Militarization: A Witch’s War Brew?
How military power affects authoritarian regimes’ conflict behavior
AUTHOR: LINA SIGURDH FALL TERM 2020
Bachelor Thesis Peace and Conflict Studies C
Supervisor: Sophia Hatz Course examiner: Chiara Ruffa
DEPARTMENT OF PEACE AND CONFLICT UPPSALA UNIVERSITY
2
Abstract
The effect of regime type on conflict onset is a well-studied phenomenon, and various studies have found that variance in regime type, and within regime types, affects conflict onset. For instance, militarization in autocracies seems to be linked with increased risk of initiating conflict. However, even in the studies that disaggregate types of autocracies, the categorizations are relatively shallow.
This thesis aims to create a definition of military dictatorships which captures their complexity more fully, to determine whether militarization truly does increase the risk of conflict onset. Military dictatorships are here defined as a state that achieves and maintains power through threat or actual use of force, is outwardly or effectively controlled by military officers, and places high value on maintaining a powerful armed force to protect constitutional and territorial integrity. The method used is a logistic regression, where the independent variable is military dictatorship, and the dependent variable is directed dyads. The results show that when a state is a military dictatorship, the log odds of it initiating conflict is 0.73; military dictatorships are indeed more likely to initiate conflict than autocracies in general.
3 CONTENTS
ABSTRACT ... 2
1. INTRODUCTION ... 5
2. THEORETICAL FRAMEWORK ... 6
2.1.CENTRALCONCEPTSANDPREVIOUSRESEARCH ... 6
2.1.1. Previous literature ... 6
Research Question ... 8
2.1.2. The shape of authoritarianism ... 9
2.1.3. Military dictatorships... 10
2.2.THEORY ... 12
Hypothesis ... 12
Figure 1: Proposed theory flow chart ... 15
3. RESEARCH DESIGN ... 16
3.1.METHODOLOGICAL APPROACH ... 16
3.1.1. Null hypothesis ... 16
H0 ... 17
3.2.VARIABLES OF INTEREST ... 18
3.2.1. Independent variable ... 18
3.2.2. Dependent variable ... 20
3.2.3. Control variables ... 21
4. RESULTS AND IMPLICATIONS ... 21
Table 1: Summary Statistics ... 23
Table 2: Model 2 converted to odds-ratios ... 24
5. FINAL REMARKS ... 25
5.1.RESEARCH DESIGN: POTENTIAL WEAKNESSES, VALIDITY, AND RELIABILITY ... 25
5.2.IN CONCLUSION ... 27
6. REFERENCES ... 28
4 This page is left intentionally blank
5
1. Introduction
What increases and decreases risk of conflict is a hot topic within Peace and Conflict research and researchers have spent much time and effort investigating and explaining potential triggers of conflict. A multitude of factors have been found, and among them one of the most tantalizing puzzles lie: regime type. The democratic peace thesis states that democracies rarely, if ever, go to war against each other, and research into other regime types, such as different varieties of autocracies, has further proven that variation in regime type affects conflict onset. One such variation is militarization: higher levels of military power in autocracies seem to be linked with increased risk of conflict onset (Choi and James 2008; Peceny, Beer, and Sanchez-Terry 2002;
Reiter and Stam 2003; Sechser 2004; Teorell and Lindberg 2019). Importantly, though, previous research has yet to thoroughly investigate this observation—although surface level categorizations have been made, most studies looking into the effect of regime type on conflict onset stick to a dichotomous statement of democracies and non-democracies. Those who delve deeper into non-democracies often fall into the same, shallow categorizations of regime:
monarchies, military, or party-type regimes. None of the types are fully disaggregated, even though there are clear differences in belligerency among them. I argue that to predict conflict onset more accurately, a more sophisticated typology must be developed.
This thesis aims to create a deeper understanding of conflict onset from the perspective of militarization. I believe that increased military power in a country also increases conflict risk and endeavour to investigate whether this is true by distinguishing military dictatorships from other autocracies. By creating a typology where military dictatorships are conceptually and operationally separated from autocracies, I will determine whether my claim is true. Determining what regime characteristics increase or decrease conflict risk would help our understanding of conflict onset, and deeper comprehension might also lead us to be able to predict conflict onset accurately; in this thesis, I describe whether a connection between military power and belligerency is found, and what might be the cause of such a connection. The paper is structured as follows: first, I present previous research on the topic, then I map out my conceptualization of autocracies and, most importantly, military dictatorships. Then I present my research design, how I operationalized and coded my variables, and finally, I present and reflect over the results of the analysis.
6
2. Theoretical Framework
2.1.CENTRALCONCEPTSANDPREVIOUSRESEARCH
First, I must establish that this study operates under the liberal paradigm: I work under the assumption that the domestic politics of a state affect how they operate in the international system. This is in contrast to the realist paradigm, where it is assumed that domestic politics matter little, if at all, to how the international system operates (Petersen 2008).
2.1.1. Previous literature
It is perhaps the democratic peace thesis that has sparked the most interest in research on regime type and its effects on conflict onset. As mentioned, the democratic peace states that, first, democracies rarely, if ever, go to war against each other, and second, they are not less likely to participate in interstate conflict with states of a differing regime type (Peceny, Beer, and Sanchez-Terry 2002; Russett and Oneal 2001). In fact, there is evidence that mixed dyads—
one democracy and one non-democracy—are the most conflict prone types of dyads (Quackenbush and Rudy 2009). Thus, a large part of the field of research on conflict onset focuses on explaining the democratic peace and, by extension, explaining what factors—if any—encourages and discourages conflict in the international arena. Several intriguing theories for how and why regime type affects conflict onset exist, but many only succeed in explaining part of the puzzle: democratic peace research is often plagued by homogeneity. Although all explanations of the democratic peace attribute the phenomenon to variation in domestic policies of states, most fail to fully disaggregate regime type in a meaningful way; broadly, it is assumed that all autocracies and all democracies act the same (Petersen 2008). This is an issue because the impact of domestic policy on states’ behavior need not only be important when explaining the democratic peace, it might also explain why some states engage in interstate conflict while others do not in general.
Petersen (2008) addresses this issue and turns to comparative politics to delve more deeply into how regime type might differ and, if it does, if it matters. She finds that autocracies vary greatly in how they carry out both domestic and foreign policy, and that it is important to watch for these differences when predicting and explaining behavior. Peceny, Beer, and Sanchez-Terry (2002) pose a similar question to Petersen to investigate whether the democratic peace is simply an effect of homogenous governments, and thus if there is a similar
‘dictatorial peace’. Like Petersen, they turn toward comparative politics to examine the conflict
7 behavior of three types of authoritarian regimes: personalist dictatorships, single-party regimes, and military regimes. Although they do not find significant evidence for a dictatorial peace, they do find that different types of authoritarian regimes exhibit varying patterns of conflict behavior and conclude that using a more sophisticated concept of authoritarianism is crucial to determine its impact on foreign policy (Peceny, Beer, and Sanchez-Terry 2002).
There are several explanations for why regime type affects the behavior of states, especially with regards to conflict onset. One subset of theories attribute the democratic peace to international trade and economic involvement, while another subset attribute it to the norms of ‘contingent consent’ and ‘bounded competition’ fostered in democracies. (Dixon 1994;
Souva and Prins 2006). A third subset of theories attribute the variance in conflict behavior to institutional constraints, i.e. in democracies, mobilization is constrained by domestic institutions in a way it is not in autocracies. For instance, if politicians in a democratic state initiates a conflict they then lose, it is likely the loss is detrimental to their career; who trusts a politician who incites a war they cannot win? On the other hand, in autocracies, politicians might have more leeway when it comes to losing a war, even if they initiated it (Peceny, Beer, and Sanchez-Terry 2002; Petersen 2008). If it is true that certain institutional constraints can impede or expedite mobilization in democracies and non-democracies, it might also explain the variance in behavior among autocracies.
In their study Identifying the Culprit, Reiter and Stam (2003) analyze the data provided by Peceny, Beer, and Sanchez-Terry (2002) but with one important addition:
accounting for direction of initiation. They find that different types of autocracies vary not only in likelihood of participating in conflict, they also vary in likelihood of initiating conflict (Reiter and Stam 2003, 336). The authors attribute their results to the institutional constraints explanation of the democratic peace, suggesting that not only does certain institutional factors explain variation in participation, but they also explain variation in conflict initiation.
Choi and James (2008) investigate whether increased military power in a state also increases likelihood of initiating conflict. They contend that the higher domestic influence of the military, the higher the likelihood of conflict initiation (Choi and James 2008, 39). If this is true, it is implied that there are inherent attributes to military regimes that increase their belligerency. The authors find evidence that non-militaristic regimes are less likely to initiate conflict; however, due to the nature of the study, it remains inconclusive if it is explicitly
8 militarization that increases likelihood of initiating conflict, and not simply the fact that higher military power means a less democratic system (i.e. they did not exclude the possibility of multicollinearity).
Sechser (2004) also endeavors to test if, and how, militarization affects likelihood of initiating conflict. Like Choi and James, he finds that military men favor the use of force in foreign policy, and unlike Choi and James, Sechser also accounts for the issue of multicollinearity; he finds that civilian control does not correlate with democracy, and it is therefore reasonable to assume that it is, in fact, militarization (and not authoritarianism in general) that leads to more belligerent behavior. Sechser concludes that on average, when military officers have the authority to initiate militarized conflicts, they do so at a much higher rate than civilians (Sechser 2004, 770). However, both Sechser and Choi and James’s study use the same source data to create their military regime variable, and I argue that a more sophisticated definition of military dictatorships is necessary to develop a theory surrounding the effect of militarization on conflict onset. Furthermore, in contrast to the results presented by the authors, is the fact that it stands to reason that fighting interstate wars might weaken a military autocracy’s control of internal politics because the army is ‘busy elsewhere’ (Ezrow and Frantz 2011). Thus, it can be argued that military regimes are unlikely to initiate conflict because they stand to lose their mode of maintaining internal power (the military) if they go to war. Militarism might not actually be the driving factor behind the observed variation in conflict behavior in autocracies.
In sum, regime type seems to affect conflict behavior and likelihood of initiating conflict, especially autocracies are at risk. However, what characteristics of autocracies that affect conflict onset remains inconclusive, as previous research has failed to fully separate types of autocracies from each other. It is in this context we arrive at the question at the heart of this study:
Research Question
Why are some authoritarian states more likely than others to initiate conflict?
9 2.1.2. The shape of authoritarianism
In their book, Dictators and Dictatorships: Understanding Authoritarian Regimes and Their Leaders, Ezrow and Frantz (2011) begin by providing a short and sweet definition of dictatorships: “democracies are regimes in which ‘those who govern are selected through contested elections; dictatorships are ‘not democracies’” (p. 2). This parsimonious definition provides a good starting point for the conceptualization of authoritarianism: although autocracies1 can take on many forms, as will be presented below, there is one thing they can never be: democratic. Some autocracies might adopt democratic political processes and institutions (elections, for example), but they are always inherently different from democracies.
In a democracy, it is the citizens who both choose and dismiss head of state, either through direct or indirect election. Contrarily, in autocracies, it is another body of power who does—be it hereditary tradition, ‘royal councils’, the military, or a ruling party (Teorell and Lindberg 2019). Furthermore, democracies are defined by institutions such as universal suffrage, freedom of expression, and independent media. Not only must there be free and fair elections, but citizens must also be able to form and express political opinions without coercion or retribution from the state. Autocracies, then, do not provide their citizens with the means to, or allow them to, form and express their own political opinions.
Ezrow and Frantz (2011) present two contemporary typologies of authoritarianism: continuous and categorical. The continuous typology focuses on level of authoritarianism and put different types of regimes on a scale of democratization. This is useful for defining ‘grey zone’ regimes that possess both democratic and autocratic characteristics.
However, a linear scale of regime type implies that the less autocratic a regime becomes, the more democratic it is. Democratization is a complex phenomenon and does not necessarily depend on how authoritarian a regime is (Ezrow and Frantz 2011). Besides, defining authoritarian states in the way they differ from democracy diminishes differences within autocratic states: they all are not the same, and differ in other ways than ‘level’ of authoritarianism.
The categorical typology is more appropriate for defining authoritarian states separate from democracy, as it ignores levels of authoritarianism in favor of creating multidimensional categories where autocracies are disaggregated through the meaningful ways
1 I use the terms autocracy, authoritarianism, and dictatorships interchangeably
10 in which they differ (Ezrow and Frantz 2011). Hadenius and Teorell (2007) present five types of authoritarian regimes: monarchial, military, no-party, one-party, and multi-party regimes.
They are categorized through how they achieve and maintain political power; monarchies do so through hereditary succession, military through threat of or use of force, and the three ‘party’- categories are electoral autocracies who maintain power through popular (non-democratic) elections. There might also be some hybrid regimes—for instance, a military government might hold popular elections—but generally, these five categories hold true (Hadenius and Teorell 2007). The categories, and the way they are determined, point out an important aspect of authoritarianism: autocracies possess inherent characteristics which separate them from one another, and in turn, these characteristics determine how and by whom internal and foreign policy is handled and carried out (Geddes 1999).
For this study, I create two categories of authoritarianism: military dictatorships, and dictatorships in general. Notably, all categories that Hadenius and Teorell present, except for military regimes, are ‘dictatorships in general’. This is not to say that military dictatorships are the only types of regimes that carry characteristics which affect how they behave in the international forum, in fact, I argue that all autocracies possess unique characteristics which separate them from one another, but because the phenomenon of interest in this study is the effect of militarization on authoritarian regimes, the typology is necessary. Thus, authoritarian regimes are regimes that achieve and maintain power through non-democratic means, for instance hereditary tradition, the military, or through a ruling party. In the following section I will discuss how military dictatorships differ from autocracies.
2.1.3. Military dictatorships
A military dictatorship mainly operates on actual or threatened use of force to gain and maintain internal power (Hadenius and Teorell 2007). However, a regime monopolizing violence does not a military dictatorship make; arguably, upholding a monopoly of violence within the country is a condition of sovereign statehood. Therefore, for an autocracy to be defined as a militarized, military personnel themselves must carry the most political power.
Both Choi and James (2008) and Sechser (2004) separate military power in a country in three categories, and Sechser labels these strong civilian, weak civilian, and military government. When a state is controlled by civilian(s), it is a strong civilian state. A weak civilian regime is outwardly controlled by civilians but effectively so by the military: for example, Japan
11 in the period leading up to and during World War II, where the Emperor and choice civilian members of cabinet were retained, but the country was effectively controlled by the military (Choi and James 2008; Sechser 2004). Lastly, a state is categorized as a military government when it is explicitly controlled by a group of military officers (Sechser 2004). While I agree that there might be a grey zone where the military has a stronghold on the government but not complete control of policy, or there is an officer as head of state but the military is not necessarily in control, I argue militarization affects both persons in the government and the way the government operates as a unit. Even small amounts of military control (in the form of an officer head of state, for instance) affect foreign policy (Geddes 1999). Underscoring this, Sechser (2004) found no substantive behavioral difference between weak civilian and military government states. Therefore, in this thesis, no distinction between the types ‘weak civilian’
and ‘military government’ is made.
Furthermore, military dictatorships are characterized by the idea of themselves as
“guardians of the constitution” and a means of bringing stability and restoring order to a country (Ezrow and Frantz 2011, 7). They value territorial integrity of the nation highly and believe that they cannot adequately pursue goals of territorial integrity and internal order without a unified and capable military (Geddes 1999). In fact, it has been found that many institutionalized military regimes adhere to a “national security” ideology, which boils down to the belief that the military is ideally suited to be the guarantor of the nation. In turn, this belief legitimates aggressive foreign policy (Peceny, Beer, and Sanchez-Terry 2002, 19). Militaries in charge might take on the role of an arbitrator or a ruler. The regimes who act as arbitrators do not intend to stay in power but are ready to ‘go back to the barracks’ after disputes are settled.
Military regimes who take on the role as ruler, on the other hand, have no desire to return to the barracks and no confidence in the existing order of society. The goal for a military ruler is to stay in power for as long as possible, and to maximize power for themselves (Ezrow and Frantz 2011).
In sum, a military dictatorship is defined in three ways: in their mode of maintaining power, their persons in government, and their policy. A military dictatorship achieves and maintains power through threat or actual use of force, is outwardly or effectively controlled by military officers, and places high value on maintaining a powerful armed force to protect constitutional and territorial integrity.
12
2.2.THEORY
The independent variable in this study is militarization and the dependent variable conflict initiation. In the following part, I will attempt to formulate a theory of how the variables are causally connected. Importantly, though, the goal of this thesis is not to test the causal mechanism I present and hence, this part mainly serves as an explanation as to why I expect the outcome I do.
Hypothesis
On a societal level, military regimes tend to have weak roots in society, which means they find it difficult to control or withstand popular protest. They also strongly desire maintaining internal cohesion and hierarchy, which means they favor avoiding, negotiating, or diverting internal conflict whenever possible (Hadenius and Teorell 2007). Fighting interstate wars are effective for uniting a country, as it unifies the population in an ‘us versus them’ mindset, where ‘them’
are the population or military of another country entirely (Malesevic 2011). Thus, the military might have incentives, on a societal level, to incite conflict to maintain internal cohesion.
Furthermore, there might be some personal biases that affect behavior of officers in the regime and thus how military dictatorships act in certain situations.
First, persons who act the same way over a long period of time develop distinctive and persistent habits of thought, which in turn creates a unique relation to the world, which provides a differing outlook and avenues for rationalizing behavior and role. The ‘military mind’ is one such distinctive and persistent habit of thought, meaning that military professionals carry biases with them when they interact with the world (Huntington 1981). For instance, survey evidence from military professionals and civilians in the U.S. show that professional military officers and civilians with a veteran background share views on foreign policy which differ quite succinctly from the views of their civilian, non-vet counterparts (Feaver and Gelpi 2005, 52). I argue that these biases are the cause of military bellicosity in autocracies, and can be grouped into three categories: parochial biases, perceptual biases, and decision-making biases (Sechser 2004).
Military dictatorships are more likely to initiate violent conflict than autocracies in general.
13 Parochial biases suggest that if left unchecked, military organizational interests raise the risk of war in three ways: firstly, officers want direct battlefield experience, as it brings glory and excitement not only to individuals, but also to the military as a whole: no one cares about the military until they have won a war (Sechser 2004). Secondly, offensive operations require large arms expenditures and bolster the prestige field of commanders. The field of military strategy relies heavily on secrecy: understandably, it is of high military interest to allow as few ‘outsiders’ into the organizational and operational capabilities of the military as possible (Walt 1987). Therefore, even in democracies, those with expertise about the military’s organizational and operational capabilities are part of the professional military complex and are thus the people who might reap the fruits of higher military expenditure and a bolstered prestige field. Third, from a purely strategical point of view, offensive operations are the most advantageous; the element of surprise allows the attacker to set the stage, skewing the fight in their favor. As Sun Tzu himself said in The Art of War: “attack is the secret of defense, defense is the planning of an attack” (Sechser 2004; Tzu and Machiavelli 2013). In military doctrines this kind of strategic thinking is apparent: for instance, in the 2017 U.S. doctrine, it is stated that “major operations and campaigns feature a balance among offensive, defensive, and stability through six phases: shape, deter, seize initiative, dominate, stabilize, and enable civil authority.” (Joint Chiefs of Staff 2017, I–16; my emphasis).
Perceptual bias means that officers see the world through a lens which impacts the choices they make. For one, national security is viewed as a strictly military problem. The importance of economic and diplomatic efforts are undervalued, while officers simultaneously exaggerate external threats and opportunities to destroy them (Sechser 2004; Walt 1987). This
‘national security’ mindset creates a steppingstone for officers to overstate their own capabilities and generate optimistic casualty estimates and time frames, even more so when they make the case for striking first. After all, Napoleon eventually failed not because he was a bad military strategist, but because he was a bad political strategist. He failed to realize that his aggressive acts would unite the rest of Europe against him (Walt 1987). Furthermore, officers might overstate the probability of war. Military officers are realists in the sense that they believe war to be inevitable (Huntington 1981; Sechser 2004). Sooner or later, there will be conflict:
otherwise, in peacetimes, a military man’s line of work would be obsolete. This manner of framing reality cause officers to put too much weight into preventative war efforts. If the enemy will attack tomorrow—why not attack today and reap the rewards of having struck first?
14 Lastly, decision-making biases suggest that officers are ‘specialists on violence’
and will revert to what they know when pressured. Officers are trained to make quick, drastic decisions with immediate effect—at the battlefield, a quick decision on whether to shoot or not will have real consequences (Sechser 2004). Furthermore, officers see violence as a legitimate and effective policy instrument—they are accustomed to employ military solutions and will draw on them even when they are poorly suited to the problem at hand. Diplomatic or economic options are more arcane and because officers are used to quick decision-making with immediate effect and because they see violence as a legitimate policy instrument, they prefer violent solutions (ibid.).
To summarize, I argue that the biases presented above are inherent to the professional military. This is not to say that all individual soldiers are belligerent and violent, because no one is affected solely by one kind of bias (be it military, religious, cultural). I do argue, however, that because these biases are inherent to the professional military, military professionals, when released from civil constraints, will act in a more belligerent way than civilians released from the same types of constraints. Military officers in an autocratic regime are a risk to conflict onset, as they are conditioned to view the world in a certain way and because they are released from the ‘red tape’ that surrounds democratic institutions. Civilians in autocratic regimes, however, do not pose a similar risk, as they do not carry the biases of professional military men, despite also being released from institutional constraints.
15 Figure 1: Proposed theory flow chart Parochial biases:
If left unchecked, military organizational interests raise the risk of
war
Combat brings glory and excitement;
officers want direct battlefield experience
Offensive operations require arms expenditures and allow military to fight on its own, preplanned terms
Clear strategic advantages in
‘striking first’
Perceptual biases:
Officers see their world through a lens which
civilians do not use
National security is a strictly military problem; officers undervalue economic and diplomatic impact
Views on combat:
overstate advantages of striking first and generate optimistic cost
estimates
Overstate the probability of war and
thus put too much weight into preventative war
Decision-making biases Officers are ‘specialists
on violence’
Officers are trained to make
quick, drastic decisions with immediate effect
Officers see violence as a legitimate and effective policy
instrument;
civilian leaders have a more Officers are accustomed to
employing military solutions
Increased risk of conflict initiation
Higher military influence relative to civilian influence
H: Military dictatorships are more likely to initiate violent conflict than autocracies in general.
Absence of mitigating institutional constraints
16
3. Research Design
3.1.METHODOLOGICAL APPROACH
Due to the complicated nature of interstate conflict, and the many potential sources of conflict, a quantitative approach to the issue at hand is the natural alternative. In order to make assumptions about the causal connection between the variables, covariance must first be established (Kellstedt and Whitten 2018). A quantitative study is advantageous in controlling for observable alternative explanations and generalizing the results. The trade-off, however, is nuance and causal mechanism. The variables I use in the regression are dummy variables and can only take on the value of 0 and 1 and thus logistic regression is used to determine the probability of militarization affecting conflict onset.
This thesis is a replication of the study by Reiter and Stam (2003). I use their data and add my own independent variable, ‘military dictatorship’, to analyze. A replication study’s main goal is to build on existing research by improving on their data or methodology (King 1995). This is not to say that Reiter and Stam fails to do what they set out to do, I simply want to dig deeper into a subject their analysis only touches upon, which is the effect of militarization.
I use the same dependent variable they do, directed dyads. Notably, the Reiter and Stam study is a replication of Peceny, Beer, and Sanchez-Terry's (2002) study, and Peceny et al. coded most of the data. However, their dependent variable is different, and the data and code provided by Reiter and Stam is the source material for the analysis presented here, so for simplicity the Reiter and Stam study is treated as the main source of data. The unit of measure for the study is directed dyad-years and the data are limited to the post-World War II period, 1949–942.
3.1.1. Null hypothesis
There is a possibility that there is no significant difference between how likely military dictatorships and autocracies are to interstate conflict. While already existing studies have shown that militarist regimes might be more conflict prone, they also have shown a variance in all types of authoritarian regimes, and thus militarism might not be a driving factor, although it looks like it at first glance. For instance, it might be true that military governments do not want to weaken their military by fighting abroad as it is their main instrument for internal power, and therefore are not more likely than others to instigate interstate conflict. Furthermore, in security
2 The Reiter and Stam study covered the period between 1945 and 1994, but because other data I need are limited to 1949, I chose to exclude the first four years of Reiter and Stam’s study.
17 studies, the most accepted theory is ‘military conservativism’: due to various factors, such as civilian naïveté and ignorance when faced with military issues, it is argued that military officers are less likely to advocate the use of force than their civilian counterparts (Walt 1987). For instance, Feaver and Gelpi (2005) surveyed American military personnel, civilian veterans, and civilian non-veterans and found that civilians “tend to have a more interventionist foreign policy agenda, and that they are willing to use the military to achieve those foreign policy goals” (51).
In civilian regimes in the U.S., Latin America and Easter Europe, the military do not act more hawkish than civilians (Choi and James 2008). It also stands to reason that because it is military men who risk their life in conflict, they would be more conservative in recommending use of military force (Huntington 1981).
Although the theory of military conservativism is supported by many within the field of security studies, I argue that it is not applicable to autocratic subjects. The majority of empirical work which evidences military conservativism has been done in democratic states, primarily in the U.S. (Sechser 2004). I do not deny that U.S. military officers have acted dovish in modern times; I do, however, question that their behavior has any implication on how military officers might act in military dictatorships, where they are released from the constraints of democratic institutions. I also assume that even though the regime consists of, or are appointed by, military officers, it is not the regime who will be on the front lines of war and risk losing their life.
H0
There is no significant difference in the likelihood of initiating conflict between military dictatorships and autocracies.
18
3.2.VARIABLES OF INTEREST
3.2.1. Independent variable
The subject of interest in this study is authoritarian states. To create an Independent Variable (IV) for testing, two things must be done. First, who is (and who is not) an authoritarian state must be determined. That will be the population. Then, to determine if, and how, militarization might affect conflict onset, the concept of military dictatorships must be operationalized.
In 2.1.2., authoritarian states were conceptualized as ‘regimes which achieve and maintain power through non-democratic means, for instance hereditary tradition, the military, or through a ruling party’. Thus, autocratic regimes’ chief executive is not determined through elections, they do not have free and fair elections or universal suffrage, there is no or limited access to political and civil society organizations, and freedom of expression is limited. These are the criteria which qualifies or disqualifies states. To measure this, I use the ‘additive polyarchy index’ (API) from the Varieties of Democracy (V-Dem) project’s data (Coppedge et al. 2020; Teorell 2019). It is derived from a series of observations on freedom of speech, clean elections, freedom of expression, elected executive, and suffrage and, as such, determines the extent to which the ‘electoral principle of democracy’ is achieved. The variable is an interval which varies from low to high, 0–1, and is aggregated through a formula where the elected executive and suffrage weigh half as much as the other factors (ibid.).
I have already discussed the issues of a sliding scale between democracy and autocracy, in that it implies that the less autocratic a government is, the more democratic it is. Although this is true, this measurement is appropriate for determining my population as it is a statement on certain aspects of governance, where presence or absence of an aspect is typical for a regime type. The goal of this study is not to make a claim that regime type is solely decided on these aspects of governance, nor as to where, exactly, the line between autocracy and democracy is drawn, nor claim which states are unequivocally autocratic/democratic. I do, however, want to For a state to be classified as ‘autocratic’ they must score lower than 0,35 on the API from V-Dem.
F i g : A s i m p
19 make sure that the states in my population do not determine policy through democratic means and to err on the side of caution and exclude autocracies rather than include democracies, the chosen cut-off value is slightly lower than the mean of the sample, ~0.4.
The crucial part of designing the IV is determining what qualifies as a military dictatorship. In 2.1.3., I defined military dictatorships as autocracies that gain and maintain power through threat or use of force, are effectively controlled by military officers, and place high value on maintaining a powerful armed force. From this definition, three criteria that define military dictatorships are derived.
To determine the military dictatorship variable, I use data from two sources: the ‘military dimension index’ (MDI) Teorell and Lindberg (2019) developed for the V-Dem dataset, and the SIPRI military expenditure data3 (SIPRI 2019). The MDI is determined on whether the chief executive of a state was a) appointed through a coup, rebellion, or by the military and b) can be dismissed by the military. The variable is an interval variable, from low to high (0–1). In the V-Dem data, both a) and b) are coded as present (1) or not (0) and then averaged4 (Coppedge et al. 2020; Teorell and Lindberg 2019). I consider a state to score on the MDI if it receives a value of 0.5 or higher in V-Dem5. Second, the SIPRI military expenditure data shows military
3 The SIPRI data was treated before use: I cleaned it up by removing unnecessary elements (such as comments) and converted percentage to decimals.
4 In the V-Dem dataset, they distinguish between ‘head of state’ and ‘head of government’. They might be the same person, but when they are not, the chief executive is determined through who has most executive power.
Neither, however, has sole power, and this is reflected in the interval nature of the MDI variable.
5 I also do robustness checks where I increase or decrease the cutoff for MDI by 0.2 (Model 3 and 4) to see whether the result changes significantly.
An autocracy is a ‘military dictatorship’ if:
1. The government prioritizes military expenditure
2. The regime gained power through use or threat of military force 3. The regime is determined by military officers
The variable military dictatorship is coded 1 if one state in the dyad fulfills all three of the criteria listed. It is coded 0 otherwise.
20 expenditure as share of GDP, which I will use to determine whether military expenditure (milexp) is prioritized (1) or not (0). To summarize the milexp data quickly, the median of military expenditure as share of GDP is 2.66%, while the mean is higher, at 4.03%; most countries spend less than 3% of the GDP on the military, but some countries spend much more than others and therefore draw up the mean. In peacetimes, it is rare for countries to spend more than 5% of their GDP on the military (Roser and Nagdy 2013; SIPRI 2019). I consider a state to have spent above average on the military if their military expenditure share of GDP exceeds 5%. If the MDI is 0.5 or higher, it fulfills criteria 2 and 3. If milexp exceeds 0.05, it fulfills criteria 1. Thus, to score on the military dictatorship variable, a state must receive 1 on the MDI variable and 1 on the milexp variable.
3.2.2. Dependent variable
The Dependent Variable (DV) is directed dyads, and I test if there is a difference in conflict initiation in dyads where at least one state is a military dictatorship, as opposed to when it is not. ‘Conflicts’ are here defined as Militarized Interstate Disputes (MIDs), “united historical cases of conflicts in which the threat, display, or use of military force […] by one member state is explicitly directed towards the government, official representatives, official forces, property, or territory of another state” (Peceny, Beer, and Sanchez-Terry 2002, 20). A directed dyad means it shows who the initiator and who the target of the MID is. Each dyad appears twice in the data set, once as the potential initiator (state A) and once as the potential target (state B);
when the potential initiator is the initiator, the variable is coded as 1, and when it is instead the target, the variable is coded as 0 (Reiter and Stam 2003). Note that in my analysis, the independent variable only appears once, as the ‘military dictatorship’ variable. In the Reiter and Stam study, each dyad appeared twice (as described above). This does not affect my results: I am only interested in one direction, which is military dictatorships initiating conflict. The variance in the DV is initiator and target of an MID. Accounting for direction is, naturally, crucial for my analysis, as the theory and hypothesis are that military dictatorships are more likely to initiate conflict. In Table 1, the dependent variable (directed dyads) is labeled ‘sideaa’.
21 3.2.3. Control variables
The field of conflict onset is large, and there are several factors that have been shown to affect when, why, and how states initiate conflict with one another. To be sure that I can draw conclusions about how militarization affects conflict onset, other, observable factors that also affect conflict onset must be accounted for.
First, there is a strong correlation between contiguity and war onset—countries that lie close to each other are more likely to be in conflict with each other (Bremer 1992).
Therefore, contiguity is included as a control variable. Second, ‘major power’ is included because there is empirical evidence that major powers are more conflict-prone than minor powers. This is often attributed to bargaining theory—large powers make large demands because they have a favorable bargaining position (Chiba, Machain, and Reed 2014). The U.S., Soviet Union/Russia, U.K., France, and China from 1950 are considered major powers (Peceny, Beer, and Sanchez-Terry 2002). Third, states formally allied with one another are less likely to engage in MIDs, especially in the post-World War II world (Russett and Oneal 2001). Ally is therefore a control variable. Fourth, ‘power ratio’ refers to the evidence that the greater power disparity between states, the less likely conflict is. Fifth, economic development is commonly attributed to predict conflict and thus included as a control. Lastly, Reiter and Stam (2003) account for potential autocorrelation bias by including ‘time since last dispute’ and cubic spline variables; logit analysis assumes independence of variables, and cubic spline is a method to ensure that assumption is not violated (Beck, Katz, and Tucker 1998; Peceny, Beer, and Sanchez-Terry 2002).
4. Results and implications
In the analysis I did four logistic regression models: first, in Model 1, I replicated the results of Model 4 by Reiter and Stam. Model 2 is the logistic regression with the added variable military dictatorship. Model 3 and 4 are robustness checks, where I increased respectively decreased the cut-off for the MDI with 0.2. The results of Model 1 reflect the results of Model 4 in Reiter and Stam’s study, which means the regression worked and I can move on to the other models. In Model 2, 3, and 4, the military dictatorship variable remains positive and significant: in Model 2, a one-unit increase in the IV yields a 0.73 change in the dependent variable, i.e., when a state is a military dictatorship, the log odds of it initiating conflict is 0.73. Table 2 is Model 2 converted to odds-ratios for easier interpreting of the results. As is apparent in the table, the
22 odds of a military dictatorship initiating conflict is 2.07 times greater than the odds of another type of autocracy initiating conflict. Thus, military dictatorships and conflict initiation covary in a way that aligns with the hypothesis that military dictatorships are more belligerent than autocracies in general.
Earlier studies have already established that militarization and conflict onset covary (Choi and James 2008; Peceny, Beer, and Sanchez-Terry 2002; Reiter and Stam 2003;
Sechser 2004). However, they have failed to establish clear boundaries of what militarization is, and thus failed to create a full picture of what the indicators of increased belligerency are.
The results of this study, then, have narrowed this gap and exposed three predictors of increased belligerency: if a regime has gained power through use or threat of force, is prioritizing military expenditure, and is appointed by military officer(s), it is more likely to initiate conflict. These three factors, condensed into the military dictatorship variable,
provides at least part of an answer to the research question brought up at the beginning of this paper: why some autocracies initiate conflict, while others do not. Although this study, due to the nature of quantitative research, cannot draw any conclusions on the causal mechanism between conflict initiation and military dictatorships, the results do not disprove the
theoretical framework presented in 2.2.. Another interesting observation in the results is how the ‘mil/democ’ (military regimes targeting democracies) variable changed from Model 1 to Model 2; when plugging in the military dictatorship variable, military regimes (as defined by Reiter and Stam) seem to be less likely to target democracies in MIDs. Note also that the control variable ‘economically advanced’, becomes insignificant when the military
dictatorship variable is plugged in. Adopting a broader perspective, the results of the analysis proves an important point about conflict onset and regime type overall: distilling states into broad categories of regime does a disservice to our comprehension of the phenomenon at hand—in the face of complex issues, it is fruitful to attempt to disaggregate categories, because it might provide a more detailed view of how the variables covary. The clearest evidence of this is that the military dictatorship variable is significant and positive, meaning we might be able to draw closer conclusions about what predicts conflict in militaristic states.
Furthermore, that the ‘economically advanced’ variable turned insignificant shows that even though the effect of economical advancement on conflict onset is well-observed, there is still other unseen factors at play, affecting its variance.
23 Table 1: Summary Statistics
Table 1: Summary statistics Dependent variable:
Directed Dyads
(1) (2) (3) (4)
Military dictatorship 0.727*** 0.736*** 0.749***
(0.137) (0.110) (0.167) Pers/democ 1.045*** 1.128*** 1.053*** 1.100***
(0.114) (0.126) (0.135) (0.121)
Democ/pers 0.105 -0.090 -0.067 -0.117
(0.170) (0.511) (0.512) (0.511) Democ/single -0.081 0.446** 0.473** 0.409**
(0.100) (0.187) (0.188) (0.186) Single/democ 0.179* -0.099 -0.047 -0.014 (0.093) (0.123) (0.126) (0.115)
Democ/mil -0.455 -1.194 -1.186 -1.238
(0.294) (1.007) (1.007) (1.006)
Mil/democ 0.642*** 0.291 0.180 0.297
(0.177) (0.223) (0.245) (0.218)
Personal 0.296 0.264 0.294 0.327
(0.220) (0.246) (0.247) (0.230)
Military -0.307 -0.478 -0.422 -0.487
(0.587) (0.590) (0.591) (0.590) Single -0.651*** -0.634*** -0.581*** -0.688***
(0.122) (0.143) (0.146) (0.139)
Democ -1.047*** 0.439** 0.441** 0.397*
(0.131) (0.212) (0.212) (0.211) Contiguous 2.914*** 3.070*** 3.055*** 2.963***
(0.061) (0.084) (0.087) (0.081) Major power 2.164*** 2.023*** 1.976*** 2.163***
(0.070) (0.104) (0.106) (0.099)
Ally 0.085 0.058 0.047 0.102
(0.068) (0.088) (0.090) (0.086) Power ratio -0.316*** -0.308*** -0.293*** -0.332***
(0.021) (0.030) (0.030) (0.029) Economically advanced -0.187** -0.080 -0.080 -0.115 (0.094) (0.142) (0.143) (0.136) Time since last dispute -0.381*** -0.419*** -0.416*** -0.416***
(0.019) (0.026) (0.027) (0.025) Cubic spline 1 -0.004*** -0.004*** -0.004*** -0.004***
(0.0004) (0.001) (0.001) (0.0005) Cubic spline 2 0.002*** 0.003*** 0.003*** 0.003***
(0.0003) (0.0004) (0.0004) (0.0004) Cubic spline 3 -0.001*** -0.001*** -0.001*** -0.001***
(0.0001) (0.0002) (0.0002) (0.0002) Constant -4.805*** -4.730*** -4.754*** -4.620***
(0.076) (0.107) (0.111) (0.102) Observations 753,456 366,457 347,536 377,702 Log Likelihood -7,950.922 -4,431.316 -4,230.364 -4,708.042 Akaike Inf. Crit. 15,941.840 8,904.632 8,502.727 9,458.083 Note: *p<0.1; **p<0.05; ***p<0.01
24 Table 2: Model 2 converted to odds-ratios
Table 2: Odds ratios of Model 2
OR 2.5 % 97.5 % Intercept 0.009 0.007 0.011 Military dictatorship 2.068 1.569 2.684 Pers/democ 3.091 2.402 3.931 Democ/pers 0.914 0.279 2.187 Democ/single 1.563 1.065 2.221 Single/democ 0.906 0.709 1.149 Democ/mil 0.303 0.017 1.364 Mil/democ 1.338 0.840 2.020 Personal 1.302 0.778 2.052 Military 0.620 0.152 1.663 Single 0.531 0.399 0.698
Democ 1.552 0.999 2.302
Contiguous 21.535 18.276 25.395 Major power 7.560 6.161 9.263
Ally 1.060 0.891 1.258
Power ratio 0.735 0.692 0.778
Economically advanced 0.924 0.695 1.214
Time since last dispute 0.658 0.625 0.692
Cubic spline 1 0.996 0.995 0.997
Cubic spline 2 1.003 1.002 1.004
Cubic spline 3 0.999 0.999 0.999
25
5. Final remarks
5.1.RESEARCH DESIGN: POTENTIAL WEAKNESSES, VALIDITY, AND RELIABILITY
Whenever complex phenomena, such as those that occur in peace and conflict research, is to be examined, some trade-offs are required. I have already mentioned that a quantitative study trades nuance for generalizability. Furthermore, what is a problem—and always is in quantitative studies—is the concept of validity: to what extent does the variable measure the concept I want to examine? In general, quantitative studies rely on proxies to measure unobservable indicators. For instance, I use policy decisions to make an argument on military bias, but I do not (cannot) measure the actual biases. Arguably, however, biases are only interesting once they translate into actual behavior: a bias that does not manifest outside a soldier’s head has no consequences for anybody. Thus, using policy choices as a proxy for bias is defensible (Sechser 2004).
Moreover, my IV design has some drawbacks to address. First, military expenditure might not be an indicator of militarization as I define it, but rather a response to a perceived threat. If a state believes it is threatened by another, or that the international system is less safe than before, it might prioritize arms expenditure. For instance, in Sweden, the government has in recent years begun prioritizing the military higher (through more funds and the reinstation of conscription), but that does not necessarily mean they, as a state, are becoming militarized. Additionally, states do not need to be autocratic to prioritize military expenditure and doing so does not make them ‘more’ autocratic. SIPRI estimates the United States to be the top military spender in the world (SIPRI 2019). This does of course not mean they are a military dictatorship: it is not only military dictatorships that value and prioritize the military. The military expenditure criterion does not stand on its own. It is balanced by other criteria, and I do maintain that it provides an important dimension to militarization; prioritizing the military is intricately connected to the ‘national security ideology’ where the military is viewed as the sole guarantor of the nation. Expenditure has previously been disregarded when defining military dictatorships, yet the results of this thesis shows that it, when combined with other factors, raises likelihood of conflict initiation.
Moreover, the empirical reality of military dictatorships cannot be captured by the three militarization criteria presented here. For instance, it is likely that personalist autocracies with a military officer in charge vary in policy behavior from autocracies with a military ‘party’
26 in charge. Such a military junta dictatorship might act more restrained than personalist military dictatorships, as the power is spread out over several persons, or it might be the other way around, that military juntas are more belligerent than personalist military regimes. These, and other possible dimensions, are not captured in this thesis’s definition of military dictatorships, because it was not within the scope of the analysis.
Reliability of a study means to what extent repeated tests of the variables of interest yield the same result. Generally, quantitative studies score highly on reliability as quantifiable data is easy to use and replicate when needed. To corroborate my work and ensure replicability of my claims, source code is included in the attachments. Model 1 also shows that when testing the method and data from Reiter and Stam (2003), the test yielded the same results as the original. However, because I use data I have not myself collected and compiled for my analysis, I cannot for sure know how the data has been collected and who coded it. Data are always at risk of random and systematic errors; random errors refer to errors that might occur because a coder made a typo, or because two states were mixed up, or because two numbers switched place. Due to the vast amount of information within the data, every piece cannot be corroborated as accurate; in Model 2, the total number of observations amount to 336,457. A few random errors among that number of observations should not affect the results of the analysis. Systematic errors are errors that do not occur only once, or randomly, but skew data because they occur systematically throughout. These errors are difficult to circumvent if they are present but are less likely to occur. To mitigate the risk of systematic error in my analysis, I have used reputable sources to gather data. In all, I feel confident the research design captures what I set out to capture, with no errors that skew the results in a significant manner.
27
5.2.IN CONCLUSION
In the realm of conflict onset research, regime type is of great interest. In security studies, the effect of militarization is. Here, the two subjects are united to create deeper understanding of what increases the risk of conflict. A deeper understanding of what factors drive conflict might, in turn, aid in predicting, explaining, and perhaps even preventing, conflict. Although the two fields have previously been combined, the concept of military dictatorships is underdeveloped, even in studies where militarism is the variable of interest. Notably, the studies most similar to this one, by Sechser and by Choi and James, used the same source data to define military dictatorships (Choi and James 2008; Sechser 2004). It was perhaps not too surprising, then, that they received similar results. In this thesis, I deliberately chose to create a military dictatorship variable from different source data than the other authors and chose to expand the definition of military dictatorships. This is the main strength, and contribution, of this thesis. The purpose was to answer the question why some autocracies are more likely than others to initiate conflict, and the results of the analysis show that militarization significantly affects autocracies’
likelihood to initiate conflict. This finding is in line with earlier findings on militarism, such as those presented by Choi and James (2008), Reiter and Stam (2003), Sechser (2004), and Teorell and Lindberg (2019), and aligns with the hypothesis presented.
The results provide some exciting avenues for future research. First, the results have shown there is value to further disaggregating regime types. Monarchies, personalist, and party regimes could benefit from a more sophisticated conceptualization—what, exactly, are they, how do they differ from each other, and how do they act differently in comparison to each other? This might even provide avenues to deter or encourage certain behaviour, or at the very least more accurately predict conflict onset. Moreover, further dissection of the concept of military dictatorships would be exciting, for instance in order to determine whether personalist military regimes and military juntas behave differently, although both are here categorized as military dictatorships. Doing the same analysis as in this thesis, but on data from a longer time span, or from the 21st century, would be another interesting research path to take. One last avenue for future research is to do a qualitative study to determine what the causal mechanism between militarization and belligerency is. Determining causality is, after all, another piece of the puzzle: now we know that militarization does increase risk of conflict initiation, but we have yet to figure out why.
28
6. References
Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker. 1998. “Taking Time Seriously:
Time-Series-Cross-Section Analysis with a Binary Dependent Variable.” American Journal of Political Science 42(4): 1260–88.
Bremer, Stuart A. 1992. “Dangerous Dyads: Conditions Affecting the Likelihood of Interstate War, 1816-1965.” Journal of Conflict Resolution 36(2): 309–41.
Chiba, Daina, Carla Martinez Machain, and William Reed. 2014. “Major Powers and Militarized Conflict.” Journal of Conflict Resolution 58(6): 976–1002.
Choi, Seung-Whan, and Patrick James. 2008. “Civil—Military Structure, Political
Communication, and the Democratic Peace.” Journal of Peace Research 45(1): 37–53.
Coppedge, Michael et al. 2020. “V-Dem Codebook V10.”
Dixon, William J. 1994. “Democracy and the Peaceful Settlement of International Conflict.”
American Political Science Review 88(1): 14–32.
Ezrow, Natasha M., and Erica Frantz. 2011. Dictators and Dictatorships: Understanding Authoritarian Regimes and Their Leaders. Bloomsbury Publishing USA.
Feaver, Peter D., and Christopher Gelpi. 2005. Choosing Your Battles: American Civil- Military Relations and the Use of Force. Princeton, UNITED STATES: Princeton University Press. http://ebookcentral.proquest.com/lib/uu/detail.action?docID=793225 (November 27, 2020).
Geddes, Barbara. 1999. “WHAT DO WE KNOW ABOUT DEMOCRATIZATION AFTER TWENTY YEARS?” Annual Review of Political Science 2(1): 115–44.
Hadenius, Axel, and Jan. Teorell. 2007. “Pathways from Authoritarianism.” Journal of Democracy 18(1): 143–57.
Huntington, Samuel P. 1981. The Soldier and the State: The Theory and Politics of Civil–
Military Relations. Harvard University Press.
Joint Chiefs of Staff. 2017. “Joint Publication 1, Doctrine for the Armed Forces of the United States.” : 176.
Kellstedt, Paul, and Guy Whitten. 2018. The Fundamentals of Political Science Research. 3rd ed. New York: Cambridge University Press.
King, Gary. 1995. “Replication, Replication.” PS: Political Science and Politics 28(3): 444–
52.
Malesevic, Sinisa. 2011. “Nationalism, War and Social Cohesion.” Ethnic and Racial Studies 34.
29 Peceny, Mark, Caroline C. Beer, and Shannon Sanchez-Terry. 2002. “Dictatorial Peace?”
American Political Science Review 96(1): 15–26.
Petersen, Karen K. 2008. “There Is More to the Story than ‘Us-Versus-Them’: Expanding the Study of Interstate Conflict and Regime Type Beyond a Dichotomy.” Peace
Economics, Peace Science and Public Policy 14(1).
https://www.degruyter.com/view/j/peps.2008.14.issue-
1/peps.2008.14.1.1103/peps.2008.14.1.1103.xml (November 12, 2020).
Quackenbush, Stephen L., and Michael Rudy. 2009. “Evaluating the Monadic Democratic Peace.” Conflict Management and Peace Science 26(3): 268–85.
Reiter, Dan, and Allan C. Stam. 2003. “Identifying the Culprit: Democracy, Dictatorship, and Dispute Initiation.” American Political Science Review 97(02).
http://www.journals.cambridge.org/abstract_S0003055403000704 (December 30, 2020).
Roser, Max, and Mohamed Nagdy. 2013. “Military Spending.” Our World in Data.
https://ourworldindata.org/military-spending (December 14, 2020).
Russett, Bruce, and John Oneal. 2001. Triangulating Peace: Democracy, Interdependence, and International Organizations. New York: Norton.
Sechser, Todd S. 2004a. “Are Soldiers Less War-Prone than Statesmen?” Journal of Conflict Resolution 48(5): 746–74.
SIPRI. 2019. “Data for All Countries from 1949-2019.”
https://www.sipri.org/databases/milex (December 2, 2020).
Souva, Mark, and Brandon Prins. 2006. “The Liberal Peace Revisited: The Role of Democracy, Dependence, and Development in Militarized Interstate Dispute Initiation, 1950–1999.” International Interactions 32(2): 183–200.
Teorell, Jan. 2019. “Additive Polyarchy Index (D) (V2x_api).”
Teorell, Jan, and Staffan I. Lindberg. 2019. “Beyond Democracy-Dictatorship Measures: A New Framework Capturing Executive Bases of Power, 1789–2016.” Perspectives on Politics 17(1): 66–84.
Tzu, Sun, and Niccolo Machiavelli. 2013. The Complete Art of War. Simon and Schuster.
Walt, Stephen M. 1987. “The Search for a Science of Strategy: A Review Essay.”
International Security 12(1): 140.
