Content uploaded by Bradford H. Morrison
Author content
All content in this area was uploaded by Bradford H. Morrison on Nov 28, 2024
Content may be subject to copyright.
THE RELATIONSHIP BETWEEN POLITICAL LEADERS’ INTEGRATIVE
COMPLEXITY AND THE USE OF VIOLENCE IN INTERNATIONAL CRISES
by
Bradford H. Morrison
B.A. Honours, University of British Columbia, 2009
M.A., McGill University, 2013
A THESIS SUBMITTED IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
in
The Faculty of Graduate and Postdoctoral Studies
(Psychology)
THE UNIVERSITY OF BRITISH COLUMBIA
(Vancouver)
August 2024
© Bradford H. Morrison, 2024
ii
The following individuals certify that they have read, and recommend to the Faculty of Graduate
and Postdoctoral studies for acceptance, the dissertation entitled:
The Relationship Between Political Leaders’ Integrative Complexity and the Use of Violence in
International Crises
submitted by Bradford H. Morrison in partial fulfillment of the requirements for
the degree of Doctor of Philosophy
in Psychology
Examining Committee:
Dr. Peter Suedfeld, Professor Emeritus, Psychology, UBC
Supervisor
Dr. Kristin Laurin, Associate Professor, Psychology, UBC
Co-supervisor
Dr. Paul Quirk, Professor, Political Science, UBC
Supervisory Committee Member
Dr. Alan Jacobs, Professor, Political Science , UBC
University Examiner
Dr. Steven Lee, Associate Professor, History , UBC
University Examiner
Dr. David Winter, Professor Emeritus, Psychology, University of Michigan
External Examiner
Additional Supervisory Committee Members:
Dr. Frances Chen, Professor, Psychology, UBC
Supervisory Committee Member
iii
Abstract
Political leaders are responsible for making decisions that can lead to, or avert,
international violence and war. The goal of this dissertation is to better understand how the
cognitive complexity, specifically integrative complexity, of political leaders relates to processes
by which international crises or confrontations become violent. It tests two hypotheses that point
in opposing directions. The strategic judgment hypothesis posits that leaders with high
integrative complexity are better able to solve coordination problems, and therefore to avoid
violence. The demonstration-of-resolve hypothesis posits that leaders with low integrative
complexity are better able to deter aggression, and therefore to avoid violence. This investigation
tests these hypotheses by scoring Auto IC, an automated measure of integrative complexity, from
a large corpus of texts from American, British, and Russian/Soviet heads of government from the
19th to 21st centuries. It uses regressions to test whether, controlling for multiple variables, the
integrative complexity of leaders, prior to and during international confrontations and crises, is a
predictor of variables relevant to the use of violence. These include the initiation of violence,
degree of reliance on violence, level of hostility, number of fatalities suffered, and degree of
success achieved at the resolution of the crisis. Chapter 6 does this using the data in the
Militarized Interstate Confrontations (MIC) dataset. Chapter 7 does this using the data in the
International Crisis Behavior (ICB) dataset. The results show that lower integrative complexity is
associated with greater reliance on violence, hostility, and number of fatalities suffered, giving
more support to the strategic judgment hypothesis. The initial results concerning initiation of
violence and success achieved, while not statistically significant, suggested additional
exploratory analyses, and modifications to the hypotheses. Theoretical and methodological
contributions, limitations, and opportunities for future research are discussed.
iv
Lay Summary
International violence and war cause immense destruction, suffering, and death. So why
do political leaders make decisions that lead to violence? Specifically, I test how the complexity
of thinking of political leaders relates to whether international crises and confrontations escalate
to become violent. Do leaders that think with higher complexity solve problems better, enabling
them to avoid violence? Or do leaders that think with lower complexity dissuade adversaries
from attacking, enabling them to avoid violence? The results show that lower complexity
thinking is associated with greater hostility, reliance on violence, and fatalities. Although there
are important qualifications, this gives more support to the idea that leaders with higher
complexity solve problems better, enabling them to avoid violence.
v
Preface
I am the primary contributor of the work in this dissertation. I am responsible for
designing the research, doing or directing most of the work collecting texts, writing the
dictionaries to code topics relevant to foreign policy and foreign countries, coding who initiated
violence against whom in the international crises, performing the statistical analyses, and writing
the manuscript.
I received important intellectual and research advice, and editing advice, from my
supervisors and committee members.
The collection of texts from several of the Russian/Soviet leaders was carried out in
collaboration with a colleague, Zlatin Mitkov.
A substantial portion of the texts from Vladimir Putin were collected as part of a project
of Dr. Peter Suedfeld’s lab, which was funded by the Strategic Multilayer Assessment (SMA).
That research has been published as a white paper for the SMA. I was involved in directing the
text collection for that project.
The coding of Auto IC was performed by the Auto IC, through the Auto IC website, on
texts that I provided to it for that purpose.
None of the research presented in this dissertation involved performing experimental
procedures on, or otherwise accessing human subjects, or the use of non-public materials
concerning human subjects, and therefore Behavioral Research Board approval was not required.
I intend to publish this research, or a future iteration of it, but it has not yet been
submitted to a journal or publisher.
vi
Table of Contents
Abstract ..................................................................................................................................... iii
Lay Summary ............................................................................................................................. iv
Preface ........................................................................................................................................ v
Table of Contents ...................................................................................................................... vi
List of Tables .............................................................................................................................. xi
Acknowledgements .................................................................................................................. xii
Chapter 1: Introduction .................................................................................................................. 1
1.1 Dissertation Overview .......................................................................................................... 6
Chapter 2: Integrative Complexity .................................................................................................. 8
2.1 Scoring Integrative Complexity from Verbal Materials ........................................................ 9
2.2 Correlates of Integrative Complexity that are of Central Importance to the Hypotheses . 10
2.2.1 Cognitive Ability ........................................................................................................... 10
2.2.2 Closed-Mindedness ..................................................................................................... 12
2.3 Other Aspects of Integrative Complexity to be Aware of ................................................... 15
2.3.1 Internal cognition versus impression management .................................................... 15
2.3.2 The state versus trait components of integrative complexity ..................................... 17
2.3.3 Stress ........................................................................................................................... 20
2.4 The Cognitive Manager Model ........................................................................................... 23
2.5 Integrative Complexity, Political Decision-Making, and Political Violence ......................... 25
2.5.1 Integrative Complexity and Political Decision Making ................................................. 25
2.5.2 Integrative Complexity and Violence ........................................................................... 28
2.5.3 Limitations of Previous Research Concerning the Relationship Between Integrative
Complexity and Violence ...................................................................................................... 30
Chapter 3: Two Classes of Problem that Can Lead to Political Violence ...................................... 33
3.1 Coordination Problems ....................................................................................................... 33
3.2 Problems of Uncertain Resolve .......................................................................................... 36
3.3 Psychological characteristics may function to address these classes of problem .............. 43
3.3.1 Honor ........................................................................................................................... 43
3.3.2 Religion and sacred values ........................................................................................... 46
3.3.4 Psychological attachment to territory ......................................................................... 47
3.3.5 Integrative complexity? ............................................................................................... 49
Chapter 4: My Hypotheses ........................................................................................................... 51
vii
4.1 The Strategic Judgment Hypothesis ................................................................................... 51
4.2 The Demonstration-of-Resolve Hypothesis ........................................................................ 53
Chapter 5: Method ....................................................................................................................... 56
5.1 Scoring Integrative Complexity Using Auto IC .................................................................... 56
5.2 The Sample of Countries, Leaders, and Verbal Materials ................................................... 58
5.2.1 The sample of countries .............................................................................................. 58
5.2.2 The sample of leaders .................................................................................................. 58
5.2.3 The sample of Verbal Materials ................................................................................... 60
5.3 Preparing and Cleaning Paragraphs .................................................................................... 62
5.4 Coding Texts for the Topic of Foreign Policy ...................................................................... 63
5.5 The Datasets of Confrontations and Crises ........................................................................ 64
5.6 The Analytical Approach – Regressions with Controls ........................................................ 66
Chapter 6: Leaders’ Integrative Complexity and Militarized Interstate Confrontations .............. 69
6.1 Coding of the Militarized Interstate Confrontation (MIC) Variables .................................. 69
6.1.1 Country ........................................................................................................................ 69
6.1.2 Time period .................................................................................................................. 69
6.1.3 During world war ......................................................................................................... 70
6.1.4 The side that initiated the first military action ............................................................ 70
6.1.5 The level of hostility employed by the leader’s country .............................................. 71
6.1.6 Fatalities (in thousands) suffered by the leader’s country .......................................... 71
6.1.7 Degree of Success at Outcome .................................................................................... 71
6.1.8 Avoidance of Defeat dummy ....................................................................................... 72
6.2 Does a Leader’s Integrative Complexity Before the Confrontation Predict Whether Their
Side Initiated the First Militarized Activity? .............................................................................. 73
6.2.1 Predictions ................................................................................................................... 73
6.2.2 Results ......................................................................................................................... 74
6.2.3 Discussion .................................................................................................................... 77
6.3 Does a Leader’s IC During a Confrontation Predict the Level of Hostility Employed by Their
Country? ................................................................................................................................... 78
6.3.1 Predictions ................................................................................................................... 78
6.3.2 Results ......................................................................................................................... 78
6.3.3 Discussion .................................................................................................................... 81
viii
6.4 Does a Leader’s IC During a Confrontation Predict the Number of Fatalities Suffered by
Their Country? .......................................................................................................................... 81
6.4.1 Predictions ................................................................................................................... 81
6.4.2 Results ......................................................................................................................... 82
6.4.3 Discussion .................................................................................................................... 84
6.5 Does a Leader’s IC During a Confrontation Predict the Degree of Success Achieved by the
Leader’s Country at the Outcome? .......................................................................................... 84
6.5.1 Predictions ................................................................................................................... 84
6.5.2 Results ......................................................................................................................... 85
6.5.3 Discussion .................................................................................................................... 90
Chapter 7: Leaders’ Integrative Complexity and International Crisis Behavior ............................ 93
7.1 Coding of the International Crisis Behavior (ICB) Variables ................................................ 93
7.1.1 Country ........................................................................................................................ 93
7.1.2 Time period .................................................................................................................. 93
7.1.3 During world war ......................................................................................................... 94
7.1.4 Protracted conflict ....................................................................................................... 94
7.1.5 Distance ....................................................................................................................... 94
7.1.6 Gravity of threat .......................................................................................................... 95
7.1.7 Power discrepancy ....................................................................................................... 96
7.1.8 The actor who initiated the first violence, and the actor against whom violence was
first initiated ......................................................................................................................... 97
7.1.9 Centrality of violence ................................................................................................... 99
7.1.10 Degree of Success .................................................................................................... 100
7.1.11 Avoidance of Defeat ................................................................................................ 100
7.1.12 Satisfaction of the leader’s country with the outcome ........................................... 100
7.1.13 Satisfaction of the adversary with the outcome ..................................................... 101
7.2 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Was the Target of Initial Violence? ......................................................................................... 101
7.2.1 Predictions ................................................................................................................. 101
7.2.2 Results ....................................................................................................................... 102
7.2.3 Discussion .................................................................................................................. 105
7.3 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Initiated the First Violence? .................................................................................................... 105
7.3.1 Predictions ................................................................................................................. 105
ix
7.3.2 Results ....................................................................................................................... 106
7.3.3 Discussion .................................................................................................................. 108
7.4 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Uses Nonstate Proxies to Initiate Violence? ........................................................................... 108
7.4.1 Predictions ................................................................................................................. 108
7.4.2 Results ....................................................................................................................... 109
7.4.3 Discussion .................................................................................................................. 112
7.5 Does a Leader’s Integrative Complexity During a Crisis Predict the Degree That Their
Country Relies on Violence Relative to Nonviolent Strategies? ............................................. 113
7.5.1 Predictions ................................................................................................................. 113
7.5.2 Results ....................................................................................................................... 114
7.5.3 Discussion .................................................................................................................. 117
7.6 Does a Leader’s Integrative Complexity During a Crisis Predict the Degree of Success
Achieved by the Leader’s Country at the Outcome? .............................................................. 118
7.6.1 Predictions ................................................................................................................. 118
7.6.2 Results ....................................................................................................................... 119
7.6.3 Discussion .................................................................................................................. 125
7.7 Does a Leader’s Integrative Complexity During a Crisis Predict the Satisfaction of the
Adversary at the Outcome, Statistically Controlling for the Degree of Success Achieved by the
Leader’s Country? ................................................................................................................... 126
7.7.1 Predictions ................................................................................................................. 126
7.7.2 Results ....................................................................................................................... 127
7.7.3 Discussion .................................................................................................................. 131
Chapter 8: General Discussion .................................................................................................... 134
8.1 Summary and Interpretation of the Results ..................................................................... 134
8.1.1 Level of hostility, reliance on violence, and fatalities ................................................ 134
8.1.2 The initiation of violence ........................................................................................... 134
8.1.3 Degree of success achieved at the resolution of the crisis ........................................ 136
8.2 Theoretical Contributions ................................................................................................. 137
8.2.1 A new theoretical framework .................................................................................... 137
8.2.2 The empirical results mainly support the strategic judgment hypothesis ................. 138
8.3 Methodological Contributions .......................................................................................... 140
8.3.1 A new corpus ............................................................................................................. 140
8.3.2 A new dataset ............................................................................................................ 140
x
8.3.3 An approach that integrates data from different domains of research .................... 140
8.4 Limitations ........................................................................................................................ 140
8.5 Opportunities for Future Research ................................................................................... 142
8.5.1 Add more countries to the dataset ........................................................................... 142
8.5.2 Distinguish between the state and trait components of integrative complexity ...... 142
8.5.3 Do qualitative investigations of crises ....................................................................... 143
8.5.4 Conduct experiments ................................................................................................ 144
References .................................................................................................................................. 146
Appendix A – American Presidency Project Exclusions .............................................................. 160
Appendix B – Dictionaries for LIWC to Code Foreign Policy and Countries ................................ 161
xi
List of Tables
Table 1 Regressions predicting whether violence was initiated by the leader’s side .................. 76
Table 2 Regressions predicting the level of hostility of the leader’s country ............................... 80
Table 3 Regressions predicting the number of fatalities suffered by the leader’s country .......... 83
Table 4 Distribution of values for the degree of success and avoidance of defeat ..................... 86
Table 5 Regressions predicting the degree that the leader’s country achieved success .............. 88
Table 6 Regressions predicting whether the leader’s country avoided defeat ............................ 90
Table 7 Regressions predicting whether the initial violence targeted the leader’s country ...... 104
Table 8 Regressions predicting whether violence was initiated by the leader’s side ................ 107
Table 9 Regressions predicting whether violence was initiated by a nonstate proxy of the
leader’s country .................................................................................................................. 111
Table 10 Regressions predicting the degree that the leader’s country relied on violence relative
to nonviolent strategies ...................................................................................................... 116
Table 11 Distribution of values for the degree of success and avoidance of defeat ................. 120
Table 12 Regressions predicting the degree that the leader’s country achieved success ......... 122
Table 13 Regressions predicting whether the leader’s country avoided defeat ........................ 124
Table 14 Regressions predicting adversary satisfaction, controlling for the leader’s success ... 130
xii
Acknowledgements
I am grateful to everyone who has given me help and support throughout this process. I
am more fortunate in my family, friends, and colleagues than I know how to express. I have been
given more support than I am able to list, and probably more than I am aware of, so anything I
write here will be inadequate. I have restricted the following acknowledgements to academic and
professional support, because an attempt to describe personal and emotional support would never
end. If you have given me your support, even if it is not explicitly mentioned here, thank you.
I am grateful to my advisor, Dr. Peter Suedfeld, for introducing me to the idea that the
psychology of political leaders could be rigorously studied using content analysis, for
introducing me to research on cognitive complexity, for giving me the opportunity to do applied
research concerning the psychology of political leaders, and for his patience and support. I am
grateful to my co-advisor, Dr. Kristin Laurin, for suggesting statistical approaches, for her
thoughts concerning how my research might relate to a broad range of recent research in social
psychology, for reminding me of bureaucratic requirements of the dissertation process, and for
pointing out when my pace was too slow to meet my timeline. I am grateful to all of my
committee members, both supervisory and external, who have been exceedingly generous in
giving questions, comments, and suggestions on how to improve the work.
I am grateful to those research assistants in Dr. Suedfeld’s lab who did work for this
research project: Jeranuhi Alakhverdiants, Borna Atrchian, Anastasiia Averianova, Aisha
Begimbetova, Aya Cottle, Jessica Flores, Valeriia Laut, Lisa Li, and Cole Wong. This research
would not have been possible without their work. I am also grateful to those research assistants
who worked on my other research projects – I look forward to listing your names when those
papers are published.
xiii
I am grateful to my colleague Zlatin Mitkov, who helped with the collection of verbal
materials for the Russian leaders, and with whom I am building a broader corpus and dataset.
And I am grateful to his research assistants who helped him with those tasks.
I am grateful to Lucian Gideon Conway, III, and to Kathrene Conway, for making Auto
IC available to me. And a special thank you to Kathrene for taking the time to process the many
files that I sent her.
I am grateful to Carol Louie, for writing and running the web scraper. Without it, this
research would not be possible.
I am grateful to all of my cohort in the psychology department. I give an especial thank
you to: Audrey Aday, Rachele Benjamin, Patrick Dubois, Ryan Dwyer, Holly Engstrom, Yvette
Graveline, Gordon Heltzel, Lia Kendall, Anita Schmalor, Cindel White, and Zak Witkower.
And, to my former lab mates: Irina Della-Rossa, Lindsy Grunert, Lisa Shiozaki, Amir Taheri,
and Ryan Cross (and to his company, CrossTech).
More broadly, I am grateful to those professors who, throughout my development as a
scholar, have inspired my interest in the political psychology of crisis and conflict, even if they
did not know it. This includes Drs. Alan Jacobs, Benjamin Nyblade, Yves Tiberghien, Philip
Resnick, Michael Brecher, Hudson Meadwell, T. V. Paul, and Stephen Saideman.
I am also grateful to the people at the Strategic Multilevel Assessment (SMA), including
Dr. Hriar Cabayan, and at NSI, including Dr. Larry Kuznar, for giving me the opportunity to do
applied work researching the psychology of political leaders, and to meet many interesting
people doing fascinating research.
1
Chapter 1: Introduction
This is a study of how the cognitive complexity, specifically the integrative complexity,
of the heads of government of the preeminent great power countries of the modern international
system, namely the United States, Soviet Union/Russia, and the United Kingdom, relates to their
use of violence in international crises and confrontations. This study tests two hypotheses, which
are not mutually exclusive because they operate through different causal pathways, but that tend
to point in opposite directions. The strategic judgment hypothesis, drawing from coordination
games such as the prisoner’s dilemma, posits that leaders with higher integrative complexity will
be less likely to employ violence, because they are better able to solve international coordination
problems, such that they can achieve their goals without resorting to violence. The
demonstration-of-resolve hypothesis, drawing from deterrence-theory and games of chicken,
posits that leaders with lower integrative complexity will be less likely to be the target of violent
acts, because they are better able to communicate resolve to other political actors, deterring them
from aggressing against the leader’s country. These hypotheses were used to generate specific
predictions concerning how political leaders’ integrative complexity relates to multiple variables,
including the initiation of violence, degree of reliance on violence, and the number of fatalities
suffered.
The study tests how consistent the predictions of these hypotheses are with empirical data
using large N, real-world data. It measures the real-world cognitive complexity of political
leaders by coding the integrative complexity of most of the heads of government of the United
States, Soviet Union/ Russia, and the United Kingdom, during the modern international system
2
of the 19th and 20th centuries (the details on dates and the leaders included are given in the
Method chapter).
The study uses variables relevant to international crises and confrontations, including the
initiation of violence, severity of violence, and number of fatalities, and variables concerning the
situational features of crises/confrontations, such as the time period and the relative power of the
countries involved. These variables are coded from data in two existing international relations
datasets, namely the International Crisis Behavior dataset (ICB), and the Militarized Interstate
Confrontations dataset (MIC) (Gibler & Miller, 2023; ICB Project - International Crisis
Behavior, n.d.).
These existing datasets are both attempts to include, as nearly as is practically possible,
all of the recorded international crises/confrontations in the modern international system
(although they use different ranges of dates, which are, respectively, post WWI, and post War of
1812). Similarly, I have attempted to code the integrative complexity, as nearly as is practically
possible, of all of the heads of government of the paramount great power countries of the modern
international system. Therefore, this study is a close approximation of the universe of cases that
are practically available to researchers. As a result, while the theoretical or causal interpretation
of results are up for dispute (as is always the case with correlational research), any empirical
relationships that are uncovered are highly likely to be representative of the real-world
relationships, because the study uses such a close approximation of the universe of cases.
The results, while not always giving answers in clear support of one hypothesis over the
other, were more consistent with the strategic judgment hypothesis than with the demonstration-
of-resolve hypothesis. In particular, the strategic judgment hypothesis was supported by the
findings that a leader’s integrative complexity during a crisis/confrontation: (a) is negatively
3
associated with the degree that their country relied on violence as a means to deal with both
international crises and confrontations; (b) is negatively associated with fatalities suffered by
their country during confrontations; and, (c) after controlling for the degree of success achieved
by the leader’s country at the resolution of an international crisis, is positively associated with
the post-crisis satisfaction of the adversary country.
Although, as a correlational study, causality cannot be determined with certainty, this
pattern of results is consistent with the theoretical model that, among the heads of government of
the paramount great powers of the international system, integrative complexity is associated with
a greater ability to find solutions to international coordination problems, and by doing so, to
avoid costly violence, and to achieve successful resolution, in ways that are relatively more
likely to be satisfactory to the adversaries.
Not all of the results were consistent with the strategic judgment hypothesis. For instance,
they were not with respect to the relationship of the leader’s integrative complexity with the
initiation of violence, or with achieving success at the resolution of the crisis. However, these
other results, while interesting and suggestive of possible modifications to the hypotheses, which
I discuss in their substantive chapters, were not statistically significant, so I do not discuss them
further in the introduction.
As well as the above substantive contributions, this study makes an important theoretical
and methodological contribution. It does so by incorporating a psychological variable, the
leader’s integrative complexity, into models of foreign policy interactions, namely prisoner’s
dilemma games and games of chicken, and tests the resulting models using large N real-world
data, controlling for multiple variables from foreign policy and international relations theory.
4
There are great advantages to this particular combination of theory and method. In terms
of theory, because there can be ironic effects of foreign policy interactions, it is possible that
psychological characteristics that make a leader more likely to be willing to use violence would
result in their country being involved in less violence. For instance, if a country is a global or
regional hegemon, then the willingness of the leader to use violence to punish aggression, and to
maintain the global or regional system, might result in overall lower rates of violence in the
system (see Gilpin, 1981; Ikenberry et al., 2009; Keohane, 1984; Krasner, 1976; Wohlforth,
1999). Or, to give another example, the willingness of a leader to use violence might deter
aggression against their country, resulting in their country being involved in fewer violent
confrontations (see Achen & Snidal, 1989; Huth, 1999; Schelling, 1980).
In terms of methods, if small N studies or case studies are used, and cases are sampled
because they are interesting, or impactful, or involved violence, or deterrence success or failure,
then the results might show different patterns from the real general relationship (Huth & Russett,
1990). To put theory and method together, and to give a more concrete example, even though
case studies have found that when political actors use violence, this is associated with low
integrative complexity, because these studies use a small n of notable cases, they cannot be used
to conclude that political leaders who exhibit low integrative complexity are more likely to be
involved in violence.
Alternatively, it may be the case that political leaders can, by exhibiting low complexity,
deter adversaries from escalating a conflict to violence, or it may be that the actual cases of
violence deterred later conflicts from escalating to violence (thereby reducing the overall
frequency of violence). Or, it may be that the cases that researchers sample as being
psychologically interesting cases of the use of violence, happen to be those cases in which the
5
leader’s integrative complexity happens to be low. In order to identify how the integrative
complexity of political leaders works, through foreign policy interactions and the international
system, to influence the frequency and severity of violence, in theorizing it is valuable to include
both the leader’s integrative complexity and foreign policy interaction into models. Similarly, in
empirical tests it is valuable to use a large N representative sample of real-world international
crises/conflicts.
The current study does this. The theoretical models of the study, the strategic judgment
hypothesis and the demonstration-of-resolve hypothesis, incorporate both the individual-level
psychological variable of integrative complexity, and the level of foreign policy interaction. The
empirical tests in this study use a large N of cases, approximating the universe of practically
accessible real-world cases of international crises/conflicts that involved the preeminent great
power countries of the modern era. The tests statistically control for multiple variables that
international relations theories suggest are relevant, so that they test how consistent the models
are with real-world behavior within the international system.
The current study does this by using all of the crises in the International Crisis Behavior
(ICB) dataset (ICB Project - International Crisis Behavior, n.d.), and all of the confrontations in
the Militarized Interstate Confrontations (MIC) dataset (Gibler & Miller, 2023), that involve the
United States, the Soviet Union, or the United Kingdom as primary actors.
An important feature of the ICB dataset is that it defines an international crisis as: a
disruptive change in the interactions between countries that involves a heightened probability of
military hostilities, and that destabilizes their relationship and the international system (Brecher,
2008; Brecher & Wilkenfeld, 1997). Although militarized confrontations are conceptually
distinct from crises, the MIC dataset similarly defines them as involving the threat of military
6
hostilities (Gibler & Miller, 2023). This is important because the presence of violence is not a
defining feature of international crises or militarized international confrontations, but is more
likely to occur during these crises/confrontations, such that they present an opportunity to test the
relationship between leaders’ integrative complexity and international violence.
The study does this by using the integrative complexity of the leaders of these countries
as a statistical predictor of the initiation of violence, degree of reliance upon violence, and
intensity of violence in international crises, while controlling for country-level and system level
variables. The study tests two hypotheses, with effects in opposing directions. The first is the
“strategic judgment hypothesis,” that leaders with high complexity are less likely to be drawn
into violence because they have the strategic judgment to get what they want without resorting to
costly violence. The second is the “demonstration-of-resolve hypothesis,” that leaders with low
integrative complexity are less likely to be drawn into violence because they can better
demonstrate resolve, enabling them to deter aggression.
1.1 Dissertation Overview
The remainder of this paper is organized in six chapters. Chapter 2 describes integrative
complexity, and how it has been used to study the psychology of political decision-making and
political violence. Chapter 3 introduces two classes of problem that can lead to political violence
and war, namely coordination problems and problems of uncertain resolve, and suggests that
different levels of integrative complexity could be appropriate to addressing each of these.
Chapter 4 introduces my hypotheses, by using the games of prisoner’s dilemma and chicken to
generate hypotheses concerning how a political leader’s integrative complexity affects the
likelihood that their country will be drawn into political violence, whether as initiator or target,
and the severity of that violence. Chapter 5 describes the method used in detail. This includes
7
describing the MIC and ICB datasets, the advantages of using crises/confrontations rather than
wars, the collection of verbal materials attributed to political leaders, how these verbal materials
were scored for integrative complexity and for whether the topic of the text was relevant to
foreign policy. It also describes the statistical approach, statistical tests, and choice of controls.
Chapters 6 and 7 describe the various statistical tests using the MIC and ICB data, respectively,
including the coding of variables, the statistical tests, results, and discussions. Chapter 8 is the
overall discussion, including the novel contributions of this study, its limitations, and avenues for
future research.
8
Chapter 2: Integrative Complexity
Integrative complexity (IC) is a psychological characteristic, and more precisely is a type
of cognitive complexity, i.e., the complexity of the cognitive structure that underlies a person’s
information processing and decision making (Suedfeld, 1992, 2010).
Integrative complexity is defined as a state variable (Suedfeld, 2010). It is derived from a
predecessor variable called “conceptual complexity,” which is defined as a trait variable
(Schroder et al., 1967). Like most state variables, integrative complexity has a trait component,
i.e., across a range of situations some persons could tend to have higher integrative complexity
than others.
Integrative complexity is scored on an ordinal scale from 1 to 7, with the lowest scores
indicating the least cognitive complexity, and higher scores indicating greater cognitive
complexity (Baker-Brown et al., 1990). Specifically, as scores increase from 1 to 3 it indicates
increasing recognition of multiple perspectives or dimensions of an issue, and as scores increase
from 3 to 7 it indicates increasing integration of those perspectives or dimensions.
Previous research has found two correlates of integrative complexity that are particularly
relevant to my hypotheses: It is positively correlated with some measures of cognitive ability
(Suedfeld & Coren, 1992), and is negatively correlated with closed-mindedness (Suedfeld,
2010). I will address both of these in further detail later in this section.
Previous research has also found correlates of integrative complexity that, while not of
central importance to my hypotheses, are worth being aware of: It is negatively correlated with
high levels of stress (Suedfeld, 2010); and it is positively correlated with accountability to
persons who might disagree with one’s position (Tetlock, 1983). These suggest two broader
issues to be aware of, namely the distinction between the state and trait components of
9
integrative complexity, and the distinction between internal cognition and impression
management. While these issues are not central to my hypotheses, they are relevant as potential
confounds to control for, and as opportunities for future research concerning the nature of the
causal pathways being studied. I will address these issues in further detail later in this section.
2.1 Scoring Integrative Complexity from Verbal Materials
In the present study, integrative complexity is scored using an automated system, namely
Auto IC, which I describe below, but first I give the background of the manually scored version
of the variable. Integrative complexity has traditionally been hand scored, from verbal materials,
such as writings, interviews, or speeches, produced by the subject/participant, by trained scorers
who follow the rules in a detailed scoring manual (Baker-Brown et al., 1990), and who have
passed a reliability test (Suedfeld, 2010).
Integrative complexity is scored on a scale from 1 to 7. The even-numbered scores
indicate levels that are transitional between the odd-numbered scores, so I only give descriptions
of the latter. Scores from 1 to 3 are distinguished by increasing degrees of conceptual
differentiation concerning the topic, such as recognition of more than one legitimate perspective
or more than one relevant dimension. A score of 1 indicates no differentiation, while a score of 3
indicates full differentiation. Scores from 3 to 7 are distinguished by increasing degrees of
integration across the differentiated perspectives or dimensions, which, for instance, can take the
form of recognition that they interact, or are part of a system, or produce emergent features
(Baker-Brown et al., 1990). A score of 3 indicates no integration (despite the full differentiation),
5 indicates full integration, and 7 indicates full integration at more than one level, i.e., a second
order integration (Baker-Brown et al., 1990).
10
Unfortunately, the traditional approach of manual scoring has the disadvantage that it is
time-consuming, and cannot practically be scaled up to more than a few thousand documents. To
address this problem, researchers have developed automated (i.e., computerized) scoring
systems, allowing rapid scoring of large corpora. One such automated scoring system, Auto IC,
is designed to mimic manually scored integrative complexity by distinguishing between
differentiation and integration, achieves satisfactory reliability with manually scored integrative
complexity (Conway et al., 2014; Houck et al., 2014), and has been validated in the political
domain (Conway et al., 2016, 2020). Because Auto IC has these advantages, as well as the
ability to rapidly score large corpora, I use it to score integrative complexity for the current
project.
2.2 Correlates of Integrative Complexity that are of Central Importance to the Hypotheses
There are two variables that previous research has found are correlated with integrative
complexity, and that are particularly relevant to the hypotheses tested by this dissertation. These
variables are cognitive ability and closed-mindedness (including extremism).
2.2.1 Cognitive Ability
The strategic judgment hypothesis assumes that higher integrative complexity is
associated with greater cognitive ability, and that leaders can use this cognitive ability to achieve
their goals, with less use of costly violence. The hypothesis therefore requires that integrative
complexity be positively associated with some measure of intelligence or cognitive ability.
Previous research has found evidence for this positive association. A study of how
conceptual complexity (the trait-level variable from which integrative complexity was
developed) relates to various measures of cognitive ability found moderate positive correlations
for several variables (Suedfeld & Coren, 1992). In particular, there were significant positive
11
correlations between integrative complexity and three measures of divergent thinking, the CAB-
F Test (r = .11), the CAB-O Test (r = .19), and the Alternative Uses Test (r = .28). There were
also weak, but significant, positive correlations between integrative complexity and a broad
measure of verbal ability, the Quick Test of verbal comprehension, and a broad measure of
crystalized intelligence, the Wonderlic Personnel Test. However, there were no significant
relationships between integrative complexity and a measure of fluid intelligence, the Figure
Classification Test, or a narrow test of verbal ability, a grammatical test, or a narrow test of
crystalized intelligence, a spelling test.
This pattern of results suggests that conceptual/integrative complexity is not a measure of
intelligence, as such, but that it is consistently positively correlated with divergent thinking, i.e.
thinking outside of the box. This is consistent with the idea that, when facing a challenging
situation, persons with low integrative complexity could struggle to identify relevant variables
(or interactions across variables) if they are not a part of the person’s usual mental model. On the
other side of this coin, it is consistent with persons with high integrative complexity being better
able to apply new or unusual mental models, or to identify relevant variables and interactions
that they are unfamiliar with.
Research concerning how integrative complexity relates to the real-world performance of
eminent persons suggests that there is a positive relationship between integrative complexity and
broad patterns of success. Political leaders with notably long tenures in office, such as Andrei
Gromyko and the Duke of Wellington, maintained high levels of integrative complexity during
times of high stress (Wallace & Suedfeld, 1988).
A study of political and military geniuses has also found that they more often showed
increases in integrative complexity during times of stress, especially among the military geniuses
12
(Suedfeld, 2014). By contrast, another study found that a sample of American and Soviet
politicians who did not have a record of outstanding success while in office showed a marked
decrease in integrative complexity during times of stress (Wallace & Suedfeld, 1988).
These results suggest that leaders who exhibit high integrative complexity during periods
of intense challenge or stress, are better able to cope with those challenges. Although we cannot
be certain of the causal relationship, this is consistent with integrative complexity being
associated with a cognitive ability that can be employed to solve challenging problems.
The overall weight of evidence is in support of integrative complexity being positively
associated with some kind of cognitive ability (or with the exercise of that ability), in particular
divergent thinking, and that this is broadly associated with greater degrees of success over a
person’s political career.
2.2.2 Closed-Mindedness
Closed-mindedness is relevant to both the strategic judgment hypothesis and to the
demonstration-of-resolve hypothesis. In the case of the strategic judgment hypothesis, leaders
who are more open-minded (less closed-minded) may be better able to understand the
perspective of their adversary, and how it relates to their own perspective, and find solutions that
meet the needs of both sides. (The logic about this is similar to that with respect to the cognitive
ability of divergent thinking, which is addressed in the subsection above.)
In the case of the demonstration-of-resolve hypothesis, leaders who are more closed-
minded may be better able to convince their adversaries that they will carry out their threats. This
is because closed-minded persons should be expected to adhere to a course of action once they
have decided on it, and to be less likely to understand complex reasons why the adversary does
not comply to their demands, and to be less likely to have a plan B.
13
Previous research has found that integrative complexity is negatively associated with
closed-mindedness, or to put it another way, that people with higher integrative complexity tend
to be more open-minded. Therefore, if it is true that closed-minded leaders are better able to
convince adversaries that they will carry out their threats, then it is likely that leaders who
exhibit low integrative complexity will also be better able to do so.
2.2.2.1 Extremism
I am discussing extremism as part of the topic of closed-mindedness because the
ideological extremes tend to require fewer compromises across competing values, i.e., for
extreme socialists the value of equality can consistently trump freedom, while for extreme
libertarians the value of freedom can consistently trump equality. Consistent with this, previous
research has found that cognitive closure and cognitive rigidity are associated with extreme
partisanship (Zmigrod et al., 2020) and ideological extremism (Zmigrod, 2020; Zmigrod et al.,
2021).
Persons who place themselves on political extremes (e.g., far left, far right), are lower in
complexity than persons who are politically moderate (Tetlock, 1986).The finding that persons
on the ideological extremes have lower integrative complexity than moderates has been
replicated in the context of British Columbia, where student members of the Liberal and
Progressive Conservative parties were higher in complexity than members of the New
Democratic Party (the most left-wing of the four parties), or the Social Credit Party (the most
right-wing of the four parties) (Suedfeld et al., 1994).
This pattern has also been replicated in a historical study concerning the debate
concerning slavery and abolitionism prior to the American Civil War. Moderates, who wanted to
limit slavery to the states in which it was already established and prohibit it in new states, had
14
higher complexity than did both the abolitionists and the anti-abolitionists/states’ rights
advocates (Tetlock et al., 1994).
Previous research has also found evidence that the issue being discussed, and how that
issue relates to the ideology of the person, is relevant to integrative complexity. A laboratory
study, in which participants were asked to write responses to prompts found that political liberals
and conservatives exhibited higher complexity in response to different prompts (Conway et al.,
2016). For instance, conservatives were more complex than liberals concerning the topics of the
death penalty and refugees, and liberals were more complex than conservatives concerning the
topics of alcohol and censorship.
The researchers in this study suggested that an explanatory variable is attitude strength.
The logic is that liberals and conservatives hold strong attitudes about different issues, and both
tend to exhibit lower integrative complexity concerning issues about which they hold strong
attitudes.
This is relevant to closed-mindedness because when people hold strong attitudes about an
issue, they are more likely to be closed-minded concerning that issue. The same paper, for
instance, found a similar pattern with respect to dogmatism as it did for integrative complexity
(Conway et al., 2016). This suggests that it is likely that, for participants of a given ideological
orientation (e.g., liberal or conservative), those topics that elicit low integrative complexity are
also likely to elicit high dogmatism (although the paper in question did not directly test this).
The overall weight of evidence is strongly in support of integrative complexity being
negatively associated with extremism and closed-mindedness, both at the trait level and at the
state level.
15
2.3 Other Aspects of Integrative Complexity to be Aware of
There are a number of additional aspects of integrative complexity that are not of central
importance to the hypotheses being tested in this paper, but that one should be aware of. These
could, for instance, be variables worth controlling for, or could be potential mediators or
moderators that, in future research, could shed further light on the pathways being studied.
These additional aspects include: the distinction between internal cognition and
impression management, the distinction between the state and trait components of integrative
complexity, and stress.
2.3.1 Internal cognition versus impression management
The current study does not test whether the integrative complexity score is indicative of a
leader’s internal cognitive complexity, or whether it is indicative of the leader’s external
communication style. The latter may not necessarily reflect their internal cognition, for instance,
because the leader might choose simple or complex messages in order to achieve desired effects
upon their audiences. In the literature, integrative complexity is defined as reflecting internal
cognitive complexity (Suedfeld, 1992), but that does not guarantee that the operational
measurement of integrative complexity is not influenced by external communication strategies.
That said, there are reasons to believe that the integrative complexity of public
communications tends to track the integrative complexity of a person’s private cognitive
processes. A theoretical reason to take this position is that integrative complexity is unlikely to
be consciously manipulated by the speaker, because it is so low in salience for the speaker
(Suedfeld & Rank, 1976).
Some empirical evidence that public integrative complexity reflects private cognition is
that previous archival research has found that, in a sample of eminent writers, the integrative
16
complexity of their private communications decreased during times of war, illness, and other
high stress periods (Porter & Suedfeld, 1981). Another study found that the integrative
complexity of editorials in the Bulletin of Atomic Scientists was lower in periods of time when
the members of the bulletin considered there to be an increased risk of nuclear catastrophe (as
indicated by their symbolic Doomsday Clock) (Suedfeld, 1980). Similarly, the annual speeches
of American Psychological Association presidents have been found to be lower in integrative
complexity when the United States was at war, compared to when it was not (Suedfeld, 1985).
While not strictly private in a narrow sense, these speeches are intended for a small audience
(relative to the country at large or the world), and provide no incentive to strategically alter
communications based on whether the country is at war.
These patterns with respect to the integrative complexity of private, or semi-private,
communications are similar to the findings of archival studies, which tend to find a pattern of
decreases in political leaders’ public integrative complexity in times of war and high stress
periods (Conway et al., 2001; Suedfeld et al., 1977; Suedfeld & Rank, 1976; Suedfeld & Tetlock,
1977). The similarity in public and private patterns in integrative complexity is consistent with
public complexity reflecting private complexity.
There is also experimental evidence that public integrative complexity reflects private
cognition. Experimental research has found that, when participants were induced to produce both
private and public verbal materials, under the same experimental control and treatment
conditions, the private and public integrative complexity reflected the same pattern of results
(Gruenfeld et al., 1998). This suggests that, at least in the types of situations studied, the
integrative complexity of public communications is reflective of the speaker’s internal cognition.
17
I believe that future research, using large N datasets, such as the one that I have built,
would help to provide stronger evidence concerning to what extent, and under what
circumstances, external communication strategies influence the integrative complexity score.
However, to do this would require coding and statistical analyses beyond the scope of the current
project. Furthermore, it is not necessary to do so to test my two hypotheses, as both could
plausibly operate through either the pathway of internal cognitive complexity, or through the
pathway of external communication style. (In the case of the “strategic judgment hypothesis,” if
the pathway were purely through external communication style, then a more appropriate name
might be the “strategic coordination hypothesis,” but the hypothesized link between expressed
integrative complexity and game theory would remain the same.) Therefore, the present study is
concerned with expressed integrative complexity, regardless of to what degree it reflects internal
cognitive complexity or external communication style.
2.3.2 The state versus trait components of integrative complexity
The current study does not test whether it is the state or trait components of integrative
complexity that is driving any empirical relationships between integrative complexity and
violence; that would be much more difficult to statistically test, and doing so may not be possible
without an even larger sample size.
In any case, the distinction between the pathways is not relevant to testing my two
hypotheses. Either hypothesis could operate through the pathway of the state or trait components
(or both, acting together). However, the distinction between the state and trait components is
relevant to the discussion of the results, and to potential avenues for future research, because the
distinction is relevant to the precise causal pathways involved.
18
Because of the danger of misinterpreting the results and discussion if the distinction
between the state and trait pathways is not well-understood, and because of the value of doing
future research concerning this distinction, I discuss it in detail here.
Some possible pathways would be primarily driven by the state component of integrative
complexity, for instance a shared association between violence, stress and integrative
complexity, because it is the stressfulness of the situation that is causally relevant. Other possible
pathways could be driven primarily by the trait component of integrative complexity. For
instance, this could be the case if violence is related to aspects of reputation, and if reputations
are built up over decades of a leader’s career.
To give an example of variation in the trait component of integrative complexity, the
lifetime integrative complexity of the leaders in this study varies – for the American presidents
for whom we have sufficient data, Lyndon B. Johnson, Donald Trump, and Joseph Biden are the
lowest, and John Quincy Adams is the highest. These differences may be due to variation in the
individuals themselves (a pure trait component), but may also be partly due to different
environments that the leaders found themselves in.
For instance, it is plausible that greater democratization would select for leaders who
exhibit lower integrative complexity. This is because in order to win elections they must
communicate to a public of political non-experts.
The public often lacks important knowledge relevant to politics (Americans Know
Surprisingly Little about Their Government, Survey Finds, 2014; Delli Carpini & Keeter, 1996),
and relies on heuristics such as using framing cues to identify experts (Druckman, 2001). This
suggests that political leaders who need to routinely win the support of the general public, rather
than that of a narrow elite with relative political expertise, may benefit from communicating at
19
lower levels of complexity. A finding that is consistent with that is that the integrative
complexity of American presidents is lower now that it was historically (Conway & Zubrod,
2022; this finding is replicated in the data in the current dissertation).
It is also plausible that situations with greater political polarization would be associated
with politicians exhibiting lower integrative complexity. This is because in highly polarized
political environments there is little chance of winning the support of people who are not already
on one’s side, so there is little incentive to justify one’s political positions to them. Being
required to justify one’s positions to people who disagree with them is associated with higher
integrative complexity (Tetlock, 1983), so highly polarized politics are likely to present reduced
incentives to communicate with high complexity.
The offices and career paths of leaders may also be relevant to their lifetime integrative
complexity, for instance, previous research has found that military leaders have lower cognitive
complexity than political leaders (Suedfeld, 2014; Suedfeld & Morrison, 2015). This is not to
suggest that the lower integrative complexity of military leaders is inferior, or that they are
always low in complexity. Rather, it may be that the military domain tends to call for lower
integrative complexity. I discuss this possibility in more detail below, in the subsection on
Integrative Complexity and Political Decision Making.
Regardless of a leader’s tendency in the trait component of complexity, situations that
occur in a more narrow and clearly defined range of time, such as international crises, could be
associated with changes in integrative complexity. For instance, situations that are highly
stressful or that impose time pressure tend to elicit lower integrative complexity (Suedfeld,
2010), while situations in which one must balance conflicting values, or in which one is held
20
accountable to people who might disagree, tend to elicit higher integrative complexity (Tetlock,
1983, 1986).
With respect to specific leaders or specific situations, previous research has found that,
for Bashar al-Assad, the President of Syria, his integrative complexity decreased when, during
the Syrian civil war, the threat from ISIS was most extreme (Suedfeld et al., 2014). Or, for Xi
Jinping, the paramount leader of the People’s Republic of China, his integrative complexity
decreased when there were protests and unrest in the ethnic-minority regions of Tibet and
Xinjiang (Suedfeld & Morrison, 2015). We cannot draw certain causal conclusions from these
case studies, but one possibility is that the stressfulness of these situations was associated with
decreased integrative complexity, while another possibility is that decreased integrative
complexity is associated with ceasing to try to find a mutually acceptable solution with one’s
adversaries, or that it is associated with taking a rigid hardline position in order to attempt to
deter or compel one’s adversaries (these three possibilities are not mutually exclusive).
2.3.3 Stress
The current study does not test whether the stress experienced by a leader is part of the
causal pathway that relates their integrative complexity to the use of violence in international
crises and conflicts. To give an example of one possible such causal relationship, it may be that
stress can cause a leader to exhibit decreases in integrative complexity, and that low integrative
complexity can be a causal contributor to international crises escalating to violence (Suedfeld &
Tetlock, 1977).
If stress is a cause of a change in integrative complexity, this is not inconsistent with
either of the two hypotheses. For instance, the causal relationship in the above paragraph is
21
consistent with the strategic judgment hypothesis. As a result, testing the role of stress is not
necessary in order to test the two hypotheses.
On the other hand, understanding the relationship between stress and integrative
complexity may help to better understand the results, and will help to point out areas for future
research. With this in mind, the relationship between stress and integrative complexity is worth
discussing in some detail.
In early complexity research, laboratory studies of the relationship between induced
stress and the conceptual complexity participants found an inverted-U relationship, in which the
complexity of responses was highest at the mid-levels of stress (Schroder et al., 1967). This is
consistent with the idea that there is an optimal level of stress that induces a person to recruit the
most cognitive resources to addressing a problem. At lower levels of stress, there may not be
enough challenge to induce the use of more cognitive resources necessary to reach higher levels
of complexity. At higher levels of stress, cognitive resources (or motivation to be complex) may
become exhausted, resulting in lower levels of complexity (Suedfeld, 1992).
A difficulty with researching an inverted-U relationship is that it is difficult to identify
what level of stress would be optimal, and this level may change depending on the person, the
issue, and the time-scale. As a result, much of the research concerning the relationship between
integrative complexity and stress for real world political leaders focuses on the stressful side of
the inverted-U – i.e., whether very high levels of stress are associated with decreases in
integrative complexity.
Most of this research has found that, at very high levels of stress, political leaders exhibit
decreases in integrative complexity. For instance, Robert E. Lee exhibited high integrative
complexity for much of the Civil War, but exhibited decreases late in the war, when the
22
Confederacy was losing the ability to continue the war (Suedfeld et al., 1986). During the first
Persian Gulf crisis, Iraqi President Saddam Hussein’s integrative complexity decreased
drastically when coalition troops began the ground attack against Iraq (Suedfeld et al., 1993;
Wallace et al., 1993). Gorbachev exhibited high integrative complexity until late 1989, when the
challenges to the USSR were arguably revealed to be insurmountable, at which point his
integrative complexity decreased (Wallace et al., 1996).
In previous research that the author contributed to, our research team have found that
Syrian President Bashar al-Assad’s integrative complexity decreased during a period of time in
which the threat from ISIS was most extreme (Suedfeld et al., 2014). In other previous research,
we have also found that Xi Jinping’s integrative complexity decreased during protests in Tibet
and Xinjiang, although it did remain stable during technical challenges such as an epidemic and
an earthquake (Suedfeld & Morrison, 2015).
As the inverted-U model would suggest, there are exceptions to this pattern. A study of
political and military geniuses found that more than half of them exhibited increases in
integrative complexity during high stress periods, and that military geniuses were particularly
likely to do so (Suedfeld, 2014). The research that I contributed to, mentioned above, concerning
Xi Jinping, may suggest that societal opposition to the Communist regime, such as ethnic
minority protests, induces a low integrative complexity responses in Xi, while technical
challenges do not seem to do so. In some of our other previous research, we have also found that
Russian President Vladimir Putin’s integrative complexity did not decrease in association with
the Russian invasion of Crimea (Suedfeld & Morrison, 2016), suggesting either that Putin did not
find this particularly stressful, or that the stress did not induce decreases in his integrative
complexity.
23
Although I do not directly test the role of stress in the current research, in order to prevent
the results from being dominated by effects of stress, whether as a confound or a mediating
variable, I do statistically control for variables that should be strongly associated with stress. This
includes variables such as the gravity of the threat, the power disparity between the two sides of
the conflict, the distance of the crisis from the leader’s country, and whether the crisis is part of
an enduring rivalry.
2.4 The Cognitive Manager Model
The cognitive manager model (Suedfeld, 1992, 2010) can explain many of the correlates
of integrative complexity discussed above. According to this model, a person has limited
cognitive resources, and therefore rather than devote maximum cognitive resources to all
problems, they should allocate cognitive resources to problems, based on the resources currently
available, the importance of the problem, and the amount of cognitive resources required to
adequately solve the problem. Higher levels of integrative complexity require greater cognitive
resources, and so, if a person does not allocate cognitive resources to a problem, they are
unlikely to exhibit high integrative complexity concerning that problem.
This explains why some measures of cognitive ability are positively associated with
integrative complexity. If higher levels of integrative complexity require more cognitive
resources, then persons with higher levels of the relevant cognitive abilities will be more able to
reach higher levels of integrative complexity. An important caveat is that the ability to reach
higher levels of integrative complexity does not mean that that level will actually be exhibited –
the person must also be motivated to do so, and persons may vary in trait tendency for that
motivation, and situations may vary in the degree to which they elicit that motivation.
24
The relationship between cognitive resources and integrative complexity can also explain
why very high levels of stress are associated with decreases in integrative complexity. At very
high levels of stress, a person’s cognitive resources are depleted (e.g., fatigue, burnout), so that
the level of integrative complexity that they exhibit decreases. This is called “disruptive stress”
(Suedfeld, 2010).
The finding that a person’s attitude strength concerning an issue is negatively associated
with their integrative complexity concerning that issue (Conway et al., 2016) can also be
explained by the cognitive manager model. If a person feels strongly that they already know the
correct answer concerning an issue, then they should have little motivation to apply cognitive
resources to analyzing it. (This application of the cognitive manager model is my own.)
The model suggests that adequately solving some problems might require complex
analyses, while other problems might be adequately solved by very simple analyses. This
suggests that it would be valuable to theorize concerning the kinds of cognitive challenges that
political leaders face (and more broadly that humans have faced throughout our biological and
cultural evolution), and whether these challenges call for high or low cognitive complexity.
This is the background framework for my theorizing concerning my two hypotheses. If
certain challenges, such as coordination problems, call for high integrative complexity, then
leaders who exhibit high integrative complexity should be more likely to be successful when
facing this type of challenge. This suggests the strategic judgment hypothesis. On the other hand,
if certain challenges, such as problems of uncertain resolve or of deterrence, call for low
integrative complexity, then leaders who exhibit low integrative complexity should be more
likely to be successful when facing this type of challenge. This suggests the demonstration of
resolve hypothesis.
25
2.5 Integrative Complexity, Political Decision-Making, and Political Violence
This section covers the ways in which existing integrative complexity theory suggests
that the integrative complexity of political leaders is relevant to the likelihood that they become
involved in political violence (e.g., war). It starts by describing the broad theory concerning how
a person’s integrative complexity relates to their decision-making, and to how that person is
perceived by others. Then it describes the results of existing empirical research concerning the
relationship of political leaders’ integrative complexity with aspects of their decision-making
that may be relevant to causal processes leading to political violence, and with violence itself.
The section concludes by discussing how the theoretical and methodological approach of the
current study addresses possible confounds, and the other limitations of previous research.
2.5.1 Integrative Complexity and Political Decision Making
The theory of integrative complexity asserts that persons with higher integrative
complexity are better able to take into account other persons’ perspectives, to incorporate
multiple variables into their understanding of a situation, and to take into account that these can
form interactive systems (Baker-Brown et al., 1990; Suedfeld, 1992, 2010). This suggests that
they may take longer to make decisions, be more likely to change aspects of their beliefs and
plans, and to be perceived by others as more flexible/less decisive.
Conversely, integrative complexity theory asserts that persons with lower integrative
complexity are more likely to focus on their own perspective, and to incorporate a smaller set of
variables in their understanding of a situation, and to focus on simple causal relationships rather
than taking into account complex interactions and emergent features of systems (Baker-Brown et
al., 1990; Suedfeld, 1992, 2010). This suggests that they may make decisions more rapidly, and
be less likely to change aspects of their beliefs and plans (but that, if they do so, to be more likely
26
to rapidly convert to a consistent opposing position, rather than gradually adopt compromise
positions), and to be perceived by others as less flexible/more decisive.
While integrative complexity is a not a measure of cognitive ability as such, most of the
existing empirical research provides support for considering integrative complexity to be
positively associated with some aspects of cognitive ability (Suedfeld & Coren, 1992). For
instance, in a study in which participants competed to make accurate geopolitical forecasts,
higher cognitive complexity was one of the strongest predictors of more accurate forecasting
(Karvetski et al., 2021). Previous research, coding the integrative complexity of political leaders
from texts, has found that political revolutionaries who exhibited increases in integrative
complexity after taking power were more successful in government than those who did not
exhibit this increase (Suedfeld & Rank, 1976); and that many (but not all) political and military
geniuses exhibited stable, or increasing integrative complexity during political or military crises
(Suedfeld, 2014). This suggests that higher integrative complexity is associated with higher
levels of a cognitive ability, or of a cognitive resource, which enables political leaders to better
achieve their goals. In so far as political violence and war are highly costly, it is plausible that
political leaders would, all things being equal, prefer to avoid violence, and that therefore higher
integrative complexity would be associated with greater avoidance of political violence.
However, integrative complexity theory also asserts that low integrative complexity can
contribute to political leaders achieving their goals, such as in situations in which a leader must
act quickly and decisively and remain committed to their course without wavering (Suedfeld,
1992, 2010), or in situations in which they must convince other persons that they have the
resolve to do so. For instance, the example of political revolutionaries exhibiting increases in
integrative complexity after taking power (see above paragraph) cuts both ways, because it
27
suggests that the violent revolution, in which they seized power by force, was a situation that
called for low integrative complexity.
Previous research has also found that the integrative complexity of political leaders is
lower than usual during elections (Conway et al., 2012), suggesting the possibility that lower
integrative complexity is useful in elections, perhaps by making the leader appear more
committed to the cause, or helping them to castigate adversaries, or to motivate their supporters.
In addition, the integrative complexity of challengers is lower than that of incumbents,
suggesting that criticizing an adversary calls for relatively low complexity, while defending
one’s political record, with its imperfections and compromises, calls for relatively higher
complexity.
Previous research has found that military leaders have lower integrative complexity than
political leaders (Suedfeld, 2014; Suedfeld & Morrison, 2015), suggesting the possibility that
lower integrative complexity may be more effective for dealing with military issues than broader
political issues. There are more possible reasons for this than could be listed here, including that:
military issues are a more restricted domain than all of politics; taking military action may
require taking actions that are inconsistent with recognition of the adversary’s perspective (in the
extreme case, killing the adversary); and deterring adversaries may require presenting oneself as
someone who cannot be easily pressured into changing course. Furthermore, at the tactical level,
successful military action often requires rapid and decisive decision-making, going by the book,
formulating simple plans, and executing them to completion without being distracted by new and
potentially adverse information (Suedfeld et al., 1986).
The advantages to being perceived as someone who cannot be pressured into changing
course may, under some circumstances, also apply to political leaders. For example, Kim Jong-
28
un has unusually low integrative complexity for a political leader (Suedfeld & Morrison, 2019),
and his regime survives, in part, by threatening other states with destruction, and by using these
threats to deter challenges to the regime, and to coerce concessions, such as food aid. Although
we cannot be certain of causality from the evidence available, it is plausible that his low
integrative complexity makes his threats more credible, thereby making them more likely to
succeed.
2.5.2 Integrative Complexity and Violence
As we can see from the above, there is evidence that integrative complexity is empirically
associated with different roles and situations, and that it is plausible that high or low complexity
could be adaptive under different circumstances. In so far as each level of complexity might,
under certain circumstances, enable leaders to achieve their goals without resorting to costly
violence, or to deter adversaries from employing violence against them, each could theoretically,
under the appropriate circumstances, be associated with less violence. However, the existing
empirical evidence lends more support to high integrative complexity being associated with less
violence.
As mentioned above, among a sample of leaders who took power in a violent revolution,
the leaders who succeeded in government tended to be those who demonstrated higher
complexity while in government than during the revolution (Suedfeld & Rank, 1976). This is
evidence that the adaptive pattern for these leaders was to exhibit relatively low integrative
complexity during the revolution, and relatively high integrative complexity while in power.
Also as mentioned above, military leaders, who by definition must be involved in
preparing for violence, if not executing it, have lower integrative complexity than political
leaders (Suedfeld, 2014; Suedfeld & Morrison, 2015).
29
Research also provides evidence that low integrative complexity is associated with the
use of violence. A study has found that leaders exhibited decreases in integrative complexity
during international crises that ended in war, such as the outbreak of the First World War or the
Korean crisis of 1950, but not prior to crises that were resolved without war, such as the Agadir
Crisis or the Cuban Missile Crisis (Suedfeld & Tetlock, 1977). Another study has found that
leaders exhibited decreases in integrative complexity prior to carrying out surprise attacks, such
as the Pearl Harbor attack (Suedfeld & Bluck, 1988).
Although this may not generalize across all nationalities and political orientations,
research has found that extremist1 groups that engage in violence have lower integrative
complexity than those that merely support violence, which are themselves lower in integrative
complexity than extremist groups that reject violence, e.g., the IRA versus Sinn Fein versus the
SDLP (Suedfeld, 2022).
Another example is that of Syrian President Bashar al-Assad. In the Syrian Civil War,
Bashar al-Assad exhibited decreases in integrative complexity during the period of time that the
extremely violent expansion of ISIS was most threatening to the survival of his regime (Suedfeld
et al., 2014). Although we cannot confidently draw conclusions concerning causality, this pattern
is consistent with an association between violence and low integrative complexity.
The Gulf War of 1990 provides another example of a negative association between
integrative complexity and political violence. A study of the integrative complexity of political
leaders involved in that conflict found that during the conflict, two leaders who were particularly
involved, namely the American President Bush, and Iraqi President Hussein, exhibited lower
integrative complexity than leaders who were less involved in the conflict (Suedfeld et al., 1993;
1 Here the word “extremist” is meant in the value neutral sense that the group seeks to bring about extreme or
radical change in society, and is not meant pejoratively.
30
Wallace et al., 1993). Furthermore, they found that Hussein’s integrative complexity was low
immediately prior to his invasion of Kuwait, increased during the early stages of the occupation
of Kuwait, and then decreased prior to the UN Security Council’s deadline for the withdrawal of
Iraqi forces (Suedfeld et al., 1993; Wallace et al., 1993). Presumably this latter decrease was
associated with the threat that coalition forces, led by the United States, would go to war against
Iraq, which is ultimately what happened.
There is also evidence of high integrative complexity being associated with moving from
violence towards peace - in peace negotiations between the Chiapas rebels and the Mexican
government, higher integrative complexity was associated with periods of greater progress in
negotiations (Liht et al., 2005).
Overall, the body of previous research supports the claim that the integrative complexity
of political leaders is negatively associated with violence, although none of the previous research
has directly tested this using a large N or representative sample.
2.5.3 Limitations of Previous Research Concerning the Relationship Between
Integrative Complexity and Violence
As discussed previously, there is evidence that very high levels of stress are associated
with decreases in integrative complexity. If this is the case, then it is theoretically possible that
stress is a confounding variable, for instance, stress would be a confound if it were to cause both
violence and low integrative complexity. Although this is theoretically possible, a more plausible
causal pathway is that stress, once it becomes so extreme that it exhausts cognitive resources,
contributes to decreases in integrative complexity, which in turn make leaders more likely to
employ violence. However, given the theoretical possibility that stress is a confound, it is a
31
weakness of previous research that it does not tend to control for variables that would be
expected to be associated with stress.
To address this limitation, the current research indirectly controls for stress by
statistically controlling for variables that should contribute to the leader’s level of stress, such as
the power disparity between the sides in the conflict, the gravity of the threat, and the distance of
the leader’s country from the conflict.
An additional limitation of previous research is that, because it tends to be based on small
samples, or case studies, of cases that are interesting to researchers, it cannot take into account
the complex structures of interactions among states, such as game theoretical structures in the
interactions between states, or the structure of international systems. It is theoretically possible
that any instance of violence, after its effects work through these international structures, reduces
the incidence of future violence. Or, conversely, that a given instance of failure to employ
violence increases the incidence of future violence. It is also theoretically possible that a
characteristic of a political leader, such as their integrative complexity, could make them less
likely to be drawn into violence in any isolated instance of conflict, but more likely to be drawn
into violence as an ironic effect of interactions in the international system.
As a result, any finding that a political leader’s integrative complexity is low, or
decreases, in association with a specific violent event, does not necessarily demonstrate that
political leaders who exhibit low integrative complexity are more likely to be involved in
violence, as an overall pattern within the international system.
To address this limitation, the current research includes a statistical control for the
international relations time period, which takes into account the different structures of the
international system (i.e., unipolar, bipolar, multipolar) across different time periods.
32
In the next chapter, I discuss how violence can emerge from coordination problems and
problems of uncertain resolve. Then I use the game theoretical models of prisoner’s dilemma and
the game of chicken to generate hypotheses concerning how the integrative complexity of
political leaders, operating within these games, could affect the likelihood of political violence
occurring.
33
Chapter 3: Two Classes of Problem that Can Lead to Political Violence
The cognitive manager model proposes that different levels of integrative complexity
may be appropriate for addressing different types of problem. Following Fearon’s bargaining
model of war (Fearon, 1995), I focus on two classes of problem that political leaders (and more
broadly human beings) face when interacting with adversaries in situations that could escalate to
violence. These two classes of problem are coordination problems and problems of uncertain
resolve.
3.1 Coordination Problems
There can be coordination problems (also called “commitment problems”) in which
leaders, by acting in the interests of their respective countries, increase the likelihood of war. In
game theory, a noteworthy game that models a coordination problem is the prisoners’ dilemma.
In this game, regardless of the strategy chosen by the other player, each player receives a higher
payoff by defecting than they would by cooperating. However, when a player defects, it lowers
the payoff of the other player, such that the lowest mutual payoff is if both players defect. The
result is that it is always in each party’s limited self-interest to defect, such that the dominant
strategy is that both parties defect, resulting in all parties being worse off than they would be if
they could commit to cooperating (Jervis, 1978).
A prominent example of a coordination problem in international relations is the security
dilemma, in which actions that a state takes to increase its security, say by arming itself,
undermine the security of other countries, giving them incentive to also arm themselves, leading
to an arms race that threatens the security of all countries (Jervis, 1978). A historic example of
this is how the early 20th century invention of the dreadnought – a battleship so superior to
previous designs that it was considered to have made them obsolete – set off a naval arms race,
34
because countries needed to build dreadnoughts in order to defend against any other country that
was building dreadnoughts.
Another example is the nuclear arms race between the United States and Soviet Union in
the mid to late 20th century. Because the adversary country could use its nuclear weapons to pre-
emptively attack and destroy one’s nuclear weapons, whenever the adversary increased its
nuclear arsenal this presented an incentive to make one’s own nuclear arsenal larger and more
difficult to destroy, in order to maintain a second-strike capability.
Another example of a coordination problem occurs when it is believed that whichever
side attacks will have the advantage over the defending side. In that case, even if no one wants to
go to war, a leader might prefer to pre-emptively attack to gain the attacker’s advantage, rather
than risk being attacked and having the defender’s disadvantage (Fearon, 1995; Jervis, 1978).
There is an iterative logic to this, in which if a leader knows that the other side is under the same
pressure to pre-emptively attack, this gives the leader an additional incentive to attack first,
which in turn gives the other side an additional incentive to do so (and so on).
This process has been proposed as another factor contributing to the outbreak of the First
World War. At the time, political leaders believed that the attacker would have the advantage, as
had been the case in the previous Franco-Prussian War (Jervis, 1976, 1978). The Germans, in
particular, relied upon taking the attacker’s advantage, because they expected to fight a two-front
war with France and Russia, and their plan for doing so (the Schlieffen Plan), required that
Germany rapidly conquer France before Russia could invade Germany. Tensions escalated after
the assassination of Archduke Franz Ferdinand, and the leaders of Austria-Hungary, France,
Germany, and Russia chose to mobilize their armies, so that they could avoid the defender’s
disadvantage by being ready to attack. When the Russian government mobilized its army, the
35
German political leaders perceived this as an unacceptable threat because the German war plans
could not handle an early Russian offensive, and so the German government declared war.
Another type of coordination problem occurs when making a concession to the other
country would increase their ability to attack in the future (Fearon, 1995). If the other country
could make a truly guaranteed commitment that if they are given what they want they will not
make further demands, and will not attack, then it might be rational to accede to their demands.
But because countries cannot give guaranteed commitments like this, any concession that
increases the adversary’s ability to attack in the future weakens one’s own future bargaining
position, which could result in an unacceptable spiral of concessions or military losses.
For instance, prior to the Second World War, Czechoslovakia had extensive fortifications
along its border with Germany. Hitler coveted Czechoslovakian territory, and promised the
European powers of Britain, France, and Italy that he would make no further territorial claims if
he were allowed to annex the Sudetenland region of Czechoslovakia. However, after the Munich
agreement, in which the European powers allowed Hitler to annex the Sudetenland - thereby
taking the defensive fortifications from Czechoslovakia without a fight – he proceeded to occupy
most of Czechoslovakia.
In this example war was avoided (for about a year), but it illustrates how that first
territorial concession put Czechoslovakia in a position in which it could not effectively resist
future demands. For this reason, it would not have been irrational for Czechoslovakia to fight a
war over the Sudetenland, even if it were not in and of itself worth the costs that would be
suffered from fighting a war with Germany.2
2 In the actual case, the leaders of the Czechoslovakian government were effectively denied the choice, because
they were excluded from participating in the Munich meeting.
36
In principle, all of these coordination problems would be solved if countries could truly
commit to activities that result in the mutual benefit of all countries involved. Arms races could
be prevented if states could all commit to limiting their arsenals. The pressure to attack pre-
emptively, when it is believed that the attacker has the advantage, would be solved directly if
states could commit to not attacking first. Or, less directly, and perhaps less unrealistically, it
would be solved if states committed to defense pacts in which they agreed to declare war on any
country that pre-emptively attacked another country. Wars fought to prevent the exchange of
territories that are crucial to the existing balance of power could be prevented if countries could
commit to not use those territories for strategic advantage. However, countries cannot truly
commit to any of these things by merely promising to do so, rather this often requires a complex
understanding of the interests of multiple actors, and of how those actors interact.
To anticipate my theorizing in the next chapter, this line of reasoning suggests that
coordination problems can be solved, and war averted, by political leaders who can take one
another’s perspectives into account, and recognize how they are part of a complex interactive
system, and find ways to shape that interactive system in order to facilitate coordination. As was
the case with the previous explanation for war, this suggests that the psychological characteristic
of high cognitive complexity would make political leaders less likely to go to war.
3.2 Problems of Uncertain Resolve
Another explanation of war is that it arises because leaders underestimate the resolve of
the leaders and citizens of other countries (or overestimate that of the citizens of their own
country). The resolve of a country’s leaders or citizens is an internal mental characteristic, such
that one cannot directly observe the resolve of another leader, or of citizens. Nor can one trust
leaders when they say that they have high resolve, because they have an incentive to fake it to
37
influence (e.g., compel or deter) other countries. However, their resolve can be observed
indirectly, by observing whether they send costly signals – in other words, by observing them
engage in costly behaviors that suggest that they would be willing to suffer further costs.
War can therefore be conceptualized as a costly signal, through which leaders and
countries learn about the resolve of their adversaries, and communicate their own resolve to
those adversaries (Fearon, 1995). If the leaders of a country say that they will fight rather than
accept a compromise, they might be bluffing, in which case war would compel them to settle
rather than to suffer the losses of war. But if they are not bluffing, then they can demonstrate
their resolve by continuing to fight despite these losses. Historical examples in which war
demonstrated resolve include the British in WWII (who Hitler believed would sue for peace), the
Viet Cong and North Vietnamese in the Vietnam War, and the Taliban in Afghanistan.
Strategic interactions in which leaders of adversary countries attempt to demonstrate
resolve, in order to convince the other side to back down, can lead to war if neither side succeeds
in conveying enough resolve to cause the other to back down. One such strategic interaction is
brinkmanship, in which the adversaries push the conflict to the brink of disaster, in order to
pressure the other side to back down (Jackson, 2005; Schelling, 1980). Brinkmanship can be
explained as a strategy for demonstrating resolve when each side has incomplete information
(Powell, 1988). It can be an effective strategy if it convinces the other side to bear the costs of
backing down first. But it is an inherently risky strategy, as it necessarily involves taking actions
that lead to the brink of war, with the potential to go past the brink.
A concrete example of brinkmanship is the Cuban missile crisis (G. Allison, 1971). In
this crisis, the Soviet Union had placed nuclear missiles in Cuba, within striking distance of the
United States. In response, President Kennedy put a blockade around Cuba, in order to pressure
38
the Soviet Union to remove the missiles. The Soviet Union had sent vessels towards Cuba which,
if they had continued on their course, would have been intercepted by the American blockade.
Ultimately General Secretary Khrushchev ordered the Soviet vessels to stop, and agreed to
remove the nuclear missiles from Cuba, in exchange for the United States removing its nuclear
missiles from Turkey.
This was a case of brinkmanship because both sides pushed the confrontation to the point
at which, if neither had backed down, the outcome could have been a naval battle, followed by
escalation to nuclear war. But, if the United States had backed down first, without gaining
concessions from the Soviet Union, this would have enabled the Soviet Union to continue to base
nuclear missiles in Cuba, within striking distance of the United States. In this situation, neither
side knew the resolve of the other side, and both used strategies of brinkmanship in order to gain
the upper hand, with the attendant risk of escalation to war.
In game theory, brinkmanship (and more generally contests of resolve) are modelled
using the game of Chicken (Jervis, 1978; Lebow, 2020; Nalebuff, 1986; Schelling, 1980). In
Chicken, players can choose to co-operate or to defect (i.e., not co-operate). If both players
defect the result is catastrophic, but a player can win maximum benefits by defecting if the other
player cooperates. This payoff structure might sound similar to that of prisoner’s dilemma, but a
crucial difference is that unlike in prisoner’s dilemma, it rewards players for pressuring the other
side to back down (i.e., through brinkmanship).
In the classic analogy for a game of chicken, the two players are driving their cars
towards one another. If a player turns away to avoid a collision (i.e., co-operates), then that
player loses, while the player who continues to drive straight (i.e., defects) wins. But if both
continue to drive straight (i.e., both defect), then the result is a catastrophic collision. In such a
39
game, one can improve one’s chances of winning by giving the impression that one will not back
down, no matter the risk, so that it is up to the adversary to pull back from disaster.
Another classic metaphor that illustrates this logic is the scenario of two adversaries who
are chained together and standing next to a cliff edge. In this scenario, one can dance on the cliff
edge until the adversary concedes defeat (Schelling, 1980). This highlights the metaphor behind
the term “brinkmanship.”
Similarly, one can improve one’s chances of winning by publicly tying one’s hands in
advance, so that it is known that one will always defect. To return to the metaphor of a game of
chicken, this could be by making a show of blindfolding oneself before starting the car.
In politics, a leader could tie their hands by whipping up popular opinion in their own
country against making concessions. By doing so, the leader could create a situation in which if
they made concessions, they would lose the next election (or in a non-democracy, face a revolt).
This enables the leader to credibly claim that their hands are tied, and that they cannot afford to
make concessions, which strengthens their position in negotiations with the foreign adversary
(Bueno De Mesquita & Smith, 2012; Putnam, 1988).
If the adversary sees that one’s behavior signals resolve, then they are more likely to
believe that they must back down to avoid disaster. However, with these strategies of
brinkmanship a risk is that if the other player is not convinced of one’s commitment to drive
straight (defect), then the result may be that they also drive straight (defect), resulting in a
catastrophic collision (war).
According to this logic, in order to convince the other side to back down, and thereby to
avoid war, it is necessary to convince the adversary that one has an uncompromising,
40
unswerving, and total commitment to hold to the path that one has taken, no matter how
dangerous and unreasonable this may appear.
Deterrence failure is a related route by which the failure to credibly communicate resolve
can lead to war. Deterrence is a type of coercion, in which an actor threatens to impose high
costs in the event that another actor commits an act, in order to prevent them from committing
that act (Schelling, 1980). For instance, in the Cold War, the United States had a policy of
deterring the Soviet Union from invading West Germany, by threatening to use nuclear weapons
against the Soviets if they invaded (Gerzhoy, 2015; Huntington, 1985; Mearsheimer, 1985).
A problem with deterrence is that, once the adversary has committed the act, the
deterrence has already failed, so carrying out the threat would not have the benefit of making the
deterrence successful, nor would it have any immediate material benefit, but it would still be
highly costly. Viewed this way, it is irrational to actually carry out the threat, because the costs
of doing so (e.g., nuclear war) greatly outweigh the benefits, which may be practically none
(Jervis, 1978; Schelling, 1980).
There is the benefit of establishing or maintaining the reputation as an actor that will
carry out its threats (this is crucial to the logic of resolve and deterrence). However, in many
conflicts the governments in question are destroyed, rendering their future reputations irrelevant.
In the current example, this would be a likely outcome of nuclear war between the two
superpowers, so that there would be little reason to expect any benefit to carrying out the
deterrent threat.
However, if the Soviet Union knew that the United States would not carry out the threat,
then the Soviet Union would have no reason to be deterred by it. As a result, in order to deter, it
is necessary to convince the other side that you have the resolve to carry out your threats, even
41
when doing so would be seemingly unreasonable. A failure to convince the other side of this
resolve could result in a failure to deter their aggression, which, if one is in fact resolved to fight,
would lead to war. According to this logic, in order to successfully deter aggression, and thereby
avert war, it is necessary to convince the other side that one will unswervingly carry out one’s
threats. One must have the reputation as someone who carries out one’s threats, no matter how
costly and fruitless this would be, and no matter how many dimensions of the issue suggest that,
in the actual event, carrying out the threat would be pointless and unreasonable.
Again, it is true that, because the countries in question will continue to interact in the
future, it may be worth fighting over reputation. The problem remains that, in such a situation,
the reputation that is of value is the reputation of someone who, when challenged, will not back
down from the brink of catastrophe.
This problem of unknown resolve is exacerbated by “salami tactics,” which is a strategy
that countries can use to circumvent attempts by an adversary country to deter them. The essence
of salami tactics is that the aggressive behavior is kept to a low enough level that, according to a
cost-benefit analysis that only takes into account the immediate material results, it would not be
worth it for the would-be deterrer to actually carry out the punishment.
A classic analogy is that if a parent tells a child to stay out of a pool, the child starts by
dipping their toes in the water, and then their feet, and their legs, and so on, and at each stage
they look to confirm that the parent has observed it. Each stage in the process is so gradual that it
may seem unreasonable for the parent to punish it, but in the end the child is in the pool
(Schelling, 1980).
A recent example of this is how China has gradually expanded its military presence in the
disputed maritime territory of the South China Sea, so that although there was no specific time
42
when they could be said to have invaded, they have now built several military bases on islands
throughout the territory. Another example might be the Russian invasion of Crimea, and Russian
involvement in the rebellion in eastern Ukraine, which have been small enough that they have
not triggered a military response by the United States, but that potentially establishes a precedent
that undermines the ability of the United States to credibly deter aggression against its eastern
European allies. Salami tactics can be deterred, but only if one convinces the adversary country
that one would punish their aggression, even when that aggression seems so minor that doing so
would seem to be an unreasonable and disproportionate response.
The logic of contests of resolve, such as brinkmanship, deterrence, and salami tactics,
suggest that resolve can be communicated, and war averted, by political leaders that are good at
convincing others that they would uncompromisingly carry out their threats, even when doing so
would seem costly and pointless, and even when doing so would seem like an unreasonable
response relative to the behavior that provoked it.
To anticipate my theorizing in the next chapter, it is plausible that the leaders will be
most effective in doing so when their cognitive structure (or their expression of it) is consistent
with this uncompromising and seemingly unreasonable behavior, such that they think, talk, and
act like they would not compromise, or second guess, or be swayed by the persuasiveness of
what would be “reasonable” in the situation. Unlike in the case of the previous explanation for
war, this suggests that high cognitive complexity would make political leaders more likely to be
drawn into war, because low cognitive complexity leaders would be more effective at deterring
aggression.
43
3.3 Psychological characteristics may function to address these classes of problem
Seemingly irrational psychological characteristics can play a role in processes leading to
political violence and war. These include a commitment to spiritual and cultural values, religion,
honor/standing/reputation, and revenge (Lebow, 2010), and a deep attachment to territory
(Vasquez & Henehan, 2001). I argue that these may serve a function with respect to addressing
coordination problems and problems of uncertain resolve.
If this is the case, then it provides further theoretical evidence that variation in integrative
complexity, whether in the trait component across persons, or in the state component across
situations, may also serve a function with respect to addressing these different classes of
problem.
3.3.1 Honor
Honor is an important aspect of international violence. In a survey of historical wars and
their causes, Lebow (2010) found that standing (i.e., reputation, honor) and revenge were the
motives behind 68% of wars. For instance, in the events leading to the outbreak of the First
World War, the Austro-Hungarian government was motivated by the desire to avenge the
assassination of the Archduke Franz Ferdinand, and to maintain Austro-Hungary’s standing as a
great power. Lebow found that the desire to avenge losses in previous wars motivated the 18th
century wars of the Austria-Prussia rivalry, as well as the Ottomans going to war against Austria
and Russia, and Louis XIV’s Dutch War.
I argue that a concern with honor can function to solve the problem of uncertain resolve.
In a rational actor model, if a good is not worth fighting over, then a predatory actor should take
it from their potential victim, unless that victim is known to also value their reputation as
someone who is willing to fight to defend what is theirs. The international system is largely an
44
environment of anarchical self-help, i.e., an environment with no central authority that provides
rules and police to enforce them. In such an environment, if a person has a reputation as someone
who is not willing to fight to defend what is theirs, then predators can successively take goods
from the target (keeping each instance below the value that would be worth fighting over) until
the victim has nothing left. This is the salami tactics mentioned previously. In order to avoid
falling victim to this, a person can establish a reputation as someone who is willing to fight to
defend what is theirs. This is a core feature of honor.
According to this reasoning, regardless of whether it is a conscious strategy, or culturally
evolved, or biologically evolved, a concern with honor can function to protect a person from
predation.
Although coming from the discipline of social psychology rather than political science,
the seminal work on the culture of honor reflects similar logic (Cohen et al., 1996; Cohen &
Nisbett, 1994; Vandello et al., 2008). The explanation for why the American South has a culture
of honor is that its inhabitants are descended from people who lived in areas that relied on
livestock herding. It is not worth dying to prevent the theft of a cow. But, if one has the
reputation of someone who will not fight to defend their livestock, then cattle thieves can
successively steal one’s cattle until there are none left. If one has the reputation of being willing
to fight to defend their livestock – a reputation of honor – then one can avoid this fate. So,
according to this reasoning, people who rely on livestock herding for survival will, over
generations, culturally evolve to value honor, because it promotes survival in this environment.
The international system is a similar environment, in that it is an anarchical system, in the
sense that countries must rely on self-help, rather than a central authority. The international
system does not have an impartial police force with a monopoly on violence that enforces the
45
law in disputes between countries. The role of structural “anarchy” is the theoretical keystone of
the international relations school of structural realism (Mearsheimer, 2001; Walt, 1990; Waltz,
1979), which has been discussed in the earlier section concerning structural approaches in
international relations.
To return to the analogy, the leader of a country is in a similar situation to a cattle herder,
in that the leader of a country often cannot prevent an act of international aggression, e.g.,
annexation of territory, and may only be able to punish it after the fact. Bearing costs to punish
an act of aggression, after the fact, is irrational, unless the logic of reputation and honor is taken
into account.
A caveat is that I am not committed to any particular pathway by which the above logic
could operate. For instance, the leaders involved might be consciously aware of this logic, or
they might not be. Their cognitive processes are also influenced by the distal causes of biological
evolution and cultural evolution, and the logic of reputation could also operate through those
pathways. The theory does not imply a commitment to any one of these over the others.
The logic of reputation, in an anarchical system, implies that concern with honor would
be a frequent aspect of international violence. If the leaders of countries, and their populations,
lose the reputation as people who are willing to fight to defend what is theirs, then they could
ultimately be preyed upon until they lose everything. While the proximal cause of conflicts over
honor could be framed as due to an irrational psychological characteristic, the distal cause of
why human beings have a tendency to feel concern over honor can be explained using the reason
concerning the political mechanism of uncertain resolve.
46
3.3.2 Religion and sacred values
Religion and the perception that certain goals are sacred are another likely causal
contributor to international violence and war. Religious conflict plays a role in the Northern
Ireland conflict (Catholic and Protestant), the Arab-Israeli conflict (Muslim and Jewish), the
conflicts in the former Yugoslavia (Orthodox, Catholic, and Muslim), and the India-Pakistan
conflict (Hindu and Muslim), to choose just a few examples. Superficially, religion and
perception of sacredness may appear to be causes of war that do not arise out of political
mechanisms like coordination failure or uncertain resolve.
However, similarly to a concern with reputation and honor, the emergence of a concern
with religious and sacred values makes sense in the context of environments in which actors are
faced with the problems of coordination failure and uncertain resolve.
Religion and sacred values can serve a function (not necessarily consciously) of
promoting resolve in the ingroup, and of communicating that resolve to both the ingroup and the
outgroup. For instance, field research has found that, among religious extremist groups,
behavioral cues of religiosity serve the function of promoting and communicating commitment,
and that the strategic value of this signaling is a likely reason for the success of these groups
(Berman & Laitin, 2008; Carvalho, 2019, 2020). In general, religious displays can serve a
function (among many possible functions) of a costly signal that reveals who is, or is not,
committed to the cause (Atran & Henrich, 2010; Atran & Norenzayan, 2004). This enables
religious groups to internally solve coordination problems, recruit and promote people who are
the most committed, and externally project credible resolve.
Researchers have also found that, across multiple countries and groups, perceptions of the
spiritual formidability of an armed group are an important aspect of the group’s military efficacy
47
and will to fight (Gómez et al., 2017, 2023; Tossell et al., 2022). This research included studies
of ISIL/DAESH fighters, which found that they perceived their group to have high spiritual
formidability. This spiritual formidability helps to explain how this group achieved military
successes out of proportion to their relatively low material strength.
This same reasoning could be applied to the Taliban, or to the Viet Cong, who had far
less material strength than the United States military, but nonetheless ultimately overcame
American opposition and conquered their respective countries. In the end, these conflicts ended
not with the Viet Cong or Taliban outfighting the American army, but with them outlasting the
American government’s will to fight. This is consistent with the argument that, in conflicts that
are tests of resolve, religion and sacred values can serve to promote and credibly project resolve.
3.3.4 Psychological attachment to territory
A psychological attachment to a territory, including a perception that the territory is
indivisible, is another common aspect of international violence and war (Senese & Vasquez,
2005; Vasquez, 1996, 2001; Vasquez & Leskiw, 2001). This association is so prominent that
some international relations researchers even argue that humans are biologically predisposed to
be territorial and to be willing to fight to defend territory (Senese & Vasquez, 2003; Vasquez,
2010; Vasquez & Henehan, 2001).
In many cases of international conflict, a territory is described as sacred or indivisible by
one or both sides of the conflict (Goddard, 2006; Toft, 2002). The claim that a territory is
indivisible has played a prominent role in conflicts over Northern Ireland, Jerusalem, Kashmir,
and Kosovo, and this list could be added to indefinitely.
Similarly to honor and religion, the tendency to feel a psychological attachment to a
territory, and to hold the attitude that that territory is indivisible, can be explained in terms of
48
game theoretical interactions (Johnson & Toft, 2014, specifically, these authors used the hawk-
dove game).
Territory itself is often a useful resource for defending against attacks, or for staging
attacks, e.g., as a source of war materials, or by providing the high ground. Giving up land with
these characteristics renders one even more materially vulnerable to future attacks and predation
(Fearon, 1995).
These conditions are, in the animal kingdom, an explanation for the evolution of
territoriality (Johnson & Toft, 2014). This is because, under these conditions, there is a value
asymmetry in which the value of the territory that one holds is greater than the value of territory
that one does not hold. With respect to territory that one holds, one has the bonus of a defender’s
advantage, while with respect to territory that one does not hold, one has an attacker’s
disadvantage. This logic suggests that individuals who are particularly willing to fight over
territory on which they are well-established will have a competitive advantage.
Under conditions in which annexing part of a territory makes it easier to annex more of
that territory, the territory can become effectively indivisible. Either a country combats any
attempt to annex any part of their territory, or the country risks losing all of it. This logic
provides a distal reason for why people have a tendency to consider territory to be indivisible,
especially under conditions of intergroup conflict.
In addition, territory is something that an adversary can take in pieces, such that it is not
worth fighting to defend any particular piece (in itself, as such). Similarly to the logic concerning
honor, if a group of people allow pieces of their land to be taken, without fighting, then they lose
reputation, and will be more likely to be predated upon again.
49
Again, although I have cited literature concerning biological evolution and biological
predispositions, I do not have any theoretical commitment as to whether these distal causes of
attachment to territory are conscious, or cultural evolution, or biological evolution.
3.3.5 Integrative complexity?
As discussed above, several psychological characteristics may serve functions with
respect to addressing coordination problems and problems of uncertain resolve. Given this
pattern, it is plausible that variation in the level of cognitive complexity with which human
beings address problems, both as a trait component across persons, and as a state component
across situations, may also serve functions with respect to addressing coordination problems and
problems of uncertain resolve.
Consistent with this, the cognitive manager model, which proposes that human beings
exhibit adaptive (but not necessarily optimal) changes in integrative complexity, depending on
what the situation calls for, suggests that different classes of problem could call for different
levels of integrative complexity (Suedfeld, 1992).
The reasoning in the previous subsections suggest that the psychological characteristics
of honor, religion, and attachment to territory, can be explained in terms of how they function to
address the problem of projecting credible resolve. These psychological characteristics can be
thought of in terms of domain-specific content. However, one could also ask whether there is a
broader psychological characteristic that is associated with them in general. If there is such a
characteristic, then this would be consistent with it serving (or it having served under previous
developmental, evolutionary, or cultural conditions) a function with respect to projecting
credible resolve. I suggest that low integrative complexity may be such a psychological
characteristic.
50
People tend to be unwilling to compromise vis-à-vis questions of honor (Tetlock, 2003),
or to compromise or agree to live-and-let-live with respect to violations of sacred values (Haidt,
2001; Tetlock, 2003), or to agree that adversaries have a right to a share of “indivisible” land
(Goddard, 2006), suggesting that low integrative complexity is a general psychological
characteristic across these domains. If so, then low integrative complexity may serve a function
with respect to projecting credible resolve.
On the other side of this coin, given that integrative complexity is not uniformly low
across all situations, or across all persons, it is worth asking whether high integrative complexity
might serve a function in other contexts. The evidence that high integrative complexity is
associated with cognitive ability, and with open-mindedness, suggests that it could contribute to
a person’s ability to solve complex problems. This may be especially the case in contexts that
involve complex interactions, or that require understanding other persons’ perspectives. If so,
then high integrative complexity may serve a function with respect to solving coordination
problems.
In the chapter below, I use this reasoning to develop two hypotheses concerning how the
level of integrative complexity of political leaders relates to their ability to solve coordination
problems and problems of uncertain resolve, and by doing so, to avoid conflicts becoming
violent.
51
Chapter 4: My Hypotheses
4.1 The Strategic Judgment Hypothesis
The first hypothesis, which I call the “strategic judgment hypothesis,” is that leaders who
are higher in integrative complexity are less likely to be involved in violence in international
crises, because their strategic judgment enables them to resolve coordination problems without
resorting to violence.
By definition, high integrative complexity is characterized by the recognition of the
relevance of multiple variables, or of the legitimacy of multiple perspectives (such as those of the
leaders of the adversary country), and the recognition that these could interact (Baker-Brown et
al., 1990). We should therefore expect a high complexity leader to be better able to identify
relevant variables, understand the perspective of the adversary, and appreciate the complex
structure of interaction between the two countries, and to use this knowledge to find ways to
achieve their goals without the use of violence.
This effect of high integrative complexity could operate through the first two
explanations of war: (1) High integrative complexity, because it is characterized by recognition
of the legitimacy of multiple perspectives, should reduce a leader’s tendency to see an issue as
not admitting to rational debate or to compromise, making peaceful resolution more likely; (2)
High integrative complexity should allow a leader to better understand the structure of
interactions involved in international coordination problems, and therefore to de-escalate when
countries fail to commit to coordinated solutions, and to successfully find strategies to peacefully
encourage commitment to coordinated action.
There is empirical evidence that is consistent with the hypothesis that high integrative
complexity is associated with less use of violence in international crises.
52
Research on leaders’ state integrative complexity has found that complexity decreases
prior to surprise attacks and war (Conway et al., 2001; Suedfeld et al., 1993; Suedfeld & Bluck,
1988; Suedfeld & Tetlock, 1977, 1977). Conversely, research has found that in peace
negotiations between the Mexican government and Chiapas rebels, complexity was stable or
increased during periods when the negotiations were making progress (Liht et al., 2005). This
suggests that high state complexity is associated with a decreased propensity to use violence, as
well as with success in negotiating an end to violence.
A study of revolutionary leaders found that, after taking power, and therefore finding
themselves in a situation that requires forging compromises across multiple classes and interest
groups, those leaders who succeeded in this new role exhibited an increase in their integrative
complexity. In a sample of political and military geniuses from throughout history, the geniuses’
baseline integrative complexity was higher than usual for modern leaders, and some of the
geniuses exhibited stable or rising integrative complexity during crises (Suedfeld, 2014). These
findings are consistent with the mechanism that persons with higher integrative complexity are
better able to understand and navigate complicated interactions, such as international
coordination problems.
Among extremist groups, low integrative complexity is associated with increased
acceptance of violence (Cross et al., 2018; Smith et al., 2008; Suedfeld et al., 2013, 2020), and
there is evidence that interventions that aim to increase complexity can prevent violent
extremism (Liht & Savage, 2013). This is consistent with the mechanism that persons with
higher integrative complexity are less likely to frame issues in black-and-white terms that admit
no compromise, and are therefore less likely to resort to violence.
53
These findings lend support to the strategic judgment hypothesis, and suggest that it
operates both through the mechanism that leaders with high complexity are less likely to frame
issues such that they admit to no compromise, and through the mechanism that leaders with high
complexity are better equipped to successfully solve complicated coordination problems.
4.2 The Demonstration-of-Resolve Hypothesis
The second hypothesis, which I call the “demonstration-of-resolve hypothesis” is that
leaders who are higher in integrative complexity are more likely to be involved in violence in
international crises, because they are less able to credibly demonstrate resolve, making them less
able to dissuade adversaries from challenging them.
In contrast to high integrative complexity, low complexity is characterized by black-and-
white thinking, the failure to recognize the legitimacy of alternative viewpoints, and inflexibility,
or in positive terms, firmness and steadfastness (Suedfeld, 2010). This suggests that, relative to
low complexity leaders, high complexity leaders should be less able to demonstrate a credible
commitment to carry out dangerous or costly strategies. As a result, integrative complexity
should operate through the third explanation of war: (3) Low integrative complexity should
increase a leader’s ability to credibly communicate resolve, increasing their ability to deter
aggression, and in games of brinkmanship increasing their ability to convince the adversary to
back down.
Compared to the strategic judgment hypothesis, there is less empirical evidence for the
demonstration-of-resolve hypothesis, but there is some evidence.
Prior to the Second World War, when each of them was discussing Germany,
Chamberlain had consistently higher integrative complexity than Churchill (Tetlock & Tyler,
1996). In the counterfactual in which Churchill were Prime Minister instead of Chamberlain, it is
54
plausible that Churchill would have been able to deter Nazi aggression, or more likely that his
steadfastness would have brought the conflict to a head earlier, while the allies held the
advantage. While this would not have prevented war, it might have contained it to a limited and
local war in central Europe, reducing violence in that way.
In the actual fact, Chamberlain attempted to appease Hitler by agreeing to allow him to
annex the Sudetenland region, which Hitler later used to destabilize and annex much of the rest
of Czechoslovakia. When he announced the agreement, Chamberlain declared that he had
“secured peace for our time” (Wheeler-Bennett, 1963), while Churchill described the agreement
as “a total and unmitigated defeat.” Churchill went on to articulate that it was a defeat because,
among other reasons, it was a failure to demonstrate resolve: “£1 was demanded at the pistol's
point. When it was given, £2 were demanded at the pistol's point….[T]he maintenance of peace
depends upon the accumulation of deterrents against the aggressor” (Churchill, 1938). Although
only a single case, this illustrates how a leader who is relatively low in cognitive complexity
could more credibly communicate resolve, and therefore more effectively deter aggression.
There is one recent study that used the cognitive complexity of Presidents of the United
States as a predictor of violence in those crises in the International Crisis Behavior dataset that
involved the United States (Keller et al., 2021). This study used the Leadership Trait Analysis
(LTA) system to score conceptual complexity (a trait cognitive complexity variable), from The
Public Papers of the Presidents of the United States. It found that, after controlling for relevant
structural variables, presidents’ conceptual complexity was a significant positive predictor of the
centrality of violence in the crisis (i.e., that violence was more important relative to other crisis
management strategies). This pattern is particularly striking because the United States was so
much more militarily powerful than most of its opponents in international crises, such that these
55
opponents knew that the United States had the capability to win in a military confrontation,
reinforcing the argument that escalation of violence was driven by the perception that the
American President lacked the resolve to use this military capability.
This is consistent with the hypothesis that, for political leaders, integrative complexity is
negatively associated with demonstration of resolve, and that as a result it is positively associated
with adversaries employing greater escalations of violence.
56
Chapter 5: Method
5.1 Scoring Integrative Complexity Using Auto IC
The explanatory variable in this research is integrative complexity (IC). Traditionally,
integrative complexity is manually coded from texts by trained coders, following detailed
instructions in a coding manual (Baker-Brown et al., 1990). However, the current research uses a
sample size of paragraphs that is several orders of magnitude larger than is practicable using
manual scoring. In order to manage this magnitude of scoring, this project uses an automated
scoring system, namely Auto IC (Conway et al., 2014, 2020).
There have been questions concerning the validity of automated scoring of integrative
complexity (Suedfeld & Tetlock, 2014; Tetlock et al., 2014). Previous research has found that, at
the paragraph level, there was a modest relationship between manually scored integrative
complexity and Auto IC, with an average correlation of .46, and that this remained statistically
significant even after controlling for superficial indicators of complexity (Conway et al., 2014).
While this average correlation is modest, it is at the paragraph level, and more aggregated data
could have much higher correlations, as non-biased random error averages out.
In my own previous research, I have found that, when sufficiently aggregated, Auto IC
does indeed have much higher correlations with manually scored integrative complexity. For
instance, in a study of Bashar al-Assad’s integrative complexity, I found that when the data are
aggregated to the month level, the correlation between manually scored integrative complexity
and Auto IC is .80 (this analysis is currently unpublished, but the original collection and scoring
of texts was for Suedfeld et al., 2014). In a study of the integrative complexity of the Irish
Republican movement during the Northern Irish peace process, I found that, when aggregated to
the two-month level, the correlation between manually scored integrative complexity and Auto
57
IC was .66, and when aggregated to discrete stages of the peace process, the correlation between
them was .94 (currently unpublished research).
These correlations suggest that, when aggregated to the level of international crises or
militarized interstate conflicts, and with a sufficiently large loading of paragraphs per
crisis/conflict, Auto IC is a valid measure of the same construct as manually scored integrative
complexity. Further evidence of the validity of Auto IC is that it has passed several tests of
construct validity, including that classical philosophical works have higher Auto IC than texts
written by the general population, and that it replicates several findings from manually scored
integrative complexity, such as the relationship between political ideology and integrative
complexity (Conway et al., 2020).
The Auto IC software allows for two approaches to the scoring unit: (a) using the natural
language paragraph as the unit that is scored, or (b) breaking texts up into 75 word chunks, and
using these chunks as the unit that is scored (Conway et al., 2014). An advantage of
standardizing the length of the unit scored is that it removes the possibility that variance in the
paragraph length contributes to variance in IC scores, i.e., that long paragraphs have higher
scores than short paragraphs. On the other hand, using natural language paragraphs has the
advantage of being more closely linked to units of meaning in the text, and of being consistent
with how IC is manually scored. In order to capture the advantages of both approaches, I use
natural language paragraphs, but limit the range of paragraph length by dropping paragraphs with
fewer than 10 words, and by dropping, or breaking up, paragraphs of over 300 words. I describe
in more detail how this was done in the subsection titled “Preparing and Cleaning Paragraphs.”
58
5.2 The Sample of Countries, Leaders, and Verbal Materials
5.2.1 The sample of countries
The three countries included in this study are the United States, the Soviet Union/Russia,
and the United Kingdom. These three countries were selected because they are the three
countries that were involved in the most crises in the International Crisis Behavior dataset (ICB
Project - International Crisis Behavior, n.d.). This is therefore the largest sample of international
crises that could be achieved from only three countries.
Another advantage is that these are the three countries that have been pre-eminent great
powers in the 19th and 20th centuries, with the United Kingdom as the pre-eminent great power
before the First World War, and the United States and Soviet Union the pre-eminent great
(super) powers after the First World War. As a result, using these three countries gives results
that are representative of the pre-eminent great powers of the modern international system.
5.2.2 The sample of leaders
I attempted to sample, as nearly as was practicable, all of the de facto heads of
government of these countries, from 1816 (the first year of the MIC dataset) to the near-present.
For the United States this is the President. For the United Kingdom this is the Prime Minister.
For the Soviet Union this is the person in charge of the Communist Party, which is usually the
General Secretary, although this is not always the case, e.g., in the case of Lenin.3 In situations in
which the Soviet Union was ruled by a troika, the member who ultimately established
themselves as the sole head of government was coded as the de facto head of government during
the troika. In post-Soviet Russia, the President is coded as the head of government, with the
3 The office that is the de facto head of government is not the de jure head of government, the Premier of the
Soviet Union, whose powers were largely nominal.
59
exception of when Putin was Prime Minister (and Medvedev was President), in which case Putin
is coded as the de facto head of government.
In cases in which power was transferred on a given day, such that there were two heads
of government on that day, e.g., one in the morning and another in the evening, then the latter
was coded as the head of government on that day. The reason for this is because the incoming
leader is more likely to speak on that day, and is the one who will make subsequent policy
decisions that could potentially be predicted by their integrative complexity on that day. There
can be days in which there is no head of government, such as days when a Prime Minister has
not yet been appointed.
Every American president is included who held office from 1816 to 2019, which is
Madison to Trump, and accounts for 41 presidents.4 Every Soviet/Russian de facto head of
government is included who held office from November 1917 to 2019, and accounts for 10
leaders. These are Lenin, Stalin, Malenkov, Khrushchev, Brezhnev, Andropov, Chernenko,
Gorbachev, Yeltsin, and Putin. Russian leaders from before Lenin were not included due to the
limited availability of texts from before the Russian revolution. Every British prime minister is
included who held office from 1816 to July 2019, which is Liverpool to May, and accounts for
40 prime ministers.5
4 There are actually data from all American Presidents from George Washington to Biden, but because the
crisis/confrontation datasets do not cover prior to 1816, or after 2019, the presidents who held office outside of
those dates are not relevant to the analyses.
5 There are actually data from all British Prime Ministers as far back as William Pitt the Younger in 1804, with the
exception of the Duke of Portland whom I could not find in the Hansard, but because the crisis/confrontation
datasets do not cover prior to 1816, or after 2019, the prime ministers who held office outside of those dates are
not relevant to the analyses.
60
5.2.3 The sample of Verbal Materials
The verbal materials from the American presidents were collected by web-scraping the
American Presidency Project (American Presidency Project, n.d.) website for all documents
attributed to each person who has held the office of president. The corpus held in this website is
extremely extensive and includes a wide range of types of publicly available documents,
including for the earlier presidents. To give an idea of how extensive the corpus is, as of the
current writing it contains 161,511 documents.
Not all documents that the American Presidency Project attributes to a given president
contain text that was entirely or primarily written by him. For instance, it includes conversations
and interviews, in which there are words from other speakers. It also includes types of text that
are written in such legalistic language that it is highly unlikely that the president was responsible
for the choice of words. I addressed this by filtering by type of document, and by attributes of
documents, to remove types and attributes that are associated with verbal materials that appear to
not be primarily words chosen by the president (for the list of exclusions, see Appendix A).
The verbal materials from the Russian/Soviet leaders were collected from a number of
sources. The Marxists Internet Archive (Marxists Internet Archive, n.d.) was web-scraped for
texts attributed to Lenin and Stalin. The Digital Archive of the Wilson Center was manually
searched and collected from, for texts attributed to several Russian/Soviet leaders (Wilson Center
Digital Archive, n.d.).
Another type of source was anthologies of speeches and writings from individual leaders
– this type of source was particularly important for Brezhnev, Gorbachev, and Yeltsin.6 The
website of the President of Russia (President of Russia, n.d.) was manually searched and
6 The list of anthologies is available by contacting the author.
61
collected from, for texts attributed to Putin. My colleague, Zlatin Mitkov, played a vital role in
the collection of texts from the anthologies of speeches and writings, and from the website of the
President of Russia, as part of a collaboration between us to build a corpus of political texts.
Finally, for Putin, texts were also used that had been collected in an earlier project that I
worked on in Peter Suedfeld’s lab (Suedfeld & Morrison, 2016), and which used a number of
sources, including the website of the President of Russia and transcripts of speeches in media
reports. In addition, the Open Source Center (now called Open Source Enterprise) kindly
allowed us to access their archive to search for political texts. Having found these texts, our
policy was to trace them back to their original source online.
The verbal materials from the British prime ministers were web-scraped from the
Hansard of the Parliament of the United Kingdom (Hansard, n.d.), for all texts attributed to
persons who held the office of Prime Minister. The Hansard can refer to the person speaking by
their office, e.g., “Prime Minister,” or by their name, which could be either the family name, or a
name of nobility, e.g., “Liverpool,” “Portland,” “Wellington,” which can change throughout the
person’s life. In order to be certain to capture all instances in which a person spoke, I scraped the
Hansard for each of their names, in the range of dates during which they held the given name,
and for the term “Prime Minister,” during the range of dates during which they were prime
minister.
Then, for each person, I wrote a customized R-script to do the following: Remove
duplicates. Remove false positives, such as cases in which the speaker’s first name, or the office
that they held, indicated that they were the wrong person. Remove instances in which the House,
i.e., Commons or Lords, did not match the House that the person of interest was currently sitting
in.
62
5.3 Preparing and Cleaning Paragraphs
The verbal materials from the Russian/Soviet leaders, because they came from several
types of sources, sometimes required additional preparation and cleaning. Some of the
documents collected for the Russian/Soviet leaders were only available as PDF files. These were
converted to plaintext files using the optical character recognition software ABBYY Finereader.
A small number of documents were only available in Russian, and these were translated to
English using Google Translate. All of the documents from Russian/Soviet leaders were
manually cleaned. This cleaning involved removing the words of speakers other than the leader
of interest, and removing words, sentences, paragraphs, or pages that were, for whatever reason
(e.g., poor scanning, poor optical character recognition, or failure to translate), too garbled to be
of use.
In order to prevent the length of paragraphs from having an undue influence on the Auto
IC scoring, I set a limit to the paragraph length of 10 to 300 words. To enforce this limit, I
removed paragraphs with fewer than 10 words prior to performing analyses. In the case of the
limit of 300 words, for the verbal materials from the Russian/Soviet leaders, the research
assistants who manually cleaned these materials were instructed to break up paragraphs that were
over 300 words (or over a pager long) into smaller natural language paragraphs. In addition, for
verbal materials from both the American and Russian/Soviet leaders, any remaining paragraphs
with more than 300 words were removed before performing analyses. This accounted for a very
small percentage of paragraphs in these cases.
Unfortunately, I later found that this simple solution would not work for the British
leaders. In the verbal materials scraped from the British Hansard there were no natural language
paragraph breaks, resulting in paragraphs with more than 300 words occurring with very high
63
frequency. To address this, I wrote an R-script to break up paragraphs with more than 300 words
into chunks of 85 words,7 and used these chunks as new paragraphs. I left paragraphs with fewer
than 300 words unchanged.
The result of the above process is that the Auto IC scores that are included in the analyses
are all from paragraphs with word counts from 10 to 300 words.
5.4 Coding Texts for the Topic of Foreign Policy
It is plausible that leaders exhibit different levels of integrative complexity with respect
to different topics, and it is also plausible that dynamic changes in a leader’s integrative
complexity with respect to one topic might not be mirrored with respect to a different topic.
Given that the hypotheses that the current study is testing are about leaders’ cognitions
concerning foreign policy, or about how adversaries make inferences about their likely foreign
policy behavior, it is the leader’s integrative complexity concerning the topic of foreign policy
that is most relevant.
In order to remove paragraphs that have no relevance to the topic of foreign policy, I
coded paragraphs for whether they contained a word relevant to foreign policy, or to a foreign
country, and removed paragraphs that did not contain at least one of these words. I used the
content analysis software LIWC-22 to do this coding (Boyd et al., 2022), using dictionaries that I
wrote for (a) words relevant to foreign policy, and (b) words relevant to various countries (see
Appendix B).
In the case of words relevant to various countries, the scores for words relevant to the
leader’s own country had to be removed, because they are not words referring to a foreign
7 I set it to 85 words because in the Auto IC documentation, paragraphs of 75 words is described as ideal, and after
empirical testing I found that, when measuring word counts for the same paragraphs, my R-script gave slightly
higher word counts than Auto IC does, such that 85 words in the former is a rough approximation of 75 words in
the latter.
64
country. For American presidents, this was the scores for words relevant to the United States,
and for the British prime ministers, this was the scores for words relevant to the United
Kingdom. For leaders of Russia/the Soviet Union, scores for words relevant to Russia, or the
Soviet Union simpliciter, were removed, for all time periods. In addition, scores for words
relevant to the non-Russian constituent republics of the Soviet Union, such as “Belarus” and
“Ukraine,” were removed for time periods when the Soviet Union was in existence, and were
retained (as indicators that the paragraph mentions a foreign country) for time periods after the
dissolution of the Soviet Union.
5.5 The Datasets of Confrontations and Crises
This study uses two pre-existing datasets, the Militarized Interstate Confrontations (MIC)
(Gibler & Miller, 2023) dataset and the International Crisis Behavior (ICB) dataset (ICB Project
- International Crisis Behavior, n.d.). For the purposes of this study, both of these datasets have
the advantage of coding the characteristics of international interactions, namely militarized
confrontations and international crises, that do not necessarily involve violence, but that do
involve a heightened risk of violence. This makes them ideal for testing how the psychological
characteristics of political leaders predict the use of international violence, under conditions in
which it is particularly likely.
The Militarized Interstate Confrontations dataset is relatively new, having been published
in 2023 (Gibler & Miller, 2023). However, it began as a modification of the Militarized Interstate
Disputes dataset (Palmer et al., 2022), which is a part of the broader Correlates of War Project
(Sarkees & Wayman, 2010), and which is well-established in the research literature. The reason
why I chose to use the MIC dataset, rather than its predecessor, is that the MIC has numerical
estimates of the number of fatalities in every confrontation in the dataset.
65
The MIC includes confrontations from 1816 to 2014. Examples of militarized interstate
confrontations include: the Andrew Jackson Raids During Seminole War (1818); the Anglo-
French Blockade of Argentine Confederation (1845-1846); the Crimean War of 1853-1856; the
Anglo-Persian War of 1856-1857; the Start of the American Occupation of Haiti (1914-1915);
the Soviet Seizures of British Ships (1923); the British Moves to Keep a Presence in the Suez
Canal (1951-1952); the Soviet Blockade of Estonia (1939); the Soviet Invasion of Poland (1939);
the Soviet Takeover of Latvia (1940); the Berlin Blockade (1949); the Allied-Soviet Air
Incidents (1953); the Taiwan Straits War of 1958; the U2 Flights [Gary Powers] (1960); the Bay
of Pigs (1960-1961); the Berlin Crisis (1961); the Cuban Missile Crisis (1962); and the Russo-
Georgian War over South Ossetia and Abkhazia (2001-2010).
The International Crisis Behavior dataset is well-established in the research literature
(Brecher, 1993, 2008, 2018; Brecher & Wilkenfeld, 1997). An advantage of this dataset is that it
codes for variables both at the level of the foreign policy crisis (characteristics of a particular
country and of its participation in the crisis) and at the level of the international crisis
(characteristics of the international system and international interactions relevant to the crisis).
Another advantage is that it codes for a large number of variables, allowing for more statistical
tests and more statistical controls than is the case for the MIC.
The ICB includes crises from 1918 to 2019. Examples of international crises include: the
Russian Civil War I (1918); Baltic Independence (1918); Entry Into WWII (1939); Soviet
Occupation of the Baltic (1939); Pearl Harbor (1941); Communism in Hungary (1947); the
Truman Doctrine (1947); the Berlin Blockade (1948); Soviet Bloc-Yugoslavia (1949); Suez
Canal (1951); Taiwan Strait I (1954); the Suez Nationalization-War (1956); the Hungarian
Uprising (1956); the Berlin Deadline (1958); the Bay of Pigs (1961); the Berlin Wall (1961);
66
Soviet Note to Finland II (1961); Cuban Missiles (1962); Cod War I (1973); Cod War II (1975);
US Hostages in Iran (1979); Contras I (1981); Invasion of Granada (1983); Able Archer (1983);
Iraq No-Fly Zone (1992); North Korea Nuclear III (2006); Russo-Georgian War (2008); Syria
Chemical Weapons (2013); and Crimea Donbass (2014).
5.6 The Analytical Approach – Regressions with Controls
The analytical approach of the statistical tests in this study is to carry out regressions in
which the leaders’ integrative complexity is used as a predictor of a variable of interest (e.g.,
initiation of violence, severity of violence, number of fatalities). These regressions include
controls for multiple variables that international relations theories, and integrative complexity
theory, suggest could be relevant. These include the time period / structure of the international
system, the power discrepancy between the two sides, and whether the crisis occurs in the
context of a protracted conflict (also known as an enduring rivalry). In addition, the regressions
include controls for multiple variables that are relevant to the level of stress experienced by the
leader, including (again) the power discrepancy between the two sides, the gravity of the threat,
and the distance of the location of the crisis from the leader’s country.
The unit of these regressions is the crisis/confrontation. In other words, the N is the
number of crises/confrontations, and in the dataset on which the regression is performed, each
row is associated with a single crisis/confrontation.
A possible drawback of this approach is that if the relationship between the leader’s
integrative complexity and the predicted variable is mediated by whether a crisis/confrontation
has occurred, then limiting the analyses to only cases in which a crisis/confrontation has
occurred is blocking the causal pathway in which crises/confrontations did not occur (see Rohrer,
2018).
67
While real and worth being aware of, this drawback is not particularly severe. This is
because international crises, and militarized interstate disputes, are by definition the situations in
which the predicted variables can occur, or at least the situations in which they are likely to
occur. For instance, the degree of success achieved at the outcome of a crisis means relatively
little in situations in which there was no crisis. The same is true of the degree of reliance on
violence as a crisis management strategy. International violence, simpliciter, can and does occur
outside of international crises, but if that international violence reaches such a high level that it
disrupts the previous pattern of relations between countries, then a crisis has been initiated. As a
result, limiting the analyses to only crises/confrontations is not blocking a causal pathway that
can account for much of the variance in the predicted variables.
In addition, alternative analytical strategies that do not focus on crises/confrontations
would have severe drawbacks. One of these drawbacks is that high levels of international
violence are relatively rare. When it comes to periods of time, or dyads of countries, in which
there is no (or very little) international violence, the integrative complexity of the leaders of the
countries will still change over time, presumably in association with other events and behaviors.
Failing to limit the data to situations in which violence is particularly likely, e.g., crises or
confrontations, would therefore be including a large number of observations in which there is
little theoretical reason to expect a relationship between the leaders’ integrative complexity and
international violence.
The analytical strategy with less costly drawbacks is therefore to use the
crisis/confrontation as the unit of observation in the regressions, as was done in the research
presented in this paper.
68
A final caveat concerning this approach is that it does not tell us what the relationship is
between a leader’s integrative complexity and the likelihood that they will become involved in a
crisis/confrontation. That question, while also interesting and important, is left to future research.
Any empirical relationships identified in this research therefore might not apply when
broadening the scope outside of crises/confrontations. For instance, a variable that is associated
with more violence within a crisis could, hypothetically, also be associated with a decreased
incidence of crises, and therefore with less violence. When interpreting the results, the reader
should keep in mind that the present research takes the presence of a crisis/confrontation as a
given.
69
Chapter 6: Leaders’ Integrative Complexity and Militarized Interstate Confrontations
6.1 Coding of the Militarized Interstate Confrontation (MIC) Variables
6.1.1 Country
In the MIC dataset this is coded from the variable “ccode”, which is the numeric country
code, and which uses the same codes as the Correlates of War (COW) project. The original
numeric country codes are: 002 = United States, 200 = United Kingdom, 365 = Russia/USSR. I
abbreviate the countries using the International Standards Organisation (ISO) three letter country
codes, which are USA = United States, GBR = United Kingdom, RUS = Russia/USSR.
6.1.2 Time period
The coding of this variable is based on the coding of the same variable in the analyses of
the International Crisis Behavior data, and uses the same categories.
This variable has categories for three time periods: Pre Cold War (before 1945-09-03);
Cold War (1945-09-03 to 1989-12-31); and Post Cold War (after 1989-12-31). Which category a
confrontation falls into is based on the day that the confrontation began.
If there are any effects of time, as such, e.g., if, in recent years, leaders’ mean IC were
lower, or international crises were more frequent, then this coding of time period would also
capture these effects.
As well as controlling for time, as such, these categories control for the structure of the
international system, with Pre Cold War having a multipolar structure of multiple great powers,
the Cold War having a bipolar structure of competing superpowers, and Post Cold War being
relatively (not completely) unipolar, with the United States as the predominant power. If the
70
structure of the international system is associated with different patterns in the use of military
activity, force, or war, then these categories will take that association into account.
It is important to control for the time period, and for the structure of the international
system, because these involve variables that are of fundamental importance in international
relations theory. International relations realism emphasizes the power structure of the
international system, e.g., unipolar, bipolar, or multipolar (Deutsch & Singer, 1964; Gilpin,
1981; Ikenberry et al., 2009; Mearsheimer, 2001; Waltz, 1979; Wohlforth, 1999). International
relations liberalism emphasizes domestic and international institutions (Doyle, 2005; Doyle &
Sambanis, 2000; Keohane, 1984, 1988; Maoz & Russett, 1993; Oneal & Ray, 1997; Russett,
1993), which change over time. International relations constructivism emphasizes norms (Reus-
Smit, 1999; Ruggie, 1982; Wendt, 1999), which also change over time.
6.1.3 During world war
This is coded as No = 0, and Yes = 1. It is coded as yes (1) if the confrontation began
during either the First World War (1914-07-29 to 1918-11-11), or the Second World War (1939-
09-02 to 1945-09-02).
6.1.4 The side that initiated the first military action
This variable is coded in the MIC dataset, from the variable “sidea” which is coded as 1 if
the country was on the side that initiated the first militarized action, and 0 if not. This includes
cases in which the military action was initiated by an actor that was not the leader’s country, but
that was on the same side of the confrontation as the leader’s country.
71
6.1.5 The level of hostility employed by the leader’s country
This variable is coded from the MIC dataset, from the variable “hostlev”. I coded it so
that 1 = No use of force (this merges three categories from the original MIC variable, namely No
Militarized Action, Threat to use force, and Display of force); 2 = Use of force; and 3 = War.
6.1.6 Fatalities (in thousands) suffered by the leader’s country
This variable is coded from the MIC dataset, by taking the mean of two variables:
“fatalmin”, which is the minimum estimate of the number of fatalities that the leader’s country
suffered during the confrontation, and the variable “fatalmax”, which is the maximum estimate
of the number of fatalities that it suffered. This mean was divided by 1,000, in order to express
the number of fatalities in terms of thousands.
6.1.7 Degree of Success at Outcome
This code is generated from the MIC variables “outcome”, and “sidea”. The variable
“sidea” indicates whether the leader’s country was on the side that initiated the militarized
activity. The variable “outcome” indicates what the outcome was with respect to countries on
each side. These outcomes were: Victory for Side A, Victory for Side B, Yield by Side A, Yield
by Side B, Stalemate, Compromise, Released, Unclear, Joins Ongoing War. I coded Defeat for
the leader’s country if the leader’s country is Side A, and the outcome was Victory for Side B, or
if the leader’s country is Side B, and the outcome was Victory for Side A. I coded Victory for the
leader’s country for the reverse, i.e., if the side of the leader’s country A or B, and Victory for A
or B, had the same letter. The same logic of coding was applied to code Yield by the leader’s
country, or Adversary Yielded.
I recoded Unclear, Released, and Joins Ongoing War to NA. The Released category is for
confrontations in which one side took persons or materiel from the other side, and in which the
72
confrontation ended with them being returned. These were coded as NA for two reasons. The
first reason is that the category confounds the issue of the conflict, strategies employed in the
conflict, and the outcome of the conflict, and does not give a code for cases in which persons or
material were taken and never given back. The second reason is that it is not clear which side
took persons/materiel, and which side had them taken (it appears to usually be Side A that did
the taking, but not always), so it is impossible to use this category to code which side was
defeated. Joins Ongoing War was coded as NA because it is impossible to say whether this
ultimately resulted in defeat.
This resulted in the following categorical codes: Defeat, Yield, Stalemate, Compromise,
Victory, Adversary Yielded, as well as NA.
To convert this categorical coding to an interval variable, it was recoded as follows:
Defeat = 1, Yield = 1, Stalemate = 2, Compromise = 2, Victory = 3, Adversary Yielded = 3.
6.1.8 Avoidance of Defeat dummy
I coded a dichotomous outcome variable for defeat for two reasons. The first reason is
that the MIC manual gives a warning that, for the coding of the outcome variable, the coding
rules were not consistently applied, and that the outcome variable is currently being recoded. The
second reason is that, after qualitatively reviewing a pseudorandom sample of confrontations
involving the United States, Russia, and United Kingdom, I came to the conclusion that the
distinction between defeat and not defeat is the most valid distinction for my purposes. It appears
that stalemates, compromises, and yielding, at the level of the confrontation, can be followed by
bigger-picture strategic victories.
This is not necessarily wrong coding on the part of the dataset. However, for my purposes
it undermines the validity of the variable as a measure of the ability of the leader to achieve
73
strategic successes, especially given that one type of complex strategy that may require high
cognitive complexity is a strategy that accepts short term losses for long term gains. Another
problem with this coding, for my purposes, is that in some circumstances avoiding the costs of
conflict by yielding may bring about better overall outcomes than a costly victory would. Among
the outcomes, defeats are the category that are least undermined by this problem, so coding
outcome as a dichotomous variable for defeat is the least noisy and most valid approach. (I made
the same analytical decision for the analyses of the International Crisis Behavior data.)
I coded it as Avoidance of Defeat, rather than defeat, so that the sign of any relationships
with this variable have the same meaning as the sign concerning the variable coding the Degree
of Success at the Outcome.
This Avoidance of Defeat dummy was coded starting with the same categories as were
described above (concerning the coding of the Degree of Success at the Outcome). It was
recoded as follows: Defeat = 0, Avoidance of Defeat (all other categories) = 1, NA = NA.
6.2 Does a Leader’s Integrative Complexity Before the Confrontation Predict Whether
Their Side Initiated the First Militarized Activity?
6.2.1 Predictions
According to the strategic judgment hypothesis, the leader’s pre-confrontation IC should
be negatively associated with the leader’s side initiating military activity, because higher
integrative complexity should be associated with finding ways to achieve one’s goals without
resorting to costly military conflict.
According to the demonstration-of-resolve hypothesis, the leader’s pre-confrontation IC
should also be negatively associated with the leader’s side initiating military activity. To
understand why, keep in mind that the coding of which side initiated militarized activity is
74
dichotomous. It therefore follows that increasing initiation by the adversary side is
mathematically identical to decreasing initiation by the leader’s side. According to the
demonstration-of-resolve hypothesis, IC is negatively associated with adversaries’ perception
that the leader has high resolve, such that leaders who express higher IC will be perceived as
having less resolve, making adversaries more likely to initiate military activity against their side.
In the dichotomous coding of the initiation variable, this is the same as making the leader’s
country less likely to initiate military activity. This is why the demonstration-of-resolve
hypothesis also predicts that pre-confrontation IC should be negatively associated with the
leader’s side initiating military activity.
This is a less direct test of the demonstration-of-resolve hypothesis than it is of the
strategic judgment hypothesis, because the former is more directly about whom the military
activity is initiated against, rather than who initiated it, and the variable codes for which side
initiated it. Although clearly connected, these are not identical: when the initial military activity
was against the side of the leader’s country, it was not necessarily against the leader’s country
itself, e.g., it could be against an ally country.
6.2.2 Results
6.2.2.1 Descriptive statistics
The following analyses have an N of 222 confrontations. The number of paragraphs
coded for integrative complexity, per confrontation, ranges from a minimum of 20 to a maximum
of 1,176, with a mean of 163.62, and a standard deviation of 147.56.
Across the confrontations, the leaders’ pre-confrontation integrative complexity ranges
from a minimum of 1.46 to a maximum of 2.95, with a mean of 2.06, and a standard deviation of
0.26.
75
The variable that represents the initiation of violence by the leader’s side is binary (No =
0, Yes = 1). Across the confrontations it has a mean of .61, and a standard deviation of 0.49.
6.2.2.2 Inferential statistics
In the regression below, the predictor variable is the leader’s mean IC, in the month prior
to the first day of the confrontation. The response variable is whether the leader’s side in the
confrontation was the side that first carried out a militarized action in the confrontation, e.g.,
violence, mobilization of troops, or verbal threat to use violence.
The regression only included confrontations in a subset that met three conditions. The
first condition: That the confrontation have at least 20 paragraphs, from the pre-confrontation
period, that have IC scores. This is so that the regression without IC as a dependent variable
includes the same confrontations as the regressions with IC, making them more directly
comparable. This condition (or the analogous condition requiring at least 20 paragraphs from
during the crisis) applies to all of the regressions in this chapter. The second condition: The
leader’s country was an original participant in the confrontation. This subset was used to exclude
confrontations in which the leader’s country entered later in the confrontation, as in these cases
they may have had little or nothing to do with the interactions that caused the initial use of
militarized action. The third condition: The confrontation did not begin during a world war. This
was because, during world wars, all three of the countries were active participants in large-scale
warfare, as were many of their allies and adversaries, such that their orientation towards the use
of military activity would be radically different from the normal state of affairs.
I included covariates for country (United States, Russia, and Great Britain), and Great
Britain is the baseline country. I did not include covariates for time period, because the average
76
value of the variable cannot vary over time – whenever there is an interstate confrontation, one
of the two sides was the side that initiated it. The results are in Table 1 below.
Table 1
Regressions Predicting Whether Violence was Initiated by the Leader’s Side
(1)
(2)
(Intercept)
0.34
-1.19
p = .25
p = .35
(0.29)
(1.27)
Country RUS
-0.10
-0.33
p = .81
p = .48
(0.42)
(0.46)
Country USA
0.21
0.23
p = .54
p = .50
(0.34)
(0.35)
Preconfrontation
0.76
IC
p = .22
(0.61)
Num.Obs.
222
222
AIC
302.4
302.8
BIC
312.6
316.4
Log.Lik.
-148.185
-147.406
F
0.464
0.808
RMSE
0.49
0.49
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is
in parentheses.
In Table 1, and in all of the subsequent tables, the coefficient is unstandardized. This
analytical decision was made because many of the control variables, and often the predictor
variable and/or predicted variable, are of a type that does not lend itself to standardization (e.g., a
categorical variable, a dichotomous variable).
The statistics in the bottom lines of the tables may require some explanation. The number
of observations (Num.Obs) is the number of confrontations/crises included in the regression. The
Akaike Information Criterion (AIC) is an estimator of prediction error, and therefore lower
77
values indicate better model fit (Burnham & Anderson, 2002). When comparing the AIC across
two similar regression models, a rule of thumb is that a difference of two or greater indicates a
potentially meaningful difference in model fit.
The Bayesian Information Criterion (BIC), log-likelihood (Log.Lik), F value (F), root
mean square error (RMSE), R-squared (R2), and adjusted R-squared (R2 adj.) are not interpreted
in this paper, but are included in the tables for any interested readers.
We can see in Table 1 that there is not a significant relationship between the leader’s pre-
crisis IC and the initiation of militarized activity by their country. In the model that includes IC
as a predictor (model 2), as compared to the model that does not (model 1), neither the AIC nor
the log likelihood are notably different, suggesting that the inclusion of IC does not improve the
model fit.
6.2.3 Discussion
The relationship between the leader’s pre-confrontation IC and whether their country
initiated militarized activity is not statistically significant, so this test provides no support for
either hypothesis.
Judging by AIC and log-likelihood, the inclusion of IC as a predictor variable does not
improve the model fit. There is, therefore, no support for the claim that a leader’s pre-
confrontation IC is associated with their country being the one that subsequently initiated
militarized activity.
The positive relationship between pre-confrontation IC and initiation of violence by the
leader’s side, while not statistically significant, is the same sign of relationship found in similar
tests in the chapter below, using the International Crisis Behavior data. One of those other tests
found a significant positive relationship between the leader’s pre-crisis IC and the initiation of
78
violence by proxies of their country. It may be that the same pattern would hold with respect to
the initiation of militarized action by proxies, but unfortunately it is not possible to test this using
the existing MIC variables.
Overall, these results do not provide support for either the strategic judgment hypothesis
or the demonstration-of-resolve hypothesis.
6.3 Does a Leader’s IC During a Confrontation Predict the Level of Hostility Employed by
Their Country?
6.3.1 Predictions
According to the strategic judgment hypothesis, the leader’s IC during a confrontation
should be negatively associated with the level of hostility employed by their country, because
higher integrative complexity should enable leaders to find ways to achieve their goals, without
costly violence and war.
According to the demonstration-of-resolve hypothesis, the leader’s IC during a
confrontation should be positively associated with the level of hostility employed by their
country, because leaders who express higher levels of IC will be perceived by adversaries as
being less resolved, and will therefore have to employ more extreme measures in order to
convince adversaries to back down.
6.3.2 Results
6.3.2.1 Descriptive statistics
The following analyses have an N of 347 confrontations. The number of paragraphs
coded for integrative complexity, per confrontation, ranges from a minimum of 20 to a maximum
of 22,767, with a mean of 1,010.27, and a standard deviation of 2,222.09.
79
Across the confrontations, the leaders’ confrontation integrative complexity ranges from
a minimum of 1.19 to a maximum of 2.72, with a mean of 2.05, and a standard deviation of 0.23.
The variable that represents the level of hostility exhibited by the leader’s side has the
coding: 1 = No use of force, 2 = Use of force, and 3 = War. Across the confrontations it has a
mean of 1.60, and a standard deviation of 0.69.
6.3.2.2 Inferential statistics
In the regression models below, the explanatory variable is the IC of the leader during the
confrontation. The response variable is the level of hostility. I included covariates for country
(United States, Russia, Great Britain), with Great Britain as the baseline. I also included
covariates for time period (Pre Cold War, Cold War, and Post Cold War), and for whether the
confrontation began during a world war (No = 0, Yes = 1). The regressions only include those
confrontations that have at least 20 paragraphs scored for IC. The results are in Table 2 below.
80
Table 2
Regressions Predicting the Level of Hostility of the Leader’s Country
(1)
(2)
(Intercept)
1.74***
2.71***
p < .01
p < .01
(0.07)
(0.33)
Country RUS
0.13
0.15
p = .21
p = .13
(0.10)
(0.10)
Country USA
-0.09
-0.11
p = .26
p = .20
(0.08)
(0.08)
Period Cold War
-0.28**
-0.32***
p < .01
p < .01
(0.09)
(0.09)
Period
-0.22*
-0.22*
Post Cold War
p = .02
p = .02
(0.10)
(0.10)
During
0.63***
0.58***
World War
p < .01
p < .01
(0.15)
(0.15)
Confrontation IC
-0.46**
p < .01
(0.15)
Num.Obs.
347
347
R2
0.144
0.167
R2 Adj.
0.132
0.152
AIC
683.6
676.4
BIC
710.6
707.2
Log.Lik.
-334.817
-330.178
F
11.489
11.340
RMSE
0.64
0.63
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is
in parentheses.
We can see that the leader’s IC during an interstate confrontation is a statistically
significant (p < .01) predictor of the level of hostility that their country engaged in during the
confrontation, and that the sign of the relationship is negative. In the model that includes IC as a
81
predictor (model 2), as compared to the model that does not (model 1), the AIC is substantially
lower, and the log likelihood is substantially higher, suggesting that the inclusion of integrative
complexity improves the model fit. The contribution of IC to the adjusted R-squared, over that
given by only the covariates, is .02.
6.3.3 Discussion
The statistically significant negative relationship between the leader’s crisis IC and the
level of hostility that their country engaged in is consistent with the strategic judgment
hypothesis, and is not consistent with the demonstration-of-resolve hypothesis.
The lower AIC and higher log likelihood in the model that includes IC as a predictor
suggest that the inclusion of IC improves the model fit. While the contribution that the leader’s
crisis IC makes to the R-squared of the model may appear small, the level of interstate hostility is
generally difficult to predict (Gartzke, 1999), and, relative to the small amount of variance
accounted for by the model without IC, the inclusion of IC is a substantial improvement.
6.4 Does a Leader’s IC During a Confrontation Predict the Number of Fatalities Suffered
by Their Country?
6.4.1 Predictions
According to the strategic judgment hypothesis, the level of IC that a leader expresses
during a confrontation should be negatively associated with the number of fatalities that their
country suffers during the confrontation. This is because, according to this hypothesis, leaders
with higher integrative complexity will be better able to achieve their goals in a confrontation,
without suffering high costs, such as fatalities.
According to the demonstration-of resolve hypothesis, the level of IC that a leader
expresses during a confrontation should be positively associated with the number of fatalities
82
that their country suffers. This is because, according to this hypothesis, adversaries see leaders
with higher IC as having less resolve, and therefore the adversaries will be more likely to
continue to impose fatalities on the leader’s country, in order to pressure the leader into making
concessions.
6.4.2 Results
6.4.2.1 Descriptive statistics
The descriptive statistics for the N of confrontations, number of paragraphs per
confrontation, and the leaders’ confrontation IC are identical to the previous section, so I do not
repeat them here.
Across the confrontations, the number of fatalities suffered by the leader’s country ranges
from a minimum of 0 to a maximum of 6,642, with a mean of 30.71, and a standard deviation of
376.18.
6.4.2.2 Inferential statistics
In the regression below, the explanatory variable is the IC of the leader during the
confrontation. The response variable is the estimated number of fatalities suffered by the leader’s
country. I included covariates for country (United States, Russia, Great Britain). I also included
covariates for time period (Pre Cold War, Cold War, and Post Cold War), and for whether the
confrontation began during a world war (No = 0, Yes = 1). The regressions only include those
confrontations that have at least 20 paragraphs scored for IC. The results are in Table 3 below.
The leader’s IC during the confrontation is a statistically significant (p < .001) predictor of
estimated thousands of fatalities, with a negative sign of relationship. In the model that includes
IC as a predictor (model 2), as compared to the model that does not (model 1), the AIC is
substantially lower, and the log likelihood is substantially higher, suggesting that the inclusion of
83
IC improves the model fit. The adjusted R-squared contributed by IC, above that given by only
the covariates, is .038.
Table 3
Regressions Predicting the Number of Fatalities Suffered by the Leader’s Country
(1)
(2)
(Intercept)
-24.14
677.85***
p = .57
p < .01
(42.00)
(182.50)
Country RUS
114.61*
132.39*
p = .04
p = .02
(56.77)
(55.78)
Country USA
23.71
14.38
p = .61
p = .75
(46.81)
(45.90)
Period Cold War
-7.22
-36.42
p = .88
p = .46
(49.15)
(48.70)
Period
-16.73
-16.19
Post Cold War
p = .76
p = .77
(55.29)
(54.15)
During
425.29***
389.76***
World War
p < .01
p < .01
(83.59)
(82.35)
Confrontation IC
-335.41***
p < .01
(84.96)
Num.Obs.
347
347
R2
0.096
0.136
R2 Adj.
0.083
0.121
AIC
5078.0
5064.5
BIC
5105.0
5095.3
Log.Lik.
-2532.005
-2524.228
F
7.281
8.925
RMSE
357.06
349.15
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is
in parentheses.
84
6.4.3 Discussion
The statistically significant negative relationship between the leader’s IC during the
confrontation, and fatalities suffered by their country, is consistent with the strategic judgment
hypothesis, and is not consistent with the demonstration-of-resolve hypothesis.
The lower AIC and log likelihood in the model that includes IC as a predictor, relative to
the model that does not, suggests that IC improves the model fit. The inclusion of IC as a
predictor accounts for a good amount of additional variance, relative to the variance accounted
for by the model without IC, and for a variable as difficult to predict as fatalities in real-world
interstate confrontations.
6.5 Does a Leader’s IC During a Confrontation Predict the Degree of Success Achieved by
the Leader’s Country at the Outcome?
6.5.1 Predictions
According to the strategic judgment hypothesis, the level of IC expressed by a leader
during an interstate confrontation should be positively associated with the degree of success
achieved by their country at the outcome of the confrontation. This is because, according to this
hypothesis, higher integrative complexity enables leaders to find and employ strategies that
result in greater success.
According to the demonstration-of-resolve hypothesis, the level of IC expressed by a
leader during an interstate confrontation should be negatively associated with the degree of
success achieved by their country at the outcome of the confrontation. This is because, according
to this hypothesis, higher levels of a leader’s IC are associated with adversaries perceiving the
leader as having less resolve, so that they are more likely to persist in imposing costs on the
85
leader’s country, in order to pressure the leader to make concessions, and are less likely to back
down themselves.
These predictions apply not only to defeat, but also to the spectrum of outcomes, from
defeat, to no net gains, and partial gains, and finally to victory. However, there are strong reasons
to believe that the coding of the variable is noisy (see the methods section on the coding of defeat
for more details), and also reasons to believe that, for my purposes, the most valid distinction is
between defeat and not defeat. For these reasons, I do two sets of analyses, one in which the
response variable is the degree of the success achieved by the leader’s country as an interval
variable, and another in which it is a dichotomous variable for the avoidance of defeat.
6.5.2 Results
6.5.2.1 Descriptive statistics
The following analyses have an N of 274 confrontations. The number of paragraphs
coded for integrative complexity, per confrontation, ranges from a minimum of 20 to a maximum
of 22,767, with a mean of 1,103.83, and a standard deviation of 2,433.58.
Across the confrontations, the leaders’ confrontation integrative complexity ranges from
a minimum of 1.51 to a maximum of 2.72, with a mean of 2.05, and a standard deviation of 0.20.
The variable that represents the degree of success (an interval variable) achieved by the
leader’s side at the resolution of the confrontation has the coding: Net Losses = 1, Neutral or
Ambiguous = 2, Net Gains = 3.8 Across the confrontations it has a mean of 2.16, and a standard
deviation of 0.58.
8 Confrontations with outcomes that were coded as the confrontation joining into an existing war could not be
given a value on the new scale, and were coded as NA, and therefore excluded from the analyses. Confrontations
with outcomes that were coded as “return”, meaning that persons or materiel that were taken (i.e., stolen,
kidnapped) were returned were also coded as NA, and therefore excluded, because the MIC coding did not
indicate which country played which role, so it is not clear which country successfully achieved a return of what
was taken, and which country did the taking and returning.
86
The variable that represents the avoidance of defeat is a binary variable (Defeat 0,
Avoidance of Defeat = 1). Across the confrontations it has a mean of .95, and a standard
deviation of 0.21.
In the following inferential statistics, the response variables, namely Degree of Success
and Avoidance of Defeat, have unevenly distributed values. In particular, very few
confrontations are coded as ending Defeat, while roughly two thirds of confrontations are coded
as ending in Compromise or Stalemate. This should be kept in mind when interpreting the
results, because it reduces the amount of variance in the response variables, and therefore
reduces the amount of variance available for the regression models to work with.
Table 4
Distribution of Values for the Degree of Success and Avoidance of Defeat
Outcome Categorical
Degree of
Success
Avoidance of
Defeat
n Confrontations
Valid
Percent
Defeat
1
0
13
0.05
Yield
1
1
15
0.05
Compromise
2
1
32
0.12
Stalemate
2
1
142
0.52
Adversary Yielded
3
1
29
0.11
Victory
3
1
43
0.16
NA
NA
NA
50
NA
Total
324
1
The data presented in this table are for the same subset of crises as for the regressions, namely excluding
crises with fewer than 20 paragraphs scored for IC, and excluding crises that began during a world war.
6.5.2.2 Inferential statistics: Predicting success, as an interval variable
In the regression below, the predictor variable is the IC expressed by the leader during the
interstate confrontation. The response variable is the degree of success achieved by the leader’s
country at the outcome of the confrontation. The regression models include covariates for
country (United States, Russia, and United Kingdom), and for time period (Pre cold War, Cold
87
War, and Post Cold War). Confrontations with fewer than 20 paragraphs scored for IC are
excluded from the regression models.
Confrontations that began during a world war are also excluded from the models, because
neither of the hypotheses makes much sense with respect to predicting outcomes of
confrontations during world wars. With respect to the strategic judgment hypothesis, it may be
that during a world war a strategy that requires higher integrative complexity involves accepting
losses or retreats in a confrontation, as part of the broader strategy to win the war. Or, strategies
that require high integrative complexity could involve limiting one’s successes for the sake of an
ally, such as providing them with resources and equipment, or moving troops to defend them,
even when doing so reduces one’s own capacity to fight. These strategies would result in a
flipping of the relationship that this hypothesis normally predicts.
With respect to the demonstration-of-resolve hypothesis, once a conflict has escalated to
total war on a global scale, there is little opportunity for the adversaries to demonstrate resolve in
order to deter or compel one another, because they are already using their full capacity to destroy
one another.
The results are in Table 5 below.
88
Table 5
Regressions Predicting the Degree that the Leader’s Country Achieved Success
(1)
(2)
(Intercept)
2.25***
2.06***
p < .01
p < .01
(0.07)
(0.37)
Country RUS
-0.17
-0.18+
p = .11
p = .10
(0.11)
(0.11)
Country USA
-0.05
-0.05
p = .52
p = .52
(0.08)
(0.08)
Period Cold War
-0.12
-0.11
p = .17
p = .20
(0.08)
(0.09)
Period
0.08
0.08
Post Cold War
p = .42
p = .41
(0.10)
(0.10)
Confrontation IC
0.09
p = .62
(0.18)
Num.Obs.
274
274
R2
0.029
0.029
R2 Adj.
0.014
0.011
AIC
485.4
487.1
BIC
507.1
512.4
Log.Lik.
-236.698
-236.569
F
1.973
1.625
RMSE
0.57
0.57
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is
in parentheses.
There is a positive sign of relationship between the leader’s IC during the confrontation,
and the degree of success achieved by their country at the outcome of the confrontation, but this
relationship is not statistically significant. In the model that includes IC as a predictor (model 2),
as compared to the model that does not (model 1), the AIC is higher, and the log likelihood is not
substantially different, suggesting that the inclusion of IC as a predictor does not improve the
89
model fit. In the model that includes IC as a predictor, as compared to the model that does not,
the adjusted R-square is lower.
6.5.2.3 Inferential Statistics: Predicting avoidance of defeat, as a dichotomous variable
In the regression below, the predictor variable is the IC expressed by the leader during the
interstate confrontation. The response variable is the avoidance of defeat. The regression
includes covariates for country (United States, Russia, and United Kingdom), and for time period
(Pre cold War, Cold War, and Post Cold War). Confrontations with fewer than 20 paragraphs
scored for IC are excluded from the regression models. Confrontations that began during a world
war are also excluded, for the same reasons as given for the regressions predicting the degree of
success (above). The results are in Table 6 below.
There is a positive sign of relationship between the leader’s IC during the confrontation,
and their avoidance of defeat at the outcome of the confrontation, but this relationship is not
statistically significant. In the model that includes IC as a predictor, as compared to the model
that does not, the AIC is higher, and the log likelihood is not notably different, suggesting that
the inclusion of IC as a predictor does not improve the model fit.
90
Table 6
Regressions Predicting Whether the Leader’s Country Avoided Defeat
(1)
(2)
(Intercept)
2.12***
-0.65
p < .01
p = .83
(0.44)
(3.08)
Country RUS
-0.91
-0.95
p = .17
p = .16
(0.66)
(0.67)
Country USA
0.91
0.81
p = .29
p = .35
(0.86)
(0.88)
Period Cold War
1.85*
1.94*
p = .02
p = .02
(0.81)
(0.83)
Period
1.83+
1.77
Post Cold War
p = .09
p = .11
(1.09)
(1.10)
Confrontation IC
1.37
p = .37
(1.51)
Num.Obs.
274
274
AIC
98.4
99.6
BIC
116.4
121.2
Log.Lik.
-44.186
-43.778
F
3.125
2.729
RMSE
0.20
0.20
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is
in parentheses.
6.5.3 Discussion
For both the regressions predicting the degree of success at the outcome, and the
avoidance of defeat at the outcome, the leader’s integrative complexity during the confrontation
is not a significant predictor of the response variable. Furthermore, in both sets of regressions,
the inclusion of the leader’s integrative complexity did not improve the model fit, judging based
on the AIC and log likelihood. There is therefore little evidence that the leader’s integrative
91
complexity during a confrontation is an important predictor of their success, or avoidance of
defeat, at the outcome of the confrontation.
In both types of regression, while not statistically significant, there was a positive sign of
relationship between the leader’s integrative complexity during the confrontation, and their
degree of success, or avoidance of defeat. This sign of relationship is consistent with the
predictions of the strategic judgment hypothesis, and is not consistent with the demonstration-of-
resolve hypothesis. So, while these results are not significant and are far from conclusive, they
are slightly more supportive of the strategic judgment hypothesis.
It is possible that better coding of the response variables (success / avoidance of defeat),
or the inclusion of controls not currently available, or the inclusion of leaders from more
countries, might yield more definitive results. In particular, the current analyses suffer from the
coding of the outcome variables. The Militarized Interstate Confrontation (MIC) Codebook notes
that the coding of the outcome is particularly problematic, that the coding rules for this variable
were not consistently applied, and that they are currently working on recoding the variable.
Another problem is that, in the set of confrontations for the three countries included in
this study, for which I have a sufficient number of paragraphs scored for integrative complexity
(20 paragraphs or more), the distribution of the values for the outcome variable is so
predominantly “stalemate,” that it leaves relatively little variance in my response variables. In
the variable Degree of Success, the value of 2 (stalemate and compromise) accounts for 64% of
the confrontations included in the regressions, while the value of 1 (Defeat and Yield) accounts
for only 10%. This problem is even worse for the variable Avoidance of Defeat, for which the
value of 0 (defeat was avoided) accounts for 95% of the data.
92
Given that the mic coding manual flags the Outcome variable as requiring recoding, and
states that future iterations of the dataset will include a recoded version of this variable, in future
research it would be worth redoing the analyses in this section using the recoded Outcome
variable.
93
Chapter 7: Leaders’ Integrative Complexity and International Crisis Behavior
7.1 Coding of the International Crisis Behavior (ICB) Variables
7.1.1 Country
In the ICB dataset this is coded from the variable “CRACID”, which is the numeric actor
code, and which, similarly to the MIC dataset, for actors that were countries, uses the same codes
as the Correlates of War (COW) project. The original numeric country codes are: 002 = United
States, 200 = United Kingdom, 365 = Russia/USSR. This paper abbreviates the countries using
the International Standards Organization (ISO) three letter country codes, which are USA =
United States, GBR = United Kingdom, RUS = Russia/USSR.
A potential source of confusion is that the ICB dataset also has its own three letter code,
in the variable “ACTOR”, which assigns the same codes as the ISO does to the United States and
Russia, but a different code to the United Kingdom, which has the ICB ACTOR code of “UKG”,
while the ISO Three-Letter Country code is “GBR”. This paper uses the ISO Three-Letter
Country codes.
7.1.2 Time period
This variable is coded from the ICB System-Level variable “PERIOD”, which was coded
according to the structure of the international system at the time of the initiation of the
international crisis. The original ICB variable had six categories: (1) Multipolarity: 1918-1939;
(2) World War II: 1939-1945; (3) Bipolarity: 1945-1962; (4) Polycentrism I: 1963-1989; (5)
Unipolarity: 1990-2010; (6) Polycentrism II: 2011-Ongoing. In order to reduce the number of
categories, and to separate the effect of world war from the effect of the polarity of the structure
94
of the international system, I recoded this as: Pre Cold War (containing 1 and 2); Cold War
(containing 3 and 4), and Post Cold War (containing 5 and 6).
I discussed the theoretical importance of the time period in the section concerning the
coding of the variable for the Militarized Interstate Confrontations analyses, so I do not repeat it
here.
7.1.3 During world war
This is coded as No = 0, and Yes = 1. It is coded as yes (1) if the international crisis
began during either the First World War (1914-07-29 to 1918-11-11), or the Second World War
(1939-09-02 to 1945-09-02), and no (0) otherwise.
7.1.4 Protracted conflict
This variable codes for whether the crisis occurred within a protracted conflict
(sometimes referred to as “enduring rivalry”) in which countries have repeated conflictual
interactions over an extended duration of time, e.g., Israel-Arab States, India-Pakistan, or the
West-East Block conflict in the Cold War. It is coded as No = 0, and Yes = 1. The codes for this
variable are taken, without any recoding, from the ICB System-Level variable “PROTRAC”.
This variable is relevant for predicting behavior in international crises because it is likely
that a pattern of conflictual interactions between countries will be associated with a greater
tendency for future conflictual interactions, even after accounting for other covariates.
7.1.5 Distance
This variable is referred to as “distance” for simplicity and brevity, but it would more
accurately be referred to as the location of the crisis actor relative to the location of the crisis,
because it is an ordinal variable indicating their relative locations. It takes the codes, without any
recoding, of the ICB Actor-Level variable “CRACTLOC”, which has the following values:
95
Home Territory (the crisis occurred on the crisis actor’s home territory) = 1; Sub-Region (the
crisis occurred in the same region, within a continent, as the home territory of the crisis actor) =
2; Same Continent (the crisis occurred in the same continent, but not the same region, as the
home territory of the crisis actor) = 3; Elsewhere (the crisis occurred in a different continent as
the home territory of the crisis actor) = 4.
This variable is relevant for predicting behavior in international crises because, with
decreasing distance, a crisis is likely to be more salient to a country’s leaders and population, and
also to provide more opportunities to take action, including military or violent action.
For similar reasons, low distance may be associated with the leader experiencing greater
stress, so controlling for this variable helps to control for a contributor to the leader’s stress.
7.1.6 Gravity of threat
This variable is a dummy, with the coding: qualitatively less severe threats = 0;
qualitatively more severe threats = 1. It is coded from the ICB Actor-Level variable “GRAVTY”,
with recoding to convert this nominal variable to a dummy variable. A code of 0 was assigned to
the following categories: (0) Economic threat; (1) Limited military threat; (2) Political threat; (3)
Territorial threat. A code of 1 was assigned to the following categories: (4) Threat to influence;
(5) Threat of grave damage; (6) Threat to existence. A code of NA was assigned to the category
(7) Other.
In the coding of the original ICB variable, where more than one type of threat applied, the
more severe (higher number was coded). So, while the variable was not a truly ordinal scale, the
recoded dummy variable is an indicator of the severity of the qualitative type of threat.
It may seem odd to code “threat to influence” as a high severity type of threat. However,
this category of threat is associated with power politics, such as competition between the United
96
States and USSR in the Cold War, and loss of influence could result in a cascade of further loss
of influence, or loss of reputation, such that it tends to be a qualitatively severe type of threat.
Examples include Communism in Poland (1946), Communism in Hungary (1947), Truman
Doctrine (1947), Communism in the Czech Republic (1948), Berlin Deadline (1958), Berlin
Wall (1961), Tet Offensive (1968), and, in the post Cold War period, multiple crises involving
Iranian nuclear weapons, North Korean nuclear weapons, and Syrian chemical weapons.
Researchers with the ICB project have, in their own publications, treated “threat to influence” for
a great/super power as a highly severe category of threat (Brecher & Wilkenfeld, 1997), and I
follow them in doing this.
More grave threats should be associated with higher levels of stress. Therefore,
controlling for this variable helps to control for a contributor to the leader’s stress.
7.1.7 Power discrepancy
This variable is a measure of the discrepancy in power between the crisis actor, and their
tight alliance partners, as against their principal adversary(ies) in the crisis. It is taken, without
any recoding, from the ICB Actor-Level variable “POWDIS”. According to the ICB Actor-Level
manual, a power score was calculated for the crisis actor and their tight alliance partners, and on
the other hand their principal adversary(ies), on the basis of six separate scores measuring
population size, GNP, territorial size, alliance capability, military expenditure, and nuclear
capability. The final power discrepancy score is the result of comparing the power of the crisis
actor (and tight alliance partners) with the power of the principal adversary(ies). Although the
manual does not explicitly say this (it says that the power of the two sides is “compared”), it
appears that the power discrepancy score is the result of subtracting the power of the
97
adversary(ies) from the power of the crisis actor (and tight alliance partners), because it has
negative values when the adversaries were relatively more powerful.
This variable has a large number of missing values, so the models that include it suffer a
reduction in the N.
Power discrepancy should be controlled for because it is a fundamental variable in
international relations theory, including hegemonic stability theory and power transition theory,
which suggest that violence should be more likely, and more extreme, when power discrepancy
is lower (G. T. Allison, 2017; Gilpin, 1981; Kindleberger, 1986; Krasner, 1976).
The less powerful the leader’s side is relative to the adversary’s side, the more the crisis
should contribute to the level of stress experienced by the leader. Therefore, controlling for
power discrepancy also helps to control for something that contributes to the leader’s level of
stress.
7.1.8 The actor who initiated the first violence, and the actor against whom violence
was first initiated
My first attempt to code these did not work. I originally attempted to code these by
writing an RScript that would automatically assign codes based on the values of existing ICB
variables (e.g., the variable for the timing of violence, the act that was the earliest violent trigger,
the act that was the earliest violent major response). Unfortunately, the automated approaches
that I attempted resulted in unacceptably high rates of false codes.
As a result, it was necessary to manually code the act that initiated violence. Given time
and resource constraints, I had to code it myself, rather than train research assistants to do so.
98
In order to reduce the potential for scorer bias, I wrote a sequence of steps that primarily
involves using the values of ICB variables, but if problems occur, they are addressed using
manual recoding.
Step 1: If the variable Timing of Violence (TIMVIO), or the variable Violence (VIOL),
indicated that there was no violence during the crisis, then I coded that there was no initiation of
violence. Therefore, no actor initiated violence, and no actor was the target of initial violence.
These are substantive codes (not codes of NA), because violence could have been initiated.
Step 2: If there was violence during the crisis, but the variable Timing of Violence
indicated that there was already violence prior to the crisis, then I coded this as Violence Prior to
and During the Crisis. In this case, because initiation of violence was impossible, the actor that
initiated violence, and the target of initial violence, were both coded as NA.
Step 3: If the variable Timing of Violence indicated that the earliest violence was the act
that triggered the international crisis, then I used the ICB code for the actor that performed the
triggering act as the code for the actor that initiated violence. Similarly, I used the ICB code for
the actor that was the target of that triggering act as the code for the target of the initial violence.
There were exceptions that required me to manually code variables. For instance, in
eleven cases the Timing of Violence indicated that the crisis trigger was the earliest violence, but
the variable for the type of crisis trigger (BREAK) indicated that it was not a violent act. In these
cases, based on the ICB descriptions of the crises, I manually coded the act that initiated the first
violence, and the actor and target of that initial violence.
Usually the ICB gave country codes (e.g., USA, RUS) for the actor and target of the
initial violence. However, in some cases it gave the codes 995 (internal actor), 996 (nonstate),
99
and 997 (unknown or multiple). In these cases, based on the ICB description of the crisis, I
manually coded the precise actor.
Step 4: If the variable Timing of Violence indicated that the earliest violence occurred
after the event that triggered the crisis, then based on the ICB description of the crisis, I manually
coded the act that initiated violence, the actor that initiated the violence, and the target.
Step 5: I went through every crisis and judged whether the existing codes made sense. I
identified patterns of problems, and decided how to fix them. Then I went through every crisis
and, where the problems occurred, I manually recoded them. For example, one such problem is
that crises that occurred during wars were sometimes coded as not having prior violence, which,
for my purposes, is clearly incorrect. I recoded these as instances in which there was prior
violence, and therefore violence was not initiated.
For every actor that initiated violence, and for every target of initial violence, I also
coded whether they were on the same side as the leader’s country (i.e., USA, RUS, GBR), or on
an opposing side.
If a non-state actor that was on the same side as the leader’s country performed the act
that was the initial violence in the crisis, this is coded as the initiation of violence by a non-state
proxy of the leader’s country.
7.1.9 Centrality of violence
This variable is a measure of the importance that the country’s foreign policy decision-
makers, i.e., usually a decision-making group lead by the head of government, attached to their
use of violence as a strategy to achieve their goals, relative to other non-violent strategies. It is
taken, without recoding, from the ICB Actor-Level variable “CENVIO”. It has the following
values: No violence = 1; Violence minor = 2; Violence important = 3; Violence preeminent = 4.
100
7.1.10 Degree of Success
This is coded from the ICB variable Outcome (OUTCOM). I somewhat simplified this
coding so that Defeat = 1, Stalemate or Compromise = 2, and Victory = 3.
Note that the terms “stalemate” and “compromise” are misleading. The code for
Stalemate does not indicate a long period of time, and the code for Compromise does not
indicate that there was a mutual trading of concessions with the opposing side. Stalemate means
that the leader’s country did not achieve any of its core goals, but also did not suffer defeat.
Compromise means that the leader’s country achieved some, but not all, of its core goals. I
collapsed them into one category, for cases in which the leader’s country neither definitively
won, nor definitively lost.
7.1.11 Avoidance of Defeat
This is also coded from the ICB variable Outcome (OUTCOM). It is a dummy variable in
which Defeat = 0, and Avoidance of Defeat (Stalemate, Compromise or Victory) = 1.
7.1.12 Satisfaction of the leader’s country with the outcome
This is a dummy variable indicating whether the post-crisis government of the leader’s
country is satisfied with the outcome, with No = 0, and Yes = 1.
It was coded from the ICB Actor-Level variable “OUTEVL”, a categorical measure
indicating whether the leader’s country was satisfied with the outcome, and whether the principal
adversary was satisfied with the outcome, such that there were four categories: (1) All parties
satisfied with content of outcome; (2) Crisis actor satisfied, adversaries dissatisfied; (3)
Adversaries satisfied, crisis actor dissatisfied; (4) All parties dissatisfied. I converted this to a
dummy for the satisfaction of the leader’s country by recoding categories 3 and 4 to 0 (leader’s
country dissatisfied), and recoding categories 1 and 2 to 1 (leader’s country satisfied).
101
The ICB manual says little about how satisfaction was coded, but it appears to be based
on whether the government of the country expressed satisfaction or dissatisfaction with the
outcome, and whether it subsequently attempted to change the outcome, which would indicate
dissatisfaction. This is a measure of whether the post-crisis government of the leader’s country is
satisfied with the outcome. This is not a measure of whether the general population of the
leader’s country was satisfied. Nor does it always indicate whether the leader, or the leader’s
government (in the sense of the cabinet or other executive body lead by them), were satisfied
with the outcome, because the post-crisis government is not always the same as the government
during the crisis.
7.1.13 Satisfaction of the adversary with the outcome
This is a dummy variable indicating whether, following the termination of the crisis, the
leader’s principal adversary was satisfied with the outcome, with No = 0, and Yes = 1. In those
cases in which the principal adversary was a country, this is whether the post-crisis government
of that adversary country was satisfied with the outcome. The details of the coding of this
variable are the same as for the coding of the satisfaction of the leader’s country (see above),
except that I coded the OUTEVL categories 2 and 4 to 0 (adversary dissatisfied), and categories
1 and 3 to 1 (adversary satisfied).
7.2 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Was the Target of Initial Violence?
7.2.1 Predictions
According to the strategic judgment hypothesis, the leader’s pre-crisis IC should be
negatively associated with having violence initiated against their country, because higher
102
integrative complexity should be associated with finding solutions to complex problems, such as
coordination problems, to the satisfaction of both sides of a dispute.
According to the demonstration-of-resolve hypothesis, the leader’s pre-crisis IC should
be positively associated with having violence initiated against their country, because higher IC
should make it more likely that adversaries perceive the leader as lacking resolve, increasing the
probability that they will use violence to pressure the leader to make concessions.
The initiation of violence against the leader’s country is more directly relevant to the
demonstration-of-resolve hypothesis than it is to the strategic judgment hypothesis, because the
demonstration-of-resolve hypothesis is about how the leader’s expression of IC affects the
perceptions, and therefore behavior, of adversaries, while the strategic judgment hypothesis
operates through the behavior and decision-making of the leader. So, although this tests both
hypotheses, it in particular is a test of the demonstration-of-resolve hypothesis.
7.2.2 Results
7.2.2.1 Descriptive statistics
These descriptive statistics describe the simplest model that includes integrative
complexity (i.e., model 2 in the regression tables). If models with additional control variables
have missing data in those variables, then the N will be reduced in those models. This applies to
all subsequent descriptive statistics in this chapter.
These analyses have an N of 79 crises. The number of paragraphs coded for the leader’s
pre-crisis integrative complexity, per crisis, ranges from a minimum of 21 to a maximum of 518,
with a mean of 141.16, and a standard deviation of 110.60.
Across the crises, the leaders’ pre-crisis integrative complexity ranges from a minimum
of 1.5 to a maximum of 2.67, with a mean of 2.01, and a standard deviation of 0.22.
103
The variable that represents that the act that initiated violence targeted the leader’s
country is binary (No = 0, Yes = 1). Across the crises it has a mean of .20, and a standard
deviation of 0.40.
7.2.2.2 Inferential statistics
In the regressions below, the explanatory variable is the leader’s mean IC, in the month
prior to the international crisis. The response variable is whether the leader’s country was the
target of the initial violence in the crisis.
As with all of the tests, crises that have fewer than 20 paragraphs scored for integrative
complexity, in this case from the pre-crisis period, were excluded. In addition, crises that began
during a world war were excluded, because neither of the two theories makes clear predictions
concerning this variable for crises that began during a world war. There is little room for
deterrence once a conflict has escalated to total war at a global scale. And, once one’s country is
engaged in a total war at a global scale, there is little reason to expect that higher integrative
complexity would be associated with the capability to avoid the violence in that war bleeding
into other conflicts, or that in any given interaction, doing so would avoid costs with respect to
the greater effects of the war.
All of the regression models include covariates for Country (USA, RUS, GBR), and Time
Period (Pre Cold War, Cold War, Post Cold War). In addition to these two covariates, models 3
to 6 each includes one of the following covariates: Protracted Conflict, Gravity of Threat,
Distance, or Power Discrepancy. The results are in Table 7 below.
104
Table 7
Regressions Predicting Whether the Initial Violence Targeted the Leader’s Country
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
-1.50*
1.07
4.06
3.60
3.64
0.81
p = .03
p = .72
p = .21
p = .29
p = .27
p = .79
(0.70)
(2.96)
(3.25)
(3.37)
(3.33)
(2.98)
Country
0.88
1.10
1.98+
0.94
0.40
1.08
RUS
p = .35
p = .27
p = .09
p = .40
p = .71
p = .32
(0.95)
(0.99)
(1.17)
(1.11)
(1.08)
(1.08)
Country
0.38
0.23
1.02
0.64
0.20
0.28
USA
p = .62
p = .78
p = .25
p = .46
p = .81
p = .73
(0.77)
(0.79)
(0.88)
(0.87)
(0.81)
(0.80)
Period
-0.60
-0.45
-0.87
-0.70
-0.39
-0.39
Post CW
p = .36
p = .51
p = .26
p = .34
p = .58
p = .60
(0.66)
(0.68)
(0.77)
(0.73)
(0.71)
(0.74)
Period
-0.47
-0.50
-0.81
-0.99
-1.00
-0.32
Pre CW
p = .65
p = .62
p = .50
p = .41
p = .36
p = .76
(1.02)
(1.02)
(1.19)
(1.21)
(1.09)
(1.04)
Pre-crisis IC
-1.28
-1.32
-1.96
-1.49
-1.14
p = .38
p = .38
p = .21
p = .32
p = .44
(1.44)
(1.51)
(1.55)
(1.49)
(1.46)
Protracted
-2.13**
Conflict
p < .01
(0.69)
Gravity of
-1.66+
Threat
p = .06
(0.88)
Distance
-0.63*
p = .04
(0.31)
Power
0.00
Discrepancy
p = .81
(0.02)
Num.Obs.
79
79
79
79
79
78
AIC
88.0
89.2
79.8
87.6
86.9
90.6
BIC
99.8
103.4
96.4
104.2
103.5
107.1
Log.Lik.
-38.995
-38.606
-32.897
-36.782
-36.470
-38.304
F
0.389
0.465
1.772
0.836
1.017
0.410
RMSE
0.40
0.40
0.37
0.39
0.38
0.40
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
105
We can see in the table above that there is a negative sign of relationship between the
leader’s pre-crisis IC and their side being the target of initial violence in the crisis, but that this
relationship is not statistically significant for any of the regression models.
In the model that includes IC as a predictor (model 2), as compared to the model that
does not (model 1), neither the AIC nor the log likelihood are notably different, suggesting that
the inclusion of IC as a predictor does not improve the model fit.
7.2.3 Discussion
The results of this test do not provide support for either hypothesis. Although the
negative sign of relationship between the leader’s pre-crisis integrative complexity, and the
initiation of violence against their country, is consistent with the strategic judgment hypothesis,
and not with the demonstration-of-resolve hypothesis, the relationship is not significant, and the
inclusion of pre-crisis integrative complexity does not improve the model fit. The results
therefore do not give strong support for either hypothesis.
7.3 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Initiated the First Violence?
7.3.1 Predictions
According to the strategic judgment hypothesis the leader’s pre-crisis IC should be
negatively associated with their country initiating the first violence in the crisis. This is because,
according to this hypothesis, higher integrative complexity enables leaders to achieve their goals
without resorting to costly violence.
The demonstration-of-resolve hypothesis does not suggest a prediction concerning
initiation of violence by the leader’s country.
106
7.3.2 Results
7.3.2.1 Descriptive statistics
The descriptive statistics for the N of crises, number of paragraphs per crisis, and pre-
crisis IC are identical to that of the previous section, so I do not repeat them here.
The variable that represents that the leaders’ country initiated the first violence is binary
(No = 0, Yes = 1). Across the crises it has a mean of .14, and a standard deviation of 0.35.
7.3.2.2 Inferential statistics
In the regressions below, the explanatory variable is the leader’s mean IC in the month
prior to the international crisis. The response variable is whether the leader’s country initiated the
first violence in the crisis.
As with all of the tests, crises that have fewer than 20 paragraphs scored for integrative
complexity, in this case from the pre-crisis period, were excluded. As with the tests in the section
above, and for the same reasons, crises that began during a world war were excluded.
All of the regression models include covariates for Country (USA, RUS, GBR), Time
Period (Pre Cold War, Cold War, Post Cold War). In addition to these two covariates, models 3
to 6 each includes one of the following covariates: Protracted Conflict, Gravity of Threat,
Distance, or Power Discrepancy. The results are in Table 8 below.
We can see in the table that there is a positive sign of relationship between the leader’s
pre-crisis IC and their side initiating the first violence in the crisis, but that this relationship is not
statistically significant for any of the regression models. In the model that includes IC as a
predictor (model 2), as compared to the model that does not (model 1), neither the AIC nor the
log likelihood are notably different, suggesting that the inclusion of IC as a predictor does not
notably improve model fit.
107
Table 8
Regressions Predicting Whether Violence was Initiated by the Leader’s Side
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
-2.69**
-9.85*
-10.10*
-8.43
-9.18+
-9.08+
p < .01
p = .04
p = .04
p = .10
p = .07
p = .06
(0.92)
(4.79)
(4.96)
(5.13)
(5.01)
(4.81)
Country
1.91+
1.38
1.37
1.04
1.10
1.38
RUS
p = .05
p = .20
p = .20
p = .37
p = .38
p = .26
(0.98)
(1.07)
(1.07)
(1.17)
(1.25)
(1.22)
Country
-0.27
0.29
0.26
0.43
0.29
0.17
USA
p = .78
p = .79
p = .81
p = .69
p = .79
p = .87
(0.98)
(1.07)
(1.08)
(1.07)
(1.06)
(1.06)
Period
1.11
0.61
0.64
0.39
0.61
0.53
Post CW
p = .18
p = .48
p = .47
p = .67
p = .48
p = .59
(0.82)
(0.88)
(0.88)
(0.91)
(0.88)
(0.97)
Period
0.64
0.83
0.83
0.45
0.69
0.47
Pre CW
p = .55
p = .46
p = .46
p = .72
p = .56
p = .70
(1.06)
(1.13)
(1.13)
(1.25)
(1.18)
(1.22)
Pre-crisis IC
3.44
3.44
3.24
3.45
3.07
p = .12
p = .12
p = .17
p = .12
p = .17
(2.20)
(2.21)
(2.34)
(2.22)
(2.23)
Protracted
0.16
Conflict
p = .83
(0.74)
Gravity of
-1.20
Threat
p = .18
(0.89)
Distance
-0.19
p = .65
(0.43)
Power
0.01
Discrepancy
p = .76
(0.02)
Num.Obs.
79
79
79
79
79
78
AIC
63.8
63.0
64.9
63.2
64.8
63.6
BIC
75.6
77.2
81.5
79.8
81.4
80.1
Log.Lik.
-26.897
-25.488
-25.465
-24.619
-25.392
-24.780
F
2.309
2.075
1.723
1.910
1.724
1.449
RMSE
0.31
0.31
0.31
0.30
0.31
0.31
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
108
7.3.3 Discussion
The results of this test do not provide support for the strategic judgment hypothesis. The
positive sign of relationship between the leader’s pre-crisis integrative complexity, and the
initiation of the first violence in the crisis by their side, is the opposite to the sign predicted by
this hypothesis, which may even suggest an effect in the direction opposite to that predicted.
However, because the results are not statistically significant, and the inclusion of the leader’s
pre-crisis integrative complexity does not improve the model fit, judging by AIC and log
likelihood, to any notable degree, there is little evidence for any consistent relationship between
the leader’s pre-crisis integrative complexity and the initiation of violence by their side.
After qualitatively exploring the data, it appears that the positive sign of this relationship
is driven by violence initiated by nonstate proxies of the leader’s country. In the next section, I
do exploratory statistical tests of this more specific relationship.
7.4 Does a Leader’s Integrative Complexity Before a Crisis Predict Whether Their Country
Uses Nonstate Proxies to Initiate Violence?
7.4.1 Predictions
This is an exploratory test, that was inspired by the results in the previous section, and
therefore there were no theoretical predictions, prior to running the initial analyses. After running
those analyses, there was some evidence that a leader’s pre-crisis integrative complexity is
positively associated with violence being initiated by nonstate proxies of the leader’s country. A
possible theoretical explanation for this relationship is that the use of nonstate proxies is a
relatively complex strategy for using violence to achieve one’s goals, while attempting to avoid
suffering the costs that are usually associated with directly employing violence, e.g., that the
other side retaliates, including by employing violence against one’s country. This suggests a
109
modification of the strategic judgment hypothesis, which is that higher integrative complexity
enables leaders to employ complex strategies to attempt to initiate violence while avoiding
bearing the costs of that violence, i.e., by using proxies.
Another possible theoretical explanation focuses on the expression of integrative
complexity, not as a reflection of internal cognition, but rather as a communicative strategy.
According to this explanation, leaders who are planning to use nonstate proxies to initiate
violence will tend, as part of this strategy, to express themselves in a high integrative complexity
manner, either in order to make themselves seem more reasonable and less likely to be
responsible for encouraging violence, or in order to depict the situation as highly complicated, in
order to discourage the listener from making the simple attribution that they are responsible for
encouraging the violence.
It is not possible, with the data available, to differentiate between these hypotheses.
7.4.2 Results
7.4.2.1 Descriptive statistics
The descriptive statistics for the N of crises, number of paragraphs per crisis, and pre-
crisis IC are identical to that of the previous two sections, so I do not repeat them here.
The variable that represents whether violence was initiated by a nonstate proxy of the
leader’s country is binary (No = 0, Yes = 1). Across the crises it has a mean of .06, and a
standard deviation of 0.25.
7.4.2.2 Inferential statistics
In the regressions below, the explanatory variable is the leader’s mean IC in the month
prior to the international crisis. The response variable is whether a non-state proxy of the leader’s
country initiated the first violence in the crisis.
110
As with all of the tests, crises that have fewer than 20 paragraphs scored for integrative
complexity, in this case from the pre-crisis period, were excluded. As with the tests in the section
above, and for the same reasons, crises that began during a world war were excluded.
All of the regression models include covariates for Country (USA, RUS, GBR), Time
Period (Pre Cold War, Cold War, Post Cold War). In addition to these two covariates, models 3
to 6 each includes one of the following covariates: Protracted Conflict, Gravity of Threat,
Distance, or Power Discrepancy. The results are in Table 9 below.
111
Table 9
Regressions Predicting Whether Violence was Initiated by a Nonstate Proxy of the Leader’s
Country
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
-19.48
-31.47
-32.18
-32.48
-30.07
-31.83
p = .99
p = .99
p = .99
p = .99
p = .99
p = .99
(2389.12)
(2270.65)
(2261.87)
(2263.54)
(3657.28)
(3538.14)
Country
18.66
17.71
17.63
17.81
18.16
18.65
RUS
p = .99
p = .99
p = .99
p = .99
p = 1.00
p = 1.00
(2389.12)
(2270.64)
(2261.86)
(2263.53)
(3657.28)
(3538.13)
Country
16.24
16.97
16.93
16.99
17.98
17.66
USA
p = .99
p = .99
p = .99
p = .99
p = 1.00
p = 1.00
(2389.12)
(2270.64)
(2261.86)
(2263.53)
(3657.27)
(3538.13)
Period
0.26
-0.88
-0.83
-0.70
-1.30
-1.04
Post CW
p = .82
p = .52
p = .55
p = .64
p = .44
p = .51
(1.11)
(1.39)
(1.39)
(1.49)
(1.67)
(1.58)
Period
-0.56
-0.41
-0.37
0.06
-1.01
-18.51
Pre CW
p = .71
p = .82
p = .84
p = .98
p = .62
p = 1.00
(1.53)
(1.79)
(1.80)
(2.45)
(2.01)
(4535.63)
Pre-crisis IC
5.78+
5.83+
5.97+
5.67+
5.55
p = .08
p = .08
p = .08
p = .08
p = .15
(3.31)
(3.35)
(3.39)
(3.27)
(3.81)
Protracted
0.34
Conflict
p = .78
(1.21)
Gravity of
0.56
Threat
p = .78
(2.02)
Distance
-0.63
p = .28
(0.58)
Power
0.01
Discrepancy
p = .80
(0.03)
Num.Obs.
79
79
79
79
79
78
AIC
39.2
37.4
39.3
39.3
38.2
35.0
BIC
51.1
51.6
55.9
55.9
54.8
51.5
Log.Lik.
-14.608
-12.675
-12.634
-12.636
-12.119
-10.509
F
1.267
1.390
1.172
1.176
1.151
1.023
RMSE
0.22
0.20
0.20
0.20
0.20
0.18
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
112
We can see in the table above that there is a positive sign of relationship between the
leader’s pre-crisis IC and a non-state proxy of their country initiating the first violence in the
crisis. This relationship is near significant (p < .10) for the regression models 2 through 5, which
include the most basic model that includes integrative complexity, as well as models that
additionally include covariates for protracted conflict, gravity of threat, and distance. The
relationship was neither significant nor near significant for model 6, which included a covariate
for power discrepancy.
In the model that includes IC as a predictor (model 2), as compared to the model that
does not (model 1), the AIC is lower, and the log likelihood is higher, suggesting that including
the leader’s pre-crisis integrative complexity as a predictor improves the model fit. However, the
magnitude the difference is only 1.8 for the AIC, and 1.9 for the log likelihood, which is
borderline with respect to whether the degree of improvement should be considered important
(Burnham & Anderson, 2002).
7.4.3 Discussion
The results of these analyses provide some support for the claim that a leader’s pre-crisis
integrative complexity is positively associated with the subsequent initiation of violence by
nonstate proxies of their country. However, because these results are only near significant for
five of the models, and not even near significant for one of the models, and because the inclusion
of the leader’s pre-crisis integrative complexity only marginally improves the model fit, the
evidence is quite weak.
In addition, the confidence that we can have in this evidence is weakened by the fact that
these are exploratory tests that were inspired by inspection of patterns in the data, rather than by
prior theory. Therefore, while the results are suggestive of the possibility that a leader’s pre-crisis
113
integrative complexity is positively associated with the initiation of violence by nonstate proxies
of their country, and this possibility should be tested further in future research, it cannot be
drawn as a firm conclusion here.
If it were the case that a leader’s pre-crisis integrative complexity is positively associated
with the initiation of violence by nonstate proxies of their country, then this would suggest two
theoretical possibilities. The first hypothesis is a modification of the strategic judgment
hypothesis, and is that leaders with higher integrative complexity are more likely to find complex
strategies that employ violence, while attempting to avoid suffering the costs of directly
employing violence, namely by using nonstate proxies to initiate violence. In this modification of
the hypothesis, it is not necessarily the case that higher integrative complexity enables leaders to
achieve their goals without the use of costly violence. Instead, their higher integrative
complexity enables them to achieve their goals by employing strategies to use violence while
avoiding bearing the costs.
The second hypothesis is that the expression of high integrative complexity is used by the
leader as a strategy of communication, in order to try to prevent the listener from coming to the
conclusion that the leader is responsible for the violence carried out by the nonstate proxy.
It is not possible to distinguish between these possibilities with the data available here.
7.5 Does a Leader’s Integrative Complexity During a Crisis Predict the Degree That Their
Country Relies on Violence Relative to Nonviolent Strategies?
7.5.1 Predictions
According to the strategic judgment hypothesis the leader’s IC during the crisis should be
negatively associated with the degree to which they rely on violence relative to nonviolent
strategies, in their attempts to deal with the crisis. This is because, according to this hypothesis,
114
higher integrative complexity enables leaders to achieve their goals without resorting to costly
violence. Furthermore, even in cases where the leader decides to employ violence, or even high
levels of violence, integrative complexity should be positively associated with the use of
nonviolent strategies, in addition to the violence, so that even in these cases the leader’s IC
should be associated with less reliance on violence relative to nonviolent strategies.
According to the demonstration-of-resolve hypothesis, the leader’s IC during a crisis
should be positively associated with the level of hostility employed by their country, because
leader’s who express higher levels of IC will be perceived by adversaries as being less resolved,
and will therefore have to employ more extreme measures in order to convince adversaries to
back down.
7.5.2 Results
7.5.2.1 Descriptive statistics
These analyses have an N of 119 crises. The number of paragraphs coded for integrative
complexity, per crisis, ranges from a minimum of 24 to a maximum of 6,525, with a mean of
613.82, and a standard deviation of 998.86.
Across the crises, the leaders’ crisis integrative complexity ranges from a minimum of
1.41 to a maximum of 3.33, with a mean of 2.06, and a standard deviation of 0.28.
The variable that represents the reliance on violence by the leader’s side (the centrality of
violence) has the coding: No violence = 1; Violence minor = 2; Violence important = 3;
Violence preeminent = 4. Across the crises it has a mean of 2.09, and a standard deviation of
1.33.
115
7.5.2.2 Inferential statistics
I used linear regressions to test whether the leader’s mean IC during an international
crisis is a significant predictor of the degree to which the leader’s country relied on violence
relative to nonviolence strategies, in its crisis management, as measured by the variable
Centrality of Violence.
As in all of the tests, the regression models only include crises for which there were at
least 20 paragraphs scored for integrative complexity.
All of the regression models included covariates for Country (USA, RUS, GBR), Time
Period (Pre Cold War, Cold War, Post Cold War), During World War (No = 0, Yes = 1), and
whether the crisis began due to a Violent Trigger (No = 0, Yes = 1). In addition to these four
covariates, models 3 to 6 each includes one of the following covariates: Protracted Conflict,
Gravity of Threat, Distance, or Power Discrepancy. The results are in Table 10 below.
116
Table 10
Regressions Predicting the Degree that the Leader’s Country Relied on Violence Relative to
Nonviolent Strategies
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
1.44***
3.74***
3.30**
3.83***
4.35***
4.08***
p < .01
p < .01
p < .01
p < .01
p < .01
p < .01
(0.29)
(1.06)
(1.08)
(1.06)
(1.16)
(1.09)
Country
-0.01
0.22
0.22
0.05
0.03
0.26
RUS
p = .98
p = .56
p = .55
p = .90
p = .94
p = .55
(0.36)
(0.37)
(0.37)
(0.39)
(0.40)
(0.44)
Country
0.07
-0.15
-0.27
-0.01
-0.13
-0.13
USA
p = .81
p = .61
p = .38
p = .97
p = .66
p = .68
(0.29)
(0.30)
(0.31)
(0.31)
(0.30)
(0.32)
Period
0.42
0.50+
0.65*
0.36
0.52*
0.38
Post CW
p = .11
p = .06
p = .02
p = .18
p = .05
p = .18
(0.26)
(0.26)
(0.27)
(0.27)
(0.26)
(0.29)
Period
0.44
0.43
0.48
0.31
0.31
0.46
Pre CW
p = .32
p = .31
p = .26
p = .47
p = .48
p = .31
(0.43)
(0.42)
(0.42)
(0.43)
(0.43)
(0.45)
During
0.98*
0.99*
0.69
1.13*
0.94+
World War
p = .05
p = .04
p = .17
p = .02
p = .05
(0.49)
(0.48)
(0.50)
(0.49)
(0.48)
Violent
0.81***
0.78**
0.80***
0.70**
0.79**
0.76**
Trigger
p < .01
p < .01
p < .01
p < .01
p < .01
p < .01
(0.24)
(0.23)
(0.23)
(0.24)
(0.23)
(0.25)
Crisis IC
-1.09*
-1.15*
-0.90+
-1.09*
-1.30*
p = .03
p = .02
p = .07
p = .03
p = .01
(0.48)
(0.48)
(0.49)
(0.48)
(0.50)
Protracted
0.34+
Conflict
p = .08
(0.19)
Gravity of
-0.54
Threat
p = .13
(0.35)
Distance
-0.17
p = .20
(0.13)
Power
0.01
Discrepancy
p = .24
(0.00)
Num.Obs.
119
119
119
117
119
103
R2
0.185
0.221
0.242
0.227
0.232
0.187
R2 Adj.
0.141
0.171
0.187
0.170
0.176
0.127
117
(1)
(2)
(3)
(4)
(5)
(6)
AIC
398.4
395.1
393.8
388.0
395.4
342.3
BIC
420.7
420.1
421.6
415.6
423.1
366.0
Log.Lik.
-191.213
-188.562
-186.884
-184.000
-187.676
-162.146
F
4.240
4.488
4.396
3.963
4.157
RMSE
1.21
1.18
1.16
1.17
1.17
1.17
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
The leader’s IC during the crisis is a statistically significant predictor of their country’s
reliance on violence in four of the models (models 2, 3, 5, and 6), and is a near significant
predictor in the fifth (model 4), all with a negative sign of relationship.
In the model that includes IC as a predictor (model 2), as compared to the model that
does not (model 1), the AIC is lower, and the log likelihood is higher, suggesting that including
the leader’s pre-crisis integrative complexity as a predictor improves the model fit.
In the model that includes IC as a predictor (model 2), as compared to the model that
does not (model 1), the adjusted R-square is increased by .030.
7.5.3 Discussion
The results of this test are consistent with the strategic judgment hypothesis, and are not
consistent with the demonstration-of-resolve hypothesis. The negative sign of relationship
between the leader’s crisis integrative complexity, and the degree to which their country relies on
violence relative to other strategies, in the attempts to address the crisis, is the sign of
relationship that is predicted by the strategic judgment hypothesis, and is the opposite of the sign
predicted by the demonstration-of-resolve hypothesis.
This relationship is statistically significant for all of the models, except model 4 (with
Gravity of Threat as an additional covariate), which is still near significant. Furthermore,
according to both AIC and log likelihood, the inclusion of the leader’s integrative complexity
118
during the crisis improves the model fit. While the inclusion of crisis integrative complexity as a
predictor may seem to contribute a relatively small amount to the adjusted R-square, .030, this is
fairly large contribution relative to the adjusted R-squared of the model without integrative
complexity (model 1), which is only .141, or compared to any of the regression models, of which
the highest is .187. A country’s reliance on violence during an international crisis is difficult to
predict, and relative to the amount of variance that any of the models can account for, the
inclusion of the leader’s crisis IC as a predictor contributes a substantial amount to the variance
in reliance on violence that is accounted for by the model.
Overall, these results provide strong support for the strategic judgment hypothesis.
7.6 Does a Leader’s Integrative Complexity During a Crisis Predict the Degree of Success
Achieved by the Leader’s Country at the Outcome?
7.6.1 Predictions
According to the strategic judgment hypothesis, the level of IC expressed by a leader
during an international crisis should be positively associated with the degree of success achieved
by their country at the outcome of the crisis. According to the demonstration-of-resolve
hypothesis, the level of IC expressed by a leader during an international crisis should be
negatively associated with the degree of success achieved by their country at the outcome of the
crisis. This is true both for success measured as an interval variable, and for avoidance of defeat
measured as a dichotomous variable.
These predictions are the same as in the chapter concerning Militarized Interstate
Confrontations, so I do not repeat the full theoretical reasoning here.
119
7.6.2 Results
7.6.2.1 Descriptive Statistics
These analyses have an N of 113 crises. The number of paragraphs coded for integrative
complexity, per crisis, ranges from a minimum of 24 to a maximum of 6,525, with a mean of
622.58, and a standard deviation of 1,004.11.
Across the crises, the leaders’ crisis integrative complexity ranges from a minimum of
1.41 to a maximum of 3.33, with a mean of 2.05, and a standard deviation of 0.28.
The variable that represents the degree of success achieved by the leader’s side at the
resolution of the crisis has the coding: Defeat = 1, Stalemate or Compromise = 2, Victory = 3.
Across the crises it has a mean of 2.38, and a standard deviation of 0.69.
The variable that represents the avoidance of defeat by the leader’s side at the resolution
of the crisis is a binary variable (Defeat = 0, Avoidance of Defeat = 1). Across the crises it has a
mean of 0.88, and a standard deviation of 0.32.
The response variables for these regressions, namely Degree of Success and Avoidance
of Defeat, have a relatively even distribution of crises between Victory and
Stalemate/Compromise, but far fewer crises that are coded as ending in Defeat. This presents a
problem for testing whether defeat is specifically the outcome that is predicted by the leader’s
crisis IC.
120
Table 11
Distribution of Values for the Degree of Success and Avoidance of Defeat
Outcome
Categorical
Degree of
Success
Avoidance of
Defeat
n Crises
Valid Percent
Defeat
1
0
11
0.10
Stalemate or
Compromise
2
1
48
0.42
Victory
3
1
54
0.48
NA
NA
NA
2
NA
Total
115
1
The data presented in this table are for the same subset of crises as for the regressions, namely excluding
crises with fewer than 20 paragraphs scored for IC, and excluding crises that began during a world war.
7.6.2.2 Inferential Statistics: Predicting success, as an interval variable
In the regression below, the predictor variable is the IC expressed by the leader during the
international crisis. The response variable is the degree of success achieved by the leader’s
country at the outcome of the crisis. All of the regression models include covariates for country
(United States, Russia, and United Kingdom), and for time period (Pre Cold War versus a
category that merges Cold War/Post Cold War). For the variable time period, the Cold War and
Post Cold War were merged into one category, because when treated as separate categories, the
standard error of the estimate for this predictor was magnitudes of order larger than the estimate
itself, suggesting that it was dividing crises into groups that were too small to yield a meaningful
estimate. For the regression models three through six, additional covariates were included for the
variables Protracted Conflict, Gravity of Threat, Distance, and Power Discrepancy.
As always, confrontations with fewer than 20 paragraphs scored for IC are excluded from
the regression models. Confrontations that began during a world war are also excluded, for the
same reasons as given for the analyses of outcomes in the chapter concerning militarized
interstate confrontations. The results are in Table 12 below.
121
There is a positive sign of relationship between the leader’s IC during the crisis, and the
degree of success achieved by their country at the outcome of the crisis, but this relationship is
not statistically significant. In the model that includes IC as a predictor (model 2), as compared
to the model that does not (model 1), the AIC is higher, and the log likelihood is not substantially
different, suggesting that the inclusion of IC as a predictor does not improve the model fit. In the
model that includes IC as a predictor, as compared to the model that does not, the adjusted R-
square is lower.
122
Table 12
Regressions Predicting the Degree that the Leader’s Country Achieved Success
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
2.50***
2.38***
2.59***
2.25***
2.61***
2.38***
p < .01
p < .01
p < .01
p < .01
p < .01
p < .01
(0.13)
(0.56)
(0.58)
(0.57)
(0.63)
(0.57)
Country
-0.25
-0.26
-0.23
-0.20
-0.33
-0.29
RUS
p = .21
p = .20
p = .25
p = .35
p = .14
p = .19
(0.20)
(0.20)
(0.20)
(0.21)
(0.22)
(0.22)
Country
-0.08
-0.07
-0.01
-0.09
-0.06
-0.08
USA
p = .59
p = .66
p = .95
p = .61
p = .73
p = .64
(0.16)
(0.17)
(0.17)
(0.17)
(0.17)
(0.17)
Period
-0.25
-0.25
-0.25
-0.20
-0.30
-0.23
Pre CW
p = .23
p = .24
p = .24
p = .36
p = .17
p = .31
(0.21)
(0.21)
(0.21)
(0.22)
(0.22)
(0.23)
Crisis IC
0.06
0.06
0.04
0.06
0.06
p = .83
p = .82
p = .88
p = .82
p = .80
(0.25)
(0.25)
(0.26)
(0.25)
(0.26)
Protracted
-0.15
Conflict
p = .15
(0.11)
Gravity of
0.17
Threat
p = .37
(0.19)
Distance
-0.07
p = .40
(0.08)
Power
0.00
Discrepancy
p = .79
(0.00)
Num.Obs.
113
113
113
111
113
111
R2
0.033
0.033
0.052
0.038
0.040
0.030
R2 Adj.
0.006
-0.002
0.008
-0.008
-0.005
-0.017
AIC
231.6
233.6
233.4
231.3
234.8
233.4
BIC
245.3
249.9
252.5
250.3
253.9
252.4
Log.Lik.
-110.818
-110.793
-109.689
-108.669
-110.412
-109.701
RMSE
0.65
0.65
0.64
0.64
0.64
0.65
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
123
7.6.2.3 Inferential Statistics: Predicting avoidance of defeat, as a dichotomous variable
In the regression below, the predictor variable is the IC expressed by the leader during the
international crisis. The response variable is the avoidance of defeat (Defeat = 0, Avoidance of
Defeat = 1), coded from the ICB dataset variable “outcom.” All of the regression models include
covariates for country (United States, Russia, and United Kingdom), and for time period (Pre
cold War versus a category that merges Cold War/Post Cold War). The merging of values for
time period was done for the same reasons as in the section concerning degree of success
(above). For the regression models three through six, additional covariates were included for the
variables Protracted Conflict, Gravity of Threat, Distance, and Power Discrepancy.
As always, crises with fewer than 20 paragraphs scored for IC are excluded from the
regression models. Crises that began during a world war are also excluded, for the same reasons
as given for the regressions predicting the degree of success (above). The results are in Table 13
below.
There is a positive sign of relationship between the leader’s IC during the crisis, and their
avoidance of defeat at the outcome of the crisis, but this relationship is not statistically
significant. In the model that includes IC as a predictor, as compared to the model that does not,
the AIC is higher, and the log likelihood is not notably different, suggesting that the inclusion of
IC as a predictor does not improve the model fit.
124
Table 13
Regressions Predicting Whether the Leader’s Country Avoided Defeat
(1)
(2)
(3)
(4)
(5)
(6)
(Intercept)
2.79***
2.34
3.49
2.63
2.70
2.46
p <.01
p = .41
p = .26
p = .37
p = .40
p = .38
(0.81)
(2.83)
(3.10)
(2.94)
(3.21)
(2.79)
Country
-1.00
-1.04
-0.93
-1.16
-1.14
-0.95
RUS
p = .30
p = .30
p = .35
p = .27
p = .29
p = .36
(0.97)
(0.99)
(0.99)
(1.06)
(1.09)
(1.03)
Country
-0.50
-0.45
-0.16
-0.41
-0.43
-0.48
USA
p = .59
p = .64
p = .87
p = .68
p = .66
p = .63
(0.92)
(0.97)
(1.02)
(0.99)
(0.97)
(0.99)
Period
-0.48
-0.45
-0.51
-0.61
-0.55
-0.63
Pre CW
p = .63
p = .65
p = .61
p = .57
p = .61
p = .54
(0.99)
(1.00)
(1.00)
(1.09)
(1.08)
(1.01)
Crisis IC
0.20
0.18
0.26
0.21
0.10
p = .87
p = .89
p = .83
p = .87
p = .94
(1.24)
(1.29)
(1.23)
(1.24)
(1.22)
Protracted
-0.71
Conflict
p = .21
(0.56)
Gravity of
-0.47
Threat
p = .66
(1.08)
Distance
-0.10
p = .81
(0.43)
Power
0.00
Discrepancy
p = .72
(0.01)
Num.Obs.
113
113
113
111
113
111
AIC
78.7
80.7
81.0
82.2
82.6
81.8
BIC
89.6
94.3
97.4
98.4
99.0
98.1
Log.Lik.
-35.354
-35.341
-34.517
-35.079
-35.313
-34.911
RMSE
0.29
0.29
0.29
0.30
0.29
0.29
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
125
7.6.3 Discussion
For both the regressions predicting the degree of success at the outcome, and the
avoidance of defeat at the outcome, the leader’s integrative complexity during the confrontation
is not a significant predictor of the response variable. Furthermore, in both sets of regressions,
the inclusion of the leader’s integrative complexity did not improve the model fit, judging based
on the AIC and log likelihood. There is therefore little evidence that the leader’s integrative
complexity during an international crisis is an important predictor of their success, or avoidance
of defeat, at the outcome of the crisis.
In both types of regression, while not statistically significant, there was a positive sign of
relationship between the leader’s integrative complexity during crisis, and their degree of
success, or avoidance of defeat. This is also the sign of relationship that was found for the
analogous tests, in the chapter concerning Militarized Interstate Disputes, although those results
were also not statistically significant. This sign of relationship is consistent with the predictions
of the strategic judgment hypothesis, and is not consistent with the demonstration-of-resolve
hypothesis. So, while these results are not significant and are far from conclusive, they are
slightly more supportive of the strategic judgment hypothesis.
There are reasons to believe that, at least for the countries included in this study, the most
important distinction is between Defeat and Avoidance of Defeat, because these countries are so
powerful that often, when a crisis ends in Stalemate or Compromise, they can at a later date take
actions to achieve their goals, essentially achieving the same outcome as for crises that ended
with Victory. To the extent that this is true, the most appropriate response variable for the tests in
this section is a dichotomous variable, coding for Defeat and Avoidance of Defeat. However,
because, for these countries, defeat is a rare outcome, this variable has the problem of low
126
variance, due to the preponderant majority of crises being coded as ending in Avoidance of
Defeat. As a result, for the currently available data, there are no response data that can be
expected to give compelling results. Future studies may be able to address this weakness, either
by extending the number of countries included in the study, or by including different outcome
variables, which can either be measured more objectively, or which are more continuous
measures, or more specifically targeted to the issues under contention in the crisis.
7.7 Does a Leader’s Integrative Complexity During a Crisis Predict the Satisfaction of the
Adversary at the Outcome, Statistically Controlling for the Degree of Success Achieved by
the Leader’s Country?
7.7.1 Predictions
This is an exploratory test. In previous tests I found that, while there is a positive
relationship between a leader’s integrative complexity during an international crisis and the
degree of success that their country achieved at the resolution of the crisis, this relationship was
not significant and did not improve model fit. This led me to wonder whether this is because a
leader can achieve successful outcomes through two paths, one of which is a high IC path, and
the other of which is a low IC path. For instance, achieving success through deterrence (or
compellence) might be a low IC path, while achieving success through solving coordination
problems might be a high IC path. If this were the case, then the counteracting effects of the two
paths could result in no significant overall relationship.
This possibility can be tested, because the two different pathways imply different effects
upon the adversary country. According to the demonstration-of-resolve hypothesis, a leader with
low integrative complexity will be more likely to achieve success through the pathway of
pressuring adversaries to back down, even when the adversary is not satisfied with this outcome.
127
In contrast, according to the strategic judgment hypothesis a leader with high integrative
complexity will be more likely to achieve success through the pathway of solving coordination
problems to the satisfaction of both the leader’s country and the adversary country.
Together, these hypotheses imply that, for any degree of success achieved by the leader’s
country at the resolution of the crisis, the leader’s integrative complexity should be positively
associated with the degree to which the adversary country is satisfied with the outcome.
It is important to statistically control for the degree of success achieved by the leader
because a common feature of international crises is that the goals of the two sides appear to be in
conflict, such that the more successful one side is, the less successful the other side is, and the
less satisfied it is. Leaders can sometimes transcend this appearance of conflict, for instance by
identifying and solving coordination problems. While this will often still not result in both sides
achieving success, in the sense of both of them achieving their initial goals, it can address or
ameliorate the causes of dissatisfaction, even for the side that fails to achieve its initial goals. The
model implied by this theory is that the more successful the leader’s country is, the less satisfied
the adversary will be, but that relative to this baseline, leaders who exhibit higher IC will
sometimes solve coordination problems and thereby boost the satisfaction of the adversary.
Because this is a boost to the level of adversary satisfaction that would be expected given the
degree of success achieved by the leader, it is important to statistically control for the latter.
7.7.2 Results
7.7.2.1 Descriptive statistics
These analyses have an N of 129 crises. The number of paragraphs coded for integrative
complexity, per crisis, ranges from a minimum of 24 to a maximum of 6,525, with a mean of
622.58, and a standard deviation of 1,004.11.
128
Across the crises, the leaders’ crisis integrative complexity ranges from a minimum of
1.41 to a maximum of 3.33, with a mean of 2.05, and a standard deviation of 0.28.
The variable that represents the degree of success achieved by the leader’s side at the
resolution of the crisis has the coding: Defeat = 1, Stalemate or Compromise = 2, Victory = 3.
Across the crises it has a mean of 2.38, and a standard deviation of 0.69.
The variable that represents the satisfaction of the adversary government at the resolution
of the crisis is binary (Dissatisfied = 0, Satisfied = 1). Across the crises it has a mean of 0.46, and
a standard deviation of 0.50.
7.7.2.2 Inferential statistics
I used logarithmic regressions to test whether, controlling for the degree of success
achieved by the leader’s country at the outcome of the crisis, the leader’s mean IC during an
international crisis is a significant predictor of whether the adversary country was satisfied with
the outcome.
As in all of the tests, the regression models only include crises for which there were at
least 20 paragraphs scored for integrative complexity.
All of the regression models included covariates for Country (USA, RUS, GBR), Time
Period (Pre Cold War, Cold War, Post Cold War), and the degree to which the leader’s country
achieved Success at Outcome. All of the models, except model 6, included covariates for
whether the crisis began During a World War. This covariate was not included in model 6
because, when included in this model, it had a standard error that was orders of magnitude larger
than its estimate. In addition to these four covariates, models 3 to 7 each includes one of the
following covariates: Protracted Conflict, Gravity of Threat, Distance, Power Discrepancy, or a
129
dummy for the Leader’s Satisfaction at the outcome of the crisis. The results are in Table 14
below.
The leader’s IC during the crisis is a statistically significant predictor of the adversary’s
satisfaction at the outcome of the crisis, in all six of the models (models 2 through 7), with a
positive sign of relationship.
In the model that includes IC as a predictor (model 2), as compared to the model that
does not (model 1), the AIC is lower, and the log likelihood is higher, suggesting that including
the leader’s crisis integrative complexity as a predictor improves the model fit.
130
Table 14
Regressions Predicting Adversary Satisfaction, Controlling for the Leader’s Success
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(Intercept)
6.65***
1.14
1.60
1.83
1.16
1.19
0.73
p <0.01
p = .64
p = .54
p = .50
p = .64
p = .63
p = .77
(1.34)
(2.48)
(2.61)
(2.68)
(2.47)
(2.46)
(2.55)
Country
-0.09
-0.81
-0.75
-1.05
-0.78
-1.04
-0.74
RUS
p = .91
p = .36
p = .40
p = .27
p = .39
p = .29
p = .42
(0.75)
(0.88)
(0.89)
(0.96)
(0.90)
(0.98)
(0.91)
Country
-0.41
0.20
0.32
0.28
0.15
0.26
0.23
USA
p = .48
p = .76
p = .64
p = .68
p = .82
p = .69
p = .73
(0.58)
(0.66)
(0.69)
(0.67)
(0.68)
(0.66)
(0.67)
Period
-0.92+
-1.25*
-1.36*
-1.22*
-1.20*
-0.85
-1.39*
Post CW
p = .09
p = .03
p = .03
p = .04
p = .04
p = .16
p = .02
(0.54)
(0.58)
(0.62)
(0.58)
(0.59)
(0.60)
(0.61)
Period
0.29
0.08
0.08
-0.08
0.10
-0.08
-0.15
Pre CW
p = .74
p = .93
p = .93
p = .94
p = .92
p = .93
p = .87
(0.86)
(0.91)
(0.91)
(0.94)
(0.91)
(0.94)
(0.94)
During
-1.99+
-1.96+
-1.74
-2.08+
-1.99+
-1.57
World War
p = .08
p = .10
p = .16
p = .09
p = .09
p = .18
(1.13)
(1.18)
(1.23)
(1.22)
(1.19)
(1.16)
Degree of
-2.56***
-2.84***
-2.90***
-2.87***
-2.81***
-2.73***
-3.26***
Success
p <0.01
p <0.01
p <0.01
p <0.01
p <0.01
p <0.01
p <0.01
(0.45)
(0.51)
(0.53)
(0.52)
(0.52)
(0.55)
(0.59)
Crisis IC
2.95*
2.98*
3.01*
2.87*
2.86*
3.15*
p = .01
p = .01
p = .01
p = .02
p = .02
p = .01
(1.20)
(1.20)
(1.22)
(1.21)
(1.21)
(1.26)
Protracted
-0.24
Conflict
p = .57
(0.42)
Distance
-0.21
p = .48
(0.30)
Gravity of
0.13
Threat
p = .86
(0.74)
Power
-0.01
Discrep.
p = .28
(0.01)
Leader’s
1.23
Satisfact’n
p = .12
(0.79)
Num.Obs.
129
129
129
129
127
113
129
131
(1)
(2)
(3)
(4)
(5)
(6)
(7)
AIC
130.9
125.7
127.3
127.2
127.3
121.1
125.2
BIC
150.9
148.5
153.1
152.9
152.9
142.9
150.9
Log.Lik.
-58.427
-54.827
-54.662
-54.577
-54.674
-52.549
-53.601
F
5.697
4.859
4.217
4.224
4.152
3.776
4.462
RMSE
0.39
0.37
0.37
0.37
0.38
0.39
0.37
+ p < 0.1, * p < 0.05, ** p < 0.01, *** p < 0.001
The coefficient is unstandardized. The standard error is in parentheses.
7.7.3 Discussion
The results of this test, which show a positive relationship between the leader’s crisis IC
and adversary satisfaction, are consistent with both the strategic judgment hypothesis and with
the demonstration-of-resolve hypothesis. Because the results are consistent with either, or both,
pathways, it is not possible to conclude in favor of one over the other.
The observed relationship is statistically significant in all of the regression models, and
according to both the AIC and the log likelihood, the inclusion of the leader’s integrative
complexity during the crisis improves the model fit.
Overall, the results of these analyses, while exploratory, provide evidence that is
consistent with either, or both, of these hypotheses. While these results cannot distinguish
between the two hypotheses, they do lend support for the overarching theoretical position that
leaders with lower IC are more likely to achieve their goals in ways that are associated with
lower satisfaction of the adversary, while leaders with higher IC are more likely to achieve their
goals in ways that are associated with higher satisfaction of the adversary. This is consistent with
existing theory concerning integrative complexity, according to which low integrative
complexity is associated with inflexibility and failure to take into account the perspective of
others, while high integrative complexity is associated with flexibility and greater recognition of
other perspectives.
132
An additional implication of the theoretical model employed here is that the relationship
between the leader’s IC and adversary satisfaction should be stronger the more successful the
leader was at achieving their goals. If the leader failed to achieve their goals, and the adversary
achieved its goals, then the adversary may feel satisfied regardless of whether coordination
problems were solved, and therefore regardless of the leader’s IC. If both sides failed to achieve
their goals, whether due to coordination failure or deterrence failure, then there would be little
reason to expect that the leader’s integrative complexity would boost the satisfaction of the
adversary. In either case, if the leader failed to achieve their goals, there is less reason to expect
the leader’s IC to boost adversary satisfaction. I did some additional exploratory tests (which I do
not report in any more detail), of whether the degree of the leader’s success moderates the
relationship between their crisis IC and adversary satisfaction. The coefficient of the interaction
term (success x IC) was positive, as predicted by this theory, but was not statistically significant.
It is possible that more data could provide enough statistical power to support this analysis, but
for now the results do not justify drawing the conclusion that the leader’s success moderates the
relationship between their IC and adversary satisfaction.
A final note of caution is that higher adversary satisfaction does not necessarily imply a
reduced likelihood of a future resurgence of crises between the two sides. It is possible that an
adversary, having resolved a crisis to its satisfaction, would be emboldened to risk more crises in
the future. It is also possible that if a crisis were resolved by devastating an adversary, it would
be rendered unable or unwilling to engage in future crises. We should therefore not assume that
these results necessarily imply that leaders who exhibit high integrative complexity resolve crises
in ways that reduce the likelihood of future crises with the adversary. I did some additional
exploratory tests (which I do not report in any more detail) of the relationship between the
133
leader’s crisis IC and the incidence of subsequent crises with the adversary, and did not find
significant results.
134
Chapter 8: General Discussion
8.1 Summary and Interpretation of the Results
8.1.1 Level of hostility, reliance on violence, and fatalities
The leader’s integrative complexity during a crisis is significantly negatively associated
with the level of hostility employed by their country, and with the degree to which their country
relied on violence relative to nonviolent strategies. This is more consistent with the strategic
judgment hypothesis than it is with the demonstration-of-resolve hypothesis.
The leader’s integrative complexity during a crisis is significantly negatively associated
with the number of fatalities suffered by their country. In terms of the contribution of integrative
complexity to the adjusted R-squared, this is the largest effect found in this dissertation. Again,
this is more consistent with the strategic judgment hypothesis than it is with the demonstration-
of-resolve hypothesis.
Given that these results are not from experimentally controlled tests, it is not possible to
definitively identify causality. However, these results do hold, for both the crisis dataset and the
confrontations dataset, even after controlling for multiple theoretically relevant variables.
Whatever the precise causal pathways, there is therefore strong evidence that the leader’s
integrative complexity during a crisis is negatively associated with their country’s level of
hostility, reliance on violence, and suffering of fatalities during that crisis. This pattern of results
is predicted by the strategic judgment hypothesis, and is not predicted by the demonstration-of-
resolve hypothesis.
8.1.2 The initiation of violence
The leader’s pre-crisis integrative complexity does not significantly predict the initiation
of violence by their side. Nor does it significantly predict the initiation of violence, by others,
135
that targets the leader’s country. There is therefore little evidence that the pathway of either of
my hypotheses has a preponderant effect with respect to the causal processes leading to the first
use of violence.
However, the results of an exploratory test do suggest a qualification of this conclusion.
There is a near significant (p < .10) positive association between a leader’s pre-crisis integrative
complexity and the initiation of violence by proxies of the leader’s country. This suggests that
whether a leader’s integrative complexity is high, or low, could be associated with identifiable
distinct causal processes leading to the initiation of violence. With respect to this particular test,
the results suggest that leaders with relatively high integrative complexity are more likely to
initiate violence through the use of proxies.
Although I did not predict this, the result makes sense in that using proxies is a more
complex strategy than directly employing violence. It is possible that, rather than merely
avoiding costs by avoiding violence unqualifiedly, leaders with relatively higher integrative
complexity also use complex strategies to attempt to employ violence without bearing the costs
of doing so. To put this in the language of my theoretical framework, this suggests that leaders
with relatively high integrative complexity are more likely to use complex strategies to attempt
to circumvent the deterrent efforts of adversaries.
This should be taken as highly speculative, given that it is based on near-significant
results of an exploratory analysis. But there would be value in doing further investigation
concerning whether different levels of integrative complexity are associated with distinct types
of causal pathways that lead to violence.
136
8.1.3 Degree of success achieved at the resolution of the crisis
The leader’s crisis integrative complexity does not significantly predict the degree of
success achieved by their country (as an interval variable), or its avoidance of defeat (as a
dichotomous variable), at the resolution of the crisis. There is therefore little evidence that the
pathway of either hypothesis has a preponderant effect with respect to the causal processes
leading to achieving success at the resolution of the crisis.
Again, the results of an exploratory test suggest a qualification. Controlling for the degree
of success achieved by the leader’s country, the leader’s crisis integrative complexity is a
significant positive predictor of the satisfaction of the adversary country. This is consistent with
either, or both, of my hypotheses. The strategic judgment hypothesis is consistent with this
pattern because, according to this hypothesis, leaders with high integrative complexity are more
likely to solve coordination problems, resulting in them achieving their goals while also allowing
the adversary country to be satisfied. The demonstration-of-resolve hypothesis is consistent with
this pattern because, according to this hypothesis, leaders with low integrative complexity are
more likely to successfully pressure adversaries to back down, resulting in them achieving their
goals, but leaving the adversary country dissatisfied.
Although either hypothesis alone is theoretically consistent with the pattern of results, the
fact that the effect is so strong and robust to controls suggests that both pathways may be
operating. In other words, in international crises, leaders with relatively high complexity may be
relatively more likely to achieve their goals through pathways that involve solving coordination
problems. While, on the other hand, leaders with relatively low integrative complexity may be
relatively more likely to achieve their goals through pathways that involve demonstrating their
resolve, and pressuring adversaries to back down.
137
This should be taken as speculative, given that it is based on the results of an exploratory
analysis. But it indicates that there would be value in doing further investigation concerning
whether different levels of integrative complexity are associated with distinct types of causal
pathways by which international crises can be resolved.
8.2 Theoretical Contributions
8.2.1 A new theoretical framework
This paper contributes a new overarching theoretical framework which can be used to
generate hypotheses that relate the integrative complexity of political leaders to foreign policy
interactions. This frames international crises and conflicts as situations that present distinct types
of cognitive challenges, namely coordination problems and problems of uncertain resolve (such
as deterrence challenges). I use this framework to generate two hypotheses concerning the
relevance of a leader’s integrative complexity to how they approach these cognitive challenges.
According to the strategic judgment hypothesis, leaders with relatively high integrative
complexity will be more likely to solve coordination problems, and will therefore be more likely
to avoid costly conflicts, in particular violence, that can arise from failure to coordinate.
According to the demonstration-of-resolve hypothesis, leaders with relatively low integrative
complexity will be more likely to credibly demonstrate resolve to adversaries, thereby deterring
them from taking actions against the leaders country, and will therefore be more likely to avoid
costly conflicts, in particular violence.
Although the current paper is primarily focused on these two hypotheses, the overarching
framework can be used to generate more fine-grained hypotheses, e.g., that leaders with higher
integrative complexity may be more likely to use proxies to initiate violence, while attempting to
avoid bearing the costs of doing so.
138
8.2.2 The empirical results mainly support the strategic judgment hypothesis
I find that the empirical results support the conclusion that the pathway of the strategic
judgment hypothesis predominates over the pathway of the demonstration-of-resolve hypothesis.
This is supported by the following results: the leader’s crisis integrative complexity is
significantly negatively associated with the level of hostility employed by their country, and with
their country’s reliance on violence relative to other strategies, and with the fatalities suffered by
their country. However, this pattern of overall support for this hypothesis has several important
qualifications.
The first qualification is that the two hypotheses are not mutually exclusive. Just because
one pathway predominates in the overall pattern of results does not mean that the other pathway
could not also have a causal effect. The operation of both pathways is consistent with the finding
that the leaders’ crisis integrative complexity does not predict their success at the resolution of
the crisis.
The operation of both pathways is also consistent with the exploratory test that found
that, controlling for the degree of success achieved by the leaders’ country, the leaders’
integrative complexity is a significant positive predictor of the satisfaction of the adversary
country. Although this could be consistent with either of the hypotheses, it is a particularly good
fit with a model in which both pathways operate.
If both pathways operate, then: (a) leaders exhibiting high integrative complexity would
be particularly likely to address crises by solving coordination problems, allowing them to
achieve success while also satisfying the adversary country; and (b) leaders exhibiting low
integrative complexity would be particularly likely to demonstrate resolve to the adversary,
getting the adversary to back down, resulting in outcomes in which the leader’s country has
139
achieved success, but the adversary is dissatisfied. The observed results are consistent with this
pattern.
Even if both pathways operate to some degree, the evidence supports the conclusion that
the pathway of the strategic judgment hypothesis has the predominant effect with respect to
hostility, reliance on violence, and suffering of fatalities.
The second qualification is that it is possible that those situations that especially call for
low integrative complexity are also those situations that are particularly violent. In other words,
contests of resolve may be associated with high violence and high fatalities, and also be
associated with low integrative complexity. But it does not logically follow that in the
counterfactual in which a leader were to have higher integrative complexity, that the result would
be less violence and fewer fatalities.
The third qualification is that the leader’s pre-crisis integrative complexity does not
predict the leader’s side initiating violence, or being the target of the initial violence. There is
therefore no evidence that the pathway of either hypothesis predominates with respect to the
initiation of violence. There may even be situations in which high integrative complexity is
associated with the use of violence, i.e., with the use of proxies to initiate violence. As
mentioned, this suggests a new hypothesis to explore, namely that high integrative complexity is
positively associated with the use of complex strategies for employing violence while attempting
to avoid the costs of doing so. But at this preliminary state, this new hypothesis is highly
speculative.
140
8.3 Methodological Contributions
8.3.1 A new corpus
This research has produced an extensive corpus of verbal materials attributed to the heads
of government of the United States, Russia/Soviet Union, and the United Kingdom. This corpus
can be expanded to more countries, and can be coded for any variable that can be coded using
standard methods of content analysis.
8.3.2 A new dataset
This research has produced a large N and highly representative dataset of the integrative
complexity of the aforementioned heads of government. Again, this approach can be extended to
more countries, and to more content analysis variables.
8.3.3 An approach that integrates data from different domains of research
This research has shown how content analysis data, in this case integrative complexity,
can be combined with international relations datasets, and more broadly political science
datasets. This approach has excellent ecological validity, because it uses data from real-world
situations and behaviors. It also has a much larger N than has been common in previous studies
of the integrative complexity of politicians and of other political actors. This method can be
employed to further study the integrative complexity of political leaders in international crises. It
can also be applied to different content analysis variables, e.g., motives, or moral foundations,
and to different types of political situation or interaction, e.g., elections, or domestic protests.
8.4 Limitations
This research did not include controlled experiments. Therefore, even with extensive
statistical controls, the results cannot provide certainty with respect to the direction of causal
effects, or with respect to the causal role of third variables.
141
The research is limited to only three countries. This limits the N that can be achieved,
which, while still much larger than previous research in the area, made it impracticable to test
mediation or moderation effects. This also prevents generalization to non-western countries, or to
countries that are not pre-eminent great powers.
The coding of the Militarized Interstate Confrontations (MIC) and International Crisis
Behavior (ICB) datasets is less transparent than is ideal, making it difficult to precisely interpret
some of the variables.
It was not possible to write computer code that would take the existing ICB data and
generate codes that indicate which actor initiated violence in the crisis. As a result I had to code
this myself. To do this I wrote a set of rules for how to do the coding, and I relied on existing
ICB variables as much as possible. However, I did not do the coding blind, and I had to adapt my
coding rules as problems arose. Therefore the results concerning the initiation of violence are
particularly provisional.
This research does not distinguish between the role of internal cognition and external
messaging. This is because the corpus that was coded for integrative complexity contains
predominantly verbal materials intended for a public audience. As a result, the negative
association between integrative complexity and hostilities/violence/fatalities, does not tell us
whether that effect is driven by the internal cognition of political leaders, or by their
communication strategies when they address the public. Both of these possibilities are consistent
with my framework and hypotheses, so it is not necessary to distinguish between them in the
current research. But the distinction between them is a meaningful one, which the current
research cannot address.
142
This research does not test whether the observed effects are driven by the state or trait
component of integrative complexity. In other words, it does not tell us whether leaders who are
generally low in complexity are more likely to rely on violence; or whether leaders show a state
decrease in complexity in association with the use of violence; or both. Again, it is not necessary
to distinguish between these possibilities in the current research, but the distinction between
them is a meaningful one.
8.5 Opportunities for Future Research
8.5.1 Add more countries to the dataset
I, and my colleague Zlatin Mitkov, are currently adding China and India to the dataset.
This will allow for tests of whether the results generalize to non-western countries, and to less
powerful countries. It will also allow for investigation into how domestic political institutions
moderate the relationship between the leaders’ integrative complexity and the use of violence by
their country. It is possible, for instance, that regimes in which the leader has relatively
unchecked executive power, e.g., Stalin or Mao’s regimes, show different patterns from regimes
in which the leader must operate within a system of checks and balances, e.g., modern Britain
and the United States.
8.5.2 Distinguish between the state and trait components of integrative complexity
It is possible to partially disentangle the state and trait components of leaders’ integrative
complexity. A variable that is closer to the state component can be generated by mean-centering
the integrative complexity within each leader, so that the variance is purely due to state changes.
A variable that is closer to the trait component can be generated by calculating the mean
integrative complexity of each leader. A caveat is that these are still not pure measures of state
143
and trait components, because situations that affect a person’s psychological state, such as the
prevalence of political violence, could persist over much a leader’s life.
Statistically distinguishing between the state and trait components of integrative
complexity would make it possible to test to what degree the observed effects are driven by the
state or trait components of integrative complexity.
8.5.3 Do qualitative investigations of crises
One can do qualitative investigations of those crises for which there is sufficient data that
it is feasible to track how the leader’s integrative complexity changes throughout the crisis.
These investigations can look into whether the order of changes in integrative complexity, and in
political events and actions, is the order predicted by the two hypotheses. Once can also
investigate whether the content of the leader’s verbal materials aligns with their integrative
complexity in a way that the hypotheses would predict. For instance, are messages that are low in
integrative complexity associated with content that indicates that the leader’s country cannot be
pressured into backing down? Are messages that are high in integrative complexity associated
with content that is consistent with the leader attempting to solve coordination problems?
One can also qualitatively investigate the relationship between the leader’s integrative
complexity and the satisfaction of the adversary country at the resolution of the crisis. Is there
qualitative evidence that, in cases in which leaders exhibit relatively high integrative complexity,
and who have achieved their goals at the resolution of the crisis, that the leaders solved
coordination problems? Conversely, is there qualitative evidence that in crises in which leaders
exhibit relatively low integrative complexity, and who have achieved their goals at the resolution
of the crisis, that they did so by demonstrating resolve to their adversaries, inducing them to back
down? These investigations would shed light on whether there are distinct strategies by which
144
leaders can achieve their goals, and whether these strategies are associated with different levels
of integrative complexity.
8.5.4 Conduct experiments
The strategic judgment hypothesis can be tested by using an iterated prisoner’s dilemma
game (which is a game that simulates a coordination problem). The strategic judgment
hypothesis predicts that, in an iterated prisoner’s dilemma game, the integrative complexity of
players would be positively associated with their success in the game.
The demonstration-of-resolve hypothesis can be tested by using a game of chicken
(which is a game that simulates a problem of uncertain resolve). The demonstration-of-resolve
hypothesis predicts that, in a game of chicken, the integrative complexity of players would be
negatively associated with their success in the game.
Using these games, experiments can be conducted to test, with respect to the predictions
of the two hypotheses:
(a) whether iterated prisoner’s dilemma games tend to elicit higher integrative complexity
in players than do games of chicken.
(b) whether, in games of iterated prisoner’s dilemma, players with high integrative
complexity tend to be cooperative and successful, while in games of chicken, players with low
integrative complexity tend to defect and be successful.
(c) whether experimentally manipulating a player’s integrative complexity prior to a
game, by giving them tasks that elicit high or low complexity, influences their strategic choices
and the degree of success that they achieve, in the ways predicted by each hypothesis;
145
(d) whether experimentally manipulating a written profile of a player, so that it exhibits
either high or low integrative complexity, changes how other players interact with that player, in
the ways predicted by each hypothesis.
Conducting experiments would address those limitations of the current study that are
inherent to non-experimental research. In particular, controlled experiments would allow the
researcher to identify a causal relationship between a person’s integrative complexity, and their
success in games that simulate the cognitive challenges involved in each of the two hypotheses.
146
References
Achen, C. H., & Snidal, D. (1989). Rational deterrence theory and comparative case studies.
41(2), 143–169.
Allison, G. (1971). Essence of decision: Explaining the Cuban missile crisis (1st ed.). Little Brown.
Allison, G. T. (2017). Destined for war? The National Interest, 149, 9–21.
American Presidency Project. (n.d.). [Archive]. American Presidency Project. Retrieved
September 2, 2021, from https://www.presidency.ucsb.edu/
Americans know surprisingly little about their government, survey finds. (2014). Annenberg
Public Policy Center.
Atran, S., & Henrich, J. (2010). The evolution of religion: How cognitive by-products, adaptive
learning heuristics, ritual displays, and group competition generate deep commitments
to prosocial religions. Biological Theory, 5(1), 18–30.
Atran, S., & Norenzayan, A. (2004). Religion’s evolutionary landscape: Counterintuition,
commitment, compassion, communion. Behavioral and Brain Sciences, 27(6), 713–730.
Baker-Brown, G., Ballard, E. J., Bluck, S., de Vries, B., Suedfeld, P., & Tetlock, P. E. (1990). Coding
manual for conceptual/integrative complexity. University of British Columbia and
University of California, Berkeley.
Berman, E., & Laitin, D. D. (2008). Religion, terrorism and public goods: Testing the club model.
Journal of Public Economics, 92(10–11), 1942–1967.
Boyd, R. L., Ashokkumar, A., Seraj, S., & Pennebaker, J. W. (2022). The development and
psychometric properties of LIWC-22. University of Texas at Austin. https://www.liwc.app
147
Brecher, M. (1993). Crises in world politics: Theory and reality. Pergamon Press.
Brecher, M. (2008). International political earthquakes. University of Michigan Press.
Brecher, M. (2018). A century of crisis and conflict in the international system. Springer
International Publishing.
Brecher, M., & Wilkenfeld, J. (1997). A study of crisis. The University of Michigan Press.
Bueno De Mesquita, B., & Smith, A. (2012). Domestic explanations of international relations.
Annual Review of Political Science, 15(1), 161–181.
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multimodel inference: A practical
information-theoretic approach (2nd ed). Springer.
Carvalho, J.-P. (2019). Religious clubs: The strategic role of religious identity. In Advances in the
Economics of Religion (pp. 25–41). Springer International Publishing.
Carvalho, J.-P. (2020). Religion and terrorism: The religious utility hypothesis. In Handbook of
the economics of terrorism.
Churchill, W. (1938, October 5). A total and unmitigated defeat. https://hansard.parliament.uk/
Cohen, D., & Nisbett, R. E. (1994). Self-Protection and the culture of honor: Explaining southern
violence. Personality and Social Psychology Bulletin, 20(5), 551–567.
Cohen, D., Nisbett, R. E., Bowdle, B. F., & Schwarz, N. (1996). Insult, aggression, and the
southern culture of honor: An “experimental ethnography.” 70(5), 945–960.
Conway, L. G., III, Conway, K. R., Gornick, L. J., & Houck, S. C. (2014). Automated integrative
complexity. Political Psychology, 35(5), 603–624.
148
Conway, L. G., III, Conway, K. R., & Houck, S. C. (2020). Validating automated integrative
complexity: Natural language processing and the Donald Trump Test. Journal of Social
and Political Psychology, 8(2), 504–524.
Conway, L. G., III, Gornick, L. J., Burfeind, C., Mandella, P., Kuenzli, A., Houck, S. C., & Fullerton,
D. T. (2012). Does complex or simple rhetoric win elections? An integrative complexity
analysis of U.S. presidential campaigns. Political Psychology, 33(5), 599–618.
Conway, L. G., III, Gornick, L. J., Houck, S. C., Anderson, C., Stockert, J., Sessoms, D., & McCue, K.
(2016). Are conservatives really more simple-minded than liberals? The domain
specificity of complex thinking. Political Psychology.
Conway, L. G., III, Suedfeld, P., & Tetlock, P. E. (2001). Integrative complexity and political
decisions that lead to war or peace. In Peace, conflict, and violence: Peace psychology
for the 21st century (pp. 66–75).
Conway, L. G., III, & Zubrod, A. (2022). Are U.S. presidents becoming less rhetorically complex?
Evaluating the integrative complexity of Joe Biden and Donald Trump in historical
context. Journal of Language and Social Psychology.
Cross, R. W., Duncan, L. A., Buxton, C. P., & Jolicoeur, P. (2018). Applying Integrative Complexity
and Motive Imagery to gain insight into groups operating in the Ukraine conflict (DRDC-
RDDC-2018-C115). Defence Research and Development Canada.
Delli Carpini, M. X., & Keeter, S. (1996). What Americans know about politics and why it
matters. Yale University Press.
Deutsch, K. W., & Singer, J. D. (1964). Multipolar power systems and international stability.
World Politics, 16(3), 390–406.
149
Doyle, M. W. (2005). Three pillars of the liberal peace. American Political Science Review, 99(3),
463–466.
Doyle, M. W., & Sambanis, N. (2000). International peacebuilding: A theoretical and
quantitative analysis. American Political Science Review, 94(4), 779–801.
Druckman, J. N. (2001). The implications of framing effects for citizen competence. Political
Behavior, 23(3), 225–256.
Fearon, J. D. (1995). Rationalist explanations for war. International Organization, 49(3), 379–
414.
Gartzke, E. (1999). War Is in the error term. International Organization, 53(3), 567–587.
Gerzhoy, G. (2015). Alliance coercion and nuclear restraint: How the United States thwarted
West Germany’s nuclear ambitions. International Security, 39(4), 91–129.
Gibler, D. M., & Miller, S. V. (2023). The Militarized Interstate Confrontation Dataset, 1816-
2014. Journal of Conflict Resolution, 1-25.
Gilpin, R. (1981). War and change in international politics. Cambridge University Press.
Goddard, S. E. (2006). Uncommon ground: Indivisible territory and the politics of legitimacy.
International Organization, 60(01).
Gómez, Á., López-Rodríguez, L., Sheikh, H., Ginges, J., Wilson, L., Waziri, H., Vázquez, A., Davis,
R., & Atran, S. (2017). The devoted actor’s will to fight and the spiritual dimension of
human conflict. Nature Human Behaviour, 1(9), 673–679.
Gómez, Á., Vázquez, A., & Atran, S. (2023). Transcultural pathways to the will to fight.
Proceedings of the National Academy of Sciences, 120(24), e2303614120.
150
Gruenfeld, D. H., Thomas-Hunt, M. C., & Kim, P. H. (1998). Cognitive flexibility, communication
strategy, and integrative complexity in groups: Public versus private reactions to
majority and minority status. Journal of Experimental Social Psychology, 34(2), 202–226.
Haidt, J. (2001). The emotional dog and its rational tail: A social intuitionist approach to moral
judgment. Psychological Review, 108(4), 814–834.
Hansard. (n.d.). UK Parliament. Retrieved September 2, 2021, from
https://hansard.parliament.uk/
Houck, S. C., Conway, L. G., III, & Gornick, L. J. (2014). Automated integrative complexity:
Current challenges and future directions. Political Psychology, 35(5), 647–659.
Huntington, S. P. (1985). Conventional deterrence and conventional retaliation in Europe. In
Military strategy in transition (1st ed.). Routledge.
Huth, P. (1999). Deterrence and international conflict: Empirical findings and theoretical
debates. Annual Review of Political Science, 2, 25–48.
Huth, P., & Russett, B. (1990). Testing deterrence theory: Rigor makes a difference. World
Politics, 42(4), 466–501.
ICB Project—International Crisis Behavior. (n.d.). [Dataset]. ICB Project - International Crisis
Behavior. Retrieved September 2, 2021, from https://sites.duke.edu/icbdata/
Ikenberry, G. J., Mastanduno, M., & Wohlforth, W. C. (2009). Introduction: Unipolarity, state
behavior, and systemic consequences. World Politics, 61(1), 1–27.
Jackson, M. G. (2005). Beyond brinkmanship: Eisenhower, nuclear war fighting, and Korea,
1953-1968. Presidential Studies Quarterly, 35(1), 52–75.
151
Jervis, R. (1976). Perception and misperception in international politics: New edition (Revised).
Princeton University Press.
Jervis, R. (1978). Cooperation under the security dilemma. World Politics, 30(2), 167–214.
Johnson, D. D. P., & Toft, M. D. (2014). Grounds for war: The evolution of territorial conflict.
International Security, 38(3), 7–38.
Karvetski, C. W., Meinel, C., Maxwell, D. T., Lu, Y., Mellers, B. A., & Tetlock, P. E. (2022). What
do forecasting rationales reveal about thinking patterns of top geopolitical forecasters?
International Journal of Forecasting, 38(2), 688-704.
Keller, J. W., Yang, Y. E., & James, P. (2022). Decision-making in U.S. foreign policy crises:
Presidential leadership and outcomes. Political Research Quarterly, 73(3), 562-575.
Keohane, R. O. (1984). After hegemony. Princeton University Press.
Keohane, R. O. (1988). International institutions: Two approaches. International Studies
Quarterly, 32(4), 379–396.
Kindleberger, C. P. (1986). The world in depression, 1929-1939. University of California Press.
Krasner, S. D. (1976). State power and the structure of international trade. World Politics, 28(3),
317–347.
Lebow, R. N. (2010). Why nations fight: Past and future motives for war. Cambridge University
Press.
Lebow, R. N. (2020). Between peace and war. Springer Nature.
Liht, J., & Savage, S. (2013). Preventing violent extremism through value complexity: Being
Muslim being British. Journal of Strategic Security, 6(4), 44–66.
152
Liht, J., Suedfeld, P., & Krawczyk, A. (2005). Integrative complexity in face-to-face negotiations
between the Chiapas guerrillas and the Mexican government. Political Psychology,
26(4), 543–552.
Maoz, Z., & Russett, B. (1993). Normative and structural causes of democratic peace, 1946–
1986. American Political Science Review, 87(3), 624–638.
Marxists Internet Archive. (n.d.). [Archive]. Marxists Internet Archive. Retrieved September 2,
2021, from https://www.marxists.org/
Mearsheimer, J. J. (1985). Nuclear weapons and deterrence in Europe. International Security,
9(3), 19–46.
Mearsheimer, J. J. (2001). The tragedy of great power politics. WW Norton & Company.
Nalebuff, B. (1986). Brinkmanship and nuclear deterrence: The neutrality of escalation. Conflict
Management and Peace Science, 9(2), 19–30.
Oneal, J. R., & Ray, J. L. (1997). New tests of the democratic peace: Controlling for economic
interdependence, 1950-85. Political Research Quarterly, 50(4), 751–775.
Palmer, G., McManus, R. W., D’Orazio, V., Kenwick, M. R., Karstens, M., Bloch, C., Dietrich, N.,
Kahn, K., Ritter, K., & Soules, M. J. (2022). The MID5 Dataset, 2011–2014: Procedures,
coding rules, and description. Conflict Management and Peace Science, 39(4), 470–482.
Porter, C. A., & Suedfeld, P. (1981). Integrative complexity in the correspondence of literary
figures: Effects of personal and societal stress. Journal of Personality and Social
Psychology, 40, 321–330.
Powell, R. (1988). Nuclear brinkmanship with two-sided incomplete information. American
Political Science Review, 82(1), 155–178.
153
President of Russia. (n.d.). [Government]. President of Russia. http://en.kremlin.ru/
Putnam, R. D. (1988). Diplomacy and domestic politics: The logic of two-level games.
International Organization, 42(3), 427–460.
Reus-Smit, C. (1999). The moral purpose of the state: Culture, social identity, and institutional
rationality in international relations. Princeton University Press.
Rohrer, J. M. (2018). Thinking clearly about correlations and causation: Graphical causal models
for observational data. Advances in Methods and Practices in Psychological Science, 1(1),
27–42.
Ruggie, J. G. (1982). International regimes, transactions, and change: Embedded liberalism in
the postwar economic order. International Organization, 36(2), 379–415.
Russett, B. M. (1993). Grasping the democratic peace: Principles for a post-Cold War world.
Princeton University Press.
Sarkees, M. R., & Wayman, F. (2010). Resort to war: 1816-2007. CQ Press.
Schelling, T. C. (1980). The strategy of conflict: With a new preface by the author. Harvard
University Press.
Schroder, H. M., Driver, M. J., & Streufert, S. (1967). Human information processing. Holt,
Rinehart, & Winston.
Senese, P. D., & Vasquez, J. A. (2003). A unified explanation of territorial conflict: Testing the
impact of sampling bias, 1919-1992. International Studies Quarterly, 47(2), 275–298.
Senese, P. D., & Vasquez, J. A. (2005). Assessing the steps to war. British Journal of Political
Science, 35(4), 607–633.
154
Smith, A. G., Suedfeld, P., Conway, L. G., III, & Winter, D. G. (2008). The language of violence:
Distinguishing terrorist from nonterrorist groups by thematic content analysis. Dynamics
of Asymmetric Conflict, 1(2), 142–163.
Suedfeld, P. (1980). Indices of world tension in the bulletin of the atomic scientists. Political
Psychology, 114–123.
Suedfeld, P. (1985). APA presidential addresses: The relation of integrative complexity to
historical, professional, and personal factors. Journal of Personality and Social
Psychology, 49(6), 1643–1651.
Suedfeld, P. (1992). Cognitive managers and their critics. Political Psychology, 13(3), 435–453.
Suedfeld, P. (2010). The cognitive processing of politics and politicians: Archival studies of
conceptual and integrative complexity. Journal of Personality, 78(6), 1669–1702.
Suedfeld, P. (2014). Political and military geniuses: Psychological profiles and responses to
stress. In D. K. Simonton (Ed.), The Wiley Handbook of Genius (pp. 244–265). Wiley.
Suedfeld, P. (2022). Integrative complexity: Analyzing and modifying extremist thinking [Poster].
Canadian Psychological Association Annual Convention, Calgary, Canada.
Suedfeld, P., & Bluck, S. (1988). Changes in integrative complexity prior to surprise attacks.
Journal of Conflict Resolution, 32(4), 626–635.
Suedfeld, P., Bluck, S., & Ballard, E. J. (1994). The effects of emotional involvement and
psychological distance on integrative complexity. Journal of Applied Social Psychology,
24(5), 443–452.
Suedfeld, P., & Coren, S. (1992). Cognitive correlates of conceptual complexity. Personality and
Individual Differences, 13(11), 1193–1199.
155
Suedfeld, P., Corteen, R. S., & McCormick, C. (1986). The role of integrative complexity in
military leadership: Robert E. Lee and his opponents. Journal of Applied Social
Psychology, 16(6), 498–507.
Suedfeld, P., Cross, R. W., & Logan, M. C. (2013). Can thematic content analysis separate the
pyramid of ideas from the pyramid of action? A comparison among different degrees of
commitment to violence. In H. Cabayan, V. Sitterle, & M. Yandura (Eds.), Looking back,
looking forward: Perspectives on terrorism and responses to it (pp. 61–68). United States
of America Department of Defense.
Suedfeld, P., Grunert, L., & Morrison, B. H. (2020). Deradicalization in history and psychology: A
selective review of the literature (p. 18) [White paper]. Strategic Multilayer Assessment
(SMA). https://nsiteam.com/sma-publications/
Suedfeld, P., & Morrison, B. H. (2015). Thematic content analysis of texts by relevant leaders
and groups. In Drivers of conflict and convergence in the Asia-Pacific region in the next 5-
25 years. Strategic Multilayer Assessment (SMA).
Suedfeld, P., & Morrison, B. H. (2016). Leaders under stress and change: Thematic content
analysis. In B. Bragg (Ed.), Drivers of conflict and convergence in Eurasia in the next 5-25
years. Strategic Multilayer Assessment (SMA).
Suedfeld, P., & Morrison, B. H. (2019). Negotiating with the leaders of the Democratic People’s
Republic of Korea: Inferences from thematic content analysis [White paper]. Strategic
Multilayer Assessment (SMA). NSITEAM.com
156
Suedfeld, P., Morrison, B. H., & Cross, R. W. (2014). Approach 1: Integrative Complexity. In A
multi-disciplinary, multi-method approach to leader assessment at a distance: The case
of Bashar al-Assad—Part II. Strategic Multilayer Assessment (SMA).
Suedfeld, P., & Rank, A. D. (1976). Revolutionary leaders: Long-term success as a function of
changes in conceptual complexity. Journal of Personality and Social Psychology, 34(2),
169–178.
Suedfeld, P., & Tetlock, P. (1977). Integrative complexity of communications in international
crises. The Journal of Conflict Resolution, 21(1), 169–184.
Suedfeld, P., & Tetlock, P. E. (2014). Integrative complexity at forty: Steps toward resolving the
scoring dilemma. Political Psychology, 35(5), 597–601.
Suedfeld, P., Tetlock, P. E., & Ramirez, C. (1977). War, peace, and integrative complexity: UN
speeches on the Middle East problem, 1947-1976. The Journal of Conflict Resolution,
21(3), 427–442.
Suedfeld, P., Wallace, M. D., & Thachuk, K. L. (1993). Changes in integrative complexity among
Middle East leaders during the Persian Gulf crisis. Journal of Social Issues, 49(4), 183–
199.
Tetlock, P. E. (1983). Accountability and complexity of thought. Journal of Personality and Social
Psychology, 45(1), 74–83.
Tetlock, P. E. (1986). A value pluralism model of ideological reasoning. Journal of Personality
and Social Psychology, 50(4), 819–827.
Tetlock, P. E. (2003). Thinking the unthinkable: Sacred values and taboo cognitions. Trends in
Cognitive Sciences, 7(7), 320–324.
157
Tetlock, P. E., Armor, D., & Peterson, R. S. (1994). The slavery debate in antebellum America:
Cognitive style, value conflict, and the limits of compromise. Journal of Personality and
Social Psychology, 66(1), 115–126.
Tetlock, P. E., Metz, S. E., Scott, S. E., & Suedfeld, P. (2014). Integrative complexity coding raises
integratively complex issues. Political Psychology, 35(5), 625–634.
Tetlock, P. E., & Tyler, A. (1996). Churchill’s cognitive and rhetorical Style: The debates over Nazi
intentions and self-government for India. Political Psychology, 17(1), 149–170.
Toft, M. D. (2002). Indivisible territory, geographic concentration, and ethnic war. Security
Studies, 12(2), 82–119.
Tossell, C. C., Gómez, A., De Visser, E. J., Vázquez, A., Donadio, B. T., Metcalfe, A., Rogan, C.,
Davis, R., & Atran, S. (2022). Spiritual over physical formidability determines willingness
to fight and sacrifice through loyalty in cross-cultural populations. Proceedings of the
National Academy of Sciences, 119(6), e2113076119.
Vandello, J. A., Cohen, D., & Ransom, S. (2008). U.S. southern and northern differences in
perceptions of norms about aggression: Mechanisms for the perpetuation of a culture of
honor. Journal of Cross-Cultural Psychology, 39(2), 162–177.
Vasquez, J. A. (1996). Distinguishing rivals that go to war from those that do not: A quantitative
comparative case study of the two paths to war. International Studies Quarterly, 40(4),
531–558.
Vasquez, J. A. (2001). Mapping the probability of war and analyzing the possibility of peace: The
role of territorial disputes. Conflict Management and Peace Science, 18(2), 145–173.
158
Vasquez, J. A. (2010). The significance of territory. In J. A. Vasquez & M. T. Henehan, Territory,
War, and Peace (1st ed., pp. 209–226). Routledge.
Vasquez, J. A., & Henehan, M. T. (2001). Territorial disputes and the probability of war, 1816-
1992. Journal of Peace Research, 38(2), 123–138.
Vasquez, J. A., & Leskiw, C. S. (2001). The origins and war proneness of interstate rivalries. In
Annual Review of Political Science (Vol. 4, pp. 295–316). Annual Reviews.
https://doi.org/10.1146/annurev.polisci.4.1.295
Wallace, M. D., & Suedfeld, P. (1988). Leadership performance in crisis: The longevity-
complexity link. International Studies Quarterly, 32(4), 439–451.
Wallace, M. D., Suedfeld, P., & Thachuk, K. (1993). Political rhetoric of leaders under stress in
the Gulf Crisis. Journal of Conflict Resolution, 37(1), 94–107.
Wallace, M. D., Suedfeld, P., & Thachuk, K. A. (1996). Failed leader or successful peacemaker?
Crisis, behavior, and the cognitive processes of Mikhail Sergeyevitch Gorbachev. Political
Psychology, 17(3), 453.
Walt, S. M. (1990). The origins of alliance. Cornell University Press.
Waltz, K. N. (1979). Theory of international politics. Waveland Press, Inc.
Wendt, A. (1999). Social theory of international politics (Vol. 67). Cambridge University Press.
Wheeler-Bennett, J. (1963). Munich: Prologue to tragedy. Springer.
Wilson Center Digital Archive. (n.d.). [Archive]. Wilson Center.
https://digitalarchive.wilsoncenter.org/
Wohlforth, W. C. (1999). The Stability of a unipolar world. International Security, 24(1), 5–41.
159
Zmigrod, L. (2020). The role of cognitive rigidity in political ideologies: Theory, evidence, and
future directions. Current Opinion in Behavioral Sciences, 34, 34–39.
Zmigrod, L., Eisenberg, I. W., Bissett, P. G., Robbins, T. W., & Poldrack, R. A. (2021). The
cognitive and perceptual correlates of ideological attitudes: A data-driven approach.
Philosophical Transactions of the Royal Society B: Biological Sciences, 376(1822),
20200424.
Zmigrod, L., Rentfrow, P. J., & Robbins, T. W. (2020). The partisan mind: Is extreme political
partisanship related to cognitive inflexibility? Journal of Experimental Psychology:
General, 149(3), 407–418.
160
Appendix A – American Presidency Project Exclusions
Categories to Exclude
Campaign Documents
Debates
Executive Orders
Interviews
Memoranda
Miscellaneous Press Secretary
Miscellaneous Remarks
News Conferences
Opposition Party Responses
Party Platforms
Press Briefings
Press Office
State Dinners
Transition Documents
Attributes to Exclude
Acts Approved By The President
Announcements - Regarding Appointments
Announcements - Regarding Nominations
Announcements - Regarding Resignations
Announcements
Checklists of WH Press Releases
Conferences
Democratic Party
Designations
Determinations
Digest of Other WH Announcements
Directives
Exchange With Reporters
Fact Sheets
Joint
Memo of Understanding
Nominations / Appointments
Nominations Submitted to the Senate
Notices
Party Platforms
Press Releases
Republican Party
Responses to Questions
161
Appendix B – Dictionaries for LIWC to Code Foreign Policy and Countries
DICTIONARY TO CODE FOREIGN POLICY
(Words or phrases coded as “Remove” are not counted as foreign policy words.)
Word s
Remove
Worl d
IR_and_FP
Military
Intl_Orgs
Corporal punishment
x
Crime war
x
Drug war
x
Global climate
x
Global warming
x
Inland empire
x
War on c ri me
x
War on d ru gs
x
War on p ov er ty
x
Abroad
x
Afar
x
Global
x
Offshore
x
Oversea
x
Overseas
x
The globe
x
Worl d
x
Adversary countries
x
162
Adversary country
x
Adversary nation
x
Adversary nations
x
Adversary state
x
Adversary states
x
Allied countries
x
Allied country
x
Allied nation
x
Allied nations
x
Allied state
x
Allied states
x
Ally countries
x
Ally country
x
Ally nation
x
Ally nations
x
Ally state
x
Ally states
x
Belt and road
x
Bipolar system
x
Bipolar world
x
Cold War
x
Colonial
x
Colonialism
x
Colonialist
x
Colonialists
x
163
Colonies
x
Colonise
x
Colonised
x
Colonises
x
Colonising
x
Colonize
x
Colonized
x
Colonizes
x
Colonizing
x
Colony
x
Decolonise
x
Decolonised
x
Decolonises
x
Decolonising
x
Decolonisation
x
Decolonize
x
Decolonized
x
Decolonizes
x
Decolonizing
x
Decolonization
x
Economic sanction
x
Economic sanctions
x
Empire
x
Empires
x
Enemy countries
x
164
Enemy country
x
Enemy nation
x
Enemy nations
x
Enemy state
x
Enemy states
x
Favored nation
x
Favored nations
x
Favoured nation
x
Favoured nations
x
Foreign
x
Friendly countries
x
Friendly country
x
Friendly nation
x
Friendly nations
x
Friendly state
x
Friendly states
x
Globalise
x
Globalised
x
Globalises
x
Globalism
x
Globalist
x
Globalists
x
Globalising
x
Globalisation
x
Globalize
x
165
Globalized
x
Globalizes
x
Globalizing
x
Globalization
x
Great power
x
Great powers
x
Hegemony
x
Hegemonies
x
Imperial
x
Imperialise
x
Imperialism
x
Imperialist
x
Imperialistic
x
Imperialists
x
Imperialize
x
International
x
Internationalise
x
Internationalism
x
Internationalist
x
Internationalisation
x
Internationalists
x
Internationalize
x
Internationalization
x
Other nation
x
Other nations
x
166
Other countries
x
Other country
x
Pariah countries
x
Pariah country
x
Pariah nation
x
Pariah nations
x
Pariah state
x
Pariah states
x
Punitive sanction
x
Punitive sanctions
x
Rogue countries
x
Rogue country
x
Rogue nation
x
Rogue nations
x
Rogue state
x
Rogue states
x
Super power
x
Super powers
x
Superpower
x
Superpowers
x
Treaties
x
Treaty
x
Unipolar system
x
Unipolar world
x
Admiral
x
167
Admirals
x
Armada
x
Armadas
x
Armistice
x
Armistices
x
Attack helicopter
x
Air force
x
Air forces
x
Air strike
x
Air strikes
x
Armed combat
x
Armed fighter
x
Armed fighters
x
Armed force
x
Armed forces
x
Armies
x
Arming
x
Army
x
Artilleries
x
Artillery
x
Battle helicopter
x
Battle tank
x
Battle tanks
x
Battalion
x
Battalions
x
168
Bazooka
x
Blockade
x
Blockaded
x
Blockades
x
Blockading
x
Border defence
x
Border defense
x
Border incursion
x
Border incursions
x
Border violation
x
Border violations
x
Bomb
x
Bombed
x
Bombs
x
Bombing
x
Bomber
x
Bombers
x
Border clash
x
Border clashes
x
Border dispute
x
Border disputes
x
Border skirmish
x
Border skirmishes
x
Brigade
x
Brigades
x
169
Cannon batteries
x
Cannon battery
x
Cavalry
x
Cavalries
x
Combatant
x
Combatants
x
Corporal
x
Court martial
x
Court martials
x
Courts martial
x
Defence budget
x
Defence budgets
x
Defense budget
x
Defense budgets
x
Defence contractor
x
Defence contractors
x
Defense contractor
x
Defense contractors
x
Defence funding
x
Defense funding
x
Defence pact
x
Defense pact
x
Defensive pact
x
Defensive pacts
x
Defence industry
x
170
Defence industries
x
Defense industry
x
Defense industries
x
Deterrence
x
Democide
x
Democides
x
Drone attack
x
Drone strike
x
Embargo
x
Embargoes
x
Ethnic cleansing
x
Ethnic cleansings
x
Fighter jet
x
Fighter jets
x
Fighter plane
x
Fighter planes
x
Flotilla
x
Flotillas
x
Genocide
x
Genocides
x
Genocidaire
x
Genocidaires
x
Grenade
x
Grenades
x
Gun batteries
x
171
Gun battery
x
Home defence
x
Home defense
x
Homeland security
x
Hussar
x
Hussars
x
Infantry
x
Intifada
x
Intifadas
x
Invade
x
Invaded
x
Invades
x
Invasion
x
Invasions
x
Lieutenant general
x
Major general
x
Marines
x
Marine defence
x
Marine defense
x
Martial
x
Militia
x
Militias
x
Militaries
x
Military
x
Militarise
x
172
Militarised
x
Militarises
x
Militarising
x
Militarize
x
Militarized
x
Militarizes
x
Militarizing
x
Militaristic
x
Missile
x
Missiles
x
Monroe doctrine
x
Mutually assured destruction
x
National defence
x
National defense
x
National security
x
Naval
x
Navies
x
Navy
x
Noncivilian
x
Peace agreement
x
Peace agreements
x
Peace negotiation
x
Peace negotiations
x
Peace pact
x
Peace pacts
x
173
Platoon
x
Platoons
x
Privateer
x
Privateers
x
Rule the seas
x
Regiment
x
Regiments
x
Rocket attack
x
Rocket attacks
x
Rocket launch
x
Rocket launcher
x
Rocket launchers
x
Roosevelt Corollary
x
Sea force
x
Sea power
x
Security dilemma
x
Soldier
x
Soldiers
x
Soldiery
x
Sphere of interest
x
Spheres of interest
x
Strike force
x
Strikeforce
x
Territor ial de fence
x
Territor ial de fense
x
174
Terroris t
x
Terroris ts
x
Terroris m
x
Troops
x
Truman Doctrine
x
War
x
Wars
x
Warr ing
x
Warl ike
x
Warm ong er
x
Warm ong ers
x
Warm ong eri ng
x
Weap on
x
Weap ons
x
Axis countries
x
Axis powers
x
Axis states
x
Comecon
x
Commmonwealth of independent
x
Commonwealth of nations
x
European Economic Community
x
European union
x
Five eyes
x
League of nations
x
Nato
x
175
North Atlantic treaty
x
nonaligned movement
x
nonaligned country
x
nonaligned nation
x
nonaligned state
x
non-aligned movement
x
non-aligned country
x
non-aligned nation
x
non-aligned state
x
United nations
x
Seato
x
Southeast asia treaty
x
Wars aw p ac t
x
DICTIONARY TO CODE COUNTRIES
(Words or phrases coded as “Miscellaneous”, i.e., 01, are not counted as words indicating
countries. The formatting of how the dictionary is presented here is different from the one above,
because I originally made the two dictionaries for different versions of LIWC, and the process
that I used to make them changed across the versions.)
%
01 Miscellaneous
02 China
03 India
04 USA
176
05 Indonesia
06 Pakistan
07 Brazil
08 Nigeria
09 Bangladesh
10 Russia
11 Mexico
12 Japan
13 Ethiopia
14 Philippines
15 Egypt
16 Vietnam
17 DRCongo
18 Iran
19 Turkey
20 Germany
21 France
22 UK
23 Italy
24 SouthKorea
25 Canada
26 Spain
27 SaudiArabia
177
28 Australia
29 Taiwan
30 Poland
31 Thailand
32 Ukraine
33 Belarus
34 Afghanistan
35 Cuba
36 Iraq
37 NorthKorea
38 Venezuela
39 Israel
40 NATO
41 EU
%
Afghan dog 01
Afghan hound 01
Afghan rug 01
Chicken kiev 01
Chinese checkers 01
Cuban cigar 01
Cuban cigars 01
English breakfast 01
178
English breakfasts 01
French bean 01
French beans 01
French dressing 01
French fry 01
French fries 01
French roast 01
Italian dressing 01
Mexican standoff 01
Mexican standoffs 01
Russian dressing 01
Sevastopol tuning 01
Sebastopol tuning 01
Spanish fly 01
Spanish flu 01
China 02
Chinese 02
Chinaman 02
Chinamen 02
Peoples Republic of China 02
People’s Republic of China 02
Chinese Communist Party 02
Beijing 02
179
Shanghai 02
Chongqing 02
Tianjin 02
Guangzhou 02
Shenzhen 02
Chengdu 02
Wuhan 02
Nanjing 02
India 03
Indian 03
Indians 03
East India 03
East Indian 03
East Indians 03
Native Indian 01
Native Indians 01
Amerindian 01
Amerindians 01
Red Indian 01
Red Indians 01
American Indian 01
American Indians 01
Desi 03
180
Desis 03
Punjab 03
Punjabi 03
Punjabis 03
Sindh 03
Sindhi 03
Sindhis 03
Bengal 03
Bengali 03
Bengalis 03
Hindi 03
Gujarat 03
Gujarati 03
Gujaratis 03
Assam 03
Assamese 03
Bihar 03
Bihari 03
Biharis 03
Mumbai 03
Bombay 03
Delhi 03
Bangalore 03
181
Ahmedabad 03
Chennai 03
Kolkata 03
Calcutta 03
America 04
American 04
Americans 04
Americas 01
North America 01
North American 01
North Americans 01
South America 01
South American 01
South Americans 01
Central America 01
Central American 01
Central Americans 01
Latin America 01
Latin American 01
Latin Americans 01
United States 04
United States of America 04
USA 04
182
Yankee 04
Yankees 04
California 04
Californian 04
Californians 04
Texas 04
Texan 04
Texans 04
Florida 04
Floridan 04
Floridans 04
New York 04
New Yorker 04
New Yorkers 04
Pennsylvania 04
Pennsylvanian 04
Pennsylvanians 04
Illinois 04
Ohio 04
North Carolina 04
Michigan 04
New Jersey 04
Washington 04
183
Arizona 04
Massachusetts 04
Tennessee 04
Indiana 04
Maryland 04
Missouri 04
Wisconsin 04
Colorado 04
Minnesota 04
South Carolina 04
Alabama 04
Louisiana 04
Kentucky 04
Oregon 04
Oklahoma 04
Connecticut 04
Utah 04
Iowa 04
Nevada 04
Arkansas 04
Mississippi 04
Kansas 04
New Mexico 04
184
Nebraska 04
Idaho 04
West Virginia 04
Hawaii 04
New Hampshire 04
Maine 04
Rhode Island 04
Montana 04
Delaware 04
South Dakota 04
North Dakota 04
Alaska 04
District of Columbia 04
Vermont 04
Wyoming 04
New York City 04
Los Angeles 04
Chicago 04
Houston 04
Philadelphia 04
Dallas 04
Indonesia 05
Indonesian 05
185
Indonesians 05
Sumatra 05
Sumatran 05
Sumatrans 05
Jakarta 05
Jakartan 05
Jakartans 05
Pakistan 06
Pakistani 06
Pakistanis 06
Karachi 06
Lahore 06
Islamabad 06
Brazil 07
Brazilian 07
Brazilians 07
Sao Paulo 07
Rio de Janeiro 07
Brasilia 07
Nigeria 08
Nigerian 08
Nigerians 08
Lagos 08
186
Bangladesh 09
Bangladeshi 09
Bangladeshis 09
Dhaka 09
Dhakan 09
Dhakans 09
Russia 10
Russian 10
Russians 10
Rusky 10
Ruskie 10
Ruskies 10
Siberia 10
Siberian 10
Siberians 10
Moscow 10
Muscovite 10
Muscovites 10
Saint Petersburg 10
Petrograd 10
Leningrad 10
Novosibirsk 10
Soviet 10
187
Soviets 10
Soviet Union 10
Union of Soviet 10
USSR 10
Yekaterinburg 10
Kazan 10
Mexico 11
Mexican 11
Mexicans 11
Monterrey 11
Guadalajara 11
Puebla 11
Toluca 11
Tijuana 11
Japan 12
Japanese 12
Nippon 12
Nipponese 12
Hokkaido 12
Honshu 12
Shikoku 12
Kyushu 12
Okinawa 12
188
Okinawan 12
Tokyo 12
Yokohama 12
Osaka 12
Nagoya 12
Fukuoka 12
Kyoto 12
Ethiopia 13
Ethiopian 13
Ethiopians 13
Addis Ababa 13
Philippine 14
Philippines 14
Filipino 14
Filipina 14
Pinoy 14
Quezon 14
Manila 14
Manila envelope 01
Egypt 15
Egyptian 15
Egyptians 15
Cairo 15
189
Vietnam 16
Vietnamese 16
Viet Nam 16
Can Tho 16
Da Nang 16
Haiphong 16
Hanoi 16
Ho Chi Minh 16
Saigon 16
Democratic Republic of the Congo 17
Congo-Kinshasa 17
DR Congo 17
The Congo 17
The DRC 17
Zaire 17
Kinshasa 17
Iran 18
Iranian 18
Iranians 18
Persia 18
Persian 18
Persians 18
Tehran 18
190
Mashhad 18
Isfahan 18
Turk 19
Turks 19
Turkey 19
Turkish 19
Turkic 01
Ottoman Empire 19
Ottoman Dynasty 19
Ankara 19
Istanbul 19
German 20
Germans 20
Germany 20
Germanic 01
Berlin 20
Hamburg 20
Hamburger 01
Munich 20
Frankfurt 20
Frankfurter 01
France 21
French 21
191
Frenchman 21
Frenchmen 21
Frenchwoman 21
Frenchwomen 21
Francais 21
Francaise 21
Paris 21
Parisian 21
Marseille 21
Toulouse 21
United Kingdom 22
The UK 22
Britain 22
Great Britain 22
British 22
England 22
Englander 22
English 22
Englishman 22
Englishmen 22
Englishwoman 22
Englishwomen 22
Wales 22
192
Welsh 22
Welshman 22
Welshmen 22
Welshwoman 22
Welshwomen 22
Scotland 22
Scottish 22
Scottishman 22
Scottishmen 22
Scottishwoman 22
Scottishwomen 22
Scots 22
Scotsman 22
Scotsmen 22
Scotswoman 22
Scotswomen 22
Scotchman 22
Scotchmen 22
Scotchwoman 22
Scotchwomen 22
Cornwall 22
Cornish 22
Cornishman 22
193
Cornishmen 22
Cornishwoman 22
Cornishwomen 22
Northern Ireland 22
Ulster 22
Italy 23
Italian 23
Italians 23
Rome 23
Roman 23
Romans 23
Milano 23
Milanese 23
Naples 23
South Korea 24
South Korean 24
South Koreans 24
Republic of Korea 24
Seoul 24
Busan 24
Incheon 24
Canada 25
Canadian 25
194
Canadians 25
Canuck 25
Canucks 25
Ottawa 25
Toronto 25
Montreal 25
Vancouver 25
Calgary 25
Quebec 25
Spain 26
Spanish 26
Mexican Spanish 11
American Spanish 01
Madrid 26
Barcelona 26
Valencia 26
Seville 26
Bilbao 26
Saudi 27
Saudi Arabia 27
Saudi Arabian 27
Saudi Arabians 27
House of Saud 27
195
Al Saud 27
Riyadh 27
Jeddah 27
Medina 27
Madinah 27
Australia 28
Australian 28
Australians 28
Aussie 28
Melbourne 28
Canberra 28
Brisbane 28
Taiwan 29
Taiwanese 29
Taipei 29
Keelung 29
Republic of China 29
Poland 30
Polish 30
Polishman 30
Polishmen 30
Polishwoman 30
Polishwomen 30
196
Polack 30
Warsaw 30
Krakow 30
Thai 31
Thailand 31
Bangkok 31
Ukraine 32
Ukrainian 32
Ukrainians 32
Kiev 32
Kyiv 32
Kharkiv 32
Odesa 32
Odessa 32
Dnipro 32
Donetsk 32
Sevastapol 32
Sebastopol 32
Luhansk 32
Lugansk 32
Donbas 32
Donbass 32
Crimea 32
197
Crimean 32
Belarus 33
Belarusian 33
Belarusians 33
Byelorussia 33
Byelorussian 33
Byelorussians 33
Belorussia 33
Belorussian 33
Belorussians 33
Minsk 33
Afghan 34
Afghani 34
Afghans 34
Afghanis 34
Afghanistan 34
Afghanistanis 34
Dari 34
Kabul 34
Kandahar 34
Cuba 35
Cuban 35
Cubans 35
198
Havana 35
Havanan 35
Havanans 35
Havanians 35
Iraq 36
Iraqi 36
Iraqis 36
Baghdad 36
Baghdadi 36
Baghdadis 36
Basrah 36
Erbil 36
Fallujah 36
al-Fallujah 36
Kirkuk 36
Mosul 36
Najaf 36
Communist Korea 37
DPRK 37
Democratic Peoples Republic of Korea 37
Democratic People's Republic of Korea 37
North Korea 37
North Korean 37
199
North Koreans 37
Hamgyong 37
Pyongyang 37
Venezuela 38
Venezuelan 38
Venezuelans 38
Barquismeto 38
Caracas 38
Maracaibo 38
Maracay 38
Israel 39
Israeli 39
Israelis 39
Haifa 39
Jerusalem 39
Jerusalemite 39
Negev 39
Tel Aviv 39
Zionist 39
Zionists 39
NATO 40
North Atlantic Treaty Organization 40
North Atlantic Treaty Organisation 40
200
The EU 41
European Parliament 41
European Union 41
The Euro 41
Eurodollar 41
Eurozone 41
