One-Sided Violence and External Support: The effects of government one-sided violence on external state support provision

(1)

One-Sided Violence and External Support

The effects of government one-sided violence on external state support provision

Tova Flisberg

Bachelor’s Thesis

Department of Peace and Conflict Studies Uppsala University

(2)

characterised by external support, also referred to as external intervention; the United States (U.S.)-led coalition interventions in Afghanistan and Iraq in the early 2000s, followed by the wars in Syria, Yemen and Ukraine in the 2010s. Common characteristics for the latter three conflicts, are external intervention and high levels of civilian fatalities. In Syria, both pro- and anti-regime forces, as well as the Syrian regime itself, are to a high degree reliant on external sponsors, the conflict constituting a battle ground for regional geopolitical rivals as major powers deepen their involvement (Laub, 2019). Similarly, the conflict in Yemen is

characterised by interventions by regional powers, including Iran and the Gulf states led by Saudi Arabia. Meanwhile, the U.S. have conducted continuous counterterrorism operations in the country (CFR, 2020). Lastly, the Russian annexation of Crimea in Ukraine is by many international actors considered a grave violation of international principles of non-

intervention and sovereignty (CFR, 2014).

At the same time, in Syria, government forces as well as rebel groups have regularly been found to target civilians. Notably, the use of chemical weapons and aerial bombardments as collective-punishment tactics by the government have resulted in massive civilian casualties (Laub, 2019). In Yemen, currently the world’s worst humanitarian crisis, an estimated 11,700 civilian fatalities in connection with direct civilian targeting since 2015 were reported by the Armed Conflict Location & Event Data Project (ACLED) in June of 2019 (Almosawa et al., 2017; Sulz, 2019). Lastly, in Ukraine, according to the U.N. the civilian death toll since the onset of the conflict exceeded 3,300 in May of 2019, attributable to both the Government and rebel groups (OHCHR, 2019: 6f). Given these observations, and considering their policy implications, the research question is formulated as follows; how does violence against civilians affect external support?

Previous research on external support has primarily been focused on causes and effects of support provision, with a main concern for rebel group support (see for instance Salehyan et al., 2011; Balch-Lindsay and Enterline, 2000; Cunningham, 2010; Lacina, 2006; Gleditsch et al., 2008; Gent, 2008). The general consensus of the field is that external intervention and support under most conditions has a detrimental effect on conflict dynamics, making civil

(4)

Balch-Lindsay and Enterline, 2000: 637f; Karlén, 2017a). As noted by Karlén (2019: 733), limited knowledge exists on why and when external support is withdrawn. In a first effort to fill this gap in the literature, Karlén (2019) carried out the first large-N analysis of causes for rebel support termination, finding conventional explanations for support provision to fall short of explaining the withdrawal of said support. While Karlén’s findings open up for new

avenues of research, an increased understanding of state support termination is called for by the empirical observation that most external support is in fact channelled to governments (Högbladh et al., 2011: 13). With the purpose of examining the relationship between one- sided violence against civilians and the termination of external support, this thesis aims to complement previous research by exploring one possible cause for state support termination.

Relying on theories of cost-benefit analyses and reputation costs (see for instance Smith, 1996; Zartman, 2001; Cunningham, 2010; Brutger and Kertzer, 2018), I argue that one-sided violence against civilians committed by governments involved in intrastate conflicts cause external state actors to withdraw support provided to those governments. The decision to provide external support is assumed to be based on a cost-benefit analysis, in which strategic benefits are weighed against costs of support. As the government receiving support commits one-sided violence, the utility of continued support provision is expected to decrease, as reputation costs are incurred on the supporting state for providing support to a government violating international laws and norms against civilian targeting. The initial cost-benefit analysis is thus re-evaluated, and the cost of continued support in terms of reputation costs is expected to outweigh its strategic benefits, causing a termination of support.

The findings suggest a significant positive relationship between government one-sided

violence and the probability of external support withdrawal, according to which an increase in civilian fatalities increases the probability of the termination of external support. Furthermore, the analysis indicate that the effects are larger at higher levels of one-sided violence, in line with the theoretical argument presented in the theoretical framework.

1.1 Disposition

The thesis is structured as follows; the following section, 2. Theoretical framework, consists of an overview of the previous research relevant to the phenomena of interest, followed by a discussion on the central concepts. The theoretical argument is then proposed, as is the

(5)

hypothesis derived from the theory. Section 3. Research design introduces the method of analysis, as well as the data. The operationalizations of the independent, dependent, and control variables are subsequently presented, including discussion of reliability and validity when appropriate, followed by a discussion on data collection reliability. Then follows section 4. Results and analysis, in which the results of the logistic regressions are presented and substantially interpreted. Lastly, section 5. Summary and conclusion provides a summary of the thesis and its findings, as well as the conclusions and avenues for future research.

2. Theoretical Framework

2.1 Previous research

Previous literature on the subject of external support in intrastate conflict has been primarily focused on two main strands of research; implications of third-party intervention for conflict dynamics, and causes of external support, primarily to rebel groups. For clarity, it should be noted that the terms intervention and external support are used here interchangeably, due to the lack of conceptual clarity in the previous literature1. A further discussion on the problems attached to this conceptual unclarity is found below.

The research on the effects of third-party intervention on conflict dynamics explores a multitude of aspects affected by intervention, including but not exclusive to; the duration of conflict, the severity of conflict, the likelihood of peaceful settlements, and the risk of recurrence of conflict (see for instance Balch-Lindsay and Enterline, 2000; Cunningham, 2010; Lacina, 2006; Karlén, 2019; Karlén, 2017a). Many aspects of the literature are disputed, including the effects of third-party intervention on conflict duration. For instance, while Collier et al. (2004: 268) find rebel-biased military interventions to decrease civil war duration, and government-biased interventions to not have a significant effect on duration, other scholars such as Balch-Lindsay and Enterline (2000: 637f) find external intervention to increase the likelihood of the emergence of a stalemate, resulting in a significant prolongation of the civil war.

(6)

However, a majority of the scholars studying third-party intervention and external support appears to be in agreement on the detrimental effects of external intervention on conflict dynamics. In short, external intervention is found to increase the risk of interstate conflict (Gleditsch et al., 2008: 499f), and the duration of civil wars (Cunningham, 2010: 125; Balch- Lindsay and Enterline, 2000: 637f; Regan, 2002). External intervention/support increases the military severity of conflict in terms of battle related deaths (Lacina, 2006: 286), and decrease the likelihood of a negotiated settlement (Karlén, 2019: 733). Furthermore, rebel sponsorship has been found to increase violence against civilians, decrease the likelihood of rebel groups entering into agreements, as well as their willingness to embrace democracy once in power post-conflict (ibid.).

Considering the main explanatory factor explored in this thesis, one-sided violence, it is worth looking further into the previous literature on the relationship between foreign involvement and civilian targeting. The shift in balance of capabilities caused by external resources has been argued to influence the strategies of violence adopted by the belligerents (Wood et al., 2012: 653; Lockyer, 2010; Salehyan et al., 2014: 636f; Karlén, 2019: 736). While as

mentioned, rebel-biased support has been found to increase violence against civilians, it is less clear what type of bias affects one-sided violence in what ways. Salehyan et al. (2014: 639, 656) find that external support to rebel groups increase the level of one-sided violence perpetrated by the group, arguing that the increase is caused by a diminishing necessity for fostering close ties with the civilian population. Furthermore, they find that support from democratic states greatly decreases this effect as opposed to support from states less concerned with human rights, and that an increasing number of sponsors providing support decreases the so-called democratic effect (ibid: 656). Wood et al. (2012: 653) on the other hand, argue that a biased intervention in the form of armed support decreases the number of intentional killings of civilians by the actor receiving support, while the number of killings perpetrated by the adversary increases. They find this to be true regardless of whether the support is provided to a rebel group or the government (ibid: 655f).

Turning to the literature at large, there appears to be a lack of consensus on the effects of external support on government perpetrated one-sided violence. Nonetheless, it is possible, and perhaps probable considering the findings cited above, that the provision of external support in itself causes increased one-sided violence. The dependent variable, external

support, would then affect the independent variable, one-sided violence, resulting in a case of

(7)

possible reversed causality (Kellstedt and Whitten, 2013: 83). In an attempt to tackle the risk of reversed causality by addressing the issue of time-order, a methodological choice to add a constant of +1 to the year of the one-sided violence variable has been made, as will be further discussed in section 3.1.1.

One of the reasons for diverging results is the definition of intervention and external support, by some defined simply as military intervention, by others as military and economic

interventions, and yet others including concepts such as diplomacy, etc. For instance, the findings presented by Regan and Aydin (2006: 754) run counter to previous research due to the inclusion of diplomacy as a form of intervention. One of the limitations of the literature at large, however, is a focus on military intervention. Considering the diverge types of external support observed in the empirical world, an exclusive focus on direct military support results in a lack of nuance. By including non-military support, such as funding, intelligence

provision, training, weapon provision, etc, all important aspects of conflict duration, a greater understanding for the nature of external support and thus possible causes for termination can arguably be achieved.

Furthermore, diverging results also stem from the study of military interventions and external support with different biases, i.e. government-biased, neutral, and rebel-biased intervnetion.

External resources affect the relative power between the parties in an intrastate conflict, with state support to rebel groups reducing the asymmetric power relation between the non-state and the state actor (Karlén, 2019: 736), and support to governments commonly having the opposite effect. The consequences on conflict dynamics can thus be expected to differ

depending on the recipient of support, necessitating a well-grounded understanding of support of all biases, which leads to the second strand of research.

The second strand of research on external support in intrastate conflict, that characterized by a focus on support to rebel groups, has been concerned primarily with the causes of support (Karlén, 2019: 733). Different explanations pertaining to the motivations behind support provision have been found and debated in the literature, such as for instance transnational ethnic kinship ties, relative military power, and interstate rivalries (Saleyhan et al., 2011: 734;

Karlén, 2019: 734). While of great academic interest to the understanding of external support provision and intervention, Karlén (2019: 734) finds that factors that are robust predictors of

(8)

cluster of predictors included in Karlén’s study, only ethnic kinship ties and the Cold War are found to consistently affect both support provision and support termination (ibid.). However, these findings provide little insight into the mechanisms behind government-biased support termination. As stated by Karlén (ibid: 735), support to governments differ from support to rebel groups in terms of the decision-making process as well as the associated potential costs, thus necessitating a focus on different factors in the study of support termination depending on the bias of the support. This conclusion makes the scholarly focus on rebel-biased support provision problematic, considering the empirical observation that most external support in the world is channelled to governments rather than rebel groups (Högbladh et al., 2011: 13).

In summary, while the causes and effects of external support in intrastate conflict are relatively well-established, the scholarly field on external support is lacking a fundamental understanding of the causes for support termination (Karlén, 2019: 733). Karlén (2019) makes a significant contribution to the field by conducting the first large-N study of rebel-biased support termination and concluding that most predictors of support fail to explain the

termination of said support. However, the causal mechanisms underlying support termination are far from established, and a study on the causes of government-biased support termination has to the best of my knowledge yet to be conducted. As mentioned, this constitutes a

considerable gap in the literature, considering the empirical observation that more external support is channelled to governments than rebel groups (Högbladh et al., 2011: 13).

Consequently, the purpose of this thesis is to explore one possible cause of the termination of government-biased external support; namely one-sided violence perpetrated by the

government receiving support. By moving away from the scholarly focus on military support and direct military intervention and including more indirect types of support in the analysis, I hope to avoid the limitations of a narrower approach to external support.

2.2 Central concepts

2.2.1 External support

The theoretical definition of external support is derived from Karlén (2017b: 500) and is herein defined as “a unilateral intervention by a third-party state in an internal armed conflict in favor of either the government or the opposition movement involved in that conflict”

(Karlén, 2017b: 500). In this thesis only support in favour of the government of interest. This follows from the logic presented above, that different factors are of interest when analysing

(9)

rebel-biased as opposed to government-biased support (Karlén, 2019: 735), in combination with the empirical observation cited above; that more support is provided to governments than to rebel-groups (Högbladh et al., 2011: 13).

The external support may take the form of direct participation of troops, or more indirect forms of support, including funding, training, military intelligence, weapons, sanctuaries, etc.

The terms used in previous literature for external support provision includes, but is not limited to, ‘biased interventions’, ‘secondary support’, and ‘external/third-party intervention’ (Karlén, 2017b: 500). As mentioned, what appears to be a discrepancy in results in previous research may be due to a lack of clarity and consensus on the definition of the terms intervention and support. In this thesis, the term external intervention is used interchangeably with the term external support. When the term military intervention is used, it refers to the direct

participation of troops.

Furthermore, external support is a theoretical and empirical given in the analysis that is to follow. The dependent variable of this thesis is the change in external support, more specifically Withdrawal of external support. The withdrawal of support is defined as a complete termination of all types of external support. For the purpose of facilitating data collection and retaining high validity in the operationalization of the dependent variable, the definition is highly minimalistic at the cost of nuance. If time constraints where no issue, it would be preferable to study the overall variation in the provision of support following from one-sided violence; including aspects such as a change in the type of support provided, a relative decrease in support, a turn to more covert provision of support, etc, rather than simply the complete termination of support. For instance, in theory, it is possible that different types of support are more or less likely to be withdrawn than others in the face of increasing one- sided violence. One such example could be that states would be less willing to openly contribute their own state troops that risk being associated with government perpetrated violence against civilians, and instead would turn to more indirect forms of support that are less overt, such as funding or intelligence. The logic of this argument will be clarified in the following sections. Regardless, for reasons already mentioned, a minimalist approach is taken in this thesis, only studying the complete termination of support.

(10)

2.2.2 One-sided violence

One-sided violence is defined by the Uppsala Conflict Data Programme (UCDP) as “the deliberate use of armed force by the government of a state or by a formally organised group against civilians which results in at least 25 deaths in a year” (UCDP, n.d.). Similar to the previous concept, a narrow definition facilitates data collection and increases the reliability of the resulting operationalisation. In this case, the definition additionally allows for reliable data collection through the utilisation of the UCDP One-sided Violence Dataset, which relies on the same definition. However, several points are worth consideration. First, the threshold at 25 deaths in a year excludes any lower levels of fatalities. Arguably, this does not affect the conceptual validity considerably, as it may be argued that lower levels of violence are unlikely to spark international reactions regardless, or in many cases not considered

significant enough to gain the attention of the international community. Hence, it is presumed that lower levels of fatalities are unlikely to affect the decision-making process of

international actors. Related to this issue is not only the well-established media selection bias, which prioritizes novelty and proximity (Höglund and Öberg, 2011: 56), but also the general notion that civilian deaths taking place in the Global North sparks more reactions

internationally than civilian deaths in the Global South. Hence, it is possible that lower levels of civilian killings perpetrated by governments would in fact spark international reactions, were they to take place in the Global North. However, this is something that falls without the spatial scope of this thesis yet constitutes a possible avenue for future research.

Second, the definition is narrow in its exclusive focus on deadly violence. Other types of violence, such as threats of violence, abductions, property damage, conflicted related sexual violence, and psychological violence, may also be of theoretical relevance to the relationship between one-sided violence and external support withdrawal. Different types of violence could in theory affect external support provision differently. For instance, while consistently under-reported, government perpetrated conflict related sexual violence in particular could have major international consequences in terms of reputation costs, partly due to the very taboo nature of the crime (UN Secretary-General, 2017: 11, 61). However, for data collection purposes, again, the narrow definition is necessary. This due to under-reporting of many types of violence, and the intangible nature of for example psychological violence. Thus, a focus on fatalities is relatively well-suited for a large-N study, leaving little room for subjective

interpretations and facilitating the use of UCDP’s extensive data collection.

(11)

2.3 Theoretical argument

2.3.1 External support provision

The theoretical framework draws on theories of cost-benefit analyses, which are based on the assumption that states are rational actors basing their foreign policy decisions on cost-benefit analyses (see for instance Smith, 1996; Zartman, 2001; Cunningham, 2010). Essentially, a party to an armed conflict is expected to pick the strategic alternative it prefers, and change that decision only when the pain associated with the present course is increased (Zartman, 2001: 8). The increased pain is then weighed against the expected utility of three options;

ceasing the fighting, entering into a negotiated agreement, or continuing the armed struggle. A rational actor is presumed to pursue the option that returns the highest expected utility

(Cunningham, 2010: 116).

Based on the assumption of states being rational actors, the causal mechanism builds on the second assumption that a majority of states choose to provide external support to a

government involved in intrastate conflict for strategic reasons. Though officially stated causes for support may vary considerably, the basic assumption is that the underlying, primary motivation is a strategic one, may it be economic, military or political. According to Gent (2008: 731), the primary goal of external interveners in intrastate conflict is presumed to be the policy outcomes, a notion supported by studies on intervention in intrastate as well as interstate conflict (Gartzke and Gleditsch, 2004; Smith, 1996; Werner, 2000). Arguing that policies generally are dictated by the winners of conflict in intra- as well as interstate conflicts, Gent (2008: 731) states that external actors regardless of the type of conflict are primarily concerned with the outcome of the conflict. More specifically, Cunningham (2010) points to three possible goals of military intervention in civil conflict; to help one of the belligerents reach military victory, to contribute to a resolution through a negotiated settlement, or to “pursue independent objectives in the war outside of the goals of the domestic combatants” (Cunningham, 2010: 116). One such empirical example, is the South African regime providing military support to neighbouring states and insurgents during the Cold War. They did so with the purpose of keeping anti-apartheid governments from being established in the region, rather than due to any ideological affinity with the supported groups (ibid.). Arguably, this supports the notion of an underlying strategic motivation of intervening actors.

(12)

2.3.2 Initial cost-benefit analysis

The initial strategic motives are then weighed against the costs of intervening, directly or indirectly. Cunningham (2010: 118) provides a clear overview of the costs involved in fighting for external actors, arguing that the total cost of fighting is lower for external than domestic actors for several reasons. First, the fighting takes place outside the territory of the intervening state. Second, civil wars primarily incur human costs, in the form of loss of life or health, and economic costs, including the destruction of infrastructure, the loss of livelihood, and the disruption of local trade. However, these costs are borne by the host state, i.e. the state in which the fighting takes place, and primarily so by the civilian population. The external state, on the other hand, mainly bears the human costs limited to the intervening military personnel (ibid.). Furthermore, when support is limited to non-military support, such as funding, weapons, intelligence, etc, the human costs for the intervening state are kept at a minimum. Third, in cases of direct as well more indirect support, economic costs are lower for the external state. While civil wars may have regional effects, the destruction of

infrastructure and disruption of domestic trade is primarily confined to the country in which fighting takes place (ibid.). Lastly, the ability of the intervening state to exploit local resources in the state where fighting takes place serves as a possible offset to some of the intervention costs in cases of direct support (ibid.).

Another possible cost of fighting is the risk of international backlash (Cunningham, 2010:

118f). If the international community responds to the intervention with sanctions or similar economic disruption in order to pressure the supporting state into withdrawal, costs of fighting are effectively increased, affecting the expected utility of war (ibid.). However, as argued by Karlén (2019: 735), whereas external support to rebel groups violates the principle of non-intervention, state support to governments is often justified under provisions of security assistance, and thus welcomed by the government receiving support. Consequently, although the reactions from the international community may vary depending on the

circumstances under which support is provided, these costs are not expected to outweigh the strategic benefits from intervention at the initial stages of support provision.

Essentially, the argument thus far holds that external support is provided based on a rational cost-benefit analysis by the intervening state, according to which strategic motives for support outweigh costs pertaining to for instance economic and military resources. Motives may for instance include geopolitical concerns, such as the example of South Africa. Considering the

(13)

low costs borne by the external state providing support, as discussed above, it is of interest what change to the conflict dynamics may alter the initial cost-benefit analysis to such a degree that support is withdrawn. As one of the primary potential costs of intervention according to the preceding theoretical argument is reputation costs inflicted by the international community, one possible trigger for such cost is herein explored; one-sided violence.

2.3.3 The causal mechanism

The first step of the causal mechanism is the independent variable; one-sided violence perpetrated by the government receiving external support. According to the Stockholm International Peace Research Institute (SIPRI) Yearbook of 2009, one-sided violence significantly increased from the early 1990s and on. The violence is perpetrated by all types of armed actors, including government forces, and found to be more fatal when perpetrated by state troops than by armed groups (Stepanova, 2009). Nevertheless, one-sided violence

constitutes a grave breach of international humanitarian law, as established in Article 51 of Additional Protocol I and Article 13 of Additional Protocol II to the Geneva Conventions (ibid: 40f). Although violations of the laws protecting civilians are far from uncommon (ICRC, 2010), according to the International Committee of the Red Cross (ICRC), state practice establishes the prohibition of violence against civilians as a norm of customary international law, applicable in interstate as well as intrastate armed conflict (ICRC, n.d.).

Rules of customary law bind all states, and all parties to a conflict, with no requirement of formal adherence (Henckaerts and Doswald-Beck, 2005: xvi). Due not only to the

conventional prohibition against one-sided violence, but also the strong customary nature of the prohibition, it is herein argued that the association with and support of a government perpetrating one-sided violence, in breach of customary and conventional law, is expected to incur increasing costs.

Following from the onset of one-sided violence, the second step of the causal mechanism are the increased costs of continued support. One-sided violence committed by the government receiving support is expected to necessitate a re-evaluation of the initial cost-benefit analysis by the supporting state due to increased reputation costs. The theoretical argument of this thesis builds on the notion that reputations essentially are beliefs about a trait of an actor, informed by past behaviour (Brutger and Kertzer, 2018: 696). When policy decisions have a

(14)

negative effect on other states’ beliefs about a characteristic valued by an actor, reputation costs are incurred (ibid.). Arguably, one such possible characteristic is an aversion to one- sided violence against civilians, grounded in international norms against civilian targeting in armed conflict, manifested in the customary status of the prohibition against one-sided violence (ICRC, n.d.).

Increasing reputation costs are presumed to drive a re-evaluation of the cost-benefit analysis based on previous literature on the effects of domestic audience costs on foreign policy decisions. The domestic component of reputation costs is a matter of second-order belief;

“what domestic constituencies think others think about the country or leader’s characteristics”

(Brutger and Kertzer, 2018: 696). Previous research holds that the mass public is indeed concerned with state reputation, to such a degree that reputation costs have notable political consequences (Brutger and Kertzer, 2018: 695; Fearon, 1995; Saunders, 2015). The concern of domestic constituencies for a state’s reputation costs, are expected to affect the decision- making process of the leadership of that state due to one of the core assumptions of theories on reputation costs; that policymakers respond to public preferences about reputation (Brutger and Kertzer, 2018: 695). This has not only been found to be true for democratic leaders, which are incentivized to adhere to public opinion by the democratic mechanism of political accountability in the form of elections, but also for authoritarian regimes, in which other mechanisms of political accountability are present, such as domestic political groups willing to coordinate against leaders (Saunders, 2015: 469; Weeks, 2008: 59f).

Furthermore, previous research has found that naming and shaming by international actors has a significant effect on the repressive tactics states use against their own population (Karlén, 2019: 740), indicating that a concern for the state’s international reputation has an impact on the decision-making of the leadership. For instance, DeMeritt (2012: 616) finds that naming and shaming substantially decreases both the likelihood and the severity of one-sided government violence. She argues that it does so by entailing a risk of legal and economic punishment, for instance in the form of sanctions. Karlén (2019: 746), on the other hand, finds threats of sanctions to have no impact on the termination of external support to rebel groups.

Considering these diverging findings, among others, the plausibility of naming and shaming or the threats of sanctions to have an effect on external support provision to governments, is not dismissible.

(15)

To summarise, increased costs are expected to follow from continued support of a government committing one-sided violence, primarily in the form of reputation costs.

Reputation costs are expected to affect the behaviour of states by incurring domestic audience costs on leaders, by the general naming and shaming of states, and possibly by implying or resulting in the direct threat of sanctions. As the costs of continued support increases, leaders are expected to re-evaluate the initial cost-benefit analysis of the expected utility of continued support.

The third step of the causal mechanism, the re-evaluation of the cost-benefit analysis, is affected by the perceived costs and the expected strategic benefits of continued support.

Presumably, this entails some type of threshold in terms of accepted levels of one-sided violence. Arguably, it is not plausible that the killing of 25 civilians in a year would spark the same international reactions as the killing of 10,000 civilians in a year. However, it is difficult to theorize about any precise threshold, any precise number of killings, that would incur enough reputation costs to outweigh the strategic benefits of continued support. As previously mentioned, different levels of one-sided violence may be accepted in different situations, depending for instance on the expected international reactions to one-sided violence in different parts of the world. Furthermore, it is a question of proportionality. A sudden and large increase in one-sided violence may result in more reactions than would consistent yet relatively low levels of killings, although they may amass to the same levels over time. Also, as previously discussed the cost-benefit analysis is highly dependent on the importance the supporting state places on the outcome of the conflict, i.e. the weight of the strategic benefits of support. Unfortunately, finding any such threshold lies outside the scope of this thesis. It is however expected that at some point, the reputational costs of providing external support to a government responsible for increasing levels of one-sided violence will outweigh the strategic benefits of continued support. This in turn leads to the last step of the causal mechanism, the dependent variable; withdrawal of external support.

2.4 Causal diagram

Figure 1. Flow Chart of Causal Mechanism

(16)

2.5 Hypothesis

Based on the theoretical framework, the following hypothesis is proposed:

H1: One-sided violence committed by a government receiving support from a state external to the conflict increases the probability of said support being withdrawn.

3. Research design

The purpose of this thesis is to present and test a theory on the possible effects of one-sided violence on external support. The thesis aims at examining whether a generalisable correlation exists between one-sided violence and the termination of external support, building on the Uppsala Conflict Data Programme (UCDP) External Support - Disaggregated/Supporter Level Dataset (hereinafter referred to as the UCDP External Support Dataset), and the UCDP One-Sided Violence Dataset. As suggested by the theoretical framework, I expect one-sided violence committed by a government receiving support to cause the intervening state to terminate said support. The unit of analysis is dyad years, the dyad consisting of the provider and the recipient of external support. For the purpose of exploring the relationship under study, the quantitative method of logistic regression will be utilised, based on a merging of the two UCDP datasets mentioned. Considering that, to the best of my knowledge, little to no research has explored state support termination, a quantitative study is well suited for a first explorative study. First, a bivariate logistic regression will be carried out in order to test the

IV: One-sided violence

Increased reputation costs

DV: Withdrawal of external support

Re-evaluation of the cost-benefit analysis

(17)

relationship between the two phenomena, one-sided violence and withdrawal of support. This will be followed by multivariate regressions aimed at establishing the effects of one-sided violence on external support while controlling for other possible explanatory factors.

The merged dataset includes 780 observations of 56 countries receiving support from 61 external actors, covering the years 1989-2009. The temporal scope of the dataset is limited to the 20-year period due to the time frames of the merged datasets. However, the starting point at 1989 is not necessarily a disadvantage, as external support provision before the end of the Cold War was heavily influenced by the dynamics of geopolitical and ideological conflicts between the West and the East (Karlén, 2019: 745). The cut-off point at 2009 is set only for practical reasons, motivated by data availability, as the UCDP External Support Dataset only covers the years 1975-2009. Although a larger sample including the last decade would increase generalisability, the results are not expected to be different for the excluded years based on the proposed theory. Due to scope conditions set by the theory, cases of external support provided as part of a coalition have been excluded from the dataset, such as the interventions in Afghanistan and Iraq in 2003 and 2004 respectively. The argument is that reputation costs for a single state are assumed to be significantly lower when acting as part of a large coalition.

3.1 Operationalizations

3.1.1 The independent variable

The independent variable, One-sided violence, is operationalized in accordance with the UCDP definition cited in section 2.2.2; “the deliberate use of armed force by the government of a state or by a formally organised group against civilians which results in at least 25 deaths in a year” (UCDP, n.d.). This allows for an unaltered use of the continuous “best estimate”

variable of the UCDP One-Sided Violence Dataset. The variable consists of the aggregated most reliable figures for all incidents of one-sided violence in a calendar year. If reports differ in number of fatalities, an analysis of the reliability of sources is made by the coders,

according to which the lower number is included if no distinction in reliability can be made (Pettersson, 2019: 4). Relying on the rigorous work around the UCDP operationalization and UCDP’s data collection, the reliability of the measurement is arguably relatively high.

Although the relationship between reliability and validity involves a trade-off (Matters and

(18)

the narrow theoretical definition of one-sided violence cited in section 2.2.2 and above. If the aim was to capture general violence against civilians, the measurement would lack validity, as other types of violence than that resulting in death are excluded. It should be noted

nonetheless, as discussed in section 2.2.2, that other types of violence also could be of

theoretical interest. Essentially, the operationalization of the independent variable enjoys high validity in that it overlaps with the theoretical definition of the variable of interest.

The issue of time-order, or reversed causality, is of relevance to the study, as the analysed data is year-specific, rather than month-specific. In order to ensure that withdrawal of support that occurs in January, for instance, followed by a sudden upsurge of one-sided violence in December that same year, is not interpreted as an increase in one-sided violence resulting in withdrawal the same year, a constant of (+1) has been added to the year variable in the One- Sided Violence Dataset before merging the two datasets. By doing this, civilian fatalities for 2001 are matched with the status of support for 2002, etc. It is however an imperfect way of dealing with the issue of time-order. The analysis does not control for whether one-sided violence begun before or after external support was first provided to an actor, and so does not completely solve the issue of reversed causality.

3.1.2 The dependent variable

The operationalization of the dependent variable, Withdrawal of external support

(Withdrawal), builds on the UCDP External Support Dataset operational definition of external support, which includes secondary warring support, as well as secondary non-warring

support. The former refers to sending troops to assist a primary party in an ongoing conflict, and the latter refers to providing other types of support to assist a primary party in an ongoing conflict. Such non-warring support may include but is not excluded to the provision of

financial assistance, weapons, sanctuary, and logistics support (Croicu et al., 2011: 4). The UCDP dataset differentiates between cases of confirmed support, and alleged support. In the merged dataset utilised for this thesis, cases of alleged support have been removed, as the removal did not affect the sample size considerably but increased the generalisability of the results.

As the phenomenon of interest is the withdrawal of external support, rather than the existence of external support, the variable in the merged dataset has been re-coded in order to indicate whether support has been withdrawn or not. The dichotomous variable takes the value (0) to

(19)

indicate continued support, including the first year of observed support provision, and the value (1) when support is withdrawn. In cases where an actor provides support for at least one year, then stops providing support for two or more consecutive years, and then resumes to providing support, the last year of the first period of support is coded as (1) withdrawal. In instances of support being provided for only one year, that year will be coded as (1) withdrawal. Furthermore, the data has been cross-checked with the UCDP Conflict

Termination Dataset version 2-2015, so that support that ceases the same year as a conflict ends is coded as (0) continued support, rather than (1) withdrawal. However, as the merged dataset is not month- or date-specific, it is possible that cases in which support was withdrawn months before the end of the conflict are coded as (0) continued support.

The validity of the measurement is arguably quite high, as it includes both the direct and more indirect types of support included in the theoretical definition of external support.

Furthermore, similarly to the independent variable, the possibility to rely on the UCDP operationalization and data collection increases the reliability of the measurement. As

discussed in section 2.2.1, it would be of theoretical interest to not only focus on the complete withdrawal of support, but in terms of measurement reliability and validity, is most definitely an advantage that the phenomena of interest is simply support termination.

3.2 Control variables

With little previous research to rely on in terms of control variables, the variables have been derived from the proposed theory. Variables that logically can be assumed to affect the independent and dependent variable will thus also be tested in the multivariate model. Firstly, as suggested by the scope condition set out above, the reputation costs for an individual state are assumed to be lower when a coalition of states intervene. A similar argument can be made that reputation costs are lower when just a few more states provide support to the same

government. It is thus expected that as the number of supporters increase, the probability of Withdrawal = (1) decreases. Additionally, previous research suggest that the number of external supporters affects the level of rebel one-sided violence (Salehyan et al., 2014: 656), indicating that number of supporters may be a confounding variable explaining the variation in the independent variable as well. Therefore, a continuous variable accounting for the total number of intervening states during the same year will be controlled for. States that withdraw

(20)

Secondly, previous literature reveals possible differences in the expected relationship depending on the regime type of the external supporter. First, as briefly touched upon,

domestic audiences have been found to have a large impact on the foreign policy decisions of democratic leaders (Gottfried and Trager, 2016: 245; Brutger and Kertzer, 2018: 713). While similar conclusions have been drawn in regard to autocratic regimes, the notion that the same effects are expected in said regimes is less well established (Weeks, 2008: 36). Also, it is possible that different state characteristics are valued by domestic audiences in democracies as opposed to autocracies, which would affect the expected probability of withdrawal.

Second, autocratic governments have been found to undertake higher levels of one-sided violence than democratic ones, and rebel support from autocratic regimes has been found to cause a larger increase in the level of rebel perpetrated one-sided violence than support from democratic governments (Eck and Hultman, 2007: 234; Salehyan et al., 2014: 656). These findings indicate a greater tolerance for one-sided violence in autocratic regimes, but also the possibility that support from an autocratic would increase the level of government one-sided violence. This would affect both the value of the independent variable, the level of one-sided violence, and the dependent variable, the probability of withdrawal. Consequently, in order to control for possible effects of regime type on both the independent and dependent variable, a control variable indicating the level of democracy/autocracy of a supporting government is included in the multivariate model.

The Regime type variable is derived from the Polity IV Project polity indicator, a regime score ranging from +10 (full democracy) to -10 (full autocracy) (Marshall et al., 2019: 13).

Democracy is measured based on “the competitiveness of political participation /.../, the openness and competitiveness of executive recruitment /.../, and constraints on the chief executive” (ibid: 14), resulting in a democracy score ranging from 0-10 (ibid.). Likewise, autocracy is measured based the same three variables, and the additional variable “the regulation of participation” (ibid: 16). The polity indicator, used to measure Regime type in this thesis, is derived by subtracting the autocracy score from the democracy score, which provides a regime score ranging from +10 to -10 (ibid: 13).2

2For further information on the Polity indicator, please refer to the Polity IV Project Political Regime Characteristics and Transitions, 1800-2018, Dataset Users’ Manual, available at

<http://www.systemicpeace.org/inscr/p4manualv2018.pdf>

(21)

If a country is under foreign occupation during war, terminating the old polity, and then re- establishes a polity once occupation ends, Polity codes the years of intervention as an

interruption period, which in the dataset used for this thesis is coded as a missing value (ibid:

19). A so-called interregnum period, a period of complete collapse of central political authority, is coded as a neutral 0 in accordance with a modified Polity variable added to the Polity IV data series in order to convert the value of such periods into the conventional polity score (ibid: 8ff). In accordance with that same modified variable, transition periods are prorated across the transition period (ibid.) I.e., each year receives a score between the value before the transition and the value after the transition, increasing or decreasing each year depending on the direction of the transition. If such a modified value is missing in the Project IV data series, it is coded as a missing value in the merged dataset used for this thesis as well.

Lastly, the type of support provided may affect the outcome of the dependent variable. As previously mentioned, different types of support may be more or less likely to be withdrawn in the face of one-sided violence. For instance, troops may be more likely to be withdrawn in order to avoid being associated with, or even accused of, civilian targeting than for example intelligence or funding, which can be provided more covertly. Furthermore, it is possible that some types of support may be indicative of stronger incentives for continued support than others. For example, it is plausible that the contribution of one’s own troops indicates a stronger resolve than the provision of sanctuary, thus decreasing the probability of Withdrawal = (1). Therefore, based on the UCDP External Support Dataset, ten dummy variables indicating the type of support provided by an external actor are used to control for type of support; Troops as secondary warring party (Troops), Access to military or

intelligence infrastructures/Joint operations (Infrastructure/Joint operations), Access to territory (Territory), Weapons, Materiel/Logistics support (Materiel/Logistics), Training and expertise (Training/Expertise), Funding/Economic Support (Funding), Intelligence Material (Intelligence), Other forms of support (Other), and Support of unknown type (Unknown).

3.3 Data collection and reliability

The primary issue of data reliability in this thesis, pertains to the risk of under-reporting of one-sided violence. As the UCDP One-Sided Violence Dataset relies on news reports, the

(22)

receive a great deal of independent coverage, others do not. Normally following a pattern of centre versus periphery, within and between countries, news media presence is usually strong in geographic, economic and political centres, while weaker in the peripheries. Armed conflict on the other hand, usually follows the opposite pattern (Höglund and Öberg, 2011: 54).

Additionally, levels of reporting are affected by harsh terrains and poor infrastructure making the physical access of journalists to conflict areas limited (ibid.). Moreover, freedom of press is often restricted during armed conflict. Not only by authoritarian governments, but also by democratic governments, especially concerning conflict information (ibid: 55). The fact that the phenomenon of interest is government perpetrated one-sided violence, further exacerbates the risk of under-reporting of civilian fatalities following from censorship, especially in countries where the primary media outlets are under state ownership.

As previously touched upon, selection bias is also a risk, due to which one-sided violence in some conflicts risk going under-reported. The likelihood of an event becoming a news item is affected by several factors, including its frequency, continuity, and reference to ‘elite nations’, to name a few (ibid: 57). This selection bias affects the reporting of one-sided violence. For instance, conflicts during which civilian fatalities increase slowly over time are less likely to be reported on than those where a sudden increase occurs, or where one or a few major incidents of one-sided violence occur. Another example is that conflicts with featuring one- sided violence are less likely to be covered if no powerful nations are involved, resulting in a possible under-reporting of one-sided violence in less powerful nations. Overall, the UCDP datasets are considered highly reliable, but the risk of bias should be kept in mind, as it may affect the representativeness of the sample. The same applies to the reliability of the data on support provision. While there is a risk of covert external support being unreported,

government support is often provided overtly, even regulated by bilateral or multilateral agreements (Karlén, 2019: 735).

4. Results and analysis

The following section will proceed as follows; first, a summary of the descriptive statistics of all variables will presented. Second, the results of the bivariate and the multivariate logistic regressions respectively will follow. Third, the predicted probabilities are calculated and discussed based on the multivariate model, followed lastly by a robustness check. A substantial interpretation of the results will be made and discussed.

(23)

4.1 Descriptive statistics

An overview of the descriptive statistics of the variables reveal a number of 780 observations, with the exception of 765 for regime type due to missing values, see section 3.2.

Table 1. Descriptive statistics.

Statistic N Mean St.Dev. Min Pctl(25) Median Pctl(75) Max One-Sided

Violence

780 164.1 574.9 0 0 0 31 5,801

Withdrawal 780 0.2 0.4 0 0 0 0 1

Regime Type 765 3.6 7.3 -10 -4 9 10 10

No. Of Supporters 780 1.6 1.1 0 1 1 2 5

Troops 780 0.2 0.4 0 0 0 0 1

Infrastructure/

Joint Operations

780 0.1 0.2 0 0 0 0 1

Territory 780 0.1 0.2 0 0 0 0 1

Weapons 780 0.5 0.5 0 0 1 1 1

Materiel/Logistics 780 0.6 0.5 0 0 1 1 1

Training/Expertise 780 0.5 0.5 0 0 1 1 1

Funding 780 0.4 0.5 0 0 0 1 1

Intelligence 780 0.1 0.2 0 0 0 0 1

Other 780 0.01 0.1 0 0 0 0 1

Unknown 780 0.003 0.1 0 0 0 0 1

Note: St. Dev. = standard deviation. Pctl(25) = 1^st quartile range. Pctl(75) = 3^rd quartile range.

The statistics for the independent variable, One-sided violence, reveal a clear positive

skewness in the distribution, with a mean value of 164.1, significantly larger than value at the 3rd quartile range, 31. The value of the mean compared to the value at the 3rd quartile range indicates the existence of outliers, including the maximum value observation of 5,801 fatalities. Figures 2. and 3. below, clearly display the skewed distribution of number of fatalities, with the majority of observations taking the value of 0. The distribution indicates that the majority of conflict years during which a government receives external support, less than 25 civilians per year are killed as a result of one-sided violence. Figures 2. and 3. also reveal the outlier(s) in the sample expected from the interpretation of Table 1; one observation of 5,801 fatalities (the maximum value) and one of 4,160 fatalities.

(24)

Figure 2. Histogram of One-Sided Violence. Figure 3. Boxplot of Number of Fatalities.

The descriptive statistics of the dependent variable, withdrawal of external support, indicate a mode of 0, i.e. continued support. A mean of 0.2 and a value of 0 at the 3rd quartile range reveal that less than 25% of the observations take on the value of 1, i.e. withdrawal of support, indicating that continued support is far more common than withdrawal.

The statistics for regime type reveal that a majority of the supporting states in the sample have a polity score above 0, and at least 25% of the states are ‘full democracies’ with a regime score of +10. The Number of supporters variable has a mean of 1.6, and a median of 1, indicating the most common number of supporters during one year is one state. Furthermore, with the value at the 3rd quartile range being 2, it is indicated that the maximum value of 5 is uncommon, perhaps an outlier. Noteworthy is also the distribution of the first type of support variable, Troops, for which the statistics indicate that a majority of the sample observations, more than 75%, did not include contribution of troops. Instead, support more commonly took the form of indirect support. An overview of the statistics reveal that the most common type of external support provided is Materiel/Logistics, followed by Weapons and

Training/Expertise, all with a mean larger than or equal to 0,5.

4.2 Regression results

To summarise, the tested hypothesis expects that one-sided violence committed by a

government receiving support from a state external to the conflict increases the probability of

(25)

said support being withdrawn. The hypothesis is derived from the theoretical framework, according to which government one-sided violence leads to increased costs of continued support, which in turn cause the supporting state to re-evaluate the initial cost-benefit analysis, resulting in the withdrawal of the external support. In order to examine the relationship between the two phenomena, a logistic regression has been carried out. First, I present the results of the bivariate logistic regression in Model 1, including only the

independent variable, One-sided violence, and the dependent variable, Withdrawal. Second, two control variables are introduced in Model 2, Regime type and Number of supporters, followed by a third model introducing the types of support variables, thus including all control variables. The results from the logistic regressions are found below in Table 2.

In the first model (M1), the coefficient for the IV, one-sided violence, indicates a positive relationship between the IV and the DV. However, the results fail to achieve statistical significance at the 95% confidence level, and so the results based on the sample cannot be generalised to the underlying population. Following social sciences standard practice, the results are considered statistically significant when p < .05, which roughly equates a

confidence level of 95% (Kellstedt and Whitten, 2013: 149). The p-value is a measurement of the level of confidence in a systematic correlation between two variables; the lower the p- value, the higher the confidence level (ibid: 147). The values p < .10, p < .05, and p < .01 roughly corresponds to the respective confidence levels of 90%, 95% and 99%.

In the second model (M2) however, all three variables are statistically significant at the 99%

confidence level. The coefficient for one-sided violence shows a positive correlation between the IV and the DV, revealing that the probability of withdrawal increases with the increase in one-sided violence, when controlling for total number of supporters and regime type. The both control variables, Number of supporters and Regime type, are found to have a

statistically significant negative effect on the dependent variable. In line with the theoretical argument, this indicates that as the number of supporters increase, the probability of

withdrawal decreases, holding one-sided violence and regime type constant. Interestingly enough, it also reveals that as the Polity score increases, the probability of withdrawal decreases, indicating that more democratic states are less likely to withdraw support when one-sided violence increases than are less democratic ones.

(26)

Table 2. The Effect of One-Sided Violence on Withdrawal of External Support, Including Control Variables (Regression Analysis)

Dependent variable:

Withdrawal

M1 M2 M3

One-sided violence 0.0001 0.001*** 0.0005**

(0.0002) (0.0003) (0.0002)

Number of supporters -1.040^*** -1.144^***

(0.151) (0.163)

Regime type -0.080^*** -0.036^**

(0.014) (0.017)

Troops -0.513

(0.385)

Infrastructure/Joint operations 0.183

(0.473)

Territory -1.824^***

(0.606)

Weapons -0.392

(0.291)

Materiel/Logistics -1.015^***

(0.284)

Training/Expertise -1.313^***

(0.271)

Funding -1.155^***

(0.302)

Intelligence 0.258

(0.564)

Other 0.154

(1.250)

Unknown 12.656

(624.033)

Constant -1.578*** -0.099 1.676***

(0.099) (0.199) (0.341)

Observations 780 765 765

Log Likelihood -359.241 -300.246 -256.387

Akaike Inf. Crit. 722.483 608.491 540.774

Note: ^*^p<0.1; >**p<0.05; >***p<0.01

(27)

In Model 3 (M3), all control variables are introduced, making it the main model of analysis.

The coefficient for the independent variable, One-sided violence, drops to 0.0005, yet remains statistically significant at the 95% confidence level. Similarly, the two control variables Number of supporters and Regime type remain statistically significant, having a negative effect on withdrawal. Four types of support have a statistically significant effect on the probability of withdrawal, all negative; namely, Territory, Materiel/Logistics,

Training/Expertise, and Funding. These results indicate that the provision of each of the four types of support, i.e. the dummy variables taking the value of (1), decreases the probability of withdrawal of support.

Although no commonly accepted agreement exists on how to assess the goodness-of-fit of a logistic regression, some approaches are available. One widely accepted tool is the Akaike information criterion (AIC) (Cavanaugh and Neath, 2019: 1). The AIC estimates the out-of- sample prediction error, and thus the relative quality of a statistical model. The resulting AIC score rewards models achieving a high level of goodness-of-fit, while penalising them if becoming overly complex, thereby dealing with the trade-off between the goodness-of-fit and the parsimony of the model (ibid.). The AIC score by itself is of little use, if not compared with the score of a competing model. The model with the greater balance between its ability to fit the data and its ability not to over-fit the data is expected to have a lower AIC score (Date, 2019). In other words, the model with the lower AIC score has a higher goodness-of- fit. From Table 2., it is evident that Model 3 is the model of highest goodness-of-fit, with the lowest AIC score of the three models, although it does not tell us how well the model fits the data.

4.3 Predicted probabilities

Due to it being a logistic regression, the magnitude of the relationships is not directly

interpretable from the coefficients. Thus, the predicted probabilities have been calculated and presented in Table 3.

(28)

Table 3. Predicted Probability of Withdrawal of External Support

One-Sided Violence

Mean Min Pctl(25) Median Pctl(75) Max

Probability of Withdrawal (M3)

0.051 (5,1%)

0.047 (4,7%)

0.048 (4,8%)

0.453 (45,3%)

Note: continuous control variables are held at the mean, dichotomous control variables are held at the mode.

When the value of One-sided violence is set to its minimum value, i.e. 0, the continuous control variables are set to their respective mean values, and the dichotomous control variables to their mode, the predicted probability of Withdrawal = 1 is 4,7%. This indicates that in a year during which no one-sided violence is recorded, the probability of withdrawal in the following year is 4,7%. As the value of One-sided violence increases to its mean value of 164,1, the predicted probability of Withdrawal = 1 increases slightly to 5,1%. Note that the predicted probability of withdrawal remains at 4,7% when One-sided violence is set to its minimum, 2nd quartile, and median value, due to the skewness of distribution displayed in Table 1; all three have the value 0 fatalities.

A considerable increase is revealed when One-sided violence is set to its maximum value of 5,801 fatalities, increasing the predicted probability of withdrawal of support to 45,3%.

Firstly, these results support the hypothesis that as one-sided violence increases, the probability of withdrawal increases (H1). Secondly, and perhaps more interestingly, these results clearly indicate that the effect on the DV of a one-unit increase in the IV varies across values of the IV. At lower levels of one-sided violence, the probability of withdrawal is very low. At higher levels of one-sided violence on the other hand, the probability of withdrawal is significantly higher. For clarifying purposes, Figure 4. shows a graph displaying the predicted probability of withdrawal across all possible values of one-sided violence within the sample range. Similar to the Table 3., the graph indicates that the effect of one-sided violence on the probability of withdrawal of support is larger the higher the level of one-sided violence.

(29)

Figure 4. Predicted probability of Withdrawal = 1

4.4 Robustness check

Considering the uneven distribution of values of the DV across values of the IV, and the previously mentioned existence of two outliers, a fourth model has been tested as a robustness check. The observations for Afghanistan in 1999 (5,801 fatalities) and the Democratic

Republic of Congo (DRC) in 1998 (4,160 fatalities) have thus been removed from the sample, rendering a sample distribution seen in Figure 5.3 The distribution remains skewed, with a clear majority of observations taking on values lower than 500 fatalities, although the distribution is more even compared to Figure 3.

Figure 3. Boxplot of Number of Fatalities (Left). Figure 5. Boxplot of Number of Fatalities, Outliers Removed (Right).

3Note that a constant of +1 has been added to the year variable for one-sided violence, i.e. these are the fatality

(30)

Table 4. The Effect of One-Sided Violence on Withdrawal of External Support, Including Control Variables (Regression Analysis)

Dependent variable:

Withdrawal

M1 M2 M3 M4

One-sided violence 0.0001 0.001^*** 0.0005^** 0.002^***

(0.0002) (0.0003) (0.0002) (0.0004)

Number of supporters -1.040^*** -1.144^*** -1.271^***

(0.151) (0.163) (0.174)

Regime type -0.080*** -0.036** -0.038**

(0.014) (0.017) (0.018)

Troops -0.513 -0.491

(0.385) (0.393)

Infrastructure/Joint operations 0.183 0.365

(0.473) (0.479)

Territory -1.824^*** -1.661^***

(0.606) (0.616)

Weapons -0.392 -0.467

(0.291) (0.300)

Materiel/Logistics -1.015^*** -0.946^***

(0.284) (0.291)

Training/Expertise -1.313^*** -1.270^***

(0.271) (0.276)

Funding -1.155^*** -1.058^***

(0.302) (0.305)

One-Sided Violence and External Support: The effects of government one-sided violence on external state support provision