.79220 0.00852
## 2.12345 -0.02742
## 1.25474 -0.03282
## 1.94280 -0.02541
## 1.91097 0.03086
## 1.83021 -0.05292
## 1.68070 0.03415
## 1.06908 -0.01566
## 1.36299 0.02776
##
## Metropolis-Hastings acceptance rates =
## 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
##
## MCMCordfactanal iteration 11 of 100
## Lambda =
## 2.71406 -0.33921
## 0.97355 -0.78553
## 0.92494 -0.93218
## 1.67764 -0.81537
## 2.25694 -0.88303
## 1.24258 -0.68528
## 1.99112 -0.07853
## 1.75077 -0.63878
## 1.77741 -0.76298
## 2.07955 -0.92991
## 1.08967 -0.79624
## 1.32726 -0.29604
##
## Metropolis-Hastings acceptance rates =
## 0.00 0.09 0.09 0.09 0.36 0.09 0.09 0.18 0.09 0.18 0.00 0.00
##
## MCMCordfactanal iteration 21 of 100
## Lambda =
## 2.65211 -0.36289
## 0.89372 -0.92863
## 1.16403 -1.15671
## 1.96163 -1.05189
154
## 2.38601 -0.95452
## 1.25767 -0.73300
## 1.92151 -0.11670
## 1.69804 -0.73513
## 1.78628 -0.82727
## 2.05410 -1.09954
## 1.31221 -1.01331
## 1.30884 -0.27945
##
## Metropolis-Hastings acceptance rates =
## 0.00 0.05 0.10 0.14 0.29 0.05 0.05 0.10 0.05 0.10 0.10 0.00
##
## MCMCordfactanal iteration 31 of 100
## Lambda =
## 2.66048 -0.36304
## 0.94693 -0.93694
## 1.16798 -1.23896
## 2.02776 -1.09611
## 2.48101 -0.97409
## 1.29213 -0.77504
## 1.82397 -0.15711
## 1.76823 -0.75664
## 1.90097 -0.85007
## 2.06222 -1.13248
## 1.39695 -1.22082
## 1.35104 -0.31588
##
## Metropolis-Hastings acceptance rates =
## 0.00 0.03 0.06 0.10 0.23 0.03 0.06 0.10 0.10 0.06 0.13 0.00
##
## MCMCordfactanal iteration 41 of 100
## Lambda =
## 2.63664 -0.39832
## 0.99755 -1.03761
## 1.34560 -1.20745
## 1.98306 -1.10021
## 2.49493 -1.00866
## 1.23575 -0.74023
## 1.91502 -0.16112
## 1.85599 -0.82566
## 1.89476 -0.81178
## 2.05163 -1.14547
## 1.39012 -1.18702
## 1.33789 -0.32687
##
## Metropolis-Hastings acceptance rates =
## 0.00 0.05 0.10 0.07 0.17 0.02 0.07 0.10 0.07 0.07 0.10 0.00
##
## MCMCordfactanal iteration 51 of 100
## Lambda =
155
## 2.68540 -0.41060
## 1.09656 -1.00474
## 1.58851 -1.36696
## 2.16588 -1.11384
## 2.64571 -0.99430
## 1.26664 -0.78111
## 1.91566 -0.11498
## 2.05974 -0.88938
## 1.93340 -0.80904
## 2.13431 -1.12767
## 1.39660 -1.15277
## 1.32495 -0.36169
##
## Metropolis-Hastings acceptance rates =
## 0.04 0.04 0.12 0.08 0.16 0.02 0.06 0.10 0.06 0.06 0.10 0.00
##
## MCMCordfactanal iteration 61 of 100
## Lambda =
## 2.74677 -0.42352
## 1.22004 -0.99785
## 1.68358 -1.45113
## 2.23811 -1.14339
## 2.71358 -1.03196
## 1.34379 -0.76416
## 1.93464 -0.09972
## 2.09065 -0.87397
## 1.99311 -0.80269
## 2.21685 -1.11322
## 1.55641 -1.24329
## 1.34473 -0.34572
##
## Metropolis-Hastings acceptance rates =
## 0.03 0.05 0.13 0.07 0.13 0.02 0.05 0.08 0.05 0.05 0.08 0.00
##
## MCMCordfactanal iteration 71 of 100
## Lambda =
## 2.73400 -0.33292
## 1.24417 -1.09854
## 1.59509 -1.38476
## 2.19797 -1.13165
## 2.70969 -1.02340
## 1.36706 -0.76276
## 1.88426 -0.14659
## 2.26286 -0.90215
## 1.97979 -0.91718
## 2.33299 -1.15601
## 1.47946 -1.23394
## 1.36738 -0.30871
##
## Metropolis-Hastings acceptance rates =
156
## 0.04 0.07 0.11 0.06 0.11 0.01 0.04 0.08 0.04 0.06 0.07 0.00
##
## MCMCordfactanal iteration 81 of 100
## Lambda =
## 2.73753 -0.39448
## 1.22206 -1.10756
## 1.65108 -1.49352
## 2.16808 -1.09443
## 2.67077 -1.05755
## 1.34106 -0.85479
## 1.92259 -0.16352
## 2.27073 -0.93054
## 1.96334 -0.91470
## 2.43065 -1.26167
## 1.45721 -1.18984
## 1.36145 -0.33714
##
## Metropolis-Hastings acceptance rates =
## 0.04 0.06 0.10 0.05 0.10 0.01 0.04 0.09 0.06 0.06 0.06 0.00
##
## MCMCordfactanal iteration 91 of 100
## Lambda =
## 2.70408 -0.39845
## 1.15740 -1.12543
## 1.72085 -1.56164
## 2.11665 -1.10307
## 2.66495 -1.03082
## 1.31508 -0.74532
## 1.91365 -0.13574
## 2.20876 -0.90355
## 1.87819 -0.85299
## 2.41261 -1.23670
## 1.45854 -1.16305
## 1.35429 -0.29487
##
## Metropolis-Hastings acceptance rates =
## 0.04 0.07 0.10 0.04 0.09 0.02 0.03 0.08 0.05 0.07 0.05 0.00
##
## Acceptance rates:
## culturaleduc neighbors natanthem borders proud optimistic fear buyus fp
## 0.04 0.07 0.09 0.05 0.09 0.03 0.03 0.08 0.05
## central flagburn natrank
## 0.06 0.05 0
# plot(posterior)
# summary(posterior)
157
Extracting and storing model output
These values correspond to the latent trait estimates of each respondent’s level of identity
salience, represented as theta (θ) in the model equation.
posterior2 <- posterior[,58:1576] #extracting the simulated estimates of the
latent trait
thetas <- apply(posterior2, 2, mean) #taking posterior means of latent trait
for each respondent
# write.csv(thetas, "idsalience_thetas.csv")
158
APPENDIX 2: ARTICLE 2 SUPPLEMENTARY MATERIAL
2.1: IRB Protocol
Study Title: National Identity Salience and Attitudes Toward Political Violence: Survey
Experimental Evidence
PI Name: Sarah Cueva Egan
Study Procedures
1. Background/Rationale
a. The proposed study falls within the theoretical framework of previous research situated
within social identity theory (i.e. Tajfel 1981, Tajfel and Turner 1986), which seeks to
understand how people sort themselves into groups and how they relate to people within
the same group and to members of outgroups. Many studies have tested and
experimentally manipulated different identities to understand the relationship between
the salience of these identities and various types of political behavior (i.e. Brewer 2001;
Huddy 2001; Stryker et al 2000; Brady and Kaplan 2009; Hatemi and McDermott 2016).
The investigator aims to understand the relationship between changes in the salience of a
particular type of identity—national identity—and attitudes toward a particular type of
political behavior—political violence. In the context of the proposed study, the main
research questions are the following.
1) What is the relationship between identity salience and expressed attitudes toward
political violence? Does a higher baseline level of identity salience predict a higher
likelihood of expressed support for political violence waged on behalf of that identity
group?
2) Would priming a particular identity—in this case, an American national identity—
lead to an increased propensity to express support for, or willingness to engage in,
political violence?
Recent studies have demonstrated connections between priming identity and different
political outcomes, such as support for anti-democratic policies (i.e. Pan and Xu 2019),
pro-social behavior toward ingroups (i.e. Chardysh 2015), and revanchist attitudes
among religious minorities (i.e. Nair and Sambanis 2018). I build upon the theoretical
findings and empirical methods of such work to determine whether a similar
phenomenon occurs when the outcome in question is political violence
100
.
Understanding whether, how, and under what conditions changes in identity salience
100
I define political violence as any act meant to injure, kill, or maim others as a result of perceived differences and
political goals. The term as I define it encompasses acts with political motivations that use violent tactics—for example,
shooting, bombing, kidnapping, or any act that harms people or property (Chenoweth and Stephan 2011). Specifically,
this could take the form of mob violence, rioting, armed insurrection or insurgency, or acts of terror committed by
individuals or groups. These acts could be directed at members of a perceived outgroup (i.e. a minority ethnic group) or
against the government itself (for example, an armed insurrection against a sovereign government). Political violence
does not include random acts of violence with no clear political goal or motivation, nor does it include instances of
interstate conflict.
159
might lead to dangerous behavior between groups is important not only for the sake of
academic knowledge, but for potentially staging policy interventions that mitigate the risk
of identity-related political violence before it begins.
2. Purpose/Objectives/Aims/Research Questions
a. The proposed study is part of my larger dissertation project that examines the
relationship between identity salience and outcomes of political violence. The core
objective here is to understand how exposure to primes meant to increase the salience of
national identity affects expressed attitudes toward political violence. Broadly speaking,
the theoretical expectation is that exposure to a treatment that invokes a sense of a
strong common American national identity that is under threat from an outgroup will
result in more permissive attitudes toward the commission of political violence on behalf
of the ingroup that is primed. I will test the effect of a treatment that primes national
identity salience on attitudes toward political violence. Those in the treatment group will
be shown an image that is meant to do two things: 1) prime their sense of American
national identity, and 2) evoke a sense of their country and identity being threatened.
Thus, the study tests the following hypotheses:
i. H1: Those with higher baseline levels of national identity salience will, all else
being equal, be more likely to express permissive attitudes toward political
violence.
ii. H2: Exposing individuals to a threatening prime connected to American
national identity will lead to an increased willingness to support or engage in
political violence on behalf of that identity.
b. Objectives: Substantively, the broader goals of the proposed study are to 1) understand
the relationship between individual attachment to group identities and political
violence and to 2) determine whether certain types of appeals to these identities are
more likely to lead to an expressed willingness to either engage in or support political
violence waged on behalf of an identity group. The findings of this study would hold
value not only within the field of political science, but could inform policy
interventions meant to reduce identity-related triggers of political violence. This
holds true not only with national identities, but ethnic/ethnoracial,
religious/sectarian, partisan, or any combination of these identity classifications.
101
Understanding the effects of divisive rhetoric from politicians, political figures, and
ordinary citizens is of vital importance to society due to its potential to reduce
radicalization that could lead to political violence, especially in an era when such
rhetoric is easily manipulated and virally spread through social and news media.
c. Potential for future research: If the results of the proposed study lead to a statistically
significant effect, this could also inform further studies as to whether the increased
salience of different types of identity—for example, ethnonational or religious—are
more or less prone to lead to political violence. Understanding the differential effects
across identity groups in different contexts, both within the United States and
outside of it, would be a potential benefit of further studies that could be informed
by the current one.
3. Participants (sample)
101
These would require further study and fall outside of the scope of the current study.
160
a. Inclusion/exclusion criteria: The sample excludes minors and anyone who is not currently
located within the United States. The aim of imposing these exclusion criteria is to
recruit a sample of American adults who are of voting age and most likely to be affected
by and potentially involved in political activity in the United States.
b. Rationale for excluding children from the study: I exclude children from the proposed study for
two reasons. First, the experimental treatment is meant to evoke a sense of threat, and
thus I exclude children to avoid exposing them to anything that could be detrimental to
their wellbeing. Second, children are unlikely to have or develop politically meaningful
attachments to national identity—or even to fully understand them— and are even less
likely to have the motivation or the means to engage in political violence in this context.
c. Participants are not part of any special subject populations.
4. Recruitment/Screening Process (sampling strategy)
a. Recruitment location: Study participants will be recruited from Amazon’s Mechanical
Turk (MTurk) online task platform. These are paid “workers” who complete tasks
on the platform and receive compensation for their time. They remain fully
anonymous to the researcher.
b. Recruitment method: Once activated as a Human Intelligence Task (HIT) on MTurk,
MTurk workers will have the option to view basic information about the study and
can choose whether or not to opt in. Once 1500 workers complete the survey, it will
be deactivated and results collected and analyzed.
c. Recruitment materials/screening tools: No materials or screening tools will be used for
recruitment.
d. Screening process: Amazon MTurk has an option to specify inclusion and exclusion
criteria prior to activation of the survey. Anyone below the age of 18 is automatically
excluded, as MTurk workers must verify through the platform that they are 18 or
older in order to register and receive payment for completed tasks. To recruit only
those participants located in the United States, I instruct MTurk to exclude any
worker whose IP address is not in the US.
i. No screening documents will be used.
ii. No screening data will be retained.
5. Methods
a. The proposed study is an online survey experiment conducted on Amazon’s Mechanical
Turk (MTurk) platform. The survey includes 29 questions, and respondents assigned to
the experimental group will be shown an image and asked to comment on it. I expect
that the study will take fifteen minutes maximum, though most likely it will be far less
for a majority of respondents. The survey instrument is attached separately within this
application for the board’s review.
i. Sample and recruitment: I will recruit a sample size of 1500 adult Americans who are
MTurk users. Anyone under the age of eighteen or not living in the United States
are ineligible to participate and will be screened out by MTurk. Once participants
are recruited into the sample, they are directed to follow a link to the survey. There
is no pre-screening beyond the initial exclusion criteria. Once selected, participants
will be shown a short consent form and be given the option to opt out of the study
or to continue their participation.
ii. Study design: I first design my survey instrument on Qualtrics (attached in section
21.2 of this application) and embed an anonymous link to the survey within the
task created through my MTurk account. The survey contains 29 questions with
161
scaled response options. The questions in the pre-treatment block measure
baseline identity salience, as well as collect key demographic information and gauge
political activity as control variables. Respondents are randomized into two groups,
an experimental and control (750 in each). The first, the control group, does not
receive a treatment and instead will simply be asked to complete the survey. Those
assigned to the experimental group will be shown an image immediately after
completing the demographic questions, and will then be asked to caption the
photo and explain their caption. They will then be directed to answer the
remaining questions in the post-treatment block. The post-treatment block
contains seven questions that measure attitudes toward political violence.
iii. Timeline of activities: I will field the survey upon receipt of IRB approval, and will
keep it active for however long it takes for 1500 individuals to complete the study.
Each study participant’s engagement with the study ends upon completion of the
survey. Once the responses are complete, the investigator will close the survey and
analyze the results. The fielding of the study itself will not exceed two weeks, and
start date is contingent upon IRB approval. Note that respondents will be
automatically compensated three days following their completion of the survey, on
a rolling basis.
Task Duration Target completion date
Data collection via
MTurk
2 weeks June 30, 2022
Analysis and writeup of
results
1 month July 15, 2022
iv. Experimental intervention: One of the two groups will receive an intervention in the
form of an image that is meant to prime their American national identity salience
and evoke a sense of threat. To avoid introducing partisan bias that could affect
the results and lead to attribution errors, I lightly edited the photo and removed
any information or landmarks that could lead respondents to immediately associate
it with a particular event. All respondents are randomized into one of the two
groups. After answering the pre-treatment questions, respondents will be shown
the treatment within the survey and will then be directed to respond to the post-
treatment questions. The purpose of the intervention is to manipulate the salience
of respondents’ (American) national identity. The treatment is embedded in the
survey for those respondents who are randomly assigned to that group. Figure 1
outlines the treatment. It is also shown in the survey instrument attached to this
amendment submission.
162
Figure 1: Experimental Treatment Image
v. Privacy: All respondents will remain fully anonymous from the beginning to the end
of the study: no personally identifiable information will be collected or maintained,
nor is it possible for the researcher to obtain any personal identifying information
from Amazon per company policy.
vi. Data management/storage: All data collected will be downloaded from the PI’s
private, password-protected Qualtrics account onto her local hard drive. The data
will be stored in her password-protected Dropbox account in a private folder.
vii. Dissemination of findings: The results from this study will be presented in a chapter of
the PI’s dissertation manuscript. The primary audience will be her dissertation
committee. Again, survey responses and participants are fully anonymous.
b. Instrumentation
i. Questionnaires/Survey Measures
A. The survey instrument was designed in Qualtrics and includes 29 multiple
choice, rank order, or (in one case) open-ended questions. All survey questions
are developed by the investigator (see the attached instrument in section 21.2).
ii. Qualitative instruments
A. Two open-ended questions are included in the survey instrument. The first,
which asks respondents to write a caption for an image and explain their
caption, is part of the experimental treatment condition and is meant to
encourage effortful information processing to maximize the potential that the
treatment will have an effect. The second open-ended question is only
163
displayed to respondents who indicate that political violence may be justified in
some form, and asks them to specify when they think it would be acceptable. A
response to this question is optional.
c. Data Analysis
i. Once responses are collected from the sample of 1500 respondents, the researcher
will deactivate the HIT on MTurk and download the data in CSV format onto her
local hard drive. The data will then be loaded into the R program
102
for data
cleaning and statistical analysis. Primary tests will consist of regression models,
tests of statistical significance, and a difference in means test for the treatment
groups.
ii. No secondary data analysis will be conducted for this project.
102
R Core Team (2020). R: A language and environment for
statistical computing. R Foundation for Statistical
Computing, Vienna, Austria. URL
https://www.R-project.org/.
164
2.2: MTurk Survey Instrument
A brief research survey
Start of Block: Please answer the following questions.
Q1 Have you participated in any political activity (for example, a protest, campaign, march, or rally)
in the past two years?
Yes
No
Prefer not to say
Q2 How important is it that K-12 education include instruction on American history, culture, and
values?
Not at all important
Slightly important
Moderately important
Very important
Extremely important
Q3 How important is it that your neighbors are US citizens?
Not at all important
Slightly important
Moderately important
Very important
Extremely important
Q4 How important is it that people stand for the national anthem at sporting events?
Not at all important
Slightly important
Moderately important
Very important
Extremely important
165
Q5
I am going to show you a series of statements. Please indicate the extent to which you agree or
disagree with each.
One of the US government's top priorities must be to protect its borders at all costs.
Strongly disagree
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q6
I am proud to be an American.
Disagree strongly
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q7 I am optimistic about the direction that America is headed.
Disagree strongly
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q8 I fear for future generations of Americans.
Disagree strongly
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
166
Q9 Americans should make every effort to buy goods made in the US, even if it costs them more
money to do so.
Strongly disagree
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q10 Pursuing America's own interests abroad must always come before considering the interests of
its allies.
Strongly disagree
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q11 Being an American is central to who I am.
Disagree strongly
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q12 People who burn the American flag should be subject to a fine or legal punishment.
Strongly disagree
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Q13 Many individuals have multiple components of their overall identity that make up who they are.
For example, an individual may consider themselves a member of a religious group, national group,
167
ethnic group, etc. Thinking about each of the following types of identity and how they relate to you,
please rank them from most important to least important.
______ Religion
______ Race/ethnicity
______ Nationality
______ Gender
______ Sexual Orientation
______ Political/partisan (i.e. Democrat or Republican)
______ Class
Q14 I am going to list a number of colors and will ask you a question. If you are paying attention, do
not answer this question and instead select "orange."
Which color typically represents the Green Party?
Red
Orange
Green
Blue
Yellow
Q15 Are you currently registered to vote?
Yes
No
I don't know
Display This Question:
If Are you currently registered to vote? = Yes
Q16 Please complete the following statement:
"I am registered as a/an..."
Republican
Democrat
Libertarian
Independent
Other ________________________________________________
168
Q17 What is your gender?
Male
Female
Non-binary
Other
Prefer not to say
Q18 What is your age?
18-24 years old
25-34 years old
35-44 years old
45-54 years old
55+ years old
Q19 What is your ethnicity?
White
Black or African American
American Indian or Alaska Native
Asian
Native Hawaiian or Pacific Islander
Hispanic or Latino/a
Other ________________________________________________
Q20 What is your highest level of completed education?
Some high school
High school
Some college
College graduate
Graduate or professional school
169
Q22 If applicable, what is your religious affiliation?
None
Christian
Jewish
Muslim
Buddhist
Hindu
Other
Prefer not to say
Q23 What is your annual household income?
$0-$20,000
$20,001-$40,000
$40,001-$60,000
$60,001-$80,000
$80,001-$100,000
$100,001 or over
Prefer not to say
End of Block: Please answer the following questions.
Start of Block: T0
Q24 Now, please answer the following questions.
End of Block: T0
Start of Block: T1
Q25
Now, please take a look at the picture below.
170
Now, imagine you need to write a caption describing what is happening in this picture. Please write
your caption in the box below, then explain in 1-2 sentences why you chose this caption.
End of Block: T1
Start of Block: Q20
Q23 If you felt your country was in danger, how likely would you be to engage in violence against
other Americans who were threatening it?
Extremely unlikely
Very unlikely
Somewhat unlikely
Neither likely nor unlikely
Somewhat likely
Very likely
Extremely likely
171
Q24 If people you know felt the country was in danger, how likely do you think they would be to
engage in violence against other Americans who were threatening it?
Extremely unlikely
Very unlikely
Somewhat unlikely
Neither likely nor unlikely
Somewhat likely
Very likely
Extremely unlikely
Q25
To what extent do you agree or disagree with the following statement?
"Using violence against others to achieve political goals is sometimes justified."
Disagree strongly
Moderately disagree
Neither agree nor disagree
Moderately agree
Strongly agree
Page Break
Q26 In your opinion, when, if ever, would political violence be justified?
Q27 When, if ever, is it okay for someone to send threatening messages to people they disagree with
politically?
Never
It depends
Always
172
Q28 Since the events at the US Capitol on January 6, 2021, over 600 Americans have been arrested
for their involvement. Many of them have been sentenced to time in jail for their actions. In your
opinion, are these punishments...
Too harsh
Somewhat harsh
Just right
Somewhat lenient
Too lenient
Q29 Do you ever think that we would be better off as a country if some members of the public just
died?
Yes
No
Maybe
End of Block: Q29
173
2.3: Additional Regression Analyses with Constituent Indicators of Political Violence Index
as Dependent Variables
This subsection shows supplemental regression results using the main constituent indicators
comprising the political violence index as dependent variables. The table shows results of all
individual political violence questions regressed against the main predictors. The results seen in the
body of the dissertation (using the political violence weighted index as a predictor) are largely the
same as those seen here.
Political Violence Constituent Indicators
Dependent variable:
violentself violentothers betteroff pvjustified threatmsg jan6
(1) (2) (3) (4) (5) (6)
Identity Salience 1.045
***
0.941
***
0.081
***
0.319
***
0.193
***
0.130
***
(0.060) (0.058) (0.030) (0.047) (0.023) (0.038)
Assigned to
Treatment
-0.102 0.037 -0.020 -0.011 -0.031 0.095
*
(0.080) (0.078) (0.041) (0.059) (0.031) (0.050)
Salience*Treated -0.141 -0.186
**
-0.007 0.040 0.010 0.089
*
(0.086) (0.084) (0.044) (0.066) (0.033) (0.054)
Constant 4.177
***
4.308
***
2.008
***
2.713
***
1.555
***
2.565
***
(0.056) (0.055) (0.029) (0.041) (0.021) (0.035)
Observations 1,914 1,914 1,897 1,654 1,908 1,910
R
2
0.213 0.178 0.007 0.060 0.071 0.024
Adjusted R
2
0.212 0.177 0.005 0.058 0.070 0.023
Residual Std.
Error
1.749 (df =
1910)
1.706 (df =
1910)
0.885 (df =
1893)
1.188 (df =
1650)
0.666 (df =
1904)
1.096 (df =
1906)
F Statistic
172.704
***
(df =
3; 1910)
138.180
***
(df =
3; 1910)
4.250
***
(df =
3; 1893)
35.003
***
(df =
3; 1650)
48.597
***
(df =
3; 1904)
15.748
***
(df =
3; 1906)
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
174
2.4: A Note on Political Violence Question Wording
A recent and ongoing debate on question wording for studies of political violence brought
renewed attention to how survey questions can be best designed to gather information from
respondents that accurately represents their attitudes toward political violence without introducing
unintended bias through poorly-worded questions. An ongoing exchange between Lily Mason and
Nathan Kalmoe on the one hand and Sean Westwood et al. on the other highlighted ongoing
disagreement among scholars of political violence as to the optimal design of such questions.
Kalmoe and Mason’s most recent published work found a high level of support in the general US
population for political violence, especially partisan political violence; Westwood et al. (2022a) argue
that Kalmoe and Mason’s results are fundamentally biased by overly generic question wording
(Kalmoe and Mason 2022, Westwood et al. 2022a). An additional response by Westwood et al.
argues that the best way to reduce upward bias in questions measuring attitudes about “low-
prevalence events” is to ask specific questions about specific violent acts, as the range of different
acts of political violence is sufficiently broad as to make it difficult to assume that all respondents
have the same conception of what political violence—and, by extension, whether or not they
support it (Westwood et al. 2022b).
In designing the questions on attitudes toward political violence in the survey experiment, I
considered these points and their methodological merits. Ultimately, I decided that the concern
about generic questions, regardless of the merits of these concerns, would not be an issue in the case
of this survey experiment. Because the main goal was to determine a) whether there was a short-
term effect on attitudes given the treatment and b) to get an understanding of general attitudes
toward violence conducted on behalf of an ingroup identity, it was less of a concern that
respondents would have differing conceptions as to what exactly constitutes political violence. Still, I
added some more specific questions—for example, respondent opinions of sentences for
175
defendants involved in the J6 political violence—as additional proxies for attitudes toward political
violence that would also provide more specificity as to what was meant by political violence.
Robustness checks testing the relationship between key predictors and the individual questions that
make up the political violence index indicated that although there were slight differences in effect
sizes, these were negligible. Overall, the results reported using the weighted political violence index
as the main dependent variable were substantively similar to those reported here: there was a strong
and statistically significant result between identity salience and permissive attitudes toward political
violence, while the relationship between assignment to treatment and attitudes toward political
violence are both substantively and statistically insignificant. The similarity between these results
suggests that the main weighted index is an acceptable measure of attitudes toward political violence
for our purposes. At the same time, the difference in effect size between respondents’ estimation of
their own acceptance of political violence and the other questions does provide some anecdotal
support to Westwood et al.’s claim of overestimation of individual acceptance of political violence.
Future research should continue to examine the internal and external validity of survey questions
measuring attitudes toward political violence, as well as the potential for overestimation of
acceptance of political violence.
176
2.5: Results of Covariate Balance Test
177
APPENDIX 3: ARTICLE 3 SUPPLEMENTARY MATERIAL
3.1 Dyadic Identity-Based Political Violence Codebook
Introduction, Scope and Purpose of Data
This document outlines the coding process and variables for the dyadic identity-based political
violence (IBPV) dataset. We recode existing data taken from the Uppsala Conflict Data Program
(UCDP’s) Georeferenced Event Dataset (GED), a disaggregated dataset documenting all episodes
of state-based, one-sided, and nonstate armed conflict that has resulted in a minimum of 25 battle-
related deaths from 1989 through 2020.
The main aim of creating this dataset is to use existing disaggregated data to create a new dataset of
aggregated instances of identity-based political violence, with events ranging from lower-level
stochastic instances of political violence to sustained insurgencies to civil wars.
Because it is disaggregated by event, the UCDP GED as of this writing contains nearly 300,000
events connected to 1288 unique conflicts. To narrow the scope of the data that we analyze here and
to avoid any bias introduced by the onset of the global coronavirus pandemic in 2020, all
observations after 2019 were removed from the dataset, which reduces unique conflict events to
1240. After reviewing these 1240 conflicts, we removed any interstate conflicts, where the two main
sides in the conflict dyad were two different countries. Interstate conflicts include large scale wars
with two or more sovereign states (i.e. the Iraq War), smaller-scale cross-border conflicts between
two sovereign states, such as ongoing hostilities between India and Pakistan over the Kashmir
region. Additionally, we exclude conflict dyads where at least one of the actors is heavily composed
of foreign fighters or is a proxy force for an external power, as was the case with some dyads active
in the war in former Zaire. We also exclude conflicts where one side in the dyad is a major
transnational terror group, namely Al-Qaeda and its affiliates (AQAP and AQIM) and ISIS. Because
such groups have transnational aspirations and have appeared in conflict dyads in multiple countries,
these have been excluded to avoid bias posed by transnational actors that do not necessarily operate
under the same dynamics seen in purely domestic-level conflicts. We also removed any conflict
dyads where the event in question involved an anticolonial campaign. After removing ineligible
dyadic events from the initial dataset, the sample size for initial analysis was 726.
Rather than tracking fatalities within these conflict dyads over time, I summed all fatalities per dyad
to create one unit of observation per dyad. This was done for two reasons. First, the scope of the
data and limited time and resources made it impossible to code all unique dyad-year events,
necessitating a method to streamline the data coding and collection process. Second, because the
aim of the analysis was to understand the difference in outcome severity by different actors, rather
than within the same dyads over time, I chose to make this a cross-national, time-invariant dataset.
However, future iterations of this effort could certainly expand on this work to create a time-series
dataset.
EXCLUSION CRITERIA:
There are some important exclusion criteria we apply where identity groups may be involved in a
conflict dyad, but this dyad is not coded as ibpv=1. If the conflict dyad involves groups from the
178
same ethnic or religious identity group–for example, intra-ethnic conflict dyads or intra-sectarian
dyads would not be considered ibpv=1. There are, for instance, some cases where the actors in
dyads involving communal violence are from the same overarching ethnic group, but are fighting
over communal divisions. We exclude any such instances from the data, and code ibpv=0. As
another example, there are conflict dyads related to the Syrian Civil War wherein the actors are both
identity actors, but we code as ibpv=0 because they belong to the same identity and thus can’t be
said to be fighting an expressly identity-based conflict. Relatedly, we code any dyads that involve
actors splintered from the same group as ibpv=0 (i.e. leadership disputes between Taliban and
Taliban splinter groups). We exclude from the dataset outright any instances where the actor has
been active in multiple countries and is composed of nationals from multiple different countries (for
example, some nomadic ethnic groups in sub-Saharan Africa). We also exclude any dyads for which
there is neither information on that dyad nor information on the actors within the dyad offered via
UCDP’s online platform, as well as any dyads that include more than two actors. Finally, there are
about 60 dyads comprised of a national government and civilians (i.e. Government of Pakistan -
Civilians). I exclude these from the dataset, as the theory being tested does not apply to incumbent
state governments. Though excluding these particular events removes some variation at the upper
bound of the severity scale, the range of severity in the random sample that is coded is sufficiently
broad as to cover minimally-severe conflict dyads (minimum = 25 deaths) and maximally-severe
conflict dyads (maximum = 12,443 deaths).
Defining Identity and Identity-Based Political Violence
We consider an actor to be identity-based if it explicitly advocates for or organizes based upon goals
related to a common ethnicity, religion, or any combination of these categories. Following are the
definitions we use for ethnicity and religion:
· Ethnicity: A group that shares a common ethnicity could organize by race or by a more specific
ethnic group—for example, a white supremacist movement would fall under the ethnic group
category, as would groups like the Chechens in Russia or the Dinka in certain African countries.
Additionally, a group that forms based on a common language—such as Anglophone groups in
Cameroon—would be considered an ethnic group by our definition. It is sometimes the case that
groups with ethnic commonalities also belong to the same religion or nationality. Many ethnic groups
aspire to establish their own autonomous states or regions, such as the Basques in Spain. We do
NOT consider groups based on kinship or clan ties to be ethnic, unless they organize specifically
along ethnic/ethnolinguistic or religious lines. For example, the conflict dyad between different
subclans of the Takima group in Papua New Guinea is not considered identity-based, as both actors
are descended from the same ethnic group and the dispute is primarily kinship based. In other words,
communal groups not based on ethnicity or religion (i.e. Ikot-Offiong and Oko Iboku) or skirmishes
between clans or subclans are not considered to be identity-based.
· Religion: A group that shares a common religion may either subscribe to a broader religious
category (i.e. Judaism, Buddhism, Christianity, etc.) or will belong to a specific sect or
denomination—for example, Shi’a Muslim groups operating in Iraq, Theravada Buddhist monks in
Thailand, or Catholics in Northern Ireland. However, to meet the criteria of an identity group based
on religion, the group must be explicitly limited to include individuals of a religious ingroup.
179
Coding Process
There are two main steps to this coding process. The first is to determine whether the events in the
data were or were not identity-based, resulting in the creation of a dichotomous variable (ibpv).
Once we have determined which events were identity-based, we continue to step 2: coding the main
predictive variables associated with identity coherence and elite support for the perpetrating actor in
the dyad. The criteria and instructions for both steps are below.
Step 1: Determining Identity-Based Political Violence
All political violence events need to be coded as either being identity-based or not identity-based.
This will require coding the following variable:
ibpv – binary variable indicating whether an event is one of identity-based political violence (1 =
yes, 0 = no) – If the perpetrating actor in the conflict dyad (coded as Side A in UCDP data) is
identity-based, or organized along explicitly ethnic or religious lines, then we consider it to be an
identity-based group and ibpv should be coded as 1. The identity component does not need to be
the only motivating factor in a conflict in order for it to be considered identity-based. For example,
an ideologically leftist group advocating for a socialist state can be considered identity-based if this
state is also organized along ethnic lines—one example of this would be ETA in Spain, a Basque
separatist group seeking to create an independent socialist Basque state. Both sides in a conflict do
NOT need to be identity-based in order for the event to be identity-based: one side being identity-
based is sufficient for ibpv to be coded as 1.
If neither or none of the actors in the conflict are identity-based, we code ibpv = 0.
*Once we have coded ibpv for all events in the dataset, we move on to coding the following
variables for only those cases coded as ibpv = 1.
Step 2: Coding Identity Coherence and Elite Support
Once we have narrowed down the initial data to dyads involving identity-based political violence
(we drop any events coded as ibpv=0), we are ready to begin coding the different variables
associated with what will become the two main variables of identity coherence and elite support.
Descriptions of the different indicator variables that make up identity coherence and elite support
are given below.
In coding these variables, we rely primarily on qualitative assessments of the different actors and
conflict dyads provided via the UCDP interactive data platform, as well as data and information
from the Ethnic Power Relations (EPR) data project. In cases where we are unable to make an
informed coding determination using this information, we cross-reference with peer-reviewed
academic journals and reputable data sources that cover civil conflict, including Stanford’s Mapping
Militants Project, the International Crisis Group’s CrisisWatch, and the University of Maryland’s
Global Terrorism Database (GTD).
180
Identity coherence
Identity coherence is a composite measure of multiple theoretically important indicators. These indicators include
original variables coded by the author, as well as three variables adapted from the Ethnic Power Relations (EPR)
Core Dataset and EPR’s ACD2EPR data, a dataset designed to be compatible with UCDP dyadic data.
Variables borrowed from EPR may be slightly modified to match methodology for this project.
idtype – Categorical variable indicating the identity category or categories to which an identity-based
group subscribes.
1 = ethnic
2 = religious
3 = combination (i.e. ethnic and/or religious)
splinter – Dichotomous variable indicating whether an actor is a splinter group, or whether it was
formed as a result of a split with a pre-existing group. [add note for whether group has itself resulted
in formation of splinter groups]
0 = no
1 = yes
goals – A binary variable indicating whether the group has a clear common goal (i.e. separatist state,
the establishment of religious law, removal of incumbent government, etc.). If no (i.e. purely around
grievance or convenience), then this would hint at less identity coherence. This is linked in part with
fractionalization of group/intragroup disagreements. NOTE: This variable refers to whether there
was a clear goal at the outset of the group’s formation, even if later actions by the group are not
related to that goal.
0 = no
1 = yes
goaltype – Only code this variable if goals=1. Categorical variable indicating what type of
goal the group shares. For example, ETA shared the goal of an autonomous and independent
Basque nation-state (3, secession), the Provisional IRA in Ireland shared the goal of reunification
with a majority Catholic and independent nation-state of Ireland (3, independence), the Muslim
Brotherhood in Egypt prior to the current government sought to establish an Islamic government
and Shari’a code of law (4, regime change). If the actor has more than one goal, code the goal that is
highest in severity (i.e. if a group begins defensively but then aims for regime change, code this actor
as a 4). NOTE: If the actor expanded or changed its goals after the time period during which the
conflict dyad was active, we do not consider those goals in making a coding determination. We only
consider what their highest goal was at the time the fighting within that dyad occurred.
1 = defense/survival/protection
This category can include any group/actor that organizes defensively (i.e. in response to an attack
against the group, including disputes over land or livelihoods).
2 = greater political representation/political disagreements
3 = secession, independence, or territorial control
4 = regime change
181
5 = genocide/cleansing/achieving ethnic domination*
*In cases where the actor targets civilians on the basis of an outgroup identity, we code goaltype as 5. We
would primarily see this type of coding for dyads where one side is civilians, and the civilians are targeted on the basis of
their ethnicity or religion.
claims – Taken from ACD2EPR, this is an ordinal indicator of whether an actor has made an
exclusive claim to fight on behalf of an identity group. Reference EPR documentation to see original
categories used by EPR in its dataset. I code this variable as follows:
0 = no exclusive claim made
1 = indirect evidence for claim (i.e. group name)
2 = direct evidence for claim
recruitment – From ACD2EPR, an ordinal indicator of whether an actor recruits members from a
specific ethnic or religious group. Reference EPR documentation to see original categories used by
EPR in its dataset. I code this variable as follows:
0 = no recruitment
1 = ethnic group members are recruited both by the actor and by the government
2 = recruitment
support – From ACD2EPR, indicates whether the actor or group is supported by at least 50% of
the members of the larger ethnic group. Reference EPR documentation to see original categories
used by EPR in its dataset. I code this variable as follows:
0 = little or no support
1 = one ethnic group supports both the actor and the government
2 = large support
Elite Support
Elite support is a composite measure of three theoretically relevant indicators. Data for the first variable was collected
and coded by the author, and the two other variables were adapted from EPR Core data.
polparty – Ordinal variable indicating whether an actor in a conflict has a formal or associated
political party.
0 = no, no political party affiliation
1 = yes, loose or informal affiliation with a political party
2 = yes, formal political party
status – Adapted from EPR Core, an ordinal variable indicating the degree of access of a group to
central state power. As in the EPR data, state power refers solely to executive power–access to
judicial and legislative institutions are not factored into the coding of this variable.
0 = irrelevant - no distinguishable elites with access to power, or groups that were
previously relevant and became irrelevant
182
1 = excluded
2 = shares power as either junior partner or senior partner in a formal or
informal power-sharing agreement
3 = group rules alone, either as a monopoly or as dominant player in the executive with
limited inclusion of outgroup members
reg_aut – Adapted from EPR Core, a dichotomous variable that measures access to executive
power at the regional level.
0 = no meaningful access to executive power at the regional level
1 = meaningful (de jure and de facto) access to executive power at the regional level
notes – Optional category wherein you can include any notes that you consider relevant to your
coding decision.
183
3.2: Table of Variables
Variable Question Scale Measurement Variable
Name
Notes
Identity-
based event
Was the
event/conflict
based on an
ethnic or
religious
identity?
dichotomous 0 = no
1 = yes
ibpv Look first at
UCDP conflict
dyad page to
see if there is
mention of
ethnic or
religious
reason. If not,
look at actor
data: if at least
one actor in the
dyad is
identity-based,
code idbased =
1. If you are
unsure, flag it.
Perpetrating
actor
identity-
based
Was actor 1 in
the event based
on an ethnic or
religious
identity?
dichotomous 0 = no
1 = yes
actor1_idbased
Identity type Which type of
identity was the
actor organized
around?
Ordinal 1 = ethnic
2 = religion
3 = ethnoreligious
idtype This variable
will be coded
for at least one
actor. If only
one actor is
id_based, then
code this and
the following
variables
accordingly. If
an actor is
NOT id-based,
code all of the
following
variables as
NA.
Splinter
group
Was the
actor/group
formed as the
result of
splintering
from a pre-
existing group?
Dichotomous 0 = no
1 = yes
splinter Same rule as
above. Only
code this
variable if
id_based=1.
Goals Does the
actor/group
Dichotomous 0 = no
1 = yes
goals Same rule as
above. Only
code this
184
have a clearly
stated goal?
variable if
id_based = 1.
Goal type What is the
main type of
goal that the
actor/group is
pursuing?
Ordinal 1 =
defense/survival/protection
2 = greater political
representation or political
disagreements
3 = secession, independence,
or territorial control
4 = regime change
5 =
genocide/cleansing/achieving
ethnic domination
goaltype ONLY CODE
THIS
VARIABLE IF
GOAL=1.
Same rule as
above. Only
code this
variable if
id_based = 1.
Claims Does the actor
make an
exclusive claim
to fight on
behalf of a
particular
ethnic or
religious
group?
Ordinal 0 = no claim
1 = indirect claim
2 = direct claim
claims Adapted from
ACD2EPR
Recruitment
Does the actor
recruit from a
particular
ethnic or
religious
group?
Ordinal 0 = no
1 = members of the group
recruited by both actor and
government
2 = yes, recruited by the actor
recruitment Adapted from
ACD2EPR
Population
support
Does the actor
have the
support of at
least 50% of
the ethnic or
religious
group’s
population?
Ordinal 0 = no
1 = divided support
2 = yes
popsupport Adapted from
ACD2EPR
Political
party
Does the
actor/group
have a political
party?
Ordinal 0 = no
1 = yes, but loosely or
unofficially
2 = yes, formal political party
polparty Same rule as
above. Only
code this
variable if
id_based = 1.
Status Does the actor
have access to
state power at
the executive
level?
Ordinal 0 = no
1 = excluded
2 = shares power
3 = group rules alone
status Adapted from
EPR Core
Regional
Autonomy
Does the actor
have access to
regional
Dichotomous 0 = no
1 = yes
reg_aut Adapted from
EPR Core
185
autonomy and
political power
at the regional
level?
Resources
UCDP Dashboard: UCDP has an interactive dashboard where you can search for qualitative
information about actors, countries, and conflicts. I would recommend starting here to find
information for coding, as they will have information on all the actors in the dataset.
Mapping Militants Project (MMP): Stanford’s Center for International Security and Cooperation
(CISAC) maintains a great resource that profiles militant groups, their organization, strategy,
involvement in violence, etc. Though they do not cover all (or even most) of the actors we are
researching, they do have extremely helpful profiles that can help in coding. It would be worth a
quick look to see if any actor you are coding has a profile on MMP.
International Crisis Group CrisisWatch: The International Crisis Group regularly publishes reports
on ongoing violent conflicts around the world. Their reports and infographics tend to be very
informative and detailed regarding the different actors in conflicts.
UMD’s Global Terrorism Database (GTD): The University of Maryland’s National Consortium for
the Study of Terrorism and Responses to Terrorism (START) collects and updates information on
terrorist/extremist groups, some of which are listed as actors in the data that we are coding. You can
search for specific actors on their website (the GTD portal, linked above), and search by
“Perpetrator.”
186
3.3: Supplementary Analyses
Ordinal Logit, H 2
Dependent variable:
Severity
logistic
(1) (2)
Identity Coherence 0.045 -0.225
(0.361) (0.575)
Elite Support 0.617 4.074
***
(0.652) (1.523)
Group Size
-2.127
(1.944)
Constant -0.257 -1.946
***
(0.279) (0.625)
Observations 120 72
Log Likelihood -82.624 -43.551
Akaike Inf. Crit. 171.249 95.102
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
187
Analyzing Regime Type in IBPV
Dependent variable:
Severity (Total Deaths and Outcome Type)
negative
logistic
binomial
(1) (2) (3) (4)
Identity Coherence 0.868
***
1.044
***
0.641 0.524
(0.303) (0.302) (0.549) (0.563)
Elite Support 0.157 -0.303 1.201 -0.106
(0.849) (0.887) (1.513) (1.728)
Regime Type (P5) -0.120
***
-0.193
***
-0.091
**
-0.235
**
(0.025) (0.048) (0.046) (0.100)
Group Size -0.113 -0.057 1.200 1.373
(1.074) (1.054) (2.015) (2.063)
Elite Support*Regime Type
0.275
*
0.484
*
(0.143)
(0.281)
Constant 6.445
***
6.443
***
0.259 0.558
(0.313) (0.312) (0.550) (0.626)
Observations 79 79 79 79
Log Likelihood -577.503 -575.461 -46.770 -45.220
theta 0.645
***
(0.087) 0.670
***
(0.091)
Akaike Inf. Crit. 1,165.006 1,162.923 103.540 102.440
Note:
*
p<0.1;
**
p<0.05;
***
p<0.01
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
The strategy of firms towards political elites and consumers in polarized western democracies
PDF
Political identity and political participation: China's post-80's generation
PDF
Authoritarian religion: explaining when and why authoritarian governments instrumentalize religion
PDF
A theory of status-quo terrorism: democracies in conflict and their proclivity to outsource repression
PDF
The changing state of tolerance and tolerance in changing states
PDF
Flinging the boomerang: locating instability and the threat potential of identity-bias in US national security policy
PDF
The practice of everyday politics: lifestyle and identity as radical activism
PDF
Politics behind appointments: policy, patronage, and party systems in the Andes
PDF
Voices unheard, stories untold: Black women, police violence and political participation
PDF
The erosion of democracy in the post–Cold War world: understanding the phenomenon and its causes
PDF
Contemporary art and "post-black" identity politics
PDF
Neo-liberal hegemony and re-construction of Islamic community discourses in Turkey: the cases of the Gülen and Erenköy communities
PDF
The politics of transformation: mass media and the Northern Ireland peace process
PDF
Social policy and presidential ideology in Latin America: the political economy of social spending and anti-poverty programs
PDF
The political representation of Kurdish, Kemalist, and conservative Muslim women in Turkey (1990-2010)
PDF
Window Seat: examining public space, politics, and social identity through urban public transportation
PDF
Essays on political economy and corruption
PDF
The role of local actors in international development: how incentive types affect foreign aid effectiveness
PDF
Identity and regional integration: case studies on central Asia and southeast Asia
PDF
Economic policy-making in a changing environment: the politics of success in Peru vis-à-vis China
Asset Metadata
Creator
Cueva Egan, Sarah
(filename)
Core Title
Identity salience and political violence: centering social identity in the study of intrastate conflict
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Political Science and International Relations
Degree Conferral Date
2022-12
Publication Date
11/04/2022
Defense Date
10/17/2022
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
identity as a variable,OAI-PMH Harvest,political attitudes,political violence,social identity
Format
theses
(aat)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Brand, Laurie (
committee chair
), Agius-Vallejo, Jody (
committee member
), Lo, James (
committee member
), Munck, Gerardo (
committee member
), Schwartz, Stephanie (
committee member
)
Creator Email
sarahcuev@gmail.com,scueva@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-oUC112305794
Unique identifier
UC112305794
Identifier
etd-CuevaEganS-11304.pdf (filename)
Legacy Identifier
etd-CuevaEganS-11304
Document Type
Dissertation
Format
theses (aat)
Rights
Cueva Egan, Sarah
Internet Media Type
application/pdf
Type
texts
Source
20221107-usctheses-batch-990
(batch),
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Repository Email
cisadmin@lib.usc.edu
Tags
identity as a variable
political attitudes
political violence
social identity